Research on CASA-WOFOST tomato yield dynamic prediction model integrating light energy utilization algorithm and crop physiological mechanism

This study addresses the core contradiction in regional-scale tomato yield estimation,which includes the pure mechanistic models lack accuracyand the pure empirical models lack biological constraints. It proposes the CASA-WOFOST coupled yield estimation model that integrates the remote sensing light energy utilization mechanisms with the crop physiological process simulation. The study constructs a data-driven system based on the drones multispectral remote sensing, meteorological observations and ground yield quadrant data. First, the CASA model is used to obtain the net primary productivity (GPP) through the APAR and light energy utilization rate calculations, which establishes the remote sensing-driven framework for the canopy light energy absorption and carbon assimilation. Then, the WOFOST crop model is introduced to simulate the tomato phenological development, photosynthesis, respiration consumption, and dry matter distribution to organs, mapping GPP to fruit dry matter formation, which achieves the closed-loop physiological mechanism from the energy absorption to the carbon accumulation of yield formation. At the parameter level, the DREAM algorithm is used for the global calibration; the optimization strategy is constructed by using the LAI and yield as joint likelihood functions. At the state level, the EnKF is used to assimilate the remote sensing LAI, which updates the model's growth trajectory to reduce the structural errors. This model constructs a multi-layered coupled system of "remote sensing light energy constraint + physiological process constraint + dynamic assimilation correction," which makes the productivity calculation both spatially sensitive and consistent with biological principles. Experimental results show that the CASA-WOFOST achieves higher coefficients of determination and lower errors under various vegetation indices and texture features, which demonstrates the significant improvement in its stability and generalization ability in regional-scale crop yield prediction.