Abstract
Accurate knowledge of rice growth stages underlies effective crop health assessment, productivity estimation, and food-security planning, since phenology data supports timely agricultural management and sustainable production practices. Synthetic Aperture Radar (SAR) provides a practical alternative to optical monitoring in this context, as its cloud-penetrating capability allows continuous data acquisition regardless of weather conditions. This study examines the capability of Sentinel-1 SAR data, combined with machine learning, for estimating rice growth stages and plant age. Vertical Transmit–Horizontal Receive (VH) backscatter served as the primary predictor, analyzed together with field observations across ten BBCH growth-stage classes; radar decomposition and machine learning were then applied to quantify growth stages and their relationship to crop age. VH backscatter sensitivity shifted considerably over the growing season, ranging from −24.03 to −17.98 decibels during early planting and increasing to −11.00 to −6.69 decibels at harvest. The resulting machine learning model classified growth stages with satisfactory accuracy, while the VH-to-age relationship achieved a Root Mean Square Error (RMSE) of 1.41, reflecting reliable estimation performance for operational rice growth monitoring