The study was conducted in Mahabubnagar district, Telangana, during
kharif, 2024, where accurate rice yield estimation is crucial for food security planning and agricultural management, particularly under diverse agro-climatic conditions and fragmented landholdings. A semi-physical model (SPM) was adopted, integrating multi-source datasets: Sentinel-1 synthetic aperture radar (SAR) and Sentinel-2 NDVI for rice area mapping, meteorological inputs (rainfall, temperature, solar radiation), and physiological crop parameters. Sentinel-1A backscatter data were processed using temporal filtering and supervised machine learning (Random Forest) classification, combined with Sentinel-2A/2B optical data, to delineate rice cultivation areas with an accuracy of 98.33% and a kappa coefficient of 0.97 during the
kharif season, showing only a 0.76% deviation from district statistics.
Rice yield estimation was driven by net primary productivity (NPP), derived from meteorological and crop growth parameters, then converted to grain yield using a crop-specific harvest index. Calibration and validation with ground-truth field survey data yielded strong predictive performance (R² = 0.79, RMSE = 1159 kg ha-1, nRMSE = 24.19%), demonstrating the model’s effectiveness in capturing spatial and temporal yield variations at the regional scale.
