Evaluating predictor strategies for regression-based downscaling with a focus on glacierized mountain environments

Abstract

his study explores the potential of different predictor strategies for improving the performance of linear regression-based downscaling approaches. The investigated local-scale target variables are precipitation, air temperature, wind speed, relative humidity and global radiation, all at a daily time scale. Observations of these target variables are assessed from three sites in close proximity to mountain glaciers: (1) the Vernagtbach station in the European Alps, (2) the Artesonraju measuring site in the tropical South American Andes, and (3) the Mount Brewster measuring site in the Southern Alps of New Zealand. The large-scale data set being evaluated is the ERA interim reanalysis data set. In the downscaling procedure, particular emphasis is put on developing efficient yet not over-fit models from the limited information in the temporally short (few-years) observational records of the high mountain sites. For direct (univariate) predictors, optimum scale analysis turns out to be a powerful means to improve the forecast skill without the need to increase the downscaling model complexity. Yet the traditional (multivariate) predictor sets show generally higher skill than the direct predictors, for all variables, sites and days of the year. Only in the case of large sampling uncertainty (identified here to particularly affect observed precipitation), the use of univariate predictor options is justified. Overall, we find a range in forecast skill between the different predictor options applied in the literature amounting up to 0.5 (where 0 indicates no skill, and 1 represents perfect skill). This highlights the importance of using sophisticated predictor selection in the downscaling process, particularly when focusing on glacierized mountains environments.