Estimating geothermal heat flow (GHF) is critical for understanding Earth's thermal regime and facilitating geothermal exploration. Machine learning techniques offer a promising approach for deriving GHF estimates based on various geological and geophysical features. While various studies have employed different data preprocessing methods for global GHF datasets, the influence of these methods on GHF predictions has not been thoroughly investigated. In this study, we assess the impact of four distinct preprocessing techniques applied to GHF datasets on GHF prediction accuracy within China using the Gradient Boosted Regression Trees (GBRT) algorithm. The results showed that more features are needed for the model performance to converge after the dataset is processed on average. Moreover, averaging GHF data considerably influences the model's assessment of feature importance, resulting in lower GHF estimations and a reduced prediction range. We also find that the prediction accuracy of GBRT models on the test set can be substantially improved by applying a low-pass filter to the GHF data. Upon further comparison, the model derived from the low-pass filtered dataset exhibits the highest predictive capability for GHF in China. Lastly, we present eight new GHF maps for China, which consistently indicate high GHF values in Tibet and eastern China, and low GHF values in northwest China. Our models also identify several unobserved high-heat regions, potentially offering valuable insights for geothermal exploration.

Press the Back button in your browser, or search again.

Copyright 2024, Stanford Geothermal Program: Readers who download papers from this site should honor the copyright of the original authors and may not copy or distribute the work further without the permission of the original publisher.