Demonstration of self-updating landslide hazard maps with dynamic crowd-sourced data in Rwanda

Mitigating landslide impacts in regions experiencing rapid land cover conversion requires a good understanding of these hazards. Inventories of previous landslide events are necessary starting points but are often lacking or inadequate, especially unpopulated areas. Growing satellite data availability and lower barriers-to-entry in cloud computing and machine learning offer opportunities to augment inventories and rapidly re-assess hazard conditions. We demonstrate self-updating landslide hazard assessments of Rwanda, informed by a continually growing catalog of landslide events derived via visual image interpretation. We use multiple techniques including logistic regression and random forest with a set of ~20 potential explanatory variables in Google Earth Engine to quickly quantify the influence of expert crowd-sourced landslide data points on the accuracy of landslide hazard mapping. Coupled with socio-economic information describing populations' vulnerability to landslides, this approach can potentially update depictions of landslide risk at a finer temporal scale and aid in land use planning when used with other decision support systems. This work responds to the Programme of Action for the Implementation of the Sendai Framework for Disaster Risk Reduction 2015-2030 in Africa. It specifically focuses on inventorying and mapping of different approaches and methods used for risk assessment & analysis; and developing and publicizing awareness products, including risk mapping products, in communities. Further, such knowledge helps prioritize high risk zones and contributes to strategies for greenhouse gas emissions reduction and carbon capture, also ultimately reducing the impact of landslide disasters.