Within the Horizon 2020-funded project NEXT – New Exploration Technologies, a method called self-organizing maps (SOM) was used for geoscientific data integration and mineral prospectivity analysis. The Geological Survey of Finland (GTK) and Beak Consultants GmbH (BEAK) developed tools for implementing SOM, including the open-source standalone application GisSOM (developed by GTK) and SOM as an option for predictive modeling in BEAK’s advangeo® 2D Prediction software.
Currently, the versions of GisSOM and SOM in advangeo 2D Prediction only compute and report quantization error as a quality metric. However, several metrics for SOM quality assessment will be added as new results in the GisSOM and advangeo 2D and 3D Prediction software, including topological errors, SOM maps with the spread of variables within nodes, SOM map with mean QE of data points within each node, and map embedded accuracy for calculating the population-based test statistics.
The goal is to incorporate SOM-related uncertainties into the subsequent supervised classification workflow. The evaluation and interpretation of the SOM results can be used as input parameters for prediction modeling with ANN using the SOM codebook vectors. Two approaches are proposed for achieving this: Proposed Method 1 quantifies uncertainties using the variation of the weights for each neuron, while Proposed Method 2 calculates the Z-score (parametric) and/or rank-based score, e.g., U-statistic (non-parametric) of the best matching unit concerning the corresponding data points for each variable.
Moreover, advangeo 3D Prediction is developed for 3D integration of 3D geophysical data volumes or 3D geophysical models with a focus on mineral exploration, but it is also useful for more general geological data integration problems and application fields of the 3D potential modeling like geo risk analyses, hydrogeological or geothermal analyses. It was originally developed by Beak Consultants GmbH between 2014 and 2017, but it is not commercialized yet due to missing installation packages. In the frame of DroneSOM, it is planned to be improved.
Different measures can be used as convergence criteria for monitoring the training of the SOM and optimization of the SOM results. The quality measures that will be used and developed in DroneSOM for GisSOM and Advangeo software are described in the report (DroneSOM, D5.1 – Documentation of the SOM quality measures & 3D visualization tools, including technical specifications). Additionally, a methodology for assessing SOM-related uncertainties and using these as input parameters for prediction modeling with ANN using the SOM codebook vectors is also presented in the report.
Lastly, to be able to reach more clients and sell the software, it is important to be independent of additional commercial software packages. To achieve this, it is necessary to test a variety of different open-source software that can be used to create such Voxet files and visualize them. If there are differences in the data format that the open-source software reads and writes, advangeo 3D Prediction needs to be extended to read and write other data formats. A review of existing open-source 3D geomodelling software yielded the following 5 software packages: 1. Mira Geoscience ANALYST 2. ParaView 3. GemPy 4. VisIt 5. Mayavi
