Long-term trends of precipitation in Europe: a comparison across multiple datasets

Different types of climate datasets (station, gridded, reanalyses) and even individual datasets have been shown to differ in how they capture statistical properties of climate variables. Here we compare trends in precipitation totals in Europe between station data (taken from the ECA&D database), gridded data (E-OBS and CRU TS), and reanalyses (20CR, JRA-55, and NCEP/NCAR) for period 1961-2011, both annually and for individual seasons. Theil-Sen non-parametric trend estimator is used for the quantification of the trend magnitude; Mann-Kendall test is used to evaluate the significance of trends.

On the annual basis, station data indicate precipitation increases in northern Europe and decreases in southern and southeastern Europe. Whereas trends in the gridded datasets roughly agree with station data, although tend to overestimate them, reanalyses provide much more negative trends with a different geographical distribution. There is a tendency for reanalyses to overestimate precipitation in the beginning of the period at some places, whereas they underestimate precipitation near the end of the period elsewhere. Particularly notable is an excessive, and likely unrealistic, drying trend in central, southwestern, and southeastern Europe in NCEP/NCAR in most seasons. Reanalyses thus do not appear to be suitable data sources for estimation of precipitation trends.  

Reasons for the disagreement are identified by a detailed examination of local or regional time series. The reasons are varied and depend on the specific type of dataset: Station series may suffer from inhomogeneities; gridded data may be affected by different sets of stations entering the interpolation procedure at different times; while reanalyses may be affected by different kinds of data being assimilated into them in different periods.