Regional, seasonal and predictor-optimized downscaling to provide groups of local scale scenarios in the complex structured terrain of Austria

The aim of this study is to derive local scale climate change scenarios for temperature and precipitation at approximately thirty stations in Austria and to discuss the scenarios' dependency on combinations of large scale fields used as predictors in empirical downscaling. We distinguish between the seasons and different climatic provinces in Austria. To accomplish this task we utilize large scale, monthly NCEP/NCAR reanalysis data and station data provided by the Austrian weather service for the second half of the 20th century. These datasets are analyzed by means of Empirical Orthogonal Functions (EOF) and combined via transfer functions derived by Canonical Correlation Analysis (CCA). The performance of the transfer functions is validated in several validation experiments and used to determine a group of best performing large scale predictors. These predictors are extracted from the IS92a 'greenhouse gas only' (GHG) and IS92a 'greenhouse gas plus aerosols' (GHG+ars) scenarios, performed with the ECHAM4/OPYC3 climate model for the first half of the 21st century and projected onto the local scale by the transfer functions. We discuss the local scale impact of different predictors. In order to assess the bulk properties of these ensembles, those variables which generate the wettest/driest or coolest/warmest changes are further discussed. All changes mentioned below refer to scenarios for the first half of the 21st century compared to observations recorded during the second half of the 20th century. Findings depend on seasons, regions and the scenario actually used. In the case of temperature and the IS92a GHG scenario an overall increase of about 1° to more than 3°C is found. The temperature increase introduced by the IS92a GHG+ars scenario shows around 0.8°C lower values. Differences between the warm and cool realizations are in the same range. In the case of precipitation the performance of the empirical models is substantially lower. For the IS92a GHG differences between wet and dry realizations are in some cases large and do not even allow to define the sign of change. The range of the IS92a GHG+ars ensemble is smaller and indicates moderate seasonal precipitation reductions. Possible reasons that may cause different local scale responses are discussed and an approach of how to deal with them, based on BUSUIOC and VON STROCH, (1996), is proposed.