Semi-automatic paddy rice intensity mapping for southern China with multiple cropping systems

• A new semi-automatic paddy rice intensity mapping framework is proposed. • This framework applies to diverse climatic and geographical conditions. • A rice feature set is developed by considering multiple phenological stages. • An automated method for dividing the growing season is proposed. Accurate paddy rice cropping intensity (PRCI) information is crucial for yield estimation, food policy formulation, and water resources management. However, the high cultivation intensity, fragmented farmland, complex crop rotation systems, and overlapping crop growth stages in southern China make it challenging to differentiate rice varieties from other crops. To address this limitation, we propose a semi-automatic framework for mapping PRCI in southern China. This framework first fused MODIS and Sentinel-2 data to generate high-spatiotemporal-resolution vegetation indices, forming the data foundation to identify the key phenological stages of paddy rice. Then, an automated division approach is designed to divide crop-growing seasons. Based on that, we developed a discriminant feature set for rice and identified potential rice-planting regions using an unsupervised Gaussian mixture model. Finally, we obtained reliable 10-m PRCI maps from PRP regions by integrating statistical data, Sentinel-1 data, limited rice samples, and the time-weighted dynamic time warping algorithm. Experimental results showed that our framework consistently achieved a high mapping accuracy of 87 % across various regions with diverse landforms and agricultural practices. Furthermore, the framework showed a strong correlation with municipal-level statistical data (R 2 = 0.99) and ideal adaptability to complex landscapes and crop rotation systems.