Avalanche activity in an extreme maritime climate: The application of classification trees for forecasting

Classification trees were trained to determine periods of significant avalanche activity in terms of an avalanche day, based on meteorological parameters for the Milford Road, Fiordland, New Zealand. Using a 10 year data set of meteorological parameters and over 1800 individual avalanche occurrences from the Transit New Zealand Milford Road Avalanche Programme we have described and statistically explored this avalanche regime and the relationship between storm and avalanche activity in this extreme climatic region. lford Road is located in the south western corner of New Zealand and is dominated by a maritime climate delivering in excess of 8 m water equivalent per year in precipitation, while winter storms can deposit up to 2 m of snow in one storm. As the avalanche climate is dominated by direct action avalanching, the meteorological parameters up to a maximum of 72 h preceding a significant avalanche period were examined. Standard meteorological parameters including air temperature, air pressure, wind speed and direction, snow depth and precipitation were obtained from two automatic weather stations located in the starting zone and at road level. These parameters as well as two derived wind drift parameters were used as the variables for predicting the avalanche days. fold cross validated classification trees were created, and suggested for use in forecasting. The classification tree with highest accuracy of 85% predicted avalanche days less well at 79%. An alternative tree using only wind speed and wind speed and precipitation combined in a temperature sensitive wind drift parameter resulted in a lower overall accuracy of 78%, but permitted a higher rate of correct prediction for avalanche days at 86%. The alternative, more conservative tree also reduced the number of false negative cases (observed as avalanche days, but predicted as non-avalanche days) from 31 to 20 at a cost of increasing the false positive or false alarm rate.