Landslide inventories: Leveraging big earth data for natural hazard management

Challenge: Develop approaches for automated landslide mapping, a prerequisite for better prediction and management of landslide risks. 

Landslides are among the most disastrous natural hazards, and lead to both a significant loss of life and infrastructure damage with massive socio-economic impacts. Unplanned and incautious development may increase the risk of landslides by removing stabilizing vegetation and creating steep slopes more prone to landslides, but better management may minimize these risks. However, such management requires improved estimates of local landslide risk—which will be of great interest in the infrastructure and risk sectors as well as environmental planning. 

Progress in earth-observation makes global high-resolution data of landslides and potential landslide-triggering factors available in near real-time. Currently, the absence of landslide inventories limits the ability to leverage such data for training and validating novel, high-resolution, data-driven landslide models to provide improved estimates.

This project would address the lack of such landslide inventories by building automated approaches for identifying potential candidate sites and intuitive interfaces for their initial, expert based validation. It requires bringing together some of the latest earth observation data from Planet Labs with state-of-the art computer vision approaches to prototype semi-automated landslide identification. Successful development of such a tool would help overcome some of the greatest challenges in natural hazard management and mitigation.