Approaches to Modeling Geothermal Heat Flow from Various Datasets

Geothermal heat flow models are currently developed for remote areas like Antarctica and parts of Africa. Due to the sparsity of actual geothermal heat flow measurements, indirect data based on various quantities related, e.g., to gravitational and magnetic information are used for predicting heat flow in such regions and quantifying its uncertainty. Here we present and compare two common approaches, a data driven random forest approach that is trained with several covariates (magnetic and gravitational anomalies, lithospheric thickness, topography, seismic velocities) and a physics based approach relating magnetic anomalies to Curie depth and subsequently to geothermal heat flow (requiring various simplifications and a priori assumptions on the underlying physics).