Item: Operational decision tree avalanche forecasting
Title: Operational decision tree avalanche forecasting
Proceedings: 2002 International Snow Science Workshop, Penticton, British Columbia
Authors: Walter Rosenthal, Donald Bren, School of Environmental Science and Management, University of California, Santa Barbara, Kelly Elder, Rocky Mountain Research Station, USDA Forest Service Fort Collins, Colorado Robert, E. Davis
Abstract: Decision tree models ofmaximum avalanche size class run daily at Mammoth Mountain, California. A classification tree grown on an eight-year subset of all weather and avalanche records shows an absolute accuracy on avalanche control days of from about 60-70% in a given year; accepting overestimates increases this to 70-80%. Errors arise from the rarity of large events, exclusion of the smallest most frequent events, and tree sensitivity to small changes in key predictor variables. A complete 19-year data set yields a pair of decision trees forecasting both maximum size class and maximum crown size over the entire mountain. Tested against a twentieth year, the size class tree may be more accurate for extreme events but performed slightly worse overall than the original tree. Coupling the size class and crown trees identified both class 5 avalanches during the test year. A third set of trees, driven by hourly data from a remote instrument network, distributes maximum class and crown sizes over geographic sub-regions of the mountain. These are striking for both small size and low misclassification rate. If a major source of error is chaotic avalanche behavior, decision trees may prove most valuable for providing probability estimates from given sets of initial conditions.
Language of Article:
Keywords: avalanche, decision trees, avalanche forecasting
Digital Abstract Not Available