 |
| Titel |
Reduction of uncertainty of hydrological modelling using different precipitation inputs |
| VerfasserIn |
T. Pluntke, D. Pavlik, C. Bernhofer |
| Konferenz |
EGU General Assembly 2012
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 14 (2012) |
| Datensatznummer |
250067046
|
|
|
|
|
|
| Zusammenfassung |
| Precipitation is one of the main sources of uncertainty in hydrological modelling, due to its
high temporal and spatial variability. A dense network of rain gauge stations or a combination
with, e.g., radar data is needed to account for the - in comparison to other climatic elements -
pronounced variability. The density of existing station-networks is low in many countries
worldwide. Alternative approaches that use additional information should be applied to
improve the estimation of areal precipitation.
Within the project „International Research Alliance Saxony“ (http://www.iwas-sachsen.ufz.de/),
one subproject aims at a system analysis of a meso-scale catchment of the Western Bug
in Ukraine. Effective and sustainable measures have to be identified to improve
the water quality of the Western Bug under the premise of upcoming changes of
climate, land use and socio economy. An exact quantification of the water balance
is needed as a pre-requisite for a matter balance. This contribution demonstrates
possibilities to reduce the uncertainties of water balance modelling of the catchment
Kamianka-Buzka/ Western Bug (2560Â km2) by applying and combining alternative
precipitation inputs.
Available precipitation data were undergone an extensive quality check and were bias
corrected. The Soil and Water Assessment Tool (SWAT, http://swatmodel.tamu.edu/)
was used for water balance modelling. By default, meteorological observations are
incorporated into SWAT using the station that is nearest to the centroid of each
sub-catchment. Two alternative precipitation inputs were applied: 1) Data of 20 stations were
regionalized using kriging methods. 2) The output of the Regional Climate Model
CCLM that was set up for the region was used. After a pre-calibration of the model,
three models - having different precipitation inputs - were set up and calibrated
independently applying the auto-calibration procedure Sequential Uncertainty Fitting
(Abbaspour et al. 2004). The performance of the models was evaluated with the
Nash-Sutcliff-Efficiency coefficient (NSE) and the R2 between observed and modelled
runoff.
The model Stations performed better (R2/NSE: 0.66/0.61) than CCLM and Regionalized
(0.54/0.54 and 0.57/0.53). Uncertainty of the hydrologic modelling (POC and d-factor) could
not be reduced applying the alternative models. A promising method to improve the model
performance and reduce the uncertainty is model averaging. Two model averaging methods
were tested: arithmetic mean of the ensemble and a weighted mean (depending on NSE). The
results show that the model performance could be improved (R2/NSE: 0.67/0.67)
and the uncertainty reduced. Differences between the applied model averaging
methods were marginal. Although not all observations could be reproduced, neither
by the single models nor the ensemble averages, it was illustrated that combining
different precipitation inputs improved the hydrologic predictions. Further calibration
runs as well as the application of Bayesian Model Averaging are envisaged as next
steps.
Reference:
Abbaspour, K. C., Johnson, C., & van Genuchten, M. T. (2004). Estimating uncertain
flow and transport parameters using a sequential uncertainty fitting procedure. Vadose Zone
Journal, (3), 1340-1352. |
|
|
|
|
|
|