International Journal of Environment and Climate Change

Accurate mapping of croplands and crop types is critical for food security assessment and climate-resilient agricultural planning. However, discrimination of individual crop types and cropping patterns using spaceborne observations remains challenging due to spectral similarity among crops. This study aims to evaluate a robust crop inventory mapping using both optical and microwave remote sensing datasets within the Google Earth Engine (GEE) cloud computing platform in the Kommamuru canal command area during the Kharif and Rabi seasons from 2022 to 2024. This study developed a machine-learning methodology for cropland and crop-type classification using time-series Sentinel-2 optical data and Sentinel-1 synthetic aperture radar (SAR) data implemented on the GEE platform. The methodology involved (i) Preprocessing and temporal composition of Sentinel-2 visible, near-infrared, red-edge, and shortwave infrared (SWIR) bands; (ii) Extraction of vegetation indices; (iii) Integration of Sentinel-1 backscatter features; and (iv) Supervised classification using a transfer learning-based approach. The framework was applied to identify cropland and non-cropland in the first step and to classify specific major cropping patterns: paddy, tobacco, black gram, maize, green gram, chilli, jowar, and groundnut. The comparison was based on sensor characteristics, classification behaviour, and seasonal crop condition accuracies. Cropland and non-cropland classification achieved producer’s accuracies across different feature combinations. Quantitative analysis showed that inclusion of SWIR bands significantly improved classification accuracy compared to visible and near-infrared bands alone, while the addition of red-edge bands further enhanced crop discrimination. Optical classification during Kharif achieved high overall accuracies ranging from 80.24% to 84.32%, whereas SAR-based classification recorded lower accuracies ranging from 35.28% to 46.41%. Similarly, optical classification during Rabi produced overall accuracies ranging from 74.34% to 79.41%, while SAR classification showed comparatively low performance between 27.42% and 37.41%. The results demonstrated that integrating optical and microwave remote sensing data within a transfer learning framework on GEE enables more accurate crop-type and cropping-pattern mapping, providing robust support for precision agriculture and food security monitoring.