Computer assisted avalanche prediction using electronic weather sensor data

doi:10.1016/j.coldregions.2009.07.006

2009-11-01

PRE

DOI：10.1016/j.coldregions.2009.07.006

原文链接

原文求助

摘要

En 中文

A nearest neighbour model was used to predict avalanches in two highway corridors in British Columbia, Canada. The model accepts hourly electronic weather sensor data dynamically to produce automated predictions of the probability that avalanches will occur in the next 12 h. Output includes a list of nearest neighbours calculated by a Euclidian metric which provides information on patterns of avalanche activity in similar situations in the past. New variables are automatically generated from the hourly interval sensor measurements, including information about accumulated precipitation and maximum and minimum temperatures. A jackknife cross-validation routine generates fitness statistics by selecting test cases that are not temporally autocorrelated. The avalanche prediction system described here was applied operationally in Kootenay Pass, near Salmo, BC, and also at Bear Pass, near Stewart, BC, where accuracies of 76 and 72% were achieved respectively. (C) 2009 Elsevier B.V. All rights reserved.

Keyword:

Avalanche

Prediction

Electronic weather sensor

Nearest neighbour

Cross-validation