Mapping Urban Vegetation Change Using RGB-Based K-Means Clustering In Baubau City
Abstract
Urban ecological monitoring is increasingly essential in rapidly developing tropical regions. This study investigates land-cover changes and vegetation dynamics in Baubau City, Indonesia, from 2019 to 2025 using an RGB-based unsupervised classification framework. The primary aim is to develop a practical, cloud-enabled approach for analyzing multitemporal satellite imagery where access to multispectral data is limited. RGB composites were processed using K-Means clustering, Cosine Similarity matrices, and Elbow validation to extract dominant land-cover classes over time. The analysis revealed four major land-cover classes: dense vegetation, mixed vegetation, open land, and built-up areas. A significant ecological disturbance was identified in 2022, characterized by a sudden decline in vegetation and increased spectral similarity across clusters, suggesting both urban expansion and atmospheric interference. By 2023–2025, vegetation began to recover, though in a more fragmented pattern. Sub-clustering of dense vegetation confirmed this shift, with very dense classes giving way to transitional zones. The findings demonstrate the capability of RGB-only methods to produce ecologically meaningful classifications when paired with robust analytical techniques. The study also highlights the benefits of using cloud platforms like Google Earth Engine and Colab for accessible environmental monitoring. These insights support urban planning efforts and encourage the integration of lightweight remote sensing approaches into decision-making processes in resource-constrained settings.