Theoretical and Applied Climatology

, Volume 91, Issue 1, pp 129-147

First online:

Statistical downscaling of hourly and daily climate scenarios for various meteorological variables in South-central Canada

  • C. S. ChengAffiliated withMeteorological Service of Canada (MSC) Branch-Ontario, Environment Canada
  • , G. LiAffiliated withMeteorological Service of Canada (MSC) Branch-Ontario, Environment Canada
  • , Q. LiAffiliated withMeteorological Service of Canada (MSC) Branch-Ontario, Environment Canada
  • , H. AuldAffiliated withAdaptation and Impacts Research Division, MSC Branch, Environment Canada

Rent the article at a discount

Rent now

* Final gross prices may vary according to local VAT.

Get Access


A regression-based methodology was used to downscale hourly and daily station-scale meteorological variables from outputs of large-scale general circulation models (GCMs). Meteorological variables include air temperature, dew point, and west–east and south–north wind velocities at the surface and three upper atmospheric levels (925, 850, and 500 hPa), as well as mean sea-level air pressure and total cloud cover. Different regression methods were used to construct downscaling transfer functions for different weather variables. Multiple stepwise regression analysis was used for all weather variables, except total cloud cover. Cumulative logit regression was employed for analysis of cloud cover, since cloud cover is an ordered categorical data format. For both regression procedures, to avoid multicollinearity between explanatory variables, principal components analysis was used to convert inter-correlated weather variables into uncorrelated principal components that were used as predictors. The results demonstrated that the downscaling method was able to capture the relationship between the premises and the response; for example, most hourly downscaling transfer functions could explain over 95% of the total variance for several variables (e.g. surface air temperature, dew point, and air pressure). Downscaling transfer functions were validated using a cross-validation scheme, and it was concluded that the functions for all weather variables used in the study are reliable. Performance of the downscaling method was also evaluated by comparing data distributions and extreme weather characteristics of downscaled GCM historical runs and observations during the period 1961–2000. The results showed that data distributions of downscaled GCM historical runs for all weather variables are significantly similar to those of observations. In addition, extreme characteristics of the downscaled meteorological variables (e.g. temperature, dew point, air pressure, and total cloud cover) were examined.