Prediction and interpretation of high-temperature heat damage risk zones in the Eastern Tibetan Transport Corridor based on a location-information-fusion convolution neural network

High-temperature heat damage (HTHD) is a common geological hazard in underground engineering, and it poses a huge threat to the lives of the construction personnels and the safety of the project. The Eastern Tibetan Transport Corridor is an HTHD prone area. To address this hazard, an HTHD spatial database was built based on multi-source data such as Sustainable Development Goals Satellite-1 (SDGSAT-1) image. The spatial location relationship between geothermal samples, which is often overlooked in traditional HTHD risk zone prediction, was explored. An interpretable location-information-fusion convolutional neural network (LI-CNN) model was proposed to accurately predict and classify HTHD risk zones. The proposed LI-CNN model achieved a consistently superior performance across all evaluation metrics, with an AUC value of 0.81. The model significantly outperformed five traditional models, including logistic regression (AUC: 0.72), decision tree (AUC: 0.72), k-nearest neighbour (AUC: 0.72), extreme gradient boosting (AUC: 0.77), and convolutional neural network (AUC: 0.77). Finally, two deep learning interpretation models were combined to conduct both global and local interpretation. The results revealed that fault density, land surface temperature, and river density exerted the greatest influence on HTHD occurrence. Overall, our work contributes to the development of sustainable and disaster-resistant infrastructure in accordance with Sustainable Development Goal 9.1 to support economic development and improve human well-being.