SPRC: A novel Sentinel-1/-2 Phenology-enhanced framework for automated paddy rice mapping

• A new automated training sample generation method. • A new process for time-series feature selection and optimization. • Robust multi-layer probabilistic Random Forest classification framework. Accurate mapping of rice cultivation regions is essential for addressing global food challenges and developing climate adaptation strategies. In large-scale paddy rice mapping, obtaining sufficient training samples that capture the range of phenological variations is difficult. Cloud contamination and missing data complicate the collection of spatially and temporally complete observations for reliable rice mapping missions. This study introduces the Sentinel-1/-2 phenology-enhanced paddy rice (SPRC) mapping framework. Automated rice mapping is achieved by integrating Sentinel-2 phenology, Sentinel-1 V-shape features, Jeffries–Matusita distance, SPRC-based feature selection, and multi-layer probabilistic random forest classification with voting. The method demonstrated robust performance across varying weather conditions, achieving overall accuracy and F1 scores of 96.21 % and 96.66 in Heilongjiang Province (HLJ), China, and 95.76 % and 95.62 in Sakata (Sak), Yamagata, Japan, respectively, outperforming previous approaches. When compared with statistical data in HLJ, the maps exhibit a highly linear correlation and achieve an R 2 value of 0.996, corresponding to a paddy rice area of just 1.1 % less than the reported average. The results reveal that the optimal time-series feature combinations for paddy rice mapping are strongly affected by spatial, temporal, and weather-related variations. The SPRC framework successfully produced rice maps for HLJ from 2019 to 2021 and for Sak in 2020. These findings illustrate the SPRC framework’s efficacy and precision in mapping paddy rice. The 3-year rice data are invaluable for food surveillance, regional agricultural monitoring, and policy-making.