Stanley's ROSES proposal selected for funding

"Landslides are an influential hazard in Nepal, resulting in both extensive loss of life and diverse indirect impacts on human well-being. Much research has already been undertaken on this subject, including numerous case studies, susceptibility maps, and rainfall analyses. However, no forecasting system exists for the whole country to enable dynamic characterization of increased landslide hazard potential. This type of information is critical for improved awareness of cascading hazards and anticipation of impacts they cause on natural and human systems. Recent progress in research geared towards predicting landslides in the Karnali Basin of far-western Nepal and the Lower Mekong River Basin as demonstrated the potential for broader implementation.

We will assist stakeholders in Nepal with the development of a decision support system that assesses landslide hazard with a lead time of 1-2 days. This system will build upon work that has been completed for the SERVIR-HKH [Hindu Kush Himalaya] node. The High-Impact Weather Assessment Toolkit (HIWAT) provides an ensemble of precipitation forecasts that have been converted to a probability-matched mean value for each location within a service area that includes Bangladesh and Nepal. With machine learning, these maps of precipitation will be combined with other predictors to produce a gridded probability of landslide occurrence with a daily 4-km resolution. We have already developed a prototype system using HIWAT and an existing landslide model for the Karnali Basin. This work will serve as the basis for broadening this system to a national scale, which would otherwise have been difficult to complete within the 2-year period of performance.

We will also produce a tool for rapid mapping of landslides from optical satellite imagery. Although previous research by our team has used very high-resolution imagery from commercial satellites, this tool will primarily rely on the Sentinel-2 mission as it provides data that is freely available to all stakeholders in Nepal. The landslide mapping tool will rely on deep learning (artificial neural networks) to segment the satellite images into geospatially referenced polygons that indicate the presence of recent landslides. In order to minimize the computational burden on stakeholders, the tool will rely on existing cloud-based platforms provided by NASA or commercial technology companies. The tool is intended to enable rapid assessment of damage from landslides in the days before officials can reach an affected area by ground transportation.

Finally, we will build the capacity of regional stakeholders through a series of annual workshops in Kathmandu, Nepal. Each workshop will take place after the end of the annual monsoon in order to enable attendance by stakeholders that have disaster management responsibilities. In the first workshop, prototypes of both the landslide forecast and the landslide mapping tool will be introduced. Extensive stakeholder consultation will be sought on the scientific approach, outputs, and visualization capabilities of each system.

The proposed work represents a significant need and demand by groups within the region and will impact decisions and actions by stakeholders in Nepal. It builds upon work already completed as part of the High Mountain Asia and related projects, including the mapping and modeling of landslides in Nepal’s Karnali Basin."