For one month lead time, the ACC for observations is up to 0.6 (EV < 35%) for February/March and August/September in the three regions. The skill for the simulated data is lower (up to ACC = 0.5) and its seasonal dependence differs from that of the observations. Main predictors are the preceding temperatures in the predictand region. The simulated data is split in segments to estimate the distribution of skill. For lead times up to one year there is skill (ACC >0.3) in the observations for England (spring and late summer), and Scandinavia (August-September), but none in Germany. The observed two-month mean England temperature in spring and late summer can be predicted with 6 months lead time using the first two North Atlantic SST EOF coefficients for 1970 to 1996 with 1870-1969 as training set. A leave-two-out cross-validation in 1870-1999 shows an obvious reduction of skill. Beyond one month, the skill in simulated data is much lower than in the observations. The linear regression prediction scheme shows up as a method to evaluate general circulation models.
KeyWords Plus:European Climate, Predictability, Monthly forecasts, GCM simulation
Addresses:Blender R, Luksch U, Fraedrich K, Univ Hamburg, Inst Meteorol, Bundesstr 55, D-20146 Hamburg, Germany Raible C.C., Climate and Environmental Physics, Physics Institute, University of Bern, Sidlerstr. 5, CH-3012 Bern, Switzerland.
Reprints: Raible CC, Climate and Environmental Physics, Physics Institute, University of Bern, Sidlerstrasse 5, CH-3012 Bern, Switzerland, raible@climate.unibe.ch