![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Evaluation of medium-range runoff forecasts for a 50 km2 watershed |
VerfasserIn |
David Kneis, Gerd Bürger, Axel Bronstert |
Konferenz |
EGU General Assembly 2010
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 12 (2010) |
Datensatznummer |
250034427
|
|
|
|
Zusammenfassung |
Forecasting of medium-range runoff for small mountainous catchments is a particular
challenge for a number of reasons: (1) Reaction times are short, thus hydrological forecasts
necessarily rely on the uncertain output of numerical weather prediction (NWP) models, (2)
There is a high sensitivity to positional errors in the quantitative precipitation forecast
(QPF) as well as to errors in quantitative precipitation estimation (QPE), (3) Given
that rainfall is spatially variable, small catchments are generally more likely to
experience extreme areal precipitation than larger river basins, (4) Runoff-coefficients
are typically high and flood events originate from both heavy rainfall or/and snow
melt. Nevertheless, there is a demand for medium-range hydrological forecasts for
small catchments which can be used for the purposes of early warning and the
proactive management of multi-purpose reservoirs. We studied the quality of runoff
forecasts for the Lehnmühle Reservoir watershed (50Â km2) located at the border
between Germany and Czech Republic (Weißeritz River). For this catchment, we
set up the conceptual, semi-distributed hydrological model LARSIM which is in
operational use at many flood forecasting centres. LARSIM is typically run as a
continuous model with stream flow assimilation enabled during the simulation
period. The usual time step of in- and output is 1Â h. We produced medium-range
forecasts (actually ‘hindcasts’) of runoff by driving the calibrated hydrological
model with NWP data of the past. For the period 2000–2008, we computed daily
forecasts with a maximum lead time of +120Â h based on the 51-member ensemble
prediction issued by the ECMWF (European Centre for Medium-range Weather
Forecast). Expanded downscaling, a variant of empirical downscaling, was used to
make the spatially coarse ECMWF forecast applicable to the target scale. For the
period 2005–2008, we also tried the deterministic COSMO-EU forecast (+78Â h
maximum lead time, 7Ã7Â km resolution) disseminated by the German Weather
Service as an alternative model input. The quality of the forecasts was assessed
by comparing the predicted change in stream flow ÎQ (not the flow rate itself)
to the corresponding observations using several measures of skill. According to
our results, the ensemble-based runoff forecast performs better than deterministic
forecasts for lead times greater than +48Â h. However, for the studied catchment both
the deterministic and the ensemble forecast are uncertain in such a way that they
cannot be recommended for the purpose of early warning or the dynamic control of
multi-purpose reservoirs. Even if a relative error in the predicted ÎQ of ± 30% is
tolerated, considerable fractions of the forecasts (in the order of 50%) turn out to be
under-predictions or false alarms. We can show that this is due to both the insufficient quality
of precipitation forecasts as well as deficits in hydrological modelling. We conclude
that further efforts to quantify and reduce uncertainties from multiple sources are
necessary to improve the quality of medium-range hydrological predictions for small
catchments. |
|
|
|
|
|