The 
precipitation
database
In the context of ELDAS two precipitation products were produced:
- A high resolution gauge and radar based data set covering Europe at 3-hourly resolution between 1 Oct 1999 and 31 Dec 2000
- Global 1° daily satellite estimates (GPCP-1DD) calibrated with on event bias-corrected rain gauge data.
The European data set is based on approx. 21,000 rain-gauges (see Figure 1.1). For a detailed description of the database see Rubel et al. (2003). Additional efforts have been undertaken to extend the database for future projects to Eastern European countries (Skomorowski et al., 2004).
All rain gauge measurements have been corrected for systematic measurements errors due to wind induced losses, evaporation and wetting. The corrected measurements have been analysed on a regular 0.2° lat/lon grid using an OI (ordinary block kriging) technique and saved as GRIB format in the ECMWF MARS archive. Apart from the analyses, the kriging technique provides the error variance on a daily basis, which is used to identify regions where observations are not sufficient. These have been replaced by precipitation forecast of the operational run of the ECMWF NWP model.
|
Country |
Number of gauges |
|
Germany |
4000 |
|
Spain |
3350 |
|
Portugal |
50 |
|
Great Britain |
2900 |
|
France |
4250 |
|
The Netherlands |
320 |
|
Ireland |
400 |
|
Norway |
670 |
|
Sweden |
740 |
|
Finland |
400 |
|
Denmark |
210 |
|
Latvia |
80 |
|
Lithuania |
60 |
|
Estonia |
60 |
|
Poland |
1190 |
|
Austria |
490 |
|
Italy |
200 |
|
Switzerland |
350 |
Fig.1.1: Distribution of precipitation gauges for the ELDAS precipitation database.
In addition, the Baltic radar data (BALTRAD; Michelson, 2003, 2004) as well as the Central European radar data (CERAD) have been collected for the entire ELDAS period October 1999 to December 2000. For CERAD a semi-automatic algorithm to detect and remove the various uncertainties was developed. The various errors in the radar estimates result in locally large differences between the ground truth and the radar estimates of precipitation. The sources of errors are for example ground and sea clutter, geometrical artefacts produced by solar radiation or temporal dysfunction of the radar sites. But also uncertainties based on the Z-R relationship, missing flags - to distinguish between „not measured“ and „no precipitation“ - and errors resulting from the composting algorithm made it necessary to develop such a method. Figure 1.2 depicts the single steps in analysing radar data beginning with the original CERAD precipitation estimates and ending with the final radar analyses on the ELDAS grid.
Figure 1.2: Processing line for CERAD precipitation: Daily precipitation estimated from raw CERAD data (upper left panel), CERAD data after correction (upper right), corrected CERAD interpolated to the 0.2° ELDAS grid (lower left) and for comparison analysed rain gauge data on the ELDAS grid (lower right). Date July 6, 2000, 06 UTC. Note that the scaling for precipitation on the original CERAD projection (upper panels) is different from those on the ELDAS grid (lower panels).
The upper-left panel of Figure 1.2 shows errors in the raw data due to sea clutter over the Baltic Sea, geometric artefacts in form of rings over Austria and Hungary and features of high precipitation values overlapping the ring. The upper-right panel shows the same date after having applied the semi-automatic method to reduce these artefacts. For the BALTRAD database only minor corrections had to be applied after the initial quality control at SMHI. Only faults in the calculation of reflectivity values to precipitation values below a threshold of 0.2 mm/h in the 15-minute composite image had to be adjusted due to non-realistic precipitation values in 3- and 24-hourly accumulations.
The results of the semi-automatic clutter removal algorithm have been used for the temporal disaggregation of the daily precipitation analysis to three hourly fields. The following steps have been performed to obtain these fields:
1. The 3-hourly accumulated radar-derived precipitation fields from the BALTRAD and CERAD network have been merged together on the ELDAS grid using a maximum display algorithm.
2. Meteo France provided 3-hourly SAFRAN precipitation output instead of raw gauge data. This dataset covered France and Corsica and has been blended with the radar data.
3. As background-field, an ECMWF T511 experiment run has been used in regions, were no radar or information from SAFRAN was available.
4. A linear disaggegation scheme was applied to obtain 3-hourly precipitation fields, which are consistent with the daily analysis.
Figure 1.3: Left: Merged Radar, T511 model run and SAFRAN precipitation fields with a temporal resolution of 3 hours on a 0.2° lat/lon grid; Right: the gauge data blended with the 3- hourly data from the left panel. ECMWF precipitation fields are shown in paler colours.
The left panel of Figure 1.3. shows the 3-hourly accumulated, merged BALTRAD, CERAD, SAFRAN and T511 experiment fields on the regular 0.2° lat/lon ELDAS grid. In the right panel the final result is depicted, where the daily gauge data have been disaggregated using the linear coefficients derived from the fields presented in the left panel. In addition the calibrated ECMWF t511 experiment run is displayed in brighter colour.
The BALTRAD radar fields were further analysed at SMHI to create local datasets of high resolution gauge adjusted precipitation for use in the demonstration flood forecasting studies at basin scale. In this study the radar data is calibrated to quantitative precipitation amounts using general calibration procedures. This differs to the use of radar data in the ELDAS precipitation data base, where radar intensities are only used for the temporal disaggregation of the gauge data. The applicability of the radar based quantitative precipitation rates less straightforward, as will be discussed below.
Figure 1.4: Combined satellite gauge precipitation product for July 1, 2000.
In addition, a global scale daily precipitation product based on existing multi-satellite estimates, the GPCP-1DD product, and bias-corrected rain gauge measurements has been provided in a preliminary version for the year 2000 (Kottek and Rubel, 2004). This work will be continued in the EU FP6 integrated project GEOLAND. The multi-satellite estimates of precipitation as well as the bias-corrected rain gauge measurements, based on about 6000 synoptic stations worldwide are available for the period 1997 to 2003. An example for 1st July 2000 is shown in figure 1.4. For verification purposes of this combined precipitation product over Europe the ELDAS dataset will be used.