Geo-AI for landslide susceptibility in Rwanda: integrating LiDAR, soil moisture, and historical data

Abstract

Landslide susceptibility in Rwanda’s mountainous regions requires advanced predictive tools to address increasing risks driven by climate change and human activities. This systematic review synthesizes 36 studies published between 2017 and 2025 on Geo-AI frameworks integrating LiDAR-derived terrain metrics, soil moisture data and historical landslide inventories for susceptibility mapping. Findings show that deep learning models, especially convolutional neural networks (CNNs), outperform classical machine learning approaches, achieving Area Under the Curve (AUC) values of 0.93–0.96 compared to 0.85–0.90 for random forest and support vector machines. While classical models offer greater interpretability, they are less effective at capturing complex nonlinear relationships. Key challenges include data scarcity, model opacity and climate non-stationarity. The review proposes an integrative Geo-AI framework for Rwanda emphasizing multiscale data fusion, ensemble modeling and explainable AI to support early warning systems, land-use planning and disaster risk reduction in East Africa.