Machine learning-based prediction of terrestrial heat-flow across mainland China

Abstract

Due to the limitations of interpolation methods in areas with sparse data points and highly heterogeneous spatial distributions, this study employs the Gradient Boosted Regression Tree (GBRT) method to predict the terrestrial heat-flow distribution across mainland China. The model is trained using available heat-flow data, with regional geophysical and geological information incorporated as constraints. A map of heat-flow with a resolution of 1° × 1° is obtained. Low heat-flow values (mean heat-flow of 50.7–52.2 mW/m²) are observed in the Tarim Craton, the Western Central Asian Orogen, and the Yangtze Craton, whereas the Tibetan Plateau exhibits high heat-flow values (mean heat-flow of about 79.0 mW/m²) due to the collision of the Indian Plate with the Eurasian Plate. Located at the junction of the Pacific, Eurasian and Philippine Sea plates, the Cathaysia Block shows a complex geothermal pattern with a sporadic distribution of high heat-flow due to heat transport presumably dominated by the subduction and upwelling of mantle.