|Title||Dynamical downscaling of ECMWF EPS forecasts applied to cases of severe weather in Croatia|
|Year of Publication||2007|
|Authors||Brankovic, C, Matjacic, B, Ivatek-Sahdan, S, Buizza, R|
|Secondary Title||Technical Memorandum|
Dynamical downscaling has been applied to global ensemble forecasts to assess its impact during cases of severe weather over various parts of Croatia. It was based on the 12.2-km version of the Aladin limited area model, nested in the ECMWF TL255 EPS. The 3-hourly EPS output was used to force the Aladin model over the southern Europe domain. Both global and regional 51-member ensembles were run. From three cases studied in more detail, downscaling brings improvement in one case where a well-resolved small-scale orography was important. In the second case, no improvement in the downscaled ensemble is seen - severe weather was defined at very small scales over an area with no major orographic differences between the global and regional models. In the third case, severe weather was captured well by the global EPS and Aladin yielded only a little improvement in synoptic development and probabilistic forecasts. Our results indicate that downscaling can have a large impact on clustering: when applied to both sets of ensembles the same clustering algorithm may yield differing results. This implies that downscaling may affect dynamical and physical properties of the global ensemble forecasts. It has been argued that this is due to explicitly resolved small spatial scales, in particular to those related to orography. Therefore, in the process of downscaling, it may not always be feasible to make a selection of global ensemble members that might be representative of all evolution scenarios. This notion is supported by a larger spread in the Aladin than in the EPS ensembles for the fields closely related to small-scale variations in orography. Although this paper discusses too few cases, it provides useful indications on the accuracy of both ECMWF EPS and Aladin ensembles in cases of severe weather. Since no clear-cut difference between the global and the downscaled ensembles was detected, more synoptic cases are required to be able to draw some definite conclusions.