Statistical Downscaling of Intensity and Frequency Predictions of Local and Regional Extreme Precipitation

The modeling of precipitation at local, regional and global scale remains uncertain, and for many regions, problematic. Further, the downscaling of precipitation values from gridded data to single point locations poses many challenges since precipitation in general is not a 'single variable,' but rather a composite response to a variety of atmospheric variables. Multiple linear regression using point gauge measurements (SMHI) as the dependent variable and three different sets of ECMWF reanalysis (model) output variables as independent variables for 5 locations in Sweden suggests that for statistical downscaling of modeled precipitation, three additional variables should be included in the calculation: percentage of high cloud cover, mean sea level pressure and top minus surface radiation. We investigate the suitability of usage of these variables in a downscaling procedure, develop a predictive model based on these findings, then extend the method to CMIP5 results.