![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Stepwise analogue downscaling for hydrology (SANDHY): validation experiments over France |
VerfasserIn |
Sabine Radanovics, Jean-Philippe Vidal, Eric Sauquet, Aurélien Ben Daoud, Guillaume Bontron |
Konferenz |
EGU General Assembly 2014
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 16 (2014) |
Datensatznummer |
250093063
|
Publikation (Nr.) |
EGU/EGU2014-7437.pdf |
|
|
|
Zusammenfassung |
Statistical downscaling aims at finding relationships between local precipitation (predictand)
and large-scale predictor fields, in various contexts, from medium-term forecasting to climate
change impact studies. One of the challenges of statistical downscaling in a climate
change context is that the predictor-predictand relationship should still be valid
under climate change conditions. A minimum requirement is therefore to test the
performance of the downscaling method on independent data under current climate
conditions.
The downscaling method considered is the Stepwise ANalog Downscaling method for
HYdrology (SANDHY). ERA-40 reanalysis data are used as large scale predictors and daily
precipitation from the French near surface reanalysis (Safran) as predictand. Two 20-year
periods have been selected from the common archive period of the two data sources:
1958–1978 (“early”) and 1982–2002 (“late”). SANDHY has been optimised over the late
period in terms of geopotential predictor domains individually for 608 target zones covering
France.
The validation setup consists of 4 experiments, that all use the parameters as optimised
for the late period and that are compared in terms of continous ranked probability skill score
(CRPSS) with climatology as reference:
Reference simulation. A simulation of the late period is performed using the
late period as an archive for searching the analogue dates, thus representing the
best possible case. The CRPSS shows a spatial distribution similar to the one of
the mean precipitation.
Out-of-sample validation. The early period is simulated using the late period
as an archive for searching the analogue dates. The idea is to simulate a period
whose local data is not “known” by the model as it would be the case in
any application. The average skill loss compared to the reference simulation is
reasonable with some more skill loss in the northern part of the country and no
loss in the southeastern part.
Alternative archive. The late period is simulated using the early period as an
archive for the analogue search. Using the alternative archive leads to small and
spatially uniform skill loss compared to the reference simulation.
Imperfect predictor domains. The early period is simulated using the early
period as an archive for the analogue search. The results are very similar to the
out-of-sample validation in terms of mean skill loss and spatial distribution.
The results of experiment 2 indicate that SANDHY is quite robust at most locations.
Experiment 3 shows that both archives are suitable for downscaling. Experiment 4
shows that the skill loss observed in experiment 2 stems rather from the imperfect
predictor domains than from the imperfect archive. Overall the results increase the
confidence in applying SANDHY for downscaling in various contexts over France. |
|
|
|
|
|