Integrating UAV and satellite LAI date into a modified DSSAT-Rapeseed model to improve yield predictions

Accurate early-season estimation of winter rapeseed (Brassica napus L.) yield is essential for effective pre-winter management and informed agronomic decision-making. Leaf area index (LAI) is a key indicator of canopy development, photosynthetic capacity, and yield potential, and can be efficiently monitored using remote sensing technologies. However, uncertainties in crop model inputs and limited capacity to dynamically integrate field observations constrain the accuracy of yield simulations, particularly under heterogeneous farm conditions. This study aimed to improve rapeseed yield prediction by integrating unmanned aerial vehicle (UAV) and satellite-derived LAI into a modified version of the DSSAT-CROPGRO-Canola model using data assimilation.A modified DSSAT rapeseed model was developed by introducing a novel “target LAI” (TLAI) mechanism within the pest module, enabling remotely sensed LAI to directly regulate simulated leaf area, leaf biomass, and leaf nitrogen content. This approach ensured mechanistic consistency between assimilated observations and co-varying plant state variables. Multi-year UAV-derived LAI and field measurements from two experimental stations in the Jianghan Plain, China, were used to calibrate sensitive cultivar parameters through a trial-and-error optimization procedure. Pre-winter LAI was assimilated using the Ensemble Kalman Filter (EnKF), and model performance was evaluated using independent experimental datasets. To assess applicability under real farming conditions, LAI derived from Sentinel-2A imagery via a vegetation index lookup table (VI-LUT) was assimilated into the modified model for yield prediction at actual farm fields.Assimilation of LAI substantially improved yield simulation accuracy. At experimental plots, incorporating UAV-derived LAI reduced yield mean absolute error (MAE) from 452 to 234 kg ha⁻¹ and decreased mean absolute percentage error (MAPE) from 14.3% to 7.4% compared with the original DSSAT model. At actual farm fields, assimilation of Sentinel-2A LAI reduced MAE from 443 to 259 kg ha⁻¹ and improved yield prediction accuracy by 41.5%. The modified model effectively captured temporal LAI dynamics and their co-variation with leaf nitrogen and biomass, resulting in more realistic growth trajectories and improved yield estimates.These results demonstrate that integrating multi-scale remote sensing LAI into a mechanistically enhanced DSSAT framework significantly improves early-stage yield prediction of winter rapeseed under both experimental and real-world conditions. The proposed approach provides a scalable and practical pathway for coupling UAV and satellite observations with crop growth models, supporting precision management and regional yield forecasting for rapeseed production