Best Practices in the Statistical Prediction of Time-Averaged Winds

Statistical downscaling methodologies based on multiple linear-regression that yield significant improvements to the predictability of: land surface winds (over the Canadian prairies and Ontario) and sea surface winds over the global ocean (Northeast Pacific, Northwest Atlantic and Pacific, tropical Pacific and Atlantic) are presented. We downscale mid-tropospheric predictors (wind components and speed, temperature, and geopotential height) from reanalysis products to predict historical wind observations at thirty-one airport-based weather surface stations and fifty-two moored ocean buoys. Three particular methodologies are assessed as functions of: the statistical features of the wind; the statistical averaging timescale of the wind statistics; and the wind regime (as defined by how variable the vector wind is relative to its mean amplitude).

These methodologies are informed by recent studies investigating the statistical predictability of surface winds and theoretically-derived relationships between the predictability of scalar and vector wind quantities. As these past studies found the predictability of mean quantities to be greatest on shorter averaging timescales, predictions of quantities over longer averaging timescales are constructed from predictions of daily means. Surface wind quantities such as the magnitude of the mean vector wind, which are non-linear functions of the mean vector winds, are constructed from predictions of the mean vector winds rather than predicted directly. Lastly, predictions of sub-monthly and sub-seasonal variability are improved by including predictive information contained in the predictions of the daily mean quantities.

Substantial improvements in the prediction skill for monthly-averaged wind component quantities are observed. A marked decrease of the anisotropy of predictability for monthly-averaged vector quantities is also observed, a result of strong improvements in the predictability of the poorest predicted components. Improvements in the predictability of statistical features of the wind speed are also observed, although these are less substantial for daily-averaged quantities. These results demonstrate the importance of constructing statistical predictions of monthly- and longer-timescale surface wind variability from daily predictions.