Open Collections

Applied automated numerical avalanche forecasting using electronic weather sensor data

University of British Columbia

Numerical avalanche prediction was used for Canadian highways avalanche forecasting for ten years before changes in information technology infrastructure rendered the original numerical avalanche forecasting model incompatible and therefore obsolete. Now these efforts are being renewed with greater automation by the use of electronic weather sensor data. Use of this data presents several challenges and opportunities. Automated hourly observations generate large datasets that require systems for filtering historic and current data; as well as fitness testing routines that dynamically extract independent validation samples from serially correlated datasets. These weather sensor data manipulation systems offer several advantages over traditional avalanche prediction models that are based on manually observed weather and surface snow information. Rapid dataset generation enables spatial scaling of predictions, easy generation and testing of memory variables, model comparison, and visual verification of predicted avalanche probability time series. These features will facilitate operational implementation of avalanche forecasting models for applied computer assisted avalanche forecasting-in highways avalanche control programs across British Columbia, Canada. In the winter of 2006/7, the Avalanche Forecast System (AFS) was applied in two avalanche prone transportation corridors. The AFS uses only electronic weather sensor data and incorporates all of the aforementioned capabilities. A nearest neighbour analysis is used to generate avalanche probabilities, however the AFS data management systems could also be made to operate with classical linear and modern non-linear statistical prediction methods. Automated filters eliminate erroneous data dynamically, permit investigation of various prediction targets (such as natural avalanche occurrences, or avalanches of different size classes), and a jackknife cross-validation routine generates fitness statistics by selecting test cases that are not temporally autocorrelated. The AFS was applied operationally in Kootenay Pass, near Salmo, BC, and also at Bear Pass, near Stewart, BC, where accuracy of 76% +/-2% and 71% +/-2% were achieved respectively.

Text

2011-03-09

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/32241

Geography

University of British Columbia

Graduate