GFM4MPM: Towards Geospatial Foundation Models for Mineral Prospectivity Mapping

TL;DR

This work tackles mineral prospectivity mapping under severe label scarcity by introducing a geospatial foundation-model framework trained with masked image modeling on unlabeled multi-band rasters. A single encoder backbone is pretrained in a self-supervised manner and then reused by a lightweight classifier for downstream MPM tasks, with a novel positive-unlabeled learning strategy to mitigate mislabeling. The method also integrates Integrated Gradients for per-prediction explanations and MC Dropout for epistemic uncertainty, producing both mean prospectivity and uncertainty maps. Across MVT and CD deposit datasets, the approach outperforms baselines on key metrics and demonstrates robustness to input sparsity, signaling a meaningful step toward generalizable geospatial foundation models for mineral exploration.

Abstract

Machine Learning (ML) for Mineral Prospectivity Mapping (MPM) remains a challenging problem as it requires the analysis of associations between large-scale multi-modal geospatial data and few historical mineral commodity observations (positive labels). Recent MPM works have explored Deep Learning (DL) as a modeling tool with more representation capacity. However, these overparameterized methods may be more prone to overfitting due to their reliance on scarce labeled data. While a large quantity of unlabeled geospatial data exists, no prior MPM works have considered using such information in a self-supervised manner. Our MPM approach uses a masked image modeling framework to pretrain a backbone neural network in a self-supervised manner using unlabeled geospatial data alone. After pretraining, the backbone network provides feature extraction for downstream MPM tasks. We evaluated our approach alongside existing methods to assess mineral prospectivity of Mississippi Valley Type (MVT) and Clastic-Dominated (CD) Lead-Zinc deposits in North America and Australia. Our results demonstrate that self-supervision promotes robustness in learned features, improving prospectivity predictions. Additionally, we leverage explainable artificial intelligence techniques to demonstrate that individual predictions can be interpreted from a geological perspective.

Figures (16)

• Figure 1: Overview of our Mineral Prospectivity Mapping (MPM) approach. Left-to-right: geospatial data preprocessing; geospatial Self-Supervised Learning; MPM using extracted SSL features; and validating MPM predictions using feature importances
• Figure 2: Illustration of data processing with multi-band georeferenced raster $X$. Red and green prisms are samples $x'_i \in \mathbb{R}^{m \times w \times w}$, with red ✗ and green ✓ representing absent and present center pixel labels for the samples, respectively.
• Figure 3: Illustration of masked image modeling pretraining procedure for a single explanatory feature layer.
• Figure 4: Eight masked image modeling pretraining examples showing input, 75% mask, masked input, and reconstruction left-to-right in columns.
• Figure 5: MVT Lead-Zinc prospectivity map showing likelihoods overlayed onto uncertainties. Deposits are grouped by usage: training, validation, or testing.