Machine learning discovers South American subduction zone hotter than previously predicted

GHF (Geothermal heat flow) is crucial for characterizing the Earth’s thermal state. Compared to other regions worldwide, GHF measurements of South America are relatively sparse for mapping GHF over the continent based on traditional models. Here we apply the Machine Learning (ML) techniques to predict the GHF in South America. By comparing the global model, ML finds that South American subduction zones are hotter than the global model due to large-scale magmatism, which leads to the higher shallow arc temperatures than canonical thermomechanical and global models. In combination ML model with the local singularity analysis of heat flows, active volcanoes, and igneous rock samples, it suggests that geothermal anomalies along the Andean Mountain Range are spatially correlated with magmatic activity in the subduction zone. It is believed that the ML methods have reliable GHF prediction in regions like South America, where GHF measurements are limited and uneven.