Integrating Random Forest and Spectral Similarity Index using Landsat Data for the Identification of Mining Areas in Samarinda and Surrounding Regions

Abstract

The growth of coal mining activity in Samarinda City has led to rapid land cover changes, impacting environmental conditions and surface geology. Remote sensing is an efficient method for monitoring these changes because it provides extensive spatial and temporally consistent data. This research aims to develop a more accurate mining area mapping approach by integrating the Random Forest algorithm and Spectral Index Similarity Analysis on Landsat 9 imagery from 2024. The first phase used Random Forest to map five main land cover classes: forests, crops, urban areas, water bodies, and bare land. The Random Forest algorithm demonstrated high performance with an Overall Accuracy of 94.7% and a Kappa Coefficient of 0.92. However, the bare land class still exhibited spectral confusion with mining areas due to similar reflectance in the red-SWIR channel. To address this limitation, the second phase applied NDVI, NDBI, and BSI-based analysis specifically to bare land areas to distinguish active mines from non-mining surfaces. The spectral logic of low NDVI (<0.25), high BSI, and BSI > NDBI has proven effective in identifying overburden characteristics and dry mining surfaces. The final mapping showed that the identified mining area in 2024 reached 4,246.26 hectares, or approximately 5.9% of the total administrative area of Samarinda City. Coal mines are concentrated in the eastern, southern, and partly western parts of the city, reflecting the intensity of mining activity in the Balikpapan Formation and Pulau Balang Formation lithologies. These findings demonstrate that the hybrid Random Forest and Spectral Index approach can improve the accuracy of mining area separation and can serve as a basis for continuous coal mining spatial monitoring.