Zentralanstalt für Meteorologie und Geodynamik
1. Comparison of HOM, SPLIDHOM and INTERP2. Ideas for the daily benchmark dataset (temperature)
Christine Gruber, Ingeborg Auer
Zentralanstalt für Meteorologie und Geodynamik
Intercomparison experiments
Comparison of: Della-Marta and Wanner, 2006 (HOM) Mestre et al., ???? (SPLIDHOM) Vincent et al., 2002; Brunetti et al., 2006 (INTERP)
I. Semi-synthetic data Use of parallel measurements Combination of series:
artificial but realistic breaks
the truth is known for evaluation of the methods
II. (Preliminary) Application of the methods to a test dataset (Lower Austria) Uncertainty estimation using bootstrap
temperature dependent adjustments
Zentralanstalt für Meteorologie und Geodynamik
Semi-synthetic data
Parallel measurement breaks Realistic inhomogeneities (relocation, screen change,..) Not only temperature dependence included Can be combined at given break point known position
In Austria not enough stations with long parallel measurements available…
Zentralanstalt für Meteorologie und Geodynamik
Results for 5 Stations, TMIN/TMAX, 4 seasons=40 series
Absolute differences of percentiles• Homogenized-truth• RAW-truth
Zentralanstalt für Meteorologie und Geodynamik
Benefit of the homogenization
HOM SPLIDHOM INTERP
Q10 Q50 Q90 Q10 Q50 Q90 Q10 Q50 Q90
TMIN
TMAX
Zentralanstalt für Meteorologie und Geodynamik
Conclusions
For evaluation parallel measurement data is used+ realistic breaks
- only 40 time series homogenized (*20 different samples)
- Many time series too small inhomogeneities, less temperature dependence
HOM and SPLIDHOM similar, main differences for extreme values
Improvement of HOM/SPLIDHOM compared to INTERP, in the case that: Highly correlated reference stations available Inhomogeneity is temperature dependent
Zentralanstalt für Meteorologie und Geodynamik
Lower Austria- Experiment
Preliminary analysis of the Lower Austria temperatures
Mainly to see how the methods work for real data
Influence of reference stations, magnitude of the breaks,…
Testing a bootstrap approach for estimating uncertainties
Break detection with HOCLIS and PRODIGE (annual means) Homogenization with SPLIDHOM (HOM)
Zentralanstalt für Meteorologie und Geodynamik
Lower Austria- Experiment
TMAX PRODIGE HOCLIS META
HOH homogen homogen 1971 05 Station relocation
KRM
1997
197101,198210, 199604
197101,198210,199604
Station relocationStation relocationChange to automated station
RET 19511985
1994
1983 111987 011995 06
1983 11
1995 06
Station relocation
Change to automated station
SPO
1978
1955 091971 011979 041994 01
1955 091971 011979 041994 01
Station relocation21 19 UhrStation relocationChange to automated station
WIE 1951/52
19851994
1953 011971 011980 011993 01
1953 011971 01
1993 01
Station relocation21 19 Uhr
Change to automated station
WMA 19561985(1989)
1956 01
1990 03
1956 01
1990 03
Station relocationStation relocation
ZWE 1971 1971 011980 011994 08
1971 011980 011994 08
21--> 19 UhrStation relocationChange to automated station
Zentralanstalt für Meteorologie und Geodynamik
WIE summer, SPLIDHOM
Ref=KRM
Ref=HOH
Ref=WMA
Influence of undetected breakpoints (higher order moments) in REF?Too short HSPs for KRM, WMA!
1993 1980 1971 1953adju
stm
ent [
°C]
temperature [°C]
Zentralanstalt für Meteorologie und Geodynamik
Adjustments Vienna
Error growth!!!?
1993 1980 1971
HOM
SPLIDHOM
How many values are required that breaks can be adjusted reliably?Comparison of different methods useful
Uncertainty of the adjustments seems to be reduced for earlier breaksIntroduction of a “model” easier to adjust in the following (earlier) breaks
Zentralanstalt für Meteorologie und Geodynamik
WIE winter SPLIDHOM
Ref=KRM
Ref=HOH
Ref=WMA
Zentralanstalt für Meteorologie und Geodynamik
WIE (ref=HOH)
Q10 Q90
Annual mean
All data estimateMean of bootstrap sample0.9 confidence intervalOriginal
Uncertainties in extremes of the adjustments have hardly any influence(in this case)
Zentralanstalt für Meteorologie und Geodynamik
WIE (ref=KRM)
All data estimateMean of bootstrap sample0.9 confidence intervalOriginal
Q10 Q90
Annual mean
Zentralanstalt für Meteorologie und Geodynamik
WIE (ref=WMA)
All data estimateMean of bootstrap sample0.9 confidence intervalOriginal
Q10 Q90
Annual mean
Zentralanstalt für Meteorologie und Geodynamik
Example for usefulness of uncertainty estimates
Q10
No effect of the adjustments on the 0.1 percentileBut information about the (minimum) uncertainty of the time series
Zentralanstalt für Meteorologie und Geodynamik
Example for usefulness of uncertainty estimates
Annual mean
Zentralanstalt für Meteorologie und Geodynamik
Open questions
Requirements for reference stations? correlation length of HSPs
Detection of “higher order moment”- breaks? Is it possible to adjust higher order moments?
Problems due to micro-scale climate changes (test-reference station distribution change)
Uncertainty assessment (especially for extreme values) method uncertainty sampling uncertainty representativeness (references)
Zentralanstalt für Meteorologie und Geodynamik
Benchmark daily data
Zentralanstalt für Meteorologie und Geodynamik
The nature of the problem
Extreme value studies homogenization of daily data necessary
Adjusting inhomogeneities in dependence of the weather type, physical reasons (primary effect) Adjustments as function of wind, sunshine duration, global
radiation… (difficult due to data availability)
Adjustment of the temperature dependence of the inhomogeneities (secondary effect) Adjusting the temperature distribution (e.g. Della-Marta and
Wanner, 2006) Effect of inhomogeneities on temperature percentiles/extremes is
reduced (that’s what we want in extreme value studies)
In a first step: Shall we take into account only temperature dependent breaks in the daily benchmark?
Zentralanstalt für Meteorologie und Geodynamik
Significance of temperature dependence
How often significant temperature dependence occurs?
Typical pattern and range of the magnitude pattern for synthetic inhomogeneities
Zentralanstalt für Meteorologie und Geodynamik
Possible working steps
I. Case study• Real dataset, metadata (availability?)• Classification of inhomogeneities due to their source• Examination of temperature dependence? (e.g. HOM)• Other dependencies (wind, radiation,…)• Typical pattern benchmark
II. Semi-synthetic (parallel measurement) series• Realistic inhomogeneities, but truth is known for evaluation• Dependencies to other elements could be studied (wind, radiation?)• Data availability? (too few stations in Austria)
III. Surrogate• Based on typical inhomogeneity pattern (temperature dependent)• (If other dependencies shall be treated as well benchmark multiple
series???? ( new adjustment-method multi-parameter???)
Typical pattern?
We
mus
t lea
rn m
ore
abou
t the
pro
blem