Machine Learning- and Remote Sensing-Based Lithological Mapping Using VNIR + SWIR PRISMA Hyperspectral and ASTER Multispectral Datasets in Northwest of Queensland

Lithological mapping is essential for geological studies, mineral exploration, and environmental assessment. Satellite remote sensing combined with machine learning provides a scalable, cost-effective approach for regional lithological discrimination. This study evaluates hyperspectral and multispectral satellite imagery for lithological mapping in a geologically complex region of northwestern Queensland, Australia. The study area, within the Mount Isa Inlier, comprises diverse sedimentary, volcanic, intrusive, and metamorphic lithologies. PRISMA hyperspectral and ASTER multispectral imagery were analyzed using supervised classification algorithms, including Support Vector Machine (SVM), Mahalanobis Distance (MaDC), Minimum Distance (MDC), and Maximum Likelihood (MLC). Image-derived endmembers from representative lithologies were used as training data. Classification accuracy was assessed using confusion matrices, Overall Accuracy (OA), and the Kappa coefficient. PRISMA imagery outperformed ASTER data. SVM achieved the highest performance for PRISMA (OA = 82.03%, Kappa = 0.81), whereas MLC achieved the highest performance for ASTER (OA = 33.29%, Kappa = 0.30). Classification accuracy was evaluated using an independent set of validation ROIs that were spatially separated from the training samples, providing a more reliable estimate of model performance. These results highlight the benefits of hyperspectral remote sensing with machine learning for lithological discrimination in complex terrain and emphasise the importance of spatially independent validation. The approach demonstrates strong potential for regional-scale applications and may support more efficient mineral exploration and geological mapping workflows.

Minerals

Minerals

16

2075-163X

N/A

7

33