Self-Organizing Map Classification of Magnetic-Derived Features for Ni-Cu-PGE Exploration along the Nickel Line, Northern Sweden

Sammanfattning

This study applies Self‑Organizing Maps (SOM) to airborne magnetic data from the Nickel Line in northern Sweden to identify magnetic‑texture domains that may reflect lithological contrasts, structural features, or sulphide-bearing systems within a predominantly paragneissic Paleoproterozoic terrain. The Nickel Line hosts several known Ni‑Cu‑Co/PGE occurrences associated with mafic-ultramafic intrusions and sulphide assemblages, yet many prospective zones remain poorly defined in regional geological mapping. To address this gap, a workflow was developed to extract peak-geometry and curvature-derived attributes from the SGU total magnetic intensity (TMI) grid and use these features as inputs to unsupervised SOM classification. Magnetic feature extraction involved identifying local maxima, minima, and inflection points along 500 m TMI profiles, followed by calculating peak_minus, peak_plus, and second‑derivative curvature values. For each grid cell, maxima and minima within a 2,500 m radius were summarised using mean and standard deviation statistics, producing an 8-input peak geometry dataset and an extended 12-input dataset incorporating curvature attributes. Two SOM models (5×5 architecture) were trained using these inputs, and their performance was evaluated using cluster‑validation metrics including silhouette score, Davies-Bouldin index, Calinski-Harabasz index, quantization error, same-neighbour fraction, and η² relationships with TMI. The 12-input SOM model outperformed the 8-input model across multiple validation criteria, producing more coherent spatial domains and stronger relationships with magnetic intensity. Curvature-enhanced classes (particularly Classes 4 and 5) showed the closest spatial association with mapped geological contacts, with 13-16% of grid cells lying within 1 km of a contact and median distances as low as 2.4 km. All SOM classes occur overwhelmingly within SGU-mapped paragneiss, indicating that the SOM is not identifying lithological units but instead detecting submap-scale variations in magnetic texture. These variations likely reflect differences in magnetic mineralogy, deformation intensity, or the presence of small or concealed mafic-ultramafic bodies, features known to influence Ni-Cu-PGE prospectivity. The results demonstrate that SOM-based magnetic texture analysis provides a robust framework for identifying subtle geological domains not captured in regional mapping. The preferred 12-input SOM model highlights several magnetic‑texture zones that coincide with known mineral occurrences and structural corridors, suggesting potential for delineating additional targets for Ni‑Cu‑Co/PGE exploration along the Nickel Line. Future work should integrate gravity, electromagnetic, or geochemical datasets to refine these domains and assess their mineral potential across other Swedish greenstone belts.