On using local avalanche danger level estimates for regional forecast verification

Operational verification of regional avalanche forecasts strongly relies on high quality field observations. In addition, specifically trained and experienced observers may provide local danger level estimates - a condensed, but subjective summary of current avalanche conditions. However, these estimates not only reflect local rather than regional conditions, but may also be influenced by, for example, the observers' personal experience and the ease of perceiving the hazard. We explored close to 10,000 local danger level estimates reported by more than 100 trained observers to the national forecasting service in Switzerland. Even at distances less than about 10 km, observers disagreed in their local estimate 22% of the time. Some observers had a bias towards consistently higher or lower local estimates. The hit rate when comparing local estimates (nowcasts) with the regional forecasts was 76%. It varied considerably between individual observers, but partly also among typical groups of observers (e.g. mountain guides, ski area staff or avalanche forecasters). Taking into account the uncertainty in local estimates and the reporting bias revealed a slightly lower agreement between local nowcast and regional forecast of 71%. These levels of agreement seem rather low, but are in line with previous studies. We conclude that local nowcasts can be used for forecast verification, but substantial uncertainty remains and the true avalanche danger level remains unknown.