The use of machine learning algorithms provided by the Google Earth Engine (GEE) platform has been increasingly adopted. This study applied the Random Forest (RF) algorithm using both Sentinel-1/2 imagery provided by GEE and high-resolution CAS500-1 imagery to generate Land Use and Land Cover (LULC) classification maps. Based on these maps, a detailed change detection analysis was conducted between the two time points. The study area is Osong-myeon, Pyeongtaek City, Gyeonggi Province, Korea, and two satellite images from November 2022 and November 2023 were used. Three combinations of input datasets were tested for LULC classification: Sentinel-1/2, CAS500-1, and the Normalized Difference Vegetation Index (NDVI) of Sentinel-2. Classification accuracy and Kappa coefficients were reported for each case. The combination of Sentinel-1/2, CAS500-1, and NDVI was used for the change detection analysis. The results showed that 79.01% of the vegetation zone remained unchanged, while 30.10% of the bare zone originated from previously vegetated areas. Additionally, 21.66% of the forest zone in 2023 had newly converted from the bare zone compared to 2022. This study presents the feasibility of high-resolution imagery for precise LULC classification and change detection. It highlights the practical value of the GEE platform in environmental monitoring in a small area.



Keywords Google Earth Engine, CAS500-1, Change detection, Random forest, Image classification