AUTOMATED NUMERICAL PREDICTION USING ELECTRONIC METEOROLOGICAL AND MANUAL SNOWPACK DATA

Nearest neighbour algorithms using manual observation data can provide useful and accurate predictions of avalanche activity (McClung and Tweedy 1994, Floyer and McClung 2003, Roeger et al. 2003a, Zeidler and Jamieson 2004, Purves 2003). Here, a system is proposed that will use electronic data from automated weather stations in two distinctly different avalanche prone transportation corridors: Kootenay Pass and Bear Pass in British Columbia, Canada. The goal is to create a flexible, modular framework for numerical avalanche prediction using nearest neighbours that is automated, scalable, and that can be easily applied to different forecast operations. In addition to now-casts of avalanche probability, the program will provide advanced forecasts based on numerical or human meteorological forecasts (Roeger et al. 2003a). Furthermore, two methods of incorporating snowpack information into the avalanche predictions are outlined. The first is a simple threshold sum method similar to the one proposed by Schweizer and Jamieson (2003), and the second employs a data mining algorithm called MART (multiple additive regression trees). Probabilities generated by each algorithm will be combined using a Bayesian framework (McClung and Tweedy 1994).