Verification of Probabilistic Downscaling

We have a method for downscaling daily temperature and precipitation variables that produces daily-varying probability density functions (PDFs). The advantages of probabilistic downscaling, which include maintaining information on the variance as well as the mean, tuning the autocorrelation of realizations to match that of the observations, and providing more detailed information on the extremes will be discussed. Furthermore, the output from probabilistic downscaling is flexible, so that it may be tailored to the needs of the user.

Full verification is essential before the application of any downscaling model, particularly with a multifaceted approach. Climate downscaling is used in numerous applications, so many diverse verification methods should be utilized. We apply and adapt a multitude of forecast and distribution verification techniques to test the skill and representativeness of the daily-varying downscaled PDFs. Methods are used to assess the daily distributions, as well as the characteristics of randomly-sampled realizations. Examples of the best verification methods to use for specific applications of the data will be provided. Our downscaling is shown to be skillful across a variety of metrics for many aspects of the distribution, including the representation of extreme events. We also show that observed climatological statistics are accurately reproduced in the downscaled realizations.