Discovering new sweet spots with the geothermal map renovated by machine learning in Henan Province, China

Geothermal heat flow (GHF) is crucial for the exploration and exploitation of geothermal resources. Given the technology challenges and high costs, GHF measurements are scarce in many regions. This study hybridizes a genetic algorithm (GA) with a Gradient Boosting Regression Tree (GBRT) model based on different scale datasets to predict GHF distribution in Henan Province, China, integrating 17 geological and geophysical features (GGFs). The model's performance improved with the inclusion of neighboring regional data but declined with national data, underscoring the significance of geographic data homogeneity. Surprisingly, expanding to a global dataset significantly improves prediction, achieving an R2 of 92.25 % on the test dataset. To our best knowledge, this is the highest value of R2 for GHF predictions reported in the current literature. This suggests that diverse geological features and heat flow patterns in the global dataset are crucial for learning geological patterns associated with GHF. The GA-GBRT model successfully predicted GHF trends in Henan Province even without local measured GHF data, validating its effectiveness. The renovated GHF map of Henan Province identified five new geothermal sweet spots that have not been discovered previously. This is of great significance for guiding future geothermal exploration efforts in the region.