Statistical wind forecasting for Arctic locations using recursive partitioning and regression trees

Weather forecasting for northern Canada during the winter is a challenge. Weather at a given location is highly dependent on wind – speeds will determine whether or not snow will blow around, giving blizzard conditions; direction will determine whether moisture from offshore areas will affect the area and whether winds will blow from a favoured direction for enhancement.

A statistical technique known as Recursive Partitioning and Regression Trees was used in this study. Data from 2005 and 2007 were used to determine which Canadian GEM model predictors should be used for a given location, and to create the regression trees - the resultant trees define sets of rules that act on the predictors such that cases of the predictands are placed into homogenous groups.

Verification of these forecasts for the period from January to April as well as September and October 2008 will be presented. Forecasts are provided in near-real time to the Prairie and Arctic Storm Prediction Centre as well as the Canadian Meteorological Aviation Centre - West.