Satellite-derived phenological parameterization and classification framework for joint crop intensity and type mapping

Timely and detailed crop maps are essential for agricultural monitoring, particularly in regions with multiple cropping cycles per year. This study presents a unified satellite-based framework for jointly estimating crop intensity and crop type from dense optical time series using a single vegetation index, the Green Chlorophyll Vegetation Index (GCVI). Harmonic regression was applied to reconstruct smooth GCVI time series, followed by peak-based segmentation to identify crop-growing periods. Skewed normal functions were then fitted to each segment to derive phenological parameters, which were used as inputs to a Random Forest classifier. The method was applied to the Tsukushi Plain in Japan, where double cropping of rice, soybean, wheat, and onion is widely practiced, together with lotus cultivation. The framework achieved a crop-intensity accuracy of approximately 0.905 and a crop-type classification accuracy of 0.874 across five target crops. The double-cropping-aware approach improved overall accuracy compared with the annual GCVI harmonic baseline model, which achieved an overall accuracy of 0.789. Estimated planted areas showed strong agreement with official statistics for 2023, with R 2 values of 0.98 for rice, 0.99 for soybean, 1.00 for wheat, and 0.93 for onion. The approach also captured intra-annual crop management patterns, such as soybean block rotation in double-cropping regions, which are not represented in conventional annual land-cover maps. Although demonstrated using a single vegetation index and limited reference data, the proposed phenology-based framework is scalable and adaptable to other indices, sensors, and agricultural systems, enabling transparent and reliable crop monitoring in data-limited and multi-cropping environments.