A framework of crop water productivity estimation from UAV observations: A case study of summer maize

This investigation establishes Crop Water Productivity (CWP) - quantified as yield per unit water consumption (kg/m³) - as a pivotal metric for agricultural water resource optimization. However, current methodologies face limitations in estimation accuracy and operational efficiency due to the multidisciplinary complexity integrating agronomic and hydrological expertise. To address this challenge, our research develops an innovative UAV-based monitoring framework through systematic integration of long-term multispectral/thermal infrared observations with multi-model fusion: (1) Surface Energy Balance Algorithm for Land (SEBAL) and FAO-56 Penman-Monteith models for evapotranspiration (ET) estimation; (2) Random Forest algorithm incorporating four phenotypical growth indicators for yield estimation, ultimately enabling CWP quantification. Key scientific findings demonstrate: (1) SEBAL outperformed FAO-56 in daily ET estimation (R² = 0.76 vs. 0.71, RMSE = 1.15 vs. 1.31 mm/d). (2) The machine learning yield model exhibited robust predictive capability (R² = 0.77, RMSE = 0.98 t/ha), successfully capturing yield variability across treatments. (3) Error propagation analysis validated framework reliability (CWP RMSE = 0.67 kg/m³), effectively differentiating CWP performance among management practices. This breakthrough validates the operational efficacy of UAV remote sensing for precision agricultural water assessment, providing decision-support for field-scale irrigation scheduling optimization, drought-resilient cultivar selection through CWP benchmarking and sustainable intensification strategies. The methodology establishes novel methodological benchmarks for crop-water relationship studies through its innovative fusion of multi-source remote sensing data and multiple model combination.