Data Quality Control of MAP (DAQUAMAP)

Inga Groehn, Reinhold Steinacker, Christian Häberli, Wolfgang Pöttschacher, Manfred Dorninger, Department of Meteorology and Geophysics, University of Vienna, 1190 Wien, Austria
Introduction

Before evaluating meteorological data a quality control should be carried out. For example, observational errors may have a serious impact on objective analyses. Before conducting an objective analysis, i.e. interpolating irregularly spaced observations to a uniform grid, the data should be checked up on errors carefully.

But also when using data from different data providers, it is advisable to check the data with a uniform method in order to get a more homogeneous data set irrespective of e.g. state borders. The stations of MAP are operated by several dozens of institutions. So some variety in measuring and processing procedures as well as in quality control efforts have to be expected.

For this purpose a mathematical method of quality control has been developed and continually improved during the last years at the Department for Meteorology and Geophysics (University of Vienna) in the project of DAQUAMAP in order to get a more homogeneous data set for the whole Alpine region.

Method (short overview)

The method includes a piecewise functional fitting approach which is based on a variational algorithm. Like for thin-plate splines, an integral of squares of second spatial derivatives is minimised. The second derivatives are obtained from overlapping finite elements using a polynomial approach (Steinacker et al. 2000). The basic advantages of this method are that neither a first guess or (prognostic) model field is necessary nor a priori knowledge about structure or weighting functions is required.

Implementation

The results (in terms of deviation estimates) of the year 1999 are now available on the web (as soon as the results have been approved by the respective data provider):

(You will be asked for your MAP password).

At the moment the parameters mean sea level pressure, potential temperature, and humidity of GTS-stations below 750 msl were calculated. An extension to the non-GTS surface stations of the participating institutions is under way.

 

 

obs 

n

PGE

median

IQR

mean

sdev

 all

2144

0.23

0.67

0.96

0.62

0.88

00

288

0

0.64

0.97

0.64

0.69

06

337

0.3

0.9

0.88

0.75

0.82

12

338

0.3

0.54

0.96

0.49

0.93

18

348

0.57

0.56

0.9

0.49

0.79


Figure 1. Time series of deviation estimate (reduced pressure) and statistical evaluation of station 1 (n denotes number of observations, PGE the percentage of gross errors, and IQR the interquartile range).

 

 

obs 

 n PGE median IQR mean sdev
 all 2611 0.27 -0.21 0.41 -0.24 0.37
00 353 0 -0.3 0.45 -0.32 0.36
06 347 0 -0.28 0.44 -0.3 0.34
12 350 0.86 -0.16 0.37 -0.18 0.35
18 349 0.57 -0.08 0.3 -0.12 0.3

Figure 2. Time series of deviation estimate (reduced pressure) and statistical evaluation of station 2.

 

Results

General comments:

A positive (negative) deviation means that the measured value of the station is higher (lower) than predicted from the surrounding stations. If it persists and does not show any correlation with the diurnal or seasonal cycle or with a synoptic setting, it may be caused by sensor problems (systematic errors) and can give some hints on the representativity of the stations with respect to the scale which can be resolved by the local/regional station density.

 

Example for reduced pressure

In figure 1 there is a light positive deviation with the median (dashed line) of 0.67 for nearly all observations until middle/end of October. At that time it looks as if the barometer was changed. The two strong deviations concentrated at the beginning of the year indicate most probably gross errors caused by instrumental misreading, miscoding etc.

In comparison figure 2 shows very small deviations (compare also tables with their statistics). Only at the beginning and at the end there are some `stronger' deviations due to meteorological forcing. At the end of January you also find a single gross error.

 
Example for potential temperature

A station higher than the surrounding stations will show a natural tendency for positive deviation because the annual mean potential temperature is increasing with height (approximately 0.4 K/100 m). But also in the case of two nearby stations with differences in height: This can lead to deviations with opposite sign especially during the cold season when one station is situated above the temperature inversion while the other one stays in the cold air pool. Also heat island effects of big cities (Fig.3) are represented in temperature measurements.

In comparison figure 4 shows very clear meteorological forcing and the diurnal cycles (compare median values).

 

 

obs 

 n PGE median IQR mean sdev
 all 2611 0.92 1.05 1.35 1.27 1.39
00 353 0.57 1.62 1.72 1.8 1.51
06 347 0 1.37 1.52 1.59 1.33
12 350 2 0.67 1.08

0.8

 1.14

18

 349 2.01 0.79 0.97 0.86 1.1

Figure 3. Time series of deviation estimate (potential temperature) and statistical evaluation of station 3.

 

 

obs 

 n PGE median IQR mean sdev
 all 2641 0.8 -0.02 1.06 0.05 1.06
00 356 0.84 -0.14 1 -0.05 1.08
06 355 0.85 -0.06 1.08 -0.03 1.09
12 356 0.84 0.11 1.11

0.2

 1.07

18

 345 0.58 0.05 1.07 0.1 1.03

Figure 4. Time series of deviation estimate (potential temperature) and statistical evaluation of station 4.

 

References

Steinacker R., C. Häberli, and W. Pöttschacher, 2000: A transparent method for the analysis and quality evaluation of irregularly distributed and noisy observational data. Mon. Wea. Rev., 128, 2303-2316.

Acknowledgements

DAQUAMAP is funded by the optional EUMETNET programme MAP-NWS.

 

© MAP Data Centre - April '05 - MAP WebMaster