International Journal of Plant & Soil Science

Monitoring coffee productivity is crucial for maximizing export potential, yet conventional field-based methods remain inefficient for large-scale applications. Remote sensing offers spectral and environmental insights linked to crop physiology, enabling spatially and temporally explicit yield prediction. This study aimed to develop an efficient and accurate yield-prediction model by integrating AutoML with remote-sensing–derived predictors, including vegetation indices, soil temperature, climate, and topographic variables. The study was conducted in 2023 in Bandung Regency, West Java, Indonesia, using yield observations from 73 field plots. Decision tree (DT), random forest (RF), extremely randomized trees (ExtraTrees), extreme gradient boosting (XGBoost), and light gradient boosting machine (LightGBM) as individual models and ensemble structure were implemented using the AutoML tool in ArcGIS Pro. Among them, the ensemble structure achieved the highest predictive performance (R² = 0.85; RMSE = 960.17 kg ha⁻¹), outperforming all individual models. Model interpretation indicated that Land Surface Temperature (LST) and precipitation (CHIRPS) were the most influential predictors governing yield variability across smallholder coffee plots. These findings demonstrate the potential of integrating AutoML and remote-sensing–derived variables as an efficient, scalable approach for yield prediction in heterogeneous smallholder coffee systems.