Constructing Site-Specific Climate Change Scenarios on a Monthly Scale Using Statistical Downscaling

Abstract

Monthly mean temperature and monthly precipitation totals in two small catchments in the Czech Republic are estimated from large-scale 500 hPa height and 1000/500 hPa thickness fields using statistical downscaling. The method used is multiple linear regression. Whereas precipitation can be determined from large-scale fields with some confidence in only a few months of the year, temperature can be determined successfully. Principal components calculated separately from the height and thickness anomalies are identified as the best predictor set. The method is most accurate if the regression is performed using seasons based on three months. The test on an independent sample, consisting of warm seasons, confirms that the method successfully reproduces the difference in mean temperature between two climatic states, which indicates that this downscaling method is applicable for constructing scenarios of a future climate change. The ECHAM3 GCM is used for scenario construction. The GCM is shown to simulate surface temperature and precipitation with low accuracy, whereas the large-scale atmospheric fields are reproduced well; this justifies the downscaling approach. The observed regression equations are applied to 2xCO2 GCM output so that the model’s bias is eleminated. This procedure is then discussed and finally, temperature scenarios for the 2xCO2 climate are constructed for the two catchments.