Leveraging thermal remote sensing and unmanned aerial vehicle high-throughput phenotyping for assessing and monitoring the water status of neglected and underutilised taro crops in smallholder farming systems.

Abstract

As threats posed by climate change and variability continue to intensify, smallholder farming systems are challenged by the urgent need to sustain crop production and ensure food security. Taro (Colocasia esculenta (L)), a Neglected and Underutilised Crop Species (NUS), has emerged as a promising future-smart crop due to its resilience to drought and heat stresses, holding great potential for diversifying existing cropping systems and enhancing smallholder farming resilience. Despite its reported adaptive capabilities, taro remains vulnerable to prolonged water stress. Such conditions can disrupt internal water balance, leading to reduced equivalent water thickness, increased foliar temperature and decreased stomatal conductance, which can ultimately compromise taro’s tuber quality and productivity. Therefore, accurate and robust monitoring of taro crop water status indicators is essential for the rapid detection of water deficits, facilitating proactive and targeted interventions aimed at mitigating stress impacts and maintaining optimal productivity. Cutting-edge remote sensing technologies, particularly Unmanned Aerial Vehicles (UAVs) equipped with high-resolution thermal cameras integrated with multispectral sensors, have revolutionised precision agriculture. Such technologies have emerged as invaluable tools that enable near-real-time crop monitoring at ultra-high spatial and temporal resolutions, suitable for continuous field-scale assessments of water status. Hence, this study sought to evaluate the utility of UAV thermal remote sensing in assessing and monitoring the crop water status of neglected and underutilised taro crop within smallholder farming systems. Taro is classified as a NUS owing to its limited inclusion in mainstream agricultural research and policy, particularly in Africa and many other regions worldwide, despite its potential to support food security in climate-vulnerable regions. Specifically, this study first adopted a systematic approach to review the progress, challenges, and opportunities in utilising UAV thermal remote sensing to assess and monitor the water status within crop farming systems. The findings revealed that studies utilising UAV thermal remote sensing to assess crop water status are disproportionately concentrated in the global north, with a limited focus on neglected and underutilised crops (> 4 %) and smallholder rainfed systems in the global south (2.3 %). Furthermore, results highlighted that while UAV-derived thermal datasets have gained significant traction, integrating thermal imagery with multispectral data is crucial for leveraging their complementary strengths, enhancing accuracy, and providing a more comprehensive assessment of crop water status. The findings further highlighted the importance of advanced image segmentation techniques in mitigating soil background interference, which can distort crop thermal signatures and compromise the precision of crop water status assessments. As a result, the second objective of the study was to assess the utility of indexbased image segmentation techniques and UAV thermal remotely sensed data in enhancing the estimation of smallholder taro equivalent water thickness (EWTcanopy) as a proxy of crop water status. To achieve this objective, a comparative analysis was conducted to assess the predictive performance of models with and without the thermal band, while also evaluating the effectiveness of Excess Green (ExG), Excess Red (ExR), and Excess Green minus Excess Red (ExGR) image segmentation techniques in improving taro EWTcanopy estimations. The findings revealed that incorporating the thermal band and applying image segmentation, particularly using the ExGR technique, significantly enhanced the prediction accuracy of taro EWTcanopy, leading to a substantial increase in the R² value from 0.32 to 0.92, while the rRMSE was significantly reduced from 60.51% to 15.31%. Having established the importance of integrating thermal data with the ExGR image segmentation technique, the third objective aimed to evaluate the utility of UAV remotely sensed data for high-throughput crop phenotyping of taro equivalent water thickness, fuel moisture content, stomatal conductance, foliar temperature and chlorophyll content as proxies for water status within smallholder farms. The findings revealed that a multi-modal approach, integrating thermal and multispectral data outperforms singlemodal methods, yielding R2 values greater than 0.91 and rRMSEs less than 14.15%. Notably, the thermal waveband and derived thermal indices emerged as the most influential variables for estimating stomatal conductance and leaf temperature, with R² values of 0.96 and 0.95, respectively. In contrast, for equivalent water thickness and fuel moisture content, other spectral variables ranked higher in importance. However, incorporating thermal spectral variables substantially improved the prediction accuracy for these traits, increasing R² from 0.73 to 0.95 (rRMSE reduced from 33.82 % to 14.15 %) for equivalent water thickness, and from 0.77 to 0.94 (rRMSE reduced from 6.55 % to 3.32 %) for fuel moisture content. Subsequently, the fourth objective sought to conduct a multi-temporal analysis of NUS taro crop water status using multi-modal UAV remotely sensed data and deep learning techniques to estimate stomatal conductance and foliar temperature as key physiological indicators across different growth stages of smallholder taro crops. The findings highlighted distinct trends in stomatal conductance and leaf temperature, with the emergence stage exhibiting the highest leaf temperatures and lowest stomatal conductance, while the vegetative stage showed the lowest leaf temperatures and a peak in stomatal conductance. Notably, the vegetative growth stage exhibited the highest prediction accuracies for stomatal conductance (R2 of 0.96, RMSE of 29.34 mmol m−2 s −1 and rRMSE of 12.86 %) and leaf temperature (R2 of 0.95, RMSE of 0.33 °C and rRMSE of 1.11 %). This pattern may be attributed to the limited canopy cover during the emergence stage, where exposed soil temperatures to interfere with crop thermal signatures, in contrast to the vegetative stage where increased foliage reduces soil influence and supports optimal physiological activity. Finally, the fifth objective evaluated the utility of a data-driven approach using UAV thermal and multispectral remotely sensed data, along with topographic variables, to estimate the stomatal conductance and leaf temperature of smallholder taro crops across different growth stages (emergence, vegetative, and maturity) as proxies for crop water status. While integrating multi-source datasets provides a comprehensive evaluation of crop water conditions, it is recommended that advanced feature selection and model optimisation are employed to address challenges of redundancy, multicollinearity and overfitting because of combing large feature subsets. To this end, the findings highlighted the utility of integrating diverse yet relevant datasets, including thermal, multispectral, and topographic data, into a unified data-driven framework for estimating crop water status. Additionally, this study applied critical water stress thresholds (50 mmol m−2 s −1 for stomatal conductance and 35 °C for leaf temperature) to the optimised models, enabling the spatially explicit identification of waterstressed areas within the taro field. Results revealed significant stress during the emergence stage, with 14.18 % of crops showing low stomatal conductance and 37.14 % exceeding the leaf temperature threshold. In contrast, minimal stress was observed in the vegetative growth stage (1.85 %), while the maturity stage showed a slight increase in stress, with 9.36% of the area exceeding the leaf temperature threshold. The findings of this study highlight the critical importance of early-stage monitoring and targeted interventions, especially during the emergence stage, to manage potential negative impacts caused by water stress on taro. Overall, the findings of this study demonstrated the transformative potential of integrating UAV thermal remote sensing with advanced deep learning techniques in providing rapid and robust spatially explicit information on smallholder taro crop water status for ensuring crop productivity and developing early warning systems of water stress. The findings make a significant contribution to the anecdotal knowledge of neglected and underutilised crops, such as taro. Additionally, they play a crucial role in promoting climate-smart agriculture and enhancing climate resilience within smallholder farming systems. Lastly, the implications of this study are aligned with global and regional developmental goals, including Sustainable Development Goals (2 and 13) and the African Union's Agenda 2063 Goals (5), contributing to sustainable agricultural practices that enhance food security and climate resilience. Ultimately, this study is a pathway towards transformative, data-driven frameworks and actionable solutions that could empower decision-makers to support smallholder farmers in proactively adapting to climate variability, enhancing long-term crop viability and fostering resilience and sustainability in the face of intensifying climate stress.