Artificial Intelligence-Based Lithology Classification Using Sentinel-1 Data in Amurang, Sulawesi, Indonesia

One of the biggest challenges temperate and tropical regions face is that dense forest covers much of the landscape, which can be problematic in lithological mapping. Synthetic Aperture Radar (SAR) data provides a window through heavily vegetated canopy and essential information about surface scattering that can be used to infer underlying lithology. This research proposes a new methodology for lithology classification based on Sentinel-1 SAR nested geospatial data and a hybrid Artificial Intelligence (AI) and Geographic Information Systems (GIS) technique. The purpose of this study is to demonstrate the ability of AI, GIS, and Sentinel-1 data to classify lithology in the heavy jungle of Amurang, Sulawesi, Indonesia. The results indicate the proposed method can accurately map 1:50,000 scale lithology and refine the Qv unit into young volcanic rocks (Qv) and young lava (Qvl) and further define the Qvl unit into three sub-units based on age where Qvls-1 is the younger and Qvls-3 is the older. Cross-validated results indicate our method identified lithology with an overall accuracy of 91.00%, a commission error rate of 3.03%, and an omission error rate of 2.15% compared to the 2006 X-band InSAR derived geological map of the Amurang, Sulawesi. The proposed method distinguishes and refines specific rock units and has the potential to semi-automate lithological mapping in heavily vegetated areas.