Through the use of a univariate and classification tree analysis we have identified the key meteorological variables responsible for significant avalanche activity on the Seward Highway in Alaska.
Using 28 years of snowpack, weather and avalanche data, including more than 4500 individual avalanche events we trained and developed classifi cations trees using remotely measured meteorological variables as predictors of signifi cant avalnche activity.
We generated a tree with equal misclassifi cation costs that used only three parameters; the sum
of 72 h of water, the 24 h high temperature, and the 72 h average high temperature, to obtain a POD of 0.77 within the training data set.
An alternative tree with unequal misclassifi cation costs has a higher POD, but at the cost of an unacceptably high FAR for operational considerations