A novel canopy water indicator for UAV imaging to monitor winter wheat water status

Abstract

The utilization of UAV-based imaging systems for precise assessment of crop hydration levels plays a pivotal role in optimizing irrigation strategies and enhancing the efficiency of agricultural water resource management. While canopy fuel moisture content (FMCc) serves as a key parameter for evaluating plant hydration status, its accurate quantification relies heavily on precise measurements of the leaf area index (LAI). However, the complexity involved in acquiring LAI data and the associated high costs limit the practical application of FMCc in crop water monitoring. To address this limitation, this study proposed a novel canopy water indicator, termed r-FMCc, which integrates canopy coverage and FMC. The effectiveness of FMC, FMCc and r-FMCc in assessing wheat water status were comparatively analyzed using UAV hyperspectral data. First, the hyperspectral data were processed to generate a range of vegetation indices. Subsequently, a Boruta-based feature selection algorithm was employed to identify those indices that exhibited significant correlations with the three target water parameters (FMC, FMCc,and r-FMCc). To develop robust estimation models, four machine learning algorithms were implemented across individual and combined growth stages, and their performance was validated using independent ground-measured datasets that were not used during the training process. The results indicated significant positive correlations between LAI and canopy coverage across all growth stages. Among the four estimation models, the random forest (RF) and Gaussian process regression models exhibited superior performance in estimating various water indicators. Considering variability across growth stages significantly improved the accuracy of water status quantification compared to assessments based on individual growth stages. Using RF, The R²values for the training sets of FMC, FMCc, and r-FMCc across multiple growth stages were 0.96, 0.98, and 0.98, respectively, while the corresponding R²values for the testing sets were 0.83, 0.90, and 0.89. The integration of UAV-based hyperspectral imagery with machine learning techniques enables high-throughput and precise quantification of wheat canopy water status parameters. The newly proposed wheat water indicator (r-FMCc) enhances the applicability of UAV imaging for monitoring wheat water status without compromising estimation accuracy.