Development and validation of a high-resolution monthly gridded temperature and precipitation data set for Switzerland (1951-2000)

A 5 km gridded temperature and precipitation data set was constructed for the topo- graphically complex region of Switzerland in the European Alps. The data set consists of 1961-1990 mean fields for monthly mean temperature (T -- ) and monthly total precipitation (P -- ), plus monthly anomaly fields ∆T and ∆P for 1951-2000. All data are point estimates and come with extensive statis- tics on interpolation errors as a function of geographical location, elevation and time of the year. A novel interpolation method was employed that accounted for possible orographic effects at different spatial scales and allowed for regionally and seasonally varying relief-climate relationships. The accuracy of the interpolations was quantified by means of cross-validation. The proposed method was found to be superior to linear regression employing elevation as the only predictor for P- , and better than inverse distance weighting (IDW) interpolation for September to February ∆T. It was worse than IDW interpolation for springtime ∆T and for March to September ∆P. The areal mean cross-validation errors obtained for the new method were generally close to zero. The annually averaged mean absolute error for T -- was 0.6°C and for P- it was 10.5 mm mo -1 (or 11%). The average proportion of temporal variance explained by the cross-validated monthly 1951-2000 station time series was 89% for ∆T and 81% for ∆P. The average proportion of spatial variance of the monthly anomaly fields explained was 13% for ∆T and 40% for ∆P. The largest cross-validation errors were generally found at regions of lower station density in south-southeast Switzerland and at elevations above ~2000 m above sea level. All error variances showed distinct annual cycles. The ∆T and ∆P fields and the derived trend fields showed substantial small-scale variability, which was not well reproduced and deserves further study. The individual gridpoint estimates should therefore be interpreted with care.