Statistical downscaling in European mountains: verification of reconstructed air temperature

General circulation models constrained by past meteorological, records have been run as part of a major community effort, using datasets from 1958 to the present date, to produce consistent gridded atmospheric databases known as reanalysis data. We derive linear regression models to transfer reanalysis gridded data to high elevation weather stations in Europe using daily observations from 1994 to 2001. The models are then used to reconstruct daily mean, minimum and maximum air temperatures since 1958 at the weather stations. The regression models use principal components of reanalysis temperature and pressure variables in addition to local (nearest grid point) temperature and pressure variables. An all-subsets regression technique in conjunction with cross validation is used to find the best model. The accuracy of the approach is verified using observed monthly data from 1980 to 1990 at 29 stations, and monthly data since 1958 at 8 stations. The verification results indicate that retrodiction to 1980 is good at all stations. However, validation at 3 alpine stations shows large differences between observed and reconstructed temperatures prior to 1970. Nevertheless, the basic spatio-temporal warming pattern we reconstruct for the European mountains has many similarities to that for the European 'lowlands'. We find regional climatic trends for the period 1958 to 2001 of typically 0.7degreesC per 100 yr for minimum temperatures and twice that for maximum temperatures. These trends are probably underestimated. Our reconstructions suggest that there has been an increase in the diurnal temperature range in European mountains in addition to the overall warming.