High-precision crop rotation mapping in arid agroecosystems: integrating sentinel-2 time series, SVM, and GIS

The increasing global demand for food and the progressive constraints on natural resources underscore the need for advanced remote sensing frameworks to monitor, manage, and analyze agricultural systems. This study aimed to identify dominant crop-rotation patterns and quantify their spatiotemporal dynamics in an agricultural region of Shush County, Khuzestan Province, Iran. We utilized a three-year Sentinel-2 Level-2A time series (2023-2025) and implemented a supervised classification based on the Support Vector Machine (SVM). Training samples were derived from field surveys, NDVI time-series analysis, and the local cropping calendar. Final rotation maps were generated by layer-intersection operations in ArcMap. Accuracy assessment demonstrated stable, high precision across years: overall classification accuracies for rotation class's wheat-wheat-wheat, wheat-rice-wheat, and wheat-canola-wheat were 98.55%, 98.33%, and 98.58%, respectively, confirming the algorithm's strong capability for crop discrimination and temporal consistency. Spatiotemporal analyses indicated that the three-year rotations wheat-rice-wheat and wheat-wheat-wheat occupied the largest shares of cultivated area, whereas wheat-canola-wheat represented a minor proportion. The integration of Sentinel-2 time-series, SVM classification, and GIS spatial analysis provides an accurate and operational framework for regional crop-rotation monitoring. These findings highlight the method's potential to support sustainable agricultural planning and decision-making through reliable multi-year rotation mapping.