For understanding and proper interpretation of climate projections, detailed examination of reference databases used for validating regional climate models is essential. The knowledge of the main characteristics of these datasets is important in order to correctly interpret such statements as the summer mean temperature is too high in the control period, for instance.
The following datasets were scrutinised from this angle at the Hungarian Meteorological Service: ERA-40 re-analysis data by ECMWF (European Centre for Medium-Range Weather Forecasts); two CRU (Climatic Research Unit) datasets prepared by different methodologies and (approximately 50 and 20 km) resolutions; the latest version of ECA&D (European Climate Assessment & Dataset); the gridded, so-called HUGRID data made by Climate Analysis Division of HMS, and some raw observational time series (mainly for precipitation and temperature) in Hungary.
For evaluation and assessment of the datasets it is important to understand their production method. Mostly non-homogenized observations with the use of different interpolation techniques were applied during the preparation process of CRU, ECA and HUGRID resulting in surface gridded values covering the entire globe (CRU0.5), Europe (CRU10' and ECA) or the area of Hungary (HUGRID). Datasets representing a large area often face the problem of insufficient information in data-sparse regions (such as oceans or seas) and in higher atmospheric layers (where fewer observations are available). The largest difference between the above-mentioned data and the ERA-40 re-analyses is that the latter one is prepared by using not only observational data, but also short-term numerical forecasts. The result of this data-assimilation type production is that rather precise information at higher atmospheric levels and at data-poor regions is available, as well. It is essential to note that the precipitation fields of ERA-40 are originated from six-hour model forecasts, therefore, their use for comparison is limited (which is not the case for temperature).
More details about the datasets can be found at the following references:
- ERA-40: Uppala et al. (2005)
- CRU 0.5: Mitchell and Jones (2005)
- CRU 10': Mitchell et al. (2004)
- ECA: van Engelen et al. (2008)
- HUGRID: Szentimrey et al. (2005)
One can conclude on the comparative analysis of the observational datasets that ERA-40 is rough and not suitable for fine-scale validation, but only this data can provide "perfect" initial and lateral boundary conditions for our RCMs for the past simulations. Among the CRU datasets the coarse (half-degree) resolution one contains larger number of newer and, presumably, more accurate data. ECA database is similar to the finer (10-minute) CRU, except that the former is updated regularly. As far as the quality of the internationally developed datasets for Hungary is concerned, these data may depict only an approximation of the real climate, since there were relatively few of the available observations used during their production, unlike HUGRID where all the Hungarian stations were included.
Daily data series are available only in ECA and HUGRID, therefore, extreme climate indices and their trends can be calculated uniquely from these datasets. On the other hand, several European and Hungarian institutions prefer to use CRU10' (because of its better resolution) for the evaluation of the general behaviour of their RCMs.
Below, the annual mean temperature time series observed in Hungary and its linear trend between 1961 and 2000 based on HUGRID (the increase is significant; Fig. 1) and the mean summer temperature difference for 1961-2000 between CRU10' and ECA (Fig. 2) can be seen.
Figure 1: Time series and its linear trend of mean annual temperature (°C)
observed in Hungary (based on HUGRID) for 1961-2000.
Figure 2: Mean temperature difference in summer of 1961-2000 (°C)
between (10' resolution) CRU and ECA.