Rice Yield Prediction Using Sentinel-1 Radar Vegetation Indices and XGBoost

Monitoring rice yield is crucial for sustainable and resilient agricultural systems and global supply chains that ensure worldwide food security. Optical vegetation indices (OVIs) are significant parameters for predicting rice yield via remote sensing. Previous studies show that radar vegetation indices (RVIs) exhibit correlations with OVIs and vegetation water content. However, research on RVIs' integration into rice yield prediction is currently in its preliminary stage. In this study, rice yield prediction was completed using RIVs and the XGBoost regression model. Sentinel-1 VH/VV polarization data and county-level rice statistics covering Guangdong Province, China, spanning from 2017–2021 were used. The experiment results demonstrated that the rice yield prediction performance of RVIs was comparable to that of OVIs. In early rice yield prediction, the R2 was 0.54 and the RMSE was 259.03 kg/ha based on RVIs. In late rice yield prediction, the R2 was 0.69 and the RMSE was 405.21 kg/ha based on RVIs. The experimental results demonstrated the excellent applicability of the RVIs for rice yield prediction, with potential contributions to global sustainable development.