We present a methodology for statistically downscaling daily wind speed variations on a local scale in an extended western Great Lakes region. The methodology is based on predicting the daily-varying probability distribution of local-scale wind speed, conditioned on the debiased estimate of the daily-varying large-scale wind from a climate model. The statistical model is trained for each month separately, using the NCEP/NCAR reanalysis as a large-scale predictor and local wind variations at 48 National Weather Service first-order stations as the predictand. The model is then applied to (debiased) output from the CMIP3 global climate models, resulting in projections of daily wind speed under present and future climate forcing.
Details of the statistical model will be discussed. The model is similar to a generalized linear model using the gamma distribution, but it allows the shape and scale parameters to change with the large-scale wind speed. In fact, it is the local-scale mean and standard deviation that are defined as functions of the large-scale wind via a link function, since for the gamma distribution they are related to the shape and scale parameters via simple products. This link function was chosen to ensure that the mean and standard deviation are always positive and exhibit a mostly linear relationship with the large-scale wind predictors. By downscaling to the distribution parameters, a probability density function (PDF) is created. Numerous, equally-likely realizations can be extracted, but producing a PDF makes the data more flexible and thus more useful to a variety of applications.