Combined use of Sentinel 1 and Sentinel 2 for urban land cover mapping using Google Earth Engine in the case of Adama City, Ethiopia.

Abstract

Urban land cover mapping plays a vital role in understanding urbanization patterns and their associated environmental consequences. This study investigated the effectiveness of combining Sentinel-1 Synthetic Aperture Radar (SAR) and Sentinel-2 Multispectral Instrument (MSI) satellite data for urban land cover classification in Adama City, Ethiopia, utilizing the capabilities of Google Earth Engine (GEE). The primary objectives included classifying urban land cover types, detecting changes in land cover between 2015 and 2025, and assessing the classification performance of two machine learning algorithms Support Vector Machine (SVM) and Random Forest (RF). The study employed GEE for efficient data access, preprocessing, and analysis. Preprocessing procedures included radiometric and geometric corrections, cloud masking, and image stacking. Classification was conducted using SVM and RF, and a subsequent change detection analysis was carried out to track land cover transformation over the ten-year period. Results indicated substantial land cover changes in Adama City. Built-up areas expanded by 219.37%, growing from 1,679.38 hectares in 2015 to 5,364.57 hectares in 2025, reflecting rapid urban development. Agricultural land experienced a slight decline of 2.33%, decreasing from 21,742.77 hectares to 21,235.69 hectares, likely due to conversion into urban zones. Forest areas saw a significant reduction of 50.61%, dropping from 5,201.93 hectares to 2,569.65 hectares, pointing to accelerated deforestation. Similarly, shrubland decreased by 32.75%, while bareland increased dramatically by 346.18%, rising from 87.58 hectares to 390.83 hectares. In terms of classification accuracy, the SVM algorithm yielded better results for Sentinel-2 MSI data, achieving an overall accuracy (OA) of 83% and a Kappa coefficient of 0.78, compared to RF's OA of 74% and Kappa of 0.67. For Sentinel-1 SAR, RF outperformed SVM with an OA of 85% and a Kappa of 0.81, while SVM reached an OA of 75% and Kappa of 0.68. When the datasets were combined, RF again showed stronger performance (OA = 83%) compared to SVM (OA = 77%), illustrating the benefits of data fusion for urban land cover classification. These findings demonstrate the potential of integrating radar and optical satellite data for enhanced urban monitoring and planning. The study recommends incorporating additional data sources and employing more advanced machine learning methods to further improve classification precision and support sustainable urban development in rapidly expanding cities.