The Distribution of Surface Heat Flow on the Tibetan Plateau Revealed by Data-Driven Methods

We developed a spatially intelligent method named interpretable geographically neural network weighted regression (IGNNWR) for predicting SHF that considers the spatial heterogeneity and nonlinearity of the geophysical and geological factors affecting the geothermal regime. We applied the IGNNWR model to create an accurate SHF map of the Tibetan Plateau. The results revealed the existence of the zones where the SHF values are medium to high in the southern, northeastern, and southeastern parts of the Tibetan Plateau. By analyzing the explanatory power of the IGNNWR model, we identified Moho surface depth, distance to ridges, topography, and the average curvature of satellite gravity gradients as the essential geophysical and geological features that strongly influence the formation of the zones where the SHF values are high.