Data quality control for the ENSEMBLES grid

1
Data quality control for the ENSEMBLES grid
Evelyn Zenklusen Michael Begert Christof
Appenzeller Christian Häberli Mark
Liniger Thomas Schlegel
2
(No Transcript)
3
What we have and what we aim at

• Methods based on ECAD experience
• implemented
• statement if series are homogeneous or not for a
  given period (e.g.1946-1999)
• Additional goals
• date the breakpoints
• homogeneous subperiods
• separate information for each climate variable

useful (?), doubtful (?), suspect (?)
4
THOMAS(Tool for Homogenization of Monthly Data
Series at MeteoSwiss)

• Pro
• Twelve different homogeneity tests implemented
• Includes full station history
• Based on monthly time series but daily output
  resolution possible
• Contra
• Includes a lot of manual work (construction of
  reference series, interpretation of test results)
• ? not suited for large datasets (ENSEMBLES)
• But
• the Swiss series homogenized by THOMAS provide a
  highly valuable core dataset for the testing in
  ENSEMBLES
• Reference and details
• Begert Michael, Schlegel Thomas and Kirchhofer
  Walther, 2005 Homogenous temperature and
  precipitation series of Switzerland from 1864 to
  2000, Int. J. Climatol. 25 65-80.

5
VERAQC (Vienna Enhanced Resolution Analysis
Quality Control at Univ. Vienna)

• Pro
• based on objective spatial interpolation
• designed for quality control
• applied at MeteoSwiss on daily data
• idea use VERAQC-output for homogenization
• Contra
• Not yet tested. - Does it work??
• References and details
• Steinacker Reinhold, Christian Häberli and
  Wolfgang
• Pöttschacher, 2000 "A transparent method for
  the
• analysis and quality evaluation of irregularly
  distributed
• and noisy observational data", Monthly Weather
  Review, Vol. 128, No. 7, pp. 2303-2316.

6
VERAQC for homogenizing the ENSEMBLES dataset
European monthly data
Homogeneity test (EasterlingPeterson
two-phase Regression homogeneity
test Alexanderssons standard normal homogeneity
test)
Deviations
Significant breakpoints
7
Precipitation 1960-2004 VERAQC Alexandersson
number of breakpoints detected 0(?), 1(?), 2(?),
3(?), 4(?), gt5(?)
8
Tmin 1960-2004 VERAQC Alexandersson
number of breakpoints detected 0(?), 1(?), 2(?),
3(?), 4(?), gt5(?)
9
Example series precipitation Beesel 1960-2004
Deviation series
Breakpoints detected by Easterling Peterson
Breakpoints detected by Alexandersson
10
Discovered limitations of VERAQC

• sensitivity to changes in network density
• incomplete deviation series for some stations
  (example Amiandos)

11
Changes in the station networkExample Amiandos
precipitation 1960 - 2004

• Observation series

Deviation series
12
Discovered limitations of VERAQC

• sensitivity to changes in network density
• incomplete deviation series for some stations
  (example Amiandos)
• artificial breakpoints (example Andermatt)

13
Changes in the station networkExample Andermatt
maximum temperature 1960-2004
Deviations Andermatt Tmax
Deviations Locarno Tmax
Deviations Engelberg Tmax
14
Discovered limitations of VERAQC

• sensitivity to changes in network density
• incomplete deviation series for some stations
  (example Amiandos)
• artificial breakpoints (example Andermatt)
• One step further to test the process
• analyse only complete station series of a
  desired period
• e.g. 1960-2000 (network density of complete
  climate series is high)
• Precipitation 795 stations (55)
• Tmin 527 stations (60)

15
Precipitation only complete series 1960-2000 VER
AQC Alexandersson
number of breakpoints detected 0(?), 1(?), 2(?),
3(?), 4(?), gt5(?)
16
Tmin only complete series 1960-2000 VERAQC Alexa
ndersson
number of breakpoints detected 0(?), 1(?), 2(?),
3(?), 4(?), gt5(?)
17
Precipitation Difference breakpointsall -
breakpointscomplete 1960-2000 VERAQC Alexanderss
on
Lower(?), equal(?) or higher (?) number of
breakpoints if only complete series are tested
18
Tmin Difference breakpointsall -
breakpointscomplete 1960-2000 VERAQC Alexanderss
on
Lower(?), equal(?) or higher (?) number of
breakpoints if only complete series are tested
19
Skill of VERAQCCH-stations comparison with
THOMAS

• Precipitation 1960-2000, only complete series

number of breakpoints
Total amount of breakpoints detected VERAQC_ep 7
9 VERAQC_alex 52
20
Skill of VERAQCCH-stations comparison with
THOMAS

• Tmin 1960-2000, only complete series

number of of breakpoints
Total amount of breakpoints detected VERAQC_ep 1
97 VERAQC_alex 110
21
Has VERAQC detected the large adjustments and
missed the small ones?
Precipitation(mean adjustment factors of THOMAS) Precipitation(mean adjustment factors of THOMAS) Minimum temperature(mean adjustment amounts of THOMAS) Minimum temperature(mean adjustment amounts of THOMAS)
detected missed detected missed
EP 21.0 ( 10.5) 14.0 ( 7.8) 0.81C ( 0.46) 0.62C ( 0.39)
SNHT 24.0 ( 13.9) 14.7 ( 7.3) 0.89C ( 0.46) 0.61C ( 0.38)
22
Summary and conclusions

• ECAD procedure is implemented and works
• With VERAQC an automated homogeneity test
  procedure has been implemented and tested
• method shows unsatisfying results
• significant loss of stations at the edge of
  investigated area
• sensitive to changes in the network density
• high number of undetected inhomogeneities and
  false alarms
• sensitive to inhomogeneities in reference
  series(dispersion of inhomogeneities)

23
Outlook

• Two ways to proceed
• Improvement of VERAQC test procedure
• reduce influences of the varying network
  density(anomalies as inputdata, flag breakpoints
  generated by network changes)
• reduce false alarm rate(combination of test
  results, test tuning)
• Calculation of deviation series according to
  THOMAS procedure
• selection of reference stations due to
  correlation analysis
• use a mean of chosen reference series to
  calculate the deviations

24

• Thank you for your attention

questions ?