Advanced Geothermal prospectivity mapping with Machine Learning using Multi-model Regression and Geology-dependent Multicriteria-Decision Analysis

Geothermal prospectivity mapping using multiple-criteria decision analysis is well-documented in the scientific literature. However, most previous works suffer from two drawbacks: no consideration of the geology for defining the weights of the decision criteria, and no metrics of reliability for the prospectivity prediction. In this work, we introduce a method that overcomes these drawbacks and extends on a large scale the mapping provided manually by the experts. The method is based on (i) automatic regressions of both heat flow and thermal gradient maps using a multi-model approach, (ii) introduction of geology-dependent criteria weights for the decision analysis, and (iii) combination of multiple decision maps to obtain a more robust prospectivity mapping and a quantification of the reliability of the prediction. Applied to geothermal data from British Columbia, the method appears to produce prospectivity maps with higher spatial resolution than conventional ones. Besides, it substantially attenuates the footprint of data acquisition and facilitates the identification of geographical zones where the prediction is the most uncertain, thus where a human subjective interpretation would have the most added value. The proposed workflow is easily transposable to any context of subsurface exploration or exploitation, for instance in the petroleum, mining, or hydrogeological industries.