A Physics-Informed and Transferable Machine Learning Workflow for Basin-Scale Heat Flow Prediction

This dataset contains the input data and source code required to reproduce the results presented in the research article "A Physics-Informed and Transferable Machine Learning Workflow for Basin-Scale Heat Flow Prediction".
The upload is organized into two main components:
Dataset: A compiled dataset of 16 multi-source geophysical and geological parameters used as features for predicting heat flow in the Junggar and Tarim Basins. These parameters include crustal thickness, Moho depth, lithospheric thickness, sediment thickness, gravity and magnetic anomalies, and distances to various tectonic features (e.g., ridges, trenches, volcanoes). The dataset also includes the measured heat flow values used as the prediction target. The data is provided in a structured format (e.g., CSV) suitable for direct use in the provided code.Code: The Python source code implementing the proposed physics-informed machine learning workflow. This includes:Scripts for data preprocessing and feature engineering.Implementation of the individual base learners: Ordinary Linear Regression (OLR), Gradient Boosting Regression Trees (GBRT), Light Gradient Boosting Machine (LGBM), K-Nearest Neighbors (KNN), and Support Vector Regression (SVR).The core stacking ensemble model with a Ridge Regression meta-learner, trained using nested cross-validation.The interpretability module, featuring SHAP (SHapley Additive exPlanations) analysis for quantifying feature contributions and spatial autocorrelation analysis.Scripts for generating prediction maps and reproducing the key figures from the manuscript, including the accuracy metrics comparison and the hierarchical thermal control analysis across different tectonic units.