Integrated deep learning with explainable artificial intelligence for enhanced landslide management

Landslides pose significant threats to mountainous regions, causing widespread damage to both property and human lives. This study seeks to enhance landslide prediction in the Aqabat Al-Sulbat Asir region of Saudi Arabia by integrating deep neural networks (DNNs), 1D convolutional neural networks (CNNs), and a combined DNN and CNN ensemble (DCN) with explainable artificial intelligence (XAI) techniques. These XAI techniques enhance the interpretability of these complex deep learning models, thereby facilitating better decision-making strategies. Furthermore, the DNN model is employed to incorporate game theory principles, assessing the individual impact of variables on landslide prediction. Our findings indicate high and very high landslide susceptibility zones covering 35.1-41.32 km2 and 15.14-16.2 km2, respectively. The DCN model boasts the highest area under the curve (AUC) at 0.97, followed by CNN (0.94) and DNN (0.9), showcasing DCN's superiority. XAI analysis exposes significant residuals in CNN's posterior despite its high AUC. Notably, precipitation, slope, soil texture, and line density emerge as pivotal parameters for accurate landslide prediction. Game theory results highlight line density's preeminence, trailed by topographic wetness index, curvature, and slope in landslide occurrence. By incorporating deep learning models, XAI, and game theory, this study presents a holistic approach to landslide management. This comprehensive framework equips authorities and stakeholders with valuable tools for informed decision-making in landslide-prone areas, delivering accurate predictions and insights into crucial parameters.