A comprehensive landslide early warning system for Rwanda: integrating predictive modeling, environmental parameters, and community engagement

This study aimed to develop a machine learning-based Early Warning System (EWS) tailored to Rwanda's geoclimatic conditions, identifying critical landslide-inducing factors and operationalizing a predictive model to minimize disaster impacts. Rwanda's mountainous terrain, bimodal rainfall patterns exceeding 2,000 mm annually, and accelerating land-use changes make it highly susceptible to rainfall-induced landslides, resulting in recurrent human fatalities, infrastructure losses, and environmental degradation. Current landslide monitoring systems in Rwanda are primarily reactive, lacking real-time forecasting and predictive accuracy.

Results revealed a strong correlation between rainfall and landslide occurrences, particularly when daily precipitation exceeded 100 mm, with 55% of all events occurring during high-intensity rainfall periods. Clayey-silt soils accounted for over 60% of landslides due to poor drainage and rapid saturation, while steep slopes above 35°, combined with fractured volcanic formations, were present in 60% of failure sites. Elevations above 2,000 meters and north-facing slopes experienced 20% more landslides due to persistent moisture retention. The developed EWS achieved an 85% prediction accuracy, providing automated, location-specific alerts via a web interface that allows for early evacuation and targeted responses. The study concludes that integrating machine learning with geospatial data significantly improves landslide prediction and risk mitigation. It recommends that the government of Rwanda and partners prioritize the expansion of sensor networks, institutionalize the EWS into national disaster frameworks, scale geotechnical mapping, and intensify community-based education to ensure long-term system sustainability and regional replication.