CLUES: A WEB-BASED LAND USE EXPERT SYSTEM FOR THE
WESTERN CAPE
ADRIAAN VAN NIEKERK
Dissertation presented for the degree of Doctor of Philosophy
at
Stellenbosch University
Promotor: Prof JH van der Merwe
December 2008
ii
DECLARATION
By submitting this dissertation electronically, I declare that the entirety of the work contained
therein is my own, original work, that I am the owner of the copyright thereof (unless to the
extent explicitly otherwise stated) and that I have not previously in its entirety or in part
submitted it for obtaining any qualification.
Date: December 2008
Copyright ? 2008 Stellenbosch University
All rights reserved
iii
SUMMARY
GIS has revolutionized geographic analysis and spatial decision support and has greatly
enhanced our understanding of the real world though it?s mapping and spatial modelling
capabilities. Although GIS software is becoming more powerful, less expensive and more user-
friendly, GIS still remains the domain of a sel ected few who can operate and afford these
systems. Since the introduction of web mapping tools such as Google Earth, accessibility to
geographic information has escalated. Such tools enable anyone with access to a computer and
the Internet to explore geographic data online and produce maps on demand. Web mapping
products have, however, a very narrow range of functionality. In contrast to GIS that focuses on
spatial data capture, storage, manipulation, analysis and presentation, the function of web
mapping tools is to visualize and communicate geographical data. The positive impact of web
mapping tools suggests, however, that GIS has not yet developed to a level where anyone can
use the technology to support spatial decisions and enhance productivity. A possible solution is
to close the functional gap between web mapping tools and GIS to make spatial analysis more
accessible, thereby promoting geographical awareness and supporting better spatial decisions.
In this research, a web-base d spatial decision support system (SDSS) was developed to
demonstrate how the Internet can be used to deliver low-cost, user-friendly and interactive
spatial analysis functionality to a wide audience. Although the resulting Cape Land Use Expert
System (CLUES) was specifically developed to pe rform land suitability analysis in the Western
Cape, the technology can also be applied to other regions and modified for other applications.
CLUES consists of five components: a land un it database (LUD), knowledge base, inference
engine, web map service (WMS) and graphical user interface (GUI). The LUD consists of
polygons (land units) and attributes (land properties), while the knowledge base stores each
user?s land use requirement rules. These rules are used by the inference engine to rate the
suitability of each land unit in the LUD. The re sult is then mapped by the WMS and presented to
the users as suitability maps. Users can direct the entire analysis through a user-friendly GUI.
The development and demonstration of CLUES e xposed several advantages and limitations of
current technology and has demonstrated that the Internet offers great opportunities for the
deployment of spatial analysis and modelling functionality to a wide audience.
KEY WORDS
Expert system, geographical information system, Internet, land evaluation, spatial decision
support system, suitability analysis, web mapping, Web
iv
OPSOMMING
GIS het geografiese analise en die ondersteun ing van ruimtelike besluitneming revolusion?r
verander en ons begrip van die werklike w?reld aansienlik versterk deur die karterings- en
ruimtelike modelleringsvermo?ns daarvan. Alhoe wel GIS programmatuur kragtiger, goedkoper
en meer gebruikersvriendelik raak, bly dit steeds die domein van uitverkorenes wat hierdie
stelsels kan bedryf en bekostig. Die bekendstelling van web karteringsgereedskap soos Google
Earth het toegang tot geografiese inligting grootliks verbreed. Sulke gereedskap laat enigeen met
rekenaar- en die Internettoegang toe om geogra fiese data aanlyn te ondersoek en kaarte op
aanvraag te maak. Web karteri ngsprodukte het egter ?n baie nou funksionele reikwydte. In
teenstellings met GIS, wat fokus op vasleggi ng, berging, manipulasie, analise en visuele
voorstelling van ruimtelike data, is die funksie van web karteringsgereedskap slegs om
geografiese data te visualiseer en te kommunikeer. Die positiewe impak van web
karteringsgereedskap dui egter daarop dat GIS nog nie tot die vl ak onwikkel het waar enigeen
toegang tot die tegnologie kan bekom om ruimtelike besluite te steun en produktiwiteit te
verbeter nie. ?n Moontlike oplossing is om die funksionele gaping tussen web
karteringsgereedskap en GIS uit te skakel en ruimtelike analise meer toeganklik te maak om
sodoende geografiese bewustheid te bevorder en ruimtelike besluitneming beter te ondersteun.
In hierdie navorsing is ?n web-gebaseerde ruimte like besluitsteunstelsel (RBSS) ontwikkel om te
toon hoe die Internet gebruik kan word om lae- koste, gebruikersvriendelike en interaktiewe
ruimtelik-analitiese funksionaliteit aan ?n wye gehoor te lewer. Alhoewel die voortvloeiende
Cape Land Use Expert System (CLUES) spesifiek vir grondgeskiktheidsanalise in die Wes-Kaap
ontwikkel is, kan die tegnologie aangewend word in soortgelyke gebiede of vir ander toepassings
aangepas word. CLUES bestaan uit vyf kompone nte: ?n landeenheid databasis (LED),
kennisbasis, afleidingsenjin, web kaartdiens (WKD) en grafiese gebruikerskoppelvlak (GGK).
Die LED bestaan uit poligone (l andeenhede) en attribute (land eienskappe), terwyl die
kennisbasis elke gebruiker se re?ls rakende grondgebruikvereistes hou. Die afleidingsenjin
gebruik hierdie re?ls om die geskiktheid van el ke landeenheid in die LED te skaal. Die WKD
karteer dan die resultaat en stel dit as geskiktheidskaarte aan die gebruikers voor. Gebruikers kan
die hele analise met behulp van die gebruikersvriendelike GGK rig.
Die ontwikkeling en demonstrasie van CLUES het die voordele en beperkinge van die huidige
tegnologie blootgel? en kon demonstreer dat die Internet groot geleenthede vir die ontplooiing
van ruimtelike analise en modelleringsfunksionaliteit aan ?n wye gehoor inhou.
v
TREFWOORDE
Deskundige stelsel, geografiese inligtingstelsel, Internet, grond/land evaluering, ruimtelike
besluitsteunstelsel, geskiktheidsanalise, web kartering, Web
vi
ACKNOWLEDGEMENTS
I sincerely thank:
? Helene, my wife, for her support, understanding and patience throughout this project.
? The staff of the Department of Geology, Geography and Envi ronmental Studies, for their
continued support and guidance.
? Dr Pieter de Necker, for his editorial work on this document, insightful suggestions and
for keeping me motivated.
? Mr Bennie Schloms, who was always willing to provide advice and share his expertise.
? Prof Hannes van der Merwe, my promotor, for his continued motivation, support and
timely suggestions.
vii
CONTENTS
DECLARATION..................................................................................................... ii
SUMMARY.............................................................................................................iii
OPSOMMING........................................................................................................ iv
ACKNOWLEDGEMENTS................................................................................... vi
CONTENTS ...........................................................................................................vii
TABLES ................................................................................................................xiii
FIGURES .............................................................................................................. xiv
ACRONYMS AND ABBREVIATIONS...........................................................xvii
CHAPTER 1: TOWARDS IMPROVED SPATIAL DECISION MAKING.. 1
1.1 GEOGRAPHICAL INFORMATION SYSTEMS: AN OVERVIEW .......................2
1.2 SPATIAL DECISION SUPPORT SYSTEMS AND GIS............................................4
1.3 THE INTERNET AND GIS COMMUNICATION.....................................................6
1.4 WEB MAPPING FOR SPATIAL DECISION SUPPORT .........................................7
1.5 RESEARCH PROBLEM FORMULATION ...............................................................7
1.6 RESEARCH AIM AND OBJECTIVES ....................................................................... 9
1.7 THE STUDY REGION ................................................................................................10
1.8 RESEARCH METHODOLOGY AND AGENDA ....................................................11
CHAPTER 2: THE PRINCIPLES AND PRACTICE OF LAND
SUITABILITY ANALYSIS...................................................... 14
2.1 LAND EVALUATION APPROACHES.....................................................................14
2.1.1 Set objectives........................................................................................................15
2.1.2 Collect data ..........................................................................................................16
2.1.3 Identify land uses.................................................................................................16
2.1.4 Specify land use requirements ...........................................................................16
2.1.5 Map land units.....................................................................................................17
2.1.6 Determine land properties spatially ..................................................................18
2.1.7 Analyse land units for suitability .......................................................................18
2.1.8 Present results......................................................................................................18
2.2 BOOLEAN OVERLAY................................................................................................19
2.3 MULTI-CRITERIA DECISION MAKING...............................................................21
2.3.1 The MCDM procedure .......................................................................................22
viii
2.3.1.1 Set objectives ................................................................................................ 22
2.3.1.2 Select appropriate criteria.............................................................................. 23
2.3.1.3 Map spatial criteria........................................................................................ 23
2.3.1.4 Standardize measurements ............................................................................ 23
2.3.1.5 Set criteria weights ........................................................................................ 27
2.3.1.6 Multi-criteria evaluation (MCE) ................................................................... 29
2.3.1.7 Multi-objective evaluation (MOE)................................................................ 30
2.3.2 Discussion.............................................................................................................30
2.4 EXPERT SYSTEMS.....................................................................................................32
2.4.1 The rulebase.........................................................................................................33
2.4.2 The inference engine ...........................................................................................36
2.4.3 Land unit database..............................................................................................37
2.4.4 Existing land evaluation systems .......................................................................37
2.5 SUMMARY ...................................................................................................................39
CHAPTER 3: WEB MAPPING TECHNOLOGY.......................................... 40
3.1 WEB APPLICATIONS ................................................................................................40
3.2 WEB COMPONENTS..................................................................................................41
3.2.1 Web browsers ......................................................................................................42
3.2.2 Markup languages...............................................................................................42
3.2.3 Client-side scripting ............................................................................................44
3.2.4 Server-side scripting ...........................................................................................45
3.2.5 Web servers..........................................................................................................46
3.2.6 Relational database management systems ........................................................47
3.3 WEB MAPS ...................................................................................................................49
3.3.1 Characteristics of web maps...............................................................................49
3.3.1.1 Static vs. dynamic ......................................................................................... 49
3.3.1.2 Interactive vs. view only ............................................................................... 50
3.3.1.3 Distributed..................................................................................................... 50
3.3.1.4 Animated ....................................................................................................... 50
3.3.1.5 Reusable ........................................................................................................ 50
3.3.1.6 Real-time ....................................................................................................... 50
3.3.1.7 Personalized .................................................................................................. 51
3.3.1.8 Collaborative ................................................................................................. 51
3.3.2 Formats of web maps ..........................................................................................51
3.3.2.1 Raster............................................................................................................. 51
ix
3.3.2.2 Vector ............................................................................................................ 52
3.4 WEB MAP SERVICES ................................................................................................53
3.4.1 Data storage and retrieval ..................................................................................53
3.4.2 Map production...................................................................................................55
3.4.3 Map distribution..................................................................................................56
3.5 SUMMARY ...................................................................................................................56
CHAPTER 4: REQUIREMENT ANALYSIS AND DESIGN OF CLUES .. 58
4.1 SYSTEM REQUIREMENTS ......................................................................................58
4.1.1 Functional needs..................................................................................................58
4.1.2 Operational characteristics ................................................................................59
4.1.2.1 Accessibility .................................................................................................. 59
4.1.2.2 Performance .................................................................................................. 59
4.1.2.3 Presentation ................................................................................................... 60
4.1.3 Data requirements...............................................................................................61
4.1.3.1 Operational scale ........................................................................................... 61
4.1.3.2 Climate data................................................................................................... 61
4.1.3.3 Soil characteristics......................................................................................... 63
4.1.3.4 Terrain types.................................................................................................. 65
4.1.3.5 Infrastructure attributes ................................................................................. 67
4.1.3.6 Current land cover and use............................................................................ 67
4.1.4 Summary..............................................................................................................67
4.2 SYSTEM DESIGN........................................................................................................69
CHAPTER 5: LAND PROPERTY DATA COLLECTION .......................... 72
5.1 TERRAIN DATA..........................................................................................................72
5.1.1 DEM selection criteria ........................................................................................72
5.1.2 Existing Western Cape DEM .............................................................................74
5.1.3 DEM accuracy assessment..................................................................................75
5.2 WESTERN CAPE SOIL INFORMATION ...............................................................77
5.2.1 Existing soil data..................................................................................................77
5.2.1.1 Detailed surveys ............................................................................................ 77
5.2.1.2 Semi-detailed surveys ................................................................................... 77
5.2.1.3 Reconnaissance surveys ................................................................................ 77
5.2.1.4 Soil database strategy.................................................................................... 78
5.2.2 Land types information system..........................................................................78
5.2.2.1 Digital data structure ..................................................................................... 80
x
5.2.2.2 Soil property extraction................................................................................. 82
5.3 WESTERN CAPE CLIMATE INFORMATION......................................................87
5.3.1 Existing climate data...........................................................................................87
5.3.2 Comparison of data quality................................................................................88
5.4 SUMMARY ...................................................................................................................89
CHAPTER 6: DEVELOPING THE LAND UNIT DATABASE................... 91
6.1 LAND COMPONENT MAPPING TECHNIQUES ..................................................91
6.1.1 Terrain analysis ...................................................................................................93
6.1.2 Geomorphometry ................................................................................................93
6.1.3 Automated mapping............................................................................................94
6.2 Automated component mapping with ALCoM..........................................................95
6.3 Image processing techniques........................................................................................98
6.3.1 Image clustering ..................................................................................................98
6.3.2 Image segmentation.............................................................................................99
6.4 Comparison of ALCoM and MRS.............................................................................103
6.5 SEGMENTING THE WESTERN CAPE.................................................................103
6.6 LAND PROPERTY EXTRACTION ........................................................................104
6.7 STORAGE ...................................................................................................................105
6.8 SUMMARY .................................................................................................................105
CHAPTER 7: DEVELOPING THE KNOWLEDGE BASE ....................... 107
7.1 LOGICAL DATA MODELLING METHODOLOGY...........................................107
7.1.1 Identify major modelling entities.....................................................................108
7.1.2 Determine operational relationships between entities ...................................108
7.1.3 Identify primary keys........................................................................................109
7.1.4 Define foreign keys ............................................................................................110
7.1.5 Determine key business rules ...........................................................................110
7.1.6 Add remaining non-key attributes...................................................................110
7.1.7 Normalize data structure..................................................................................112
7.1.8 Specify additional attribute business rules .....................................................113
7.1.9 Combine user views...........................................................................................114
7.1.10 Integrate with existing data models.................................................................114
7.2 IMPLEMENTATION ................................................................................................115
CHAPTER 8: DEVELOPMENT OF THE CLUES WEBSITE.................. 117
8.1 THE INFERENCE ENGINE.....................................................................................117
8.1.1 Suitability calculation procedure.....................................................................117
xi
8.1.2 Membership value calculation .........................................................................118
8.1.2.1 Represent rules as linear functions.............................................................. 118
8.1.2.2 Determine membership function equations................................................. 119
8.1.3 Suitability value calculation .............................................................................121
8.1.4 Inference engine algorithm...............................................................................122
8.2 THE WEB MAP SERVICE AND WEB SERVER.................................................. 124
8.2.1 Choice of software and hardware ....................................................................124
8.2.2 WMS configuration...........................................................................................125
8.2.2.1 ArcIMS overview........................................................................................ 125
8.2.2.2 ArcIMS and web server configuration ........................................................ 127
8.3 Development of a user-friendly GUI .........................................................................128
8.3.1 GUI structure.....................................................................................................129
8.3.2 Login & security module ..................................................................................133
8.3.3 Menu module .....................................................................................................135
8.3.4 User details module ...........................................................................................135
8.3.5 Rulebase module................................................................................................135
8.3.5.1 Land uses specification ............................................................................... 136
8.3.5.2 Land use requirements specification ........................................................... 137
8.3.5.3 Land use requirement rules specification.................................................... 138
8.3.6 Projects module .................................................................................................139
8.3.7 Analyse & map module.....................................................................................140
CHAPTER 9: DEMONSTRATIONS OF CLUES........................................ 143
9.1 SETTING RULES FOR PERENNIAL CROPS......................................................144
9.1.1 Terrain requirement rules................................................................................144
9.1.1.1 Slope gradient rules..................................................................................... 144
9.1.1.2 Aspect, curvature and elevation .................................................................. 148
9.1.2 Climate requirement rules................................................................................149
9.1.2.1 Chill units rules ........................................................................................... 150
9.1.2.2 Heat units rules............................................................................................ 151
9.1.2.3 Frost occurrence rules ................................................................................. 152
9.1.2.4 Rainfall rules ............................................................................................... 154
9.1.3 Soil requirement rules.......................................................................................156
9.1.3.1 Soil clay content rules ................................................................................. 156
9.1.3.2 Effective soil depth rules............................................................................. 157
9.1.4 Current land uses and wetlands requirement rules.......................................159
xii
9.1.4.1 Urban areas rules......................................................................................... 159
9.1.4.2 Conservation areas rules.............................................................................. 159
9.1.4.3 Wetlands rules ............................................................................................. 160
9.2 WEIGHTING SUITABILITY FACTORS...............................................................161
9.3 CASE STUDY SUITABILITY ANALYSIS AT VARYING SCALES..................163
9.3.1 Greater Cape Town...........................................................................................164
9.3.2 Stellenbosch........................................................................................................164
9.3.3 Swartland ...........................................................................................................166
9.3.4 Rural Malmesbury ............................................................................................166
9.4 DISCUSSION ..............................................................................................................167
CHAPTER 10: EVALUATION OF THE RESEARCH................................. 169
10.1 ASSESSMENT OF CLUES .......................................................................................169
10.1.1 Functionality requirements ..............................................................................170
10.1.2 Operational performance requirements .........................................................172
10.1.2.1 System accessibility .................................................................................... 172
10.1.2.2 Operational speed........................................................................................ 172
10.1.2.3 System user-friendliness ............................................................................. 173
10.1.2.4 Data storage mode and capacity.................................................................. 174
10.1.3 Data requirements.............................................................................................174
10.1.3.1 Soil data....................................................................................................... 174
10.1.3.2 Terrain data ................................................................................................. 175
10.1.3.3 Climate data................................................................................................. 176
10.1.4 Scale, scalability and flexibility requirements ................................................176
10.2 POTENTIAL OF WEB TECHNOLOGY FOR SDSS DEVELOPMENT............178
10.3 RESEARCH OBJECTIVES REVISITED ...............................................................180
10.4 CONCLUSION............................................................................................................181
REFERENCES .................................................................................................... 183
PERSONAL_COMMUNICATIONS................................................................ 201
APPENDICES ..................................................................................................... 202
xiii
TABLES
Table 2-1 Scale of analytical hierarchy process (AHP) comparisons.........................................28
Table 2-2 AHP comparison matrix for perennial crops ..............................................................28
Table 2-3 Example of five Boolean rules specifying effective soil depth requirements for
perennial crops .......................................................................................................... 34
Table 2-4 Example of two Boolean and five fuzzy rules specifying effective soil depth
requirements for perennial crops............................................................................... 35
Table 4-1 The data requirements of CLUES............................................................................... 68
Table 5-1 Vertical error in the WCDEM and the STRM DEM.................................................. 76
Table 5-2 Accuracy summary of existi ng climatic data for the Western Cape........................... 88
Table 5-3 Data sets collected for CLUES ................................................................................... 89
Table 6-1 Statistical co mparison of land components mapped using ALCoM and MRS ........ 101
Table 7-1 Relationship matrix of entities in the knowledge base ............................................. 109
Table 7-2 Entity attribut es in the knowledge base .................................................................... 111
Table 7-3 Business rules for US ER entity in the knowledge base............................................ 113
Table 8-1 Suitability level factors .......................................................................................... ... 118
Table 8-2 Interpretations of land unit suitability values ........................................................... 122
Table 8-3 Tools available for manipulating the map frame ...................................................... 141
Table 9-1 Calculated suitability values and levels for selected slope values............................ 147
Table 9-2 AHP comparison matrix of weights assigned to land use requirements................... 162
xiv
FIGURES
Figure 1-1 The Western Ca pe province, South Africa................................................................ 11
Figure 1-2 Research design for developing CLUES, a web-based SDSS for the Western Cape 12
Figure 2-1 The steps in a land evaluation process ...................................................................... 15
Figure 2-2 Intersect and uni on Boolean overlay operations .......................................................20
Figure 2-3 A step-wise procedur e for multi-criteria decision making ........................................ 22
Figure 2-4 Linear scali ng of effective soil depth for perennial crop suitability.......................... 24
Figure 2-5 Ascending (a) and desc ending (b) S-membership functions..................................... 26
Figure 2-6 Effective soil dept h (a) and soil acidity (b) membership functions for perennial crops
................................................................................................................................... 26
Figure 2-7 Boolean constraint of e ffective soil depth for perennial crops.................................. 27
Figure 2-8 Graduated shades used to visual ize suitability levels of factors and results ............. 30
Figure 2-9 Procedure for an expert system land suitability analysis........................................... 33
Figure 2-10 Levels of suitabili ty of effective soil depths for perennial crops using Boolean
classification.............................................................................................................. 34
Figure 2-11 Levels of suitabili ty of effective soil depths for perennial crops using fuzzy
classification .............................................................................................................. 35
Figure 3-1 Example of a HTML document................................................................................. 42
Figure 3-2 Example of an email stored as XML ......................................................................... 43
Figure 3-3 Example of JavaScri pt code that displays an error message when the web page is
opened. ...................................................................................................................... 45
Figure 3-4 A SQL statement usi ng the SELECT and JOIN operators........................................ 48
Figure 3-5 Classifi cation of web maps........................................................................................ 49
Figure 4-1 Soil texture triangl e showing the twelve major textural classes and particle size
scales ......................................................................................................................... 64
Figure 4-2 The components of CLUES....................................................................................... 69
Figure 5-1 Selection of reference poin ts used in the DEM accuracy assessment ....................... 75
Figure 5-2 Land type Ca6 in memoir format .............................................................................. 79
Figure 5-3 Land type polygon (a) and its associated attribute information (b)...........................80
Figure 5-4 Land type database structure .....................................................................................80
Figure 5-5 Conceptual view of land type object levels............................................................... 82
Figure 5-6 Effective soil dept h derived from land type data....................................................... 84
Figure 5-7 Soil clay content derived from land type data ........................................................... 85
Figure 5-8 Soil mechanical limitati ons derived from the land type data .................................... 86
xv
Figure 6-1 Two hypothetical hills lopes, each consisting of a sequence of five land components
................................................................................................................................... 92
Figure 6-2 The ALCoM algorithm.............................................................................................. 96
Figure 6-3 Location of the test area for ALCoM ........................................................................ 97
Figure 6-4 Detailed view of the test area, with selected terrain features indicated..................... 97
Figure 6-5 Land component boundaries mapped by ALCoM .................................................... 97
Figure 6-6 A conceptual comparison of clusters (a) and segments (b) in relation to attributes A
to H ............................................................................................................................ 99
Figure 6-7 Land components mapped by multi-resolution segmentation................................. 102
Figure 7-1 Logical data model diagram of CLUES knowledge base ....................................... 115
Figure 8-1 Levels of suitability of effective soil depths for perennial crops using linear fuzzy
classification............................................................................................................ 119
Figure 8-2 Symmetrical (a) and asymmetrical (b) fuzzy rules dec onstructed to two lines, A and
B. ............................................................................................................................. 119
Figure 8-3 Inferen ce engine algorithm...................................................................................... 123
Figure 8-4 ArcIMS components................................................................................................ 125
Figure 8-5 Main steps followed to produce a suitability map using CLUES ........................... 129
Figure 8-6 GUI pages and linkages in CLUES ......................................................................... 131
Figure 8-7 Code of the main page showing interaction between HTML, JavaScript and Visual
Basic elements ......................................................................................................... 132
Figure 8-8 Index page of the login & security module ............................................................. 133
Figure 8-9 Register page of the login & security module ......................................................... 134
Figure 8-10 Register_success page of the login & security module ......................................... 134
Figure 8-11 Main and banner menus as shown on the main page ............................................ 135
Figure 8-12 A list of land uses ow ned by the current user shown on the landuses page.......... 136
Figure 8-13 Requirements fo r land use A as listed on the req page ......................................... 137
Figure 8-14 Requirement rules for a hypot hetical land use as displayed on the rules page ..... 138
Figure 8-15 Information about a user?s projects shown on the projects page .......................... 139
Figure 8-16 CLUES map viewer............................................................................................... 140
Figure 9-1 The landuse_create page used to create a new land use called ?perennial crops? .. 144
Figure 9-2 Slope gradient (%) added as a requirement for perennial crops ............................. 145
Figure 9-3 Rules defining suitabi lity levels for slope gradient ................................................. 146
Figure 9-4 Satellite image of the AOI ....................................................................................... 147
Figure 9-5 Slope requirement map of the AOI ......................................................................... 148
Figure 9-6 Suitability resu lts overlaying a satellite image for orientation purposes................. 149
xvi
Figure 9-7 Chill units rules................................................................................................. ....... 150
Figure 9-8 Chill units requirement map of the AOI.................................................................. 151
Figure 9-9 Heat units rules .................................................................................................. ...... 152
Figure 9-10 Heat units re quirement map of the AOI ................................................................ 153
Figure 9-11 Frost requirement rules.......................................................................................... 153
Figure 9-12 Frost requ irement map of the AOI ........................................................................ 154
Figure 9-13 Mean annual rainfall requirement rules................................................................. 155
Figure 9-14 Mean annual rain fall requirement map of the AOI ............................................... 155
Figure 9-15 Soil clay content requirement rules ....................................................................... 157
Figure 9-16 Soil clay conten t suitability map of the AOI ......................................................... 157
Figure 9-17 Effective soil depth requirement rules................................................................... 158
Figure 9-18 Effective soil depth requirement map of the AOI ................................................. 159
Figure 9-19 Urban areas in the AOI considered permanently unsuitable ................................. 160
Figure 9-20 Conservation areas in the AOI considered permanently unsuitable...................... 161
Figure 9-21 Wetlands in the AOI considered permanently unsuitable ..................................... 161
Figure 9-22 List of perennial crop requirements with weights shown on the req page............ 163
Figure 9-23 Suitability map for perennial crops in the greater Cape Town AOI...................... 164
Figure 9-24 A compilation of CLUES screen ca ptures showing detailed maps of (a) the
suitability analysis result, (b) the suitability overlay, (c) land unit outlines, and (d)
the satellite image of the area north of Stellenbosch............................................... 165
Figure 9-25 A compilation of CLUES screen capt ures showing (a) the satellite image and (b)
the suitability analysis results for the Swartland area ............................................. 166
Figure 9-26 A compilation of CLUE S screen captures showing detailed maps of (a) the satellite
image and (b) the suitability analysis results for the area west of Malmesbury ..... 167
xvii
ACRONYMS AND ABBREVIATIONS
ADF Application development framework
ADO ActiveX data objects
AEZWIN Agro-Ecological Zone for Windows
AHP Analytical hierarchy process
AHPP AHP Program
ALCoM Automated Land Component Mapper
ALES Automated Land Evaluation System
AOI Area of interest
API Application programming interface
ARPANET Advanced Research Projects Agency Network
ASP Active server pages
BDE Borland database engine
CAD Computer-aided design
CAPE Cape Action for People and the Environment
CDSM Chief Directorate Surveys and Mapping
CEMS Cape Environmental Management System
CFML ColdFusion markup language
CFR Cape floristic region
CGA Centre for Geographical Analysis
CGI Common gateway interface
CGIS Canada Geographic Information System
CLUES Cape Land Use Expert System
COM Component object model
CPU Central processing unit
CR Consistency ratio
CSIR Council for Scientific and Industrial Research
CU Chill units
DBMS Database management system
DEM Digital elevation model
DHTML Dynamic hypertext markup language
DIME Dual independent map encoding
DSS Decision support systems
xviii
EROS Earth Resources Observation and Science
ESRI Environmental Systems Research Institute
FAO Food and Agriculture Organization
FMF Fuzzy membership function
GDD Growing degree-days
GIF Graphics interchange format
GIS Geographical information system
GPS Global positioning system
GRASS Geographic resources analysis support system
GUI Graphical user interface
HTML Hypertext markup language
HTTP Hypertext transfer protocol
HU Heat units
IBM International Business Machines
IIS Internet Information Services
IP Internet protocol
ISCW Institute of Soil, Climate and Water
ISLE Intelligent System for Land Evaluation
ISODATA Iterative self-organizing da ta analysis technique algorithm
JPEG Joint photographic experts group
LAN Local area network
LBS Location based services
LDM Logical data modelling
LDMD Logical data model diagram
LEIGIS Land Evaluation using an Intelligent Geographical Information System
LIDAR Light detecting and ranging
LT Land type
MADM Multi-attribute decision making
MAE Mean absolute error
MCA Multi-criteria analysis
MCDM Multi-criteria decision making
MCE Multi-criteria evaluation
MicroLEIS Mediterranean Land Ev aluation Information System
MODM Multi-objective decision making
MOE Multi-objective evaluation
xix
MRS Multi-resolution segmentation
MS Microsoft
MS-DOS Microsoft Disk Operating System
NASA National Aeronautics and Space Administration
NDA National Department of Agriculture
ODBC Open database connection
OGC Open Geospatial Consortium
OS Open source
OWA Ordered weighted averaging
PDF Portable document format
PHP PHP hypertext preprocessor
PNG Portable network graphics
RADAR Radio detecting and ranging
RAM Random access memory
RDBMS Relational database management system
RGB Red, green and blue
RLE Run length encoding
RMSE Root mean square error
SAAAC South African atlas of agrohydrology and -climatology
SANBI South African National Biodiversity Institute
SAWS South African Weather Services
SAWIS South African Wine Information & Systems
SDE Spatial data engine
SDSS Spatial decision support systems
SGML Standard generalized markup language
SGV Slope gradient variance
SQL Structured query language
SRTM Shuttle Radar Topography Mission
SVG Scalable vector graphics
SWF Shock wave flash
TIFF Tagged image file format
TOC Table of contents
TU Terrain units
TWI Topographical wetness index
UNEP United Nations Environment Programme
xx
URL Uniform resource locator
USGS United States Geological Survey
UTM Universal transverse Mercator
WCCG Western Cape climate grids
WCDEM Western Cape Digital Elevation Model (20m resolution)
WCDEM80 Western Cape Digital Elevation Model (80m resolution)
WLC Weighted linear combination
WMS Web map service
WWW World Wide Web
XHTML Extensible hypertext markup language
XML Extensible markup language
1
CHAPTER 1: TOWARDS IMPROVED SPATIAL DECISION MAKING
? G IS is an unfinished revolution?Now, more than ever, geographic awareness, thinking and
curiosity remain the key to getting the best from a tool that can help support Earth-changing
decision making. ? (Van Wyngaarden & Waters 2007: s.p.).
During their forty years of existence, geographical information systems (GIS) have
revolutionized geographical anal ysis and spatial decision support and greatly enhanced our
understanding of the real world through their mapping and spatial modelling capabilities.
Although the GIS industry is contin uing to grow as the software becomes more powerful, less
expensive and more user-friendly GIS remain th e domain of a select few who can operate and
afford these systems.
Since the introduction of web mapping tools such as Yahoo Maps , Microsoft Virtual Earth and
NASA World Wind, accessibility to geographical information has escalated. Web mapping
software enables anyone with access to a computer and the Internet to explore geographical data
online and produce maps on demand. An increasingly popular web mapping tool is Google Earth
which bundles seamless satellite imagery and other geographical information into a tool that is
simple and easy to use. Unlike GIS that require users to know how data is acquired, prepared,
stored and queried, Google Earth allows users to use the geotechnology in an intuitive,
interactive manner.
Web mapping products such as Google Earth have a very narrow range of functionality. In
contrast to GIS that focus on spatial data capture, storage, manipulation, analysis and
presentation, the function of web mapping tools is to visualize and communicate geographical
data in a highly user-friendly manner. The pos itive impact of web mapping tools on geographical
awareness, as well as the popularity of these systems, suggest that GIS have not ye t developed to
a level where all professionals can use the technology to support spatial decisions and enhance
productivity. GIS functionality is not yet a ccessible enough to significantly enhance the
geographical awareness of the general public. According to Van Wyngaarden & Waters (2007)
GIS should do more to encourage geographical th inking and to demonstrate the principles of
geographical science. The functi onal gap between web mapping tools and GIS should be closed
to make GIS functionality more accessible to a ll professionals and the public in general. The
adoption of these capabilities by a wider audience will promote geographical awareness and
ultimately lead to better spatial decisions.
This chapter sets out to provide critical pers pective concerning GIS and its development as
spatial decision support system, lately also via the Internet and web mapping. It concludes with
2
formal statements of the research problem, aim and objectives, describes the study region and
introduces the research methodology applied in the dissertation.
1.1 GEOGRAPHICAL INFORMATION SYSTEMS: AN OVERVIEW
The value of computer technology for spatial applic ations was first recognized in the early 1960s
when Canada?s Department of Forestry and Ru ral Development set out to map and compile an
inventory of its natural resources in order to manage land more effectively. Realizing that the
manual production of the maps required considerable time and funds, the project team decided to
use computer technology. In conjunction with IBM, the fi rst GIS ? called the Canada
Geographic Information System (CGIS) ? wa s developed and completed in 1963 (DeMers
2005).
In 1967 the United States government needed a sim ilar tool to support the automatic referencing
and aggregation of their census records. The dual independent map encoding (DIME) program
was developed and, although the applications for which CGIS a nd DIME were developed were
quite different, the many similarities were quickly recognized. As a result, the Harvard
Laboratory for Computer Graphi cs and Spatial Analysis, under direction of Howard Fisher,
developed the first general purpose GIS, called ODYSSEY GIS, which was completed in 1977
(Longley et al. 2002). Other GIS followed.
The computer systems on which early GIS opera ted were prohibitively expensive and only a
select few governmental and educational institutions could afford ownership. The dramatic fall
in prices of computer hardware in the 1980s changed this. Smaller institutions started to
implement GIS and natural resources agencies pa rticularly, such as the Environmental Systems
Research Institute (ESRI), drove the development of better software. During the 1990s hardware
and software continued to be improved, leading to further price reductions and an explosion in
the number of users and applications (Longley et al. 2002). Today geospatial technology, along
with nanotechnology and biotechnology, is one of the three most important emerging industries
in the United States (U.S. Department of Labor 2006).
Because of the diverse applications of GIS there is no generally accepted definition for the
technology. Opinions about what is essential for a system to be called a GIS differ according to
each user?s needs. Most users agree, however, that GIS are special computer systems that
capture, store, query, analyse and display geographically referenced data (Chang 2006; Clarke
2003; DeMers 2005). But the property which truly di fferentiates GIS from other spatial systems,
such as Computer-aided design (CAD), is their ability to analyse spatial data (Clarke 2003).
Spatial analysis is defined by Longley et al. (2 002: 278) as ?the proces s by which we turn raw
3
spatial data into useful information?[to]?add valu e, support decisions, and reveal patterns and
anomalies that are not immediately obvious.?
In order to conduct spatial analysis, a spatial database is required. Geographical data is expensive
to collect and capture and can be the most expensive component of GIS. Fortunately GIS are
maturing and more data is becoming available and many governments, including that of South
Africa, have made state-owned data freely acces sible to the public (South Africa 2000). In spite
of the improved policies and advances in technologies to aid data capture, such as remote
sensing and global positioning systems (GPS), the establishment of spatial databases still
impedes many GIS implementations.
For a GIS to operate, a computer system is required. The early GIS ran on bulky mainframes,
which were later replaced with UNIX workstations and since th e mid-1990s workstations have
made way for personal (micro) computers running Microsoft (MS) Windows. Today nearly three
quarters of GIS professionals us e MS Windows, while only one quarter use UNIX-based systems
(GISjobs.com 2006). This trend is continuing, primarily because of the affordability and
computing power of modern personal computers.
Besides hardware, a GIS also requires specialized software to process spatial data. GIS software
became considerably more affordable during the 1990s, mainly due to greater demand and
technological advances. Even some free GIS so ftware packages, such as GRASS (GRASS 2006)
and LandSerf (Wood 2006), have become available. In spite of the availability of free GIS
software, most GIS professionals use proprietary GIS software products. More than three
quarters of GIS professionals currently use eith er ArcView or ArcGIS developed by ESRI, while
Autodesk, Intergraph and other GIS software deve lopers compete for the remaining market share
(GISjobs.com 2006). The drop in prices, together with advances in providing user-friendly
interfaces, are perceived to be the major reasons why GIS software sales more than doubled from
1995 to 2000 (Longley et al. 2002).
Undoubtedly the most important and least valued component of GI S are the people, also called
?brainware? (Chang 2006), who use a nd manage these systems. GIS ar e often regarded as a set of
spatial tools (Clarke 2003) and it is the GIS opera tor?s function to choose th e correct set of tools
for a specific application and to use these tools in the correct order and manner to achieve the
desired results. A GIS is us eless without human input.
The number of GIS users, estimated in 2000 to have been about three million worldwide, is
rapidly increasing (Longley et al. 2002). GIS ha ve become more accessible mainly owing to the
lower costs of GIS software and ha rdware and the advent of intuitive, user-friendly desktop GIS.
With these systems users can perform many spa tial tasks with little GI S training, thus enabling
4
many managers, scientists and decision makers to perform routine queries and analyses
themselves instead of relying on dedicated GIS st aff. This has opened a vast new market for GIS
vendors, resulting in huge increases in software sales.
Increasing numbers of users, resulting from better access to GIS, is certainly good for the
industry, but off-the-shelf desktop GIS cannot ensu re that unskilled users choose the appropriate
procedures (?tools?) for a specific task or that they use appropriate procedures correctly. Such
failings can lead to meaningless results or errors which might remain undetected by the casual
user. One way to ensure that GIS users, skille d or unskilled, use the technology appropriately is
to specialize a GIS, i.e. set it up so that it is us ed for a specific task only. The term ?specialized
GIS? is somewhat ambiguous as GIS are by nature not specialized (i.e. can be used for many
purposes), but it signifies that some operations are automated in such a way that there is little
room for error. When GIS are customized to perform a combination of automated operations,
they are often called spatial decisions support systems (SDSS).
1.2 SPATIAL DECISION SUPPORT SYSTEMS AND GIS
SDSS are decision support systems (DSS) with a spatial component. DSS were developed in the
1960s to aid decision making when problems are ne ither well structured nor unambiguous. These
problems are referred to in the literature as ill-structured or semi-structured (Ascough et al. 2001;
Densham 1991; Goodchild & Densham 1990), i.e. they cannot be solved with an algorithm or a
predefined sequence of operations. To solve such problems, predictive analysis is often required
to present decision makers with different scenarios to explore the possible effects of their
decisions. This type of inter active exploration enables a decision maker to develop a better
understanding of a given problem. DSS are therefore not meant to provide solutions, but rather to
support decisions.
DSS distinguish themselves from other information systems in that they require a database
management system, analytical modelling capabilities, analysis procedures, and a user interface
with display and report generators. DSS often use expert systems to interpret and analyse data.
Expert systems are software programs that mathematically analyse subject-specific knowledge in
the form of rules to solve problems.
In addition to the capabilities of DSS, SDSS should include a geographical database,
mechanisms for spatial data input, representation of spatial relations and structures, spatial
analysis functions and various output forms including maps (Ascough et al. 2001). SDSS can
therefore be created by extending existing DSS to handle spatial data.
5
When the sub-systems of a GIS (i.e. input, stor age, management and analysis, and output) are
compared to those of a SDSS, many similarities are evident, although a GIS cannot be regarded
as a SDSS on its own as it lacks the interactivity and automation required for scenario
generation. However, thanks to GIS?s capability to handle spatial data, GIS and DSS are often
combined to create SDSS (Agrell, Stam & Fi scher 2004). Such GIS/DSS implementations save
development time as they make use of existing software.
Another approach to developing SDSS is to modify and customize GIS to incorporate the
functionality of DSS (Basson 2005; Bester 2004; Mlisa 2007; Van Niekerk 1997; Varma,
Ferguson & Wild 2000). Most modern GIS allow developers to extend their functionality
through programming. This approach has a cost be nefit as no DSS licensing costs are applicable,
although, as with the GIS/DSS approach, the user must still own a copy of the GIS software in
order to use the system. In many cases users are managers or decision makers for whom the GIS
software must be purchased especially to run the SDSS. This can be very costly and, because
these SDSS (and users) rarely use the full GI S functionality, it is highly inefficient.
Some GIS developers offer comp onent-based software development tools that allow spatial
functionality to be incorporated in other non-GIS software. With these tool kits, GIS functions
(components) can be seamlessly embedded in existing applications such as spreadsheets, digital
atlases, and routing systems. A programmer can also create an entirely new SDSS (or even GIS)
containing only the functionality that is needed for a specific application (Longley et al. 2002).
This lowers the implementation cost per user as the developer purchases a tool kit once and pays
a small licensing fee for each deployment (ESRI 2002c). However, the procedure requires
additional development time and resources and is only viable when the number of users is large.
Because SDSS are designed for specific applications, many of the operations can be automated.
This reduces the need for user input, which limits the chances of human error. The software can
also be developed to be highly user-friendly as the user interfaces can be simplified to include
only the functions that are necessary for the specific task. The user can also be directed through
the process and adequately informed about the available options, thereby reducing the possibility
of error.
The ease of use and increased integrity of SDSS make spatial analysis more accessible to users
with very little or even no GIS skills. Unfortunately, the higher level of sophistication of SDSS
comes at a price. Even if an organization is w illing to invest in the development of a SDSS, the
GIS licensing costs that are often involved impede its deployment.
6
1.3 THE INTERNET AND GIS COMMUNICATION
The Internet is a publicly accessible worldwide sy stem of interconnected computer networks. It
is based on the packet-switching Advanced Re search Projects Agency Network (ARPANET)
created by the US Department of Defence in the early 1970s (Longley et al. 2002; Wikipedia
2005). Since the development of the World Wi de Web (WWW) by Be rners-Lee (1989), the
uptake of WWW (or Web) technolog y has been remarkably quick. This hypertext- based service
has brought the Internet into the realm of everyday use. In 2002 it was estimated that the Web
consisted of two billion publicly-indexed web pages (netz-tipp.de 2002), while in a more recent
study this figure was estimated to be 11.5 billion (Antonio & Signorini 2005).
Since the early 1990s, the Internet has had a pr ofound effect on technology, science and society.
It has changed the way we conduct business, communicate, educate and govern (Longley et al.
2002). In the South African banking sector alone, more than one million online banking accounts
existed in 2003. In 2004 the to tal increased by 32% (Golds tuck 2004) and in 2006 by 49%
(Research Surveys 2006). Web-based automated information systems, such as online banking,
enable companies to link their data-processing syst ems in an efficient and flexible manner. By
doing so, companies can work more closely with suppliers and partners and better satisfy the
needs and expectations of their customers.
The Internet is fast becoming a primary source of information and web users increasingly resort
to it to support their decision making (Jarupathirun & Zahedi 2007). DSS are being web-enabled
owing to the increased accessibility and familiar interfaces of websites (Salewicza & Nakayama
2004; Thysen & Detlefsen 2006; Wang & Chei n 2003; Wang 2005) and because Internet
technologies offer tighter integration with existing information systems (Vahidov & Kersten
2004).
Concerning spatial technologies, the Internet has been the greatest external stimulus for GIS
since the late 1990s. It has shifted the vision and basic role of GIS, i.e. to perform spatial tasks
more efficiently, to communicating geographical information between users. The Internet
provides an easy, cost-effective way to access spa tial databases distributed worldwide. Greatly
stimulated by market demand for geographical information, web applications that serve spatial
data have grown rapidly since the first Internet mapping site was introduced by Xerox PARC in
1993. In 2002, more users made use of GIS functi onality through the Internet than through all
the other types of GIS software combined (Longley et al. 2002).
7
1.4 WEB MAPPING FOR SPATIAL DECISION SUPPORT
Since its introduction in the middle to late 1990s, web mapping has been increasingly used to
distribute geographical data over the Internet. Web mapping soft ware such as Google Maps
(Google 2005), MapMachine (National Geographic Society 2005), AlertNet (Reuters Foundation
2005), MapQuest (MapQuest 2008) and StreetMa p (MWEB 2005), enables anyone with access
to a computer and the Internet to explore geographical data online and produce maps on demand.
With these tools, users can access geographical in formation without the need for expensive GIS
software (Van Wyngaarden & Waters 2007).
The introduction of products such as Google Ea rth has highlighted the value of web mapping for
spatial decision support. Many organizations have recognized the potential of such systems for
the cost-effective distribution of maps and other spatial information within organizational
structures to improve productivity and to make better decisions. The functionality of most web
mapping applications is however limited to data display and does not support GIS functionality
such as editing, spatial analysis and modelling (Pummakarnchana, Tripathi & Dutta 2005).
Web mapping solutions that offer more advanced functionality often require supplementary
software to be installed on the user?s computer . The installation of these so-called plug-ins and
interpreters (Jiang 2003) is a de terrent to many web users who are not familiar with downloading
and installing software. In addition, many users regard software downloads as a security risk.
Consequently, for optimal accessibility, web-based SDSS should preferably be compatible with
existing web browser software.
GIS functionality is difficult to implement using web technology due to the complexity of
managing the data used, created and updated during these operations (Green & Bossomaier
2001). However, the gap between GIS and web mapping applications is expected to close as the
demand for more functionality increases. The addition of spatial analysis and modelling
capabilities to web mapping applications holds much potential (Jiang 2003), especially for the
cost-effective development of web-base d spatial decision support systems.
1.5 RESEARCH PROBLEM FORMULATION
The research addresses two probl em fields: firstly, the technical development needs of spatial
decision support systems, and secondly the application of the technology towards a specific
development challenge in the Western Cape Provi nce. Both problem elements are encapsulated
here.
8
GIS are invaluable for supporting spatial deci sion making. GIS can be regarded as toolboxes
with a large selection of operations that can be used for a wide range of applications.
Unfortunately, the flexibility of GIS makes them difficult to use because operators not only need
to know which tools are appropriate for a specific task, but they also need to know how each
operation functions. The implication is that only those skilled in GI S have access to functionality
such as spatial analysis and modelling. To make this functionality more accessible to decision
makers, GIS are often customized into easily used SDSS. The development and deployment of
SDSS have traditionally been expensive due to high development and/or software licensing
costs.
Web mapping has established itself as a highly cost-effective means of communicating spatial
information. Internet technology provides a possible solution to the high cost of SDSS, as it
eliminates the need for expensive hardware or software. However, the problem exists that, due to
the complexities involved, existing web mapping technologies do not offer the spatial analysis
capabilities required by SDSS. This limitation current ly impedes the use of the Internet for SDSS
deployments. Consequently, one purpose of this research is to answer the following question:
Can web mapping be extended with currently av ailable technology to include the spatial
analytical functionalit y required by SDSS?
The second problem element is embedded in th e particular development challenges of the
Western Cape Province ? the area of application for this research. The province is experiencing
an alarmingly high population growth rate of 2.86% which is the second highest of the nine
South African provinces (Statistics South Africa 2001). Population growt h, together with an
urbanization level of 90% (Kok & Collinson 2006) , are causing increasing needs for housing and
food which place immense pressures on the province?s land resources.
The Western Cape contributes 23% towards the national agricultural contribution to GDP and
agriculture is one of the major i ndustries and the biggest user of land in the province. More than
11 million hectares (84%) of the province?s land surface is currently producing more than 55%
of South Africa?s tota l agricultural exports, of which the principal products are fruit (27%),
winter grain (21%), white meat (18%), wine (18%) and vegetables ( 16%) (CNDV Africa 2005).
Wine and fruit are the two principal export produc ts of the Western Cape. Due to an ever-
increasing demand for quality wine and fruit, the land area under wine grapes and orchard
cultivation is steadily expanding. Nationally, the total area under vineyards increased from
84 0030ha in 1994 to 101 958ha in 2007 ? an average annual expansion of nearly 2%. Almost all
(96%) of the wine-grape vine pl antings occurred in the Western Cape, placing increased pressure
on existing land resources (SAWIS 2008).
9
The Western Cape?s natural resources cannot be managed sustainably without performing sound
land use planning. Such planning requires accurate information about the suitability of land for
specific purposes. Because of the size and variet y of existing land uses in the Western Cape,
more accessible and easily used tools are needed to support decisions about land use in the
province. It is therefore a secondary c hallenge to this research to adapt and apply the developed
spatial decision support system to demonstrate its applic ability to implementa tion in the Western
Cape Province.
Land suitability analysis is one of the first and arguably the most useful applications of spatial
technology as it strongly relies on spatial analysis techniques (Malczewski 2006). In South
Africa, and specifically in the Western Cape Pr ovince, there is a general lack of awareness
among planners of the benefits and possibilities of SDSS for land use planning (Moss 2006, pers
com). Most of the few SDSS that can be applied for land suitability analysis are prohibitively
expensive, difficult to implement, user-unfriendly, or lack interactive spatial scenario-building
capabilities. In addition, most systems require considerable resources of hardware, software and
data (De la Rosa 2002; De la Rosa et al. 2004; Kalogirou 2002; Rossi ter 2001; Rossiter & Van
Wambeke 1997).
No land evaluation systems are presently in use in the Western Cape. The only comparable
systems actively in use are C-Plan and the Ca pe Environmental Management System (CEMS).
C-Plan is being employed by the Cape Action fo r People and the Environment (CAPE) to aid
conservation planning in the Cape floristic region (CFR) (New South Wales National Parks and
Wildlife Service 2001; SANBI s.d.), whereas CEMS was developed for CapeNature to evaluate
land for its conservation potential and to determine the effectiveness of existing nature reserves
(CapeNature 2007; Van Niekerk 199 7). Because both systems are onl y concerned with one land
use, namely conservation, neither of them is a true land evaluation system. There is clearly an
urgent need for a land evaluation system that can be used to support decisions about the Western
Cape?s stressed land resources.
1.6 RESEARCH AIM AND OBJECTIVES
The primary aim of this research is to evaluate the potential of the Internet to deliver low-cost,
user-friendly and interac tive spatial analysis functionality to a wide audience. To serve the
widest possible audience, only technologies that are compatible with existing web browser
software will be considered. A web-based SDSS will be developed to better understand the
capabilities and limitations of the currently available technologies. Because SDSS are problem-
specific, an application that adequately evaluates the spatial analysis capabilities of the SDSS is
required.
10
Hence, the secondary aim of this research is to build a web-based SDSS, called the Cape Land
Use Expert System (CLUES), which can be used to perform land suitability analyses for the
Western Cape Province.
To achieve the research aims, fi ve objectives have been set:
1. Review the literature to determine the system and data requirements of a web-based land
evaluation system and to overview the technologies now available.
2. Collect and prepare fundamental data sets for testing and demonstrating a web-based land
evaluation system.
3. Design, develop and implement CLUES.
4. Demonstrate how CLUES can be used to crea te land use scenarios for the Western Cape.
5. Critically evaluate CLUES, make recomme ndations for its improvement and point out
the limitations and potentials of Internet technology for SDSS development.
1.7 THE STUDY REGION
While the spatial application is of secondary importance only, some introductory background to
the Western Cape as province is in order. The Western Cape is South Africa?s fourth largest
province, covering 11% of th e country?s land area (see Figure 1-1). In 2007 the province
accommodated approximately 4.8 million people, 10.1% of the national total (Statistics South
Africa 2007). At 129 462 km 2 it is about the same size as England or Bangladesh.
The Western Cape is well known fo r its natural beauty and its environmental and biological
diversity. It comprises most of the CFR, the only floral kingdom located entirely within the
geographical confines of one country. The CFR is recognized globally as a biodiversity hotspot
which covers only 0.05% of the earth?s land surface, but as for biodiversity it contains three per
cent of the world?s plan t species (SANBI s.d.).
Thanks to its Mediterranean climate and relatively fertile soils, agriculture is one of the main
economic activities of the Western Cape. The province generates more than 20% of South
Africa?s gross farming income while employing one quarter of all the country?s farm workers
(Statistics South Africa 2006b).
Economically, the Western Cape has been boo ming, with an average real annual economic
growth rate of 4.2% between 1996 and 2005. This is the highest of all the provinces and is
considerably better than the national rate (3.7%) over th e same period. In 2005 the Western
Cape?s growth rate increased to 5.7% (Statistics Sout h Africa 2006a).
11
Figure 1-1 The Western Cape province, South Africa
1.8 RESEARCH METHODOLOGY AND AGENDA
This research is investigative and experimental in nature, with the intended outcome being a new
online information system by which available Internet technology can be qualitatively evaluated
for its applicability to land suitability analysis specifically and SDSS in general.
The research design is shown in Figure 1-2. The planning phase i nvolves the identification and
formulation of the problem, followed by setting the aims and objectives. These aspects have
been dealt with in this chapter.
The second research activity is to conduct a literat ure review to establish what land suitability
analysis entails. An outline of the land evaluation process and related concepts is provided in
Chapter 2. The literature review also covers th e various approaches to land suitability analysis,
namely Boolean overlay, multi-criteria decision ma king and expert systems. Chapter 2 concludes
with an overview of existing SDSS for land evaluation.
12
Figure 1-2 Research design for developing CLUES, a web-based SDSS for the Western Cape
Owing to the strong focus on technology in this re search, Chapter 3 is a review of technologies
now available for online spatial information system development. The review provides a basis
for designing the web-based land evaluation sy stem reported in Chapter 4. As with most
software developments, this design phase was preceded by a requirement analysis to determine
what functionality and other characteristics the system should exhibit.
REVIEW LITERATURE
(CHAPTER 2)
? Land evaluation
? Approaches to land suitability analysis
? Existing land evaluation systems
REVIEW CURRENT TECHNOLOGY
(CHAPTER 3)
? Web applications
? Web components
? Web maps
? Web map services
COLLECT AND MANIPULATE DATA
(CHAPTER 5)
? Collect and manipulate available data
related to terrain, soil, climate,
infrastructure and current land uses
CONDUCT REQUIREMENT ANALYSIS
(CHAPTER 4)
? Functionality
? Accessibility
? Speed
? User-friendliness
? Data
CREATE KNOWLEDGE BASE
(CHAPTER 7)
? Logical data modelling
? Database implementation
PLAN RESEARCH
(CHAPTER 1)
? Research problem: Extend web mapping
? Aims and objectives: Develop a web-
based SDSS
? Research methodology: Investigative &
experimental
DEVELOP LAND UNIT DATABASE
(CHAPTER 6)
? Land unit mapping
? Land property extraction
DESIGN SYSTEM
(CHAPTER 4)
? Land unit database
? Knowledge base
? Inference engine
? Web map service (WMS)
? Graphical user interface (GUI)
DEVELOP WEBSITE
(CHAPTER 8)
? Inference engine
? WMS & web server
? GUI
DEMONSTRATE CLUES
(CHAPTER 9)
? Use CLUES to perform suitability
analyses for perennial crops in a number
of areas in the Western Cape
? Produce land use suitability maps at
varying levels of scale
EVALUATION OF CLUES
(CHAPTER 10)
? Revisit system and data requirements
? Revisit research aims and objectives
? Record potentials and limitations of web-
based SDSS
? Suggest avenues of future research
KEY: LINK COMPARISON
13
Owing to the intrinsically spatial nature of land suitability analysis, the types of geographical
data to be analysed had to be considered in the system implementation. Chapter 5 describes the
availability and collection of fundamental data sets on terrain, climate and soils. Activities
related to the collection of the data and the manipulations needed to redress the lack of data are
reported.
Data collection is followed by the implementation of the system, which comprises the creation of
the three main components of the land evaluation expert system. The development of the first
component, namely the land unit database, involves the mapping of the land units (i.e. basic
mapping unit) and the extraction of land property data (i.e. environmental and physical
information about land units). These two activities are discussed in Chapter 6.
The second component of CLUES is the knowledge base. As the name suggests, the knowledge
base is a database that stores expert knowledge in the form of land use requirement rules. The
design and implementation of the knowledge base, using the logical data modelling procedure,
are described in Chapter 7.
The knowledge base not only contains rules, but also manages the operational data needed for
the functioning of the CLUES website, as set out in Chapter 8. The website component consists
of three parts, namely the inference engine, web map service (WMS ) and graphical user interface
(GUI). The function of the inference engine is to relate the information in the land unit database
to the rules in the knowledge base to evaluate each land unit?s suitability fo r a particular use. The
suitability ratings are used to produce suitability maps, which are created and distributed through
the WMS. The most prominent element of the CLUES website is the GUI , which directs user
interaction with the system.
To demonstrate the functionality of CLUES, a number of land use scenarios are created and
discussed in Chapter 9. This involves performing a suitability analysis for perennial crops and
producing suitability maps at varying scales for two agricultural regions in the Western Cape.
Each step is described in detail and illustrated with screen captures of the user interface.
Finally, in Chapter 10 CLUES is evaluated by comparing it with each of the requirements
stipulated in Chapter 4. In addi tion, the research is critically assessed regarding the achievement
of objectives. The potentials and limitations of web technology for SDSS deployment are
discussed. The report concludes with so me suggestions for further research.
In Chapter 2 which follows, the relevant literature is reviewed.
14
CHAPTER 2: THE PRINCIPLES AND PRACTICE OF LAND
SUITABILITY ANALYSIS
Physical land suitability analysis is a prerequisite for land use planning and development as it
supports decisions regarding land use and leads to optimal use of land resources (Van Ranst et al.
1996). Owing to the vast quantitie s of data required in land suitability analysis, computer
technology is often employed to manage, store, and analyse such data (Davidson 1986; De la
Rosa et al. 2004; FAO 1984). Because most data used in suitability analysis is spatial in nature,
geographical information systems (GIS) have beco me invaluable in land evaluation (Dai, Lee &
Zhang 2001). The GIS procedures in land suitability analysis can, however, be extremely time-
consuming and laborious processes without the aid of some level of automation. Fortunately,
most modern GIS allow users to program the soft ware to repeat a series of operations. This
capability of GIS not only speeds up the process of suitability analysis, but also facilitates the
creation of land use scenarios.
GIS that are customized for scenario building are often called spatial decision support systems
(SDSS). SDSS are excellent platforms for land suitability analysis as different land suitability
scenarios can be generated by making slight changes to the land use requirements or land
properties. The true potential of SDSS, supported by GIS, lies in th e ability to incorporate large
spatial data sets so that local land use decisions can be made while considering the effects on a
regional or provincial scale. The user-friendliness of SDSS also allows planners and decision
makers, who are often incapable to use GIS, to perform land suitability assessments themselves.
The primary aim of this research is to deve lop a web-based Cape Land Use Expert System
(CLUES) to demonstrate how the Internet can be used as a platform for SDSS and how it can
make GIS functionality, in partic ular spatial analysis, more accessible and cost-effective. In this
chapter, an outline of the land evaluation procedure is provided to illustrate what functionality is
needed by such a system. This is followed by an overview of the three main approaches to
developing land evaluation systems, namely Boolean overlay, multi-criteria decision making,
and expert systems, as elements of each of these techniques are used in the system design.
2.1 LAND EVALUATION APPROACHES
Land evaluation is the interpretation of land propert ies such as climate, soils, fauna and flora in
terms of the requirements of alternative land uses (FAO 1976). Land evaluation can therefore be
defined as the process of estimating the potential of land for alternative uses (Dent & Young
1981). Land evaluation, based on the guidelin es set out by the Food and Agriculture
Organization of the United Nations (FAO 1976; 1984; 1985), is an integral part of land use
15
planning and has been established as one of the preferred methods to support sustainable land
use management. Land evaluation gives an holis tic, multi-disciplinar y approach to sound
development and conservation by combining economic and social principles with environmental,
agricultural and biological sciences (Fourie 2006).
Land evaluation is based on the principle that certain land properties (i.e. soils, climate,
topography and other environmental and social variables) influence the success of a particular
land use. In essence, the objective is to co mpare and match each potential land use with the
properties of each type of land (F AO 1984). The land evaluation process ( Figure 2-1) involves
eight interrelated steps: set objectives to be reach ed; collect appropriate sp atial data; identify land
uses to be considered; specify land use requirements; map land units; determine land properties;
analyse the match between requirements and properties; and present results. These steps are
described in the following sections.
Figure 2-1 The steps in a land evaluation process
2.1.1 Set objectives
The land evaluation process starts with a pla nning exercise in which the objectives of the
evaluation are set. The most impor tant decisions made at this stage are about the boundaries of
the study area in which the evaluation will be carried out, and the level of detail required. The
level of detail depends on the type of evaluation that will be done. Investigative and
reconnaissance investigations are conducted at the largest map scale available to cover large (e.g.
national or provincial) areas. The scal es of such surveys vary from 1:2 00000000 to 1:120 000.
(2) COLLECT DATA
(3) IDENTIFY LAND USES
(4) SPECIFY LAND USE
REQUIREMENTS
(7) ANALYSE
(6) DETERMINE LAND
PROPERTIES
(5) MAP LAND UNITS
(8) PRESENT RESULTS
ITERATION
(1) SET OBJECTIVES
ITERATION ITERATION
Adapted from Dent & Youn g (1981 )
16
Semi-detailed investigations are perf ormed at scales ranging from 1:100 0000 to 1:30 0000,
usually for smaller regions such as districts, municipalities or catchment areas. For applications
at local or farm level, intensive and detailed investigations are needed. Such evaluations are
usually carried out at scales larger than 1:10 000 (FAO 1976; Lambrechts & Ellis s.d.).
Careful consideration is needed about the type of investigation envisaged because it determines
how elaborate the evaluation will be and also dictates what data is required.
2.1.2 Collect data
The data requirements are determined during the second step of the evaluation process. This
involves inventorying the available data to determine, if necessary, what additional data will
have to be captured or purchased. If it is too costly to collect additional data, the evaluation
objectives will have to be revisited. As most of th e data used in land evaluation is spatial in
nature, a GIS is often used to capture, store and prepare the data. Analysts must ensure that the
data sets are of an appropriate scale and that they conform geographically (i.e. are registered and
projected to the same coordinate system). Ex amples of phenomena captured in data might
include topography, geology, soils, hydrology, vegetation, land use and climate parameters.
2.1.3 Identify land uses
Once the data has been collected, land uses that ar e worth considering for their suitability in the
specified study area need to be identified. Land us es can be described in terms of ?major land
uses? or ?land utilization types?. When an eval uation is done for a large area it is probably
sufficient to specify broad or major land use t ypes, such as rain-fed agriculture, irrigated
agriculture, urban, and conservation areas. For mo re detailed studies, more specific and
demanding subdivisions or land utilization types such as grains , deciduous fruit, residential, and
wilderness area are more appropriate. Information about the production of goods (timber, crops
or livestock) or the offering of services (recreational, sewage or refuse) is often used to describe
land utilization types. Additiona l attributes to consider include market orientation, capital
intensity, labour intensity, and type of technology employed. Commercial sugarcane production,
on large privately owned properties, with low labour intensity, high capital inputs and level of
mechanization, would be an example of a land utilization type (FAO 1985).
2.1.4 Specify land use requirements
In land evaluation, land is described according to land characteristics and land qualities. Land
characteristics are properties of land that can be measured. Examples are slope gradient, slope
aspect, soil depth and land cover. Several land characteristics can be combined to form a more
17
complex land quality, such as fertility, available moisture supply or erodibility. Land qualities
are often qualitative in nature, whereas land characteristics are usually quantitative (FAO 1976).
Because either land characteristics or land qualities can be used to describe and classify land, the
term land property is used in this research to encompass both terms.
The set of land properties needed to sustain a particular land use is called a land requirement. To
determine land suitability, land properties are compared with land requirements (Burrough,
MacMillan & Van Deursen 1992), for instance th e production of a certain crop needing deep,
well-drained soils on gentle slopes, with an average annual rainfall of 300-500mm. In step five
of the land evaluation process these requirements are specified for each of the land uses
identified in step three (cf. Figure 2-1). Not only must the land use requirements be identified,
but the values that will be considered suitable (S) or not suitable (N) should be specified. The
principle classification of S and N is mainly based on technical, environmental or economic
factors. In most cases a further classification is required to differentiate between highly suitable
(S1), moderately suitable (S2), marginally suitable (S3), unsuitable at present (N1), and
permanently unsuitable (N2).
2.1.5 Map land units
In step five of the land evaluation process, land units are demarcated. The term ?land? is often
associated with any portion of the earth?s surface not covered by oceans or water bodies, but the
concept of land in the land evaluation context is much wider. Besides terrain and soil, land
includes the total physical environment (e.g. climate, hydrology, vegetation) and the results of
present and past human activity (e.g. salinization, vegetation clearance). Fo r the purposes of land
evaluation, social and economic characteristics are not included in the concept of land (FAO
1976).
Land units, or land-mapping units, are areas with pr operties that differ sufficiently from those of
other land units to affect their suitability for different land uses. Although any parcel of land can
be considered a land unit, it is more efficient and meaningful to use parcels that can be
adequately described in terms of one or a combination of land properties. A land unit should
therefore represent an area that is, in terms of predetermined properties, different from the
surrounding land and can be assumed to have homogeneous land properties (FAO 1984). The
degree of homogeneity or internal variation will vary depending on the scale and intensity set out
in the evaluation objectives (FAO 1976). When a reconnaissance evaluation is carried out over a
large region at a small map scale (i.e. 1:500 000 or smaller), large generalized land units such as
climatic zones would be sufficient. For more deta iled studies carried out at large map scales (i.e.
1:25 000 or larger), smaller map units such as so il types would be more appropriate. Landforms
18
are often used as land units in medium-scale studies (1:25 000 to 1:500 000) because many
physical land properties, including soil, climate and biology, are related to terrain (MacMillan,
Jones & McNabb 2004; Speight 1977). Examples of landforms and terrain units include crests,
cliffs, terraces, footslopes, pediments, pediplains and alluvial plains (McDonald et al. 1984).
Although the size of the la nd units should be kept as small as possible to limit generalization, too
many units can become unmanageable as each individual land unit is considered individually
regarding its land properties and requirements. Fo rtunately, capacities to handle large numbers of
land units have increased considerably with the use of computer technology and often the
decision about the size, number and delineation of land units is determined by data availability.
While soil type boundaries would probably be th e most suitable delineation of land units for
agricultural land uses, soil information is often not available at the required scales. In such cases
other available data sets, such as landforms and terrain units, can be used instead.
2.1.6 Determine land properties spatially
The properties (defined in Section 2.1.4) of each land unit are determined spatially during step
six. Using GIS, this essentially involves the co nversion of vector land property data to raster
format. Next, the land units are sequentially overlaid onto each raster to calculate the average
land property values (e.g. annual rainfall, effective soil depth, slope gradient) for each land unit.
The values are then added to th e land unit layer as attributes.
2.1.7 Analyse land units for suitability
During analysis, each land unit?s land properties are considered individually and compared with
the land use requirements to classify a unit into its appropriate suitability level (i.e. S1, S2, S3,
N1 or N2). Land suitability measurement can be as simple as determining whether a land unit
meets all the land use requirements, or it might involve complex mathematical calculations to
produce a suitability index used to find the optimal land use for a specific area. The chosen
methodology depends on the required outcome and the classification method used when the land
use requirements were set.
2.1.8 Present results
To conclude the evaluation process, the results are presented in the form of ?suitability? and
?solution? maps. Suitability maps are usually chor opleth maps depicting the level of suitability of
each land unit for a single land use using colour shading. Solution maps are simple qualitative
thematic maps showing only the land uses that are most suitable for any given land unit.
19
The land evaluation approach is versatile as it ca n be applied in rural or urban land use planning
(Bosshard 2000; Dai, Lee & Zha ng 2001; Fourie 2006; L?tz & Bast ian 2002) and it can be done
on national (Mantel et al. 2000), regional (Ceballos-Silva & L?pez-Blan co 2003b; Igu?, Gaiser
& Stahr 2004), watershed (Nisar Ahamed, Rao & Murthy 2000) or local (Cools, De Pauw &
Deckers 2002) levels. The procedur e has been shown to be highly suitable for forestry (Thwaites
& Slater 2000; Twery et al. 2005), agricultura l management (Dendgiz, Bayramin & Y?ksel
2003; Mantel, Zhang & Zhang 2003; Mongkolsawa t, Thirangoon & Kuptawutinan 1997; Smith,
McDonald & Thwaites 2000; Wandahwa & Van Ra nst 1996) and conservation planning (Phua &
Minowa 2005).
Land evaluation is part of a larger land use pla nning process and the results should be used to
support decisions about land use change. While la nd evaluation focuses on the suitability of land
units for different uses, land use planning examines the relationships between uses. Factors such
as social and economic needs of the community, as well as the environmental stability of an area,
should also be considered during the planning process. Environmental conservation is always an
objective of land evaluation and it is assumed that no form of land use will be judged suitable
unless it can be sustained on a long-term basis without significant detriment to the land (FAO
1984).
It is obvious that suitability analyses lie at the centre of land evaluation and although computer
processing is not a prerequisite for suitability analysis, it has become indispensable, especially
where large numbers of land units are considered. The next three sections focus on the main
approaches to land suitability analysis using computer technology, namely Boolean overlay,
multi-criteria decision making, and expert systems.
2.2 BOOLEAN OVERLAY
Overlay procedures play a centr al role in land suitability mapping. Many agree that overlaying
was introduced by McHarg (1969) when he superim posed individual transparent maps of natural
and man-made environmental phenomena to pr oduce overall land use suitability maps. This
manual procedure was soon incorporated into GIS and has become one of their most useful
operations.
The Boolean intersect and uni on operators are the two fundamental overlaying operations
available in GIS and can be perfor med on raster or vector data. When raster data is used, a raster
layer is required for each input feature (e.g. suitable soils) so that the position of features is
represented by cell values of 1 where the feature is present, while cells where the features are
absent have a value of 0. When the Intersect operation is carried out on two Boolean layers, an
20
AND comparison is made between the two layers. Th is means that if a cell has a value of 1 in
both input layers, it will be given a value of 1 in the output laye r. Cells that do not meet this
requirement will be assigned a value of 0. In the union operation, cells for which at least one
input value is equal to 1 are awar ded a value of 1. In logical arithme tic this is equivalent to the
OR operation. The output of the intersect and union operations is illustrated in Figure 2-2. In
suitability analysis, Boolean layers are created by assessing each criterion?s thresholds of
suitability (Ceballos-Silva & L ? pez-Blanco 2003a).
Figure 2-2 Intersect and union Boolean overlay operations
Land suitability often requires multiple input criter ia. When Boolean overlay is used to analyse
numerous input layers, two distinctly different results are obtained. Boolean intersection results
in a very ?hard? decision as a region will be excluded from the result if any single criterion fails
to meet its threshold. Conversely, the Boolean union operator implements a very liberal mode of
aggregation: a region will be chosen in the result as long as a single criterion meets its threshold.
By using the intersect operation the risk of producing an inaccurate classification is minimized as
no trade-off between criteria is allowed (i.e. hi gh suitability of one criterion cannot compensate
for low suitability in another), while risk is maximized when the union operation is used as one
criterion can override all other criteria (Ceballos-Silva & L ? pez-Blanco 2003a).
Boolean overlay is popular among GIS users as it is a standard feature in most proprietary (off-
the-shelf) GIS. Boolean layers (i .e. true/false binary layers) representing threshold values of land
properties can easily be analysed using standard intersect (logical AND) overlay operators
(Malczewski 2004). The operator used to combin e criteria using Boolean overlay should be
carefully considered as an inappropriate choice could result in inappropriate results. Another
problem with Boolean overlay is that all criteria are of equal importance (Ceballos-Silva &
L ? pez-Blanco 2003a). Boolean overlay is also a very discrete or ?h ard? decision strategy and this
increases the risk of error due to inappropriate thresholds or inaccurate data (Eastman 2000).
21
These limitations of Boolean overlay are ad dressed by multi-criteria decision making as
discussed in the next section.
2.3 MULTI-CRITERIA DECISION MAKING
Decision making regarding land suitability is often difficult as it involves numerous
stakeholders, multiple factors, and sometimes conflicting objectives (Traintaphyllou 2000).
Because the alternative land uses for any particular parcel of land are potentially unlimited, many
land use evaluation systems employ multi-criteria decision making (MCDM) techniques.
MCDM (also referred to as multi-criteria evalua tion (MCE) or multi-criteria analysis (MCA) in
the literature) essentially divides a problem into smaller understandable parts and evaluates each
part independently. The results of the individual evaluations are integrated to provide an overall
solution to the original problem (Malczewski 199 9). By using MCDM, solutions can be found to
decision making problems with multiple alternatives, evaluated by decision criteria (Jankowiski
& Nyerges 2001).
The analyst can choose from a range of MCDM methodologies for a particular application. The
available methodologies can be organized into three major dichotom ies (Bester 2004):
? multi-objective versus multi-attribute decision problems;
? individual versus group decision makers; and
? decisions taken under certainty (deterministic) versus uncertainty (probabilistic and
fuzzy).
Where multi-attribute decision making (MADM) pr oduces alternatives based on attributes,
multi-objective decision making (MODM) disti nguishes between alternatives based on the
objectives of the analysis. Because MODM require s alternatives it is usually executed as an
optional additional step after MADM is completed (Malczewski 1999).
MADM and MODM can be performed by an indi vidual or by a group. When a group of decision
makers is involved, competitive or independent conflict can occur. Competitive conflict arises
when preferences of different decision makers are in direct conflict, while independent conflict
occurs when actions of one decision have indirect consequences on another (Eastman 2006;
Malczewski 1999).
Decision making frequently involves a degree of uncertainty caused by inappropriate
information, unforeseen circumstances, or invalid methods. To co mpensate for this, probabilistic
or fuzzy approaches can be used instead of deterministic (Boolean) methods (Malczewski 1999).
22
MCDM can support decisions concerning spatial and non-spatial problems. One spatial
application for which it is routinely used is land suitability evaluation. The MCDM procedure is
discussed in the next section.
2.3.1 The MCDM procedure
Van der Merwe (1997) suggests a seven-step pr ocedure for applying MCDM to perform land
suitability analysis (Figure 2-3). A brief overview of each st ep is provided in the following sub-
sections.
Adapted from Van der Merwe (1997)
Figure 2-3 A step-wise procedure for multi-criteria decision making
2.3.1.1 Set objectives
The objectives of the MCDM exerci se must be set before the evaluation can be carried out as
they dictate which methodology or decision strategy will be used in the evaluation (i.e. multi-
attribute, multi-objective, individua l, participative, deterministic, and probabilistic). For instance,
an objective might be to find areas most suitable for perennial crops, in which case a multi-
attribute evaluation would suffice as there is only one alternative (or land use). Another objective
may be to find 100 hectares most suitable for pe rennial crops and 50 hectares most suitable for
grain production: an objective ca lling for two land uses to be weighed against each other. Such
an analysis requires multi-objective evaluation.
GIS APPLICATION FIELD
(Technical specialists)
POLICY- AND DECISION MAKING FIELD
(Officials, planners and community)
1. SET OBJECTIVES
2. SELECT APPROPRIATE CRITERIA
3. MAP SPATIAL CRITERIA 4. STANDARDIZE MEASUREMENTS
5. SET CRITERIA WEIGHTS
6. MULTI-CRITERIA EVALUATION
7. MULTI-OBJECTIVE EVALUATION
G
E
N
E
R
A
TE
A
LT
E
R
N
A
TI
V
E
S
C
E
N
A
R
IO
S
EVALUATE RESULTS
23
If the requirements for each alternative are well established, the decision rules can be set by an
individual expert. Group partic ipation is often employed for semi-structured problem solving
(Densham 1991). Due to the degree of uncerta inty about data quality and precise land use
requirement thresholds, probabilistic methods are the most frequently used, especially if multi-
objective evaluation is required.
Land suitability analysis objectives should e volve from a problem statement based on
discussions with stakeholders and/ or a literature study. It is important that the study area and the
scale at which the analysis will be carried out are clearly defined before the analysis is done. All
the land uses considered in the evaluation must be specified, and assumptions and limitations due
to data availability and time constraints should be acknowledged.
2.3.1.2 Select appropriate criteria
In step two of the MCDM process, the appropriate criteria for measuring land suitability are
defined. Criteria can be either factors or constraints. Factors re fer to criteria that enhance or
detract from a land use?s overall suitability, while constraints are meant to limit or exclude areas
for consideration (Malczewski 1999).
2.3.1.3 Map spatial criteria
Once the criteria are selected, the factors and co nstraints for each criterion are mapped, usually
using GIS. Due to the continuous nature of many criteria, a raster format is often chosen for
MCDM. Depending on availability of existing data, new data may have to be captured. If
existing data is used, reformatting and manipulation may be necessary as some GIS require raster
data sets to have the same extent and resolution. Analysts must also ensure that the appropriate
map projection and datum are used because criter ion layers will be overlaid. Map projections
that are true to area (i.e. equal- area projections) are recommended (DeMers 2005).
2.3.1.4 Standardize measurements
Once the spatial data is in the appropriate format, the level of suitability must be specified and
incorporated into the data. Th is procedure of preparing the data for analysis is called
measurement standardization. B ecause factors can be continuous and measured in different
scales, step four of the MCE process requires all factors to be reformatted to a common
measurement scale. To do so, linear scaling (Equation 2-1) is commonly used (Malczewski
1999).
m
RR
RR
X ii ??
?=
)( minmax
min Equation 2-1
24
where R i is the raw score;
R min represents the minimum score;
R max is the maximum score; and
m is an arbitrary multiplier representing the upper standardized range
value.
To demonstrate the use of Equation 2-1, it is known that perennial crops require soils of at least
300mm deep and that suitability increases as soil depth increases due to the better resilience of
perennial plants to withstand adverse climatic conditions. All soils with an effective depth of less
than 300mm can therefore be regarded as unsuita ble (Schloms 2008, pers com). In this example,
the parameters of Equation 2-1 are set to: R min = 300; R max = 1200; and m = 1. Soils of 300-
1200mm depth are rescaled to va lues ranging from 0 to 1, w ith 1 representing the highest
possible suitability and 0 the lowest. Figure 2-4 illustrates th is example graphically.
Figure 2-4 Linear scaling of effective soil depth for perennial crop suitability
Linear scaling can be regarded as a form of fuzzy classification as it incorporates a gradual
transition between thresholds. Fuzzy classifica tion is based on fuzzy set theory (Zadeh 1965)
which resembles human reasoning when approximate information is used to make decisions. It
was specifically developed to mathematically represent uncertainty and can be used to deal with
the imprecision intrinsic to many spatial problems (Argialas 1995).
A fuzzy set can be defined mathematically as follows: if X = [ x ] denotes a space of objects, then
the fuzzy set A in X is the set of ordered pairs
{ })(, xxA A?= Xx ? Equation 2-2
where )( xA? is known as the ?grade of membership of x in A ? and
Xx ? signifies that x is contained in X .
25
As with linear scaling, )( xA? is by definition a number ranging from 0 to 1, with 1 representing
full membership of the set and 0 non-memb ership. The level of membership of x in A does not
represent probability but possibility: )( xA? of x in A specifies the degree to which x belongs
to A (Burrough 1989; Sicat, Carranza & Nidumolu 2005).
A fuzzy membership function (FMF) is used to de termine the suitability value of a mapping unit.
A popular FMF for land suitability analysis is the S-membership function (Huajun et al. 1991;
Huajun & Van Ranst 1992; Sicat, Carranza & Ni dumolu 2005) expressed in Equations 2-3 and
2-4. Graphic representations of S-membership f unctions are shown in Figures 2-5a and 2.5b. In
Equation 2-3, when x is equal to or exceeds ?, a full membership (i.e. ),,;( ???? x = 1) is
achieved. Lower values of x will result in a partial membership (i.e. ),,;( ???xS < 1). Half-
membership (i.e. ),,;( ???? x = 0.5) is achieved when x equals ? and non-membership (i.e.
),,;( ???xS = 0) is reached when x is equal to or less than ?. Equation 2-4 is the descending
version of Equation 2-3 and can be similarly described.
???
???
?
???
??=
,1
,)]/()[(21
,)]/()[(2
,0
),,;(
2
2
???
???????
x
x
x
],[
],[
],[
],[
+??
?
?
???
?
??
??
?
x
x
x
x
Equation 2-3
where ? is the lower limit of attribute x ;
? is the upper limit of attribute x ; and
? is (? + ?)/2.
???
???
?
??
???=
,0
,)]/()[(2
,)]/()[(21
,1
),,;(
2
2
???
???????
x
x
x
],[
],[
],[
],[
+??
?
?
???
?
??
??
?
x
x
x
x
Equation 2-4
where ? is the upper limit of attribute x ;
? is the lower limit of attribute x ; and
? is (? + ?)/2.
In the example of perennial crops an effective soil depth of at least 300mm is required, while a
depth of 900mm or more is c onsidered equally suitable. Equation 2-3 can be used in this scenario
by setting ? = 300 and ? = 900 millimetres (see Figure 2-6a). In this case an asymmetrical
function
26
Figure 2-5 Ascending (a) and descending (b) S-membership functions
is more appropriate as perennial crops require soils having a pH between 5 and 7 (Fourie 2006).
In this situation, the ascending and descending asymmetrical S-membership functions can be
combined to form a symmetrical function (see Figure 2-6b) by setting ? = 5 and ? = 6 in
Equation 2-3 and ? = 6 and ? = 7 in Equation 2-4.
Figure 2-6 Effective soil depth (a) and soil acidity (b) membership functions for perennial crops
Huajun & Van Ranst (1992) have shown that fuzzy methods are more accurate than Boolean
classification. Another advantage of the fuzzy approach is its to lerance to inexact resource data.
This is especially important in land evaluation as suitability analysts are often forced to use low-
quality data or large mapping scales because no other data exists. To demonstrate so-called
?fuzzy tolerance?, suppose the effective soil de pth data in the perennial crops example is
inaccurate and overestimates depths by 150 millimetre s (i.e. a true depth of 300mm is indicated
as being 450mm). In this situation, a true de pth of 375mm is considered suitable using Boolean
classification (compare Figure 2-7). In Equation 2-3, however, )750,450,150;375(? = 0.03125
indicating that the suitability of a depth of 375cm is almost negligible in this scenario.
27
The main disadvantage of fuzzy rules is their co mplexity. Setting the appropriate parameters for
each requirement can be challenging, especially without specialized software to graph and
visualize the effects of different parameters. In addition, programming is often necessary to
implement fuzzy functions in GIS as more complex procedures are needed (Davidson,
Theocharopoulos & Bloksma 1994; Fourie 2006; Hall, Wang & Subaryono 1992; Jiang &
Eastman 2000; Malczewski 2006; Nisar Ah amed, Rao & Murthy 2000; Wang, Hall &
Subaryono 1990). Some software packages provide specialized tools usi ng linear or non-linear
functions to scale factors. The FUZZY module of IDRISI A ndes is a good example (Eastman
2006).
In the perennial crops example, a constraint must be set for values of less than 300mm ( Figure
2-7) to prevent other factors (s uch as soil acidity) from compensating for effective soil depth.
Constraints are given a suitability value of 0 and are represented by Boolean (or mask) layers
(Eastman et al. 1995). All soils with a depth of less than 300mm can ther efore be regarded as
unsuitable. Figure 2-7 graphically demons trates Boolean classification of this threshold.
Figure 2-7 Boolean constraint of effective soil depth for perennial crops
Boolean classification is easy to implement using computer technology as simple logical (AND,
OR) operations are required. Such operations are supported by structured query language (SQL),
which is standard in most database management systems. Most GIS packages offer this
functionality (Burrough, MacMillan & Van Deursen 1992).
2.3.1.5 Set criteria weights
By nature different criteria do not have equal importance for a particular objective or land use.
Effective soil depth might, for instance, be considered more important for wine grape production
than slope gradient. To take this into considerat ion, each criterion must be weighted according to
its relative importance. Weights can either be a ssigned by the analyst or in consultation with
stakeholders. Weight values of criteria range from 0 to 1 and shoul d be specified so that their
sum is 1. Deciding on which weights to allocate to each criterion becomes more difficult as the
28
number of criteria increases. Fortunately, a method called the an alytical hierarchy process (AHP)
supports this task (Saaty & Vargas 1991). AHP employs a pairwise comparison of criteria to
arrive at a scale of preferences. Complex unstructured problems are broken down into their
component parts, which are then arranged into hierarchical order. The relative importance of
each pair of criteria is subjectively judged and numerical values (see Table 2-1) are assigned
accordingly. These values are placed in a comparison matrix and evaluated.
Table 2-1 Scale of analytical hierarchy process (AHP) comparisons
Numerical values Description
1 Equal importance
3 Moderate importance
5 Strong or essential importance
7 Very strong or demonstrated importance
9 Extreme importance
2, 4, 6, 8 Intermediate values
Reciprocals Inverse comparison
For instance, if effective soil depth, soil acidity , slope gradient and slope aspect are the criteria
used to select the most suitable areas for perennial crop production, one might decide that
effective soil depth is slightly more important than soil acidity. A value of 2 is given to this
pairwise comparison and placed in the effective soil depth (row) and soil acidity (column)
position of a comparison matrix (see Table 2-2).
Table 2-2 AHP comparison matrix for perennial crops
Effective soil
depth
Soil acidity Slope gradient Slope aspect Priorities (%)
Effective soil depth 1 2 3 3 44
Soil acidity 1/2 1 3 3 31
Slope gradient 1/3 1/3 1 1/2 10
Slope aspect 1/3 1/3 2 1 15
If effective soil depth is slightly more important than soil acidity, it follows that the inverse is
true (that soil acidity is slightly more important than effective soil depth) and a value ? is placed
in the soil acidity (row) and effective soil depth (column) cell. Effective soil depth is also
moderately more important than both slope gradient and slope aspect as indicated by the value of
3 in the appropriate cells.
Source Saaty & Vargas (1991)
Adapted from Saaty & Vargas (1991)
29
Once finalized, the comparison matrix can be used to determine the priorities of each criterion by
using the AHP program (AHPP) developed by the Canadian Conservation Institute (2005).
AHPP employs the principle Eigen value method as described in Saaty (1998) and was used to
calculate the priorities for the example of perennial crops in Table 2-2. The resulting priorities
are 44%, 31%, 10%, 15% for effective soil depth, soil acidity, slope gradient and slope aspect
respectively.
Because comparison matrixes are created by human reasoning, they can contain inconsistencies.
For instance, criterion A may be regarded as more important than criterion B, while B might be
considered more important than criterion C. An inconsistency will occur if criterion C has been
defined as being more important than criterion A (Marinoni 2004). To guard against such
inconsistencies, Saaty (1977) in troduced a consistency ratio (CR) which can be calculated from
the principle eigenvector of the comparison matrix. A comparison matrix is considered
inconsistent when its CR value is 0.1 or more. The CR for Table 2-2 is 0.045 according to
AHPP, indicating that there are no significant logical inconsistencies present in the matrix.
AHP has been successfully applied in many MCDM applications and shown to be especially
useful when public participation is incorporated in the weighting process (Mau-Crimmins, De
Steiguer & Dennis 2005). However, the setting of AHP scales can be confusing. According to
Van der Merwe (2008, pers com) the pre-ranking of factors in order of importance can simplify
the process considerably. Whichever method is used, many agree that the setting of weights
without applying some kind of method to ensure consistency, could have adverse effects on
suitability analyses (Malczewski 2004).
2.3.1.6 Multi-criteria evaluation (MCE)
In the sixth step of the MCDM procedure, the criteria are analysed to produce suitability maps.
In MCE, factors, constraints and weights are combined using weighted linear combination
(WLC). This essen tially involves calculating a suitability value for a particular land use using
Equation 2-5.
?? ?= jii cxwS Equation 2-5
where S is the suitability value;
w i is the weight of factor i ;
x i is the criterion score of factor i ;
cj is the Boolean criterion score of constraint j ; and
30
? is the product of criteria.
In contrast to the high-risk Boolean inters ect (AND) and union (OR) operations, WLC produces
a risk-averse (Eastman 2000) and full trade-off solution (Mahini & Gholamalifard 2006). If more
control over the level of trade-off is required, ordered weighted averaging (OWA) can be applied
as it employs an additional set of weights, called order weights, that are assigned on a location
basis to manipulate trade-off (Malczewski 2006).
The result of MCE is a set of maps showing the level of suitability for each land use analysed.
Suitability values are provided as a ratio scale from 0 to m (see Equation 2-1). Graduated shades
are often used to help visualize increasing suitability. Figure 2-8 demonstrates the result of a
simple MCE involving two factors (A and B) of equal weight.
Figure 2-8 Graduated shades used to visualize suitability levels of factors and results
2.3.1.7 Multi-objective evaluation (MOE)
In multi-objective evaluation (MOE), alternative objectives (i.e. land uses) are compared to find
the best solution according to the objectives set dur ing the first step of the MCDM process. Land
uses can either be conflicting or complementary (or non-conflicting). Conf licting land uses occur
when a land parcel is suitable for two or more land uses, but can be used for only one purpose. If
a parcel of land is suitable for more than one land use and can accommodate multiple uses (e.g.
recreation and forestry), it is considered to be complementary (Eastman 2006).
2.3.2 Discussion
The use of MCDM for land suitabil ity analysis is well established. Three recent applications are
those by Van der Merwe & Steyl (2005), Agrell, Stam & Fischer (2004) and Ceballos-Silva &
L?pez-Blanco (2003b). However, MCDM is not limited to solving land suitability problems.
Other applications are in economics (Al-Najj ar & Alsyouf 2003), noise pollution (Van der
Merwe & Von Holdt 2006), forestry (Bruno et al. 2006; Varma, Ferguson & Wild 2000),
conservation (Phua & Minowa 2005; Wood & Dragic evic 2007), flood vulne rability (Yalcin &
Akyurek 2004), transportation (Vreeker, Nijk amp & Ter Welle 2002) and tourism potential
31
determining (Van der Merwe, Ferreira & Va n Niekerk 2008). Because various methods are
available, no two implementations of MCDM are identical. This versatility and flexibility makes
MCDM a remarkably popular spatial decision support methodology.
Most of the MCDM-based SDSS reported in the li terature were not specifically developed for
land evaluation applications. The exception is Agro-Ecological Zone for Windows (AEZWIN),
developed by Fischer, Granat & Makowski (199 8). Although AEZWIN boast s an interactive and
intuitive user interface, it lacks a spatial component. To use th e system, the data must be
prepared in a GIS, imported to AEZWIN for analysis, and then exported to a GIS for visual
interpretation. Such loosely coupled approaches are not suitable for SDSS as they cannot provide
an environment that enables interactive scenario building.
A recent approach is to integrate MCDM with GIS to improve interactivity. An example is
MCDM_AV, which is an extension for ArcView GIS that allows users to carry out multi-criteria
evaluation on both raster and vector data in an existing GIS environment (Bester 2004). The
advantage of incorporating MCDM into an existing GIS is that less programming is needed
because much of the functionality needed by MCDM is inherent in GIS. The disadvantage of this
approach is that the user needs to own a copy of the GIS software which can be prohibitively
expensive. The user must also be proficient in GI S in order to operate them in such applications.
GIS are often used to conduct MCDM owing to th e formers? ability to spatially integrate and
compare multiple geographically referenced data sets. Figure 2-3 highlights the importance of
GIS in the MCDM procedure, with GIS being inst rumental in four of the seven steps (i.e. map
spatial criteria, set criteria weights, multi-cr iteria evaluation and multi-objective evaluation). Due
to their complexity, GIS analyses should preferab ly be carried out by GIS experts. Practitioners
involved in MCDM can be separated into two groups: those doing the GI S analyses (GIS
analysts) and those making the policies and decisions (officials, planners, and community
members). It is perceivable that a GIS analyst might not always have sufficient insight into the
problem at hand. Good communication between th e specialist and the decision makers is
therefore essential for the MCDM process to be successful as misunderstandings can lead to
incorrect and possibly conflicting results (Van der Merwe 1997).
Although MCDM is highly effective when participation by many stakeholders is required (Mau-
Crimmins, De Steiguer & Dennis 2005) , it can also be useful in projects where the policy- or
decision maker is an individual such as an official, planner or scientist. Decisions regarding the
weights of the criteria can be based on studies reported in the literature, rather than on multiple
stakeholder participation. If the decision maker is sufficiently competent in GIS, the entire
procedure can be carried out by one person (Van der Merwe 2006). An advantage of an
32
individual (or small group) approach is that the MCDM procedure becomes more explorative if
the analyst can experiment with the weightings without having to consult many stakeholders.
This means that different weighting schemes can easily be applied to interactively build
scenarios. Interactive scenario building is essential for decision support, especially where
objectives are vague and problems semi-structured (Clarke 1990). Special so ftware is required to
automate the MCDM process to enable interactive scenario building. Unfortunately, the MCDM
process is not easily automated because it constantly requires input from the analyst or decision
makers. A more effective method is to allow the operator to set criteria standardization rules and
weights once during each suitability analysis so that all the other steps can be automated. The
setting of rules not only reduces user input, but also facilitates the storage of decisions for future
reference and use. Such a rule-based appr oach is explored in the next section.
MCDM provides a highly flexible methodology for land suitability analysis. The wide range of
methods available in MCDM can be confusing to some decision makers, especially if they are
unfamiliar with the fundamental concepts. Using an inappropriate method can considerably
influence the outcome of the evaluation (Joerin, Theriault & Musy 2001). Care must be
exercised when suitability values resulting from MCE are interpreted (Proctor & Qureshi 2005).
For practical implementations, suit ability values are often converted from a ratio scale to an
ordinal scale to produce more user-friendly suitability levels (i.e. low, medium, high). To do so,
the analyst and/or stakeholders set minimum and maximum limits for each suitability class. Due
to the way in which MCDM standardizes and combines criteria, the thresholds are often set
without any consideration of the underlying factors and may therefore be inconsistent with the
original criteria settings.
2.4 EXPERT SYSTEMS
Expert systems are computer systems that emulate human decision making (often called artificial
intelligence) by considering a set of predefined rules. The main co mponents of an expert system
are a knowledge base which stores information in the form of rules, and an inference engine that
contains protocols about how the rules in the knowledge base are applied ( Encyclopaedia
Britannica 2007).
The rules in the knowledge base are usually obtained by interviewing experts in a particular
subject. The interviewer, or knowledge engineer , organizes the information obtained from the
experts into a collection of rules, typically in an ?if-then? structure. The inference engine is then
used to make deductions from the rules to solve complex problems (Encyclopaedia Britannica
2007).
33
The ability of expert systems to support problems involving many different factors, makes it
highly suitable for SDSS. Together with the capabil ities of GIS to store, manipulate, analyse and
present spatial data, expert systems are powerful tools for supporting complex spatial decisions
such as the optimal use of land. Because land use requirements can be formulated easily as ?if-
then? rules, the rule-based approach of expe rt systems is eminently appropriate for land
suitability analysis. For instance, if a land unit ha s an effective soil depth of more than 300mm, it
can be regarded as suitable for the production of perennial crops. By applying this rule to the
effective soil depths of all land parcels (i.e. land units) in a GIS database, the system can select
all the land units that meet this requirement. Similar rules can be created for other land properties
and combined using a predetermined methodology (i.e. inference engine) to calculate a
suitability index.
While the requirements for some land uses are well researched, many land use requirements are
unknown. In such cases, land management decisions are often made by experts like agricultural
consultants, farmers or engineers. By capturing and safeguarding expert knowledge and
experience in a database of rules, computers can be used to carry out land use evaluation over
very large areas.
The expert system procedure for doing a suitability analysis is shown in Figure 2-9.
Figure 2-9 Procedure for an expert system land suitability analysis
The process starts with the establishment of la nd-requirement rules, wh ich are then evaluated
against the land unit properties. During evaluation, the inference engine assigns a suitability
rating to each land unit, which is used to produce a suitability map. Different scenarios can be
generated by repeating the process with altered rule sets. The subsequent sections consider the
rulebase, inference engine and land unit data base and overview some existing systems.
2.4.1 The rulebase
Land use requirement rules are central to the expe rt system analysis procedure. Both Boolean
and fuzzy rules are usually accommodated to provide as much flexibility in rule creation as
possible ( Encyclopaedia Britannica 2007). For Boolean rules, simp le thresholds are used to
define land property suitability. Soils can, for instance, be reclassified into binary values (0 and
1), where a value of 1 represents su itable, and 0 unsuitable soils (see Figure 2-7). To implement
34
levels of suitability (i.e. S1, S2, S3, N1, N2), a se t of thresholds can be specified as illustrated in
Figure 2-10.
Figure 2-10 Levels of suitability of effective soil depths for perennial crops using Boolean classification
The individual threshold values needed to implement the land requirement shown in Figure 2-10
are provided in Table 2-3.
Table 2-3 Example of five Boolean rules specifying effective soil depth requirements for perennial crops
# SUITABILITY LEVEL LOWER (mm) UPPER (mm)
1 Permanently unsuitable (N2) 0 150
2 Unsuitable at present (N1) 150 300
3 Marginally suitable (S3) 300 525
4 Moderately suitable (S2) 525 750
5 Highly suitable (S1) 750 1350
For rule 1 the lower and upper thresholds are specified as 0mm and 150mm respectively and the
lower threshold for rule 2 is set equal to the upper threshold of rule 1. A similar procedure is
followed for the rest of the suitability levels. Because all soils of 750mm or more are considered
in this example to be highly suitable, the upper value of this suitability class should actually be
infinity. To represent such ?open-ended? rule s, the upper threshold is set to the maximum
effective soil depth value in the database, which is 1350mm in this case.
In reality, suitability increases as effective soil depth increases. In Figure 2-10, 300mm-deep
soils are regarded just as suitable as 450mm-deep soils. A fuzzy set function (see Section 2.3.1.4)
therefore better represents the effective soil depth requirement. To implement different levels of
suitability, multiple fuzzy set functions are requir ed. For instance, a total of seven S-membership
functions are used in Figure 2-11 and parameters ? (lower threshold) and ? (upper threshold) are
set so that levels overlap. In this example effective soil depths of 300mm are regarded as either
marginally suitable (S3) or not suitable at present (N1). As depths increase from 300mm
35
Figure 2-11 Levels of suitability of effective so il depths for perennial crops using fuzzy classification
to 450mm the suitability value of level S3 increase s, while the fuzzy value of N1 decreases. This
provides a ?softer? transition betwee n suitability levels when compared to Boolean classification
(see Figure 2-10).
The rules needed to implement the example in Figure 2-11 are provided in Table 2-4.
Table 2-4 Example of two Boolean and five fuzzy rules specifying effective soil depth requirements for perennial
crops
# SUITABILITY LEVEL ?
(lower threshold)
?
(central value)
?
(upper threshold)
RULE TYPE
1 Permanently unsuitable (N2) 0 - 150 Boolean
2 Permanently unsuitable (N2) 150 150 250 Fuzzy
(asymmetrical)
3 Unsuitable at present (N1) 150 240 330 Fuzzy (symmetrical)
4 Marginally suitable (S3) 250 390 520 Fuzzy (symmetrical)
5 Moderately suitable (S2) 330 470 620 Fuzzy (symmetrical)
6 Highly suitable (S1) 450 600 600 Fuzzy
(asymmetrical)
7 Highly suitable (S1) 600 - 1350 Boolean
For symmetrical fuzzy rules, an additional central value ( ?) needs to be specified to indicate the
centre of the membership function. For instance, in rule 4 membership increases from 250mm
and reaches a maximum membership at ? (i.e. 390mm). The membership value then decreases
with increasing depth until it reaches 520mm effec tive soil depth. For asymmetrical functions,
the central value is set equal to ? or ?, depending on the slope direction of the function. Rule 6,
for instance, has a positive slope (i.e. membership increases as effective soil depth increases) and
is specified by setting the central value equal to ? (i.e. 600mm).
36
2.4.2 The inference engine
All seven rules in Table 2-4 constitute one land use requirement (effective soil depth) for
perennial crops. Because the success of most land uses will depend on multiple requirements,
each land property value pi must be sequentially tested against the fuzzy set function
),,;( ???xS of each requirement ri by setting x = pi in Equation 2-3 and Equation 2-4. The
resulting membership values are combined and averaged using Equation 2-6.
n
p
S iij
),,;( ?????= Equation 2-6
where Sj is the overall suitability value for land unit j ;
?i is the membership value for land property pi ;
? is the upper limit of land requirement ri ;
? is the central value;
? is the lower limit of land requirement ri ; and
n is the number of land properties.
The result is an overall suitability value ranging from 0 to 1. In Figure 2-6, if p1 is effective soil
depth and p2 soil acidity of a particular land unit u j , with p1 set equal to 600mm and the pH of p2
set to 6, then Sj = (0.5 + 1.0)/2 = 0.75. To differentiate between the importance of land
properties, weightings can be set and multiplied by individual membership values as expressed
in Equation 2-7.
),,;( ?????= iiij pwS Equation 2-7
where S is the overall suitability value for land unit j ;
w i is the weight of land property pi ;
?i is the membership value of land property pi ;
? is the upper limit of land requirement ri ;
? is the central value; and
? is the lower limit of land requirement ri.
Suppose that effective soil depth (p1 ) in Figure 2-6 is twice as impor tant as soil acidity (p2 ) for
perennial crop suitability and that p1 is equal to 600mm and the pH for p2 is 6, then Sj =
((0.66*0.5) + (0.33*1.0)) = 0.66. It is important th at the land property weights sum to 1 and that
they are consistent when compared in pairs. To ensure consistency, Saaty?s (1977) analytical
hierarchy process (AHP) is recommended when weighting land properties (refer to Section
2.3.1.4).
37
Figure 2-11 shows how the upper and lower limits of several membership functions can be
manipulated to represent the different suitability levels (S1, S2, S3, N1, N2). Suitability values
must be calculated for each level using Equation 2-7. The level with the highest suitability value
is identified and used to classify the land unit into that suitability level. In addition, to reclassify
land units into suitability levels, the suitability values of different land uses can be compared to
determine the most suitable land use for a particular land unit.
2.4.3 Land unit database
In the expert system approach, land units are usually stored as vector polygons in a GIS
database. The vector data model is preferable b ecause multiple attributes (i.e. land properties)
can be linked to each land unit. This method is e fficient because the geometrical properties (i.e.
boundaries) of each land unit are stored only once, saving space and processing time. Another
advantage of storing land properties as attributes to polygons is that suitability analysis merely
involves the comparison of different fields in the database, which means that suitability values
can be calculated by performing simple tabular operations. Although the use of tabular
operations instead of spatial operations (i.e. overlaying) has significant benefits concerning
processing speed, the main advantage of such operations is that the entire suitability analysis can
be performed using standard database management operations (see Section 3.2.6). This means
that GIS software is not required for the analysis part of the procedure. The only step in the
expert system procedure involving spatial functionality is the mapping of the resulting suitability
levels. Mapping is, however, a relatively simple operation and can be incorporated without the
use of expensive GIS software. The next sectio n overviews how mapping is handled in existing
expert systems for land suitability analysis.
2.4.4 Existing land evaluation systems
The Automated Land Evaluation System (ALES) is one of the first and most popular land
evaluation expert systems in existence. ALES, developed at Co rnell University (Rossiter 2001;
Rossiter & Van Wambeke 1997), is an expert sy stem based on Microsoft?s Disk Operating
System (MS-DOS) that allows users to build a data base of rules to evaluate land according to the
methods presented by the Framework for land evaluation (FAO 1976). The system includes a
framework for a knowledge base, a land unit database and an inference mechanism that relates
the knowledge base to the land unit database (Rossiter 2001).
ALES has been applied in thirteen cases of whic h the most recent examples include applications
for land use planning in the Philippines (Lantican et al. 1998), the potential of sustainable wheat
production in Uruguay (Mantel et al. 2000), banana production potential on Hainan island, China
38
(Mantel, Zhang & Zhang 2003) and for finding alternative uses for forested land (Fernandez
Ruiz 2003). The frequency of appearance of publica tions in which ALES is used is declining,
most probably because the last version (4.65) wa s released in 1996. A major drawback of ALES
is that it is programmed in MS-DOS, for which support was disc ontinued with the release of
Windows 2000. ALES also lacks a user-fri endly graphical interface, which makes its
implementation difficult. Another disadvantage is that an external GIS is required to view the
results, which reduces the ability of the system to interactively produce different land use
scenarios.
Another DSS developed since the early 1990s is the Mediterranean Land Evaluation Information
System (MicroLEIS) which has evolved into an agro-ecological decision support system with a
range of tools for land use decision making. It focuses on soil protection by improving
agricultural soil use, its planning and its management. The toolkit in cludes a database, statistics,
expert system and a neural network. As with ALES, the original Mi croLEIS version was non-
spatial, which means that a separate GIS was needed to input data and display results.
Fortunately, a GIS version has re cently been implemented in ArcView GIS to overcome this
limitation. Another useful development in MicroLEIS is the establishment of a website to which
users can upload tabular data, carry out suitability analysis, and download the results for
visualization on any GIS. In spite of its lack of an integrated spatial component, the website is
reportedly very popular with more than seven hundred registered users in 2004 (De la Rosa
2002; De la Rosa et al. 2004).
An expert system called the Intelligent System for Land Evaluation (IS LE) was developed in
Borland Delphi as an MS Windows-based stand- alone system. The ISLE components include a
user-friendly graphical interface, an existing FLEX expert syst em and the Borland database
engine (BDE). Meanwhile, a major improvement over ALES has been the use of ESRI?s
MapObjects to include a fully interactive mapping component, allowing users to easily generate
different scenarios and to see the results spatially (Tsoumakas & Vlahavas 1999). MapObjects is
a powerful collection of embeddable mapping and GIS components which developers can use to
create applications that include dynamic maps and GIS capabilities (ESRI 2002c).
MapObjects was also applied in a SDSS called LEIGIS (Land Evaluation using an Intelligent
Geographical Information System) to provide interactive mapping functionality. LEIGIS is
similar to ISLE in most aspects, except that it was developed in MS Visual Basic and it uses the
CLIPS expert system instead of FLEX (Kalogirou 2002).
The literature review indicates a distinct move ment towards fully integrated systems that
incorporate the required spatial functionality without the reliance on GIS software packages. The
39
main advantage of such systems is that access is not limited to existing GIS users. However, the
use of the Internet as a platform for fully integrated expert systems has not yet been attempted.
Given the popularity of the MicroLEIS website, ther e is a clear need for a fully integrated web-
based land evaluation expert system.
2.5 SUMMARY
Land suitability analysis involves the evaluation of land properties against land use requirements
to determine if a parcel of land is suitable for a particular use. Because the number of properties
(also called land characteristics or qualities) is potentially large and geographical in nature,
spatial techniques and technologies such as Boolean overlay, multi-criteria decision making and
expert systems have become fundamental tools to support the decision-making process.
Although Boolean overlay has the advantage of being simple to understand and easy to
implement with standard GIS software, MCDM has established itself as a technique for
supporting complex decisions involving numerous spatial factors with varying levels of
importance. A major advantage of MCDM for de veloping SDSS is its ability to generate
alternatives by modifying the importance levels (i.e. weights) of individual criteria. MCDM also
supports fuzzy classification, which is a le ss discrete or ?hard? decision strategy.
A convenient approach to land suitability analysis is to use expert systems. They rely on expert
knowledge stored in a database (or knowledge base) in the form of land requirement rules.
Boolean or fuzzy rules are defined according to known land use requirements or they can be
obtained from land use experts. The rules are a pplied to a database of land units using an
established protocol (i.e. inference engine) to calculate suitability values.
The main advantage of the expert system appro ach is the way in which the rules are separated
from the data. Very little data preparation is needed for expert systems because measurement
standardization is inherent in the rules. This enables the use of standardized rules that are
independent of the geographical area in which the land suitability assessment is carried out. In
addition, it facilitates system automation and implementation, especially in a multi-user, read-
only environment such as the Web where data-e diting and creation capabilities are limited.
Expert systems are also compatible with database management systems, web scripting and web
mapping services for carrying out suitability analysis. The next chapter e xplores the possibilities
of developing a Web-base d land evaluation system.
40
CHAPTER 3: WEB MAPPING TECHNOLOGY
Since its inception in 1989 the World Wide Web (WWW) or Web, has undergone many changes
to the technologies on which it relies. In spite of forecasts to the contrary in the 1990s, hardware
and infrastructural improvements have significantly speeded data transfer via the Internet. In
terms of software, web browsers have become increasingly sophisticated and now have
capabilities for rendering almost any type of information. These technological improvements
have demonstrably impacted the way geographical information is communicated by making the
visualization of spatial data on the Internet a reality. Web mapping tools, such as Google Maps
(Google 2005), MapMachine (National Geogra phic Society 2005) and StreetMap (MWEB
2005), enable anyone with access to a computer a nd the Internet to explore geographical data
online and produce maps on demand. As a result, more people access geographical information
through the Internet than via any other medium (Longley et al. 2002). It is this popularity of the
Internet as vehicle for delivering spatial information that this research aims to exploit.
In order to develop a Web-ba sed land evaluation system, a thorough understanding of the
available technology is required. This chapte r focuses on the current Web technologies and
concepts related to Internet mapping applications. This is done by first examining each of the
major web components, followed by a descripti on of the type of web maps published on the
Internet. The chapter concludes with an overview of existing web mapping services.
3.1 WEB APPLICATIONS
A web application is a web browser-based appli cation that is accessed via a computer network
such as the Internet or an intranet. Essentially, a web application is a website that provides a
specific function such as Internet banking, online stores, electronic discussion groups and web
mail (e.g. Microsoft Hotmail and Google Gmail).
The main benefit of a web appli cation is that no software other than a standard web browser is
required to use the application. This gives de velopers the ability to update and maintain
applications without the need to distribute (and redistribute) software to potentially millions of
clients. This client-server architecture is not a novel idea as it is the basis on which mainframes
and minicomputers have been functioning since the early 1970s. More recently, the dumb
terminals used by these systems have given way to multiple-task personal computers running
web browsers.
In addition to cost savings effected by the distribution and maintenance of web applications,
users also benefit because only standard web browser software is needed to run such
41
applications. Web browser software is not only free, but it also supports most platforms (i.e.
Windows, Linux, Macintosh etc.) making web app lications platform-independent. Users also do
not need powerful computers as most of the processing takes place on the web server. Another
significant benefit of web applications is that users require little additional skills or training to
use them as the interface consists of standard web pages and components such as text, images,
form fields and buttons, with which most users are familiar. This means that users are more
likely to adopt and use web applications than other forms of implementations.
The main limitation of web applications is that users need an Internet connection to use them.
The speed of the connection (i.e. bandwidth) is also a limiting factor because complex graphics
and large volumes of information are time-cons uming and costly items to download, making
such systems sluggish and unresponsive. Although many of these limitations can be overcome
with high-speed (i.e. broadband) Internet access, this technology is unlikely to become readily
available to all South Africans.
While the familiarity of web interfaces improves user acceptance and adoption, the interfaces
also limit the flexibility of web applications. The graphic capability of web applications is
especially restrictive. Consequently, web mapping applications are difficult to implement and
they are usually limited to viewing and manipulating existing drawings or maps. There are,
however, several web applications, such as Google Maps (G oogle 2005), MapMachine (National
Geographic Society 2005) , AlertNet (Reuters Foundation 2005) and StreetMap (MWEB 2005),
that allow users to view, edit and create maps online.
3.2 WEB COMPONENTS
Information on the Internet is mainly stored as web pages, which can be downloaded and viewed
by Internet users. The web pages are physica lly stored on computers, or web servers,
permanently connected to the Internet. Specialized web browser software is used to access web
pages. To initiate a request for a web page, the user enters a uniform resource locator (URL) into
the web browser. A URL is simply an easily r ecognizable representation of the web server?s
Internet protocol (IP) address, which is stored in a distributed Internet database. Once the
browser has looked up the IP address, the browser sends a hypertext transfer protocol (HTTP)
request to the server to retrieve the required page. The page is then sent back to the web browser
to be rendered along with any files referenced by it. Six interrelated web components crucial to
its application for web mapping are examined in the following sections.
42
3.2.1 Web browsers
Much of the process involved in retrieving a web page is handled by web browser software
mostly hidden from the user. Popular browsers such as Microsoft Internet Explorer, Mozilla
Firefox and Opera can furnish many types of informa tion including text, graphics, audio,
interactive multimedia, and applets (i.e. small software components).
Although the basic file format for a web page is HTML (hypertext markup language), most
browsers natively support a variety of additional image file formats, such as JPEG, PNG and
GIF, and they can be extended to support more through the use of plug-ins. The combination of
HTTP content type and URL protocol specificati on allows web page designers to embed images,
animations, video, sound, and streaming media into a web page, or to make them accessible
through the web page. The way in which text, imag es and other media are presented to the web
user is defined by a markup language.
3.2.2 Markup languages
A markup language combines text and information about how text should be structured and
presented. The best-known markup language is HTML, a derivative of standard generalized
markup language (SGML) developed in the 1960s . The first version of HTML (called HTML
Tags) was introduced by (Berners-Lee 1991) to de scribe the structure of a web page and has
since been revised and extended several times to its current version 4.0.
The structure of web pages is defined through th e use of a hierarchy of parent- and child-HTML
elements (or tags) that are interpreted by the web browser. An example of a HTML document is
shown in Figure 3-1.
Figure 3-1 Example of a HTML document
In the first line the HTML label is used to indicate that the document is in HTML format. The
HTML element is called the root element as it can have no parent elements. All element labels
must be enclosed in corner brackets (i.e. < and >) and termin ated by a corresponding end label.
End labels are identified by the forward slash character ?/? before the label. For instance, the
simple pageSTELLENBOSCH UNIVERSITY
Click here to open the
page to Stellenbosch University website
43
HTML label needs a corresponding / HTML label to end its influence, as shown in the last line of
the document.
An HTML document usually consists of a heading and a body section. The heading section
contains meta-data and is identified with the HEAD element. The heading can have a number of
child elements, including the title of the document as specified using the TITLE label. The body
section is defined using the BODY label and includes the visible contents of the web page. HTML
elements not only define the structure of the page, but can also be used to describe how the
content will be presented. For instance, the I element can be used to display text content as italic
as implemented in line four of the example. Elements can also be used to link parts of a web
page to other web pages. This is done through the A element, which specifies a hyperlink to
another web page that will open when selected. The A element is an example of an element that
requires attributes to influence its behaviour. In this example, the HREF attribute is used to
specify the address of the hyperlink.
Although there are currently 91 HTML elements fr om which a web developer can choose, there
is a constant demand for more elements with more functionality (W3C 2008). This demand has
led to the development of extensible markup language (XML), a general-purpose markup
language through which additional elements can be created as needed (W3C 2006). Each
element is defined in a document-type definiti on which is interpreted by the browser. It is
important to note that XML is not a replacem ent of HTML, but merely an extension.
Figure 3-2 illustrates how XML can be used to st ore the contents of an email. The first line
declares that the document is in XML format (version 1.0) using a Latin/West European
character set (ISO-8859-1). The root element EMAIL is specified in the second line and defines
the type of object that the XML document is desc ribing. The next three lines specify three child
elements, namely ADDRESS, SUBJECT and BODY. Each child element contains the relevant
data and is terminated using syntax similar to HTML. The end of the document is defined by the
EMAIL element.
Figure 3-2 Example of an email stored as XML
Although the examples above make XML and HT ML seem similar, the two markup languages
have two different purposes. Where HTML inst ructs web browsers how content should be
avn@sun.ac.za
XML example
This is an example of an e-mail stored in XML
44
interpreted, XML is used for data storage and transfer. XML syntax is more restrictive than
HTML and is therefore more eas ily parsed by web browsers.
With the continuous elaboration of HTML to include progressively more functionality, the
computer memory and processing capabilities needed to render web pages have also increased.
This is problematic for small devices such as cellphones and palmtops that have limited
processing capabilities. As a result, XML was used to devel op an additional markup language
called extensible hypertext markup language (XHTML ). Essentially, XHTML is a more efficient
version of HTML due to the formatting restricti ons placed on its structure and it is widely used
for small devices (W3C 2002).
Web pages can be created using a combina tion of HTML, XML and XHTML code. Once the
code has been parsed and translated by the web browser, it is displayed as text and images.
Unfortunately, web content created by markup languages is static, which has led to the
development of client-side scripting.
3.2.3 Client-side scripting
Client-side scripting refers to a class of programs executed by a web browser to alter the display
and behaviour of static HTML web-page conten t. The term ?client-side? indicates that the
operations are carried out on the client?s (user?s) computer or de vice, enabling web pages to react
faster to users? actions. Client-side scripting is often used to make web pages more interactive
and dynamic. Because of this effect, web pages that use a combination of HTML and client-side
scripting are said to be created using dynamic hypertext markup language (DHTML)
(W3Schools 2008a).
Another important function of client-side scripti ng is to improve the robustness of websites and
applications. With scripting, users can be lim ited to perform only certain actions, depending on
the current status of the application (K?bben 2001) . For instance, client-side scripting may be
used to restrict users to enter only numbers into a form field. Client-sid e scripting can also be
used to guide users through interactive dialogs such as warnings and choice menus, thereby
limiting errors and improving system usability (Canter 2004).
The most popular client-side scripting language is JavaScript (Mozilla Foundation 2008). This
language, introduced in 1995 by Netscape, can be us ed to manipulate web page objects such as
windows, documents, links and forms to perform a wide range of tasks. JavaScript code (see
Figure 3-3) can either be incorporated in a web pa ge or it can be called from a separate file. The
advantage of keeping code separate from the web page is that the same code can be reused by
different web pages, thereby limiting duplication and coding time.
45
Figure 3-3 Example of JavaScript code that displays an error message when the web page is opened.
Whether JavaScript code is imbedde d into a web page or loaded from a separate file, all the code
that is referenced by the web page must be downloaded from the server before it can be
executed. The more complex applications become, the more code needs to be downloaded,
which can slow down the overall performance of a website. In addition, JavaScript can only
perform operations on information that has already been downloaded from the web server. This
not only limits its capabilities, but also poses security risks as sensitive information or code, such
as financial records and passwords, can be viewed by any Internet user. Client-side scripting
should therefore be limited to operations concerning the user interface and should not be used to
retrieve sensitive information from the web server. Instead, for secure, dynamic information
retrieval, server-side sc ripting should be used.
More advanced client-side functionality can be implemented using Java, an object-orientated
programming language that is platform independent (Sun Microsystems 2008). However, in
contrast to JavaScript which is inherently supported by most web browsers, applications
developed in Java require special interpreters that need to be installed on the client. Similarly,
web browser plug-ins can also be installed on clie nts to extend the functionality of existing web
browsers (K?bben 2001).
3.2.4 Server-side scripting
The development of web pages using HTML, XML a nd JavaScript can be very costly and time-
consuming processes, especially for websites with large volumes of information that must be
updated regularly. Using server-side scripting, web pages can be created automatically on
request. In essence, the main function of server-side scripting is to instantly generate HTML,
XML and JavaScript code to meet the requireme nts of a specific user. For instance, when an
Internet banking user requests to view a financial statement, the web server generates unique
code to display the relevant information. A new web page, based on the information supplied by
the user, is created and once the page has been downloaded, it is removed from the server.
A major advantage of server-side scripting over client-side scripti ng is that operations are carried
out on the server before any information is sent to the client. Because the scripts and operations
simple page
46
are hidden from the web user, less data needs to be sent over the Internet. This improves website
responsiveness and security. The down side of server-side scripting is that the server carries more
load due to the additional processing required to execute the scripts and retrieve the necessary
information. Fortunately, technological advances in computer processing alleviate much of the
demand for additional processing power.
Fundamentally, any programming language can be used for server-sid e scripting, although
several languages have been specifically developed for this purpose. Of these, PHP (PHP
hypertext preprocessor) (35% ) and ASP (active server pages) (21%) are the most popular (Nexen
2007). Although PHP can be run on any web serv er, it is most often used for Linux
implementations, while ASP, developed by Microsoft, runs only on Windows-based web servers.
Other server-side solutions include common gateway interfaces (CGI) and application
programming interfaces (API). While CGI scripts are used to develop interfaces between
existing software (such as a GIS package) a nd web server software, API are server-side
programs (often written in Java) to extend the functionality of web servers (Jiang 2003).
3.2.5 Web servers
A web server is a combination of hardware and software. As for hardware, essentially any
computer can be used as a web server. However, most servers are dedicated computers with
special hardware configurations to handle requests from a large number of users. Web servers
are generally equipped with powerful processors and large volumes of computer memory. Most
server hardware is currently supplied by IBM (31%) and Hewlett-Packar d (28%), while DELL
(12%) is becoming a strong conten der with a market share growing by more than six per cent
annually (Modine 2007).
A number of web server software programs are available. Although more than half of all web
servers run Apache software, current trends indicate that Apache will be replaced in 2008 by
Microsoft?s Internet Information Services (IIS) so ftware as the most popular web server software
(Netcraft 2007).
Many web servers include a database component to manage the large volumes of data that some
web applications require. Information is extracted from the database and dynamically converted
to web pages using server-side scripting. Data bases can either be hosted directly on the web
server or they can be stored on a dedicated database server to relieve the load on the web server.
The configuration will greatly depend on the size of the database and the database management
system being used.
47
3.2.6 Relational database management systems
A database management system (DBMS) is a complex set of software programs specifically
designed to control the organiza tion, storage and retrieval of data from a database. Although
several types of DBMS exist, relational DBMS (RDBMS) are the most popular. The relational
database model, first proposed by Codd (1970), uses predicate logic and set theory to represent
all data as mathematical relations.
The relational model consists of three data com ponents: structure, manipulation, and integrity.
Data structure refers to the organization of data , while data manipulation signifies the types of
operations users can perform on the structure. Th e set of ?business? ru les governing how data
values behave on these operations, is known as data integrity (Fleming & Von Halle 1989).
Relational data is organized into two-dimensional tables (also called relations or entities). Each
table consists of a set of uniquely named columns (also called attributes or fields) and unnamed
rows (also called tuples or records). To be relati onal, the data must be organized in the table so
that each row is unique. Entries in columns must be single-valued and one kind, while the
sequence of columns (left to right) and rows (top to bottom) is insignificant.
Data in tables can be manipulated by relational assignment. Although relational assignment is
similar to variable assignments in computer programming, in relational databases the variable is
a table and the assignment expression involves other tables. Eight operations, namely SELECT,
PROJECT, PRODUCT, JOIN, UNION, INTE RSECTION, DIFFERENC E and DIVISION are
available for relational assignments (Fleming & Von Halle 1989).
Data integrity, the third component of the relational model, is governed by rules that constrain
permissible values in the table columns and the actions that should be taken to remove records.
For example, the entity integrity rule states that no null values (i.e. empty or zero) are allowed in
primary keys, that is the column or set of columns that uniquely identify each row. Another
important rule to ensure data integrity is the referential integrity rule which addresses the
integrity of foreign keys. A foreign key is a column or set of columns functioning as a primary
key in another table. The rule states that the valu es in a foreign key must be either null or must
have values matching the values in its corresponding primary key. There are many other rules
meant to deal with the integrity of all columns, including primary and foreign keys. These so-
called ?domain integrity ru les? restrict column en tries to values that correspond to each column?s
domain. A domain refers to a logical pool of permissible data types (e.g. text, number), lengths
and ranges as well as settings such as default values, uniqueness, and nullability. Although the
relational model does not dictate how data integrity is implemented, integrity is a logical and
48
integral part of any relational database and should be defined and endorsed without involving the
user in the technical implementation (Fleming & Von Halle 1989).
Users and database developers usually interact with databases through a standard database
language called structured query language (SQL), which has two ma in functions. First, it can be
used to manipulate the data in a database through the use of SQ L relational operators such as
SELECT, INSERT, UPDATE, DELETE, JOIN and UNION. An example of a SELECT
statement combined with a JOIN is shown in Figure 3-4.
Figure 3-4 A SQL statement usin g the SELECT and JOIN operators
The second function of SQL is to al ter the structure of an existing database or create an entirely
new database using operators such as CREATE, ALTER and DROP. Although some database
software includes additional operators, most relational database software can be accessed and
manipulated using the same basic SQL statements (Fleming & Von Halle 1989).
A number of RDBMS software packages is available. According to Pettey (2007), Oracle
currently produces the most popular database software which commands 47% of the market
share, while other popular DBMS developers include IBM (21%) and Microsoft (17%). Oracle
and IBM?s software offerings ar e aimed at the enterprise level market, while Microsoft also
caters for the small business and home office applications with their Microsoft Access software.
Owing to its portable file structure, Microsoft A ccess is also widely used for web applications
and it is the database of choice for rapid application development because it is so easy to set up
and manipulate. Due to its user-friendly inte rface and compatibility with other Microsoft
products, many developers prefer to use Microsoft Access to design and implement prototype
databases. Once the prototype is stable in terms of its structure, or when the size of the database
nears the software?s limitations , it is usually replaced by a more robust RDBMS such as Oracle
or SQL Server. The latter is Microsoft?s en terprise database solution (Microsoft 2007).
The choice of a RDBMS largely de pends on the type, size and comp lexity of the web application
for which it will be used. However, it is essential that the server-side scripting language and the
RDBMS are compatible. Fortunately, many operati ng systems or third-party developers offer
standard open database connection (ODBC) interfaces to translate requests form web
applications into a format that DBMS can interpret.
SELECT field1, field2, field3
FROM first_table
INNER JOIN second_table
ON first_table.keyfield = second_table.foreign_keyfield
Source: W3Schools (2008b)
49
The web components described above illustrate the complexity of web applications. While
RDBMS and server-side scripting are used to dynamically generate web pages, client-side
scripting, HTML, images and ot her media are used to create dynamic web pages. An additional
level of complexity is added by web applications that produce dynamic web maps.
3.3 WEB MAPS
Web maps refer to all the types of maps distri buted via the Internet. Although there are various
types of web maps, nine types, distinguished by their characteristics, are overviewed in the next
section, followed by an exposition of their formats.
3.3.1 Characteristics of web maps
Web maps can be characterised as being either st atic or dynamic. In addition, each of these main
web map types can be further classified as view only and interactive web maps (Figure 3-5).
These, as well as some additiona l characteristics of web maps (distributed, animated, real-time,
personalized, reusable, and collaborative), are considered below.
Figure 3-5 Classification of web maps
3.3.1.1 Static vs. dynamic
Static web maps are similar to paper maps in that they are created once and are infrequently
updated (Kraak 2001). Although many static maps are created specifically for Internet
distribution, some static maps are simply digital (scanned) versions of hard-copy paper maps.
Such maps are not always suitable for Internet use because the high resolution needed to
accurately represent the quality of the original printed map often results in file sizes being too
large to download (Peterson 2003). Consequently , some mapping applications allow users to
select sections of maps for downloading. In addition, client-side a nd server-side scripting is used
to let users change (i.e. pan) the extent and position of these sections interactively, or enlarge and
reduce (i.e. zoom) the map scale as needed. Us ually, such dynamic web maps are created each
time it is downloaded. The map is therefore not stored on the server, but is dynamically
Web Maps
Static maps
Interactive
Dynamic maps
View only
Interactive
View only
Adapted from Kraak (2001)
50
generated from a database or GIS on request. Sp ecialized server-side programs, called web map
service (WMS) software, are required to c onvert the source data into map format.
3.3.1.2 Interactive vs. view only
Client-side scripting can be employed to enhan ce view only or non-interactive web maps by
providing functions like hyperlinking and active areas through which users can open other
spatially related web pages or multimedia by clicking on different places on a map (Taylor
2005). Although static maps are frequently used as bases for producing these so-called
?clickable? maps, many interactiv e maps are created dynamically.
3.3.1.3 Distributed
Some dynamic web maps are created from distributed data sources located on different WMS.
Such maps, called distributed maps, are requested through a standardized protocol understood by
all the relevant WMS. Maps are delivered along with service-level metadata and map feature
attributes (Tsou 2003). A list of availa ble WMS are listed on wms-sites.com .
3.3.1.4 Animated
Although dynamic or even static maps can be made to appear animated (i.e. with animated
symbols or panning) the term animated web map refers to a map that illustrates spatial change
over time using animation (Peterson 2003). A good example of such a map is a weather map
showing the movement of a weather system. Animated web maps are produced by displaying a
sequence of static maps to give the appearance of movement (Cartwright 2003).
3.3.1.5 Reusable
Due to the cost of producing and serving web maps, some companies sell maps to web
developers who cannot produce maps themselves. For instance, Google Maps allows other
websites to use (or reuse) maps that are dynamically generated on G oogle?s servers. Although
Google maps appear to be part of the website in which they are displayed, the actual map content
is downloaded from a Google Maps server each ti me the particular web page is accessed.
3.3.1.6 Real-time
A major advantage of the Web over other means of map publication is that distribution can be
nearly instantaneous. This enables the mapping of phenomena in real-time. An example of such
an application is satellite tracking of vehicles which permits owners to view the position of their
vehicles at any given moment (Altech Netstar 2008). Real-time maps can also provide
functionality such as location based services (LBS) (Gartner 2005).
51
3.3.1.7 Personalized
Some web applications give users the functionality to produce personalized maps. Examples are
occasion maps (e.g. location of a function or meeting) or maps showing directions to a particular
location (e.g. conference venue, office, shop, hotel). With such applications, users can specify
the map?s extent, scale and visible features (i.e . layers) as well as colour schemes and unique
symbolizations.
3.3.1.8 Collaborative
A collaborative web map is a map created and edited by various users (Caquard 2003). An
example of a collaborative mapping application is WikiMapia, an implementation of reusable
Google Maps that allows users to add info rmation to any location on earth. Although the
technology is still immature and complex, it has much potential, especially for data collection
and spatial applications in which participation is required.
The above discussion of web map characteristics is not exhaustive, but it provides a good idea of
the types of web maps now in use. Additional types of maps and features are expected to appear
as web mapping develops and new uses for web maps are found. Although different types of
technology are used to create and distribute web maps, the format of maps on the Web is
restricted by the formats recognized by web browsers. The next sec tion is a synopsis of the types
of formats used for web maps.
3.3.2 Formats of web maps
Like GIS data, web map formats are of two ki nds ? raster and vector ? each considered
separately below.
3.3.2.1 Raster
Raster files, or images, are the most common data format for web maps. A raster is a two-
dimensional array of grid cells or pixels (short for ?picture elements?) with each cell representing
a colour-intensity value. By combining three rasters representing red, green and blue (RGB)
respectively, a full-colour image can be formed. The number of distinct colours that can be
rendered by combining three rasters (colour bands) depends on the number of intensity values
allowed in each grid cell. For instance, if each gr id cell can store 256 values (8 bits), then a RGB
image can store a total of 256 x 256 x 256 = 16 777 216 (or 2 24 ) distinct colours. Such images
are called 24-bit images because they can store 8 + 8 + 8 = 24 bits of data (colours).
Most web maps are stored as either graphics interchange format (GIF ) or joint photographic
experts group (JPEG) images (Peterson 2003). The major difference between GIF and JPEG
52
images is the number of colours that each can represent. Where 8-bit GIF files can only display
256 distinct colours, JPEG im ages can represent up to 224 colours due to its 24- bit capability. In
general, images require large disk storage space and are slow to download via the Internet. To
limit file size, GIF uses run length encoding (R LE), a popular compression technique that can be
reversed (i.e. uncompressed) without any loss in quality. RLE compression is only efficient for
images that include large homogeneous areas. Fo r photographs containing a high level of colour
variability, JPEG format is more suitable.
To sufficiently compress the additional colour information stored in JPEG images, a specifically
designed compression technique is used that generalizes colour vari ation. This so-called ?lossy?
compression technique is non-reversible, meaning that the quality of the original image is lost
once compression has been carried out. However, the level of compression can be limited to
restrict the loss of quality to a degree that it is unnoticeable in most applications.
Although JPEG and GIF are the mo st popular raster formats on the Web, a third format called
portable network graphics (PNG) is rapidly gaining popular ity. PNG was designed in 1996
specifically for transferring images over the Internet and is similar to GIF. The major advantage
of PNG over GIF is that it has bit depth of 24, which enables it to display up to 2 24 colours. PNG
therefore combines the advantages of ?lossless? GIF and the colour range of JPEG. As with GIF,
PNG format is not suitable for images with a high level of colour variability.
Raster-based web maps are popular because they ar e compatible with computer data structures
(e.g. arrays) and digital monitors. Web maps in this format can also be rendered by web browsers
without the need for any additional software. The ma in drawback of raster-based web maps is the
inflexibility of the scale at which they can be displayed. Images are usually created for the
resolution at which they will be viewed (e.g. 96dpi for the Mi crosoft Windows operating system)
and cannot be enlarged without loss of detail. This limitation can be overcome by using vector-
based graphics for creating web maps.
3.3.2.2 Vector
The vector data model stores spatial features as points, lines and polygons. Points are represented
by individual coordinate pairs referenced to a common coordinate system. Lines are sequences of
points that are connected, while polygons (areas) are closed lines (i.e. their starting and ending
points coincide).
Three vector formats, namely portable document fo rmat (PDF), scalable vector graphics (SVG),
and shock wave flash (SWF), are currently used for web maps (Peterson 2003). Each has
different capabilities. Due to their portability and compatibility with desktop publishing, PDF is
53
very popular for the distribution of static maps. SWF (also called Flash) is a powerful medium
for producing interactive, animated web maps. Because maps in PDF and Flash format need to
be downloaded fully before they can be viewed, they are not suitable for dynamic, distributed,
personalized or collaborative web mapping. The only vector format that can be used to create
web maps with these characteristics is SVG, an open standard for vector graphics on the Web.
SVG is based on XML and is therefore highly compatible with HTML and related formats.
Unfortunately, not all web browsers are capable of rendering SVG data and special software
(plug-ins) is required. The lack of web br owser support for the SVG format has limited its
adoption by users and developers. However, SVG is a relatively new format and it is expected
that all web browsers will support the format in the near future.
3.4 WEB MAP SERVICES
As discussed in Section 3.3.1.1, the function of a web ma p service (WMS) is to produce and
serve dynamic web maps. Requests for dynamic maps are usually received in a standard WMS
format, as specified by the Open Geospatial Consortium (OGC) (De la Beaujardiere 2004), and
interpreted by the software to extract the necessary data from a GIS database. Based on the
user?s requirements, the spatial data is then converted to a map and served through a web server
as an image or a set of vector features viewable with a web browser.
A WMS comprises three function al components: (1) data storage and retrieval; (2) map
production; and (3) map distribution. Each of th ese components is described in the following
sections.
3.4.1 Data storage and retrieval
Web applications allow users to download data, but rarely allow them to upload data. This
limitation mainly exists to prevent users from uploading any harmful content such as computer
viruses and it ensures that hackers cannot attack and corrupt systems. Allowing users to upload
data also causes logistical problems for storage space and access rights.
Internet mapping has additional limitations regarding users uploading data because access to the
map server software ? running on the web server ? is required to set up and design the maps that
will be served. For security reasons, access to this software is usually limited to an administrator
because anyone else with access could edit or even delete the maps being served. To reduce the
security risk, a collection of spatial data sets is available for viewing and users needing specific
data for their analyses can request it to be manually loaded by the system administrator.
54
The GIS data used by a WMS can be stored in se veral formats. Popular formats for vector data
include SVG and shape file format, while tagged image file format (TIFF) or JPEG 2000 format
are often used for raster data. Storing large GIS da ta sets in these file structures is sometimes
inefficient because the entire data set needs to be loaded onto the server?s memory. Although this
limitation can be overcome by using spatial indexing (i.e. divide the data into smaller spatial
units and only load the necessary areas into memory), many WMS applications rely on RDBMS
(see Section 3.2.6) for the storage and retrieval of spatial data. RDBMS not only improves
efficiency, but also protects data integrity by managing events like simultaneous requests from
multiple users. RDBMS has other useful data management functionality such as regular backups
and versioning (the ability to centrally monitor changes and to roll back or undo to a previous
version).
Unfortunately, standard RDBMS are not suitable for storing geographical features such as lines
and polygons requiring variable length records (a line can theoretically consist of an unlimited
number of coordinate pairs). Because most RDBMS are designed to mainly store text strings and
numbers, they are also not good repositories for raster data. Standard RDBMS can however be
modified by adding additional software to manage the conversion of spatial data into non-spatial
data structures. The need for this software ha s prompted several commercial RDBMS, such as
Oracle and Informix, to offer spatial extensions which handle these conversions automatically
(IBM 2007; Oracle 2007). ESRI has produced separate spatial data engine (SDE) software,
called ArcSDE, which manages the storage of spatial data in Oracle, Informix, DB2 and
Microsoft SQL Server databases, without th e need for spatial extensions (ESRI 2007c).
Regarding hardware, the main requirement for the storage and retrieval of spatial data is hard-
drive space. WMS implementati ons such as Google Earth and the United States Geological
Survey?s (USGS) Earth Resources Observati on and Science (EROS) data centre require
enormous volumes of hard-drive space. Multiple , dedicated data servers are also needed to
process the continuous requests for data. Other ha rdware requirements include tape drives for
backing up data as well as network infrastructure to connect to the servers that are responsible
for map production.
55
3.4.2 Map production
The production of dynamic maps for Web distri bution has special software and hardware
requirements. WMS software can be categorized into open source (OS) a nd proprietary software.
Whereas the intellectual propert y of proprietary software is owned by an individual or a
commercial company, the intellectual property of OS software is relaxed or non-existent. This
means that the source code of OS software is available to anyone and the software can be
obtained and used for free. According to the Open Source Initiative (Open Source Initiative
2007: s.p.), OS is ?a development method for soft ware that harnesses the power of distributed
peer review and transparency of process?[which promises]?bette r quality, higher reliability,
more flexibility, lower cost, and an end to predatory vendor lock-in.?
Examples of OS WMS software are ALOV Map, GeoServer, GeoTools, MapIt!, MapServer and
MapZoom. Of these, MapServer is currently the most popular with 49 active public WMS listed
on wms-sites.com and more than one hundred implementations listed on the MapServer website
(Lime 2006). MapServer, origina lly developed by the University of Minnesota in cooperation
with the National Aeronautics and Space Administration (NASA) and the Minnesota Department
of Natural Resources, is now maintained by a number of developers internationally. The
software is extremely versatile regarding the server-side scripting (e.g. PHP, Python, Perl, Ruby,
Java, and C#) and the platforms (e.g. Windows, Linux, Mac OS X and Solaris) it supports.
MapServer is also compatible with several RDBMS including Oracle Spatial, MySQL and
PostGIS (Lime 2006).
Probably the largest MapServer implementation is the United Nations Environment Programme
(UNEP) GEO Data Portal, an online database of more than 450 environmental variables. Other
MapServer applications include travel maps (e.g. Komotini City Guide; Yosemite National Park
Hiking Map; Winnipeg Restaurants), online atlase s (e.g. Atlas of Eastern and Southeastern
Europe; Atlas Amazonas; Atlas of Canada) and local authority implementations (Bayawan City
Online Project Monitoring System; Naga City, Philippines; Minnesota Land Use and Land Cover
Map).
Whereas OS WMS software are mostly stand- alone packages focused on serving web maps,
proprietary WMS software packages are usually extensions of existing GIS. Examples are
Autodesk?s Mapguide, Intergraph?s GeoMedia WebMap and ESRI?s ArcIMS. Of these
packages, ArcIMS is currently the most popular with 121 public WMS listed on wms-sites.com .
Due to its scalability and robustness, ArcIMS is regularly used for extensive applications such as
National Geographic?s MapMachin e (nationalgeographic.com), the USGS EROS data centre
56
(edc.usgs.gov), and Canada?s Geoscience Data Repository (gdr.nrcan.gc.ca ). ArcIMS is popular
in South Africa, where it is used by the Department of Water Affairs and Forestry map services
(dwaf.gov.za ), the Council for Scientific and Industrial Research (CSIR) web map server
(spatial.csir.co.za ), and the South African National Department of Agriculture?s Agricultural
Geo-referenced Information System (agis.agric.za ).
While the data storage and retrieval component of WMS is highly dependent on large volumes of
hard-drive space, the conversion of GIS data in to map format requires high processing capacity.
To improve response times, multiple servers are often used to balance the processing load.
However, the use of multiple map servers increases the complexity of the software needed to
manage and perform the requests. To further reduce the load on map servers, separate servers are
frequently used to distribute the maps to web users.
3.4.3 Map distribution
Once GIS data has been extracted and converted into a map format that is web browser
compatible, the data is placed on a web server (Section 3.2.5) for distribution. The web server
requires little disk space as the maps are stored only until a user sends a request for a different
map. Web servers not only let users download maps but also handle requests for new maps.
Requests are received as URL, which are passed to the map server software. Although each map
server handles requests differently, many WMS have adopted Open Geospatial Consortium
(OGC) specifications (Open Geospatial Cons ortium 2007) which enable users to send
standardized requests to severa l WMS simultaneously to produce distributed maps (see Section
3.3.1.3).
3.5 SUMMARY
Web mapping applications are esse ntially web applications that serve, along with other online
content, dynamic and interactive web maps. Web mapping applications are developed by using
standard web components such as markup languages, client-side sc ripting, server-side scripting
and relational database management systems along with special server-side software that
produces maps on demand. The function of web ma pping software is to accept requests for maps
using a standardized protocol a nd to dynamically create maps from GIS data sets. The maps are
converted into a format compatible with standard web browsers and are temporarily stored on a
web server along with other web content (i.e. text, images and forms) for downloading.
Succinctly, web mapping applications comprise a WMS and web pages in which the created
maps are displayed. The next chapter explains how these components are incorporated into the
57
Cape Land Use Expert System (CLUES) desi gn to enable online suitability analysis
functionality.
58
CHAPTER 4: REQUIREMENT ANALYSIS AND DESIGN OF CLUES
As with most large software developments, the design of the Cape Land Use Expert System
(CLUES) was preceded by a requirement analysis, which entails establishing and expressing the
needs and constraints placed on a software product (Kotonya & Sommerville 1998) while
bearing in mind that a software requirement is a property the developed or adapted software must
exhibit to solve a particular problem (Abran et al. 2004).
The requirements for CLUES outlined in this chapte r were identified from the relevant literature
by studying the functionality needed in order to perform land suitability analysis (see Chapter 2)
and by examining the architectures and data used in similar existing systems. Good functionality,
accessibility, user-friendliness and speed were singled out as the major factors contributing to the
success of a land evaluation system. The discussion of th e requirements is followed by a
description of the system design which includes an overview of the implementation of the
CLUES components.
4.1 SYSTEM REQUIREMENTS
The properties that CLUES should exhibit are divided into two categories. Essentially, a
distinction is made between properties that are related to functional needs (i.e. what the system
should do) and operational characteristics (i.e. how the system should do it). Because of the
strong reliance of SDSS on spatial data, an additional category, data requirements, has been
added.
4.1.1 Functional needs
The functional requirements of CLUES are directly related to the secondary aim of this research,
which specifies that the system must enable users to perform spatial land suitability analysis for
anywhere within the Western Cape. The literature review determined that, in order to perform
land suitability analysis, users must be able to:
? collect and prepare data;
? identify the land uses to be evaluated;
? specify land use requirement rules;
? map land units;
? determine land properties;
? calculate (analyse) land suitability;
59
? summarize and tabulate; and
? create suitability maps.
These main functions are similar to the land evaluation steps shown in Figure 2-1. They are
however not exhaustive as additional functionality will be required to support these operations.
4.1.2 Operational characteristics
How the system?s functionality is accessed, perf ormed and presented are part of its operational
characteristics. These characteristics are expounded in the following sections.
4.1.2.1 Accessibility
The research?s aim stipulates that users must be able to access CLUES functionality via the
Internet. This requirement has far-reaching logistical implications because the system must be
able to handle multiple users concurrently. It must preferably do so without any user being aware
of other users. This means that an individual wo rking environment must be created for each user
enabling them to input their own data, define their own land uses and set up their own land use
requirement rules. Users must not only be allowed to perform individual suitability analyses;
they must also be able to store suitability maps and parameters so that they can continue with a
project at a later stage.
4.1.2.2 Performance
One of the main requirements of a SDSS is that it must facilitate scenario building. Users must
be able to explore the data and interactively see the possible effects of different decisions. The
system must be responsive to any changes in the criteria or rules and suitability maps must be
created on demand without long delays.
Research about acceptable waiting times for web pages to open has provided inconsistent
findings. According to Nah (2004), users are not willing to wait more than two seconds for web
pages to download, while Dennis & Taylor (2006) showed that se ven seconds was considered to
be acceptable. Selvidge (1999) reported that litt le difference in users? tolerance was observed
between one-second waiting ti mes and 20-second waiting times, but that there is a marked
difference between one-second and 30-second waiti ng times. Users? willingness to wait longer
depends on the perceived complexity of the information requested (Bailey: s.d.). Longer waiting
times are acceptable for content such as Internet searches, imagery and software downloads or
when the requested information is of high importance (Dobbs 2004).
60
Unfortunately, no research has yet been done on acceptable waiting times for maps. A useful
guideline is that the response time of CLUE S should match or improve upon those of GIS
software. GIS users often have to wait long for maps to be drawn or for analyses to complete.
For example, a simple overlay (union) ope ration of Western Cape farm boundaries ( ? 50 000
polygons) and land uses (? 16 0000 polygons) in ArcGIS takes more than one minute to
complete. Because land suitability analyses often involve multiple overlay operations (one for
each land property being considered), it is expected that a suitability evaluation will require
processing times of several minutes. Such long delays might however dissuade users from
exploring different scenarios. A possible solution to reducing delays might be to limit the extent
of individual analyses.
To encourage scenario building, a system design goal of a maximum delay of one minute was set
for a suitability analysis to be completed using CLUES. Response times for non-spatial
functions, such as setting up parameters and rules, must preferably be less than seven seconds.
To improve response times, provision of unn ecessary information and imagery should be
limited. The graphical user interface (GUI) must therefore be simple and functional, rather than
elaborate.
4.1.2.3 Presentation
Because the system will be developed to allow access through the Internet, the interface through
which the user will control the system will be a website. Using a website as interface can
improve usability as most users are familiar with web pages and their components (i.e. text,
images, forms and buttons).
To ensure that the CLUES website is intuitive, Nielsen?s (1994) guidelin es for user interface
design will be followed, namely to:
? provide feedback to users in order to keep them informed about what the system is doing;
? use language, phrases and concepts that are familiar to the user;
? control user freedom in order to prevent users from choosing inappropriate options;
? allow users an ?emergency exit? so that they can return to a previous state, especially if
the user has made an error;
? be consistent in the words, situations, or actions used;
? prevent errors by presenting users with confirmation dialogs;
? minimize the user's memory load by maki ng objects, actions, and options visible;
61
? introduce ways in which actions can be accelerated using shortcuts;
? not include irrelevant information in dialogs; and
? express error messages in plain language that precisely indicates the problem and
constructively suggests a solution.
The CLUES user interface should not only confor m to user-interface guidelines, but should also
follow web-design principles. The web-de sign guidelines document developed by U.S.
Department of Health and Human Services (2006), titled Research-based web design & usability
guidelines , was identified as the most suitable source for the purposes of this study. This
comprehensive document (216 pages) includes stra tegies regarding the design process, makes
suggestions on how to optimize the user?s expe rience, and provides practical guidance about
page layout, navigation, scrolling, paging, headings, links, and text.
4.1.3 Data requirements
Land evaluations conducted by others were studie d to discover existing data sets essential for
inclusion into CLUES. A wide range of applica tions was considered. These include suitability
involving agricultural, forestry, environmental and urban land uses. Most of the applications
used climate, soil, terrain, infrastructure and land use data. After stipulating an operational scale,
the value of each of these data types is discussed in the next five subsections.
4.1.3.1 Operational scale
CLUES must allow users to work at the largest possible scale. Although Lambrechts & Ellis
(s.d.) suggest that scales ranging from 1:2 000 000 to 1: 120 000 are appropriate for areas as
large as the Western Cape (see Section 2.1.1), the application of CLUES for suitability analysis
will be more valuable if it can be used for semi-detailed (i.e. 1:100 000 to 1:30 000)
investigations. Because South Africa has standardized on a scale of 1:50 000 for its largest-scale
topographical map series covering the Western Cape (CDSM 2007b), and because much relevant
data is mapped at this scale, it was decided that CLUES must have the ability to carry out
suitability analyses at a scale of 1:50 000. All the data collected for use in CLUES should
therefore preferably be at a scale of 1:50 000 or larger.
4.1.3.2 Climate data
Climatic data such as rainfall, temperature, humidity, and solar radiation are used in most of the
land evaluations consulted. Rainfall and temperature are considered to be the most important
climatic parameters (Ceballos- Silva & L?pez-Blanco 2003b; De la Rosa et al. 2004). Rainfall is
especially significant for suitability evaluations concerning vegetated land uses because water is
62
the most important factor in plant development. Water is essential for the maintenance of
physiological and chemical processes in plants and acts as an exchanger of energy and a carrier
of nutrients (Schulze 1997).
Rainfall is an essential factor for agricultural land uses, especially for determining the suitability
of land for the production of specific crops (Ceballos-Silva & L ? pez-Blanco 2003a). Rainfall
also influences the distribution and occurrence of natural vegetation types, which in turn relate to
the geography of ecosystems and fauna (Du Toit et al. 2002; Mucina & Rutherford 2006). Most
evaluations employ information about the total amount of rainfall per year as well as rainfall
averages per month.
Temperature is frequently used as an index of the energy status in the environment and it affects
all forms of life. Because of its influence on human comfort, temperature determines our demand
for energy and therefore our dependence on resources (Schulze 1997). Temperature also strongly
affects animal behaviour and responses, such as hibernation and migration (Walther et al. 2007),
and it has a significant effect on community distribution and size (Botes et al. 2006). The
occurrence of natural vegetation is strongly related to temperature as all plants have upper and
lower temperature limits above or below which their growth development processes cease. Crops
have different optimum temperature requirements for development processes such as
photosynthesis, respiration and flowering. For instance, the optim um average temperature range
for wheat is 25?C to 31?C, while sorghum need s temperatures between 31?C and 37?C (Schulze
1997).
Three additional temperature-based variables we re found to be popular and were considered
fundamental to land evaluation. The first, heat units (also called growing degree days), is a
heuristic index that is frequently used in phenology to predict crop maturity or bloom dates
(Schulze 1997). Consequently, heat units are often used to identify areas suitable for crop
production (Carey 2005). A second indicator often used in agricultural applications is chill units
(also called positive chill units or chill hours). Chill units are defined as the accumulative
number of hours that plants are exposed to temperatures ranging from 2.4?C to 9.1?C during
winter. Chill units is an important factor in land evaluation as most deciduous plants require a
minimum number of chill units to satisfy dormancy, to stimulate growth, develop leaves, flower
and set fruit (Reiger 2006; Schulze 1997). Because too low temperatures can be damaging to
plants, especially during the growing season, frost data was introduced as a third temperature-
based variable. Frost data is often used in la nd evaluations to identify areas that should be
avoided for the production of particularly perennial crops.
63
Frost, chill units, heat units, mean annual temperature, as well as monthly mean, minimum and
maximum temperatures were included in most of the suitability analyses consulted and were
consequently established as indispensable climatic data sets for CLUES.
Climatic data is obtained from long-term obser vations taken at weather stations and because
weather stations are relatively sparsely situated, especially so in mountainous regions or areas
with low population densities, climatic parameters at any given location are usually determined
by interpolating values from the nearest weather station or stations (Ceballos-Silva & L?pez-
Blanco 2003b). Owing to the effects of topogr aphy on climate (in particular orographic
precipitation and cooling), more realistic values are obtained when elevation is considered in
interpolation processes (Joubert 2007).
4.1.3.3 Soil characteristics
Soil information is especially important in land suitability analyses that are concerned with
vegetated land uses as most plants require soil for support (anchorage), water, oxygen and
nutrients. Most plants also prefer soils with specific characteristics. In addition, soil data is
significant for urban land uses, as sandy soils are more suitable as foundations for roads and
buildings than clayey soils (Brown 2003).
Soils are formed through the combined effect of physical, chemical, biological and
anthropogenic processes on the parent material. Parent materials consist of geological materials
that have undergone some degree of physical or chemical weathering. The result is the formation
of soil horizons or layers with distinctive co lour and texture properties. These properties,
together with the thickness and arrangement of soil horizons, are studied by soil scientists and
classified into soil types. During a typical soil survey, classifications are conducted at several
locations in the surveyed area and, together with terrain maps, used to produce a soil map.
Terrain has a strong influence on soil formation as it not only influences erosion and drainage,
but also determines where weathered materials are deposited (Schloms 2007, pers com; Soil
Survey Division Staff 1993; Van Niekerk & Schloms 2001).
Soil maps in their native form are of little use in automated land suitability analysis because soil
types need to be interpreted by a soil scientist for a particular land use. Usually this involves
creating groupings or ratings of soils according to their limitations, suitability and potentials for
particular land uses (Soil Survey Division Staff 1993). To avoid this additional step, quantitative
soil properties such as texture, depth and chemical characteristics are often extracted from soil
type data and used in suitability calculations (Ceballos-Silva & L?pez- Blanco 2003b; Cools, De
Pauw & Deckers 2002; De la Rosa et al. 2004).
64
Soil texture is one of the most important characteristics of soil as it directly relates to many other
land properties such as available water for plants, permeability, infiltration, plant nutrients,
erodability, tillage danger, and tillage strength requirements. While finer-textured soils are
generally more fertile, contain more organic matter and retain moisture and nutrients better, too
clayey soils are likely to be too difficult to manage and unstable between dry and wet periods.
Sandy soils are more stable, but need frequent fertilization and good water management (Brown
2003; Lambrechts & Ellis s.d.).
Soil texture describes the relative proportion of different grain sizes of mineral particles.
Particles are grouped into soil separates (sand, silt and clay) according to their size. Sand
particles are large (0.1mm to 2mm in diameter), while silt and clay particles are small (0.002mm
to 0.05mm and less than 0.002mm re spectively). The percentage of each soil separate in a soil is
used to classify soil texture into 12 major textural classes ( Figure 4-1).
Figure 4-1 Soil texture triangle showing the twel ve major textural classes and particle size scales
Although soil fertility can be partly ascribed to texture, soil reaction or pH is another factor
strongly affecting the nutrient availability in a soil. Soil pH is an indication of how alkaline or
acid a soil is and ranges from 0 (very acidic) to 14 (very alkaline) and mainly depends on the
type of parent material from which the soil was formed. Rainfall also affects pH as basic
nutrients such as calcium and magnesium are often leached and replaced by more acidic
elements such as aluminium and iron when water passes through soil. Soils in high-rainfall
regions are therefore usually more acidic than those formed under arid conditions. Soil pH is
important for land evaluations because many nutrients become soluble below a pH of 5 and they
Source: Soil Survey Division Staff (1993: s.p.)
65
are available to plants only at these levels of acidity. Some plants need these nutrients in order to
develop, while others have adapted to more alkaline soils. Too high ac idity or alkalinity may
become toxic (Soil Survey Division Staff 1993).
Another soil characteristic frequently used in land suitability analysis is effective soil depth or
rooting depth (Cools, De Pauw & Deckers 20 02; Dendgiz, Bayramin & Y?ksel 2003). This
refers to the depth, measured from the surface, at which root penetration would be strongly
inhibited due to physical characteristics such as contact with bedrock, dense clay or permanent
water, or due to contact with soils with extreme chemical properties (Soil Survey Division Staff
1993). Effective soil depth not only determines root growth, but also influences the water-
retaining capacity of soils. Deep soils usually have more available water and nutrients than
shallow soils, although the relative advantage of deep soils varies with climate, duration of
growth season and type of plant (Lambrechts & Ellis s.d.).
An important soil property to consider in suitability analysis involving vegetation is moisture
content as it is an indication of available water for plant use. Water is the major constituent of the
physiologically active tissues of plants and serves as a reagent in photosynthetic and hydrolytic
vegetation processes. It is also a solvent for salts, sugars and other solutes and is essential for the
maintenance of turgidity necessary for cell enlargement and growth (Mweso 2003). Water also
alters soil development and its chemical properties, induces periods of drought stress and
modifies temperatures that catalyze biotic processes. Owing to its strong relation to topography,
soil moisture can be estimated by using indices such as the topographical wetness index (TWI).
As a function of upslope area and local slope, TWI is relatively easy to generate from terrain
data. Although most types of soil data are valuable for land suitability analysis, soil texture, soil
pH, soil effective depth, and TWI were distingu ished as fundamental variables for CLUES.
4.1.3.4 Terrain types
Terrain type is used in land evaluation during the land unit mapping phase of the land evaluation
process (refer to Section 2.1). Due to its strong re lationship with soil and climate, terrain type is
often used in the absence of climatic and soil data or as an additional parameter in land
suitability assessments (Ceballos-Silva & L?pez-Blanco 2003b; Cools, De Pauw & Deckers
2002).
Terrain analysis is the study of the nature, origin, morphological history and composition of
landforms, the usual result being a landform map (Argialas 1995). Terrain analysis has wide
applications in pure sciences such as hydrology, botany, zoology, and ecology as well as applied
sciences such as agriculture, forestry, civil and military engineering, and landscape planning
66
(Mitchell et al. 1979). Because landform interp retation and mapping are time consuming, labour
intensive and costly operations, and seeing that the skills required are a product of lengthy,
expensive training and experience (Argialas 1995), landform maps may be substituted by basic
terrain derivatives such as slope gradient, aspect and curvature. These parameters can be
generated easily from a digital elevation model (DEM) using standard GIS operations (Van
Niekerk & Schloms 2001).
A DEM records elevations of the earth?s surf ace for each cell in a grid, hereby converting a
continuous data variable to a discrete representation (DeMers 2005). This simple model is
extremely versatile and highly efficient for computer analysis (Longley et al. 2002).
Slope gradient is defined as the angle between the surface tangent and the horizontal and it
controls the gravitational force available for geomorphic work (Van Niekerk & Schloms 2001).
Slope gradient is especially useful for evaluating urban and agricultural land uses as it imposes
limitations on construction and cultivation (Lambrechts & Ellis s.d.; Mitchell 1991). Most
governments, including South Africa?s, have laws that prevent agricultural and urban
developments on steep slopes (James 2001). Slope gradient is also used in environmental
modelling owing to the strong relationship between slope gradient and land cover (Adediran et
al. 2004; Hoersch, Braun & Schmidt 2002; Pickup & Chewings 1996).
Slope aspect is the direction in which a slope faces and therefore determines its exposure to
illumination from the sun. In the southern hemisphere, northern slopes receive more solar
radiation than southern slopes, especially during winter. Slope aspect, in combination with
gradient, determines the amount of solar radiation that reaches an area. It affects the temperature
of the soil, the rate of temperature change, vegetation composition, evapotranspiration and other
influences on soil properties (Irvin, Ventura & Slater 1997). Solar radiation is essential for plant
development due to its role in photosynthesis making it an important factor to consider in
agricultural and forested land uses. Solar radiation also affects the distribution and occurrence of
some animal species (Du Toit, Mouton & Van Niekerk 2006).
Curvature is the rate of change of slope gradient over a given distance and is an indication of
where surface runoff will accumulate or disperse. Because of the three-dimensional nature of
terrain, slopes can curve in infinite directions. For suitability anal ysis it is usually sufficient to
know whether an area is concave or convex along the slope direction (profile curvature) and/or
perpendicular to the slope (plan curvature) (Irvin, Ventura & Slat er 1997; Van Niekerk &
Schloms 2001).
Owing to the developmental limits imposed by slope gradient, the effect of aspect on plant
growth (Dendgiz, Bayramin & Y?ksel 2003) and the influence of curvature on hydrological and
67
soil formation processes, these two terrain derivatives were confirmed as essential data sets for
CLUES. Zhou & Liu (2004) showed that the accuracy of these derivatives is highly dependent
on the quality of the DEM from which they are generated. Care should therefore be taken in the
selection of an appropriate DEM.
4.1.3.5 Infrastructure attributes
The availability of existing infrastructure affect s the cost and potential to develop land. Many
types of infrastructure can be considered in land evaluations. These include roads, railways,
airports, electricity, dams, irrigation, and storage facilities. Roads are probably the most
important as they provide access to many of the other types of infrastructure. Infrastructure will
be incorporated in CLUES by using roads as a fundamental data set.
4.1.3.6 Current land cover and use
The suitability of a parcel of la nd for a particular land use is affected by its current land cover.
Built-up or urban areas, for instance, are less suitable for conservation than wetlands. The land
cover classes considered to be of fundamental importance for suitability analysis using CLUES
are urban areas, agriculture, wetlands, permanent rivers, and permanent water bodies. Nature
conservation areas (consisting of national parks and provincial reserves) were singled out as an
important land use to include as a principal data set.
4.1.4 Summary
This section aimed to document the functional and operational requirements of CLUES by
considering the content of the land evaluation procedure and existing systems. The literature
review established that CLUES should be fast enough to enable interactive scenario building,
and that a waiting time of less than one minute was appropriate for maps, while other functions
should be completed within seven seconds.
The web-based nature of the system poses challenges given that multiple users will access and
update the system simultaneously. The system should therefore be designed to allow multiple
users access to suitability analyses in such a way that they will be unaware of one another. In
addition, user-interface guidelines, as well as web-design principles, must be adhered to in order
to ensure that the system is as user-friendly as possible.
A range of land suitability analysis applications were studied to determine what data would be
needed to demonstrate the functionality of CLUES. It emerged that data relating to climate, soil,
terrain and infrastructure has to be included in the CLUES spatial database. Table 4-1
summarizes the specific data se ts identified to be fundamental for inclusion in CLUES.
68
Table 4-1 The data requirements of CLUES
TYPE REQUIRED DATA SETS
Climate Annual rainfall
Monthly rainfall
Mean annual temperature
Minimum temperatures (per month)
Maximum temperature (per month)
Mean temperature (per month)
Heat units
Chill units
Frost
Soil Soil effective depth
Soil texture
Soil pH
Topographical wetness index (TWI)
Terrain Elevation (DEM)
Slope gradient
Slope aspect
Slope curvature (plan and profile)
Infrastructure Roads
Current land
cover and use
Urban areas
Agriculture
Wetlands
Permanent rivers
Permanent water bodies
Nature conservation areas
Quality-wise, the data sets used in CLUES for suitability analysis must be as detailed and
accurate as possible and must cover the entire We stern Cape. Data should preferably be collected
at scales ranging between 1:100 000 and 1:30 000 to enable semi-detailed investigations
(Lambrechts & Ellis s.d.), but should be standardized, if possible, at 1:50 000 scale for
comparison purposes. This scale is fitting as it is consistent with South Africa?s largest scale
topographical map series.
Because data collection is not the focus of this research, only existing and available data was
considered for inclusion. An inventory and detailed descriptions of the existing climate, soil,
terrain, infrastructure, land cover and land use data for the Western Cape are provided in the next
chapter.
69
4.2 SYSTEM DESIGN
CLUES has been designed to demonstrate the pot ential of the Web as platform for spatial
decision support systems. Because web technology has not yet been optimized for spatial
analysis, several challenges were presented for system design. The major one was to
accommodate the simultaneous data creation and updating needs of spatial analysis in a multi-
user web environment. Where GIS software are aimed at individual users, web applications can
potentially be accessed by millions of users simultaneously. Special consideration must be given
to the effective sharing of computer resources and data. The sharing of data is especially
challenging in design because users must be allowed to concurrently edit and update the same
data. This is particularly important in land suitab ility analysis where generic spatial data layers
are examined and compared to produce new suitability information. Each suitability analysis
generates new data to be managed for each user. Because the establishment of suitability analysis
parameters (i.e. land use requirement rules and weights) can be a time consuming process, each
user?s settings must be managed individually to allow users to build their own rule sets which
they can modify and reuse for different projects. The management of data and user settings must
be done so that all users are unaware of one another.
To satisfy the requirements set out above, CLUES is designed as a web-based expert system. The
design involved combining the components of expert systems with those of web mapping
applications. Expert systems that perform suitability analysis usually consist of a land unit
database, a knowledge base and an inference engine (see Section 2.4), while web mapping
applications (see Section 3.4) normally comprise a spatial database, a WMS and a website.
Figure 4-2 shows that all of th ese components are incorporated into CLUES, with the land unit
database acting as the WMS spatial database.
Figure 4-2 The components of CLUES
INFERENCE ENGINE
SUITABILITY ANALYSIS
LAND UNIT
DATABASE
LAND UNITS
LAND PROPERTIES
SUITABILITY VALUES
KNOWLEDGE
BASE
LAND USES
LAND REQUIREMENTS
USER DETAILS
USERUSER USER
GUI
ASP
WMS
ArcIMS
WEBSITE
70
As in other expert systems, the two main components of CLUES are the land unit database and
the knowledge base. The third major component is the website, which consists of three main
elements namely, the graphical user interface (GUI), the web map service (WMS) and the
inference engine.
CLUES is designed so that suitability analyses can be done by following the expert system
procedure (see Figure 2-9). The function of the GUI is to dynamically generate a collection of
web pages through which users can store land requirement rules in the knowledge base and
invoke the inference engine to carry out a suitability analysis. During analysis, the inference
engine compares the land properties of each land unit in the land unit database with the land
requirement rules in the knowledge base. The suita bility values calculated by the inference
engine are temporarily stored in the land unit database for mapping purposes. To produce
suitability maps, the WMS extrac ts the necessary data from the land unit database and converts
the GIS data to dynamic web maps (see Section 3.3.1.1). The maps are temporarily stored on the
web server along with the other web page content generated by the GUI. This process is repeated
for each suitability analysis.
The suitability analysis procedure is a simplifie d rendition of the land evaluation procedure. The
major difference is that, due to security risk (see Section 3.4.1), CLUES cannot allow users to
upload their own spatial data sets. Instead, a collection of existing spatial data sets that are
essential to suitability analyses, is available to all users. Should users need specific data for their
analyses, it can be loaded by the system administrator.
The system design is based on the principle of loose coupling, a design goal employed in most
enterprise and web systems. L oose coupling ensures that a component can be changed or even
replaced without affecting the functionality of other components. Th is is deemed to be important
for CLUES because it is expected that the system will be modified and expanded in future. For
instance, the two databases (i.e. the land unit database and the knowledge base) are entirely
independent of one another. Therefore, the syst em can easily be applied to another region by
simply replacing the land unit database with one of another area. Another advantage of the loose
coupling design approach is that it lends flexibility to hardware configuration as changes to
hardware will not dramatically affect the individual components.
The implementation of each of the components shown in Figure 4-2 is detailed in the chapters to
follow. The first component developed was the land unit database as it contains the spatial data
on which the entire suitability analysis is based. To do this, the requisite land property data sets
had to be collected. A lack of appropriate data demanded some data manipulations and
preparations to be carried out as described in Chapter 5.
71
The next implementation activity was to delineate land units so that the land properties could be
imported into the database as attributes. Chapter 6 describes the different techniques considered
for mapping land units and provides an overview of how the fundamental land properties were
assigned to each land unit.
To store the land use requirement rules used to ra te the land units according to their suitability
for particular uses, the logical data modelling (LDM) procedure was used to design and
implement the knowledge base. In addition to the rules, the knowledge base must also
accommodate other operational information such as user details and land uses. Chapter 7
describes the LDM procedure for designi ng and implementing the knowledge base.
The rules in the knowledge base are used by the infe rence engine to carry out suitability analysis,
thus the inference engine acts as an interface between the knowledge base and the land unit
database. Users can view and edit the knowledge base through the GUI, but do not have direct
access to the land unit database. However, users can access the information in the land unit
database through the WMS, which also acts as a type of user interface. In effect, all the
components of the website function as a combined interface between the users and the two
databases. Each of the elements of the website is described in more detail in Chapter 8.
The suitability maps generated by the infe rence engine and the WMS are based on the
environmental and physical properties contained in the land unit database and the land use
requirements in the knowledge base. The next ch apter describes the activities to collect and
manipulate the appropriate land property data for inclusion in CLUES.
72
CHAPTER 5: LAND PROPERTY DATA COLLECTION
Data collection, the second step in the land evaluation process, involves the capturing, gathering
and preparation of the data for use in the suitability analysis. Although there are infinite variables
that can be used in suitability assessments (De la Rosa et al. 2004), a number of fundamental
data sets relating to climate, soil and terrain were identified during the requirement analysis
phase (Section 4.1.3). This chapter describes the activitie s to acquire the data. The chapter gives
an inventory of existing data as well as the motives for selecting specific data sets. The
manipulations to prepare the data sets for analysis are also described. The next three sections
give details of the selection, collection and manipulation of terrain, soils and climate data.
5.1 TERRAIN DATA
Terrain attributes is one of the most useful types of data for land evaluation. Elevation, slope
gradient, slope aspect, plan curvature and profile curvature were identified by the requirement
analysis to be fundamental data sets for land evaluation. Because all this information can be
easily derived from a digital elevation model (DEM), the focus of terrain data collection was to
find an appropriate DEM for the Western Cape regi on. The next three sections concentrate on the
criteria for selecting a suitable DEM, inventorying existing models and assessing the accuracy of
the chosen DEM respectively.
5.1.1 DEM selection criteria
Digital elevation models are essentially elevation rasters generated by interpolating the elevation
of a given raster cell from nearby cells with known elevations. The known elevations are
typically digitized from topographi cal maps, but they can also be surveyed elevations (including
GPS measurements) obtained using photogrammetr y or by processing RADAR (radio detecting
and ranging) or LIDAR (light detecting and ranging) data (Campbell 2006; DeMers 2005).
The requirement analysis determined that data only at a map scale of 1:50 000 (or larger) will be
considered for inclusion in CLUES. Howeve r, because DEM are sometimes derived from
primary data sources (such as RADAR and LIDAR ), the map scales of DEM are not always
known. Map scale is therefore an unsuitable measure for selecting an appropriate DEM for
inclusion in CLUES. This also applies to other ra ster data sets such as climate data (see Section
5.3) derived from primary data sources. In order to select appropriate primary data, raster cells
should be smaller than the minimum mapping unit at a scale of 1:50 000. This is specified by
McDonald et al (1984) to be 150x150 metres.
73
Resolution should not be regarded as the only measure of DEM quality. According to
Thompson, Bell & Butler (2001), the quality of a DEM is influenced by four factors: (1) the
interpolation method and algorithm; (2) source da ta; (3) resolution; and (4) terrain roughness and
complexity.
When selecting an interpolation algorithm, the intended use of the DEM and the nature of the
source data must be considered. For instance, if the DEM is to be used for small-scale-mapping,
a simple inverted linear distance interpolator will suffice, but if the DEM is to be used for
hydrological modelling at a local catchment level, a more complex interpolator is required.
Different interpolators are often used for different types of input (source) data due to differences
in the density and location of the known elevations. However, some more advanced DEM
generation software include several interpolators and can accommodate various source data
types, including contours, elevation points, river lines and water bodies as input. Special care
must be taken when contours are used as source data because artefacts such as ?rice terraces? and
?tiger stripes? may be created, especially when inferior interpolators are used (Burrough &
McDonnel 1998). The source data also directly influences the quality of the resulting DEM as
inaccurate input data will result in an inaccurate DEM ? the rule of ?garbage in, garbage out?
applies ? and minor errors in the source data will be propagated and cause very noticeable
artefacts such as spurious sinks and peaks in the DEM (Hengl, Grube r & Shrestha 2004).
Because resolution can be easily manipulated or changed, it is not always a good measure of the
detail contained in a DEM because a high-resolu tion DEM will not necessarily be more accurate
than one at lower resolution derived from the same sample points (Zhou & Liu 2004). Resolution
must be considered in combination with the other three factors (i.e. algorithm, source data, and
terrain roughness) when describing a DEM?s quality. Since only one elevation value is stored per
raster cell, the resolution (i.e. width and height of raster cells) of a DEM has a noticeable
influence on the accuracy of a DEM and its derived products (i.e. slope, aspect, curvature).
Because each cell occupies a specific area, a reduction in the cell area (i.e. an increase in
resolution) can potentially represent surfaces more accurately. This is especially true for terrain
with a high degree of topographical complexity as in mountainous areas. One should, however,
be aware that a 100% increase in resolution will result in a 400% increase in storage volume as
the number of cells will increase fourfold.
Due to the lack of a single quantitative measure of quality, absolute vertical accuracy is
frequently used to compare DEM. Vertical accuracy is determined by statistically comparing
DEM values with known elevations, usually obtained through highly accurate surveying
techniques. Another factor that should be taken in consideration when selecting a DEM is the
74
accuracy of derived products such as slope gradient and curvature. Thompson, Bell & Butler
(2001) have shown that a low-resolution DEM produces smoother, less-detailed slopes while
smaller variations in slopes can be observed when slope gradients are derived from a high-
resolution DEM. Because slope gradient and curvature are essential for land suitability analysis,
vertical accuracy and resolution are both considered when selecting a DEM for CLUES.
5.1.2 Existing Western Cape DEM
Three existing DEM were identified that cover the Western Cape. The first is the official South
African DEM produced by the Chief Directorate Surveys and Mapping (CDSM) using
photogrammetry techniques. The CDSM DEM is not a single DEM, but a combination of three
separate DEM with resolutions of 50, 200 and 400 metres respectiv ely. Each of these DEM has a
limited coverage that, in combination, covers the entire Western Cape. The 50-metre DEM is
available for urban areas only, while the 200-metre DEM only re presents mountainous areas.
The rest of the Western Cape is covered by the 400-metre DEM. As for quality, CDSM has
estimated the vertical accuracy of its product to be 10 metres or better (CDSM 2007a).
Unfortunately, due to its variable resolution, the CDSM DEM was disqualified for inclusion in
CLUES as the required resolution of at least 150 metres is not met for most of the Western Cape.
In addition, the transitions between the three constituent DEM would cause artefacts and
unrealistic values if used to derive products such as slope gradient and slope aspect (Thompson,
Bell & Butler 2001).
The second DEM available for the Western Ca pe is the SRTM DEM, developed in 2000 by
NASA during the Shuttle Radar Topography Mission (SRTM) (NAS A 2005). In contrast to the
variable resolution of the CDSM DEM, the elevation values in the SRTM DEM are regularly
spaced at 90-metre intervals, which is considerab ly better than the minimum requirement of 150
metres. The DEM is also reported to have a verti cal accuracy of less than nine metres (Rodriguez
et al. 2005).
The third DEM that covers the entire Western Cape is the Western Cape Digital Elevation Model
(WCDEM), developed by the Centre for Geogr aphical Analysis (CGA) at Stellenbosch
University (Van Niekerk 2001). This 20-metre resolution DEM was generated from contours
digitized from the 1:50 000 national topographical map series using ANUDEM software.
Although the resolution of the WCDEM is considerably higher than any of the other available
DEM, no other information about its accuracy was available. To ensure th at the most accurate
DEM is included in CLUES, an independent accuracy assessment was conducted on both the
WCDEM and the SRTM DEM using the same refe rence data and methods for both products.
75
5.1.3 DEM accuracy assessment
To determine the accuracy of the WCDEM a nd the SRTM DEM, elevation values were
systematically compared to reference elevations. Highly accurate (sub-metre) elevation points,
obtained from CDSM, were used as reference data. This data was not used in the generation of
the WCDEM or SRTM DEM and is therefore a suitable data set to use for the accuracy
assessment. To restrict the reference data to a manageable size, a 5% sample was extracted from
the database of reference points. Due to the quarter-degree structure of the database, reference
points were selected as blocks of 15x15 arc minutes. Of the 262 blocks covering the Western
Cape, 13 (i.e. 5%) were selected for the accuracy a ssessment. To ensure that the selected blocks
are representative, a stratified sample was drawn by randomly selecting a proportional number of
blocks in each stratum. Regions of one-degree wi dth were used as horizont al strata, resulting in
the selection of 2620 reference points. The loca tion of the reference points and the 15x15 arc-
minute blocks are shown in Figure 5-1.
Figure 5-1 Selection of reference poi nts used in the DEM accuracy assessment
76
To determine vertical accuracy, the mean absolute error (MAE) and root mean square error
(RMSE) for each DEM were calculated using Equations 5-1 and 5-2 respectively (Bolstad &
Smith). The results of the accuracy assessments are summarized in Table 5-1.
n
xx
MAE
ji? ?= Equation 5-1
where MAE is the mean absolute error;
x i is the DEM?s elevation value;
x j is the reference point?s elevation value; and
n is the number of reference points.
( )
n
xx
RMSE ji? ?= 2 Equation 5-2
where RMSE is the root mean square error;
x i is the DEM?s elevation value;
x j is the reference point?s elevation value; and
n is the number of reference points.
Table 5-1 Vertical error in the WCDEM and the SRTM DEM
DEM RESOLUTION (m) MAE (m) RMSE (m)
SRTM DEM 90 15 24
WCDEM 20 7 10
The WCDEM performed significantl y better than the SRTM DEM in terms of MAE and RMSE,
both indicators pointing to the WCDEM as th e more accurate DEM. Consequently, the WCDEM
was selected for CLUES.
Although elevation is a vital land property for suitability analysis, its derivatives are more
frequently used in land evaluation. Due to the relatively moderate terrain of the Western Cape
(i.e. average slope gradient of 6.5?), the Ho rn algorithm (Chang 2006) av ailable in ArcGIS was
used to create slope gradient, slope direction (aspect) and curvature (plan and profile) rasters
from the WCDEM.
A data set closely related to terrain and often used in suitability analysis is soil. The next section
explores the Western Ca pe soil data sources.
77
5.2 WESTERN CAPE SOIL INFORMATION
The importance of soil data for land evaluation was discussed in Section 4.1.3.3. The
requirement analysis established that soil type data has little value for suitability analysis and
that generic soil properties such as effective soil depth and texture are more useful owing to their
quantitative nature. The following se ctions overview the soil data available for the Western Cape
and also describe the preparation of the data for use in CLUES.
5.2.1 Existing soil data
According to Lambrechts & Ellis (s.d.) soil su rveys can be categorized according to map scales
into detailed (1:1 000 to 1:2 500), semi-deta iled (1:10 000 to 1:100 000), reconnaissance (1:120
000 to 1:500 000) and investigativ e (<1:500 000) surveys. Invest igative surveys were excluded
from the soil inventory as they are usually conducted rapidly along specific routes and are
generally unsuitable for land evaluation purposes (Schloms 2007, pers com).
5.2.1.1 Detailed surveys
Detailed soil surveys are expensive to carry out because they involve the collection of extensive
field samples and thorough laboratory analysis (McSweeney et al. 1994). Due to the high costs,
soil surveys at 1:1 000 to 1:2 500 map scales ar e usually only conducted for agricultural areas.
This is true for the Western Ca pe where detailed soil maps generally only exist for areas of
intensive agriculture (i.e. vineyards and orchards). As most of these surveys are funded by
private landowners for farm management and planning purposes, the data is not stored centrally
and is therefore not easily accessible (Schloms 2007, pers com).
5.2.1.2 Semi-detailed surveys
Only a limited amount and coverage of semi-detaile d soil data is available for the Western Cape.
Apart from a peri-urban survey conducted in great er Cape Town at a scale of 1:10 000, a number
of 1:50 000 scale surveys have been carried out fo r the major agricultural areas in the Breede,
Berg, Doring and Olifants River catchments. The so ils of the rest of the Western Cape remains
uncharted at this level of detail (Schloms 2007, pers com).
5.2.1.3 Reconnaissance surveys
Although a number of reconnaissance soil surveys have been conducted for selected areas in the
Western Cape (such as the Karoo), a land type surv ey of South Africa is the only one that covers
the entire province. Although this survey was originally conducted at 1:50 000 scale, it was
published at a scale of 1:250 000 and cannot be regarded as a semi-detailed survey.
78
The land type survey was initiated in 1971 by the Institute of Soil, Climate and Water (ISCW) to
provide an inventory of soils, terrain forms and macro-climate for South Africa. It was
completed in 2002. Funded by the National Depart ment of Agriculture (NDA), the survey was
based on factual observations, soil analysis and long-term climatic records so that it can be
reliably applied to determine land use potential, land sustainability, and best management
practices (Patterson 2005; Turner 2005).
5.2.1.4 Soil database strategy
The soil inventory exercise established that detail ed soil data of the West ern Cape is lacking and
that data at the required scale of 1:50 000 is available for certain areas only. Because CLUES
requires a continuous layer of soil data covering the entire province, the land type data was
chosen as the fundamental soil data source in spite of its generalized nature. This data will be
substituted by more detailed information when it becomes available. This strategy means that
differently scaled soil data will be used for different areas ? a less than ideal option. However, it
does ensure that a suitability analysis conducted anywhere in the Western Cape is based on the
best available data.
The next section overviews the land type data and documents how it was manipulated to extract
soil information in a suitable format for land evaluation purposes.
5.2.2 Land types information system
The ISCW land type information is published as memoirs. An example is the memoir for land
type Ca6 as shown in Figure 5-2. This specific land type covers an area of 421 200 hectares and
includes four terrain units. Soil information is related to each terrain unit as illustrated by the
terrain form sketch. Each terrain unit is further described in terms of its area, slope, slope length,
slope shape, mechanical limitations and soil series. By interpreting this information, a soil
scientist can easily gain a synopsis of the soils in each land type. The land type data was never
intended for computer analysis and is therefore unsuited for making quantitative comparisons
between land types using GIS. There is, for instance, no easy way to find all the land types
having soils deeper than one metre. Consequently, innovative manipulations are needed to
quantify the land type data. To do so, knowledge of the data structure of the recently digitized
data set was required.
79
Figure 5-2 Land type Ca6 in memoir format
Source: Land Type Survey Staff (1984)
80
5.2.2.1 Digital data structure
The digital land type data consists of spatia l and tabular components. The spatial component
encompasses the location and boundaries of each land type, stored as polygons (see Figure 5-3a)
in shape file format. The spatial features are linked to an attribute table (Figure 5-3b) which, in
turn, can be linked to six informational tables, namely Tables A, B, D, E, F, and G shown in
Figure 5-4.
Figure 5-3 Land type polygon (a) and its associated attribute information (b)
Figure 5-4 Land type database structure
As with any shape file, the SHAPE field in the attribute table (Figure 5-3b) contains objects that
define the shape and position of each polygon (ESRI 2002b). Each land type consists of one or
more polygons and can be uniquely identified through the land type code stored in the
LANDTYPE field. The OPP_HA fiel d contains the area in hectares of each polygon while the
ATTRIBUTES
SHAPE
LANDTYPE
BRDSOIL
AFRDESCR
ENGDESCR
AFRCLASS
ENGCLASS
OPP_HA
TABLE A
LANDTYPE
CLIMATENO
TERRAIN_T
AREA
UNAREA
INVENTORY
TABLE B
LANDTYPE
MAPNO
MAPAREA
TABLE D
LANDTYPE
TERRAIN_U
TERRAIN_P
SLOPE_L
SLOPE_T
SLOPE_U
SLOPE_LL
SLOPE_TL
SLOPE_UL
SLOPE_SHP
TABLE E
LANDTYPE
PROFILE
TABLE F
LANDTYPE
COMPLEX
SERIES
SOIL_D_L
SOIL_D_T
SOIL_D_U
CLAY_A_L
CLAY_A_T
CLAY_A_U
CLAY_E_L
CLAY_E_T
CLAY_E_U
CLAY_B_L
CLAY_B_T
CLAY_B_U
DEPTH_L
MB
HORIZON
SKAKEL
1
m
1
m
m
m
1 1
m
m
m
1
TABLE G
LANDTYPE
TERRAIN_U
SOIL_P
SKAKEL
SHAPE Polygon
LANDTYPE Lc126
OPP_HA 12577.067
BRDSOIL Lc
AFRDESCR Rots met min of geen grond
ENGDESCR Rock with little or no soil
AFRCLASS Diverse landklasse
ENGCLASS Miscellaneous land classes
(a) (b)
81
remaining fields BRDSOIL, LAFRDESCR, ENGDESCR, AFRCLASS, and ENGCLASS
provide general descriptive information bilingually.
The LANDTYPE field in the attribut e table relates to Table A in Figure 5-4 containing
information about the climate (CLIMATENO), general terrain type (TERRAIN_T), area
available for agriculture (AREA) and area unavailable for agriculture (UNAREA), as well as the
name of the surveyor who inventoried (INVENT ORY) each land type. Table A is related one-to-
many with the attribute table; in other words, every row in Table A is related to many rows
(polygons) in the attribute table through its primary key, LANDTYPE.
Table B stores information about the original land type map sheet (MAPNO) from which each
land type was digitized. As a land type could have been digitized from one or more sheets, this
table has a many-to-one relationship with Ta ble A. When combined, the LANDTYPE and
MAPNO fields act as a primary key. The table al so stores the area each land type covers on the
original sheet or sheets in MAPAREA.
Similarly, Table E relates many-to-one to Table A. In Table E the modal profiles (PROFILE) of
each land type are stored, providing a general indication of the type of soil occurring in a land
type.
Information about a land type?s te rrain units is stored in Table D. This table has a many-to-many
relationship with the attribute table as multiple polygons representing any land unit can relate to
multiple terrain units (TERRAIN_U field). The LANDTYPE field in Table D is therefore not
unique. Rows in Table D can, however, be uni quely identified by combining the LANDTYPE
and TERRAIN_U fields. The percentage area that each terrain unit covers is stored in the
TERRAIN_P field. The slope grad ient (%) is stored as a range in fields SLOPE_L, SLOPE_T
and SLOPE_U, where the suffix L indicates lower limit, T the operand (<, -, >) and U the upper
limit. Similarly slope length (in metres) is stored in fields SLOPE_LL, SLOPE_TL and
SLOPE_UL. The SLOPE_SHP field i ndicates whether the shape of the slope is convex, straight,
or concave.
Tables G and F store informa tion about each soil series found in each terrain unit
(TERRAIN_U). Table G can be linked many-to -one to Table D using the LANDTYPE and
TERRAIN_U fields combined and Table F is re lated many-to-one by the SKAKEL field. Table
F contains details of each soil series, including effective soil depth, clay content per horizon and
mechanical limitations. Similar to slope in Table D, effective soil depths (mm) are stored as
ranges in fields SOIL_D_L, SOIL_D_T a nd SOIL_D_U, while CLAY_A_L, CLAY_A_T and
CLAY_A_U store clay content (%) in the A horizon in Table F. The clay content of horizons E
and B21 is provided in the same manne r in CLAY_E_L, CLAY_E_T, CLAY_E_U,
82
CLAY_B_L, CLAY_B_T, and CLAY_B _U. The clay content of hor izons E and B21 is absent
for many land types.
In addition, mechanical limitations are stored in the MB field as integers ranging from 0 to 4
indicating no mechanical limitations (0); many stones, but ploughable (1); large stones and/or
boulders, unploughable (2); very shallow soils on rock (3); and no soil (4).
Figure 5-5 shows how this structur e can be conceptualized in levels of detail. Each land type
(LT) can have one or more terrain units (TU) in level 2 and each terrain unit can have one or
more soil series (SS) in level 3. A terrain unit cannot be shared between two land types, but a soil
series (SSj ) can occur on two different terrain units (TU i and TU j ) if they occur in the same land
type (LT i). Only level 1 has a spatial component. Terra in and soil information must therefore be
extracted from the tabular data and applied at the land type level in order for the data to be
suitable for map production or analysis. The ex traction procedure is explained in the next
section.
Figure 5-5 Conceptual view of land type object levels
5.2.2.2 Soil property extraction
The land type data contains soil property informati on in the form of effective depth, clay content,
and mechanical limitations. Although the requirement analysis highlighted soil texture as a
fundamental variable in land evaluation, no information about texture is available in the land
type data. However, clay content can be used as a surrogate for texture as texture tends to
decrease with increasing levels of clay.
Because it is not possible to directly link effective depth, clay content, and mechanical
limitations to the land type polygons, the land type data is not suitable for mapping or spatial
analysis. This section provides an overview of how the various soil properties were extracted
from the land type data for quantitative land suitability analysis.
LTa
TUa TUb TUc
SSa SSb SSc SSd SSf
LEVEL 1
Land Types (LT)
LEVEL 2
Terrain units (TU)
LEVEL 3
Soil series (SS)
SSe
83
Effective soil depth
Effective soil depth is stored at the soil series level (see Figure 5-5), so to use this information
for mapping or spatial analysis, the effective depths of all the soil series within a land type must
be aggregated. This is achieved by using simple averaging techniques at all three data levels. The
first step converts each effective soil depth range in Table F to its central value using Equation
5-3. The result is a value representing the aver age effective soil depth of each soil series.
2
minmax
ss
s
DD
D
?= Equation 5-3
where Ds is the average effective soil depth of soil series s ;
maxsD is the upper value of the effective depth range for soil series s; and
minsD is the lower value of the effective depth range for soil series s.
Next, the average effective soil depth of each terrain unit is calculated. In Equation 5-4 the
individual Ds values are summed using the percentage area (SOIL_P) of each soil series as
weights.
?
=
??
???
? ?=
n
s
s
st
P
DD
1 100
Equation 5-4
where Dt is the average effective soil depth of terrain unit t;
Ds is the average effective soil depth of soil series s;
Ps is the percentage area covered by soil series s within terrain unit t; and
n is the number of soil series within terrain unit t.
To calculate the average effective soil depth of the entire land type, Equation 5-5 is applied.
Similar to Equation 5-4, the relative size of each terrain unit is considered in the summation.
?
=
??
???
? ?=
m
t
t
tl
P
DD
1 100
Equation 5-5
where Dl is the average effective soil depth of land type l ;
Dt is the average effective soil depth of terrain unit t;
Pt is the percentage area covered by terrain unit t within land type l ; and
m is the number of terrain units within land type l.
By substituting Equation 5-3 and Equation 5-4 into Equation 5-5 it follows that:
? ?
= = ?
?
?
?
???
? ????
?
???
? ??=
m
t
t
n
s
sss
l
PPDD
D
1 1
minmax
1001002
Equation 5-6
Equation 5-6 could only be used for soil series with effective soil depth values in both the
SOIL_D_L and SOIL_D_U fields in Table F. Some records only contain values in the
84
SOIL_D_U field with a ?>? or ?? occurred in the SOIL_D_T field. These exceptions were
handled programmatically.
Only effective soil depths of 1200mm or less are specified in the land type data. Deeper soils are
indicated as being deeper than 1200mm. Because th e extraction process considered such soils as
being only 1200mm deep, the aver aging effect of the procedure underestimates true depths.
Values of more than 1000mm should therefore be considered to be ?1000 or deeper?.
The result of the soil propert y extraction procedure is a single value representing average
effective soil depth for each land type. In this format the effective depth information is more
suitable for spatial analysis. It can also be used to produce a choropleth map (see Figure 5-6) of
effective soil depth in the Western Cape. The ma p confirms the notion that deeper soils occur on
valley bottoms and plains, while shallow soils are more frequently found in areas with high
relief.
Figure 5-6 Effective soil depth derived from land type data
85
Average clay content
To calculate the average clay content of each land type, only the A horizon was considered as B
and E horizons are not always present. Moreover , duplex soils may cause misleading results. As
seen in Equation 5-7, the procedure to calculate the average clay content of horizon A is almost
identical to that for effective depth (Equation 5-6). The spatial ex traction result is shown in
Figure 5-7.
? ?
= = ?
?
?
?
???
? ????
?
???
? ??=
m
t
t
n
s
sss
l
PPCC
C
1 1
minmax
1001002
Equation 5-7
where C l is the average clay content of horizon A;
maxsC is the upper value of the A horizon cl ay content range for soil series s ;
minsC is the lower value of the A horizon cl ay content range for soil series s ;
Ps is the percentage area covered by soil series s within terrain unit t;
n is the number of soil series within terrain unit t;
Pt is the percentage area covered by terrain unit t within land type l ; and
m is the number of terrain units within land type l.
Figure 5-7 Soil clay content derived from land type data
86
Mechanical limitations
Equation 5-8 is used to cal culate an average mechanical limitations value (M l) for each land type
and the spatial result is displayed in Figure 5-8. The only difference between this procedure and
those followed for effective soil depth and soil clay content is that the mechanical limitations of
each series (M s) are stored as integers (0-4) and not as ranges so that no conversion is necessary.
? ?
= =
???
?
???
? ???
???
? ?=
m
t
t
n
s
ss
l
PPM
M
1 1 1001002
Equation 5-8
where M s is the mechanical limitations value for soil series s ;
Ps is the percentage area covered by soil series s within terrain unit t;
n is the number of soil series within terrain unit t;
Pt is the percentage area covered by terrain unit t within land type l ; and
m is the number of terrain units within land type l.
Figure 5-8 Soil mechanical limitations derived from the land type data
To manually carry out the calculations necessary to extract the soil properties discussed above
for all the land types is a time-consuming, if not impossible, task. Although GIS can speed up the
process, hundreds of GIS operations would be re quired to retrieve and combine the necessary
87
data from the six tables in Figure 5-4. This would not only be a time-consuming task, but is
prone to human error. An alternative ? au tomating the entire procedure using programming
techniques ? was implemented. An Avenue scri pt (see Appendix A) was written and executed in
ArcView GIS 3.3. The automation not only saves time, but enables easy repetition. It can also be
used to extract soil property information from the data base for the rest of South Africa.
Although the land type survey includes general climatic indicators, an inventory of other
available climate data sources for the Western Cape was made.
5.3 WESTERN CAPE CLIMATE INFORMATION
According to Lutgens & Tarbuck (1998), an area ?s climate is an aggregate of its weather
conditions over time. To derive reliable climate data, statistical analysis is performed on long-
term (i.e. 30 years or more) weather observa tions (Houghton et al. 2001). In South Africa,
weather data is recorded by the South African Weather Serv ices (SAWS) through a national
network of 118 automatic weather stations, 1 12 climate stations and 1512 rainfall stations.
Unfortunately, many of the weather stations are sparsely situated, especially in mountainous
regions or areas with low population densities, resulting in vast regions being insufficiently
represented by weather stations (SAWS 2007). Inte rpolation methods are employed to estimate
climate data for areas not represented by weather stations. The accuracy of such estimations is a
function of the distance between weather stations and the interpolation method employed. Based
on the distance averaging effect, a higher density of weather stations should result in greater
accuracy. Accuracy is also influenced by the interpolation algorithm used (Lynch 1999). These
factors were considered during the selection and capture of appropriate climate data for inclusion
in CLUES.
5.3.1 Existing climate data
The most commonly used climatic data for South Africa was developed by Schulze (1997) for
the South African atlas of agr ohydrology and -climatology (SAAAC). The SAAAC data was
derived from weather station observations at a resolution of one arc minute (?2km) using
stepwise multiple regression techniques. Latitude , longitude, altitude, and aspect were included
to produce a statistical goodness-of-fit model for each quaternary catchment in South Africa. The
modelled values were combined to produce more than 200 climatic and agrohydrological data
sets, including mean annual precipitation and median monthly rainfall, as well as means of daily
maximum and minimum temperature, on a national scale (Schulze 1997).
88
A second global set of climatic data was developed for WorldClim by Hijmans et al. (2005) at a
resolution of 30 arc seconds ( ? 1km). The WorldClim grids were interpolated from weather
station observations using ANUSPLIN software, th e variations between weather stations being
modelled using latitude, longitude and elevation as independent variables. The authors showed
that there is significant benefit in the higher spatial resolution of their climate grids. This is
mainly because a higher spatial resolution can accommodate more variation in terms of terrain.
The influence of resolution on terrain variability led to the development of a set of high
resolution climatic grids for the Western Ca pe (Joubert & Van Niek erk 2005). Long-term (35
years on average) weather data was collected for 125 weather stations in and around the province
and used as the data source for the interpolation of the Western Cape climate grids (WCCG).
Combinations of elevation, latitude, longitude, hillshade and distance to oceans were used as
covariates and independent variables in the ANUSPLIN interpolation algorithm. To model as
much climatic variation as possible while keeping data sets manageably small, the SRTM DEM
(see Section 5.1.2) was used as the data source for elevation and hillshade. Climate grids at a
resolution of 90 metres were interpolated fo r monthly mean daily maximum temperature,
monthly mean daily minimum temperature and mean monthly rainfall (Joubert 2007).
5.3.2 Comparison of data quality
According to the requirement analysis, climate data is needed at a 1:50 000 map scale. Because
climate grids are interpolated from weather station point data, map scale is not an appropriate
measure of suitability. Instead, a comparison of the existing grids regarding their spatial
resolution and accuracy was made.
The reported accuracies of the ex isting data are summarized in Table 5-2. Other than the slight
difference in resolution between the SAAAC and Wo rldClim data sets, no difference in accuracy
was observed between these two data sources. The WCCG data set is sign ificantly more accurate
than the other two data sets ? three times more accurate in terms of temperature and more than
twice as accurate in terms of rainfall.
Table 5-2 Accuracy summary of existin g climatic data for the Western Cape
SOURCE RESOLUTION (m) TEMPERATURE (?C) RAINFALL (mm)
SAAAC 1600 1.0 10.0
WorldClim 1000 1.0 10.0
WCCG 90 0.3 4.5
Sources: Hijmans et al. (2007) and Joubert (2007)
89
The markedly higher accuracy of th e WCCG data is attributable to the inclusion of more source
data (weather stations), the use of more-detailed elevation data, and the incorporation of more
independent variables into the interpolation algorithm. In addition, with its resolution of 90
metres, the WCCG data set is more than ten time s finer than the other two sets. This higher level
of detail is especially important for suitability analysis of land uses such as perennial crops that
are strongly influenced by local (micro) climate. Understandably, the WCCG data set was
chosen as the primary climate data source for CLUES. In addition to the WCCG data, other
climate data sets namely heat units, chill units and frost were included. These data were
considered essential for agrohydrological applications of CLUES.
5.4 SUMMARY
This chapter described the collection of suitable terrain, soil and climate data needed to
demonstrate the functionality of CLUES. A sele ction was made from existing data sets by
comparing data quality and scale. Concerning terrain, the WCDEM is presently the most
accurate DEM available for the Western Cape, while the WCCG climate data is notably superior
in comparison to other available climate data. A summary of the data sets to be used in CLUES
is given in Table 5-3.
Table 5-3 Data sets collected for CLUES
TYPE DATA SET SOURCE RESOLUTION/SCALE
Climate Annual rainfall
Monthly rainfall
Mean annual temperature
Minimum temperature (per month)
Maximum temperature (per month)
Mean temperature (per month)
Heat units
Chill units
Frost
Joubert (2007)
Joubert (2007)
Joubert (2007)
Joubert (2007)
Joubert (2007)
Joubert (2007)
Schulze (1997)
Schulze (1997)
Schulze (1997)
90m
90m
90m
90m
90m
90m
?2km
?2km
?2km
Soil Effective soil depth
Soil texture
Topographical wetness index (TWI)
Land Type Survey Staff (1984)
Land Type Survey Staff (1984)
Van Niekerk (2001)
1:250 000
1:250 000
20m
Terrain Elevation (DEM)
Slope gradient
Slope aspect
Slope curvature (plan and profile)
Van Niekerk (2001)
Van Niekerk (2001)
Van Niekerk (2001)
Van Niekerk (2001)
20m
20m
20m
20m
Infrastructure Roads CDSM (2007b) 1:50 000
Current land
cover and use
Urban areas
Wetlands
Permanent rivers
Permanent waterbodies
Nature conservation areas
Thompson (1999)
Thompson (1999)
Thompson (1999)
Thompson (1999)
Thompson (1999)
1:250 000
1:250 000
1:250 000
1:250 000
1:250 000
90
There is a paucity of soil data for the Western Ca pe, land type data being the only data set that
meets the required coverage. However, because the land type data is published at a scale of 1:250
000, it does not satisfy the 1:50 000 scale requiremen ts. In the absence of better data, land type
information is included as a fundamental soil data layer, with the recommendation that the
generic soil properties (i.e. effective soil depth, texture and pH) be updated with more detailed
soil data as it becomes available.
The data sets listed in Table 5-3 were collected to be used as land property information. Based
on the land evaluation approach, land properties are related to land use requirements to
determine whether a land unit is suitable for a particular use. As explained in Section 4.2, land
properties are stored as attributes of land units in the land unit database. The activities to develop
the land unit database are described in the next chapter.
91
CHAPTER 6: DEVELOPING THE LAND UNIT DATABASE
The first step to implement CLUES was to set up a land unit database as a repository of all the
spatial data required for suitability analysis using CLUES. The database essentially consists of
polygons (land units) with a set of attributes (land properties).
Land units were defined in Section 2.1.5 as parcels of land that differ significantly from the
surrounding land. Although any parcel of land can be considered a land unit, it is more efficient
and meaningful to use parcels that can be adequately described by one or a combination of land
properties. A land unit should represent an area that, according to predetermined properties, is
different from the surrounding land and can be assumed to be homogeneous in terms of its land
properties. The degree of homogeneity or intern al variation will vary depending on the scale and
intensity set out in the evaluation objectives (FAO 1976).
Although the size of the la nd units should be kept as small as possible to limit generalization, too
many units can become unmanageable as each individual land unit is considered individually as
to its land properties and requirements. The decisions about the size, number and delineation of
land units are often determined by data availability. Soil type boundaries are probably the most
suitable delineations of land units for most land uses, but soil information is often unavailable at
the required scales. In these cases other available datasets, such as terrain units (i.e. land
components), can be used. Land components are sometimes used as land units in medium-scale
studies (1:25 000 to 1:500 000) because many physic al land properties, including soil, climate
and biology, are related to terrain (MacMillan, Jones & McNabb 2004; Speight 1977). Due to the
unavailability of semi-detailed soil da ta in the Western Cape (see Section 5.2.1), land
components were used as basic mapping units (land units) in this study.
This chapter outlines the procedures to establish a land unit database. In the first part of the
chapter the techniques for delineating land components, in particular the ALCoM algorithm, are
explained. The rest of the chapter focuse s on populating the database with land property
information.
6.1 LAND COMPONENT MAPPING TECHNIQUES
Land components are essentially subdivisions of la ndscapes and are frequently used in suitability
analysis as a basic mapping unit (i.e. land unit). Although ?landscape? has been variously
defined, it can be conceptualized in the terrain analysis context as a hierarchical collection of
terrain forms comprising land regions, land systems, land forms, hillslopes, land components,
and land elements.
92
A land element is the smallest practical terrain unit at a given scale of mapping. McDonald et al.
(1984) suggest that such elements should not be less than 150x150 metres in size (i.e. less than
2.25 hectares) at 1:50 000 scale, but can potentially be much larger in homogeneous landscapes.
Land elements can be combined to form land components which are typically associated with
ridge crests, fallfaces, midslopes, and footslopes (Argialas 1995; Dymond, Derose &
Harmsworth 1995).
Hillslopes (also called profiles) are sequences of land components orientated in the slope
direction (see Figure 6-1). The sequences of components forming hillslopes differ according to
number and type. Fallfaces are, for instance, not present on low hills while midslopes are absent
on mesas. Complex hillslopes can include multiple occurrences of a particular type of
component.
Figure 6-1 Two hypothetical hillslopes, each co nsisting of a sequence of five land components
Landforms (e.g. hills, mesas, escarpments) ar e essentially sequences of hillslopes arranged
perpendicular to the slope direction and they are, in many cases, the main focus of terrain
analysis. However, landforms have little value in land suitability analysis because land properties
can vary considerably within a landform. In the southern hemisphere, temperatures will, for
instance, be considerably higher on north-facing hillslopes than on south-facing hillslopes, while
soils will be deeper in channel beds than on crests. Land components are thus the most
appropriate demarcations to use as the basis for suitability analysis.
Traditional techniques for demarcating land component boundaries are subjective and time-
consuming as they rely on the visual analysis of terrain data. The following sections overview
terrain analysis, geomorphometry and the automated land component mapping techniques
considered for demarcating land components in the Western Cape.
Source: Van Niekerk & Schloms (2001)
93
6.1.1 Terrain analysis
Terrain analysis was defined in Section 4.1.3.4 as the study of the nature, origin, morphological
history and composition of landforms, the result of which is a landform or land component map.
Land components can be mapped by studying t opographical maps, interpreting aerial
photographs (Speight 1977) and making field m easurements (Graff & Usery 1993). Such terrain
analysis techniques are considered to be an art without formal theory and often rely on the
interpreter?s implicit terrain-related knowledge of the area being studied (Irvin, Ventura & Slater
1997). Such skill is the product of lengthy, expe nsive training and experience (Argialas 1995).
The subjective nature of terrain analysis is a major drawback because in most cases it is
impossible to make any useful comparisons between land component maps produced by
different analysts or even by the same analyst at different times (Speight 1977). The
interpretation and mapping of land components are extremely time-consuming, labour-intensive
and expensive tasks (Adediran et al. 2004) and is difficult to verify in the field owing to the
fractal nature of topography (Hengl, Gruber & Shrestha 2004). Consequently, more objective
and automated methods are needed to map land components. Computer analysis of
geomorphometry is a convenient option.
6.1.2 Geomorphometry
Geomorphometry, the numerical representati on of topography, combines mathematics,
engineering and computer science. In the past, geomorphometry concentrated on the geometry
of terrain, but technical advances in computing, analytical algorithms, input-output devices and
large sets of topographic data have shifted the focus to digital representation of terrain, process
modelling and generalization (Adediran et al. 2004).
Recently, the increasing availability of digital elevation models has promoted the use of
computer technology for the calculation and discrimination of terrain properties. DEM-derived
data sets such as slope, aspect, hydrographical pattern and shaded relief are being increasingly
exploited in terrain analysis. These morphometric parameters are not only less prone to human
error but can be used to objec tively and quantitatively compare terrain units (Dymond, Derose &
Harmsworth 1995; Giles & Franklin 1998).
GIS is often used to support geomorphometry and land component mapping. The most common
approach is to use GIS overlaying techniques to combine DEM derivatives such as slope and
aspect to create unique, homogeneous morphological units (Adediran et al. 2004). Classification
is required to convert the continuous slope and aspect raster surfaces into regions (polygons).
Once the slope and aspect rasters have been clas sified, they are usually converted to vector
94
format and overlaid to create new polygons representing combinations of aspect and slope. The
overlay operation is, in many cases, followed by a conflation operation to get rid of
insignificantly small polygons.
The use of overlay tech niques to delineate morphological land units is simple, fast and can be
done with standard GIS software. The problem with this technique is the way in which terrain is
generalized during the classification process. Slope aspect is usually classified into nine standard
aspect classes representing north, north-east, ea st, south-east, south, south-west, west, north-
west, and no aspect (level) (Dymond, Derose & Harmsworth 1995), while slope gradient is
usually classified into a number of equal-interval classes. The effect of applying such
classification schemes over the entire extent of the slope gradient and aspect rasters (i.e. as a
global raster operation) is that class breaks will not likely coincide with local terrain transitions.
This is especially problematic for slope breaks because small transitions in slope gradient can
have drastic effects on land properties such as soil and vegetation cover.
The hypothetical slopes shown in Figure 6-1 illustrate that slope breaks do not occur at
consistent slope gradients. For in stance, there is an abrupt (33? ) transition between the midslope
and the fallface, while in the far range the transition is less acute (5?). Although the slope
gradient variance along slope breaks is exaggerated in the illustration, it demonstrates that land
components such as midslopes can differ significantly in terms of slope gradient. While these
subtle differences will be relatively easy to map using manual techniques (i.e. topographic map
and aerial photo interpretation), it is unlikely that the boundaries of land components will be
mapped along natural slope breaks using the global classification and overlay approach described
above (Dymond, Derose & Harmsworth 1995).
6.1.3 Automated mapping
The inability of GIS classifica tion and overlaying techniques to accurately identify and map
slope breaks prompted Dymond, Derose & Harmsworth (1995) to develop an algorithm to
automatically detect such transitions from a DEM. The algorithm, which was implemented in
FORTRAN, starts by classifying as pect into eight 45? classes and then splits each resulting
aspect region into two major land components re presenting upper and lower slopes. Each aspect
region is split along the five-metre vertical interval contour that is most likely to represent a
slope break. The most appropriate division is determined by comparing the variances of each
component pair created at consecutive elevation intervals. The sp lit elevation giving the smallest
average variance is used if the difference between the upper and lower slope angles is significant.
If necessary, the process is repeated for each of the upper and lower land components, to
95
potentially produce four land components representing upper-upper slope s, upper-lower slopes,
lower-upper slopes and lower-lower slopes.
According to Dymond, Derose & Harmsworth (1995), the algorithm produces results which
compare well with results obtained from manual mappings using an analytical stereoplotter.
However, some land component boundaries are not realistic because slope breaks often occur at
varying elevations. Better results are obtainable by using distance from streams instead of
elevation as the split lines. A major limitation of the algorithm is that a maximum of four land
components can be mapped per aspect region, while in reality many more slope breaks occur.
Small (5m) elevation or distance intervals are also necessary to accurately delineate land
component boundaries. This has a significant eff ect on the performance of the algorithm as
several land component pairs must be examined regarding variability for each aspect region. For
an aspect region with a 100m elevation range, 20 land component pairs have to be examined for
the first division (upper and lower slopes). If a second division is necessary, the number of
iterations doubles. This approach requires extremely intensive computer processing and is not
viable for large areas like the Western Cape. Cl early, an improved, more efficient algorithm for
automated land component mapping is needed.
6.2 Automated component mapping with ALCoM
Based on the work done by Dymond, Derose & Ha rmsworth (1995) a new algorithm, called the
Automated Land Component Mapper (ALCoM), was developed by Van Niekerk & Schloms
(2001) to automatically map land components from a DEM. ALCoM differs from Dymond,
Derose & Harmsworth?s (1995) algor ithm in that it relies on a statistical technique developed by
Jenks (1967) to identify natural slope breaks. Although Jenks? (1967) technique can be applied to
identify natural breaks within any data set, it proved to be very effective in identifying slope
breaks when applied to slope gradient data (Van Niekerk & Schloms 2001). The statistical
detection of natural breaks is relatively fast and was expected to be more efficient and accurate
than the sequential land component mapping approach of Dymond, Derose & Harmsworth
(1995).
Like the Dymond, Derose & Harmsw orth (1995) algorithm, ALCoM (see Figure 6-2) starts with
the creation of an aspect raster (step 1), which is then classified (step 2) into nine 45? aspect
classes. The resulting aspect cl assification is regionalized in st ep 3, resulting in unique polygons
representing individual aspect regions or directional hillslopes. The focus of the algorithm then
changes from a global to the local level by calculating the slope gradient for each aspect region
(steps 4 and 5). The most prom inent slope break in an aspect region is then determined by
employing Jenks? (1967) technique with the number of breaks set to one (step 6).
96
Figure 6-2 The ALCoM algorithm
In step 7, the slope gradient variance of each of the resulting land components is determined and
the detection of slope breaks is repeated with increasing number of breaks until each of the
resulting land components is homogeneous in terms of slope gradient. To determine
homogeneity, the slope gradient variance (SGV) of each land component is compared to the
SGV of the entire aspect region. A land componen t is considered to be homogeneous only if its
SGV is ten times lower than the overall SGV of the aspect region. A lower ratio results in the
mapping of smaller, more homogeneous land components, while a higher ratio produces larger,
less homogeneous land components. Once the accepta ble level of homogeneity is reached, the
next aspect region is considered. The algorithm terminates when all land components for all
aspect regions have been mapped.
ALCoM was automated in ArcView GIS using the Avenue programming language (ESRI
2002a) (see Appendix B for source code) and tested on a 15x5km area in the Stellenbosch region
(see Figure 6-3). The test area was chosen becau se it includes a variety of land components
ranging from predominantly level areas in the west to fallfaces, ridges and crests in the east
(Figure 6-4). The researcher also has implicit knowledge of the area, which is essential for the
interpretation of the results.
DEM
ASPECT
Aspect
Classify
Regionalize
ASPECT REGIONS
Slope
Jenks?
LAND COMPONENTS
ASPECT CLASSES
Select region
ASPECT REGION
SLOPE GRID
Variance
HIGH
LOW
DATA
Operation
KEY
Step 1
Step 2
Step 3
Step 4
Step 5
Step 6
Step 7
97
Figure 6-3 Location of the test area for ALCoM
Figure 6-4 Detailed view of the test area , with selected terrain features indicated
ALCoM was applied to the WCDEM (see Section 5.1.2) and produced 1057 land components
for the test area. The land components were visually compared with aerial photographs,
orthophoto maps, topographical maps and SPOT 5 satellite images. The results (see Figure 6-5)
show that ALCoM is very sensitive to small cha nges in relief, resulting in a highly detailed land
component map.
Figure 6-5 Land component boundaries mapped by ALCoM
Channel bed
Slope breaks
Fallface
Ridge
Crest
Channel bed
Footslope
Footslope
Crest
Channel bed
Slope breaks
Fallface
Ridge
Crest
Channel bed
Footslope
Footslope
Crest
98
Although the high level of detail of the resulting land components makes comparison difficult,
visual inspection shows that the boundaries of land components correspond very well with slope
breaks. Slope breaks that do not follow contour lines or that are not parallel to streams are
accurately mapped. This is an improvement over the Dymond, Derose & Harmsworth (1995)
algorithm. Statistically, the land components are relatively homogeneous with standard
deviations of 2.9? and 39.6? for slope gradient and aspect respectively. The relatively high
standard deviation of slope aspect is attributed to the nine 45? classes that were used to derive
aspect regions (see step 1 in Figure 6-2) and the larger rang e of possible values (0-359?).
The efficiency of ALCoM is considerably higher than that of the Dymond, Derose &
Harmsworth (1995) algorithm, as the number of iterations needed to detect slope breaks with
ALCoM is not related to elevation difference. Because the maximum number of iterations per
aspect region is equal to the number of slope breaks, no unnecessary iterations are required. In
spite of the improvement in efficiency, however, ALCoM still requires considerable computer
resources. Based on the 1.3 hours processing tim e that was required to generate the land
components in the test area, it is estimated that more than 100 days of processing time would be
needed to analyse the entire Western Cape provin ce. Because such a long process is unlikely to
complete successfully without interruption, a more efficient solution is required. Owing to its
ability to efficiently analyse large data sets, image processing was considered as alternative for
the automated demarcation of the Western Cape?s land components.
6.3 Image processing techniques
Image processing involves the manipulation of digital imagery to enhance its quality, change its
format or to extract various types of information. It encompasses a wide range of techniques for
which specialized computer programs are needed. Because images are essentially multilayered or
stacked rasters, image processing can be used to analyse terrain data. This section focuses on two
image processing techniques used in land component mapping, namely image clustering and
segmentation.
6.3.1 Image clustering
Image clustering is most commonly used to convert multiband imagery into regions of similar
attributes. The best-known clustering technique is the iterative self-organizing data analysis
technique algorithm (ISODATA) (Hall & Khanna 1977) frequently used to cluster multiband
satellite imagery into regions of similar spectral reflectances. The process starts with the
specification of the number of classes (clusters) needed, followed by the assignment of arbitrary
mean values to each class. Each raster cell is allocated to the closest class mean in the feature
99
space, class means are recalculated and each pixel is again allocated to the new class means. This
procedure is repeated several times until the class means stabilize (Gibson & Power 2000). The
resulting classes are regarded as unique spectral combinations and, in most cases, are combined
and converted to information classes (i.e. land cover classes) (Campbell 2006).
Instead of using ISODATA on satellite imagery, Irvin, Ventura & Slater (1997) employed the
technique on terrain attributes such as elevation, slope gradient, slope aspect (solar radiation) and
curvature to map land components. Although the approach produced relatively good results, the
histograms of many of the clusters had multimodal distributions, indicating that the land
components were not homogeneous. Clustering techniques such as ISODATA are based on
global thresholds, which often lead to overclustering (i.e. producing units that are too small)
and/or underclustering (merging regions that do not belong together) because local contrasts are
not considered or are not consistently represented (Definiens Imaging 2004). Another problem
with clustering techniques is that the analyst needs to specify the number of clusters that should
be generated (Adediran et al. 2004). Because ther e is no way in which the optimal number of
clusters can be known in advance, the only way to find an appropriate value is through
experimentation and knowledge of the area under study (Irvin, Ventura & Slater 1997). In
addition, the optimal number of clusters depends on the landscape being studied, making it
unsuitable for application in large, complex areas requiring considerable experimentation and
operator interaction.
6.3.2 Image segmentation
Like clustering, image segmentation algorithms group pixels into spatial regions (segments)
which meet predetermined criteria of homogeneity (Definiens Imaging 2004). The main
difference between clustering and segmentation is the way in which the image is regionalized.
The conceptual difference between cluste rs and segments is illustrated in Figure 6-6.
Figure 6-6 A conceptual comparison of clusters (a) and segments (b) in relation to attributes A to H
A
A
D
A
B
B
C
C
A
C
B
D
E
F
G
H
(a) (b)
100
Where clusters can consist of one or more gr oupings of pixels (polygons) that have similar
attributes (indicated as A to G) in the context of the entire image, segments are individual pixel
groupings locally different from adjacent pixels.
Among the various existing image segmentation methods, region-growing segmentation is the
best known. Region-growing segmentation clusters adjacent cells together if they have similar
attributes. The segmentation process starts with a number of seed points that are either randomly
sampled, statistically determined or specified by the user (Definiens Imaging 2004). The
advantage of using randomly sampled seed points is that the procedure is autonomous and
requires no input from the user. This approach ca n, however, lead to unpredictable results as the
segmentation is highly sensitive to the initial positions of the seed cells. The use of random seeds
also means the process cannot be repeated to produce the same segmentation result.
Better segmentation results can be obtained when seed points are statistically determined, but
such measures are related to the global feature space of the image and are therefore relative to
the specific image. Any change in the extent or position of the image will produce different seed
points, which means that the segmentation will yield different results.
Seed points can also be specified by the analyst. Miliaresis (2001), for example, used cells that
were pre-classified as ridges to discriminate mountainous and non-mountainous regions from a
DEM. In another terrain application, Giles & Franklin (1998) selected seeds based on field
surveys and aerial photo interpretation to map land components. Campbell (2006) warns,
however, that the use of training data imposes a structure on the clustering which might not
match the natural clusters that exist in the data. In addition, the selection of training data can be a
time-consuming, expensive and tedious undertaking, especially for large regions.
To overcome the limitations of region-growi ng image segmentation algorithms, De Kok,
Schneider & Ammer (1999) developed an algor ithm to extract homogeneous image objects
based on local contrasts. An important feature of this technique is that segmentation can be
repeated to produce the same results, even if the extent or position of the image is changed. The
so-called multi-resolution segmentation (MRS ) technique can operate on multiple bands
simultaneously and can produce multiscale segmentations on images having different
resolutions. MRS was first implemented in 2000 as part of the eCognition object-orientated
image analysis software package. eCognition is different from other remote sensing software in
that image classification is not performed on individual pixels, but rather on segments (also
called objects or polygons). The advantage of this approach is that important semantic
information such as shape, texture and topology can be used. Such information is not apparent
when considering individual pixels (Definiens Imaging 2004).
101
The MRS algorithm is based on a pairwise re gion-merging technique which consecutively
merges image cells. It involves an optimization pr ocedure which, for a given number of objects,
minimizes the average heterogeneity and ma ximizes their respective homogeneity. The
procedure starts with single seed cells, which are iteratively merged into larger units while the
upper threshold of homogeneity is not exceeded locally. If none of the neighbouring cells fall
within the allowed thresholds, the best candidate becomes the seed and the merger process is
repeated. This approach minimi zes variability within merged cells (Definiens Imaging 2007).
The homogeneity threshold is in directly set by the operator through a scale parameter which
determines the number of segments that will be created. Conversely, variation within each
segment increases as the scale parameter increases. Too high values could therefore lead to the
loss of important detail, while too low values can result in an unnecessarily large number of
small and almost identical segments. For efficien t storage and faster processing, the number of
segments should be kept to a minimum. The s cale parameter is not a quantitative value and
cannot be based on any scientific calculations. The only way to determine an appropriate scale
factor is through experimentation (Definiens Imaging 2004).
Although no implementations of MRS for land component mapping from DEM could be found
in the literature, the technique is expected to produce similar results to ALCoM due to the way in
which it discriminates regions based on local contrasts. To compare MRS with ALCoM, the
technique was applied to the same 10x5km test area shown in Figure 6-3. Slope and aspect
rasters were derived from the WCDEM and imported as separate layers into the latest version of
eCognition (version 7), named Definiens Developer (Definiens Imaging 2007).
To find an appropriate scale parameter, the rasters were segmented several times with increasing
scale parameters. The resulting land components for each scale level were statistically compared
with those produced by ALCoM. The comparison found that a scale parameter of 10 generated
land components exhibiting similar levels of overall homogeneity of slope and aspect to those
produced by ALCoM. While the land components produced by ALCoM are slightly more
homogeneous (i.e. have a lower overall standard deviation) for slope gradient, slope aspect
variance is slightly lower in the MRS data set (see Table 6-1).
Table 6-1 Statistical comparison of land components mapped using ALCoM and MRS
LEVEL # of components Average size
(ha)
OSD* of slope
gradient (deg)
OSD* of slope
aspect (deg)
OSD* of area
(ha)
ALCoM 892 8.4 2.9 36.7 25.6
MRS 509 14.7 3.4 33.7 13.3
* Overall standard deviation
102
Overall, the MRS and ALCoM land components ar e very similar regarding slope and aspect
variation. There is, however, a clear difference when the two data sets are compared visually
(compare Figure 6-7 with Figure 6-5).
Figure 6-7 Land components mapped by multi-resolution segmentation
It is apparent that the land components in Figure 6-7 are more simila r in size than those of
ALCoM. This observation is supporte d by the fact that the standard deviation of land component
area is 13.3 hectares for the MRS data set, while the corresponding value is 25.6 hectares for the
ALCoM data set ( Table 6-1). Although one would expect th at areas with low relief will produce
larger land components, the MRS algorithm is more sensitive to local terrain variability. In the
ALCoM algorithm, local variability is related to th e overall variability of a hillslope (i.e. aspect
area), which can include large, relatively level and homogeneous regions that will not be
subdivided into smaller land components.
ALCoM produces results that are more aligned w ith manual interpretations of terrain. However,
for land suitability analysis, the land components produced by ALCoM are not necessarily better
than those produced by MRS. The higher local sensitivity of MRS produces more detailed
components in areas of moderate terrain which are more likely to be affected by land use
changes. The ability of MRS to pick up subtle cha nges in moderate terrains is an invaluable asset
for land evaluation purposes.
A major advantage of MRS for mapping land com ponents is its efficiency. Where ALCoM took
1.3 hours to generate th e land components in the test area, MRS completed the mapping in a
fraction of a second! Definiens Developer is cl early an optimal solution for image processing
compared to ALCoM which was developed under th e constraints of an existing GIS environment
(ArcView GIS). Based on the time it took Definiens Developer to process the test area, it was
estimated that it would map the entire We stern Cape in a matter of minutes.
Channel bed
Slope breaks
Fallface
Ridge
Crest
Channel bed
Footslope
Footslope
Crest
103
6.4 Comparison of ALCoM and MRS
Due to the indiscriminate way in which terrain is generalized, classification and overlay
techniques are not suitable for delineating land components. A new technique, called the
Automated Land Component Mapper (ALCoM), wa s developed to identify slope breaks to be
used as boundaries for land components. Although the resulting land components are
representative of the terrain features, the process proved to be too slow and unsuitable for
implementation in large areas. A new image segmentation technique, called multi-resolution
segmentation (MRS), was subsequently tested as an alternative to ALCo M. MRS relies on local
contrasts and variability to produce segments (land components), irrespective of the extent and
position of the input image. In comparison to ALCoM, MRS not only produces fewer and
therefore more homogeneous land components, but completes the task in considerably less time.
The results also show that MRS produces more detailed land components in areas of moderate
terrain which are more likely to be targeted for land use changes. For land evaluation purposes,
MRS is demonstrably the most appropriate method for mapping land components in the Western
Cape.
6.5 SEGMENTING THE WESTERN CAPE
Multi-resolution image segmentation, as implem ented in Definiens Developer software, was
used to delineate land components/units from the Western Cape DEM (WCDEM). Because
hardware and software constraints prevent it from being combined into one data set, the 20m-
resolution WCDEM was obtained as 12 adjoining tiles covering the Western Cape. Although the
high resolution of the WCDEM is an asset in that it ensures accuracy, the model is too large to
be processed on a provincial level at its native resolution. To enable the combination of the data
into a single data set, the individual blocks were rescaled to 80m-resolution and merged. A
resolution of 80 metres was chosen because it is a factor of the original resolution, which meant
that no changes in the extent of the data set was not required. A cell size of 80x80m is also
considerably less than the required resolution (or minimum mapping unit) of 150m (see Section
5.1.1).
Although the volume of data of the rescaled version of the WCDEM (henceforth called
WCDEM80) remains large, it can be analysed as a unit. However, as discussed in Section 5.1.1,
resolution influences DEM quality and a reduction of resolution invariably reduces accuracy. To
determine what effect the resolution change has on accuracy, the rescaled data set was assessed
using the same method described in Section 5.1.3. The accuracy assessment shows that the
WCDEM80 has a mean absolute error of nine me tres compared to the seven metres of the
104
WCDEM. The accuracy difference of two metres betw een the degraded and original data sets is
insignificant, especially when compared to the 15m vertical accuracy of the best alternative
DEM (i.e. SRTM DEM).
Once merged, WCDEM80 was used to generate slope gradient and aspect rasters using ArcGIS
9.2. The resulting rasters were exported from ArcGIS as single-layer 8-bit TIFF images and
combined in Definiens Developer. MRS was performed on these images, using a scale factor of
10, to produce 652 704 land components covering the Western Cape. The average size of the
land components (19.8ha) is consid erably larger than the size of those produ ced by MRS in the
test area (Table 6-1). This difference is attributable to the higher average slope of the test area
(14.4?) compared to the entire Western Cape?s (6.5?) because land components are larger in
moderate terrain than in mountainous areas. The segmentation of the entire province performed
well concerning slope and aspect homogeneity ? the overall standard deviation of slope gradient
and aspect being 2.9? and 17.2? respectively. Th is indicates that the land units produced for the
whole Western Cape are more homogeneous than those mapped for the test area.
The segmentation procedure for the Western Cape was completed in less than one hour and the
result was exported from Definiens Developer as an ESRI shape file used to facilitate the
extraction of the environmental and physical land properties by a GIS.
6.6 LAND PROPERTY EXTRACTION
Once the soil, climate and terrain data was collected, it had to be prepared for importation into
the land unit database. To do so, the land property data sets not already in ESRI grid format were
converted and projected to conf orm to the land unit data. The uni versal transverse Mercator
(UTM) map projection was chosen because the enti re province is represented in a single zone
(34S) and conforms to many other da ta sets of the Western Cape.
Once all the data sets accorded with the coordinate system, the extraction process involved
calculating the average property value for each land unit. This procedure can be performed with
the Zonal Statistics tool in ArcGIS Spatial Anal yst extension. Unfortunately, the tool can only
handle data sets of 100 000 or fewer polygons a nd was therefore unable to calculate the land
properties for the more than 650 000 land units.
An alternative algorithm for extracting land properties was developed in ArcView GIS.
Essentially, the algorithm breaks the task of calculating zonal statistics into areas of manageable
size (quarter degrees) and repeats the process fo r each area consecutively. The process takes a
few hours to complete. The Avenue code develope d for this algorithm is attached as Appendix
C.
105
6.7 STORAGE
To determine the most efficient database solutio n, the land units were stored as vector polygons
(areas) in both ESRI shape file format and in ESRI geodatabase format using ArcSDE and
Oracle 9.1 RDBMS. Oracle is currently the most popular RDBMS and is widely considered to be
the most robust DBMS software available (see Section 3.2.6). Because Oracle is essentially a
non-spatial database, ArcSDE acted as a spatial database engine.
Shape files have become a de facto GIS data exchange format in South Africa and they are
compatible with most spatial software packages. A polygon shape file is a simple, non-
topological spatial database in which the geometrical location and shape of the polygons are
stored separately from the attribute information. Each polygon can have multiple attributes
which are stored as fields (columns) in a dBase table and each record (row) in the table
corresponds to one polygon (ESRI 2007a).
The geodatabase and shape file land unit databases were compared as to their response times for
queries and updates. Although the geodatabase was expected to be faster than the shape file
format database, very little difference in response times was observed during querying. Updates
to the attributes of the geodatabase were three times slower than updates to the shape file. The
slower response times of the geodatabase are ascribable to the additional ?housekeeping?
required by the versioning feature in geodatabases (ESRI 2007a). The major advantage of a
geodatabase approach is that the size of the database is limite d only by the available hardware.
Theoretically, any number of polygons can be accommodated in a geodatabase, whereas shape
files are limited to three billion polygons (ESRI 2002b). Because the number of land units was
fewer than one million (652 0704), the additional capacity of a geodatabase was not needed so
that the land unit database will be stored in shape file format. Because of the loosely coupled
design of the system, the land unit database can be migrated to Oracle when more storage space
is required without affecting the other system components.
6.8 SUMMARY
The land unit data set is an efficient representati on of land as it combines a range of terrain and
other land variables into logically delineated polygons along with their attributes. Because
CLUES uses land units as the basic analysis an d mapping unit, special care had to be taken to
ensure that land units represent homogeneous parcels of land. Several techniques were
considered for the optimal delineation of land units, with ALCoM and image segmentation
producing the best results. The latt er technique is the more efficient one and was used to map the
more than 600 000 land units covering the Western Ca pe. Once the land units were delineated, it
106
was used to extract soil, terrain and climatic properties from the collected data sets. This process
was automated in ArcView.
The land unit database is the basis for land suitability analysis using CLUES as it represents the
basic mapping units along with their land properties. To determine the suitability of each land
unit for a particular land use, land use requirements in the form of rules must be defined. The
land requirement rules are stored in a database separate from the land units. The development of
this so-called knowledge base is discussed in the next chapter.
107
CHAPTER 7: DEVELOPING THE KNOWLEDGE BASE
In the expert systems approach to land evaluation systems development, the knowledge base is a
collection of land requirement rules based on existing expert knowledge. The rules are stored in
predefined format to comply with the inference engine which carries out the suitability analysis.
Most existing expert systems are single-user systems, meaning that a database of rules is created
to explicitly carry out suitability analysis for a particular project or user. In multi-user
applications such as CLUES, each user must be able to populate their own individual knowledge
base with rules. A multi-user knowledge base is th erefore required to store and manage the rules
of each user. The design and implementation of a da tabase is discussed in this chapter in two
consecutive main sections. To ensure that th e database includes all the necessary data to
effectively support the suitability analysis procedure, a structured methodology called logical
data modelling, was used in its development.
7.1 LOGICAL DATA MODELLING METHODOLOGY
The CLUES database was designed using the l ogical data modelling approach. Logical data
modelling (LDM) is based on the ph ilosophy that ?business data [s uch as land use requirements]
have an existence that is independent of how they are accessed, who accesses them, and whether
or not such access is computerized? (Fleming & Von Halle 1989:9). The methodology is entirely
data-driven and is not biased by any application requirements or technological considerations. It
also facilitates comprehensive understanding of the business information requirements and
effective communication of these requirements to designers, developers and users. The well-
structured technique provides a foundation for the design of correct, consistent, sharable and
flexible databases using any database technology and software. Th e process involves the
following ten sequential design activities:
1. Identify major modelling entities.
2. Determine operational relationships between entities.
3. Identify primary keys.
4. Define foreign keys.
5. Determine key business rules.
6. Add remaining non-key attributes.
7. Normalize data structure.
8. Specify additional attribute business rules.
9. Combine user views.
10. Integrate the model with existing data models.
108
Each of these ten sequential steps was completed during the operational design of the CLUES
database as described in the following sections.
7.1.1 Identify major modelling entities
Entities are the key objects of interest to the user and can either be physical (e.g. user, land unit)
or abstract (e.g. land requirement, land property) in nature. Fo r the CLUES database, six objects
were identified, namely users, land uses, land requirements, land requirement rules, land
properties and land units. These objects are major entities. The user entity represents any user
who has access to the system and land use is the object being evaluated. A land requirement is
the third entity and, because requirements are sets of rules, each land requirement rule is
considered to be a discrete entity. During suitability analysis, the land requirement rules are
compared with land properties, which are related to the physical land units covering the study
area. Two other objects, namely project and data source are added for operational reasons
bringing the total number of objects to eight. The projec t entity was added as an abstract concept
to represent a specific suitability analysis because it is expected that users will need to work on
several different suitability analyses at any given time. The data source entity was added because
it is envisaged that the system will eventually include hundreds of land properties and that
metadata about each property will be needed to keep track of it.
7.1.2 Determine operational relationships between entities
The second step in logical data modelling is to determine relationships between entities. All
relationships have direction, which can be defined as one-to-one (1:1), one-to-many (1:M), or
many-to-many (M:M). A one-to-many relation be tween entities A and B signifies that one
instance of A will relate to many instances of B. For example, one land use will relate to many
land requirements.
The relationships and directions for each pair of the identified entities are specified in Table 7-1.
The relationship matrix indicates that USER (f irst row) has 1:M relationships with LAND_USE
(second column) and PROJECT (seventh column) as each user can specify many land uses and
work on several projects. The notation 1:M indicates direction, which means that an inverse (i.e.
M:1) relationship exists between, for instan ce, LAND_USE (second row) and USER (first
column). There is a many-to-many relationshi p between PROJECT and LAND USE because, in
optimal land use identification, each project can include many land uses and a particular land use
can feature in several projects.
109
Table 7-1 Relationship matrix of entities in the knowledge base
ENTITIES User Land use Land requirement
Land
requirement
rule
Land
property Land unit Project Data source
User - 1:M * * * * 1:M *
Land use M:1 - 1:M * * * M:M *
Land
requirement * M:1 - 1:M 1:1 * * *
Land
requirement
rule
* * M:1 - * * * *
Land
property * * 1:1 * - M:M * M:1
Land unit * * * * M:M - * *
Project M:1 M:M * * * * - *
Data source * * * * 1:M * * -
* No direct relationship
Each land use relates to a number of land requirements and each land requirement is related to
many land requirement rules. Although there is an indirect relationship between users and land
requirement rules (i.e. via the entities land use and land requirement), only direct relationships
are considered in the matrix. The relationship betw een land units and land use is also indirect.
In order to evaluate each land unit?s suitability, there needs to be a 1:1 relationship between land
requirement and land property. There is a many- to-many relationship between land unit and land
properties, because each land unit will likely have many properties, while each property is
represented by many land units. Concerning metadata, several land properties can have the same
data source.
7.1.3 Identify primary keys
The third step in the LDM methodology is to add ke y attributes to each entity. An attribute is an
atomic unit of data about an entity. The most important attribut e of any entity is the primary key
used to uniquely identify a specific occurrence of an entity. Because each of the entities is
created and managed by the users, identification numbers are used as the primary key for each
110
entity. Each primary key is named using the entity name and the suffix ?_ID?. For instance,
PROJECT_ID is used as the PROJECT entity?s pr imary key. The underscore character is used
instead of spaces as most DBMS do not allow spaces in attribute names. Upper-case characters
are used for uniformity. The same convention is followed for the entity names (i.e. land use is
changed to LAND_USE)
7.1.4 Define foreign keys
The definition of foreign keys invo lves the identification of keys that relate to other entities. Due
to the simplicity of the data model, all of the foreign keys for the relationships in the matrix are
defined to be the primary keys of the related entities. For instance, the USER entity is configured
to relate to the LAND_USE entity through the US ER_ID attribute, while the LAND_USE_ID is
used to relate to LAND_REQUIREMENT.
7.1.5 Determine key business rules
Key business rules govern the effects of insert, delete and update operations on relationships and
they address the integrity of attributes through placing constraints on the values of attributes (i.e.
they impose domain integrity). Insertion implies that all the relationships in Table 7-1 are
defined as being ?dependent? on one another. In other words, a new instance of a child entity
(e.g. LAND_USE) can only be created if an instance of the parent entity (e.g. USER) already
exists. Conversely, for the deletion of instances, parents can only be deleted if no child instances
occur. To ensure the integrity of the entities, prim ary keys are not allowed to be null (i.e. zero or
empty), must always be a number and may never be duplicated (i.e. two records may not have
the same value).
7.1.6 Add remaining non-key attributes
The sixth step of the LDM appr oach is to add non-key (i.e. not primary or secondary key)
attributes to each entity. The resu lt of this activity is shown in Table 7-2. The USER_ID attribute
in entity USER is the most important as it not only acts as the primary key for the entity, but also
serves as foreign key to several other entities. In addition to the USER_ID, three other (non-key)
attributes were added to the USER entity. The NAME attribute contains the name of the user and
is simply a way of identifying who each user is. The PASSWORD item is needed to restrict
access to the system and protect the users? proj ects. Email addresses are used to communicate
with users about system updates, maintenance and password change notifications.
111
Table 7-2 Entity attributes in the knowledge base
ENTITY ATTRIBUTES TYPE
USER USER_ID
NAME
PASSWORD
EMAIL
Primary key + Foreign key
Non-key
Non-key
Non-key
LAND_USE LAND_USE_ID
USER_ID
NAME
Primary key
Foreign key
Non-key
LAND_ REQUIREMENT LAND_REQUIREMENT_ID
LAND_USE_ID
LAND_PROPERTY_ID
USER_ID
WEIGHT
Primary key
Foreign key
Foreign key
Foreign key
Non-key
LAND_ REQUIREMENT_RULE LAND_REQUIREMENT_RULE_ID
LAND_REQUIREMENT_ID
SUITABILITY
LOWER_VALUE
MIDDLE_VALUE
UPPER_VALUE
CURVE_ID
Primary key
Foreign key
Non-key
Non-key
Non-key
Non-key
Non-key
LAND_PROPERTY LAND_PROPERTY_ID
DATA_SOURCE_ID
NAME
UNIT
MIN
MAX
Primary key
Foreign key
Non-key
Non-key
Non-key
Non-key
LAND_UNIT LAND_UNIT_ID
LAND_PROPERTY_ID
VALUE
Primary key
Foreign key
Non-key
PROJECT PROJECT_ID
USER_ID
LAND_USE_ID
NAME
MODIFIED
FUNCTION
MIN_X
MAX_X
MIN_Y
MAX_Y
Primary key
Foreign key
Foreign key
Non-key
Non-key
Non-key
Non-key
Non-key
Non-key
Non-key
DATA_SOURCE DATA_SOURCE_ID
NAME
SCALE
ORIGIN
Primary key
Non-key
Non-key
Non-key
112
The only non-key attribute adde d to the LAND_USE entity is th e NAME attribute which enables
users to provide a description of each land use. For th e LAND_ REQUIREMENT entity, a
WEIGHT field defines the relative importance of a specific land requirement for a particular land
use.
Five additional attributes were allocated to the LAND_ REQUIREMENT_RULE entity. The
SUITABILIY field accommodates the suitability le vel (i.e. N2, N1, S3, S2, S1) for each rule,
while LOWER_VALUE, MIDDLE_VALUE and U PPER_VALUE specify the thresholds and
central value for each rule. The CURVE_ID attr ibute differentiates between Boolean and fuzzy
rules.
Four non-key fields were added to the LAND PR OPERTY table to keep track of the different
land properties for each land unit. While the NAM E field stores a description of each land
property, the UNIT field stores the unit (e.g . degree, metres and millimetres) used to
quantitatively measure each property. MIN and MAX fields indicate the minimum and
maximum values of each property respectively ? values needed to automatically scale ?open-
ended? rules (see Section 2.4.1).
In contrast to the LAND_UNIT entity which is allocated only one additional non-key attribute
(VALUE) to store the individual land property valu es for each land unit, the PROJECT entity is
supplemented with seven non-key attributes. Th e PROJECT entity is also provided with a
NAME field for description purposes. An additional descriptive attribute, called MODIFIED, is
included to store the date and time of a project?s last updating. The rest of the attributes in the
PROJECT entity are related to operational informa tion. The FUNCTION field indicates the type
of previous analysis conducted (i.e. land use suitability or optimal land use identification), while
the MIN_X, MAX_X, MIN_Y a nd MAX_Y fields define the extent of the study area.
The final three non-key attributes added to th e logical data model make provision for the
description of each land property data source. To do so, metadata items NAME (description of
the source), SCALE (map scale) and ORIGIN (original owner/developer) are included.
7.1.7 Normalize data structure
Normalization ensures internal consistency, minimal redundancy and maximum stability of data,
without the loss of information. It is a method by which a logical data model can be optimized
through three simple steps, namely:
1. Remove repeating or multivalued attributes to a separate child entity.
2. Remove all non-key attributes that ar e not dependent on the primary key.
3. Remove attributes that depend on other non-key attributes.
113
Because no repeating or multivalued attributes were present and no attributes were found to be
independent of the respective primary keys, the CLUES logical data model was already in
normalized form and no further action was needed.
7.1.8 Specify additional attribute business rules
After normalization, the next step completed was to supplement the business rules ? specified in
Section 7.1.5 for key attributes ? with additional bus iness rules. The main activity during this
stage of the logical data modelling process is to define the domains (i.e data type, length, range,
uniqueness and null support) and triggers (i.e. insert, modify, and delete rules) for each attribute.
Table 7-3 shows the business rules for the USER entity. The business rules for the primary key
USER_ID, defined in Section 7.1.5, are documented as being numeric. Because it is a primary
key, it must be unique and non-nul l (i.e. no null values are allowed). Inserts of new instances of
USER_ID are allowed and no trig gers are necessary in such cases. The USER_ID cannot be
updated or deleted without ensuring that no child entities are present. Update and delete triggers
are needed for the USER_ID field.
Table 7-3 Business rules for USER entity in the knowledge base
ATTRIBUTE DOMAIN AND TRIGGER RULES
USER_ID Data type: number
Format: integer
Uniqueness: unique
Null support: non-null
Insert trigger: none
Update trigger: not allowed
Delete trigger: ensure no child entities are present
NAME Data type: text
Length: 100
Uniqueness: non-unique
Null support: non-null
PASSWORD Data type: text
Length: 10
Uniqueness: non-unique
Null support: non-null
EMAIL Data type: text
Length: 50
Uniqueness: non-unique
Null support: non-null
114
The NAME field stores text and can accommodate names of up to 100 characters in length. To
enable different users with the same name to use the system, uniqueness is set to non-unique (i.e.
names can be duplicated). All users must supply a name (i.e. no null support).
The rules for the PASSWORD and EMAIL attributes ar e configured to be similar to those of the
NAME field. The only difference is the width of the records, which was limited to 10 characters
for PASSWORD and 50 characters for EMAIL. No triggers are needed for the NAME,
PASSWORD and EMAIL fields as they are depe ndent on the triggers defined on the primary
key (USER_ID). The domain and tr igger rules for entities LAND_USE,
LAND_REQUIREMENT, LAND_REQUIREMENT_RULE, LAND_PROPERTY,
LAND_UNIT and PROJECT are provided in Appendix D.
7.1.9 Combine user views
The ninth step in the logical data modelling methodology is the integration of user views to
eliminate redundancy and inconsistencies across views. Because each entity can be regarded as a
view, the main task during this phase of the design process is to combine entities belonging
together. In most cases, redundant entities have the same primary keys or have supertype-
subtype relationships. Since no such entities are present in the CLUES logical data model, no
further action was needed.
7.1.10 Integrate with existing data models
The logical data modelling methodology concludes with the integration of the designed model
with existing models. Often one database needs to be integrated with other databases, a process
which may require modifications to the logical data structure. Due to the spatial nature of land
units, the land unit entity was replaced by a separate spatial database containing the geometrical
and environmental properties of each land unit.
To perform a land suitability analysis, the item s (i.e. columns) of this so-called land unit
database are related to the records (i.e. rows) of the LAND_PROPERTY entity through the
LAND_PROPERTY_ID. To effect th is relationship, each item name in the land unit database
that stores a land property was modified to reflect the LAND_PROPERTY_ID. Since item
names cannot start with a number, a ?P? prefix wa s used to differentiate land properties from
other items. For instance, if the LAND_PROPER TY_ID for effective soil depth is 8, the
corresponding item in the land unit database was renamed to ?P8?.
Once all the entities, keys, attributes and busine ss rules had been defined, a logical data model
diagram (LDMD) was created (see Figure 7-1). A LDMD is a pictor ial representation of the
115
Figure 7-1 Logical data mode l diagram of CLUES knowledge base
logical data model that clarifies information relationships and enhances communication (Fleming
& Von Halle 1989). It documents the results of all the previous decisions (steps 1-9) in a single
diagram.
7.2 IMPLEMENTATION
The CLUES database was implemented in the Mi crosoft Access database management system
(DBMS). Microsoft Access was chosen for its simplicity and because of its user-friendliness.
The software is often used for small Internet databases because they are easy to set up and are
highly portable.
The translation of a logical data structure to a database involves the following four steps
(Fleming & Von Halle 1989):
1. Identify tables.
2. Identity columns.
116
3. Adapt data structure to the product environment.
4. Implement entity, relationship and attribute business rules.
The implementation posed few challenges as the main activities involved the creation of the
tables, adding the fields and setting the domain integrity rules. Most of the domain integrity rules
were related to setting appropriate data types (e.g. numerical, text and memo) for each field.
Although it is not expected that the user database will grow beyond a few megabytes, the data
and structure can easily be exported to a more robust DBMS, such as Microsoft SQL Server or
Oracle, should it be necessary. It was not necessa ry to implement queries or views in Microsoft
Access as interaction with the database is managed through the graphical user interface,
inference engine and web map service. The desi gn and implementation of these website elements
are discussed in the next chapter.
117
CHAPTER 8: DEVELOPMENT OF THE CLUES WEBSITE
A website is a location connected to the Internet that maintains one or more web pages (Oxford
English Dictionary 2008). It is a collection of web pages and related images, videos and other
digital media hosted on one or more web servers. This standard definition fits the CLUES
website which consists of a number of web pages, but CLUES differs from most websites owing
to its ability to dynamically (i.e. on request) produce maps, making it a web mapping application,
which is a special type of website. CLUES al so includes operations related to database
management and spatial analysis. The combinati on of all these functions necessitated a unique
approach to its design and development as an application.
As discussed in Section 4.2, the website component of CLUES comprises three elements,
namely the web map service (WMS), inference engi ne, and the graphical user interface (GUI). In
combination, these components not only contain the full functionality of CLUES but also act as
an interface between the two supporting databases described in the previous two chapters. This
chapter details the development and content of each of the CLUES website?s three elements,
starting with the inference engine.
8.1 THE INFERENCE ENGINE
The inference engine is the key element in the CLUES website as it performs the essential
function of calculating a land use suitability value for each land unit using the land requirement
rules in the knowledge base. The resulting suitability values are stored in a temporary field in the
land unit database, which is then used by the WMS to produce a suitability map. The following
two subsections outline the methodology of how the inference engine algorithm was developed
to calculate suitability and how it was implemented in CLUES.
8.1.1 Suitability calculation procedure
The procedure for calculating suitability values is encapsulated in Equation 8-1. Essentially, a
land unit?s overall suitability ( S) is determined by summarizing the product of each individual
land use requirement weighting (importance value) and suitability value.
?= iij swS Equation 8-1
where jS is the overall suitability value for land unit j ;
w i is the weight of land use requirement i ; and
si is the suitability value of land property i.
118
As set out in Section 2.4.1, each land use requirement consists of multiple rules and each rule is
related to a specific suitability level (S1, S2, S3 , N1, N2). To incorporate the suitability level
into the suitability value calculation, a suitability level factor of 1 to 5 was introduced (see Table
8-1).
Table 8-1 Suitability level factors
LEVEL CODE LEVEL DESCRIPTION RANGE
S1 Highly suitable > 2.5 Sj ? 3.5
S2 Moderately suitable > 1.5 Sj ? 2.5
S3 Marginally suitable > 0.5 Sj ? 1.5
N1 Unsuitable at present > 0 Sj ? 0.5
N2 Permanently unsuitable < 0
These factors are multiplied by the membership values of each rule using Equation 8-2, the
product being a suitability value for each land use requirement.
?= kki yls Equation 8-2
where is is the suitability value for land requirement i ;
lk is the suitability level factor of rule k ; and
yk is the membership value of rule k .
8.1.2 Membership value calculation
The membership value ( yk ) of a rule is defined by a function of the lower (?), central (?) and
upper (?) values specified for each rule (see Section 2.4.2). Because a rule can be either
symmetrical or asymmetrical, each rule needs to be deconstructed into one or two membership
functions that can be used to calculate a rule?s membership value. The steps taken to do so are
discussed below.
8.1.2.1 Represent rules as linear functions
The first step in determining a methodology for cal culating membership values was to represent
rules as linear functions. Although a curved function, such as the s-function, is often preferred in
fuzzy classifications, they are difficult to understand and are quite likely to be applied incorrectly
by users. Linear functions are less comp lex and easy to implement and visualize. Figure 8-1
illustrates the effect of using linear functions for the Boolean and fuzzy rules defined in Table
2-4. Compared to Figure 2-11 in which the same rules were implemented using the s-function, it
119
is clear from Figure 8-1 that the linear function produces a very similar classification and that the
differences in the resulting membership values between these two fuzzy sets are insignificant.
Figure 8-1 Levels of suitability of effective soil depths for perennial crops using linear fuzzy classification
For computational purposes, membership values for each rule are calculated by deconstructing
each rule into one or more lines. Figure 8-2a illustrates how a sy mmetrical fuzzy rule can be
defined using two lines, A and B. In this example, the lower value (?), central value (?), and
upper value (?) were set equal to 1, 4 and 7 respectively on a range of 0-10.
Figure 8-2 Symmetrical (a) and asymmetrical (b ) fuzzy rules deconstructed to two lines, A and B.
The use of two separate line functions for each rule allows asymmetrical functions to be defined
by specifying the central value to be off-centre. Such a rule is illustrated in Figure 8-2b, the
lower (?), central (?), and upper (?) values being set to 1, 4 and 10 respectively. For rules where
? = ? or ? = ? (such as rule 6 in Table 2-4), only one line is n eeded to represent a rule.
8.1.2.2 Determine membership function equations
Because rules are defined as linear functions, membership values can be calculated using the y-
axis formula for a line (Equation 8-3). In this formula, the value of x is the land property value of
a land unit, while y represents the membership function.
120
bmxy += Equation 8-3
where y is the y coordinate value for a point on a line;
m is slope of the line;
x is the x coordinate value for a point on a line; and
b is the y-axis intercept.
To solve Equation 8-3, values for m and b are required. Because the values for ?, ?, and ? and
their corresponding y coordinates (i.e. membership values) are known, these can be substituted
into Equation 8-3 to calculate b. By setting x = ? and y = 1 into Equation 8-3 the y-axis intercept
is:
?mb ?= 1 Equation 8-4
From Equations 8-3 and 8-4 it follows that
?mmxy ?+= 1
1+?= ?mmx
1)( +?= ?xm Equation 8-5
The remaining variable to solve in Equation 8-3 is m , which can be calculated using Equation
8-6.
ji
ji
xx
yy
m ?
?= Equation 8-6
where m is the slope of the line;
yi and yj are the y coordinates of any two points i and j on the line; and
x i and x j are the x coordinates of any two points i and j on the line.
A positive slope indicates that the value of y increases as x increases (i.e. an ascending function),
while a negative slope indicates that y decreases as x increases (i.e. a descending function).
Functions for line A and line B in Figure 8-2a can therefore be regarded as being positive and
negative respectively. Because the y values for ?, ?, and ? are known, they can be introduced into
Equation 8-6 to calculate the slopes of lines A and B using Equations 8-7 and 8-8 respectively.
?? ?=
1
Am ],[ ???x Equation 8-7
where m A is the slope of the line A;
121
? is the lower limit of the fuzzy function; and
? is the central value of the fuzzy function.
?? ??=
1
Bm ],[ ???x Equation 8-8
where m B is the slope of the line B;
? is the upper limit of the fuzzy function; and
? is the central value of the fuzzy function.
By substituting Equation 8-7 into Equation 8-5 it follows that, for line A, membership values can
be calculated using Equation 8-9.
1),,;( +?
?= ??
???? xxy A
],[ ???x Equation 8-9
Similarly, Equation 8-8 can be substituted into Equation 8-5 to produce Equation 8-10, which
can be used to calculate membership values for line B.
??
???? ?
??= xxy B 1),,;(
],[ ???x Equation 8-10
Linear functions can also be used to represen t Boolean rules by setting the y-axis intercept ( b)
equal to 1 and the slope ( m ) equal to 0 in Equation 8-3. By doing so, Equation 8-3 is reduced to:
1=Cy ],[ ???x Equation 8-11
where yC is the membership value for Boolean rules.
8.1.3 Suitability value calculation
The application of Equations 8-9, 8-10 or 8-11 to the land property value relating to land use
requirement k of land unit j , produces a membership value yk , ranging from 0 to 1. By using
Equation 8-2, this value is multip lied by the suitability factor (lk ) relating to a rule?s suitability
level and added to the products of all the other rules to derive a land use requirement suitability
value (ranging from 1 to 5). The suitability values of all the land use requirements are combined
using the weighted summation procedure in Equation 8-1.
The result of the inference procedure is a va lue ranging between 1 and 5 that represents the
average suitability of a land unit for a particular use. A land unit with a suitability value equal to
1 or less than 1 can be interpreted as being permanently unsuitable, while a value greater than 4
122
can be considered highly suitable. Similarly, the other suitability levels can be derived using the
ranges in Table 8-2.
Table 8-2 Interpretations of land unit suitability values
LEVEL CODE LEVEL DESCRIPTION RANGE
S1 Highly suitable > 4 Sj ? 5
S2 Moderately suitable > 3 Sj ? 4
S3 Marginally suitable > 2 Sj ? 3
N1 Unsuitable at present > 1 Sj ? 2
N2 Permanently unsuitable > 0 Sj ? 1
Because Equation 8-12 has an averaging effect on su itability values, an overall suitability value
of 3 can be attained for a land unit with suitabil ity values of 5 and 1 for equally weighted land
use requirements A and B respectively. This mean s that although a land unit is considered to be
permanently unsuitable in terms of property B, its suitability level is elevated to moderately
suitable by property A. In reality, however, the land unit should be considered permanently
unsuitable for the particular use because it is unsuitable in terms of at least one of its properties.
To ensure that the suitab ility value of a land unit found to be unsuitable in terms of any one of its
land properties is not promoted to above 2, the overall suitability values of all land units for
which any of its properties were found to be permanently unsuitable or unsuitable at present,
were reset equal to 1 and 2 respectively. This post-classification step is implemented
programmatically as discussed next.
8.1.4 Inference engine algorithm
Based on Equations 8-9, 8-10 and 8-11, the inference engine is implemented using the algorithm
set out in Figure 8-3. Upon logging in, each active user is allocated a temporary suitability item
(i.e. column) in the land unit database in which calculations can be carried out. By denying other
users access to this item, each user?s evaluations are protected from corruption. The suitability
item is automatically initialized (i .e. set all values to zero) by the system before each suitability
analysis. Initialization effectively erases the re sults of all previous evaluations for which the
suitability item was used.
The land unit database is accessed through an ope n database connection (ODBC), which is a
standard software programming interface to connect to DBMS. To establish an ODBC, the
ActiveX data objects (ADO) object database connec tion was used. This connection facilitates the
use of structured query language (SQL) to interr ogate and manipulate any database, irrespective
123
Figure 8-3 Inference engine algorithm
of the software in which it was created. By using this approach, the land unit database can easily
be replaced by another database using a different DBMS without affecting the inference engine.
In step 1 of the inference engine algorithm, an ODBC is established with the knowledge base to
determine which land use is to be evaluated. This information is stored in the PROJECT table of
the knowledge base and is retrieved using the current project?s identification number. Once the
land use is known, it?s identification number can be used to extract the related land use
requirements from the LAND_REQUIREMENT table in step 2.
Each land use requirement is sequentially considered to calculate its individual impact on the
overall suitability values of all land units. To do so , the suitability level and the type of rule as
well as the lower (?), central (?) and upper (?) values for each rule relating to the current land use
requirement, are retrieved from the knowledge base in step 3. Together with Equations 8-9, 8-10
and 8-11, this information is used to formulat e the SQL statements that sequentially update the
values stored in the suitability item of each land unit in step 4.
As mentioned above, the overall suitability values in the suitability item are overwritten each
time an evaluation is carried out. By not storing the evaluation results, the data storage and
LAND USE
REQUIREMENTS
Step 2
Step 3
Step 4
KNOWLEDGE
BASE
Select
Select
LAND USE
REQUIREMENT
RULES
LAND USE
ID
LAND UNIT
DATA BASE
Update
Repeat for all
land use
requirements
Repeat for all
land use
requirement
rules
KEY
RECORD SET
DATABASE
DATA ITEM SQL Query
Select
PROJECT
ID
Step 1
124
management problems associated with spatial analysis are overcome. Because all the parameters
of the evaluation are stored in the knowledge base, users can easily regenerate a previous
evaluation. If the evaluation and mapping process is fast enough, this ability creates the
impression that evaluation results are being stored.
The algorithm was implemented as a Visual Basi c procedure (named CalculateSuitability), for
which the code is provided in Appendix E.
8.2 THE WEB MAP SERVICE AND WEB SERVER
The ability to view the results of a suitability an alysis as a map was identified in Chapter 2 as a
principal requirement for CLUES. To enable user s to interact with the land unit database in a
spatial manner, a web map service (WMS) was implemented. The function of a WMS (see
Section 3.4) is to convert data stored in a GIS databa se into a format compatible with a standard
web browser. The CLUES WMS implementation involved two major tasks: software and
hardware selection; and WMS c onfiguration. These tasks are discussed in the following two
sections.
8.2.1 Choice of software and hardware
ArcIMS version 9.2 was chosen as the CLUES WM S software, mainly because it is currently the
best established WMS software avai lable and is quite likely to enj oy continued support. It is also
the WMS of choice for several governmental or ganizations in South Africa, including the
Provincial Government of the Western Cape (Van der Merwe 2008, pers com). By using
ArcIMS, CLUES will therefore be more likely to be adopted and maintained by the provincial
government once the system becomes operational. Scalability is another important reason for
choosing ArcIMS as this allows for future expansion of the system. It is possible to initially set
up the WMS on a small server and re place it later with a more powerful server without affecting
the rest of the system. The WMS can even be expanded to operate from multiple servers without
making any modifications to the other components (ESRI 2003).
ESRI (2007b) recommends that separate servers be used for the spatial database (i.e. land unit
database), WMS, and web server, especially when the number of concurrent users is expected to
be large. However, only one server, a DELL Po werEdge 2650 was available for this project and
was used for all the CLUES components. The Du al Intel Xeon 2.8GHz central processing unit
(CPU) of the PowerEdge 2650 is significantly s uperior to the minimum 1.3GHz CPU required
by ArcIMS (ESRI 2007b). As for memory, the 4GB random access memory (RAM) of the
PowerEdge exceeds the minimum requirements by 2GB. This configuration serves only for
demonstration purposes and, depending on demand, the machine could eventually be replaced by
125
a more powerful server, or even multiple servers. Due to the scalability of ArcIMS and the
loosely coupled design of CLUES, this can be done with little effort.
8.2.2 WMS configuration
The configuration of the CLUES WMS entailed setting up ArcIMS and configuring the web
server. An overview of ArcIMS is provided below, followed by an account of the steps taken to
set up the WMS.
8.2.2.1 ArcIMS overview
The five constituent ArcIMS components are th e web server, connector, application server,
spatial server, and the database. Users access ArcIMS services over the Internet or intranet using
alternative clients such as a web browser, ArcExplorer, ArcMap, ArcPad or mobile devices. The
relationships of the different components of ArcIMS are shown in Figure 8-4.
Figure 8-4 ArcIMS components
The spatial data served by ArcIMS is usually stored as ESRI shape files, ARC/INFO coverages,
geodatabases, images or rasters. Requests for this data are handled by the spatial server.
Essentially, the function of the spatial server is to bundle the data into the appropriate format
before sending the information to a client. Although other formats can also be accommodated,
most web mapping applications require data to be sent as images that can be viewed using a
standard web browser (ESRI 2007b).
The retrieval and conversion of data from the spa tial databases is the most processing-intensive
activity of the WMS. For large applications, Ar cIMS can be configured to include multiple
spatial servers running on separate computers to distribute the load (ESRI 2007b).
The images generated by the spatial server are se nt to the application server connector via the
application server. For configurations where mu ltiple spatial servers are used, the application
Client
Client
Client
Web server
Connector
Application server
Spatial Server
Database
Internet
Requests
Requests Responses
Responses
Adapted from ESRI (2007b)
126
server?s function is to allocate requests for data to the spatial server with the least load. This
ensures that requests for data are evenly distributed among the spatial servers (ESRI 2007b).
Requests for data are sent from users? web browsers to the web se rver where they are interpreted
by the application server connector. The defa ult ArcIMS connector, called Servlet Engine,
converts requests from the web server into ArcXML format, a special implementation of XML
(see Section 3.2.2), and then sends the formatted re quests to the application server for
processing.
By using the Servlet Engine, ArcIMS can be rapidly configured to distribute spatial data using
the preconfigured HTML and Java applications that are shipped with ArcIMS. Although these
multipurpose applications are easy to configure, their functionality and flexibility are limited.
Fortunately, ArcIMS includes three other connect ors, namely the Web application development
framework (ADF), ColdFusion and ActiveX c onnectors, which can be used to develop
customized client applications (ESRI 2007b).
The ADF connector supports web applications and services that can be accessed from multiple
GIS servers. Compatible applications and serv ices include ESRI products such as ArcIMS,
ArcGIS Server and ArcWeb Services as well as standard WMS services. Applications using the
ADF connector can be developed in either th e Microsoft .NET framework or on the Java
platform (ESRI 2007b).
Another popular web-developing environment is ColdFusion. Applications that are developed
using ColdFusion markup language (CFML) can c onnect to the ArcIMS application server
through the ColdFusion connector which provides se veral tags to formulate requests for map
data from the spatial server (ESRI 2007b).
Developers can also create highly customized web mapping appli cations using Microsoft Visual
Basic and active server pages (ASP) programming languages (ESRI 2007b). Such applications
can send requests to the ActiveX connector by us ing more than 30 predefined component object
model (COM) objects. These COM objects can be us ed in conjunction with Microsoft?s ActiveX
data objects (ADO) to connect to existing da tabases. With ADO, developers can write
applications that access databases without any knowledge of how the database was implemented.
ADO therefore allows applic ations to be independent of the database, meaning that a database
can be replaced by another without affecting the application.
Once requests for data have been interpreted by the connector and the maps have been produced
by the spatial server, the resulting images are placed on a web server for downloading. ArcIMS
supports a range of web servers of which Apache and Microsoft IIS are the most popular.
127
Depending on the number of requests, maps can either be placed on a dedicated web server or
they could be stored with other web content on an existing server.
From the above discussion it is clear that there is a range of possible configurations of ArcIMS
and that each application requires a unique approach. The configuration of ArcIMS for the
development of the CLUES WMS is de scribed in the next section.
8.2.2.2 ArcIMS and web server configuration
Due to the level of customization needed to develop CLUES, the ActiveX connector and ASP
were used for developing the CL UES WMS. Seeing that the ActiveX connector is only available
on Microsoft Windows platforms, the DELL PowerEdge 2650 was loaded with Microsoft
Windows Server 2003. This was followed by the in stallation of the ArcIMS 9.2 software and
licenses which were obtained from Stellenbosch University.
Because the load on the web server is unlikely to be high, a single web server was set up to
support both the WMS and the GUI. Although ArcIMS is compatible with most web server
software, Microsoft?s internet information serv ices (IIS) was chosen for CLUES implementation
because it is the only web server software that supports the ArcIMS ActiveX connector. ArcIMS
was set up to store the maps produced by the spatial server in a virtual directory called Output,
which has a physical path of C:\ArcIMS\Output . Although the maps are separated from the GUI
web pages, they are combined by the user?s we b browsers to appear in the map viewer (see
Section 8.3.7)
The appearance of the maps produced by ArcIMS is determined by a map configuration file,
structured in ArcXML format. The map confi guration file for the CLUES application is
appended (see Appendix F). The file is separated into two main sections, namely environment
and map. The environment section is used to set operational parameters that are not related to the
maps, like the country from where the maps are being served and the language and fonts used by
the system.
The settings directly related to the maps bei ng produced by ArcIMS are defined under the MAP
section. The first setting defines the units of the coordinate system in which the data is stored.
Because the data is stored in the UTM Zone 34 South coordinate system, the unit parameter was
set to metres using the MAPUNITS elemen t under the PROPERTIES subsection. Another
important setting that needs to be defined in the PROPERTIES subsection is the initial extent of
the map or the area that will be visible when a user opens the application for the first time. By
using the ENVELOPE element, this was se t to the limits of the Western Cape (see Figure 8-16).
128
The next subsection in the CLUES map configurat ion file is called WORKSPACES and is used
to define the location of the data sets and descriptions of each layer shown on a map. Although
all the data is stored in a singe data folder on the server (path e:\clues\d ata), separate workspaces
had to be specified for shape files and images as the SHAPEWORKSPACE and
IMAGEWORKSPACE elements respectively. These workspaces are referenced by the
definitions of each layer, which are specified using the LAYER element.
Four layers were specified. The fi rst layer, called Suitability, is used to display the result of a
suitability analysis and relates to the land unit database. To map suitability, the values in the S1
field of the land unit database are symbolized according to five suitability levels (i.e.
permanently unsuitable, unsuitable at present, marginally suitable, moderately suitable and
highly suitable) using the values specified in Table 8-2. To set the co lours for each of these
classes, the SIMPLEPOLYGONSYM BOL element is employed.
The result of an optimal land use analysis is displayed in the second or Land Use layer. The
configuration of this layer is similar to that of the Suitability layer, the major difference being
that the Land Use layer is set to display a nomin al data set representing up to 25 land uses, as
stored in the L1 field of the land unit database.
The two remaining layers are unrelated to any an alysis and are included mainly for orientation
purposes. Two image layers, representing a satel lite image and a hillshade respectively, are
specified. Both layers are set to be slightly transparent using the IMAGEPROPERTIES element.
Once completed, the map configuration file was us ed to create an image WMS with the ArcIMS
Administrator tool. Because the map configuration file includes a full-colour satellite image for
orientation purposes, the JPEG f ile format was chosen for the output map format. This ensures
that the full range of colours in the satellite image is accommodated. A compression level of
10% ensures that the images remain small eno ugh for transfer over the Internet. Once created,
the CLUES WMS was started using the Services Manager tool.
As explained above, the WMS is configured to receive requests formulated to use a range of
predefined ActiveX controls. Since ActiveX c ontrols are particularly user-unfriendly a
mechanism was needed to enable users to generate requests through a user-friendly interface.
The development of this facility is recounted next.
8.3 Development of a user-friendly GUI
The graphical user interface (GUI) was developed as a user-friendly interface to the WMS, the
inference engine and the knowledge base (see Figure 4-2). The function of the GUI is to lead
users through the suitability analysis procedure (see Figure 8-5).
129
Figure 8-5 Main steps followed to produce a suitability map using CLUES
The process starts by specifying the land use that will be evaluated, followed by the defining of
the land requirements and their related rules. Land properties, stored as attributes to each land
unit in the land unit database, are then evaluated against the land requirements and used to
calculate land unit suitability for the chosen land use. Finally, the result of the evaluation is
presented to the user as a suitability map. The procedure can be repeated for multiple land uses
and changes to the land requirements can be interactively evaluated and mapped in order to
facilitate scenario building. For a multi-object ive solution, an optimal land use map can be
produced if the requirements of more than one land use are specified.
The steps shown in Figure 8-5 closely resemble steps 3, 4, 7 and 8 of the land evaluation process
(compare Figure 2-1). The main difference between th e two procedures is that in the CLUES
process the data is already available and the user simply selects the appropriate land properties
from a set of available data sets. Recall that in the land evaluation process the identification of
land uses is preceded by a data collection and preparation process. By eliminating this process,
users of CLUES can start with suitability assessme nts without the need to collect or prepare any
data.
The next section describes how the GUI was structur ed to facilitate suitability analysis. This is
followed by detailed descriptions of the six main GUI modules. The web pages that comprise
each module is provided in digital format in Appendix G (a Compact Disk).
8.3.1 GUI structure
The GUI, shown diagrammatically in Figure 8-6, consists of 34 we b pages (each represented by
a rectangle) written in HTML, JavaScript and Vi sual Basic code. Each of these elements is
required for any of the GUI web pages to f unction properly. As explained in Section 3.2.2,
HTML code is used to format the page struct ure and visual appearance, while the server-side
scripts or active server pages (ASP) written in Visual Basic, generate dynamic HTML code
based on the status of the suitability analysis and the user (see Section 3.2.4). The resulting
130
HTML is placed on the web server for downloading. JavaScript is used as client-side scripting
(see Section 3.2.3) to improve interactivity.
Figure 8-7 demonstrates how the three progra mming elements (i.e. HTML, Visual Basic and
JavaScript) interact in a page (the main page is used as example) and highlights each component
as a different colour. The three components ar e highly integrated within each page, often
interacting and exchanging data. To reduce redundancy, code used by more than one page is
placed in central libraries. These libraries are named lib.asp and lib.js for the Visual Basic and
JavaScript codes respectively. Refer to Appendix G (a CD) for the content of each of these files.
In Figure 8-6 it is clear that the structure of th e GUI and the linkages between pages resembles
the sequence of operations shown in Figure 8-5. The GUI can be di vided into six main modules:
login & security ; menu ; user details ; projects ; analyse & map ; and rulebase. Each module
contains one or more pages, some of which are visible and others hidden. The pages that are not
visible are intermediate pages that carry out operations not associated with input or output
(display).
The login & security module acts as gatekeeper to the system and prevents unregistered users
from entering. Once registered, users can log into the system using the password specified during
registration. This will open the menu module, containing a single page called main. The menu
acts as the hub to all the functionality of the system, directing users to the user details , projects ,
and rulebase modules.
In the user details module, users can change the information entered during registration. This is
useful for users wanting to change their passwords or to update their email addresses.
Before an analysis can be carried out, the user must create a set of rules for each of the land uses
that will be evaluated. A land use can have multiple requirements and each requirement consists
of one or more rules. The rulebase module allows users to add, edit and delete individual land
uses, requirements, and rules.
The projects module allows users to edit and delete existing projects or to create new projects. A
project keeps record of the status and characteris tics of an evaluation so that users can return to
an evaluation at a later stage. Users can also work on several projects simultaneously.
Once created, a project can be opene d as an interactive map through the analyse & map module.
This module encompasses the inference engine a nd WMS. If an evaluation has been carried out
previously under the opened projec t, the result of that analysis is displayed. If no previous
analysis is associated with a project, a new ev aluation based on the land uses and rules in the
rulebase can be initiated from within the mapping environment.
131
Figure 8-6 GUI pages and linkages in CLUES
132
Figure 8-7 Code of the main page showing interaction between HTML, JavaScript and Visual Basic elements
In the following sections, the web pages comprising each of the six modules are described in
more detail. The descriptions focus on the functions and elements of each page. No details
regarding the coding are discussed and readers are referred to the full code of each page, as
provided in Appendix G.
<%@ LANGUAGE=VBScript %>
<% Option Explicit %>
<%
if (NOT FromSite) then Response.Redirect("index.asp") end if
if (Session("theUserID") = "") then Response.Redirect("logout_inactive.asp") end if
%>
CLUES
<%
if (Session("Error") = "PRE-2") then Response.Write("")
else Response.Write("") end if
Session("Error") = ""
Dim Items
if (Session("theProjectID") <> 0) then
Items = Array("projects", "rulebase", "analysis", "help", "logout")
else
Items = Array("projects", "rulebase", "help", "logout")
end if
displaymenu "./","_self",Items
insertspace(4)
Response.Write("
<% Activity %>
KEY
HTML
JavaScript
Visual Basic
133
8.3.2 Login & security module
The login & security section consists of five web pages. Upon accessing the website, users are
taken by default to an index page, which acts as a doorway to the system. The index page (see
Figure 8-8) contains a banner, welcoming mess age and login form. The banner is a standard
element on all the GUI pages and is meant to provide a uniform identity, which is considered to
be an important attribute of web applications (U.S. Department of Health and Human Services
2006). The welcoming message introduces the user to the system and provides instructions on
how to log on. The form contains two fields where existing users can enter their username and
password respectively.
Figure 8-8 Index page of the login & security module
Upon clicking on the login button, the information in the form is sent to the login page. The
function of this page is to check whether the entered username and password combination exists
in the USER table in the knowledge base. If no username and password match is found in the
database, the user is redirected to the index page and no further action is taken by the login page.
If a matching record is identified, users are directed to the main page.
The welcoming message on the index page also informs users that those without CLUES
accounts can register by using the provided link. The link opens the register page (see Figure
8-9) on which unregistered users can create a new account. In addition to the standard header, the
register page contains a form consisting of five fields: username, full name, password, confirm
password and email address. All these fields in the form correspond to those of the USER table
in the knowledge base.
134
Figure 8-9 Register page of the login & security module
Once entered into the form, the new user?s information is sent to the register_update page which
stores the field values into the USER table. The values are, however, first compared with those in
the USER table to ensure that the username and email address do not already occur. If one or
both do exist, the user is redirected to the register page and an appropriate error message is
displayed. If no other records are found with the same username and email address, the user is
taken to the register_success page which informs the user that the registration process was
successful (Figure 8-10). The user is also provided with a link to the index page in order to log
on using the newly created account details.
Figure 8-10 Register_success page of the login & security module
135
8.3.3 Menu module
Upon successful completion of the login procedure, users are taken to the main page containing
the main menu (Figure 8-11). The main menu is a list of hyperlinks which direct the user to the
remaining three GUI modules, namely User details, Rulebase and Projects . The user can exit the
system by clicking on the Logout menu item. This will invoke the logout page, which clears all
the session variables and redirects the user to the index page.
Figure 8-11 Main and banner menus as shown on the main page
Some of the items in the main menu are duplicated in a secondary menu, displayed on the
banner. The banner menu is shown on most pages so that users can access these items directly
without having to return to the main menu. A help item is also provided, which opens an Adobe
Acrobat portable document format (PDF) file c ontaining helpful information about the system.
8.3.4 User details module
The user details module contains two pages namely user_details and user_update . The user_
details page is almost identical to the register page as it also presents a form for users to update
their details. The only difference between the user_details page and the register page is that the
current user?s details are preloaded on the former from the database so that users can view and
edit their details. Upon submission of the form, the user_update page stores the updated data in
the USER table in the knowledge base.
8.3.5 Rulebase module
The rulebase module is the most complex of all the GUI modules as it comprises 17 web pages
and more than 1500 lines of code. To structure the description of these pages, the module is
136
subdivided into three sections, namely those dealing with specification of land uses, land use
requirements, and land requirement rules. These subdivisions relate to the landuses , req and
rules pages respectively (see Figure 8-6 for system location) and are described separately below.
8.3.5.1 Land uses specification
The first page shown upon entering the rulebase is the landuses page (Figure 8-12). This page
relates to the LAND_USE table in the knowledge base and lists all the land uses owned by the
current user.
Figure 8-12 A list of land uses owned by the current user shown on the landuses page
Two items are available under the OPTIONS column for each land use. By choosing the edit
( ) option, the landuse_edit page is opened. In addition to the information shown in the landuse
page, this page contains a form through which the user can edit the current land use name. Once
submitted, the information in the form is sent to the landuse_update page where it is stored in the
LAND_USE table of the knowle dge base. The delete ( ) item in the OPTIONS column opens
the landuse_delete page, which simply removes a land use from the LAND_USE table and the
related rows of the LAND_REQUIREMENT and LAND_REQUIREMENT_RULE tables.
Users can create new land uses by using the CREATE LAND USE link. This opens the
landuse_create page, which allows the user to enter the name of a new land use into a form.
When submitted, the landuse_update page is opened inserting a new record into the LAND_USE
table and setting the NAME field to the value entered.
In addition to a link back to the main menu, a function by which users can import land uses is
also provided. The IMPORT LAND USE link opens the landuse_import page, which displays a
137
list of land uses that are stored in the LAND_USE table and owned by the Administrator
(USER_ID = 1). Once selected, the landuse_import_update page is activated to create a
duplicate row in the LAND_USE table of the know ledge base and to change the ownership of
the duplicate land use to that of the current user. The duplication process is repeated for all the
land requirements and rules related to the specific land use.
8.3.5.2 Land use requirements specification
Each of the land use names in the land use list hyperlink to the req page which contains the land
use requirements for a particular land use (see Figure 8-13). Users can add requirements by
clicking on the ADD REQUIREMENT link on the page. This opens the req_add page which lets
the user select a requirement from a list of available land use properties in the
LAND_PROPERTY table of the knowledge base. Th e selected property is added to the current
land use requirements list by the req_add_update page.
Figure 8-13 Requirements for land use A as listed on the req page
The relative importance of each requirement is shown in the WEIGHT (%) column of the land
requirements list. These weights can be edited directly and saved by clicking on the SAVE
function. The list also includes an OPTIONS column enabling users to edit or delete any of the
listed land use requirements. These two items open the rules and req_delete pages respectively.
The req_delete page is similar to the landuse_delete page in that it simply removes the relevant
rows from the LAND_REQUIREMENT and L AND_REQUIREMENT_RULE tables. To open
the relevant rules associated with the listed requirements, the user can either click on the edit
option under the OPTIONS column or simply clic k on the property name. At any time users can
138
select the QUIT link to exit the page. Exiting will discard any changes made to the land use
requirements.
8.3.5.3 Land use requirement rules specification
The rules for each land use requirement are shown on the rules page. Figure 8-14 lists six rules
defining the hypothetical Land Use A. The requirement includes one symmetrical, fuzzy rule for
the highly suitable, moderately suitable, marginally suitable and unsuitable at present suitability
categories, while the permanently unsuitable suitability level is defined by two rules. The first of
these is an asymmetrical fuzzy rule ranging from 60% to 80%, while the second constitutes a
Boolean rule specifying that all values greater than 80 are considered permanently unsuitable.
Although the lower threshold of the sixth rule is equal to the upper threshold of the fifth rule, the
value stored in the latter is interpreted by the inference engine as ?less than? 80. The two values
are therefore mutually exclusive.
Figure 8-14 Requirement rules for a hypothetical land use as displayed on the rules page
The six rules listed in the table are also illustrated diagrammatically so that users can visualize
the specified thresholds and fuzzy functions. The diagram was challenging to implement and
considerable time was invested in its development. Although the diagram is functional and is
rendered quickly, it has a number of limitations. Probably the most noticeable is the opaqueness
of the shapes representing the different suitability levels, causing some rules to be partly
obscured by others. Each fuzzy function is composed of one or more images and because current
web browsers are unable to render transparent images, this shortcoming could not be overcome.
To compensate for this limitation, the order in which the rules are drawn in the diagram can be
manipulated by moving the mouse over the property name in the table. Another problem with
139
current technology is that only rectangular and triangular shapes could be used to emulate the
forms of the fuzzy functions. Functions that curv e, such as s-functions, can not be depicted.
Rules can be deleted and edited in the same way described for requirements and land uses. New
rules can be added by using the ADD RULE functi on on the page or the user can return to the
req page by clicking on the BACK TO REQUIREMENTS function.
8.3.6 Projects module
Users can store the results of a particular suitability analysis done in a project. A user?s projects
can be viewed on the projects page (see Figure 8-15). The page lists all the projects along with
the date and precise time a project was last mo dified. Users can also edit and delete projects
using the tools in the OPTIONS column and new projects can be cr eated by using the NEW
PROJECT function.
Figure 8-15 Information about a user?s projects shown on the projects page
As mentioned in Section 8.3.3, each GUI page includes a banner with a menu that acts as a short
cut to the modules. The banner menu is customized for each page to include relevant items only.
While some items, such as RULEBASE and HELP, are standard on all pages allowing users to
open the rulebase and online manual at any time, other items are only present under certain
conditions. For instance, the ANALYSIS item which opens the analyse & map module is only
available if a project has been ope ned during the current session. The analyse & map module can
also be opened by clicking on a project name in the NAME list.
140
8.3.7 Analyse & map module
The analyse & map module consists of one page which acts as an interactive map viewer. In
reality, the viewer consists of many pages located in frames within the viewer page (called
map/index). A frame is an HTML element that facilitates the inclusion of web pages in an
existing web page. Frames are of ten used for web mapping applications to emulate GIS viewers.
Such viewers usually contain a map area, where the spatial information is rendered, and a menu
allowing users to select tools to interact with the map (e.g. zoom and pan). Many applications
also include a table of contents (TOC) listing av ailable data layers. Users can use the TOC to
select the layers to be displayed. When include d in web mapping applications, such elements are
usually implemented as separate pages loaded into predefined frames. The advantage of this
approach is that only certain frames need to be refreshed once an action is taken.
Five web pages and frames were used to de velop an integrated analysis and mapping
environment for CLUES. The first frame, called top, holds the banner and spans the entire width
of the viewer (see Figure 8-16). As with the other GUI pa ges, the banner includes menu items
that can be used to quickly access other functions. In the mapping environment, this feature is
especially useful when a land use requirement rule needs to be edited and re-evaluated.
Figure 8-16 CLUES map viewer
The menu frame, positioned directly below the banner, was implemented to accommodate the
analysis menu. The analysis menu contains tw o drop-down lists and a button. From these two
lists the user can select the type of analysis and land use to be evaluated respectively. Although
only one analysis function is currently available, it is expected that more functions will be added
141
as the system is expanded. The land use drop-down list contains all the land uses in the user?s
rulebase and the GO button initiates the analysis.
The frame containing the map information is called map and is, at 565x330 pixels, the largest of
all the frames. In addition to the map data, the map frame includes a north arrow and a line scale.
The map frame can be manipulated by the range of tools in the tool frame, which is positioned
directly below the map frame. Table 8-3 describes the functions of each of the available tools.
The tool frame includes a map scale, which can be edited by the user to change the scale of the
displayed map. If any of the land units on the map are currently selected, a report can be opened
and printed using the REPORT function, whil e the LEGEND link opens a key for the current
map. To ensure that users are informed about th e status of any action, a status indicator is
provided to the right of the tool frame.
Table 8-3 Tools available for manipulating the map frame
Tool Description
Zoom out using mouse
Zoom in using mouse
Decrease scale by a factor of two
Increase scale by a factor of two
Zoom to the Western Cape
Pan using mouse
Pan up
Pan down
Pan left
Pan right
Show information about a particular land unit using the mouse
Select land units by drawing a rectangular area and show information about all the selected land units
Build a query
Deselect all land units
Show a legend
142
The layers frame is directly above the status indicator and lists the layers currently being
displayed. In addition, this frame allows users to manipulate the visibility of layers by switching
layers on or off. A refresh map button is included to effect any changes made to the layers list.
Although the frames described here are all individual web pages, they are integrated through
programming to work as a unit. Many actions carried out in one frame will cause a reaction in
another frame. For more details about the implementation of the analyse & map module the
reader is referred to the source code in Appendix G, which includes detailed comments and
explanations.
This chapter described the development of the CLUES website, consisting of a GUI, inference
engine and WMS. The GUI enables users to stor e a set of land requirements and rules in the
CLUES knowledge base. During a suitability analysis , the knowledge base is interrogated by the
inference engine to calculate land use suitability values for each land unit in the land unit
database. The suitability values are then used to dynamically produce suitability maps which are
converted by the WMS to a format that is viewable using the GUI. The GUI, inference engine
and WMS form a coherent unit an d enable users to carry out suitability analyses in a user-
friendly manner demonstrated in the next chapter.
143
CHAPTER 9: DEMONSTRATIONS OF CLUES
CLUES was developed to demonstrate how web technology can be used to deliver spatial
analysis and modelling capabilities to Internet users. The resulting software product is a working
web-based land evaluation system for carrying out la nd suitability analysis in the Western Cape.
Land evaluation was the chosen application field because it strongly relies on geographical data
and spatial analysis. In addition, there exist a need for a system that will support decisions
regarding the optimal use of the Western Cape ?s land resources. Sound land use planning is
required to ensure that the Western Cape?s land and environmental resources are used
sustainably.
To support decisions about the optimal use of la nd, this chapter demonstrates how CLUES can
be used to generate suitability maps for perennial crops. Perennial crops constitute a major land
use (see Section 2.1.3) encompassing many types of fruit and vines, but exclude annual
commodities such as vegetables and wheat. Annual crops are excluded in this study because their
cultivation practices and requirements differ markedly from those of perennial crops (Reiger
2006). Although the requirements specifi ed in this chapter pertain to the main perennial crops
produced in the Western Cape, namely deciduous fru its, vines, citrus and olives, it can be easily
modified to include other crops.
To perform suitability analyses using CLUES, users are required to define the requirements of
each land use to be assessed. Land use requir ements are defined by setting land requirement
rules, consisting of suitability levels, functions and thresholds for each land property considered.
Because land use requirements seldom have equal importance, the relative importance of
individual land use requirements can be specified by allocating a weight to each requirement. To
conclude the suitability analysis procedure, a suitability map is produced showing the different
levels of suitability for a particular area.
The following sections demonstrate how CLUES can be used to produce suitability maps for
perennial crops. The defined rule s are based on the specific crop requirements recorded in the
literature. Only those fundamental land propert ies available in the land unit database were
considered for the requirement definitions. The steps taken in the suitability analysis are
illustrated using computer screen captures of the system?s interface. Some maps are however
supplemented with locational information (e.g. town names) for orientation purposes and do not
appear on screen. Suitability maps are also provided with a standard legend which CLUES
displays as a separate window.
144
9.1 SETTING RULES FOR PERENNIAL CROPS
In order to set rules for perennial crop suitability, a new land use needs to be created in the
knowledge base. To do so, the rulebase was opened from the main menu ( Figure 8-11) to access
the land uses page (Figure 8-12). A new land use called Perennial Crops was created using the
CREATE LAND USE link. This link opens the landuse_create page allowing one to enter the
name of a new land use (Figure 9-1).
Figure 9-1 The landuse_create page used to create a new land use called ?perennial crops?
Once the new land use has been created, the knowledge base is prepared to accommodate land
use requirements related to perennial crops. Th e steps taken to set up the necessary rules are
discussed in the following sections.
9.1.1 Terrain requirement rules
Terrain has both direct and indirect influences on most agricultural land uses because it relates to
climate and soil (see Section 4.1.3.4). The terrain-related factor s considered for perennial crops
are slope gradient, aspect, curvature and elevation.
9.1.1.1 Slope gradient rules
According to the regulations of the Conservation of Agricultural Resources Act no. 43 of 1983
(South Africa 1984), no land with slopes steeper than 20% may be ploughed without permission
from the appropriate authorities. Land units with slopes of more than 20% should therefore not
be considered for perennial crops and hence are excluded from further analysis. In general,
145
slopes with lower gradients are more suitable for perennial crop production, simply because it is
easier to work the land on gentle slopes and because soils on steeper terrain are mostly shallower
and dryer than those in areas that are gently inclined.
To install a rule that will exclude land units with slopes of more than 20% in CLUES, the
Perennial Crops requirements page was opened by clicking on the appropriate link on the
landuses page shown in Figure 9-1. The next step was to add a new item to the requirement list
by selecting the ADD REQUIREMENT link. This opens the req_add page which contains a
drop-down list of all the land prope rties in the knowledge base (see Figure 9-2). By referring to
this list a user can select properties from a range of land properties to be included in a suitability
analysis. To define a rule for slopes, the Slope gradient (%) item was selected.
Figure 9-2 Slope gradient (%) added as a requirement for perennial crops
Once created, the new rule was added to th e slope land use requirement by opening the
rules_add page using the ADD RULE link on the rules page. As explained in Section 8.1.1,
when a land unit is classified as being permanently unsuitable for a particular land property, it
cannot be promoted to a higher suitability level, no matter how well it performs any of its other
land properties. A rule with a suitability level of permanently unsuitable is therefore regarded as
a constraint, while suitability levels other than permanently unsuitable are considered as factors.
Factors are rules that enhance or detract from a land use?s overall suitability, while constraints
are meant to limit or exclude areas for consideration (Malczewski 1999).
146
To specify a constraint in terms of slope gradient it was appropriate to specify a Boolean rule
with a suitability level of permanently unsuitable. A Boolean rule was chosen because it makes a
hard distinction between land units on slopes of more than 20% and those that are not. To create
the rule, the lower threshold for permanently unsuitable was set to 20%, while the upper
threshold was set to the maximum slope value that occurs in the land unit database (i.e. 207%).
The diagram for the rule is shown in Figure 9-3. Although the middle value (i.e. 113.5%) is not
used in Boolean rules, it is automatically calculated by the system in case the membership
function is changed.
Figure 9-3 Rules defining suitability levels for slope gradient
Two additional fuzzy rules, ranging from 0% to 20%, were set using ascending and descending
fuzzy functions (see Section 8.1.2.2) to represent highly suitable and moderately suitable values
respectively. In the highly suitable rule, the lower value was set to 0% while the upper value was
set to 20%. To indicate that membership should d ecrease from 0% to 20%, the middle value is
also set equal to the lower value (0%). A similar approach was taken with the ascending
moderately suitable rule, where the middle value was set equal to the upper value (20%).
Although Boolean rules ranging from 0% to 10% and 10% to 20% could have been defined to
represent the highly suitable and moderately suitable categories, the combined effect of the two
fuzzy rules ensures a smooth transition between these categories. This is illustrated in Table 9-1
which lists the suitability values and levels for selected slope gradients. Values greater than 0 are
used for factors (i.e. highly suitable, moderately suitable and marginally suitable), while values
of less than 0 are used for constraints (i.e. permanently unsuitable and unsuitable at present).
Refer to Section 8.1 for details of suitability value calculations.
147
Table 9-1 Calculated suitability values and levels for selected slope values
SLOPE (%) SUITABILITY VALUE SUITABILITY LEVEL
0 3.0 Highly suitable
5 2.75 Highly suitable
10 2.5 Highly suitable
15 2.25 Moderately suitable
19.9 2.005 Moderately suitable
20 -2 Permanently unsuitable
To view the spatial implications of the Slope gradient (%) rules, a project was created in the
projects page (see Figure 8-15). A project is es sentially a virtual container that keeps record of
the actions during land suitability mapping. This enab les the storage of a user?s activities so that
a project can be reopened later. The mapping e nvironment is opened by clicking on the newly
created project in the projects ta ble. When a new project is opene d, a satellite image map of the
Western Cape is presented ( Figure 8-16). Because suitability an alysis results are difficult to
visualize at the scale of Figure 8-16 (i.e. 1:4 000 000), the mapping of the individual
requirements is demonstrated on a smaller area of interest (AOI) in the greater Cape Town
region (Figure 9-4). By using the zoom tool the scal e and extent of the mapped area was changed
to include Table Mountain (south of Cape Town) in the west and the Hottentots Holland
Mountains (north-east of Some rset West) in the east.
Figure 9-4 Satellite image of the AOI
Stellenbosch
Pniel
Cape Town
Mitchells Plain
Wynberg
Somerset West
Bellville
148
The suitability analysis for perennial crops based on the slope requirement was initiated by
selecting Perennial Crops from the drop-down la nd use list in the menu frame above the map and
by clicking on the ANALYSE button. Th e resulting slope requirement map (Figure 9-5)
indicates that most of the AOI is highly suitable in terms of slope and that the mountainous areas
were identified as being permanently unsuitable. Only a narrow band of hillslopes was classified
as being moderately suitable.
Figure 9-5 Slope requirement map of the AOI
The legend in Figure 9-5 is a key to the colours used for each suitability level. The standard
legend is taken from a separate window that appears when the LEGEND link (bottom-right of
map window) is opened and some categories may therefore be absent in a given analysis image.
For orientation purposes, the user can at any time toggle between the Suitability and Satimage
layers in the layers list (to the right of the map) to enable or disable the display of the suitability
result. If both the Suitability and Satimage layers are active, the suitability result overlays the
satellite image, as shown in Figure 9-6.
9.1.1.2 Aspect, curvature and elevation
Although data for slope aspect, curvature and elevation is available in the land unit database, it
was not used in the suitability analysis. More research is needed to determine the influence of
these land properties on perennial crops. It is generally accepted that slope aspect, in
combination with slope gradient, does play a role in the quality of perennial crops owing to its
influence on the amount of sunlight an area receives (Carey 2005). In the southern hemisphere,
moderately inclined northern slopes receive more solar radiation than southern or gently inclined
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
149
slopes (Buckle 1996). Consequently, temperatures on northern slopes are generally higher where
they may influence soil moisture and formation. In the relatively warm climate of the Western
Cape, aspect has little impact on the minimum requirements of land for the establishment of
perennial crops (Saayman 1981). However, slope as pect should be considered in suitability
analyses of specific crops or cultivars (Carey 2005).
Concerning curvature, it is well known that day-night temperatur e differentials for land units in
convex landscape positions are generally more stable and therefore less likely to be affected by
frost (Buckle 1996). The soils on convex land com ponents are dryer and better drained due to the
divergence of runoff and groundwater flow from such areas. As with slope aspect, gradient has
an indirect effect on crop suitability as it will affect soil formation over time. However, too little
is known about the direct effects of curvature on perennial crops to set general suitability rules.
Figure 9-6 Suitability results overlaying a satellite image for orientation purposes
Elevation has a direct effect on climate because temperature decreases on average by about
0.3?C in South Africa with every 100m above sea level (Saayman 1981). Land units on hills and
mountains are therefore generally cooler than those in low-ly ing areas. No explicit rule to
incorporate the effect of elevation on temperature was implemented in the rulebase because it is
inherent in other land properties such as chill and heat units. The rules set for the climatic
requirements of perennial crops are discussed next.
9.1.2 Climate requirement rules
Climate is the most important environmental variable affecting the production of fruit crops on a
regional scale (Jackson 1999). Four climatic attri butes relating to temperature and rainfall were
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
150
considered in the suitability analysis of perennial crops. They are chi ll units, heat units, frost
occurrence and mean annual rainfall.
9.1.2.1 Chill units rules
The cumulative number of hours that plants are e xposed to temperatures ranging from 2.4?C to
9.1?C during winter is one of the most important climatic factors to consider for perennial crops.
Most deciduous plants require a minimum number of chill units (CU) ? measured in the Western
Cape as hours from leaf-drop in May until September ? to satisfy dormancy, to stimulate growth,
develop leaves, flower and set fruit. Failure to produce the re quired CU in areas with warm
winters will prevent or reduce budding, resulting in poor crops (Reiger 2006; Schulze 1997).
Most temperate fruit crops such as pome fruit (apples and pears) require at least 1000CU each
winter (Reiger 2006), whereas peaches and apricots require between 806CU and 925CU
(Valentini et al. 2004).
Due to the wide range of the chilling requirements of different varieties of perennial crops as
well as the influence of microclimate on CU (Jackson 1999), no constraining chilling
requirements were defined for CLUES. However, ar eas with less than 40 0 hours of chilling were
considered marginally unsuitable for perennial crops, while land units with values of more than
1000CU were considered to be highly suitable for perennial crops. As in the case of slope
gradient, fuzzy functions were used to facilitate gradual transitions between chill suitability
levels (see Figure 9-7).
Figure 9-7 Chill units rules
Although no category was specified for moderately suitable, the overlapping nature of the two
fuzzy rules, representing marginally suitable and highly suitable land units respectively,
151
inherently encompasses this intermediate category. The inference engine allocates all land units
with CU values ranging from 550 to 850 to this clas s. The spatial effect of the two fuzzy rules is
illustrated in Figure 9-8.
It is clear from this requirement map that the chill factor has a strong relationship with elevation
as most of the areas identified as being moderately or highly suitable are higher-lying areas,
while the coastal and low-lying ar eas of the Cape Flats (i.e. around Mitchells Plain) are allocated
to the marginally suitable category.
Figure 9-8 Chill units requirement map of the AOI
9.1.2.2 Heat units rules
Once the chilling requirement of a temperate woody plant has been satisfied, it must receive a
certain number of heat units (HU) or growing degree-days (GDD) in order to resume growth.
This is especially important fo r perennial crops as they require a minimum number of HU for
fruits to ripen. Heat units are the number of days in a growing season (i.e. October to March in
the Western Cape) with an average temperat ure of more than 10?C (Schulze 1997). HU are
especially important for wine grapes which require a minimum of 1000HU to ripen, while areas
with more than 2222HU are considered to be suitable for mass production of dessert wine
cultivars (Saayman 1981). Some apple cultiv ars (Royal Gala) also require 1000HU (Ortega-
Farias & Leon 2002). Fruits such as some variet ies of pears and apricots, require as few as
112HU (Valentini et al. 2004), while olives require about 150HU to fl ower (Orlandi et al. 2005).
In spite of the wide range of HU requirements of different perennial crops, it is generally
accepted that land suitability for the production of perennial crops increases as HU increase and
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
152
that areas with HU of more than 1000 will be suitable for all perennial crops. Two rules
representing highly suitable were subsequently created for HU. The first uses a fuzzy function to
affect an increase in membership to the highly suitable category for land units with HU values of
0 to 1000 and the second rule allocates full member ship to all land units with HU values of more
than 1000 ( Figure 9-9).
Figure 9-9 Heat units rules
The resulting requirement map is Figure 9-10 which, owing to the relatively high temperatures
experienced in the Western Cape, clearly show s that not many land units will be negatively
influenced by this requirement. In the AOI, only the mountain peaks (compare Figure 9-4) were
rated as being moderately suitable for perennial crops while the rest of the AOI was rated as
being highly suitable.
9.1.2.3 Frost occurrence rules
Most perennial crops can withstand very low temperatures (as low as -18?C) during their
dormancy period (Reiger 2006) but th ey are less cold hardy during spring when leaf and flower
buds begin to swell and bloom (Jackson 1999). Because temperatures in the Western Cape rarely
fall below 5?C, the main concern for perennial crops is temperatures below 0?C which expose
buds and fruit to frost during the growing season. Areas prone to frost from September to March
are therefore less suitable for perennial crop production. To identify such areas, the Frost (end)
land property, indicating the date on which frost occurrence is likely to end (as averaged from
historic data), was included in the analysis. Rules for marginally, moderately and highly suitable
areas were defined as shown in Figure 9-11.
153
Figure 9-10 Heat units requirement map of the AOI
Figure 9-11 Frost requirement rules
Areas prone to late frost are considered to be marginally suitable. Because the frost (end)
property is measured as days from the beginning of the year, all values of more than 273 (i.e.
later than the end of September), are considered to be marginally suitable. Areas experiencing
frost prior to the onset of the growing season (i.e. before day 212 or end of July), are considered
highly suitable, while all other areas are considered to be moderately suitable for perennial crop
production. Four additional fuzzy rules to create gradual transitions between the suitability
categories have been formulated.
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
154
The requirement map for frost (end) is shown in Figure 9-12. Due to proximity to the sea, most
of the AOI is not adversely affected by frost and is consequently regarded as highly suitable.
Only the Hottentots Holland Mountains north-eas t of Somerset West are mapped as being
marginally suitable for perennial crop production.
Figure 9-12 Frost requirement map of the AOI
9.1.2.4 Rainfall rules
Moisture availability is one of the main factors determining the growth performance of a plant.
Moisture can be obtained from natural sources such as precipitation and soil, or it can be
provided artificially through irrigation. In the Western Cape th e amount of irrigation water
needed to sustain optimal perennial crop growth is directly related to rainfall. To accommodate
rainfall variations, seven rules relating to mean annual rainfall were added to the rulebase (see
Figure 9-13).
Two fuzzy transitions were created in the ra nges 150mm to 400mm an d 600mm to 850mm to
separate the marginally suitable, moderately suitable and highly suitable categories. Areas with
an annual rainfall of more than 850mm are considered to be highly suitable for perennial crops as
only limited irrigation is likely to be required in such areas. The rules aim to indicate that
irrigation requirements for perennial crops gradually increase as the annual rainfall decreases.
The annual rainfall requirement map (see Figure 9-14) shows that the en tire AOI is either highly
or moderately suitable for perennial crops. This can be expected because the AOI has a relatively
high annual rainfall, particularly the mountains east of Stellenbosch where some of highest
rainfall in South Africa is measured (more than 2000mm annually). As a result, most the
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
155
Figure 9-13 Mean annual rainfall requirement rules
Figure 9-14 Mean annual rainfall requirement map of the AOI
vineyards and orchards in the Stellenbosch region are irrigated only during the driest months
(January and February).
Apart from rainfall, the availability of water for plants is also affected by the moisture holding
capacity of soils. The definition of soil requirements for perennial crops is described in the next
section.
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
156
9.1.3 Soil requirement rules
Soil properties are important for suitability analysis of perennial crops as they provide the
necessary anchorage, moisture and nutrients for deciduous plants to develop. Two major
attributes of soil are considered, namely clay content and effective depth.
9.1.3.1 Soil clay content rules
The clay content of soils determines to a great extent their moisture-holding capacity because
clayey, fine-textured soils absorb more water than sandy, coarse-textured soils. Soils with
excessive clay content can become waterlogged in wet areas thereby inhibiting plant growth.
Growth can also be inhibited if there is too much fluctuation in soil moisture as is common in
some areas in the Western Cape where very sandy soils occur (Saayman 1981). Finer-textured
soils are generally also more fertile than sandy soils as they contain more organic matter and
retain nutrients better. Concerning manageability, too clayey soils can be difficult to till and
sandy soils which are more stable need frequent fertilization (Brown 2003; Lambrechts & Ellis
s.d.).
The influences of soil clay cont ent in the A-horizon on the establishment of perennial crops were
considered in defining the requirement rules (see Figure 9-15). Six fuzzy and two Boolean rules
were defined. For the highly suitable category a Boolean rule was set at 10% to 20% and two
asymmetrical fuzzy rules, ranging from 5% to 10% and 20% to 25% resp ectively, were created
to implement a softer transition between the highly suitable and moderately suitable classes. A
similar approach was taken for the transition between moderately suitable and marginally
suitable. An additional Boolean rule was used to allocate values of more than 30% to the
marginally suitable class.
The rules in Figure 9-15 were executed to produce the suitability map in Figure 9-16. By
comparing the map with the rule set diagram in Figure 9-15, a good unders tanding of the soil
clay content of the area can be deduced. Most of the areas rated as highly suitable for perennial
crops in terms of clay content correspond well with the areas currently used for this land use: the
lower slopes of the Table Mountai n range and the winelands of Stellenbosch and Somerset West
stand out. Most of the sandy Cape Flats are assigne d to the moderately suitable class, while the
high sandstone mountainous areas are classified as marginally unsuitable.
157
Figure 9-15 Soil clay content requirement rules
Figure 9-16 Soil clay content suitability map of the AOI
9.1.3.2 Effective soil depth rules
Effective soil depth, also called rooting depth, refers to the depth at which root penetration is
strongly inhibited due to physical characteristics such as contact with bedrock, dense clay or
permanent water, or due to contact with sub-soils with extreme chemical properties (Soil Survey
Division Staff 1993). The effective depth of soil great ly determines the ability of soils to deliver
adequate nutrients and moisture to perennial crops. It is generally accepted that deeper soils are
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
158
more resistant to adverse climate conditions and that they ensure more stable growing
circumstances (Saayman 1981).
As discussed in Section 5.2.2.2, the effective soil depth prop erty was derived from the land type
data which only differentiates between depths of 1200mm or less. Soil series with effective
depths of more than 1200mm we re simply indicated as being 1200mm or deeper. In such cases
the lower limit (i.e. 1200mm) was used. Due to the effect of averaging, most land units with
deep soils were calculated to be shallower than they truly are. The values in the effective depth
land property must therefore be interpreted as an index of depth rather than an absolute measure
of effective depth. For instance, soils with an effective depth of more than 1000mm should be
interpreted as being 1000mm or deeper.
Seven rules were defined for the effective soil depth property (Figure 9-17). In the first rule,
shallow soils (0mm to 200mm) are allocated to the marginally suitable category. Soils with and
effective depth of between 400mm and 500mm are c onsidered to be moderately suitable for
annual crop production, while deep soils (i.e. deeper than 700mm) are regarded to be highly
suitable. Four fuzzy rules were defined to creat e transitions between the three Boolean rules.
Figure 9-17 Effective soil depth requirement rules
The requirement map for perenn ial crops in terms of effective soil depth is shown in Figure 9-18.
As expected, the more suitable deeper soils occur in the lower-lyi ng areas. The mountainous
areas are generally not considered to be suitable for annual crop production because much of the
topsoil has been removed through erosion, the residual soil being too shallow for vineyard or
orchard plantings.
159
Figure 9-18 Effective soil depth requirement map of the AOI
9.1.4 Current land uses and wetlands requirement rules
The factors defined in the previous section relate to the natural and physical properties of land
and do not take current land uses into account. To demonstrat e how the present land uses can be
incorporated into suitability analysis, two majo r land uses, namely urban and conservation areas,
were used to define general constraints on agricultural development. Although wetlands is not a
land use, it is added to the rulebase as a constraint to demonstrate how sensitive environmental
areas can be included in a suitability analysis.
9.1.4.1 Urban areas rules
It was assumed that perennial crops cannot be established on land already transformed by urban
use. Consequently, urban areas are considered to be permanently unsuitable for perennial crops.
To implement this constraint, the Urban areas (distance to) land property was used to set a rule
that excludes land units that are within 500m from existing urban areas. A 500m bu ffer was used
around urban areas to incorporate possible future expansion of the urban edge. Figure 9-19
shows the spatial effects when this rule applies.
9.1.4.2 Conservation areas rules
Conservation areas are protected by law from any agricultural development. National parks and
nature reserves are therefore included as a constraint in the rulebase by using the Conservation
areas (distance to) land property. As with Urban areas (distance to) , all land units that are
within 500 metres of conservation areas were ex cluded from further analysis. The land units
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
160
Figure 9-19 Urban areas in the AOI considered permanently unsuitable
affected by this constraint are shown in Figure 9-20 to be permanently unsuitable for perennial
crop production.
9.1.4.3 Wetlands rules
Wetlands are among the most threatened ecosystems in the Western Cape and they must be
protected at all costs from agricultural development (South Africa 1997). To exclude wetlands
from consideration for perennial crops, a similar approach to the previous two agricultural
constraints was taken using the wetlands (distance to) land property. Figure 9-21 shows the land
units in the AOI that are affected by this requirement. Unfortunately, it is clear from the map that
only major wetlands are included in the land unit database and that it should be expanded to
include other smaller, but equally sensitive wetlands. A possible remedy is to use the information
in the national wetlands map currently being developed by the South African National
Biodiversity Institute (Dini 2007).
The constraints on perennial crop production discussed in this section do not constitute an
exhaustive list. Other land cove r and land uses such as water bodies, natural vegetation and
mines, could also be included to reduce the number of land units considered for perennial crops.
There is also a range of other factors, such as soil pH, salinity and access to irrigation water, that
may contribute to the suitability of land for perennial crops. Nevertheless, the factors discussed
in this section should suffice to demonstrate the abilities of CLUES.
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
161
Figure 9-20 Conservation areas in the AOI considered permanently unsuitable
Figure 9-21 Wetlands in the AOI considered permanently unsuitable
9.2 WEIGHTING SUITABILITY FACTORS
All the factors defined above for the suitability analysis of land for perennial crop production are
not equally important for determining suitability. The CLUES inference engine is designed to
incorporate differences in importance by means of a weighting scheme. The weights assigned to
the individual requirements were obtained by using the analytical hierarchy process (AHP) (see
Section 2.3.1.5). The resulting AHP comparison matrix is shown in Table 9-2 (refer to Table 2-1
for scale value descriptions).
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
162
Table 9-2 AHP comparison matrix of weights assigned to land use requirements
S
lo
pe
gr
ad
ie
nt
H
ea
t u
ni
ts
E
ffe
ct
iv
e
so
il
de
pt
h
S
oi
l c
la
y
co
nt
en
t o
f
ho
riz
on
A
M
ea
n
an
nu
al
ra
in
fa
ll
Fr
os
t (
en
d)
C
hi
ll
un
its
Slope
gradient - 1/2 1/4 1/4 1 1/4 1/4
Heat units 2 - 1/2 1/2 2 1/2 1/2
Effective soil
depth 4 2 - 1 4 1 1
Soil clay
content of
horizon A
4 2 1 - 4 1 1
Mean annual
rainfall 1 1/2 1/4 1/4 - 1/4 1/4
Frost (end) 4 2 1 1 4 - 1
Chill units 4 2 1 1 4 1 -
An online AHP application developed by the Canadian Conservation Institute (2005) was used to
calculate the eigenvectors from the comparison matrix to determine the overall importance of
each factor. The resulting weight (importance) values are expressed as percentages in Figure
9-22. Although very little information about the relative importance of each of the factors could
be found in the literature, most sources emphasize the importance of chil l units during dormancy,
especially for pome fruit. Frost dur ing the growing season was considered to be critical for most
perennial crops. These two factor s, along with the requirements related to soil, consequently
received the highest weightings (20% each). Th e remaining 20% were allocated to heat units
(10%), slope gradient (5%), m ean annual rainfall (5%), urban ar eas (0.01%), wetlands (0.01%)
and conservation areas (0.01%). Owing to the Western Cape?s rela tively warm climate, heat
units are generally less important than chill units, while slope gradient is considered to be
slightly more important than mean annual rainfall as it is assumed that irrigation infrastructure is
available.
163
Figure 9-22 List of perennial crop requirements with weights shown on the req page
The three constraints, namely urban areas, wetlands, and conservation areas, were given weights
of 0.01% each. Because requirements with weight s of 0 are ignored by the inference engine, low
values of slightly more than zero should be used to activate the constraints and to indicate that
they do not inordinately contribute to or detract from the resulting suitability values. This
approach to incorporating constraints is necessary because a requirement can act as a constraint
(i.e. only consist of permanently unsuitable and unsuitable at present rules) or a factor (i.e.
include marginally, moderately or highly suitable rules). A requirement, such as the slope
gradient requirement in Figure 9-3, can however act as a cons traint and a factor if it includes
rules of both types. A weight of 0.01 is used for constraints to indicate that a requirement should
not be interpreted as a factor. This ensures that land units that are permanently unsuitable in
terms of any of the land requirements are not affected by other factors, no matter how well they
perform.
It is important to note that the above weighting scheme is by no means a definitive solution for
perennial crops as many horticulturalists will quite likely disagree with the chosen weights. This
weighting scheme was merely created to demonstrate the functionality of the system. Ideally,
more realistic weights and criteria should be derived from input obtained in a series of interviews
with experts because the best solutions are often obtained when a group of experts participates in
the AHP process.
9.3 CASE STUDY SUITABILITY ANALYSIS AT VARYING SCALES
The final step in the su itability analysis procedure is the production of suitability maps. The
following sections demonstrate how suitability maps can be created for different areas and at
164
varying levels of detail (i.e. map scales) to support land use decisions. Four case studies done for
areas of decreasing size and increasing map s cale are reported, namely greater Cape Town,
Swartland, rural Malmesbury and Stellenbosch.
9.3.1 Greater Cape Town
The suitability analysis for pe rennial crops was done using the land use requirements and
weights set out in the previous two sections. The analysis and mapping procedures took less than
two minutes to execute for this AOI. Figure 9-23 is the resulting suitability map for perennial
crop production in greater Cape Town. Most of the land units in this AOI are allocated to the
permanently unsuitable suitability level. Thes e units comprise urban and conservation areas,
wetlands and terrain with slope gradients of more than 20%.
Figure 9-23 Suitability map for perennial crops in the greater Cape Town AOI
None of the land units in the AOI were identified as being marginally suitable for perennial crop
production as this category is predominantly concealed by the permanently unsuitable class
(compare Figures 9-8, 9-12, 9-16 and 9-18). This is especially noticeable in the Cape Flats area
where most of the land units are marginally suitable in terms of chill units (see Figure 9-8). Land
units in this area that were not masked by the permanently unsuitable category were upgraded to
moderately suitable, mainly because they scored high in terms of heat units, frost end and
effective soil depth.
9.3.2 Stellenbosch
The overall suitability of land for perennial cr op production was calculated to be high in the
Franschhoek-Stellenbosch-Somerset West region. This area scored high or moderate for most of
LEGEND
Highly suitable
Moderately suitable
Marginally suitable
Unsuitable at present
Permanently unsuitable
Detailed map (Figure 9-24)
165
the requirements and it coincides well with areas currently used for producing wine grapes. This
is substantiated by a larger-scaled satellite image ( Figure 9-24d) of the area north of Stellenbosch
(location indicated as boxed in Figure 9-23) in which orchards a nd vineyards are clearly visible.
By overlaying the suitability layer (Figure 9-24a) on the satellite im age one can visually compare
the suitability levels with current land uses. Observe that some areas currently under vineyards
are identified as being unsuitable for perennial crops due to the 20% slope gradient. Some
vineyards near to Stellenbosch are excluded according to the rule that land units within 500m of
urban areas are classified as being permanently unsuitable for perennial crops.
Figure 9-24 A compilation of CLUES screen captures showing detailed maps of (a) the suitability analysis result,
(b) the suitability overlay, (c) land unit outlines, and (d) the satellite image of the area north of
Stellenbosch
Although the use of land components as mapping units is effective for natural environmental
variables such as climate and soil, this example illustrates that they have limitations when used
with features unrelated to terrain. For instance, urban edge s often do not coincide with the
boundaries of the land units (i.e. land components). This is strikingly appa rent when the outlines
of the land units are superimposed on the suitability map (Figure 9-24c). A possible solution to
this limitation is to include boundaries of selected features in the land unit mapping process. But
this will substantially increase the number of land units in the database, which in turn will
negatively influence the system?s overall response times.
Highly suitable Moderately suitable Marginally suitable Unsuitable at present Permanently unsuitable
(a) (b)
(c) (d) Stellenbosch
166
It is clear from Figure 9-24 that most of the land units id entified in the Stellenbosch area as being
highly suitable for perennial crops are indeed being used for this purpose. However, the aim of
suitability mapping may also be to identify areas that are not being optimally or even illegally
used so that alternative uses can be considered or remedies undertaken. To illustrate the further
potential of suitability analysis, a proportion of the Swartland was chosen as an alternative AOI.
9.3.3 Swartland
The Swartland ( Figure 9-25a) is a major wheat-producing area in the Western Cape where the
recently improved irrigation infrastructure has fortuitously made it possible to introduce
perennial crops.
Figure 9-25 A compilation of CLUE S screen captures showing (a) the satellite image and (b) the suitability
analysis results for the Swartland area
According to the suitability analysis, most of the land units in the region are moderately suitable
for perennial crops, while some areas around Darling and Malmesbury and the hillslopes south
and north of Kasteelberg, are identified as highly suitable (Figure 9-25b). The land units
excluded by the analysis are mostly those related to excessive slope and urban land use.
9.3.4 Rural Malmesbury
A larger-scale map of the rural area we st of Malmesbury in the Swartland (Figure 9-26) reveals
that the bulk of the land units that are highly suitable for perennial crops are currently being
extensively worked for the cultivation of annual crops. Two areas are permanently unsuitable for
perennial crops: the larger one was excluded mainly due its proximity to the town, while the
small area in the north is a wetland (Figure 9-26b). As observed in Figure 9-24, some
irregularities in suitability are related to the edges of urban areas. This anomaly is attributable to
Atlantis
Vo?lvlei Dam
Malmesbury
Kasteelberg
Darling
Highly suitable Moderately suitable Marginally suitable Unsuitable at present Permanently unsuitable
(a) (b)
Detailed map (Figure 9-26)
167
Figure 9-26 A compilation of CLUES screen captures showing detailed maps of (a) the satellite image and (b) the
suitability analysis results for the area west of Malmesbury
the use of land components as the basic mapping units which do not align with existing land use
boundaries.
Because perennial crops are potentially more profitable per area unit than wheat, it is fair to
conclude that the non-urban and non-wetland areas in Figure 9-26 are cu rrently not being
optimally used. Based on this preliminary analysis, they may potentially be suitable for perennial
crops. With an annual rainfall of 460mm, this area is significantly drier than the Stellenbosch
area and it will therefore have to rely much more on irrigation. Alternatively, dry-land vineyards
or olives can be considered. A more detailed (large-scale) survey incorporating factors like local
soil properties and irrigation infrastructure could be conducted to identify specific areas suitable
for intensive perennial crop cultivation.
9.4 DISCUSSION
This chapter has aimed to demonstrate how CLUE S can be used to interactively and rapidly
conduct suitability analyses in the Western Cape. The system o ffers the functionality to set and
store detailed land use requirements and produce high-quality suitability ma ps within seconds. A
series of suitability maps of varying scales were created for the greater Cape Town, Stellenbosch
and Swartland/Malmesbury areas according to ten land requirements for perennial crop
production. Although the results can be improved by incorporating more detailed data and by
setting better rules and weights, the analyses have succeeded to illustrate the capabilities of the
system. The demonstration clearly shows that the strength of CLUES is not only its ability to
carry out suitability analyses, but that its true powers lie in its ability to facilitate the user-
friendly assimilation of expert knowledge related to land uses and to rapidly generate spatial
visualizations of the products of the analyses. The highly automated environment enables experts
to interactively see the individual and combined effects on suitability by applying rules and
weighting schemes. In addition, the web-based platform of CLUES allo ws experts to use the
Malmesbury
Highly suitable Moderately suitable Marginally suitable Unsuitable at present Permanently unsuitable
(a) (b)
168
functionality from any computer with Internet access. This not only improves usability, but also
increases accessibility as special software and licenses are not required. However, the use of web
technology does limit users to the data stored in the land unit database because users are not
allowed to include external data in their analyses. This restriction, along with other specific
limitations of CLUES and web technology genera lly, are discussed in the next chapter.
169
CHAPTER 10: EVALUATION OF THE RESEARCH
Web technology may provide a solu tion to the high cost of SDSS and GIS as it eliminates the
need for expensive hardware or software. Web mapping has shown that the Internet is a cost-
effective way in which spatial information can be delivered to many users. The spatial analysis
functionality employed by SDSS is however difficult to implement using web technology. This
is mainly due to the complexities involved in managing the data used, created and updated in
such spatial analysis operations. To date, this limitation has impeded the use of the Internet for
SDSS applications.
The demonstrations described in the previous chapter confirm that spatial analysis functionality
can be implemented using standard (i.e. web browser compatible) web technologies. The
abilities and limitations of CLUES specifically a nd web technology generally are assessed in this
chapter and in conclusion the study?s aims and object ives are revisited to critically evaluate the
research results. Suggestions for future research are finally offered.
10.1 ASSESSMENT OF CLUES
To investigate the potential of web technology fo r SDSS development, a system was needed that
incorporates the functionality most frequently associated with SDSS. The most important
property that a SDSS should exhibit is the ability to interactively generate different geographical
scenarios through automated modelling and directed spatial analysis. This ability is especially
important in land suitability analysis as it involves multiple factors of varying importance that
are often difficult to define. By interactively visualizing the effects of different land use
requirements, the user can gain a better understanding of the dynamics and complexities of land
uses and their related properties.
CLUES was designed to encompass the functionality of SDSS with particular application to land
evaluation. Given that the aim was to determ ine the potential of the Internet for SDSS
development, the system had to be implemented using existing web technology so that Internet
users would be able to carry out land evaluation using a standard web browser.
The success of a software product is usually measur ed against its ability to meet the requirements
identified during the requirement analysis. A requirement analysis expresses the needs and
constraints placed on a software product and precedes most large software developments. A
software requirement is a property that the software must exhibit to solve a particular problem or
perform a certain function.
170
Drawing on an extensive literature study, a range of functional (i.e. what the system should do)
and operational (i.e. how the system should do it) requirements was identified (Section 4.1). The
data needed to support the functional and operational requirements was also specified. The
following sections revisit these requirements in order to review the success of CLUES as a web-
based land evaluation system. In each case further system design and implementation
suggestions are offered, together with ideas for further systematic research in this field.
10.1.1 Functionality requirements
The main functional requirement of CLUES is to produce land use suitability maps. Chapter 9
records how CLUES can be used to carry out land su itability analyses and it testifies that most of
the related functional requirements were implemented. It demonstrated that one is able to define
land uses, set suitability rules, execute evaluations and create suitability maps. The only
functional requirement that could not be implemented in CLUES is the ability for users to load
and prepare their own spatial data for analysis. This functionality was excluded due the security
risk associated with uploading data to web servers. A shared spatial database (i.e. a land unit
database) was consequently developed and populated with fundamental data sets for land
evaluation.
The process of setting land requir ement rules is highly flexible as users are allowed to specify
Boolean and fuzzy rules for individual suitab ility levels. Users can also manipulate the
importance of each land property by assigning weights to individual land requirements while a
unique set of land requirements can be defined for each land use that needs to be analysed.
Careful consideration must be given to the relative weightings of land use requirements as these
strongly influence the outcome of a suitability analysis. Decisions about weightings should be
supported by techniques such as Saaty?s (2003) an alytical hierarchical process (AHP) or pair-
wise comparison (Malczewski 1999).
The mapping capability of CLUES is one of its most powerful functions in that it allows users to
spatially visualize different land use scenar ios. The mapping environment is intuitive and
provides a range of tools through which users can produce suitability maps. Users can also
produce reports on individual or selected land units.
Although the main functional requirements of CLUES have been met, much additional
functionality should be considered to improve the system. A tool whereby users can produce an
optimal land use map would greatly enhance the system?s usefulness. Such a map can be
generated by assigning to a land unit the land use having the highest suitability rating. The tool
should permit users to select the land uses to be considered in the analysis and also allow them to
171
specify targets for the areas that each land use should cover. For agricultural land uses these
targets might reflect current demand for produce or market prices. By including prices and
average yields in the knowledge base, maps of optimal income can be produced. Such maps will
be very helpful to agricultural economics planning.
In the current version a satellite image is provided with which users can orientate themselves. It
may be convenient to supplement the satellite image with additional layers such as roads and
annotations to better orientate a user. However, care must be taken to keep the mapping
environment simple and easy to use. Too many layers and additional map information might
confuse users, especially those unfamiliar with GIS or similar software. Orientation layers
should therefore be deactivated by default, but an option to activate them should be included.
The present version of CLUES does not allow for individual land properties to be viewed
spatially. Although it is possible to view the effect of a single land use requirement on a
suitability analysis, it may be helpful to allow users to produce maps of individual land
properties so that they can familiarize themselves with the data in the land unit database. The
inclusion of hundreds of land property layers in the mapping environment will inevitably cause
confusion. A possible solution to this problem is to develop a second map viewer in which the
individual land properties can be examined.
The system now offers a tool whereby users can import existing land use requirements from the
central knowledge base. These la nd requirements should be extended to include more land uses
so that users can work more efficiently. Requirement specifications should be based on expert
knowledge and relevant literature, and the sources of the requirements must be properly
referenced. CLUES will benefit from functionality that facilitates AHP and related consistency
calculations.
The reporting capabilities of CLUES need to be improved. A worthwhile augmented report
would be one that provides an overview of a land use evaluation. Such a report might include
summaries of the rules and land use requirements used during the suitability analysis as well as
statistics about the analysis results. An outline of the relative proportions (i.e. area) of each
suitability level can, for instance, be provided. In addition, information about the specific area
under consideration can be compared with that of the entire province. This kind of overview will
be especially beneficial to environmental managers who are concerned with provincial and
national conservation targets.
Although the use of land components as mapping units (land units) is effective for representing
natural variables (e.g. effective soil depth, annual rainfall, heat units), they often do not coincide
with man-made boundaries (e.g. urban edges a nd conservation area limits). This may cause
172
inconsistent results when man-made features are considered in suitability assessments. A
possible solution is to split land units along the boundaries of selected features. However, care
must be taken to only incorporate essential boundaries as the splitting process may significantly
increase the number of land units, causing increased storage requirements and longer processing
times, particularly during suitability analyses.
10.1.2 Operational performance requirements
The operational performance of a system qualitatively describes how well it performs its primary
function. The following sections evaluate CLUES in terms of its accessibility, speed, user-
friendliness, and data storage capabilities.
10.1.2.1 System accessibility
Improved accessibility is one of the main advantages of using web technology for SDSS
development. For demonstration purposes CLUES runs on an intranet and is accessible only to
computers on the Stellenbosch University local area network (LAN). However, the system can
be conveniently migrated to a web server open to all Internet users. Th is will allow anyone with
access to a computer and the Internet to use the system.
Users do not need expensive GIS or expert syst em licenses to use CLUES. The only software
required is a freely available web browser. This means that CLUES is platform independent and
can be used on all major operating systems like Microsoft Windows, Linux, UNIX or Apple
Macintosh. CLUES requires limited computer resour ces as most of the processing is carried out
on the server. Even a computer with a modest hardware configuration or a mobile device such as
a cellphone can be used.
10.1.2.2 Operational speed
Although the centralization of computer processing (i.e. a server-client approach) eliminates the
need for sophisticated hardware and software on the client-side (i.e. user), it has implications as
to the hardware and processing speed of the server. Because suitability analysis is
computationally intensive ? each record (i.e. la nd unit) in the land unit database must be
considered and rated in terms of its land properties ? an analysis can take a long time to
complete.
In the requirement analysis it is specified that the response time of CL UES should be similar to
that of a GIS. In addition, a maximum response time of 60 seconds is specified as this is the
maximum time that a web user is willing to wait for a web page to load (see Section 4.1.2.2).
Due to the size of the land unit da tabase, however, it was determined that a tabular calculation
173
that emulates a suitability analysis takes more than three minutes to complete in ArcGIS 9.2.
Although the performance of CLUES is similar to that of ArcGIS, a response time of 60 seconds
is clearly not a realistic target.
An SDSS is meant to enable users to interact with spatial data to better understand semi-
structured and complex geographical problems. A major advantage of SDSS over GIS is that it
facilitates scenario building. Because response times in excess of three minutes inhibit
interaction and will inevitably frustrate users, a measure was implemented that reduces the
number of land units considered during a suitability analysis. By limiting the evaluation to those
land units visible on the map at the time of initiating an analysis, response times were
substantially improved to less than one minute (at a scale of 1:50 000). In addition, because the
processing time increases exponentially as the map scale is reduced, analyses at scales smaller
than 1:1 i000i000 were disabled.
Response times can be improved if CLUES is migr ated to a more powerful server. The modest
server on which the system now runs is not really suitable for intensive processing. Multiple
servers may even become necessary as the number of concurrent users increases.
Another factor influencing response time is Internet bandwidth. Although very little bandwidth is
required for most of the operations for setting of land use requirements, the mapping component
of CLUES is image intensive which may cause de lays on slow Internet connections. The size of
the CLUES map frame was intent ionally made small to keep bandwidth requirements at a
minimum.
10.1.2.3 System user-friendliness
Regarding the user-friendliness of the system, all of the requirements set out in Section 4.1.2.3
were implemented in the design and construction of the GUI. The interface was kept as simple as
possible whilst including all the necessary functionality. The appearances of menus, forms and
tables are consistent on all pages to prevent confusion and users are led through the process with
appropriate feedback and help.
To support the setting of rules, graphics are used to visualize the effect of rule thresholds. The
visualization of fuzzy functions especially can be improved with the help of other web
technologies such as shock wave flash (SWF). B ecause the graphics representing individual rules
are constructed from oblique images, fuzzy rules are partly obscured by one another. With SWF,
rules can be made transparent. It will also allow for more complex fuzzy functions (such as the s-
function) to be implemented.
174
The mapping and analysis environment is intuitive even for users who are unfamiliar with GIS
software. With a little practice, users can create high-quality ma ps using a range of available
controls. These include panning, zooming and enab ling/disabling layers for display. Suitability
maps are automatically produced from within the mapping environment by simply selecting the
appropriate land use from a drop-down list. For be tter readability, suitability maps are rendered
in colours identical to the colours assigned to the suitability levels in the rulebase. Results can
therefore be related to specific rule sets, hence supporting the interpretation of results as well as
the interactive fine-t uning of rules.
10.1.2.4 Data storage mode and capacity
The use of land units as basic mapping units enabled the storage of suitability analysis results in
the land unit database. Because these results are only stored temporarily (until the user ends the
session), the storage space requirements remain constant. This means that there is no danger that
the system will become unstable due to the creation of overly large volumes of data by users.
The database is also protected from corruption by vi rtue of the storage and retrieval of data being
internally managed by the system.
10.1.3 Data requirements
Although all the data sets specified in the requirement analysis were acquired, not all of these
were available at the specified scale of 1:50 000. The following sections overview the quality of
the data in the land unit database and make suggestions on how it can be improved.
10.1.3.1 Soil data
The soil data currently used by the system was de rived from land type data, published at a scale
of 1:250 000. This data is not deta iled enough to work in land suitability analyses at a scale of
1:50 000. Ideally, it should be replaced by more detailed data mapped at scales of 1:50 000 or
larger. Unfortunately, the land type data is the only existing soil data covering the entire Western
Cape. Although semi-detailed to detailed (i.e . 1:50 000 to 1:1 000) soil surveys have been
conducted for selected areas in the Western Cape, only some of this data is available in GIS
format. Much time and financial resources will be required to collect and capture this data into a
suitable format for use in suitability analyses. Moreover, the major part of the Western Cape has
not been surveyed at large scales. This means that land type data will, for the foreseeable future,
remain the only soil data source for the larger part of the Western Cape.
The land type data can be further analysed to estimate other variables such as soil pH and
fertility. To do so, each soil series in each land type must be determined and interpreted to
175
calculate average values that are representative of all the soil series in a land type. Owing to the
large number of soil series in each land type, an algorithm will have to be developed to automate
this process.
Because most of the data contained in the land types is based on terrain descriptions, it is
possible to spatially enhance the data with the aid of terrain analysis (Van Niekerk & Schloms
2002). However, more research is needed to determine the accuracy of such enhancements.
A methodology is also needed to use soil data at different scales in CLUES, including the semi-
detailed soil data available in GIS format. This will significantly improve suitability analysis in
areas for which the semi-detailed data is availabl e. To avoid any inconsistencies occurring in the
results obtained from different soil data sources, users should be informed about the quality of
the data used in an analysis and they should also be advised to interpret the results accordingly.
Techniques should be explored to predict so il distributions using more easily measured
environmental variables for areas in which semi-d etailed data is not available. These so-called
soil-landscape models are in creasingly being employed to supplement and even replace
conventional soil maps (Park, McSweeney & Lowery 2001). Soil-landscape modelling is the
application of statistical techniques to predict the spatial distribution of soils using terrain and
other environmental variables (Hengl, Gruber & Shrestha 2004). The technique is based on the
premise that there is a strong relationship between soil and topography (Jenny 1941). Although
more research is needed to determine the accuracy of soil-landscape model predictions
(McBratney, Santos & Minasny 2003; Thwaites & Slater 2000), the technique shows potential,
especially in the Western Cape for which relatively good terrain data is available (Van Niekerk
& Schloms 2002).
10.1.3.2 Terrain data
The WCDEM was used as the terrain data s ource. According to the accuracy assessment
reported in Section 5.1.3, no DEM of the Western Cape has a higher quality than the WCDEM
and it was found to be superior in terms of both spatial resolution (20m) and vertical accuracy
(7m). The WCDEM was however too detailed to be analysed as a unit in CLUES and was
consequently rescaled to 80m resolution. The ve rtical accuracy (9m) of the resulting WCDEM80
was not significantly effected and was still superior to other DEM of the Western Cape.
The WCDEM80 was used in CLUES as the source fo r elevation information and to generate the
terrain derivatives slope gradient, slope aspect and curvature. It was also used to delineate land
components using multi-resolution image segmen tation. The land components were used as
basic mapping units (land units) for suitability analysis.
176
10.1.3.3 Climate data
The climate data included in the land unit database was generated at a resolution of 90 metres ?
by far the most detailed climate data set currently available. Unfortunately, this data set consists
of only average monthly temperature and rainfall variables. Further analysis is required to derive
other variables useful in suitability analyses, for example chill units, heat units, continentality, a
summer aridity index and precipitation seasonality. These climatic indicators can be generated in
a GIS and loaded into the land unit database. A lternatively, functionality can be developed in
CLUES that will generate the indices from the climate variables already in the database. Land
properties calculated on demand can be considered as ?virtual? propert ies as they are not
physically stored in the land unit database. This will not only save storage space but will allow
users to manipulate and fine-tune land properties to their specific needs.
10.1.4 Scale, scalability and flexibility requirements
Although the current version of CLUES can be used in an operational environment, its
application is limited by the number of land property data sets in the land unit database. The
number of available land properties should be extended so that users can have more freedom in
terms of land requirement construction.
As discussed in Section 10.1.3.1, the scale of the land unit data influences the level at which land
evaluations can be done. Users must be made aware that the suitability maps generated by
CLUES are not meant to be used to support land use management decisions on a local or farm
level. The maps provide an overview of land su itability on municipal and provincial levels and
should be used as a preliminary indication of areas that might be suitable for a particular use.
Such areas should then be targeted for detailed analyses based on data captured at larger (> 01:
10 0000) scales.
The principles by which CLUES carries out suitability analyses are not related to scale.
Consequently, it is possible to modify the system to perform analyses at more detailed levels if
the data is available. Conversely, by simply modifying the land unit database, the system can be
configured to be used for another province or even on a national or global level. If such
modifications result in a substantial increase in land units, a more robust DBMS may be
required.
It is important to note that although the size of the land unit database can be extended, any
increase in the number of land units will influence the response times of the system as there is a
direct relationship between execution times and number of land units. For implementations
where the number of land units exceeds one million, it is recommended that multiple land unit
177
databases be used in order to reduce load. An additional menu or map can be included in the GUI
that will allow users to choose the area (or land unit database) in which they would like to work.
Such a differential approach will improve response times considerably and will effectively
support an unlimited number of land units.
Another limitation of the existing land unit database is that it can only support 255 fields (i.e.
attributes). Because land properties are stored as attributes of each land unit (i.e. record), each
land property requires a field in the database. Consequently, there is a limit to the number of land
properties that can be stored in the database. The maximum number of land properties also
depends on the number of active users because each user is assigned a field for analysis
purposes. It is unlikely that the number of users will exceed 50 at any time, which means that
there are 200 (less five for other operational purposes) fields available for the storage of land
properties. If storage for more than 200 land properties is needed, the database can be replaced
by an enterprise DBMS such as Oracle, which can accommodate up to 1000 fields. Owing to
factors such as the WMS and web server loads, a better solution would be to set up separate
servers, each with a different instance of CLUES. As with most enterprise web applications, a
central user management database can be created to direct users to the appropriate instance of
CLUES.
During suitability analysis, the attributes (i.e. land properties) of each land unit in the land unit
database are considered and evaluated against the rules in the knowledge base. By doing so
CLUES essentially emulates one of the most f undamental spatial analysis operations, namely
overlaying. Although logically delineated units representing terrain morphology (i.e. land
components) are used as the basic mapping units in CLUES, other units may also be employed.
Regular tessellations of square regions can for instance be used to emulate most raster GIS
overlying operations. This approach is highly effective as it eliminates the need for complex
spatial actions usually associated with vector overlaying. Operations such as finding
intersections between overlaying futures, splitting features, and creating new records and fields
are very processing-intensive, especially for larg e and complex areas such as the Western Cape.
CLUES was developed to investigate the poten tial of web technology for SDSS development ?
in a sense it is a case study of a web-based SDSS for land ev aluation. Its development has
provided a good understanding of the available technologies and given insight into the
capabilities and limitations of web technology for SDSS development. A synopsis of the
potential of web technology for SDSS development is presented in the next section.
178
10.2 POTENTIAL OF WEB TECHNOLOGY FOR SDSS DEVELOPMENT
Web technology holds much potenti al for SDSS development as it provides a cost-effective and
intuitive way in which directed spatial analysis functionality can be delivered to a wide audience.
Web-based SDSS such as CLUES ar e attractive to users because they are highly accessible and
free. Users are also likely to find web-based SDSS intuitive and easy to use as the user interface
consists of web pages and other web components that are familiar.
Web-based SDSS offer significant be nefits to developers because much of the costs related to
the distribution and maintenance of software are eliminated with client-server web technology.
In contrast to local (i.e. desktop) SDSS, updates can be made on a continuous basis without
seriously inconveniencing users. In effect, there is only one always up-to-date version of the
system at any given time. This simplifies s upport, maintenance and training activities.
The centralized way in which web-based systems st ore and distribute data is appealing to users
and developers. Many users do not have access to the spatial data required by many desktop
SDSS. The provision of preconfigur ed data sets enables users to obtain results within minutes of
entering the system as no data collection is required. In addition, it eliminates the need for users
to carry out data manipulations such as coordinate system, map projection and datum
conversions which, when done incorrectly, can have serious consequences regarding incorrect
analysis results. The disadvantage of not allowing us ers to upload their own data is that they are
bound by the available data, restricting them to the available land properties, and also limiting
their analyses to a particular region.
A significant advantage of web-ba sed SDSS is that reformatting of data to comply with the
system requirements prior to development, reduces system development time by avoiding
functionality to handle different formats and data types. Less time can also be spent on
implementing robust data management and security measures to prevent data corruption.
Security breaches are a major risk in web appl ications. In spite of the continuous efforts by
software developers and governments to reduce Internet-related crimes, many web users have
become victims of Internet fraud or have inadvertently downloaded malicious software. By
disabling the uploading of data to the web-based SDSS, much of th e security risk is eliminated.
Administrators must nevertheless ensure that server backups are made regularly to prevent data
loss.
One of the main limitations of web technology fo r SDSS deployment is response times as some
Internet connections are slow and can cause delays, especially when large maps (i.e. images) are
frequently downloaded. However, due to the large databases involved in SDSS, the major cause
179
of delay is more likely to be the processing of the tabular and spatial information. This is true for
CLUES which is unable to carry out a suitability an alysis for the entire Western Cape within an
acceptable waiting period (i.e. 60 seconds). This limita tion is directly related to the size of the
spatial database as similar delays occur when the same operations are carried out using a desktop
GIS. The delay is a result of the insufficient da ta-processing capability of the DBMS and is not
attributable to web technology per se. In general, web applications are less responsive than their
desktop counterparts as most actions are carried out by the web server. Wh ile the processing time
on the web server is comparable with that of a desktop application, additional delays are created
when a request for data is sent to the web server and when the resulting information is
transmitted back as web pages. The combined effect of these delays can cause a latency that can
frustrate users with slow Internet connections. However, on faster connections delays are almost
unnoticeable, especially to frequent Internet users who have become accustomed to short delays.
Another limitation of web-based SDSS is the restrictions imposed by web technology concerning
user-interface sophisticat ion. Unlike desktop applications that can use almost unlimited graphics,
web applications are limited by the standard markup and scripting languages used to develop
them. Although most of the graphical requirements of CLUES are met by the use of simple
image constructs and HTML, more advanced gr aphic capabilities may be required by other web-
based SDSS. For instance, an application might require input mechanisms that allow users to
change weightings by using animated sliders. Such advanced interface functionality is not
available using HTML and will have to be im plemented using SVG or SWF. Unfortunately,
these technologies are not native to all web browsers and might not work on all systems.
In spite of the graphical restrictions of web technology, most of the mapping functionality
needed by SDSS can be implemented using existing WMS. Although ArcIMS was used to
develop CLUES, it is only one of the many WMS available. In addition to a number of other
proprietary solutions, there are several open-s ource WMS that can be freely downloaded from
the Internet. These products are frequently updated with new functionality and should be
considered for future web-based SDSS impl ementations. Because the source code of open-
source products is available, the potential for customization is virtually unlimited. However,
proprietary products such as ArcIMS are considered to be more user-friendly and better
documented, hence requiring less development time.
WMS are aimed at producing web maps and are not equipped with spatial analysis functionality.
Most of the spatial analysis operations required by SDSS can nevertheless be developed using
existing web technology. Not only can overlaying be emulated using fixed mapping units and
standard DBMS functions, but operations such as proximity and connectivity can also be carried
180
out using scripting. In addition, many enterprise DBMS, such as Oracle and Informix, have
introduced spatial extensions to their products that offer a range of GIS-like operations.
However, more research is required to investigate the potential of these extensions for the
development of web-based SDSS.
The development of CLUES has shown that we b technology offers many opportunities for the
deployment of spatial analysis and modelling functionality. The generation process exposed
many advantages and limitations of the currently available technology as discussed in this
section. In the final section the research aims and objectives are revisited in order to evaluate the
successfulness of this study.
10.3 RESEARCH OBJECTIVES REVISITED
The main aim of this research was to investig ate the potential of web technology as a platform
for delivering SDSS functionality to a wide audience. As an experiment, a web-based land use
expert system was developed called the Cape Land Use Evaluation System (CLUES). The
motive for developing the system was to gain insights into the abilities and limitations of
available web technologies.
The first objective (see Section 1.6) was to review the literature to determine what functionality
is needed by a land evaluation system. The literature on each step in the land evaluation
procedure and the three approaches to land suitability analysis, namely Boolean overlay, multi-
criteria decision making (MCDM) and expert systems, was methodically reviewed (Chapter 2).
The literature survey also revealed helpful info rmation about web technologies available for the
development of a web-based SDSS and special atten tion was given to the technologies related to
web mapping applications (Chapter 3).
The study of current technologies and existing SDSS was not only instrumental in doing the
requirements analysis and designing CLUES (Chapter 4), but also helped identify what data was
needed to demonstrate the functionality of the system. This requirement led to the second
research objective, namely to co llect and prepare fundamental data sets to test and demonstrate
CLUES (Chapter 5). Although good- quality climate and terrain data sets were obtained, a
general lack of detailed soil data necessitated the use of smaller scale (i.e. 1:250 000) landtype
data.
The third and most comprehensive objective of the research was to design, develop and
implement the web-based land evaluation system. Land evaluation was chosen as the SDSS
application as it strongly relies on spatial analysis ? the cornerstone of SDSS. An expert system
approach was taken to develop CLUES which cons ists of a land unit database, a knowledge base
181
and a website. The activities related to the development of each of these components are
described in Chapters 6, 7 and 8 respectively. Most of the requirements set out in Chapter 4 were
successfully implemented in the resulting system.
The demonstration of CLUES (Chapter 9) c onstituted the fourth research objective. The
functionality of the system was illustrated by carrying out land suitability analyses of perennial
crops in the greater Cape Town , Stellenbosch and Swartland regions. The result ing land use
scenarios are realistic and informative. They did, however, expose some limitations of the
system and the available data. This is reported in Chapter 10, which addr esses the final objective
of the research, i.e. to critically evaluate and make recommendations about CLUES, and draw
attention to the general limitations and potentials of web technology for web-based SDSS
development. The evaluation showed that available web technology offers excellent
opportunities for the deployment of spatial analysis and modelling functionality to a wide
audience.
10.4 CONCLUSION
Products like Google Earth have spotlighted the value of web mapping technology for spatial
decision support and they have demonstrated the potential of such tools for the cost-effective
distribution of maps and other spatial information to improve productivity and to aid decision
making. But the functionality of most web mapping applications is limited to data display and
does not support more advanced functionality, such as spatial analysis and modelling, needed for
SDSS development. This research has shown that web mapping technology can be extended to
include this functionality by combining standard web mapping technology and database
management systems with client-side and server-side web progra mming. The techniques
developed here can be used to implement and distribute powerful spatial analysis functionality to
Internet users. To date, such functionality has been the domain of those with access to expensive
GIS software and the necessary data. A web-ba sed SDSS such as CLUES has the potential to
dramatically increase access to spatial analysis functionality since anyone with access to a
modest computer and the Internet can use the systems.
Increased usage of web-based SD SS is likely to help improve spatial awareness because users
will gain a better understanding of the possibilities of spatial technologies. Increased
accessibility is likely to stimulate an increase in demand for additional functionality which in
turn will inspire the development of better technology. One can anticipat e that web-based SDSS
will boost the current upward trend in the online use of maps and geographical tools, and also
that new SDSS will be developed for various applications. Such online SDSS are expected to
become valuable sources of spatial information and they will also provide mechanisms through
182
which users can store, analyse and share expert knowledge to make better spatial decisions. Web
technology is a medium possessing the unique ability to act as an intermediary between
collaborating individuals to find solutions to complex geographical problems facing our
increasingly complex world. It is expected that the capacity of web technology to cost-effectively
deploy geographical information and functionality to a wide audience will bring the so-called
?unfinished GIS revolution? to an end and that web-based SDSS will help support earth-changing
decision making.
183
REFERENCES
Abran A, Moore JW, Bourque P & Dupuis R (eds) 2004. The software engineering body of
knowledge guide. Los Alamitos: IEEE Computer Societ y Professional Practices Committee.
Adediran AO, Parcharidisb I, Poscolieri c M & Pavlopoulosd K 2004. Computer-assisted
discrimination of morphological units on north-central Crete (Greece) by applying
multivariate statistics to local relief gradients. Geomorphology 58: 357-370.
Agrell PJ, Stam A & Fischer GW 2004. Inter active multiobjective agro-ecological land use
planning: The Bungoma region in Kenya. European Journal of Operational Research 158:
194-217.
Al-Najjar B & Alsyouf I 2003. Selecting the most efficient maintenance approach using fuzzy
multiple criteria decision making. International Journal of Economics 84: 85-100.
Altech Netstar 2008. Netstar [online]. Available from
http://www.netstar.co.za/content/home/home.aspx [Accessed 1 July 2008].
Antonio G & Signorini A 2005. The indexabl e Web is more than 11.5 billion pages . Paper
delivered at the WWW conference, Chiba.
Argialas DP 1995. Towards structured-knowle dge models for landform representation.
Zeitschrift f?r Geomorfologie N.F 101: 85-108.
Ascough JCI, Rector HD, Hoag DL, McMaster GS, Vandenberg BC, Shaffer MJ, Weltz MA &
Ahjua LR 2001. Multicriteria spatial decisi on support systems for agriculture: Overview,
applications, and future research directions. In Rizzoli AE & Jakeman AJ (eds) Integrated
assessment and decision support proceedings of th e 1st Biennial Meeting of the IEMSS. June
24-27 2002 , 175-180. Lugano: IEMSS.
Bailey B s.d. User interface design update [onlin e]. Insights from Human Factors International
Inc. Available from http://www.keller.com/artic les/downloadspeed.txt [Accessed 6 February
2008].
Basson FC 2005. A spatial decision support sy stem for groundwater abstraction impact
assessment and licensing. MSc thesis. Stellenbosch: Stellenbosch University.
Berners-Lee T 1989. Information management: A proposal [online]. Geneva: CERN. Available
from http://www.w3.org/History/1989/proposal.html [Accessed 18 Oct 2005].
184
Berners-Lee T 1991. Re: status. Re: X11 BR OWSER for WWW [onlin e]. Available from
http://lists.w3.org/Archives/P ublic/www-talk/1991SepOct/0003.html [Accessed 31 October
2007].
Bester FJ 2004. Multi-criteria decision maki ng and geographical information systems: An
extension for ArcView. MSc thesis. Stellenbosch: Stellenbosch University.
Bolstad PV & Smith JL Errors in GIS: Assess ing spatial data accuracy. In Lyon JG & McCarthy
J (eds) Wetland and environmental applications in GIS , 301-312. New York: Lewis.
Bosshard A 2000. A methodology and terminology of sustainability assessment and its
perspectives for rural planning. Agriculture, Ecosystems & Environment 77: 29-41.
Botes A, McGeoch MA, Robertson HG, Van Niek erk A, Davids HP & Chown SL 2006. Ants,
altitude and change in the northern Cape Floristic Region. Journal of Biogeography 33: 71-
90.
Brown RB 2003. Soil texture [online]. Gainesville : University of Florida. Available from
http://edis.ifas.ufl.edu/SS169 [Accessed 11 September 2007].
Bruno R, Follador M, Paegelow M, Renno F & Villa N 2006. Integrating Remote Sensing, GIS
and Prediction Models to Monitor the Deforestation and Erosion in Peten Reserve,
Guatemala . Society for Mathematical Geology XIth In ternational Congress, Universit? de
Li?ge, Belgium.
Buckle C 1996. Weather and climate in Africa. Harlow: Longman.
Burrough PA 1989. Fuzzy mathematical methods for soil survey and land evaluation. Journal of
Soil Science 40: 477-492.
Burrough PA, MacMillan RA & Van Deursen W 1992. Fuzzy classification methods for
determining land suitability from soil profile observations and topography. Journal of Soil
Science 43: 193-210.
Burrough PA & McDonnel RA 1998. Principles of geographical information systems. Oxford:
Oxford University Press.
Campbell JB 2006. Introduction to remote sensing. London: Taylor & Francis.
Canadian Conservation Institute 2005. Analyti cal Hierarchy Process (AHP) program [online].
Ottawa: Canadian Conservation Institute. Available from http://www.cci-
icc.gc.ca/tools/ahp/index_e.asp [Accessed 10 July 2008].
185
Canter S 2004. Understanding Client-Side Scrip ting [online]. PC Magazine. Available from
http://www.pcmag.com/article2/0,1759,1564972,00.asp [Accessed 1 June 2008].
CapeNature 2007. CapeNature reserves [on line]. Cape Town: WCNCB. Available from
http://www.capenature.com/index.php?fSectionId=3 [Accessed 31 August 2007].
Caquard S 2003. Internet, maps and public particip ation: Contemporary limits and possibilities.
In Peterson MP (ed) Maps and the Internet , 345-357. Amsterdam: Elsevier.
Carey VA 2005. The use of viticultural terroir un its for demarcation of geographical indicators
for wine production in Stellenbosch and surrounds. PhD dissertation. Stellenbosch:
Stellenbosch University, Dept of Viticulture and Oenology.
Cartwright W 2003. Maps on the Web. In Peterson MP (ed) Maps and the Internet , 35-56.
Amsterdam: Elsevier.
CDSM 2007a. Digital elevation model (DEM) [ online]. Cape Town: CDSM. Available from
http://w3sli.wcape.gov.za/ [Accessed 21 September 2007].
CDSM 2007b. Maps of the national series [ online]. Cape Town: CDSM. Available from
http://w3sli.wcape.gov.za/ [Accessed 21 September 2007].
Ceballos-Silva A & L ? pez-Blanco J 2003a. Delineation of su itable areas for crops using a multi-
criteria evaluation approach and land use/c over mapping: A case study in Central Mexico.
Agricultural Systems 77: 117-136.
Ceballos-Silva A & L?pez-Blanco J 2003b. Evaluati ng biophysical variables to identify suitable
areas for oats in Central Mexico: A multi-criteria and GIS approach. Agriculture, Ecosystems
and Environment 95: 371-377.
Chang K 2006. Introduction to geographi c information systems. 3rd ed. New York: McGraw
Hill.
Clarke KC 2003. Getting started with geographic information systems. Upper Saddle River:
Prentice Hall.
Clarke M 1990. Geographical info rmation systems and model-based analysis: Towards effective
decision support systems. In Scholten HJ & Stillwell JCH (eds) Geographical information
systems for urban and regional planning , 165-175. Amsterdam: Kluw er Academic Publishers.
CNDV Africa 2005. Western Cape provincial spa tial development framework [online]. Cape
Town: PGWC. Available from http://www.capegateway.gov.za/eng/pubs/guides/W/120505
[Accessed 11 March 2008].
186
Codd EF 1970. A relational model of data for large shared data banks. Cummiciations of the
ACM 13: 377-387.
Cools N, De Pauw E & Deckers J 2002. Towards an integration of conventional land evaluation
methods and farmers? soil suitability asse ssment: A case study in northwestern Syria.
Agriculture, Ecosystems and Environment 1968: 1-16.
Dai FC, Lee CF & Zhang XH 2001. GIS-based geo- environmental evaluation for urban land-use
planning: A case study. Engineering Geology 61: 257-271.
Davidson DA (ed) 1986. Land evaluation. New York: Van Nostrand Reinhold.
Davidson DA, Theocharopoulos SP & Bloksma RJ 1994. A land evaluation project in Greece
using GIS and based on Boolean and fuzzy set methodologies. International Journal of
Geographical Information Systems 8: 369-384.
De Kok R, Schneider T & Ammer U 1999. Object-b ased classification and applications in the
alpine forest environment. International Archives of P hotogrammetry and Remote Sensing 32:
part 7-4-3.
De la Beaujardiere J 2004. OGC Web map service interface [online]. Available from
http://www.opengeospatial.org/standards/wms [Accessed 7 February 2008].
De la Rosa D 2002. MicroLEIS DSS: A land eval uation decision support system for agricultural
soil protection [online]. Seville: Instituto de Recursos Naturales y Agrobiolog?a de Sevilla.
Available from http://irnas106.irnase.csic.es/mic rolei/manual2/overview.htm#top [Accessed
13 October 2006].
De la Rosa D, Mayol F, Diaz-Pereira F, Fern andez M & De la Rosa D 2004. A land evaluation
decision support system (MicroLEIS DSS) for agricultural soil protection with special
reference to the Mediterranean region. Environmental Modelling & Software 19: 929?942.
Definiens Imaging 2004. eCognition 4 User Manual. Munich: Definiens Imaging GmbH.
Definiens Imaging 2007. Definiens Developer (version 7) user manual. Munich: Definiens AG.
DeMers MN 2005. Fundamentals of geographic information systems. 3rd ed. New York: Wiley
& Sons.
Dendgiz O, Bayramin ? & Y?ksel M 2003. Geogra phic information system and remote sensing
based land evaluation of Beypazar ? area soils by ILSEN model. T urkish Journal of
Agriculture and Forestry 27: 145-153.
187
Dennis AR & Taylor NJ 2006. Information foragi ng on the web: The effects of "acceptable"
Internet delays on multi-page information search behaviour. Decision Support Systems 42:
810-820.
Densham PJ 1991. Spatial decisi on support systems. In Maguire DJ, Goodchild MF & Rhind
DW (eds) Geographical information system s: Principles and applications , 403-412. Harlow:
Longman.
Dent D & Young A 1981. Soil survey and land evaluation. London: George Allen & Unwin.
Dini J 2007. WfWet's year in review: March 2006 to April 2007 [online]. Cape Town: SANBI.
Available from http://wetlands.sanbi.org/resource.php?id=113 [Accessed 10 March 2008].
Dobbs D 2004. Vital statistics: Web pages that suck [online]. Electroni c review of computer
books. Available from http://www.ercb.com/feature/feature.0027.html [Accessed 13 May
2008].
Du Toit DA, Mouton PLN, Flemming AF, Van Ni ekerk A, Day JA & Schulz R 2002. Climate
and the presence of generation glands in female girdled lizards: A case study of the cordylus-
niger-oelofseni c omplex. Journal of Herpetology 39: 384-388.
Du Toit DA, Mouton PLN & Van Niekerk A 2006. C limatic correlates of melanistic cordylid
lizards . Paper presented at the 8th Herpetological Association of Africa Symposium, North-
West University, Potchefstroom.
Dymond JR, Derose RC & Harmsworth GR 1995. Automated mapping of land components from
digital elevation data. Earth Surface Processes and Landforms 20: 131-137.
Eastman JR 2000. Decision strategies in GIS. Directions Magazine Dec 2000: s.p.
Eastman JR 2006. IDRISI Andes: Guide to GIS and image processing. Worcester: Clark
University.
Eastman JR, Jin W, Kyem PAK & Toledano J 1995. Raster procedures for multi-criteria/multi-
objective decisions. Photogrammetric Engineering & Remote Sensing 61: 539-547.
Encyclopaedia Britannica 2007. Artificial intelligence [ online]. Encyclop?dia Britannica
Online. Available from http://www.britannica.com/eb/article-219098 [Accessed 23 Dec
2007].
ESRI 2002a. ArcView GIS online user guide. Redlands: ESRI.
ESRI 2002b. ArcView online help. Redlands: ESRI.
ESRI 2002c. MapObjects. Redlands: ESRI.
188
ESRI 2003. ArcIMS scalability supports heavy demand for site. ArcUser January-March: s.p.
ESRI 2007a. ArcGIS 9.2 online user manual. Redlands: ESRI.
ESRI 2007b. ArcIMS 9.2 Installation manual. Redlands: ESRI.
ESRI 2007c. ArcSDE [online]. Available from
http://www.esri.com/softwar e/arcgis/arcsd e/index.html [Accessed 23 November 2007].
FAO 1976. A framework for land evaluation. Rome: FAO.
FAO 1984. Land evaluation for development. Rome: FAO.
FAO 1985. Guidelines: Land evaluation for irrigated agriculture. Rome: FAO.
Fernandez Ruiz R 2003. Alternative land uses to fo restry in the Western Cape: a case study of La
Motte plantation . MSc thesis. Stellenbosch: Stellenbosch University.
Fischer G, Granat J & Makowski M 1998. A E Z W I N : An interactive multiple-criteria analysis
tool for land resources appraisal. Laxenburg: International Institute for Applied Systems
Analysis.
Fleming CC & Von Halle B 1989. Handbook of relational database design. New York:
Addison-Wesley.
Fourie JC 2006. Evaluating agricultural potent ial of a Cape Metropolitain catchment: A fuzzy
logic approach. MSc thesis. Stellenbosch: Stellenbosch University.
Gartner G 2005. TeleCartography : A new means of GeoCommunica tion. In Taylor DRF (ed)
Cybercartography: Theory and practice , 373-387. Amsterdam: Elsevier.
Gibson PJ & Power CH 2000. Introduction to remote sensin g: Digital image processing and
applications. London: Taylor and Francis.
Giles PT & Franklin SE 1998. An automated appr oach to the classification of the slope units
using digital data. Geomorphology 21: 251-264.
GISjobs.com 2006. User survey [online]. A ppleton: GISjobs.com, LLC. Available from
http://www.gisjobs.com/survey/resp onses.jsp?countryLoc=all&sal=N [Accessed 10 February
2006].
Goldstuck A 2004. The next big boom [online]. Pi negowrie: World Wide Worx. Available from
http://www.ghostdigest.co.za/code/A_391.html [Accessed 21 Aug 2006].
189
Goodchild MF & Densham PJ 1990. Research initiative six, spatial decision support systems:
Scientific report for the specialist meeting, Technical Report 90-5. Santa Barbara: National
Center for Geographic Information and Analysis.
Google 2005. Google launches free 3D mapping a nd search product [onl ine]. Mountain View:
Google Inc. Available from http://www.google.com/press/pressrel/google_earth.html
[Accessed 18 Oct 2005].
Graff LH & Usery EL 1993. Automated classificati on of generic terrain features in digital
elevation models. Photogrammetric Engineering & Remote Sensing 59: 1409-1417.
GRASS 2006. Geographic resources analysis support system [online]. Trento. Available from
http://grass.itc.it/ [Accessed 16 February 2006].
Green D & Bossomaier T 2001. Online GIS and spatial metadata. London: Taylor & Francis.
Hall DJ & Khanna DK 1977. The ISODATA method co mputation for the relative perception of
similarities and differences in complex and real data. In Enslein K, Ralston A & Wilf HS
(eds) Statistical methods fo r digital computers , 340-373. New York: John Wiley & Sons.
Hall GB, Wang F & Subaryono 1992. Comparison of Boolean and fuzzy classification methods
in land suitability analysis by using geographical information systems. Environment and
Planning 24: 497-516.
Hengl T, Gruber S & Shrestha DP 2004. Reduction of errors in digital terrain parameters used in
soil-landscape modelling. International Journal of A pplied Earth Observation and
Geoinformation 5: 97-112.
Hijmans RJ, Cameron SE, Parra JL, Jones PG & Jarvis A 2005. Very high resolution
interpolated climate surfaces for global land areas. International Journal of Climatology 25:
1965-1978.
Hoersch B, Braun G & Schmidt U 2002. Relation between landform and vegetation in alpine
regions of Wallis, Switzerland. A multiscale remote sensing and GIS approach. Computers,
Environment and Urban Systems 26: 113-139.
Houghton JT, Ding Y, Griggs DJ, Noguer M, Van der Linden PJ, Da X, Maskell K & Johnson
CA (eds) 2001. Climate change 2001: Th e scientific basis. Cambridge: Cambridge Press.
Huajun T, Debaveye J, Da R & Van Ranst E 1991. Land suitability classification based on fuzzy
set theory. Pedologie 61: 277-290.
Huajun T & Van Ranst E 1992. Testing of fuzzy se t theory in land suitability assessment for
rainfed grain maize production. Pedologie 42: 129-147.
190
IBM 2007. Informix spatial datablad e module [online]. Available from http://www-
306.ibm.com/software/data/informix/blades/spatial/ [Accessed 23 November 2007].
Igu? AM, Gaiser T & Stahr K 2004. A soil and te rrain digital database (SOTER) for improved
land use planning in Central Benin. European Journal of Agronomy 21: 41-52.
Irvin BJ, Ventura SJ & Slater BK 1997. Fuzzy a nd isodata classification of landform elements
from digital terrain data in Pleasant Valley, Wisconsin. Geoderma 77: 137-154.
Jackson DI 1999. Climate and fruit plan ts. In Jackson DI & Looney NE (eds) Temparate and
subtropical fruit production , 7-14. New York: CABI.
James AJ 2001. Die identifisering van ontwi kkelingsensitiewe areas teen berghellings:
Stellenbosch- en Hottentotshollandberge . MA thesis. Stellenbosch: Stellenbosch University.
Jankowiski P & Nyerges T 2001. GIS-supported colla borative decision making: Results of an
experiment. Annals of the Associatio n of American Geographers 91: 48-70.
Jarupathirun S & Zahedi FM 2007. Exploring the influence of perceptual factors in the success
of web-based spatial DSS. Decision Support Systems 43: 933-951.
Jenks GF 1967. The data model c oncept in statistical mapping. International Yearbook of
Cartography 7: 186-190.
Jenny H 1941. Factors of soil formation: A system of quantitative pedology. New York:
McGraw-Hill.
Jiang B 2003. Beyond serving maps: Serving GIS func tionality over the Internet. In Peterson MP
(ed) Maps and the Internet , 147-157. Oxford: Elsevier.
Jiang H & Eastman JR 2000. Application of fuzzy measures in multi-criteria evaluation in GIS.
International Journal of Ge ographical Information Science 14: 173-184.
Joerin F, Theriault M & Musy A 2001. Using GI S and outranking multi-criteria analysis for
land-use suitability assessment. International Journal of Geog raphical Information Science
15: 153-174.
Joubert SJ 2007. High-resolution climatic variable generation for the Western Cape . MSc thesis.
Stellenbosch: Stellenbosch University.
Joubert SJ & Van Niekerk A 2005. Enhancement of climate data in the South Western Cape
using digital elevation models. Paper presented at the Annual South African Geography
Students Conference, Cape Town, Un iversity of the Western Cape.
191
Kalogirou S 2002. Expert systems and GIS: An application of land suitability evaluation.
Computers, Environment and Urban Systems 26: 89-112.
K?bben B 2001. Publishing maps on the Web. In Kraak M-J (ed) Web cartography
developments and prospects , 73-86. London: Taylor & Francis.
Kok P & Collinson M 2006. Migration and urbanization in So uth Africa. Report 03-04-02.
Pretoria: Statistics South Africa.
Kotonya G & Sommerville I 1998. Requirements engineering: Processes and techniques. West
Sussex: John Wiley & Sons.
Kraak M-J 2001. Settings and needs for web cartography. In Kraak M-J & Brown A (eds) Web
cartography developments and prospects , 1-7. London: Taylor & Francis.
Lambrechts JJN & Ellis F s.d. Soil surveys, soil maps, survey methods and land and soil
capability evaluation. Stellenbosch: Stellenbosch University.
Land Type Survey Staff 1984. Land types of the maps 2626 Wes-Rand & 2726 Kroonstad .
Pretoria: Department of Agriculture.
Lantican CA, Grierson IT, Chittleborough DT & Lewis MM 1998. Application of the automated
land evaluation system (ALES) and remote sensi ng for land use planning in the Philippines.
Paper presented at the Ninth Australasian Remote Sensing Conference, Sydney.
Lime S 2006. MapServer [online]. Minnesota: University of Minnesota. Available from
http://mapserver.gis.umn.edu/ [Accessed 17 November 2007].
Longley PA, Goodchild MF, Maquire DJ & Rhind DW 2002. Geographic information systems
and science. Chichester: John Wiley & Sons.
Lutgens FK & Tarbuck EJ 1998. The atmosphere. New Jersey: Prentice Hall.
L?tz M & Bastian O 2002. Implementation of land scape planning and nature conservation in the
agricultural landscape: a case study from Saxony. Agriculture, Ecosystems and Environment
92: 159-170.
Lynch SD 1999. Converting point estimates of daily rainfall onto a rectangular grid. Paper
presented at the ESRI User Conference ?98, San Diego.
MacMillan RA, Jones RK & McNabb DH 2004. Defi ning a hierarchy of spatial entities for
environmental analysis and modeling using digital elevation models (DEMs). Computers,
Environment and Urban Systems 28: 175-200.
192
Mahini AS & Gholamalifard M 2006. Siting MSW landfills with a weighted linear combination
methodology in a GIS environment. International Journal of Environment, Science and
Technology 3: 435-445.
Malczewski J 1999. G IS and multi-criteria decision analysis. New York: John Wiley & Sons.
Malczewski J 2004. GIS-based land-use suit ability analysis: A critical overview. Progress in
Planning 62: 3-65.
Malczewski J 2006. Ordered weighted averaging with fuzzy quantifiers: GIS-based multicriteria
evaluation for land-use suitability analysis. International Journal of Applied Earth
Observation and Geoinformation 8: 270-277.
Mantel S, Van Engelen VWP, Molfino JH & Re sink JW 2000. Explorin g biophysical potential
and sustainability of wheat cultivation in Uruguay at the national level. Soil Use and
Management 16: 270-278.
Mantel S, Zhang X & Zhang G 2003. Identification of potential for banana in Hainan island,
China. Pedosphere 13: 147-155.
MapQuest 2008. Mapquest [online]. Denver. Available from http://www.mapquest.com/
[Accessed 1 June 2008].
Marinoni O 2004. Implementation of the analytic al hierarchy process with VBA in ArcGIS.
Computers & Geosciences 30: 637-646.
Mau-Crimmins T, De Steigue r JE & Dennis D 2005. AHP as a means for improving public
participation: A pre-post experi ment with university students. Forest Policy and Economics 7:
501-514.
McBratney AB, Santos MLM & Minasny B 2003. On digital soil mapping. Geoderma 117: 3-52.
McDonald RC, Isbell RF, Speight JG, Walker J & Hopkins MS 1984. Australian soil and land
survey. Melbourne: Inkata Press.
McHarg I 1969. Design with nature. New York: Doubleday & Company.
McSweeney K, Gessler PE, Slater B, Hammer RD & Bell J 1994. Towards a new framework for
modeling the soil-landscape continuum. Soil Science Society of America 33: 127-145.
Microsoft 2007. Microsoft Office Access 2007 t op 10 benefits [online]. Available from
http://office.microsoft.com/en-us/access/HA101650211033.aspx [Accessed 5 November
2007].
193
Miliaresis GC 2001. Geomorphometric mappi ng of Zagros Ranges at regional scale. Computers
& Geosciences 27: 775-786.
Mitchell CW 1991. Terrain evaluation. New York: Longman.
Mitchell CW, Webster R, Beckett PHT & Clifford B 1979. An analysis of terrain classification
for long-range prediction of conditions in deserts. Geographical Journal : 72-85.
Mlisa A 2007. Spatial decision support system for hydrological studies in the Table Mountain
Group aquifers . MSc thesis. Stellenbosch: Stellenbosch University.
Modine A 2007. Dell makes headway in server ma rket [online]. The Register. Available from
http://www.theregister.co.uk/2007/08/23/idc_servers_q207/ [Accessed 2 November 2007].
Mongkolsawat C, Thirangoon P & Kuptawutinan P 1997. A physical evaluation of land
suitability for rice: A methodological study using GIS . Paper presented at the 18th Asian
Conference on Remote Sensing, Malaysia.
Mozilla Foundation 2008. About Java Script [online]. Available from
http://developer.mozilla.org/en/docs/About_JavaScript [Accessed 1 July 2008].
Mucina L & Rutherford MC (eds) 2006. The vegetation of South Africa, Lesotho and Swaziland.
Pretoria: SANBI.
MWEB 2005. StreetMap [online]. J ohannesburg: MWEB. Available from
http://new.mweb.co.za/general/streetmap.jsp [Accessed 18 Oct 2005].
Mweso E 2003. Evaluating the importance of soil moisture availability (as a land quality) on
selected rainfed crops in Serowe area, Botswana. MSc thesis. Enschede: International Institute
for Geo-Information Science and Earth Observation.
Nah F 2004. A study on tolerable waiting time: How long are Web users willing to wait?
[online]. Lincoln: University of Nebraska-Lincoln. Available from
http://sigs.aisnet.org/sighci/bit04/BIT_Nah.pdf [Accessed 31 August 2007].
NASA 2005. Shuttle Radar Topography Mission [online]. Available from
http://www2.jpl.nasa.gov/srtm/mission.htm [Accessed 13 May 2008].
National Geographic Society 2005. Map mach ine [online]. Washington D.C.: National
Geographic Society. Available from http://plasma.nationalgeographic.com/mapmachine/
[Accessed 18 Oct 2005].
Netcraft 2007. August 2007 web server survey [online]. Available from
http://news.netcraft.com/arc hives/web_server_survey.html [Accessed 22 August 2007].
194
netz-tipp.de 2002. Das Internet spricht Englisch - und neuerdings auch Deutsch [online]. netz-
tipp.de. Available from http://www.netz-tipp.de/languages.html [Accessed 19 Oct 2005].
New South Wales National Park s and Wildlife Service 2001. C-Plan: User manaul. Armidale:
New South Wales National Parks and Wildlife Service.
Nexen 2007. PHP stats evolution for August 2007 [online]. Available from
http://www.nexen.net/chi ffres_cles/phpversion/17507-
php_stats_evolution_for_august_2007.php [Accessed 2 November 2007].
Nielsen J 1994. Ten usability heur istics [online]. Available from
http://www.useit.com/papers/h euristic/heuristic_list.html [Accessed 4 September 2007].
Nisar Ahamed TR, Rao GK & Murthy JSR 2000. GIS-based fuzzy membership model for crop-
land suitability analysis. Agricultural Systems 63: 75-95.
Open Geospatial Consortium 2007. OpenGIS web map service (WMS) implementation
specification [online]. Available from http://www.opengeospatial.org/standards [Accessed 24
December 2007].
Open Source Initiative 2007. Home [online]. Available from http://www.opensource.org/
[Accessed 13 May 2008].
Oracle 2007. Oracle spatial & Oracle locator: Location features for Oracle database 11g [online].
Available from http://www.oracle.com/technolo gy/products/spatial/index.html [Accessed 22
November 2007].
Orlandi E, Ruga L, Romano B & Fornaciari M 2005. Olive flowering as an indicator of local
climate changes. Theoretical Applied Climatology 81: 169-176.
Ortega-Farias SO & Leon L 2002. Models for pr edicting apple diameter by using growing-
degree days, cultivar Royal Gala. Acta Horticulturae 584: 163-176.
Oxford English Dictionary 2008. Compact Oxford English Dictionary [online]. Oxford: Oxford
University Press. Available from http://www.askoxford.com/concise_oed/website?view=uk
[Accessed 31 January 2008].
Park SJ, McSweeney K & Lowery B 2001. Identifica tion of the spatial distribution of soils using
a process-based terrain characterization. Geoderma 103: 249-272.
Patterson G 2005. Land type survey of South Afri ca completed [online]. ARC. Available from
http://www.arc-iscw.agric.z a/main/topnews/landsurvey.htm [Accessed 23-06-2005].
195
Peterson M 2003. Maps and the Internet: An introduction. In Patterson M (ed) Maps and the
Internet , 1-16. Oxford: Elsevier.
Pettey C 2007. Gartner says worldwide relationa l database market increased 14 percent in 2006
[online]. Stamford: Gartner. Available from http://www.gartner.com/it/page.jsp?id=507466
[Accessed 5 November 2007].
Phua M-H & Minowa M 2005. A GIS- based multi-criteria decision making approach to forest
conservation planning at a landscape scale: A case study in the Kinabalu Area, Sabah,
Malaysia. Landscape and Urban Planning 71: 207-222.
Pickup G & Chewings VH 1996. Correlations betw een DEM-derived topographic indices and
remotely-sensed vegetation cover in rangelands. Earth Surface Processes and Landforms 21:
517-529.
Proctor W & Qureshi E 2005. Multi-criteria ev aluation revisited [online]. Available from
http://www.anzsee.org/anzsee2005papers/ Proctor_Multi-criteria_evaluation.pdf [Accessed 13
May 2008].
Pummakarnchana O, Tripathi N & Dutta J 2005. Air pollution monitoring and GIS modeling: A
new use of nanotechnology based solid state gas sensors. Science and Technology of
Advanced Materials 6: 251-255.
Reiger M 2006. Introduction to fruit crops. New York: Haworth Press.
Research Surveys 2006. Latest SA Web user su rvey results released [online]. Cape Town.
Available from http://www.webchek.co.za/about_press%20saweb2006.html [Accessed 21
Aug 2006].
Reuters Foundation 2005. AlertNet [ online]. Reuters. Available from http://www.alertnet.org/
[Accessed 18 Oct 2005].
Rodriguez E, Morris CS, Belz JE, Chaplin EC, Martin JM, Daffer W & Hensley S 2005. An
assessment of the SRTM topographic products. Pasadena: Jet Propulsion Laboratory.
Rossiter DG 2001. The automated land evaluation system (ALES) [online]. New York: Cornell
University. Available from http://www.css.cornell.edu/landeval/ales/ales.htm [Accessed 13
May 2008].
Rossiter DG & Van Wambeke AR 1997. Automated land evaluation system ALES version 4.65
user?s manual. New York: Cornell University.
Saaty TL 1977. A scaling method for prio rities in hierarchical structures. Journal of
Mathematical Psychology 15: 234-281.
196
Saaty TL 1998. Ranking by eigenvector versus other methods in the analytic hierarchy process.
Applied Mathematics Letters 11: 212-125.
Saaty TL 2003. Decision-making with the AHP: W hy is the principal eigenvector necessary.
E uropean Journal of Operational Research 145: 85-91.
Saaty TL & Vargas LG 1991. Prediction, projection and forcasting. Boston: Kluwer Academic
Publishers.
Saayman D 1981. Klimaat, grond en wingerdbou gebiede. In Burger J & Deist J (eds)
Wingerdbou in Suid-Afrika , Cape Town: Maskew Miller.
Salewicza KA & Nakayama M 2004. Development of a web-based decision support system
(DSS) for managing large international rivers. Global Environmental Change 14: 25?37.
SANBI s.d. CAPE: Cape Action for People and the Environment [online]. Cape Town: SANBI.
Available from http://cpu.uwc.ac.za/CAPE/index.asp#CAPE [Accessed 13 October 2006].
SAWIS 2008. Statistics of wine-grape vines as on 30 November 2007. Paarl: SAWIS.
SAWS 2007. South Africa Weather Service bac kground [online]. Pretoria. Available from
http://www.weathersa.co.za/Corporate/History.jsp [Accessed 11 October 2007].
Schulze RE 1997. South African atlas of agrohydrol ogy and -climatology. Report TT82/96.
Pretoria: Water Research Commission.
Selvidge P 1999. How long is too lo ng to wait for a website to load? Usability News 1.2: s.p.
Sicat RS, Carranza EJM & Nidumolu UB 2005. Fu zzy modeling of farmers' knowledge for land
suitability classification. Agricultural Systems 83: 49-75.
Smith CS, McDonald GT & Thwaites RN 2000. TIM: Assessing the sustainability of agricultural
land management. Journal of Environmental Management 60: 267-288.
Soil Survey Division Staff 1993. Soil survey manual. Washington DC: Soil Conservation
Service, U.S. Department of Agriculture.
South Africa 1984. Act on conservation of agricultural resources, Act 43 of 1983. Government
Gazette of South Africa 9238, 25.5.1984.
South Africa 1997. White paper on the conserva tion and sustainable use of South Africa's
biological diversity [online]. Pr etoria: Department of Environmental Affairs and Tourism.
Available from
http://www.environment.gov.za/PolLeg/W hitePapers/Biodive rsity/Contents.htm [Accessed 16
August 2004].
197
South Africa 2000. Promotion of access to information act, Act 2 of 2000. Government Gazette
of South Africa 20852.
Speight JG 1977. Landform pattern de scription from aerial photographs. Photogrammetria 32:
161-182.
Statistics South Africa 2001. Census 2001: Key results. Pretoria: Statistics South Africa.
Statistics South Africa 2006a. Gross domestic product. Pretoria: Statistics South Africa.
Statistics South Africa 2006b. Survey of large scale agriculture 2005. Preliminary. Pretoria:
Statistics South Africa.
Statistics South Africa 2007. Mi d-year population estimates, 2007 [ online]. Pretoria: Statistics
South Africa. Available from
http://www.statssa.gov.za/publications /statsdownload.asp?PPN=P0302&SCH=3952
[Accessed 30 June 2008].
Sun Microsystems 2008. The Source for Java De velopers [online]. Available from [Accessed 1
July 2008].
Taylor DRF 2005. The theory and practice of cybe rcartography: An introduction. In Taylor DRF
(ed) Cybercartography: theory and practice , 1-13. Amsterdam: Elsevier.
Thompson JA, Bell JC & Butler CA 2001. Digital el evation model resolution: Effects on terrain
attribute calculation and quantitative soil-landscape modeling. Geoderma 100: 67-89.
Thompson MW 1999. South African national land- cover database project: data users manual.
Pretoria: CSIR.
Thwaites RN & Slater BK 2000. Soil-landscape resource assessment for plantations ? a
conceptual framework towards an explicit multi-scale approach. Forest Ecology and
Management 128: 123-138.
Thysen I & Detlefsen NK 2006. Online deci sion support for irrigation for farmers. Agricultural
Water Management 86: 269-276.
Traintaphyllou E 2000. Multi-criteria decision-making methods: A comparative study. London:
Kluwer Academic.
Tsou MH 2003. An intelligent software agen t architecture for distributed cartographic
knowledge bases and Internet mapping services. In Peterson MP (ed) Maps and the Internet ,
231-245. Amsterdam: Elsevier.
198
Tsoumakas G & Vlahavas I 1999. ISLE: An intelligent system for land evaluation . ACAI '99
Workshop on Intelligent Techniques for Spatio-T emporal Data Analysis in Environmental
Applications, Chania, Greece.
Turner D 2005. Transkei and Ciskei land type survey [online]. ARC. Available from
http://www.arc-iscw.agric.z a/main/projects/landtype.htm [Accessed 23-06-2005].
Twery MJ, Knopp PD, Thomasma SA, Rauscher HM, Nute DE, Potter WD, Maier F, Wang J,
Dass M, Uchiyama H, Glende A & Hoffman RE 2005. NED-2: A decision support system for
integrated forest ecosystem management. Computers and Electronics in Agriculture 49: 24-
43.
U.S. Department of Health and Human Services 2006. Research-based Web design & usability
guidelines [online]. Available from http://www.usability.gov/pdfs/guidelines.html [Accessed
6 September 2007].
U.S. Department of Labor 2006. Career voyages [ online]. U.S. Department of Labor. Available
from http://www.careervoyages.gov/index.cfm [Accessed 9 February 2006].
Vahidov R & Kersten GE 2004. Decision statio n: situating decision support systems. Decision
Support Systems 38: 283-303.
Valentini N, Me G, Spanna F & Lovisetto M 2004. Chilling and heat requirement in apricot and
peach varieties. Acta Horticulturae 636: 199-203.
Van der Merwe JH 1997. GIS-aided land evalua tion and decision-making for regulating urban
expansion: A South African case study. GeoJournal 43: 135-151.
Van der Merwe JH 2006. Multi-criteria evalua tion for land planning: Urban application.
Presentation given to the Department of Environmental Affairs and Development Planning,
Provincial Government of the Western Ca pe, Stellenbosch, Stellenbosch University.
Van der Merwe JH, Ferreira SLA & Van Niekerk A 2008. A spatial gap-analysis of tourism
development opportunity in the West ern Cape Province. Report 01/2008. Stellenbosch:
Stellenbosch University.
Van der Merwe JH & Steyl I 2005. Rural solid waste management: A planning strategy for
higher density agricultural regions. Journal of Public Administration 40: 295-313.
Van der Merwe JH & Von Holdt DS 2006. Environmen tal footprint of aircraft noise exposure at
Cape Town International Airport. The South African Geographical Journal 88: 177-193.
Van Niekerk A 1997. Die ontwikkeling van geografi ese inligtingstelsels vir omgewingsbestuur
in die Wes-Kaap . MSc thesis. Stellenbosch: Stellenbosch University.
199
Van Niekerk A 2001. Western Cape Digital Eleva tion Model: Product description. Stellenbosch:
Centre for Geographical Analys is, Stellenbosch University.
Van Niekerk A & Schloms BHA 2001. Automated mapping of la nd components from elevation
data. Paper presented at the Fourth Biennial Inte rnational Conference of the Society of South
African Geographers, Rawsonville.
Van Niekerk A & Schloms BHA 2002. A comparison of automatically mapped land components
with large-scale soil maps . Paper presented at the Regional Conference of the International
Geographical Union, Durban.
Van Ranst E, Tang H, Groenemans R & Sinthura hat S 1996. Application of fuzzy logic to land
suitability for rubber production in peninsular Thailand. Geoderma 70: 1-19.
Van Wyngaarden R & Waters N 2007. An unfinis hed revolution ? gainin g perspective on the
future of GIS. GeoWorld September: s.p.
Varma VK, Ferguson I & Wild I 2000. Decision support system for the sustainable forest
management. Forest Ecology and Management 128: 49-55.
Vreeker R, Nijkamp P & Ter Welle C 2002. A multicriteria decision support methodology for
evaluating airport expansion plans. Transportation Research 7: 27-47.
W3C 2002. XHTML 1.0 The Extensible HyperText Markup Language (Second Edition)
[online]. Available from http://www.w3.org/TR/xhtml1/ [Accessed 1 July 2008].
W3C 2006. Extensible Markup Language (XML) 1. 1 (Second Edition) [online]. Available from
http://www.w3.org/TR/xml11/ [Accessed 1 July 2008].
W3C 2008. HTML 5 [online]. Available from http://www.w3.org/html/wg/html5/ [Accessed 4
April 2008].
W3Schools 2008a. DHTML is the art of comb ining HTML, JavaScript, DOM, and CSS.
[online]. Available from [Accessed 1 July 2008].
W3Schools 2008b. SQL join [online]. Available from
http://www.w3schools.com/sql/sql_join.asp [Accessed 5 February 2008].
Walther BA, Schaffer N, Van Niekerk A, Thu iller W, Rahbek C & Chown SL 2007. Modelling
the winter distribution of a rare and endangered migrant, the Aquatic Warbler Acrocephalus
paludicola . Ibis 149: 701-714.
200
Wandahwa P & Van Ranst E 1996. Qualitative land suitability assessment for pyrethrum
cultivation in west Kenya based upon computer-captured expert knowledge and GIS.
Agriculture, Ecosystems & Environment 56: 187-202.
Wang F, Hall GB & Subaryono 1990. Fuzzy inform ation representation and processing in
conventional GIS software: Data base design and application. International Journal of
Geographical Information Systems 4: 261-283.
Wang K-J & Chein C-F 2003. Designing an Inte rnet-based group decision support system.
Robotics and Computer Integrated Manufacturing 19: 65-77.
Wang WK 2005. A knowledge-based decision support system for measuring the performance of
government real estate investment. E xpert Systems with Applications 29: 901-912.
Wikipedia 2005. Internet [online]. Wiki media Foundation Inc. Available from
http://en.wikipedia.org/wiki/Internet [Accessed 18 Oct 2005].
Wood J 2006. Landserf [online]. London: City University. Available from
http://www.landserf.org/ [Accessed 16 February 2006].
Wood LJ & Dragicevic S 2007. GIS-based multicrit eria evaluation and fuzzy sets to identify
priority sites for marine protection. Biodiversity and Conservation 16: 2539-2558.
Yalcin G & Akyurek Z 2004. Analysing flood vul nerable areas with multicriteria evaluation
[online]. Available from http://cartesia.org/geodoc/isprs2004/comm2/papers/154.pdf
[Accessed 13 May 2008].
Zadeh L 1965. Fuzzy sets. Information and Control 8: 338-353.
Zhou Q & Liu X 2004. Analysis of errors of de rived slope and aspect related to DEM data
properties. Computers & Geosciences 30: 369-378.
201
PERSONAL COMMUNICATIONS
Moss D 2006. Interview about land evaluation systems. Stellenbosch, Dennis Moss Partnership
(11 November 2006).
Schloms BHAS 2007. Interview about soil science and pedogenesis. Stellenbosch, Stellenbosch
University (18 September 2007).
Schloms BHAS 2007. Interview about the environmental requirements of perennial crops.
Stellenbosch, Stellenbosch University (29 February 2008).
Van der Merwe A (ajvdmerw@pgwc.gov.za) 2008. RE: ArcIMS. Email to A van Niekerk
(avn@sun.ac.za) (7 July 2008).
Van der Merwe JH 2008. Discussion on AHP process. Stellenbosch, Stellenbosch University (30
April 2008).
202
APPENDICES
Appendix A Avenue script to extract soil effective depth, clay content in the A
horizon and mechanical limitations from the land type data
203
Appendix B Automatic Land Component Mapper (ALCoM) Avenue script 209
Appendix C Avenue script to extract land property data from raster datasets for
each land unit in the land unit database
215
Appendix D Field and integrity rule descriptions for each entity in the knowledge
base
216
Appendix E Visual Basic procedure to calculate each land unit?s suitability based
on the land use requirements in the knowledge base
221
Appendix F ArcIMS map configuration file 223
Appendix G CD containing the CLUES website source code 224
203
APPENDIX A
Avenue script to extract soil effective depth, clay content in the A horizon and mechanical
limitations from the land type data
' *INITIALIZE COMMON VARIABLES*
theProject = av.GetProject
theView = av.GetActiveDoc
aPrj = theView.GetProjection
' *FIND INPUT THEME*
theThemes = theView.GetThemes
theTheme = MsgBox.ListAsString(theThemes, "Please select the Land Type theme","Land Type")
theFtab = theTheme.GetFTab
' *FIND LANDTYPE TABLES*
theTable1 = theProject.FindDoc("nuut_d.dbf")
theTable2 = theProject.FindDoc("nuut_g.dbf")
theTable3 = theProject.FindDoc("nuut_f.dbf")
theSoilSuitabilityTable = av.FindDoc("SoilSuit.dbf")
theIntensiveSuitField = "Int"
if (theFTab.StartEditingWithRecovery) then ' START EDITING
' *CREATE NECESSARY FIELDS*
if (theFTab.FindField("LT_AD") = Nil) then
theFTab.AddFields({Field.Make("LT_AD", #FIELD_DECIMAL, 6, 0)})
end
if (theFTab.FindField("LT_AML") = Nil) then
theFTab.AddFields({Field.Make("LT_AML", #FIELD_DECIMAL, 6, 2)})
end
if (theFTab.FindField("LT_ACA") = Nil) then
theFTab.AddFields({Field.Make("LT_ACA", #FIELD_DECIMAL, 6, 2)})
end
if (theFTab.FindField("LT_Series1") = Nil) then
theFTab.AddFields({Field.Make("LT_Series1", #FIELD_CHAR, 250, 0)})
end
if (theFTab.FindField("LT_Series2") = Nil) then
theFTab.AddFields({Field.Make("LT_Series2", #FIELD_CHAR, 250, 0)})
end
' *LINK THE APPROPRIATE TABLES*
if (theTable1 <> Nil) then
theVTab1 = theTable1.GetVTab
theFTab.Link(theFTab.FindField("LandType"),theVTab1,theVTab1.FindField("Landtype"))
IsLink = theVTab1.IsLinked
else
MsgBox.Error("1","")
return nil
end
if (theTable2 <> Nil) then
theVTab2 = theTable2.GetVTab
theFTab.Link(theFTab.FindField("LandType"),theVTab2,theVTab2.FindField("LandType"))
IsLink = theVTab2.IsLinked
204
else
MsgBox.Error("2","")
return nil
end
theSelection = theFTab.GetSelection
theSelection.SetAll
numberofunits = theSelection.Count
' *GET INFORMATION FOR EACH LAND TYPE*
for each rec in 0..(numberofunits - 1) ' FOR EACH LANDTYPE
if (theFTab.ReturnValue(theFTab.FindField("LandType"),rec) <> "") then
' *SELECT CURRENT LAND TYPE*
theSelection.ClearAll
theSelection.Set(rec)
theFTab.SetSelection(theSelection)
theSelection2 = theVTab1.GetSelection
theSelection3 = theVTab2.GetSelection
AverageDepth = 0
AverageSoilScore = 0
AverageMechanicalLimitation = 0
MechanicalLimitation = 0
TotalSeries = ""
Depth = 0
AverageClayA = 0
Clay = 0
theSoilSortList = {}
theSoilSortPercList = {}
' *FIND LIST OF TERRAIN TYPES ASSOCIATED WITH THIS LAND TYPE*
for each rec2 in theSelection2 ' FOR EACH TERRAIN TYPE
' *TEMPORARILY STORE TERRAIN INFORMATION*
thePercentageOfTerrain =
theVTab1.ReturnValue(theVTab1.FindField("Terrain_p"),rec2)
terrain = theVTab1.ReturnValue(theVTab1.FindField("Terrain_u"), rec2)
AverageSoilScoreOfTerrainType = 0
AverageMechanicalLimitationOfTerrainType = 0
AverageDepthOfTerrainType = 0
AverageClayAOfTerrainType = 0
TotalSeriesOfTerrainType = ""
' *GET LIST OF SOIL TYPES ASSOCIATED WITH EACH TERRAIN TYPE*
for each rec3 in theSelection3 ' FOR EACH SOIL SERIES
thePercOfSoil = theVTab2.ReturnValue(theVTab2.FindField("Soil_p"),rec3)
terrain2 = theVTab2.ReturnValue(theVTab2.FindField("Terrain_u"), rec3)
if (terrain = terrain2) then ' LOOK FOR CURRENT TERRAIN TYPE IN SOIL TABLE
' *GET AVERAGE SOIL DEPTH ASSOCIATED WITH EACH SOIL IN THE CURRENT TERRAIN TYPE*
if (theVTab2.ReturnValue(theVTab2.FindField("Soil_d_t"),rec3) = "-") then
Depth = theVTab2.ReturnValue(theVTab2.FindField("Soil_d_u"),rec3) +
((theVTab2.ReturnValue(theVTab2.FindField("Soil_d_l"),rec3) ?
theVTab2.ReturnValue(theVTab2.FindField("Soil_d_u"),rec3)) / 2)
elseif (theVTab2.ReturnValue(theVTab2.FindField("Soil_d_t"),rec3) = "<")
then
Depth = theVTab2.ReturnValue(theVTab2.FindField("Soil_d_l"),rec3) / 2
elseif (theVTab2.ReturnValue(theVTab2.FindField("Soil_d_t"),rec3) = ">")
then
205
Depth = theVTab2.ReturnValue(theVTab2.FindField("Soil_d_l"),rec3)
else
Depth = 0
end
AverageDepthOfTerrainType = AverageDepthOfTerrainType + ((thePercOfSoil /
100) * Depth)
' *GET AVERAGE CLAY CONTENT ASSOCIATED WITH EACH SOIL'S A HORIZON IN THE
CURRENT TERRAIN TYPE*
if (theVTab2.ReturnValue(theVTab2.FindField("Clay_a_t"),rec3) = "-") then
Clay = theVTab2.ReturnValue(theVTab2.FindField("Clay_a_l"),rec3) +
((theVTab2.ReturnValue(theVTab2.FindField("Clay_a_u"),rec3) ?
theVTab2.ReturnValue(theVTab2.FindField("Clay_a_l"),rec3)) / 2)
elseif (theVTab2.ReturnValue(theVTab2.FindField("Clay_a_t"),rec3) = "<")
then
Clay = theVTab2.ReturnValue(theVTab2.FindField("Clay_a_l"),rec3) / 2
elseif (theVTab2.ReturnValue(theVTab2.FindField("Clay_a_t"),rec3) = ">")
then
Clay = theVTab2.ReturnValue(theVTab2.FindField("Clay_a_u"),rec3)
else
Clay = 0
end
AverageCLayAOfTerrainType = AverageClayAOfTerrainType +
((thePercOfSoil / 100) * Clay)
' *GET_TOTAL_SOIL_SERIES*
theSeries = theVTab2.ReturnValue(theVTab2.FindField("Series"),rec3)
if (theSeries = "") then
theSeries = theVTab2.ReturnValue(theVTab2.FindField("Complex"),rec3)
end
TotalSeriesOfTerrainType = TotalSeriesOfTerrainType + theSeries +
"-" + thePercOfSoil.AsString + " "
' *GET_AVERAGE_MECHANICAL_LIMITATIONS*
MechanicalLimitation = theVTab2.ReturnValue(theVTab2.FindField("Mb"),rec3)
AverageMechanicalLimitationOfTerrainType =
AverageMechanicalLimitationOfTerrainType + ((thePercOfSoil / 100) *
MechanicalLimitation.AsNumber)
end ' LOOK FOR CURRENT TERRAIN TYPE IN SOIL TABLE
end ' FOR EACH SOIL SERIES
theSoilList = {}
wordpos = 0
if (TotalSeriesOfTerrainType.Count > 0) then ' IF TotalSeriesOfTerrainType
NOT EMPTY
while (TotalSeriesOfTerrainType.Extract(wordPos) <> Nil)
theSoilList.Add(TotalSeriesOfTerrainType.Extract(wordPos))
wordPos = wordPos + 1
end
for each series in theSoilList ' FOR EACH SERIES
theLength = series.count
theEndPos = theLength - 1
theStartPos = series.indexof("-")
thePercLength = theEndPos - theStartPos
theSeriesLength = theStartPos
206
thePerc = series.Right(thePercLength).AsNumber
theSeries = series.Left(theStartPos)
theSeriesLength = theSeries.Count
charPos = 0
theBreakList = {}
while (charPos < theSeriesLength) ' SEARCH ENTIRE SERIES FOR BREAKS
test = theSeries.Middle(charPos,1).AsAscii
if (theSeries.Middle(charPos,1).AsAscii <>
theSeries.Middle(charPos,1).LCase.AsAscii) then
' IF FIRST CHARACTER IS UPPER CASE
theBreak = theSeries.Middle(charPos,2)
if (theBreak.Count > 1) then
' IF THE BREAK CONSISTS OF MORE THAN ONE CHARACTER
if (theBreak.Middle(1,1).AsAscii <>
theBreak.Middle(1,1).UCase.AsAscii) then
' IF SECOND CHARACTER IS LOWER CASE
theBreakList.Add(theSeries.Middle(charPos,2))
end ' IF SECOND CHARACTER IS LOWER CASE
end ' IF THE BREAK CONSISTS OF MORE THAN ONE CHARACTER
end ' IF FIRST CHARACTER IS UPPER CASE
charPos = charPos + 1
end ' SEARCH ENTIRE SERIES FOR BREAKS
theBreakList.RemoveDuplicates
theNewSeries = ""
for each breakChar in theBreakList
theSeries = theSeries.Substitute(breakChar," "+breakChar)
end
wordPos = 0
theSoilList2 = {}
while (theSeries.Extract(wordPos) <> Nil)
theSoilList2.Add(theSeries.Extract(wordPos))
wordPos = wordPos + 1
end
theNumberOfSoils = theSoilList2.Count
for each series2 in theSoilList2
theSoilSortList.Add(series2)
'Number.SetDefFormat( "d.d" )
theSoilSortPercList.Add(((thePercentageOfTerrain * (thePerc /
theNumberOfSoils))/100).SetFormat("d.d"))
end
end ' FOR EACH SERIES
end ' IF TotalSeriesOfTerrainType NOT EMPTY
AverageDepth = AverageDepth + ((thePercentageOfTerrain / 100) *
AverageDepthOfTerrainType)
AverageClayA = AverageClayA + ((thePercentageOfTerrain / 100) *
AverageClayAOfTerrainType)
AverageMechanicalLimitation = AverageMechanicalLimitation +
((thePercentageOfTerrain / 100) *
AverageMechanicalLimitationOfTerrainType)
end ' FOR EACH TERRAIN TYPE
else
AverageDepth = -99
207
theSoilSortList = {}
AverageMechanicalLimitation = -99
AverageClayA = -99
end ' IF LANDTYPE <> ""
theDBListOfSoils = {}
theDBListOfPerc = {}
if (theSoilSortList.Count > 0) then
for each thePos1 in (0..(theSoilSortList.Count - 1))
soil = theSoilSortList.Get(thePos1)
perc = theSoilSortPercList.Get(thePos1)
thePos2 = theDBListOfSoils.FindByValue(soil)
if (thePos2 = -1) then ' SOIL IS NOT IN DB
theDBListOfSoils.Add(soil)
theDBListOfPerc.Add(perc)
else ' SOIL IS IN DB
theCurrentPerc = theDBListOfPerc.Get(thePos2)
theDBListOfPerc.Set(thePos2,theCurrentPerc + perc)
end
end
if (theDBListOfSoils.Count > 0) then
theDBListOfSoils.Sort(TRUE)
for each thePos1 in (0..(theDBListOfSoils.Count - 1))
soil = theDBListOfSoils.Get(thePos1)
perc = theDBListOfPerc.Get(thePos1)
TotalSeries = TotalSeries ++ soil+ "-" + perc.SetFormat("d").AsString
end
end
else
TotalSeries = ""
end
theFTab.SetValue(theFTab.FindField("LT_AD"),rec,AverageDepth)
if (TotalSeries.Count > 250) then
theFTab.SetValue(theFTab.FindField("LT_Series1"),rec,TotalSeries.Left(250))
theFTab.SetValue(theFTab.FindField("LT_Series2"),rec,
TotalSeries.Middle(250,TotalSeries.Count - 250))
else
theFTab.SetValue(theFTab.FindField("LT_Series1"),rec,TotalSeries)
theFTab.SetValue(theFTab.FindField("LT_Series2"),rec,"")
end
theFTab.SetValue(theFTab.FindField("LT_AML"),rec,AverageMechanicalLimitation)
theFTab.SetValue(theFTab.FindField("LT_ACA"),rec,AverageClayA)
end ' FOR EACH LANDTYPE
end ' START EDITING
theFTab.StopEditingWithRecovery(TRUE)
theFTab.UnlinkAll
theVTab1.UnlinkAll
theVTab2.UnlinkAll
208
APPENDIX B
Automatic Land Component Mapper (ALCoM) Avenue script
theView = av.GetActiveDoc
aPrj = theView.GetProjection
id = 1
'SET EXTENT AND CELL SIZE FOR CONVERSION IF NOT ALREADY SET
theAE = theView.GetExtension(AnalysisEnvironment)
theAE.SetExtent(#ANALYSISENV_MINOF,NIL)
theTheme = theView.GetActiveThemes.Get(0)
elevGrid = theTheme.GetGrid
theresultgrid = elevGrid
'GENERATE SLOPE FROM DEM
slgrid = elevgrid.Slope(1, FALSE)
slgrid = slgrid * 100
slgrid = slgrid.INT
elevgrid = elevgrid.INT
aspgrid = elevgrid.Aspect
'CLASSIFY ASPECT GRID
aspgrid = (aspgrid = -1).con(0.AsGrid, aspgrid)
aspgrid = ((aspgrid > 0) and (aspgrid < 22.5)).con(1.AsGrid, aspgrid)
aspgrid = ((aspgrid >= 22.5) and (aspgrid < 67.5)).con(2.AsGrid, aspgrid)
aspgrid = ((aspgrid >= 67.5) and (aspgrid < 112.5)).con(3.AsGrid, aspgrid)
aspgrid = ((aspgrid >= 112.5) and (aspgrid < 157.5)).con(4.AsGrid, aspgrid)
aspgrid = ((aspgrid >= 157.5) and (aspgrid < 202.5)).con(5.AsGrid, aspgrid)
aspgrid = ((aspgrid >= 202.5) and (aspgrid < 247.5)).con(6.AsGrid, aspgrid)
aspgrid = ((aspgrid >= 247.5) and (aspgrid < 292.5)).con(7.AsGrid, aspgrid)
aspgrid = ((aspgrid >= 292.5) and (aspgrid < 337.5)).con(8.AsGrid, aspgrid)
aspgrid = ((aspgrid >= 337.5) and (aspgrid < 360)).con(1.AsGrid, aspgrid)
aspgrid = aspgrid.INT
aspgrid = aspgrid.majorityfilter (True, True)
aspgrid = aspgrid.majorityfilter (True, True)
regiongrid = aspgrid.INT
regiongrid = regiongrid.RegionGroup(True, False, Nil)
regiongrid = regiongrid.majorityfilter (True, True)
regiongrid = regiongrid.majorityfilter (True, True)
finalGrid = regiongrid
finalTheme = GTheme.Make(finalGrid)
regStats = List.Make
regStats = regiongrid.GetStatistics
if (regStats.Count > 0) then
regmin = regStats.Get(0)
regmax = regStats.Get(1)
else
MsgBox.Error("Error in Region Grid","")
end
totalslopes = 0
' COUNT NUMBER OF SLOPES
for each i in regmin..regmax
209
totalslopes = totalslopes + 1
end
numberofslopes = 1
step = 0
Deleted1 = False
Deleted2 = False
`CLASSIFY EACH ASPECT REGION
for each i in regmin..(regmax)
' Make sure nothing is currently selected
theRegionVTab = regiongrid.GetVTab
theSelection = theRegionVTab.GetSelection
theSelection.ClearAll
System.BasicEcho (((numberofslopes / totalslopes)*100).Round.AsString + "% Done " +
step.AsString + " Recl ", TRUE)
' Select the current region
numberofslopes = numberofslopes + 1
step = 1
Success = theRegionVTab.Query( "([Value] = " + i.AsString+ ")
",theSelection,#VTAB_SELTYPE_NEW)
if (theSelection.Count > 0) then
totVariance = 0
' CONVERT CURRENT SELECTION TO GRID
totSlopeGrid = regiongrid.ExtractSelection
if (totSlopeGrid.HasError) then
return NIL
end
' CALCULATE TOTAL VARIANCE
' Use the grid to extract the slopes of the same area
totSlopeGrid = (totSlopeGrid / totSlopeGrid).INT * SlGrid
totVTab = totSlopeGrid.GetVTab
totVariance = 0
totStats = List.Make
totStats = totSlopeGrid.GetStatistics
if (totStats.Count > 0) then
totAverage = totStats.Get(2)
totmin = totStats.Get(0)
totmax = totStats.Get(1)
else
MsgBox.Error("Error in Slope Grid","")
end
totCells = 0
for each recno in totVTab
totCells = totCells + totVTab.ReturnValue(totVTab.FindField( "count"),
recno)
end
for each recno in totVTab
tvalue = totVTab.ReturnValue(totVTab.FindField( "value"), recno)
tcount = totVTab.ReturnValue(totVTab.FindField( "count"), recno)
totVariance = totVariance + (((tvalue - totAverage)^2)*tcount)
end
totVariance = totVariance / totCells
210
' Decide whether the area should be reclassified
if ((totCells > 50) and ((totmax - totmin) > 1) and (totVariance > 1)) then
' CREATE NEW LEGEND FOR SPOPE GRID USING NATURAL BREAKS
theSlopeTheme = GTheme.Make(totSlopeGrid)
theSlopeLegend = theSlopeTheme.GetLegend
theSlopeLegend.Natural(theSlopeTheme, "Value", 2)
' STORE GRID ACCORDING TO ITS NEW LEGEND
theField =
theSlopeLegend.GetFieldNames.Get(theSlopeLegend.GetFieldNames.Count
- 1)
aClassList = theSlopeLegend.GetClassifications
' the list of classifications and each classification needs to
' be cloned so that the labels in the original legend do not change
theClassList = aClassList.DeepClone
numberofclasses = 0
for each c in theClassList
numberofclasses = numberofclasses + 1
end
count = 1
' Add labels for each classification starting at 1
for each c in theClassList
if (count < numberofclasses) then
c.SetLabel(count.AsString)
count = count + 1
end
end
theResultGrid =
totSlopeGrid.ReclassByClassList(theField,theClassList,FALSE)
avList = List.Make
noc = 2
count = 1
DO = true
IsSmaller = true
' While the variance is high, reclassify
While (DO)
theResultVTab = theResultGrid.GetVTab
theSelection = theResultVTab.GetSelection
theSelection.ClearAll
Success = theResultVTab.Query( "([Value] = " + count.AsString+ ")
",theSelection,#VTAB_SELTYPE_NEW)
if (theSelection.Count > 0) then
CurrentGrid = theResultGrid.ExtractSelection
CurrentGrid = (CurrentGrid / CurrentGrid).INT
CurrentGrid = CurrentGrid * SlGrid
CurrentVTab = CurrentGrid.GetVTab
CurrentStats = List.Make
CurrentStats = CurrentGrid.GetStatistics
if (CurrentStats.Count > 0) then
CurrentAverage = CurrentStats.Get(2)
else
MsgBox.Error("Error in Current Grid","")
end
211
CurrentCells = 0
for each recno in CurrentVTab
CurrentCells = CurrentCells +
CurrentVTab.ReturnValue(CurrentVTab.FindField( "count"),recno)
end
CurrentVariance = 0
for each recno in CurrentVTab
tvalue =
CurrentVTab.ReturnValue(CurrentVTab.FindField( "value"),
recno)
tcount =
CurrentVTab.ReturnValue(CurrentVTab.FindField( "count"),
recno)
CurrentVariance = CurrentVariance + (((tvalue ?
CurrentAverage)^2)*tcount)
end
CurrentVariance = CurrentVariance / CurrentCells
if ((CurrentVariance < (0.5 * totVariance) or (Step > 5))) then
IsSmaller = True
count = count + 1
AvList = AvList.Add((CurrentAverage/100))
else
IsSmaller = False
noc = noc + 1
AvList.Empty
step = step + 1
end
' The variance is still high, reclassify
if (NOT IsSmaller) then
' CREATE NEW LEGEND FOR SPOPE GRID USING NATURAL BREAKS
theSlopeLegend.Natural(theSlopeTheme, "Value", noc)
' STORE GRID ACCORDING TO ITS NEW LEGEND
theField =
theSlopeLegend.GetFieldNames.Get(theSlopeLegend.GetFieldNames.Count
- 1)
aClassList = theSlopeLegend.GetClassifications
' the list of classifications and each classification needs to
' be cloned so that the labels in the original legend do not change
theClassList = aClassList.DeepClone
numberofclasses = 0
for each c in theClassList
numberofclasses = numberofclasses + 1
end
count = 1
for each c in theClassList
if (count < numberofclasses) then
c.SetLabel(count.AsString)
count = count + 1
end
end
theResultGrid =
212
totSlopeGrid.ReclassByClassList(theField,theClassList,FALSE)
count = 1
end
else
Do = False
' STORE GRID ACCORDING TO ITS NEW LEGEND
ResultTheme = GTheme.Make(theResultGrid)
theResultLegend = ResultTheme.GetLegend
theField = theResultLegend.GetFieldNames.Get
(theResultLegend.GetFieldNames.Count - 1)
aClassList = theResultLegend.GetClassifications
' the list of classifications and each classification needs to
' be cloned so that the labels in the original legend do not change
theClassList = aClassList.DeepClone
numberofclasses = 0
for each c in theClassList
numberofclasses = numberofclasses + 1
end
count = 1
for each c in theClassList
if (count <= numberofclasses) then
c.SetLabel(AvList.Get(count - 1).Round.AsString)
count = count + 1
end
end
theResultGrid =
theResultGrid.ReclassByClassList(theField,theClassList,FALSE)
end
end
theAE = theView.GetExtension(AnalysisEnvironment)
theAE.SetExtent(#ANALYSISENV_VALUE,elevgrid.GetExtent)
finalgrid = (theResultGrid.IsNull).Con(finalgrid,theResultGrid)
if (finalgrid.HasError) then return NIL end
FinalTheme = GTheme.Make(finalgrid)
else
theAE = theView.GetExtension(AnalysisEnvironment)
theAE.SetExtent(#ANALYSISENV_VALUE,elevgrid.GetExtent)
finalgrid =
(totSlopeGrid.IsNull).Con(finalgrid,(totAverage/100).AsGrid.INT)
if (finalgrid.HasError) then return NIL end
FinalTheme = GTheme.Make(finalgrid)
end
end
end
If (finalgrid.HasError) then
return NIL
else
theView.AddTheme(FinalTheme)
end
213
APPENDIX C
Avenue script to extract land property data from raster datasets for each land unit in the land unit
database.
? This script uses a blocks.shp file to break the land units into smaller more manageable units.
Any regions can be used. For CLUES, quarter-degrees were used.
theView = av.GetActiveDoc
thePrj = theView.GetProjection
theThemes = theView.GetThemes
theGridThemes = {}
theOutTheme = theView.GetActiveThemes.Get(0)
theFTab = theOutTheme.GetFTab
For each t in theThemes
if ((t.Is(GTHEME)) and (t.IsVisible)) then
theGridThemes.Add(t)
end
end
theBlocksTheme = theView.FindTheme("Blocks.shp")
theBlocksFTab = theBlocksTheme.GetFTab
if (theFTab.StartEditingWithRecovery) then
for each b in theBlocksFTab
theBlockShape = theBlocksFTab.ReturnValue(theBlocksFTab.FindField("Shape"),b)
theFtab.SelectByPolygon(theBlockShape,#VTAB_SELTYPE_NEW )
theBitMap = theFTab.GetSelection
for each t in theGridThemes
theName = t.GetName
theGrid = t.GetGrid
if (theFTab.FindField(theName) = Nil) then
theFTab.AddFields({Field.Make (theName,#FIELD_DECIMAL,5,0)})
end
aFN = "c:\proj\phd\zstat2.dbf".AsFileName
theVTab = theGrid.ZonalStatsTable(theFTab,thePrj,theFTab.FindField("Id"),FALSE,aFN)
theFTab.Join(theFTab.FindField("ID"),theVTab,theVTab.FindField("ID"))
theFTab.Calculate ("[MEAN]", theFTab.FindField(theName))
theFTab.UnjoinAll
end
end
end
theFTab.StopEditingWithRecovery(TRUE)
214
APPENDIX D
Field and integrity rule descriptions for each entity in the knowledge base
ENTITY: LAND USE
ATTRIBUTE DOMAIN AND TRIGGER RULES
LAND_USE_ID Data type: number
Format: integer
Uniqueness: unique
Null support: non-null
Insert trigger: none
Update trigger: not allowed
Delete trigger: ensure no child entities are present
USER_ID See USER entity
NAME Data type: text
Length: 100
Uniqueness: non-unique
Null support: non-null
ENTITY: LAND REQUIREMENT
ATTRIBUTE DOMAIN AND TRIGGER RULES
LAND_REQUIREMENT_ID
Data type: number
Format: integer
Uniqueness: unique
Null support: non-null
Insert trigger: none
Update trigger: not allowed
Delete trigger: ensure no child entities are present
LAND_USE_ID See LAND_USE entity
LAND_PROPERTY_ID See LAND_PROPERTY entity
WEIGHT Data type: number
Length: 4 digits
Format: decimal (two decimal places)
Uniqueness: non-unique
Null support: non-null
215
ENTITY: LAND_ REQUIREMENT_RULE
ATTRIBUTE DOMAIN AND TRIGGER RULES
LAND_REQUIREMENT_RULE_ID
Data type: number
Format: integer
Uniqueness: unique
Null support: non-null
Insert trigger: none
Update trigger: not allowed
Delete trigger: ensure no child entities are present
LAND_REQUIREMENT_ID See LAND_REQUIREMENT entity
SUITABILITY Data type: text
Length: 2 characters
Format: alphanumeric (options: N2, N1, S3, S2, S1)
Uniqueness: non-unique
Null support: non-null
LOWER_VALUE
Data type: number
Format: decimal (3 places)
Uniqueness: non-unique
Null support: non-null
MIDDLE_VALUE
Data type: number
Format: decimal (3 places)
Uniqueness: non-unique
Null support: non-null
UPPER_VALUE
Data type: number
Format: decimal (3 places)
Uniqueness: non-unique
Null support: non-null
CURVE_ID
Data type: number
Length: 1
Range: 1-2
Format: integer
Uniqueness: non-unique
Null support: non-null
ENTITY: LAND_PROPERTY
ATTRIBUTE DOMAIN AND TRIGGER RULES
LAND_PROPERTY_ID
Data type: number
Format: integer
Uniqueness: unique
Null support: non-null
Insert trigger: none
216
Update trigger: not allowed
Delete trigger: ensure no child entities are present
DATA_SOURCE_ID See DATA_SOURCE entity
NAME
Data type: text
Length: 50
Uniqueness: non-unique
Null support: non-null
UNIT
Data type: text
Length: 50
Uniqueness: non-unique
Null support: non-null
MIN
Data type: number
Format: decimal (3 places)
Uniqueness: non-unique
Null support: non-null
MAX
Data type: number
Format: decimal (3 places)
Uniqueness: non-unique
Null support: non-null
ENTITY: LAND_ UNIT
ATTRIBUTE DOMAIN AND TRIGGER RULES
LAND_UNIT_ID
Data type: number
Format: integer
Uniqueness: unique
Null support: non-null
Insert trigger: none
Update trigger: not allowed
Delete trigger: ensure no child entities are present
LAND_PROPERTY_ID See LAND_PROPERTY entity
VALUE Data type: numeric
Length: 10 digits
Range: 0.000 ? 9999999.000
Format: decimal (three decimal places)
Uniqueness: non-unique
Null support: non-null
ENTITY: PROJECT
ATTRIBUTE DOMAIN AND TRIGGER RULES
PROJECT_ID
Data type: number
Format: integer
217
Uniqueness: unique
Null support: non-null
Insert trigger: none
Update trigger: not allowed
Delete trigger: ensure no child entities are present
USER_ID See USER entity
LAND_USE_ID See LAND_USE entity
NAME
Data type: text
Length: 50
Uniqueness: non-unique
Null support: non-null
MODIFIED
Data type: text
Length: 50
Uniqueness: non-unique
Null support: non-null
MIN_X Data type: number
Format: integer
Uniqueness: non-unique
Null support: non-null
MAX_X Data type: number
Format: integer
Uniqueness: non-unique
Null support: non-null
MIN_Y Data type: number
Format: integer
Uniqueness: non-unique
Null support: non-null
MAX_Y Data type: number
Format: integer
Uniqueness: non-unique
Null support: non-null
ENTITY: DATA_SOURCE
ATTRIBUTE DOMAIN AND TRIGGER RULES
DATA_SOURCE_ID
Data type: number
Format: integer
Uniqueness: unique
Null support: non-null
Insert trigger: none
Update trigger: not allowed
Delete trigger: ensure no child entities are present
NAME Data type: text
218
Length: 50 characters
Uniqueness: non-unique
Null support: non-null
SCALE
Data type: text
Length: 50 characters
Uniqueness: non-unique
Null support: non-null
ORIGIN
Data type: text
Length: 50 characters
Uniqueness: non-unique
Null support: non-null
219
APPENDIX E
Visual Basic procedure to calculate each land unit?s suitability based on the land use
requirements in the knowledge base
Sub CalculateSuitability()
' ++ DECLARE VARIABLES
Dim ext, xMin, xMax, yMin, yMax, theExtentString, xBuffer, yBuffer, thePropertyID
Dim theWeight, theRequirementID, map, PARoRs2, theValue, SDEoRs2, SDEsql, SDEoRs
Dim PARCon, PARConnString, PARsql, PARoRs, PARsql2, SDECon, SDEConnString
' +01+ CONNECT TO LAND UNIT DATABASE
Set SDECon = Server.CreateObject("ADODB.Connection")
SDEConnString = "DBQ=" & Server.MapPath("../../data/")
SDECon.Open "Driver={Microsoft dBase Driver (*.dbf)};" & " DriverID=277;" & SDEConnString
' +02+ INITIALIZE SUITABILITY ITEM
SDEsql = "Update lu2.dbf Set S" & Session("theUserNo") & " = 0"
Set SDEoRs = SDECon.Execute(SDEsql)
' +03+ CONNECT TO KNOWLEDGE BASE
Set PARCon = Server.CreateObject("ADODB.Connection")
PARConnString = "DBQ=" & Server.MapPath("../data/db1.mdb")
PARCon.Open "DRIVER={Microsoft Access Driver (*.mdb)}; " & PARConnString
' +04+ RETRIEVE PROJECT INFORMATION
PARsql = "select FUNCTION, LAND_USE_ID from PROJECT WHERE ~
(PROJECT_ID = " & Session("theProjectID") & ")"
Set PARoRs = PARCon.Execute(PARsql)
' +05+ RETRIEVE LAND USE REQUIREMENTS FOR SELECTED LAND USE FROM KNOWLEDGE BASE
PARsql = "select LAND_PROPERTY_ID, WEIGHT, LAND_REQUIREMENT_ID from LAND_REQUIREMENT ~
WHERE ((LAND_REQUIREMENT.USER_ID = " & Session("theUserID") & ") ~
AND(LAND_REQUIREMENT.LAND_USE_ID = " & Session("theLandUseID") & "))"
Set PARoRs = PARCon.Execute(PARsql)
' +06+ CALCULATE AND SUMMARIZE SUITABILITY VALUES FOR EACH LAND USE REQUIREMENT
Do While (Not PARoRs.EOF)
thePropertyID = PARoRs(0)
theWeight = PARoRs(1)
theRequirementID = PARoRs(2)
' +06a+ GET RULES FOR CURRENT LAND USE REQUIREMENT FROM KNOWLEDGE BASE
PARsql2 = "select SUITABILITY, LOWER_VALUE, MIDDLE_VALUE, UPPER_VALUE, CURVE_ID from ~
LAND_REQUIREMENT_RULE WHERE ((LAND_REQUIREMENT_RULE.LAND_REQUIREMENT_ID = " & ~
theRequirementID & ") AND (LAND_REQUIREMENT_RULE.SUITABILITY LIKE 'S%'))"
Set PARoRs2 = PARCon.Execute(PARsql2)
Do While (NOT PARoRs2.EOF)
' +06b+ SET THE SUITABILITY FACTOR
SELECT CASE PARoRs2(0)
CASE "S1" theValue = 5
CASE "S2" theValue = 4
CASE "S3" theValue = 3
CASE "S3" theValue = 2
CASE "S3" theValue = 1
END SELECT
' +06c+ CALCULATE SUITABILITY
if (PARoRs2(4) = 1) then ' ++ BOOLEAN RULE
SDEsql = "Update lu2.dbf Set S" & Session("theUserNo") &
" = (S" & Session("theUserNo") & " + (" & theValue & " * " & theWeight & ")) ~
WHERE (P" & thePropertyID & " >= " & PARoRs2(1) & ") AND ~
(P" & thePropertyID & " < " & PARoRs2(3) & ") " & theExtentString
Set SDEoRs = SDECon.Execute(SDEsql)
elseif (PARoRs2(4) = 2) then ' ++ FUZZY RULE
' ++ LINE A (ALPHA >= X < BETHA)
SDEsql = "Update lu2.dbf Set S" & Session("theUserNo") & ~
" = (S" & Session("theUserNo") & " + (" & theValue & ~
" * ((1 / " & PARoRs2(2) - PARoRs2(1) & ") * (P" & thePropertyID & ~
" - " & PARoRs2(1) & ")) * " & theWeight & ")) WHERE (P" & thePropertyID & " >= " & ~
PARoRs2(1) & ") AND (P" & thePropertyID & " < " & PARoRs2(2) & ")" & theExtentString
Set SDEoRs = SDECon.Execute(SDEsql)
' ++ LINE B (BETA >= X < THETA)
SDEsql = "Update lu2.dbf Set S" & Session("theUserNo") & ~
" = (S" & Session("theUserNo") & " + (" & theValue & ~
" * ((1 / " & PARoRs2(2) - PARoRs2(3) & ") * (P" & thePropertyID ~
& " - " & PARoRs2(1) & ") + 1) * " & theWeight & ")) WHERE (P" & thePropertyID & " > " & ~
PARoRs2(2) & ") AND (P" & thePropertyID & " <= " & PARoRs2(3) & ")" & theExtentString
220
Set SDEoRs = SDECon.Execute(SDEsql)
end if
PARoRs2.MoveNext
Loop
PARoRs.MoveNext
Loop
Set PARoRs = PARCon.Execute(PARsql)
' +07+ RESET SUITABILITY TO 2 IF ANY LAND PROPERTY WAS FOUND TO BE UNSUITABLE AT PRESENT
Do While (Not PARoRs.EOF)
thePropertyID = PARoRs(0)
theWeight = PARoRs(1)
theRequirementID = PARoRs(2)
if (theWeight > 0) then
PARsql2 = "select SUITABILITY, LOWER_VALUE, MIDDLE_VALUE, UPPER_VALUE, CURVE_ID from
LAND_REQUIREMENT_RULE WHERE ((LAND_REQUIREMENT_RULE.LAND_REQUIREMENT_ID = " & ~
theRequirementID & ") AND (LAND_REQUIREMENT_RULE.SUITABILITY = 'N1'))"
Set PARoRs2 = PARCon.Execute(PARsql2)
Do While (NOT PARoRs2.EOF)
SDEsql = "Update lu2.dbf Set S" & Session("theUserNo") & " = 2 WHERE (P" & thePropertyID ~
& " >= " & PARoRs2(1) & ") AND (P" & thePropertyID & " < " & PARoRs2(3) & ")" & ~
theExtentString
Set SDEoRs = SDECon.Execute(SDEsql)
PARoRs2.MoveNext
Loop
end if
PARoRs.MoveNext
Loop
Set PARoRs = PARCon.Execute(PARsql)
' +08+ RESET SUITABILITY TO 1 IF ANY LAND PROPERTY WAS FOUND TO BE UNSUITABLE AT PRESENT
Do While (Not PARoRs.EOF)
thePropertyID = PARoRs(0)
theWeight = PARoRs(1)
theRequirementID = PARoRs(2)
if (theWeight > 0) then
PARsql2 = "select SUITABILITY, LOWER_VALUE, MIDDLE_VALUE, UPPER_VALUE, CURVE_ID from
LAND_REQUIREMENT_RULE WHERE ((LAND_REQUIREMENT_RULE.LAND_REQUIREMENT_ID = " & theRequirementID
& ") AND (LAND_REQUIREMENT_RULE.SUITABILITY = 'N2'))"
Set PARoRs2 = PARCon.Execute(PARsql2)
Do While (NOT PARoRs2.EOF)
SDEsql = "Update lu2.dbf Set S" & Session("theUserNo") & " = 1 WHERE (P" & thePropertyID & "
>= " & PARoRs2(1) & ") AND (P" & thePropertyID & " < " & PARoRs2(3) & ")" & theExtentString
Set SDEoRs = SDECon.Execute(SDEsql)
PARoRs2.MoveNext
Loop
end if
PARoRs.MoveNext
Loop
' ++ CLEAR TEMPORARY FIELDS
SDEsql = "Update lu2.dbf Set B" & Session("theUserNo") & " = 0"
Set SDEoRs = SDECon.Execute(SDEsql)
SDEsql = "Update lu2.dbf Set L" & Session("theUserNo") & " = 0"
Set SDEoRs = SDECon.Execute(SDEsql)
PARCon.Close
SDECon.Close
End Sub
221
APPENDIX F
ArcIMS map configuration file
222
APPENDIX G
CD containing the CLUES website source code (see back cover)