CLIMATE-INTEGRATED MULTISPECTRAL REMOTE SENSING AND MACHINE LEARNING FOR COTTON AND WHEAT CROP MONITORING: A GOOGLE EARTH ENGINE-BASED FRAMEWORK APPLIED IN UZBEKISTAN
Abstract
Accurate, scalable, and timely crop monitoring is essential for agricultural resource management in Central Asia’s intensively irrigated landscapes. This study presents a Google Earth Engine (GEE)-based framework integrating Sentinel-2 and Landsat-8/9 multispectral imagery with climate covariates—CHIRPS precipitation and ERA5 evapotranspiration—for cotton and wheat classification in the Vobkent district, Bukhara region, Uzbekistan (2022–2025). Four machine learning classifiers—Support Vector Machine (SVM), Random Forest (RF), Classification and Regression Trees (CART), and Gradient Boosting (GB)—were evaluated in combination with six spectral indices (EVI, NDVI, NDWI, PSRI, GNDVI, MSAVI). Classification accuracy was assessed using Overall Accuracy (OA), Kappa coefficient, and per-class Producer’s and User’s Accuracy on an independent 30% hold-out test set. The best-performing configuration—Sentinel-2 + EVI + SVM with precipitation and evapotranspiration covariates—achieved OA = 0.911 and Kappa = 0.795. Adding climate variables increased OA by 8–10 percentage points relative to spectral indices alone. The framework was operationalized as the Smart Crop MSI web application, providing real-time spectral index visualization, automated monthly time-series charts, and GeoTIFF/CSV export. Remote sensing and field monitoring results agreed within 3%, confirming the high reliability of the proposed methodology for multi-year crop area change detection.