![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Bias correction of EU-ENSEMBLES precipitation data with focus on the effect of sample size |
VerfasserIn |
Philipp Reiter, Oliver Gutjahr, Lukas Schefczyk, Günther Heinemann, Markus C. Casper |
Konferenz |
EGU General Assembly 2015
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 17 (2015) |
Datensatznummer |
250105403
|
Publikation (Nr.) |
EGU/EGU2015-4922.pdf |
|
|
|
Zusammenfassung |
The precipitation output of climate models often shows a bias when compared to observed
data, so that a bias correction is necessary before using it as climate forcing in impact
modeling. We expect the performance of the bias correction to strongly depend on the sample
size used for its calibration. This raises the question: how long does a time series need to be
to achieve a sufficient bias correction?
We carry out experiments using 40 years of daily precipitation data from 10 regional
climate models (RCM) of the EU-ENSEMBLES project, splitting them into a 30
year calibration period and a 10 year validation period. The RCM data are bias
corrected using decreasing sample sizes out of the calibration period. By applying skill
scores we quantify the critical sample size ncrit, at which the quality of the bias
correction becomes statistically worse compared to the correction based on 30 years.
In order to analyze whether the effect of the sample size depends on the chosen
correction method and the calibration period, we applied four variations of the
quantile matching (QM) approach and 3 different calibration/validation periods in this
study.
The results show that the spread of ncrit is large, ranging from 28 years to
approximately 10 years. This indicates that even a small decrease in sample size for the
calibration can result in a statistical significant degradation of the bias correction.
Corrections with sample sizes smaller than 10 years always perform significantly
worse than the ’best fit’ with 30 years. The chosen QM approach influences ncrit
in dependence of its degrees of freedom: the higher the degrees of freedom the
larger ncrit. We also found that the choice of the calibration period affects the ncrit
values.
In conclusion we recommend to use time series as long as possible for bias correction of
precipitation data. However, there is a large transition zone of the critical sample size where
shorter time series can perform sufficiently well, depending on the chosen correction method
and calibration/validation period. Thus, it is not possible to determine a general minimum
sample size. |
|
|
|
|
|