Frozen soil thermal conductivity prediction and its spatial distribution characteristics along Qinghai-Tibet Engineering Corridor (QTEC) – a geospatial machine learning framework

Permafrost degradation within the Qinghai-Tibet Plateau (QTP) is accelerating, posing a severe risk to the thermal stability of engineering infrastructure. Therefore, accurate estimation of soil thermal conductivity (λ) is essential for predicting ground temperature evolution and to prevent freeze–thaw-related failures. In this study, we developed a unified geospatial machine learning (ML) framework utilizing six models, extreme gradient boosting (XGBoost), random forest (RF), artificial neural network (ANN), support vector regressor (SVR), logistic regression (LR), and k-nearest neighbor (KNN) to predict frozen soil thermal conductivity (FSTC) along the Qinghai-Tibet Engineering Corridor (QTEC). In total, 639 FSTC observations and eighteen predictor variables (including in-situ, experimental, and satellite remote sensing (RS) point data) were used to develop ML models. Pearson correlation (PC) and gray relational analysis (GRA) were employed to assess the influence of predictors on FSTC. The performance of six ML algorithms was evaluated using five metrics, R2, R, RMSE, MSE, and MAE, to determine the optimal prediction model. The ensemble models, RF (R2 = 0.915) and XGBoost (R2 = 0.900), achieved the highest prediction accuracy. These two models effectively captured the complex non-linear interactions among soil, environmental, and topographic variables. Shapley Additive exPlanations (SHAP) analysis identified water content, bulk density, ground temperature, and precipitation as dominant predictors. Spatial prediction maps generated by the optimal models exhibited a south-to-north increasing pattern of FSTC. Its values range from <1.0 W/(m·K) in the south to >3.0 W/(m·K) in the northern basins. This multi-source RS and ML approach enables scalable, high-resolution mapping of frozen soil thermal properties, providing critical insights for permafrost infrastructure resilience under climate change.