Using Google Earth Engine to map landslide hazard and exposure across Nepal

Nepal is one of the most susceptible countries to landsliding, with much of the country characterised by steep topography, annual monsoon rainfall and active tectonics. Current understanding of landslides in Nepal is predominantly based on static, catchment-scale landslide inventories or centred around data from specific events, such as the 2015 Gorkha earthquakes. Whilst static inventories provide a useful snapshot of past landslide characteristics, we cannot use these to infer how long landslides persist or how the hazard posed by landslides may evolve through time. In addition, the large number of small-scale inventories that currently exist cannot be readily compared, making it difficult to assess whether trends observed in specific catchments can be applied on a national scale. In this study, we aim to utilise advances in openly accessible remote sensing of large geospatial datasets, namely Google Earth Engine, to record the spatial and temporal evolution of landslides across the full extent of Nepal. We build on an existing automated landslide detection algorithm in Google Earth Engine to compile a national scale landslide probability map, which is re-mapped annually. This allows us to capture changes in landslide hazard both spatially and temporally across the country. The algorithm uses NDVI differencing to identify possible new landslides. Our work seeks to refine this output by using landslide-specific information obtained from a series of existing manually mapped landslide inventories. This step includes applying spectral and object-based filters as well as using susceptibility metrics, such as topography, trained using manually mapped landslide inventories. By adding a landslide-specific filtering step, we aim to build on existing NDVI differencing approaches and improve per pixel landslide probability values. We present preliminary findings using this record of landslide hazard through time to better understand controls on slope failure evolution and persistence, to tackle questions such as whether new landslides evolve and runout from existing landslides, to consider how landslide mechanism changes through time, and how hazard translates into physical exposure through the use of metrics such as landslide proximity to roads and buildings.