The effects of weather classification on regression-based downscaling of daily temperature extrema in the United States

Abstract

The focus of this dissertation was on the role played by weather classification in regression-based downscaling of daily temperature extrema. Three closely related studies were conducted, each using a different criterion for weather classification. The primary objective of all these studies was to evaluate changes in downscaling model performance as meteorological properties of the training periods were varied. This objective was of interest due to potential improvements in downscaling performance when accounting for non-static relationships between predictors and predictands. The first study used the time of day of the temperature extremum as the weather classification, while the third study used the direction of the wind as the weather classification. The second study used temperature as the weather classification, with a focus on possible consequences for downscaling in warmer conditions that were not present in the training conditions. Results from all three studies indicated that downscaling performance had the potential to be affected by the weather conditions seen in the training periods.