Diagnosing water conditions timely and accurately is crucial for seasonal irrigation scheduling in crop production. The purpose of this study was to establish robust water deficit models of winter wheat in different growing seasons by combining unmanned aerial vehicle (UAV) multispectral and thermal images. In 2021 and 2022, a water deficit field experiment on winter wheat was carried out in Hebei Province, China. Five-band multispectral and thermal images were obtained with a UAV at six key growth stages of winter wheat. Fourteen vegetation indices (VIs) and two thermal indices (TIs) were calculated. Simultaneously, wheat stomatal conductance and soil water content were measured. On this basis, normalized stomatal conductance (NGS) and effective water content (EWC) were calculated. TIs had the highest correlation with NGS and EWC at early growth stages, whereas ratio vegetation index (RVI), modified simple ratio index (MSR) and normalized difference vegetation index (NDVI) were highly correlated at later stages. Partial least squares (PLS), support vector machine (SVM) and gradient boosting decision tree (GBDT) were used to predict NGS and EWC for each growth stage, with the data of 2021 as the training set and the data of 2022 as independent test set. In general, GBDT outperformed PLS and SVM, and NGS was better predicted than EWC. Including VIs and TIs effectively improved the estimation accuracy of the predictive models. The test set results of the NGS and EWC models built by GBDT achieved the best performance in flowering stage (coefficient of determination (R2) = 0.88, root mean square error (RMSE) = 0.08, normalized root mean square error (NRMSE) = 14.7%) and filling stage (R2 = 0.90, RMSE = 0.05, NRMSE = 15.9%), respectively. The models of the post-heading stage were better than those of the pre-heading stage for both NGS and EWC. This study provides a robust method for diagnosing water stress only using UAV remote sensing data.
Full Text Finder