Assimilating UAV observations and crop model simulations for dynamic estimation of crop water stress

Crop water stress (CWS) monitoring using UAV remote sensing has traditionally been limited to empirical models and specific growth stages, restricting dynamic, season-long assessment. This study proposes an integrated framework combining multispectral UAV observations with the SAFYE crop model via Ensemble Kalman Filter -based data assimilation (DA) to improve maize growth simulation and enable continuous CWS monitoring. Based on three years of field experiments, accurate inversion models for leaf area index (LAI; R2= 0.837, RMSE = 0.397) and aboveground biomass (AGB; R2 = 0.862, RMSE = 224 g m−2) were developed using a random forest algorithm. Model parameters were calibrated using particle swarm optimization, and UAV-derived data were assimilated to optimize simulations of crop growth and actual evapotranspiration (ETc act). Results show that DA significantly enhanced model performance: LAI simulation RMSE decreased from 0.29–0.61–0.11–0.36 (NRMSE: 3.57–11.56 %), AGB simulation RMSE from 148.2–255.7–49.3–136.8 g m−2 (NRMSE: 5.39–14.27 %), and agreement index (d) exceeded 0.92. ETc act simulations accurately reflected responses to irrigation and rainfall, with only 4.97 % relative error under full irrigation (W4). The developed crop water stress index (CWSI) effectively quantified water stress under different irrigation treatments. A significant negative correlation was observed between CWSI reduction and irrigation amount, while the severity of water deficit was positively correlated with the peak value of CWSI differences in terms of both timing and magnitude. This study establishes a robust UAV–crop model DA framework for dynamic, season-long CWS diagnosis and assessment.