Advancing precision water management in smallholder sugarcane farming: leveraging unmanned aerial vehicle-based remote sensing and machine learning for evapotranspiration and water stress assessment.

Abstract

This thesis addresses the imperative of optimising water resource management within smallholder sugarcane cultivation in Swayimane, KwaZulu-Natal, South Africa—a region contending with the dual pressures of escalating food demand and increasingly volatile climatic dynamics. Smallholder farmers are indispensable to advancing food security and socio-economic development, underpinning up to 80% of the region's agricultural output. Nevertheless, their capacity to mitigate food insecurity is hindered by restricted access to vital resources, including reliable water supplies and advanced tools. These limitations necessitate innovative and economically viable strategies to enhance productivity and optimise resource allocation. Precision agriculture (PA) methodologies, supported by cutting-edge technologies such as unmanned aerial vehicles (UAVs), hold promise for smallholder farmers. By enabling data-driven, resource-efficient cultivation practices, UAVs emerge as an instrument to foster sustainable agricultural systems tailored to the unique challenges of these communities. Sugarcane, a high-value commodity crop, is integral to the socio-economic fabric of smallholder farming communities, substantially contributing to employment and subsequent regional economic advancement. However, the absence of irrigation infrastructure within smallholder systems and escalating water deficits driven by rising temperatures and prolonged dry spells present formidable challenges to sustainable production. This reliance emphasises the need for resource-efficient agronomic strategies that maximise water use while safeguarding yields. By harnessing the potential of PA methodologies, including UAVs equipped with advanced multispectral sensors, farmers can acquire high-resolution insights into crop water dynamics and evapotranspiration processes. To this end, this research pioneers integrating UAV technology with machine learning (ML) to refine water management practices, focusing on enhancing evapotranspiration (ET) estimation and monitoring crop water stress. Chapter 2 undertakes a bibliometric analysis of UAV applications in precision water management, employing Biblioshiny and VOSviewer to identify key research trends and highlight potential strengths, limitations, and future opportunities. The findings reveal UAVs' potential to address the limitations of traditional ground-based and remote sensing (RS) methods, which are often labour-intensive, expensive, and lack sufficient spatial and temporal resolution for effective water management in smallholder farming systems. UAV technology, driven by advancements in high-resolution data acquisition and the proliferation of cost-effective, open-source processing platforms, offers accessible, scalable solutions tailored to smallholders. While certain factors may moderate their adoption, continuous technological progress and decreasing costs present significant opportunities for UAV applications to enhance policy formulation, strategic planning, and operational decision-making, ultimately strengthening resilience in sustainable water management for smallholders. Chapter 3 presents an empirical evaluation of vegetation index (VI)-based ET estimation methods. Using data from a smallholder sugarcane field equipped with an eddy covariance (EC) system for ground-truthing, the study assessed five actual ET (ETa) models—ET-NDVI, ET-NDVIscaled, ET-NDVIKc, ET-EVI, and ET-EVI2—alongside an ML-derived crop coefficient (Kc) prediction model correlating in-situ NDVI with Kc values. The EVI2 model demonstrated superior performance, achieving moderate to strong correlation (R² = 0.63) with lower Root Mean Square Error (RMSE = 0.67) and Mean Absolute Error (MAE = 0.52) compared to competing models. EVI2's resilience against NDVI saturation—a persistent challenge in mature sugarcane—translates to improved water use assessment, while reduced reliance on extensive in-situ data enhances its scalability for smallholder systems. Furthermore, high-resolution ETa maps derived from these models offer potential insights for optimising irrigation and improving productivity in resource-limited agricultural contexts. Chapter 4 presents a novel ML-based predictive model for the Normalised Difference Water Index (NDWI), utilising correlations with structural VIs (SVIs) from UAV and Sentinel-2 data. The Random Forest (RF) ensemble model achieves exceptional accuracy (R² = 0.95, RMSE = 0.03, MAE = 0.02), offering a precise and efficient tool for monitoring sugarcane water stress. Validated against ETa and the Water Deficit Index (WDI), the model confirms NDWI's reliability as a proxy for assessing sugarcane water status. A Principal Component Analysis (PCA) reveals complex interactions between NDWI, SVIs, and physiological parameters, further enhancing insights into sugarcane water status. Additionally, temporal analysis highlights NDWI's responsiveness to rainfall, with marked fluctuations pinpointing critical stress periods. By minimising dependence on in-situ measurements, the model offers a scalable, cost-effective solution tailored to the needs of resource-constrained smallholder farmers. In conclusion, this thesis advances PA by integrating UAV technology and ML to revolutionise water management in smallholder sugarcane farming. The VI-based ET estimation models and NDWI prediction framework deliver customised, high-resolution insights to optimise irrigation and enhance water efficiency during critical growth phases. Beyond showcasing these innovations, the study emphasises the necessity of capacity-building initiatives and user-friendly tools to enable farmer adoption, addressing climatic and socio-economic constraints. This work establishes a foundation for scalable, context-specific solutions despite its limited scope. Future research should prioritise decision support systems integrating UAV data and predictive models for real-time water stress monitoring and adaptive irrigation. By empowering smallholders, this study contributes to resilient agricultural systems, enhanced food security, and sustainable sugarcane production, laying the groundwork for global efforts against water scarcity and climate variability.