Optimization study on diagnostic methods for winter wheat water stress using UAV-borne thermal infrared imagery

To address the low estimation accuracy of the Crop Water Stress Index (CWSI) directly induced by imprecise extraction of plant canopy temperature (T c ) from thermal infrared (TIR) imagery, this study used UAV visible imagery of winter wheat under different water and nitrogen regimes to calculate the Green Leaf Index (GLI) for canopy mask construction, which was then overlaid with TIR imagery to extract T c , and subsequently multi-gradient extreme pixel elimination ratios were applied to identify the optimal method for T c extraction. Subsequently, the extracted T c are categorized into distinct pixel distribution intervals based on the standard normal distribution, and the interval-specific Crop Water Stress Index (CWSI F ) is calculated using the mean canopy temperature (T F ) of each interval. Thereafter, rigorous regression analysis was performed for the derived CWSI F variants against key crop physiological indicators to determine the most sensitive CWSI F values corresponding to each indicator for subsequent practical applications. The results indicate that proper removal of extreme pixels enhanced the consistency between UAV TIR-retrieved temperature and in-situ measured temperature. Excluding 3 % of extreme pixels from both ends of the T c distribution histogram yielded a relatively optimal level of this consistency, thus enabling more accurate characterization of the actual T c of crop. CWSI F values derived from the T F across different T c pixel distribution intervals differed significantly. Regression analysis showed that the sensitive CWSI F corresponding to stomatal conductance (G s ), transpiration rate (T r ), and net photosynthetic rate (P n ) differed significantly, requiring a comprehensive evaluation integrating multiple physiological indicators. For the scientific diagnosis of crop water status, the entropy weight method was employed to assign weights to the evaluation indicators of G s , T r , and P n . Based on these weights, a linear weighted summation model was used to obtain the comprehensive score. and the optimal CWSI F that reflects the characteristics of multiple physiological indexes was determined for each growth stage: the optimal index was CWSI -0.5 during the jointing stage and flowering stage, and CWSI -0.3 during the filling stage. This solves the problem of inconsistent evaluation of CWSI F by different physiological indicators and improves the pertinence and accuracy of water stress diagnosis. Across all growth stages, the coefficient of determination (R²) between the optimal CWSI F and plant water content (PWC) was consistently higher than that between the traditional CWSI T (CWSI calculated based on the average value of all T c ) and PWC, while the normalized root mean square error (nRMSE) of the former was consistently lower than that of the latter. This indicated that CWSI F can reflect the water status of crop more efficiently and accurately than CWSI T . The findings of this study provide a reliable technical basis for monitoring water stress and implementing staged precision irrigation in winter wheat. • An RGB–TIR UAV workflow improves T c accuracy using GLI masking and multi-gradient pixel removal. • Spatially stratified CWSI calculation enhances sensitivity to crop water stress across growth stages. • Entropy-weighted linear integration of physiological indicators determines optimal CWSI across growth stages. • The optimized CWSI method improves consistency with plant water content and precision irrigation applicability.