LAUSR

Crop rotation increases crop yield, improves soil health, and reduces plant disease. Mapping crop rotation is difficult because crop data from a single time point do not sufficiently represent the dynamics of a system. Studies have tried to map crop rotation by sequentially combining crop maps. However, this produced a large number of meaningless crop sequences, hindering the assessment of rotational benefits at regional scales. Here, we propose a crop rotation classification scheme that integrates temporal information into static crop maps and develop an innovative approach to map crop rotation. We chose a typical multiple cropping region in southern China. Given that the landscape is characterized by high crop diversity (e.g., food crops, cash crops, and permanent crops) and variable cropping intensity, our classification scheme first defines three main rotation systems, i.e., paddy, vegetable, and orchard systems, and then further divides the systems into nine subsystems according to their seasonal dynamics. Finally, we apply time series of Sentinel-1 and Sentinel-2 images to identify the systems by a hierarchical rule-based method. The map of crop rotation systems in 2020 had producer, user, and overall accuracies of 81%, 79%, and 81%, respectively. The results indicate that integrating temporal information into the classification scheme is vital to representing complex rotation systems and that remotely sensed temporal dynamics of crops are useful to characterize these systems. It also shows that crop rotation can be mapped directly rather than aggregating multiple crop layers, thus providing a new perspective for mapping and understanding crop rotation systems.