Snow Avalanche Frequency Estimation (SAFE): 32 years of remote hazard monitoring in Afghanistan

Snow avalanches are one of the most predominant natural hazards in mountain areas. Every year throughout the world, they are the cause of much material destruction and loss of life. It is therefore essential for local communities and public authorities to assess areas most vulnerable to avalanches. Here, we propose a new method for automatic avalanche detection from Landsat archives, using a snow index. This open-source and user-friendly model in Google Engine is the first to automatically inventory all the avalanches that have occurred each year across wide catchment areas, over a period of 32 years. The Snow Avalanche Frequency Estimation (SAFE) model was tested in the mountains of Afghanistan - Amu Panj Basin - one of the most remote regions in the world and one of the poorest in terms of avalanche monitoring. SAFE correctly detected 76% of the actual avalanches identified on Google Earth images and in the field. Since 1990, this region of Afghanistan has been impacted by 810,000 avalanches with an average frequency of 0.88 avalanches/km²yr^-1. With SAFE, it is now possible to clearly identify villages, roads, and rivers that are frequently affected by avalanches and thus help decision-makers in their investments in avalanche protection infrastructure. It was also found that the frequency of avalanches has not changed over the last 32 years, but SAFE has identified a northeast shift of these hazards, notably due to a slight increase in temperatures in the south at the beginning of winter. SAFE is the first robust model that can be used worldwide and is capable of filling data voids on snow avalanche impacts in inaccessible regions.