Improving Land Cover Classification Accuracy Using Ancillary Data: A Sentinel-2 and Google Earth Engine Study in 19 Mayıs District, Türkiye

The application of machine learning algorithms to remote sensing data enables the accurate classification of land cover, which is essential for environmental monitoring, land use planning, and sustainable natural resource management. In this study, enhanced land cover classification has been done using Sentinel 2A remote Sensing imagery by doing a comparison between Random Forest (RF) and Support Vector Machine (SVM) in Google Earth Engine (GEE) environment. We consider three different datasets for the performance assessment, particularly for the district 19 Mayis. Three datasets with spectral bands, spectral indices, and topographical features (elevation and slope) have been employed. We generated the evaluation metrics and calculated the overall accuracy (OA), the Kappa statistic (K), the user's accuracy (UA), and the producer's accuracy (PA). Overall, the RF model consistently outperformed the SVM model on each dataset. The SVM model in Dataset one gave an OA of 0.888 and K value of 0.849 but on the other hand the RF model in Dataset 1 gave OA of 0.927 and K value higher than SVM was 0.900. Based on Dataset 2 was RF with OA and K of 0.943 and 0.922 respectively. SVM model achieved OA of 0.912 and K of 0.880. The RF model achieved an OA of 0.965 and a K value of 0.952 according to Dataset 3 results while the OA of the SVM model was 0.927 and a K of 0.900. The results prove that integrating remote sensing data with advanced machine learning classifiers, particularly Random Forest, provides an effective approach for land cover mapping in complex and heterogeneous environments