Land Cover Classification Analysis of Sentinel-2 Multispectral Imagery Using Google Earth Engine

Abstract

Land cover represents the physical condition of the Earth’s surface, comprising both natural and human-made elements, and plays an essential role in environmental studies such as spatial planning, natural resource management, disaster mitigation, and environmental change monitoring. Semarang City has experienced rapid spatial development driven by urbanization and infrastructure expansion, which has led to significant land cover changes. This study aims to evaluate land cover classification results in part of Semarang City using Sentinel-2 satellite imagery, which provides spatial resolution of 10–60 meters and multispectral information suitable for classification purposes. To improve classification accuracy, spectral index transformations including the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Normalized Difference Built-up Index (NDBI) were applied as additional variables. Data processing was conducted through the Google Earth Engine (GEE) platform, which enables fast and efficient cloud-based computation without the need to download large datasets. Land cover classification was performed using two multispectral classification methods, namely K-Means and Random Forest, and was evaluated using Overall Accuracy and the Kappa Coefficient to assess the reliability of the classification results. The findings are expected to provide reliable spatial information on land cover conditions and serve as a basis for urban spatial planning and sustainable environmental management.