CR 20:211-222 (2002)  -  doi:10.3354/cr020211

ABSTRACT: This study concentrates on the comparison of 20 and 50 yr return values of annual maximum and minimum temperatures in (1) observations, (2) general circulation model (GCM) simulated control climates, (3) statistical downscaling from observations, and (4) statistical downscaling from GCMs. Temperature series at 4 sites in central Europe and in the nearest GCM gridpoints corresponding to the stations are analysed. Extreme value analysis was performed by fitting the generalized extreme value (GEV) distribution using L moment and maximum likelihood estimators. The comparison does not appear to be sensitive to which statistical method of the estimation of parameters of the GEV distribution is used, although individual return values are influenced by the choice of method. The skill of both GCMs in reproducing extreme high and low temperatures is limited, and statistical downscaling from GCMs tends to improve their performance, although it generally yields extremes that are too moderate compared to observed values (this holds for downscaling both from observations and GCMs). White noise addition in downscaling from observations leads to more realistic return values of annual maximum and minimum temperatures than variance inflation.

KEY WORDS: General circulation model · Statistical downscaling · Validation · Surface temperature · Central Europe · Extreme value analysis · GEV distribution · Maximum likelihood method · L moments