<![CDATA[Lithological discrimination in humid tropical islands remains constrained by dense vegetation, deep weathering, regolith cover, mixed pixels, and discontinuous bedrock exposure, which collectively weaken diagnostic spectral responses. This study evaluates the capability and limitations of Landsat-9 OLI-2 for island-scale lithological discrimination in Langkawi Island, Malaysia, using an interpretable optical-only workflow. Atmospherically corrected Landsat-9 imagery was processed through false-color composites, Optimum Index Factor (OIF)-based band selection, band-ratio enhancement, Principal Component Analysis (PCA), Normalized Difference Vegetation Index (NDVI), Jeffries–Matusita separability analysis, and Maximum Likelihood Classification (MLC). The resulting MLC lithological map was assessed pixel-by-pixel against a published geological map; consequently, the reported statistics represent map-derived agreement rather than independent field-validated lithological accuracy. Results show that the OIF-selected RGB 6-5-2 composite, selected band-ratio combinations, and PCA enhanced broad contrasts among Quaternary Alluvium, granitic terrain, and carbonate-bearing formations. The MLC classification achieved an overall map-derived agreement of 51.72% and a kappa coefficient of 0.4177. Qal, Cm-SS/Sh/Md, and OS-Ls/SS showed relatively stronger agreement, whereas PT-Ls/Mb, DP-St/Md, and Tr-Gr were more affected by spectral overlap and class confusion. NDVI-stratified assessment further confirmed that vegetation cover influences classification performance, with low-vegetation areas producing higher agreement than moderate-vegetation areas. This study establishes a reproducible full-island baseline for evaluating optical multispectral lithological mapping under humid tropical conditions. These findings demonstrate that Landsat-9 OLI-2 can support reconnaissance-level lithological discrimination in humid tropical islands but remains insufficient for precise formation-level mapping without field validation and integration with SAR, DEM-derived, or higher-resolution spectral datasets.]]>
