What question did this study set out to answer?

The aim is to enhance remote sensing-based hydrothermal alteration mapping for porphyry copper deposits using machine learning techniques.

June 9, 2026Open Access

Metaheuristic-optimized machine learning framework for remote sensing-based alteration mapping of porphyry copper systems

Key Points

The aim is to enhance remote sensing-based hydrothermal alteration mapping for porphyry copper deposits using machine learning techniques.
Integrated Boosted Trees and Quadratic Support Vector Machines with Shuffled Frog Leaping Algorithm for optimization.
Utilized ASTER and Sentinel-2 multispectral imagery to identify various alteration zones.
Conducted validation through field mapping, petrographic analysis, and X-ray diffraction.
Increased AUC values from 0.89 to 0.94 for Boosted Trees, and from 0.88 to 0.93 for Quadratic Support Vector Machines.
Achieved over 78% spatial agreement between predicted and observed alteration zones using independent validation.
Demonstrated the effectiveness of metaheuristic optimization in classifying complex mineralogical relationships.

Abstract

Remote sensing-based hydrothermal alteration mapping is a pivotal technique in critical mineral exploration, particularly for identifying porphyry copper deposits (PCDs). However, conventional multispectral approaches are often constrained by linear assumptions, spectral mixing, and suboptimal parameter selection, limiting their ability to resolve complex alteration assemblages. This study presents a metaheuristic-optimized machine learning framework that integrates Boosted Trees (BT) and Quadratic Support Vector Machines (QSVM) with the Shuffled Frog Leaping Algorithm (SFLA) to improve alteration zone discrimination using multispectral satellite imagery. ASTER shortwave infrared bands were combined with Sentinel-2 visible, near-infrared, and red-edge features to map argillic, phyllic, propylitic, and iron oxide/hydroxide alteration zones. The framework was applied to the Shahr-e-Babak district within the Urumieh–Dokhtar Magmatic Belt, Iran, a well-exposed and mineralized porphyry copper province. Metaheuristic optimization significantly enhanced classification performance relative to baseline models, increasing area-under-the-curve (AUC) values from 0.89 to 0.94 for BT and from 0.88 to 0.93 for QSVM. The optimized BT model consistently outperformed QSVM, highlighting the robustness of ensemble learning in spectrally heterogeneous geological environments. Independent validation using field mapping, petrographic thin-section analysis, and X-ray diffraction demonstrated strong geological consistency, with more than 78% spatial agreement between predicted and observed alteration zones. These results show that integrating metaheuristic optimization with ensemble- and kernel-based classifiers enables reliable learning of complex, non-linear spectral–mineralogical relationships from multispectral data. Beyond this case study, the proposed workflow is computationally efficient, scalable, and transferable, offering a generalizable decision-support tool for high-precision porphyry copper exploration in arid and semi-arid regions worldwide.

Ask AI

Helpful

Bookmark

View Full Paper