Doctoral Degrees (Soil Science)

Permanent URI for this collection


Recent Submissions

Now showing 1 - 5 of 23
  • Item
    Determination of optimal soil conditions and foliar nutrient levels in commercial rooibos tea production
    (2023-03) Smith, Jacobus Francois Naude; Hardie-Pieters, Ailsa G.; Hoffman, J. E.; Stellenbosch University. Faculty of Agrisciences. Dept. of Soil Science.
  • Item
    Evaluating soil and terrain variables in a production environment: implications for agricultural land assessment
    (Stellenbosch : Stellenbosch University, 2022-12) Barichievy, Kurt Russell; Clarke, Catherine E.; Rozanov, Andrei Borisovich; Stellenbosch University. Faculty of Agrisciences. Dept. of Soil Science.
    ENGLISH ABSTRACT: Agricultural land in South Africa is under increasing pressure to produce more food from an ever-shrinking land base, as more land is being converted to non-productive uses. Additional to these pressures, is the concept of land reform and strategic land acquisition, aimed at agrarian transform within the rural landscape. It is estimated that less than 15% of South Africa is suitable for dryland cultivation. Consequently, the sustainable utilisation of these scarce resources and preservation of agricultural land is of paramount importance, to ultimately ensure some measure of national food security in the years to come. Agricultural land evaluation is a critical tool that can achieve this goal. Unfortunately, in recent decades the development of revised or novel land evaluation methodologies has stalled for South African farm-level assessments, the scale at which land release decisions are made. Further, the relationship between productivity and individual land assessment attributes has not been adequately quantified or incorporated into contemporary local assessment procedures. It is envisaged that this study would influence and help guide in-field methodologies, as well as draft legislation and best-practice strategies, with a view of both standardising and improving agricultural land assessment techniques. By emphasising the importance of agricultural land and the accurate assessment thereof, this research also aims to increase our understanding of production-based approaches at an operational scale, though the novel combination of traditional approaches and use of newer technologies. It is anticipated that this improved understanding will be employed to not only protect more agricultural land, which may have been undervalued by historical methods, but also as an intuitive assessment tool to highlight the yield gap between potential and actual production levels. A review of pertinent literature identified the need for local verification studies to evaluate the performance of land assessment methodologies currently used in industry. To address this, five methods were verified using land assessment polygons in a commercial production environment, in the Province of KwaZulu-Natal, South Africa. The resultant classifications, derived from 225 soil observations, were compared to actual land use and precision yields achieved by dryland maize and soybean, across five growing seasons (2016 - 2020). By comparing land use with broad arability, four of the five land assessment methods were found to adequately classify arable land. Additionally, land evaluation polygons, linked to dryland precision maize and soybean yields can provide a general overview of method performance. However, it was concluded that yield performance and variation, across land evaluation methods and classes, is only explicit on or near a soil observation point where measurements are taken. Accordingly, seasonal variograms for maize and soybean were developed, to establish a representative yield buffer around individual soil observation points. This, along with yield normalisation strategies were employed, to improve verification procedures across multiple growing seasons. To determine crop productivity drivers, significant land assessment attributes inter alia slope, effective rooting depth, soil texture, soil group and soil wetness limitations were analysed against maize and soybean yields. It was found that the two crops respond differently to individual land assessment attributes and these differences should be taken cognisance of in new, crop-specific land evaluation methodologies and weighted accordingly. In an attempt to improve productivity-based land classification 78 attributes; derived from land assessment methodologies, digital terrain analysis, the pedological survey and soil colour spectrophotometry were collated. From these attributes, three new approaches, one based on biophysical scoring criteria and two based on machine learning, were developed across two commercial farming operations, in northern KwaZulu-Natal. These new methodologies were then tested on three separate commercial operations, located in different regions of the province. The biophysical scoring classification generally outperformed machine learning models and was particularly accurate when classifying observations associated with either extremely poor or extremely advantageous soil and terrain attributes. The transferability of the models to other regions, with different resources produced mixed results, highlighting the need for wider calibration in some instances. The study also found that the new productivity-based approaches can have useful applications in commercial farm management, where crop specific classification can identify underperforming areas and yields gaps, which can be ringfenced for appropriate interventions. The newly developed biophysical scoring classification was used to demonstrate the utility of these approaches in broader agricultural land release applications. The study found the new approaches better reflect production potential and should be used to supplement existing methodologies in land release assessments. Ultimately, the application of these production- based approaches can assist the land assessor to better classify the production potential of the land, as well as the decision-making authority to justify preserving more land for agricultural purposes.
  • Item
    Using remote sensing and geographical information systems to classify local landforms using a pattern recognition approach for improved soil mapping
    (Stellenbosch : Stellenbosch University, 2022-05) Atkinson, Jonathan Tom; De Clercq, W. P.; Rozanov, Andrei Borisovich; Stellenbosch University. Faculty of AgriSciences. Dept. of Soil Science.
    ENGLISH ABSTRACT: Presently, a major focus of digital soil mapping (DSM) in South Africa is unlocking the soil-landscape relationships of legacy soil data by disaggregating the only source of contiguous soil information for South Africa, the National Land Type Survey (LTS) (ARC, 2003). Each land type is best defined as a homogenous mapping unit with a unique combination of terrain type, soil pattern and macroclimate properties (Paterson et al., 2015). One of the prevailing reasons for the LTS longevity and continual temporal-interoperability is that terrain description is expressly related to a suite of catenary soil property descriptions (Milne, 1936). These terrain types are further divided into terrain morphological units (TMUs) representing a sequence of patterns based on a 5-unit landscape model of 1-crest, 2-scarp, 3-midslope, 4-footslope and 5-valley bottom. Importantly, dominant soil distribution patterns are defined by terrain units relying on an elementary terrain topo-sequence pattern approach, with much of the work done on modelling soil variation related to variation in terrain (van Zijl, 2019). Whilst the LTS remains a source of national interest, there is immense opportunity to build on the existing soil inventory data rather than only focus on “breaking it down” (disaggregation). However, what is needed is a standard operating procedure that not only leverages the ability of digital elevation models (DEM) to explicate soil-landscape associations beyond the limited 5-unit landscape model but allows better refinement of soil descriptions with landscape features. Only once the nuances of optimal DEM parametrisation under controlled conditions are fully understood can the complete scope of DSM and digital geomorphological mapping (DGM) applications be explored. This dissertation attempts to synthesise knowledge on theory, methods, and applications of using remote sensing (RS) and geographical information systems (GIS) to classify local landforms using a pattern recognition approach for improved soil mapping in the context of multiscale problems of digital terrain analysis in KwaZulu-Natal. The dissertation is divided into three parts. Part one (Chapter 2) represents the DEM pre- processing and generalisation method and establishes the protocols for soil-landscape covariate application derived from various sensor platforms and spatial scales. Part two (Chapter 3) introduces the concept of improved terrain unit mapping through the geomorphon approach and describes DEM optimisation for standardised geomorphon representation for uniformly describing soil-landscape properties for inputs to DSM applications. Finally, part three (Chapters 4 & 5) looks at applications of DEM sources and geomorphons first from a holistic landscape context by linking digital terrain and soil-landscape analysis to geodiversity. Finally, the benefit of improved RS and GIS combined with quantitative modelling approaches on improving natural resource predictions are explored by modelling soil-ecotope and soil type mapping units and proposing improvements to an existing DSS designed for KwaZulu-Natal Natal. Specifically, this research is organised into four (4) research chapters with an overview of each chapter’s contribution outlined hereafter. Chapter 2 accounts for the recognition and requirements of DEM generalisation from high to medium resolution RS platforms and the influence these pre-processing approaches have on the extraction of a wide range of terrain attributes. Digital elevation data are elemental in deriving primary topographic attributes that are input variables to various regional soil-landscape models. DEMs' utility to extract different topographic indices as primary inputs to DSM allows the generalised soil-formative relationship between topography and soil characteristics to be measured quantitatively. Traditional landscape-scale approaches to extracting and analysing soils remain subjective and an expensive last resort for large-scale regional soil distribution and variability prediction. Selecting the right DEMs is a critical step in the development of any soil-landscape model. Therefore, the ability to represent soil-landscape relationships rapidly and objectively between soil properties and landscape position using emerging technologies and elevation data in a digital environment and at varying scales is fundamental for using soil-landscape mapping as a regional planning tool. There is, however, still varied consensus on the effect of DEM source and resolution on the application of these topographic attributes to landscape and geomorphic characterisation within South Africa. However, Atkinson et al. (2017) have shown that topographic variable extraction is highly dependent on the DEM source and generalisation approach. However, while higher resolution DEMs may represent the “true” landscape surface more accurately, they do not necessarily offer the best results for all extracted terrain variables for modelling soil-landscape outputs. Given the convenience of a wide range of open-source elevation data for South Africa, there is a need to quantify the impact that DEM generalisation approaches have on simplifying detailed DEMs and compare the accuracy and reliability of results between high resolution and coarse resolution data on the extraction of localised topographic variables as a primer for soil-landscape or digital soil models. Chapter 3 explores the harmonisation of geomorphons derived from various RS platforms to define the landscape character in central KwaZulu-Natal. Robust DGM approaches that can simplify and translate the inclusion of “human knowledge” to automatic terrain classification across a broader spectrum of terrain morphological units and a range of DEM spatial scales offer great potential for improved topographic and landscape analysis and must have their utility investigated. Continual advances in quantitative modelling of surface processes, combined with new spatio-temporal and geo-computational algorithms, have revolutionised the auto-classification and mapping of landform components through the automated analysis of high-quality DEMs. Therefore, a thorough assessment of the effects that different pixel resolution (grain size) and DEM sources have on replicating observed geomorphic spatial patterns and representing selected terrain parameters using advanced automated geomorphometric mapping approaches is necessary. Specifically, it would be valuable to interrogate the self-adapting ability of these automated mapping approaches under regional conditions to quantitatively analyse how the choice of terrain model and scale influences the extraction, generalisation, and representation of digitally derived terrain attributes such as slope gradient, elevation and terrain unit feature extent. Equally important is understanding how the variation in resulting terrain unit representation is limited by spatial resolution discontinuities that ultimately influence the extraction and representation of elementary soil properties. Chapter 4 is a shift from the technical aspects of digital terrain preprocessing and modelling and instead attempts to explore the contribution of gridded soil-landscape products to the abiotic landscape development agenda. It would be worthwhile to contextualise and decode these technical aspects of terrain and soil analyses to a holistic landscape development agenda. It is argued that current global environmental problems and questions demand exploration into new scientific perspectives and improved related paradigms and methodologies. Geodiversity (abiotic complexity) has not received the same level of attention as biodiversity (biotic complexity) despite its intrinsic and indivisible linkages to ecosystem and landscape richness characterisation. The ability to better describe the substrate in which biological and human activities occur is of top standing and must have its potential explored. To date, only one landmark study has successfully investigated the influence of environmental factors on geodiversity mapping in South Africa (Kori et al., 2019). Using an array of multimodal environmental covariates, including hydrographic, lithostratigraphic, pedological, climatic, topographic, solar morphometric and geomorphic variables, I aim to provide further confirmation to regional and international geodiversity research agendas. Chapter 5 culminates in applying quantitative DSM methods, with improved terrain representation, to classify productive soil units (ecotopes) as a proposed methodology to improve the current Bioresource Report Writer (BRW) soil-landscape recommendations. In KwaZulu-Natal, it has been accepted that detailed natural resource information based on scientifically accurate and relevant criteria is required to develop spatial layers that planners, developers, local government, and other stakeholders can use to guide future development. At present, the KwaZulu-Natal Department of Agriculture and Rural Development (KZNDARD) can provide high-level crop production approximations for various crops based on BioResource Units (BRU). However, the BRW has not seen a significant revision for over two decades. Still, the natural resource information it contains provides land managers, policymakers and farmers with invaluable access to regional and farm level qualitative estimations of agricultural productivity. There is a need to preserve this information while simultaneously providing modern measures of land management recommendation at multiple scales to the end-user. Against this backdrop, access to readily interpretable soil and crop information is increasingly being prioritised by provincial planning commissions as critical inputs to DSS for sustainable land management within KwaZulu-Natal.
  • Item
    Evaluation of old store-and-release covers on discard dumps and backfilled pits to improve and predict their performance for rehabilitated mines at Mpumalanga Highveld, South Africa
    (Stellenbosch : Stellenbosch University, 2021-12) van Schalkwyk, Roeline; Hoffman, Josias Eduard; Van Zyl, Johan H. C.; Stellenbosch University. Faculty of Agrisciences. Dept. of Soil Science.
    ENGLISH ABSTRACT: Store-and-release covers (SRCs) are an important mitigation method to protect the environment at rehabilitated mines in the Mpumalanga Highveld, South Africa. The long-term performance of SRCs can be influenced by soil cover-, soil hydraulic-, and vegetation properties. Currently, a Technical Guideline for Soil Covers Development is not in place in South Africa. In addition, data sets of well- and poorly constructed covers, and the availability of data on appropriate input parameters for predicting long-term performance of such covers are limited. This need includes data for saturated hydraulic conductivity (Ksat), soil water retention curves (SWRCs), and photosynthetic active leaf- area index (LAI). Moreover, the measurement of Ksat and SWRCs is time-consuming, labour intensive and costly. Consequently, a multidisciplinary study to investigate the impact of soil cover-, soil hydraulic- and vegetation properties on long-term performance of SRCs was initiated. Most importantly, pedotransfer functions (PTFs) to predict Ksat and SWRCs were developed from particle- size distribution, soil organic matter (SOM) and bulk density. Leaf area index values for good and poor vegetation covers were determined for rehabilitated mines in Mpumalanga Highveld. Soil cover properties viz. cover configurations, soil texture, Atterberg limits, bulk density and soil nutrient availability were determined. Saturated hydraulic conductivity were measured using two types of double-ring infiltrometer, a single-ring infiltrometer, and a constant-head permeameter. Soil water retention curves were established using the pressure plate apparatus. The SRCs data-set was split into training and testing sets to validate the SWRC model. After the SRCs data-set was split into moderately- and very dense SRCs data sets, and an additional site was used to validate the moderately dense Ksat model. The data-set of very dense SRCs was also split into training and testing sets to validate the very dense Ksat model. Monthly LAI from September 2018 to August 2019 was destructively measured using a LI-3100C Area Meter. The dual-layered SRCs were constructed with sandier growth medium (top layer) underlaid by a loamy to clayey water retention layer (sub-layer). Monolithic SRCs were constructed of sandy loam or sandy clay loam soil covers. After the SRCs were split into moderately- and very dense soil cover conditions, the moderately dense SRCs performed significantly better and had acceptable bulk densities, good vegetation covers with good root distribution in the growth medium, steep slope in the desaturation function of the growth medium and high water-holding capacity (WHC) in the water retention layers. The Ksat and WHC of the moderate SRCs over 20 years had values similar to that of the soils, but the values of sandier soil cover layers were lower than critical threshold values due to low resistance to compaction. The statistical analysis of best-fit moderately- and very dense Ksat. models yielded an adjusted R2 of 0.749 and 0.999, respectively from sand-, silt- and clay content, SOM and bulk density. The statistical analysis of the best-fit SWRC model of 14 matric potentials had an adjusted R2 = 0.827 from three fractions of sand-, two fractions of silt-, clay content, SOM, and bulk density. The photosynthetic active LAI for good and poor vegetation cover of rehabilitated mines at Mpumalanga Highveld was ~1.2 and 0.8 m2.m-2, respectively. Poorly constructed soil covers result in high bulk density, low to very low Ksat and WHC values and poor vegetation properties and should be avoided at any cost. The critical threshold values for bulk density, Ksat and WHC of soils can be used to evaluate long-term soil cover performance. The developed PTFs can be used to predict soil covers’ hydraulic properties having soil physical properties similar to the old SRCs. These results can be considered as a possible amendment to the Technical Guidelines on Soil Cover Development in South Africa.
  • Item
    Digital soil mapping techniques across multiple landscape scales in South Africa
    (Stellenbosch : Stellenbosch University, 2019-12) Trevan, Flynn; Clarke, Catherine E.; Rozanov, Andrei Borisovich; De Clercq, W. P.; Stellenbosch University. Faculty of Agrisciences. Dept. of Soil Science.
    ENGLISH ABSTRACT: Digital soil mapping has seen increasing interest due to environmental concerns and increasing food security issues. Digital soil mapping offers a quantitative approach which is cost effective as less soil observations are needed to produce large area soil maps. However, digital soil mapping has only recently been addressed in South Africa. This research aimed to produce two digital soil mapping (DSM) frameworks with the available resources in South Africa. The methodologies incorporate advanced geostatistics and/or machine learning techniques to be able to produce quantitative soil maps from the farm to catchment scale. First, a framework that optimises both feature selection and predictive models was developed to produce farm-scale soil property maps. Four feature selection techniques and eight predictive models were evaluated on their ability to predict particle size distribution and SOC. A boosted linear feature selection produced the highest accuracy for all but one soil property. The top performing predictive models were robust linear models for gravel (ridge regression, RMSE 9.01%, R2 0.75), sand (support vector machine, RMSE 4.69%, R2 0.67), clay (quantile regression, RMSE 2.38%, R2 0.52), and SOC (ridge regression, RMSE 0.19%, R2 0.41). Random forest was the best predictive model for silt content with a recursive feature selection (RMSE 4.12%, R2 0.53). This approach appears to be robust for farm-scale soil mapping where the number of observations is often small but high-resolution soil data is required. Second, 24 geomorphons (landform classification) were evaluated on their association with soil classes. The geomorphon with the highest association was aggregated into a 5-unit system which was evaluated on how well the system stratified soil lightness, soil EC, SOC, effective rooting depth, depth to lithology, gravel, sand, silt, and clay. It was found that an aggregated geomorphon stratified all soil attributes except EC. Additionally, the aggregated geomorphon predicted 6 out of 9 soil properties with the greatest accuracy (RMSE) when compared to the original geomorphon (10-unit system) and a manually delineated system (5-unit system). This study shows that aggregating geomorphons can stratify the soil landscape even at the farm-scale and can be used as an initial indication of the soil spatial variability. Third, a framework to disaggregate the Land Type Survey (LTS) through machine learning was developed. Geomorphons, together with the original LTS were overlaid to produce terrain morphological units. The polygons were disaggregated further to produce a raster map of soil depth classes through a disaggregation algorithm known as DSMART. The first most probable class raster achieved an accuracy of 68% and for the two most probable class rasters, an accuracy of 91% was achieved. The two-step approach proved necessary for producing a farm-scale soil map. Forth, a study aimed to compare 10 algorithms, implemented through a modified DSMART model, in their ability to disaggregate two polygons into soil associations in two environmentally contrasting locations (Cathedral Peak, KwaZulu-Natal Province and Ntabelanga, Eastern Cape Province). At Cathedral Peak (high relief with clear toposequences), nearest shrunken centroid was the top performing algorithm with a kappa of 0.42 and an average uncertainty of 0.22. At Ntabelanga (low relief with strong geological control), the results were unsatisfactory. However, a regularised multinomial regression was the top performing algorithm, achieving a kappa of 0.17 and an average uncertainty of 0.84. The results of this study highlight the versatility of a technique to disaggregate South Africa’s national resource inventory. Disaggregation was then used to simultaneously disaggregate 20 land types in the Mvoti catchment covering 317 km2 in KwaZulu Natal province. First, the optimal geomorphon was chosen through a spatially resampled Cramer’s V test to determine the association between the soil legacy polygons and the geomorphon units. Second, feature selection algorithms were embedded into DSMART. Third, the feature selection techniques were compared using 25, 50, 100, and 200 resamples per polygon. The results indicate that the Cramer’s V test is a rapid method to determine the optimal input map. Feature selection algorithms achieved the same accuracy as using all covariates but had greater computational efficiency. It is recommended that 10 to 20 times the amount of soil classes be used for the number of resamples per polygon.