Improving early season crop type mapping using machine learning algorithms with combined SAR and optical time series

Timely and reliable crop type mapping is essential for sustainable agricultural management and food security. This study investigates the potential of integrating multi-temporal Synthetic Aperture Radar (SAR) data from Sentinel-1 and optical imagery from Sentinel-2 for early-season crop classification in Udham Singh Nagar district, Uttarakhand, India. Leveraging the Google Earth Engine (GEE) cloud platform, two ensemble machine learning algorithms, Random Forest (RF) and Gradient Boosted Trees (GBT), were implemented and evaluated for their ability to classify major land cover and crop types using combined SAR-optical datasets acquired during November–December 2022. A total of 1141 ground truth samples were collected through field surveys. The integration of structural (SAR) and spectral (optical) information substantially enhanced classification performance compared with single-sensor approaches. Among the models, GBT achieved the highest accuracy, with an overall accuracy of 93% and a Kappa coefficient of 0.90. Wheat, forest, and water bodies were mapped with high precision (F1 ≥ 0.90), whereas mustard and sugarcane showed lower accuracy due to spectral similarity and limited training data. Feature importance analysis identified late-season SAR (VV, VH) and optical (red-edge and SWIR) bands as the most influential variables. The results demonstrate that the synergy between SAR and optical time-series data, coupled with ensemble machine learning, provides a robust and scalable framework for near real-time crop monitoring. This approach offers valuable insights for precision agriculture and supports data-driven policy planning in India’s agricultural sector.