Integrating UAV and satellite LAI data into a modified DSSAT-rapeseed model to improve yield predictions

Abstract

Context

Yield estimation in the fall is crucial for effective pre-winter management of winter rapeseed. Integrating remotely sensed leaf area index (LAI) with crop models has great potential for improving simulations of crop yields.

Objective

The objective of this study was to modify the DSSAT-Rapeseed model and by integrating LAI adjustments from satellite and unmanned aerial vehicle (UAV) images to improve the accuracy of rapeseed yield predictions at early stages from both experimental plots and actual farm fields.

Methods

A new pest definition, called "target LAI," was created in the COGRO048.PST file within the pest module of DSSAT. The DSSAT model was then modified to adjust leaf weight, leaf area, and leaf nitrogen content based on remotely sensed target LAI. Field investigations and UAV-derived LAI data from two years and two experimental stations were used to calibrate model parameters through a trial-and-error method, selecting the parameter set that minimized the error between model outputs (e.g., LAI and crop yield) and observations. The model's performance was tested with yield data from a different year at the same stations, using pre-winter LAI assimilated through the Ensemble Kalman Filter (EnKF). For actual farm fields, dynamic LAI data from Sentinel-2A was integrated with the modified DSSAT model for yield simulation and compared with ground measurements.

Results

By assimilating LAI into the modified DSSAT model, the mean absolute error (MAE) for yield simulation was reduced from 452 to 234 kg/ha in the experimental plot and from 443 to 259 kg/ha in actual farm fields compared to the original DSSAT model.

Conclusions

Integrating UAV and satellite LAI during pre-winter into the modified DSSAT model using data assimilation (EnKF) improved the rapeseed yield prediction.