A Spectral–Semantic integration framework for hydrothermal alteration characterization using Landsat-8 OLI and AlphaEarth Foundations in the Eastern Tien Shan, China

Remote sensing-based hydrothermal alteration mapping plays an important role in regional mineral prospectivity analysis but remains challenging in structurally complex regions because of spectral mixing, weak peripheral anomalies, and fragmented spatial distributions. This study proposes an integrated spectral–semantic framework that combines Landsat-8 OLI multispectral imagery with AlphaEarth Foundations (AEF) semantic embeddings for hydrothermal alteration characterization and prospectivity evaluation in the Tuwu copper belt, Eastern Tien Shan, China. Conventional spectral indices and principal component analysis (PCA) were first employed to extract iron-staining and hydroxyl alteration anomalies, providing a physically interpretable spectral baseline. Subsequently, AEF semantic embeddings were incorporated into a two-stage hierarchical XGBoost framework to complement conventional spectral information by characterizing alteration-related spatial organization. The first-stage classifier achieved an AUC of 0.975, whereas the second-stage classifier yielded a macro-average AUC of 0.983 for the proximity-based semantic classes. Comparative experiments using Landsat-only, AEF-only, and Landsat + AEF feature representations further demonstrated that AEF embeddings provide substantially stronger feature representations than conventional multispectral data, while feature concatenation increased Stage 1 AUC slightly without improving the reported Stage 2 metrics. Integrated interpretation of spectral alteration anomalies, semantic patterns, fault structures, intrusive rocks, and known copper deposits delineated six Class-A and three Class-B prospective areas, with Class-A targets combining high predicted anomaly levels and mapped copper occurrences and Class-B targets representing predicted anomaly areas without mapped copper occurrences. The results demonstrate that integrating physically interpretable spectral information with foundation-model semantic representations provides an effective approach for hydrothermal alteration characterization while offering complementary spatial information for regional mineral prospectivity analysis.