Temperature trends in Europe: comparison of different data sources

Temperature trends differ markedly not only region-to-region and between seasons but also depending on the selected dataset. Only a few studies have attempted to compare temperature trends between data sources of different types. Here, one station-based (ECA&D), two gridded (E-OBS; CRUTEM) and two reanalysis (ERA-40; NCEP/NCAR) datasets are used for long-term temperature change detection over Europe. The period from 1957 to 2002 when all the datasets overlap is examined and the linear regression method is utilized to calculate temperature trends in each season separately. Raster maps illustrating differences in trends between datasets are accompanied by mean temperature series showing the causes of these discrepancies. We demonstrate that trends in reanalyses deviate considerably from the other datasets mainly because the type and amount of data assimilated into them change in time. Interestingly, whilst the ERA-40 shows lower trends due to an overestimation of the mean temperature prior 1967, the NCEP/NCAR reveal lower trends compared with other datasets owing to mean temperature underestimation at the end of the examined period. A noticeable anomaly in NCEP/NCAR data was detected in Eastern Europe in summer with temperature trends nearly twice as steep compared with other data sources. The study also reveals the weaknesses of gridded datasets, such as the unstable number of stations entering the interpolation over time. The lack of representativeness of some climate stations is the major drawback of the station data.