Beyond Data Scarcity: A Physics-Integrated Landslide Prediction for Urban Disaster Resilience
2026 IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events (PerCom Workshops), URBSENSE 2026, pp. 1094–1099
Abstract
Global climate change has increased the frequency and intensity of extreme rainfall events, causing landslides even in regions with little or no historical disaster records. Accurately predicting such events in urban environments requires integrating heterogeneous environmental data within a system-level co-design framework. However, two key challenges hinder reliable prediction: (i) limited observability due to scarce disaster records and the difficulty of measuring essential physical variables such as subsurface hydrology, and (ii) strong regional heterogeneity, where disaster data exhibit inherently non-IID characteristics across regions. Traditional physics-based models often rely on simplified assumptions because observable variables are limited, whereas deep learning models tend to overfit under scarce and imbalanced data. To address these issues, this study proposes a physics-integrated deep learning system that combines a three-layer tank model as a physics-based model with a data-driven deep learning model. The physics-based outputs are incorporated both as input features and through a skip-connection that adaptively controls their contribution under abnormal rainfall conditions. In addition, Stochastic Feature Augmentation is employed to improve robustness against regional data distribution shifts. Experiments conducted across 19 regions in Japan demonstrate that the proposed system achieves improved generalization and superior PR-AUC in even unseen regions compared with state-of-the-art methods.
Climate change has increased the frequency and intensity of extreme rainfall, causing landslides even in regions without historical disaster records. Accurate prediction of such events in urban environments faces two fundamental obstacles: (i) limited observability — many physical variables that directly govern slope stability, such as subsurface water content and geotechnical properties, cannot be measured at scale; and (ii) strong regional heterogeneity, where disaster data are inherently non-IID across regions because of differences in rainfall, terrain, vegetation, and urban structure.
We propose a physics-integrated deep learning system that addresses both challenges. A three-layer tank model, incorporated as a physics-based module, supplies intermediate hydrological state as input features and is coupled to the learned model through a skip connection that adaptively controls its contribution under abnormal rainfall. In addition, Stochastic Feature Augmentation is applied during training to improve robustness against regional distribution shifts.
Experiments across 19 regions in Japan show that the proposed system achieves superior PR-AUC and generalization, even on previously unseen regions, compared with state-of-the-art baselines. The results demonstrate the effectiveness of co-designing physics knowledge with deep learning for disaster prediction tasks where data are inherently scarce and imbalanced.
We are extending this direction toward urban disaster resilience, including distributed learning and tighter integration with meteorological and geological observation infrastructure.