Automated Geophysical Pattern Recognition for Mineral Exploration: Case Study of Carajás, Brazil Using IOCG Deposits Sofia Mantilla Salas1**, Pablo Mejia-Herrera2, Jonas Kloeckner1, Adel Asadi1 David Zhen1,Jef Caers1 ** Presenting author: sofiams@stanford.edu) 1 Mineral-X, Earth and Planetary Sciences Department, Stanford University 2Northisle Copper and Gold Abstract Mineral exploration targeting is a high-impact and high-uncertainty decision problem, and the regional-to-camp phase of exploration targeting often remains slow, manual, and difficult to reproduce. This paper develops a quantitative, reproducible, and interpretable geophysical prospecting method that identifies anomalies analogous to a known economic deposit using spatial principal component analysis (sPCA). The approach samples regional geophysical grids with sliding windows, decomposes the reference anomaly of the known deposit into spatial modes, and ranks candidate windows by deposit-weighted similarity in PCA space. We demonstrate the method in the Carajás Mineral Province, Brazil, using total magnetic intensity (TMI) and radiometric U (U) data for IOCG targeting. In a univariate case using the Paulo Afonso deposit as reference, sPCA recovered 54.1% of the total test-deposit area by prediction rank 250, compared with 10.2% for a raw pixel-to-pixel TMI value baseline. In a multivariate case using Alemão as reference, the fused TMI + U workflow recovered 77.7% of the total test-deposit area and outperformed multivariate pixel-to-pixel raw and sPCA univariate rankings. These results indicate that sPCA provides a practical framework for similarity-based geophysical prospecting in brownfield exploration. Introduction Mineral exploration targeting is a decision problem characterized by high impact and high uncertainty [1, 2, 3, 4, 5, 6, 7]. Exploration teams reduce this uncertainty through a phased process that narrows the search from regional scale to target scale, using geological, geophysical, and geochemical data to identify anomalies and reserving drilling for only the best prospects [8, 3, 9, 10]. However, the regional-to-camp targeting phase often remains slow, manual, and difficult to reproduce, and is still strongly dependent on expert judgment [3, 11, 12, 13]. This creates space for complementary data-driven methods that can rank targets faster, more transparently, and more consistently [14, 15]. Data-driven methods are of growing interest because they can help extract meaningful targeting information from large, multi-source datasets and support more systematic and reproducible exploration workflows [16, 15]. Current work is dominated by mineral prospectivity mapping
RkJQdWJsaXNoZXIy MTM0Mzk2