Incremental Correction for the Dynamical Downscaling of Ensemble Mean Atmospheric Fields

Abstract

his research was motivated by the need for an improved method compared to the conventional brute-force approach to ensemble downscaling. That method simply applies dynamical downscaling to each ensemble member. It obtains a reliable forecast by taking the ensemble average of all the downscaled ensemble members. This approach, although straightforward, has a problem in that the computational cost is too large for an operational environment. Herein a method for downscaling ensemble mean forecasts is proposed. Although this method does not provide probabilistic forecasts, it will provide the regional-scale detail at minimum cost. In this product, all of the predicted parameters are dynamically and physically consistent (i.e., most likely to occur on a seasonal time scale). It is believed that such a product has great utility for regional climate forecast and application products. The method applies a correction to one of the global forecast members in such a way that the seasonal mean is equal to that of the ensemble mean, and it then downscales the corrected global forecast. This method was tested for a 140-yr period by using the Twentieth-Century Reanalysis dataset, which is a product of ensemble Kalman filtering data assimilation. Use of the method clearly improves the downscaling skill compared to the case of using only a single member; the skill becomes equivalent to that achieved when between two and six members are used directly.