Application of Self-Organizing Maps (SOM) to 3D Geophysical Inversion Models in the DroneSOM Project

The DroneSOM project, co-funded by EIT RawMaterials, aims to develop an integrated approach to geophysical data acquisition and interpretation for mineral exploration. Drones are used to collect gravity and electromagnetic data, that is subsequently analyzed using specialized software. The three-year project is coordinated by the Geological Survey of Finland (GTK) and carried out in partnership with RADAI Oy, the Technical University of Denmark (DTU), and Beak Consultants GmbH. It comprises six work packages that include field measurements at various sites in Finland, Germany, and Sweden, as well as data integration and interpretation.

This contribution focuses on the application of Self-Organizing Maps (SOM) for the analysis of three-dimensional geophysical voxel data derived from 3D inversion of the surface-based measurements. The methodology involves preprocessing and normalization of the input data, SOM-based clustering, and subsequent evaluation and validation using sensitivity analyses, boxplots to assess data distribution within the resulting clusters, and evaluate spatial correlation with existing geological knowledge.

Results from the application of this method to a test site in Finland are presented. The analysis demonstrates how 3D-inverted geophysical data can be structured and visualized using 3D SOM to reveal geological patterns. The combination of 3D data processing and unsupervised machine learning enables the interpretation of subsurface structures and opens new perspectives for geoscientific data analysis.