Developing a hybrid deep learning model with explainable artificial intelligence (XAI) for enhanced landslide susceptibility modeling and management

Landslides in the Nainital district of Uttarakhand, India, pose a significant threat to human communities and local ecosystems. This study aims to improve landslide susceptibility modeling by integrating advanced analytical techniques with deep learning, sensitivity analysis and explainable artificial intelligence (XAI). Our approach captures the complex interaction between natural terrain and human intervention and provides a novel framework for risk assessment and management. In this analysis, we performed a multicollinearity analysis to ensure the independence of predictor variables. We optimized deep learning models, including deep neural network (DNN), convolutional neural network (CNN) and a hybrid of CNN with long short-term memory (LSTM), using Bayesian techniques. This optimization achieved a high degree of precision in parameter tuning. In the study, multicollinearity analysis showed that no parameter exceeded the multicollinearity threshold of over 9. When evaluating accuracy, the CNN-LSTM model was found to be the most effective with an Area Under the Curve (AUC) of 0.96, while DNN and CNN also had high AUCs of 0.94 and 0.95, respectively. Spatially, the CNN model identified 16.28% of the total area as highly susceptible, while the hybrid CNN-LSTM model delineated 13.39%. Sobol's sensitivity analysis emphasized critical factors such as slope, elevation and geology as well as the anthropogenic influence of distance to built-up (DTB). The SHAP analysis confirmed the importance of these factors. This integrated method offers an innovative way to understand the dynamics of landslides by combining natural and human factors and provides the basis for sustainable infrastructure planning in Nainital.