Crop Type Mapping in Smallholder and Complex Agricultural Landscapes Using Multi-Temporal Sentinel-2 Imagery on the Google Earth Engine Platform

<title>Abstract</title> <p>This study addresses the challenge of accurate crop type mapping in heterogeneous smallholder agricultural landscapes where orchards and annual crops are intermingled, resulting in significant spectral overlap and seasonal asynchrony. We develop a hybrid framework integrating multi-temporal Sentinel-2 imagery (2020–2022), topographic covariates (DEM, slope, aspect), and phenological analysis within the Google Earth Engine (GEE) platform. The study area, Saman County (Chaharmahal-Va Bakhtiari Province, Iran)—a major almond and peach production hub—was classified into nine crop classes. NDVI time-series analysis combined with the Transformed Divergence (TD) index identified May-September as the optimal temporal window for maximum class separability. Following masking of residential areas and riverbeds, and incorporating topographic layers, a Random Forest (RF) algorithm trained on 70% of 1,500 field-sampled points achieved 95% overall accuracy and strong agreement with agricultural statistics (R² = 0.96, p &lt; 0.01). This framework demonstrates that integrating phenology-guided temporal composites, physiographic variables, and machine learning within GEE provides a scalable and efficient solution for accurate crop mapping in complex smallholder systems, supporting sustainable land management and irrigation optimization.</p>