Strategies for Statistical‐Dynamical Downscaling to Urban Climate Using Global Data

ABSTRACT Statistical‐dynamical downscaling is one method to model urban climate using global and regional climate model (GCM and RCM) results. In this study, different strategies for statistical‐dynamical downscaling are derived and evaluated. For the statistical part of downscaling, a Bivariate Skill Score is developed to quantify the overlap of the joint probability densities of two meteorological variables, which helps to quantify whether the statistically selected days are representative of the full climatology. Results show that for representing the winter climate of the selected urban area (Hamburg, Germany), more days need to be simulated than for summer climate (129 days vs. 40 days simulated). For the dynamical part of the downscaling (from ~20 km to 250 m), the mesoscale atmospheric model METRAS is used with two different grid structures: (a) downscaling with three one‐way nested domains and (b) downscaling with one domain employing a non‐uniform grid. The downscaling method with the non‐uniform grid is less expensive in preparing and computing resources than the method with tree one‐way nests. The evaluation shows that similar model results are achieved in both cases in the domain of interest. The evaluation of temperature, relative humidity, wind speed and wind direction with different metrics shows that METRAS performs well for summer and winter climate, and slightly better for summer. Furthermore, METRAS performs well with both downscaling methods and the evaluation measures are slightly better for the downscaling with the non‐uniform grid. For further downscaling of GCM or RCM results to a very local scale, for example, with obstacle resolving models, using a non‐uniform grid may be a good solution to reduce the number of necessary downscaling steps and the related work load for computer and humans.