The use of technology to improve current precision viticulture practices: predicting vineyard performance

Date
2018-12
Journal Title
Journal ISSN
Volume Title
Publisher
Stellenbosch : Stellenbosch University
Abstract
ENGLISH ABSTRACT: Producing high quality grapes is difficult due to intra-vineyard spatial variability in vineyards. Variability leads to differences in grape quality and quantity. This poses a problem for producers, as homogeneous growth is nearly non-existent in vineyards. Remote sensing provides information of vineyard variability resulting in better knowledge of the distribution and occurrence thereof, leading to improved management practices. Remote sensing has been studied and implemented in several fields of research and industry, such as monitoring forest growth, pollution, population growth, etc. The potential to implement remote sensing technology is endless. Generating variability maps introduces the possibility of plant specific management practices, to alleviate problems occurring from variability. Aerial and satellite remote sensing provide new methods of variability monitoring, through spatial variability mapping of soil and plant biomass. Advances in geo-referencing and geolocations provide high accuracy precision tools for producers and researchers. New technology introduces possible means of vigour classification and stress monitoring on a plant scale, relieving the uncertainty caused by the distribution and extent of variability in vineyards. Vineyards are more difficult to analyse with remote sensing technology, due to the discontinuous canopies resulting in objects, other than plant biomass, to be monitored with the plant biomass. These objects can be soil and inter-row plant growth, along with trees close to or adjacent to the vineyards. This provides a dilemma through diluting biomass estimations and resulting in misinterpretation of the vineyard variability. These problems could be solved with the use of high-resolution multispectral imaging, providing clear classification and information of plant growth and health status. These sensing technologies have only been studied in some industries and have yet to be implemented to provide plant specific information. Introducing high accuracy plant specific information along with geolocation information will provide the producer with enough information to implement specific management practices alleviating heterogeneous plant growth and promoting homogeneous growth and yield, resulting in improved economic status through limiting input costs and environmental impact, providing better living conditions for plants along with increased plant longevity. The aim of the study was to evaluate the accuracy of leaf area index (LAI) estimations from selected remote sensing technologies with three different sensor resolutions. Imaging of the experimental site with natural variability were taken with the remote sensors. Targeted vines in the vineyard were selected as ground control points for ground truth measurements. The data acquired from the ground truth measurements were compared to the normalised difference vegetation index (NDVI) values generated from the remote sensing technologies. Grid analysis was performed on the unmanned aerial vehicle (UAV) multispectral images, mapping the LAI of individual plants. Significant differences in LAI predictions were obtained with good correlations between the ground truth data and the UAV multispectral image NDVI, r2 = 0.69. Climatic conditions proved problematic for the satellite images, where resolution also posed a problem. Variability is often caused by environmental factors, although management practices influence variability of vineyards. Management practices can be beneficial to plant growth, such as tillage promotes soil aeration and biodiversity through mixing the soil layers and providing more homogeneous soil conditions in the vineyard, or detrimental, for example saline irrigation water can lead to toxic saline concentrations in the soil and result in plant degradation over time as the symptoms are only visual when toxicity has occurred. Salinity also provides improved soil moisture conditions through reducing the rate of soil drying. Other factors result in zonal variability, such as patchy growth from nutrient deficiencies or irregular growth patterns from pests or diseases. Remote sensing technology provides several sensing methods to determine the extent and distribution of variability. These methods involve various sensors, such as multispectral, light distance and ranging (LiDAR), etc., providing enough information to make informed decisions on management practices to limit variability or improve the extent thereof. These sensors are attached to aerial, satellite or ground platforms depending on the resolution needed and the extent of the study site. Field measurements of the selected ground truth sampling points showed the presence of natural variability and the distribution thereof in the vineyard. Analysis of the UAV multispectral images revealed a good correlation between the ground truth data and the NDVI values. Soil and other objects were removed from the multispectral images, resulting in increased accuracy of biomass estimations and limited the NDVI blending effect observed in low-resolution images. Pixel based NDVI values of each plant, generated from the UAV multispectral device, were averaged to provide the NDVI per plant. Satellite images generated resolution-based area averages and blended pixel values of the soil and other objects adjacent to the vines limiting the plant-based information. Satellite images were affected by climatic conditions, especially cloud cover, along with limited image acquisitions revealed restricted image usability. UAV multispectral images provided plant-based LAI maps based on information generated through grid analysis, revealing the distribution of variability with accurate vine locations. This study provided methods of autonomous image analysis for high- and low-resolution remote sensing technology. Models with accurate plant-based estimations to monitor and evaluate management practices will improve grape production and optimise quality resulting in improved wine quality. Selective harvesting and management practices will lead to optimised yield quality for targeted wine production, feeding the consumer driven industry. This study paved the way for future research in variability estimations from remote sensing technology with emphasis on the causes of within-vineyard variability.
AFRIKAANSE OPSOMMING: Die vervaardiging van hoë gehalte druiwe is ingewikkeld as gevolg van variasie in wingerde. Variasie lei tot verskille in druiwe kwaliteit en kwantiteit. Dit hou 'n probleem in vir produsente, deur homogene groei wat byna nie in wingerde bestaan nie. Afstandswaarnemingstegnologie verskaf inligting oor wingerd variasie wat lei tot 'n beter kennis van die verspreiding en voorkoms daarvan en dus verbeterde bestuurspraktyke. Afstandswaarneming is ondersoek en geïmplementeer in verskeie navorsings- en nywerheidsvelde, soos die monitering van oerwoudgroei, besoedeling, bevolkingsgroei, ens. Die implementeringspotensiaal van afstandswaarnemingstegnologie is eindeloos. Deur variasiekaarte saam te stel, kan plantspesifieke bestuurspraktykte geïmplementeer word, wat probleme wat die voorkoms van variasie veroorsaak, kan verlig. Lugfoto’s en satelliet afstandswaarnemingstegnologie bied nuwe metodes van variasie monitering deur die kartering van grond- en plantbiomassa ruimtelike variasie. Vooruitgang in geo-verwysings en geo-ligging bied 'n hoë akkuraatheid presisie instrumente vir produsente en navorsers. Nuwe tegnologie stel moontlike maniere van groeikrag klassifikasie en stresmonitering op 'n plantvlak voor, dus verligting van onsekerheid wat veroorsaak word deur die verspreiding en omvang van variasie in wingerde. Wingerde is moeiliker om met afstandswaarneming tegnologie te ontleed, te danke aan die diskontinue lower wat lei tot monitering van voorwerpe, anders as plantbiomassa. Hierdie voorwerpe kan grond en tussen-ry plant groei wees, saam met bome wat naby of aangrensend aan die wingerd is. Dit bied 'n dilemma deur verdunde biomassa skattings en lei tot waninterpretasie van wingerdvariasie. Hierdie probleme kan opgelos word met die gebruik van 'n hoë-resolusie multispektrale kamera, deur duidelike klassifikasie en inligting van plantegroei en gesondheidstatus te verskaf. Hierdie afstandswaarnemingstegnologie is slegs in sommige nywerhede bestudeer en is nog nie gebruik om plantspesifieke inligting te verskaf nie. Die bekendstelling van hoë akkuraatheid plantspesifieke inligting saam met ligginggewing inligting sal aan die produsent genoegsame inligting verskaf om spesifieke bestuurspraktyke daar te stel ter verligting van heterogene plantegroei en die bevordering van homogene groei en opbrengs, wat lei tot verbeterde ekonomiese status deur die beperking van insetkoste en omgewingsimpak, dus beter lewensomstandighede vir plante te verskaf tesame met 'n verlengde plant lewensverwagting. Die doel van die studie was om blaaroppervlakte indeks (LAI) skattings van geselekteerde afstandswaarnemingstegnologieë met verskillende resolusies te evalueer. Lugfoto’s is van die eksperimentele terrein, wat natuurlike variasie toon, geneem met die afstandswaarnemingstegnologie. Geteikende stokke in die wingerd is gekies as grondkontrolepunte vir grond waarheidsmetings. Data verkry uit die grond waarheidsmetings is vergelyk met die genormaliseerde verskil plantegroei indeks (NDVI) metings verkry van die afstandswaarnemingstegnologieë. Matriks analise is uitgevoer op die onbemande vliegtuig (UAV) multispektrale beelde wat tot kartering van die individuele plante se LAI gelei het. Beduidende verskille in LAI voorspellings is verkry deur 'n goeie korrelasie tussen die grond waarheid data en die NDVI van die UAV multispektrale beelde. Klimaatstoestande het probleme vir die satellietbeelde aangedui, waar resolusie ook 'n probleem was. Veranderlikheid is dikwels veroorsaak deur omgewingsfaktore, hoewel bestuurspraktyke variasie van wingerde beïnvloed. Bestuurspraktyke kan voordelig wees vir die groei van plante, bv. grondbewerking wat deurlugting en biodiversiteit bevorder deur die vermenging van grondlae en meer homogene grondtoestande te verskaf in die wingerd, of nadelig, bv. sout besproeiingswater kan lei tot giftige sout konsentrasies in die grond en met verloop van tyd lei na plant agteruitgang waar visuele simptome slegs toon nadat toksisiteit plaasgevind het. Soutgehalte bied ook verbeterde grondvog toestande aan deur die tempo van grond droging te verlaag. Ander faktore lei tot sonale variasie, soos onewe groei deur voedingstekorte of onreëlmatige groeipatrone van peste of siektes. Afstandswaarnemingstegnologie bied verskeie waarnemingsmetodes aan om die omvang en verspreiding van variasie te bepaal. Hierdie metodes behels verskeie sensors, soos multispektrale kameras, “light distance and ranging” (LiDAR), ens, wat genoeg inligting verskaf om ingeligte besluite oor bestuurspraktyke te neem om variasie te beperk of verbeter. Hierdie sensors is aan lug-, satelliet- of grondplatforms geheg, afhangende van die resolusie wat nodig is en die omvang van die studie area. Veldmetings van die gekose grond waarheid monsternemingspunte het die teenwoordigheid van natuurlike variasie en die verspreiding daarvan in die wingerd. Ontleding van die UAV multispektrale beelde het 'n goeie korrelasie tussen die grond waarheid data en die NDVI waardes geopenbaar. Deur grond en ander voorwerpe uit die multispektrale beelde te verwyder, word verhoogde akkuraatheid van biomassa skattings verkry en die NDVI vermenging effek waargeneem in 'n lae-resolusie beelde beperk. Gemiddelde pixel gebaseerde NDVI waardes van elke plant, wat uit die UAV multispektrale toestel verkry is, het die totale NDVI van elke plant opgemaak. Satellietbeelde het resolusie-gebaseerde area gemiddeldes met gemengde pixelwaardes van die grond en ander voorwerpe langs die wingerd gegenereer en tot beperking van plant-gebaseerde inligting gelei. Satellietbeelde is deur klimaatstoestande beïnvloed, veral wolkbedekking, wat saam met beperkte beeld aanwinste die bruikbaarheid van die beelde verlaag. UAV multispektrale beelde, wat gebaseer is op inligting gegenereer deur matriks analise, verskaf plant-gebaseerde LAI kaarte, wat lei tot die onthulling van die verspreiding van variasie met akkurate wingerdstok posisies. Hierdie studie verskaf metodes van outomatiese beeld analise vir hoë- en lae-resolusie afstandswaarnemingstegnologieë. Modelle met akkurate plant-gebaseerde skattings wat monitering en evaluasie van bestuurspraktyke tot gevolg het, sal druiwe produksie verbeter en druiwe kwaliteit optimaliseer wat sal bydra tot verbeterde wyngehalte. Selektiewe oes- en bestuurspraktyke sal lei tot optimale opbrengsgehalte vir geteikende wynproduksie uitkomste, wat die verbruiker-gedrewe bedryf sal voed. Hierdie studie het die weg gebaan vir toekomstige navorsing in variasie skattings met afstandswaarneming tegnologie met klem op die oorsake van in-wingerd variasie.
Description
Thesis (MA)--Stellenbosch University, 2018.
Keywords
Leaf Area Index, Unmanned Aerial Vehicle, Landsat 8, Sentinel-2, Multispectral imaging, Remote sensing, Landsat satellites, Drone aircraft, UAVs (Unmanned aerial vehicles), Precision viticulture, UCTD
Citation