Forecasting snow avalanche days from meteorological data using classification trees, Grasdalen, Western Norway.

Abstract

Avalanches pose one of the most serious problems to infrastructure and people in the mountains in Norway. Processes leading to avalanche release are deterministic but the time and place of avalanche release is notoriously difficult to predict. Statistical approaches using meteorological parameters to predict the probability of natural avalanche release provide an alternative to deterministic prediction. We used classification trees to predict days with and without avalanches in the valley of Grasdalen in Western Norway based on meteorological parameters. A database with avalanche observations from almost 30 years was spatially and temporally coupled to grids of wind, precipitation and temperature. The grids were used because they provided more temporally consistent datasets than measurements from a local weather station. Avalanches were observed on 254 days and the same number of non-avalanche days was randomly selected. The optimal classification trees gave misclassification rates of 15% for all avalanche days, 18% for days with dry avalanches and 13% for days with wet avalanches. The most important meteorological parameters for the classification were the five-, one- and three-day sum of precipitation. Then followed wind speed, either measured as the maximum or mean over five days, three days or one day. Finally, daily temperature was important for the classification both alone and through a degree day parameter. Based on realistic scenarios for precipitation and temperature, our results imply that avalanche frequency will increase in the future. Further studies are needed to quantify this increase.