![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Parametrisation of initial conditions for seasonal stream flow forecasting in the Swiss Rhine basin |
VerfasserIn |
Simon Schick, Ole Rössler, Rolf Weingartner |
Konferenz |
EGU General Assembly 2016
|
Medientyp |
Artikel
|
Sprache |
en
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 18 (2016) |
Datensatznummer |
250125256
|
Publikation (Nr.) |
EGU/EGU2016-4815.pdf |
|
|
|
Zusammenfassung |
Current climate forecast models show – to the best of our knowledge – low skill in
forecasting climate variability in Central Europe at seasonal lead times. When it comes to
seasonal stream flow forecasting, initial conditions thus play an important role.
Here, initial conditions refer to the catchments moisture at the date of forecast, i.e.
snow depth, stream flow and lake level, soil moisture content, and groundwater
level.
The parametrisation of these initial conditions can take place at various spatial and
temporal scales. Examples are the grid size of a distributed model or the time aggregation of
predictors in statistical models. Therefore, the present study aims to investigate the extent to
which the parametrisation of initial conditions at different spatial scales leads to differences
in forecast errors. To do so, we conduct a forecast experiment for the Swiss Rhine at Basel,
which covers parts of Germany, Austria, and Switzerland and is southerly bounded by the
Alps. Seasonal mean stream flow is defined for the time aggregation of 30, 60, and 90 days
and forecasted at 24 dates within the calendar year, i.e. at the 1st and 16th day of each
month.
A regression model is employed due to the various anthropogenic effects on the basins
hydrology, which often are not quantifiable but might be grasped by a simple black box
model. Furthermore, the pool of candidate predictors consists of antecedent temperature,
precipitation, and stream flow only. This pragmatic approach follows the fact that
observations of variables relevant for hydrological storages are either scarce in space or time
(soil moisture, groundwater level), restricted to certain seasons (snow depth), or regions (lake
levels, snow depth).
For a systematic evaluation, we therefore focus on the comprehensive archives of
meteorological observations and reanalyses to estimate the initial conditions via climate
variability prior to the date of forecast. The experiment itself is based on four different
approaches, whose differences in model skill were estimated within a rigorous
cross-validation framework for the period 1982-2013:
The predictands are regressed on antecedent temperature, precipitation, and
stream flow. Here, temperature and precipitation constitute basin averages out of
the E-OBS gridded data set.
As in 1., but temperature and precipitation are used at the E-OBS grid scale (0.25
degree in longitude and latitude) without spatial averaging.
As in 1., but the regression model is applied to 66 gauged subcatchments of
the Rhine basin. Forecasts for these subcatchments are then simply summed and
upscaled to the area of the Rhine basin.
As in 3., but the forecasts at the subcatchment scale are additionally weighted in
terms of hydrological representativeness of the corresponding subcatchment. |
|
|
|
|
|