Integrating random forest and crop modeling improves the crop yield prediction of winter wheat and oil seed rape

The fast and accurate yield estimates with the increasing availability variety of global satellite products rapid development new algorithms remain a goal for precision agriculture food security. However, consistency reliability suitable methodologies that provide crop outcomes still need to be explored. study investigates coupling modeling machine learning (ML) improve prediction winter wheat (WW) oil seed rape (OSR) provides examples Free State Bavaria (70,550 km 2 ), Germany, in 2019. main objectives are find whether approach [Light Use Efficiency (LUE) + Random Forest (RF)] would result better more predictions compared results provided other models not using LUE. Four different RF [RF1 (input: Normalized Difference Vegetation Index (NDVI)), RF2 climate variables), RF3 NDVI RF4 LUE generated biomass variables)], one semi-empiric model were designed input requirements best predictors monitoring. indicate individual use (in RF1) variables RF2) could most accurate, reliable, precise solution monitoring; however, their combined RF3) resulted higher accuracies. Notably, suggested can reduce relative root mean square error (RRMSE) from ?8% ?1.6% increase R by 14.3% (for both WW OSR), just relying on Moreover, research compares outputs inputting three spatial inputs: Sentinel-2(S)-MOD13Q1 (10 m), Landsat (L)-MOD13Q1 (30 MOD13Q1 (MODIS) (250 m). S-MOD13Q1 data has relatively improved performance [0.80 (WW), 0.69 (OSR)], lower RRMSE (%) (9.18, 10.21) L-MOD13Q1 m) Satellite-based biomass, solar radiation, temperature found influential crops.