 |
| Titel |
Structural improvement of a simple rainfall-runoff model based on time-variable parameters |
| VerfasserIn |
Mark Honti, Christian Stamm, Peter Reichert |
| Konferenz |
EGU General Assembly 2011
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 13 (2011) |
| Datensatznummer |
250053305
|
|
|
|
|
|
| Zusammenfassung |
| Hydrological models are used in water quality and catchment modeling as backbone model
components describing the most important drivers of contaminant transport. Creating an
integrated management framework is only possible if the building model blocks are robust
and have a coverable data demand. The combination of simplicity and robustness can be
achieved when the model structure is adapted to local circumstances based on a data-driven
modification procedure.
Large environmental model frameworks are often constructed to deliver predictions on
the future behavior of the inspected system in case of an altered external forcing (like climate
change or socio-economic development). As the uncertainty present in separate building
blocks propagates through the system, the reliability of final predictions can be only
estimated properly if all identified sources of model uncertainty (input, structural/parameter
and output uncertainty in general categories) were accounted for. This is especially
true for those model blocks like hydrology, which provide crucial inputs to other
components.
Our aim was to provide reliable hydrologic inputs for a complex model system together
with proper estimates of uncertainties. We started with the evaluation of the logSPM
conceptual rainfall runoff model (Kuczera et al. 2006) on two small test-catchments (46 and
117 km2) on the Swiss Plateau. Parameter optimisation was done using an autoregressive
likelihood function that accounted for measurement errors in precipitation data based on a
stochastic rainfall multiplier concept. The model provided an acceptable goodness of fit (0.85
and 0.67 in terms of Nash-Sutcliffe index for the Mönchaltorfer Aa and the Gürbe,
respectively), but the structure of residuals suggested that there is place for structural
improvement.
The initial model was tailored to achieve a better representation of the measured
discharge series with introducing time-variable model parameters and analysing their impact
on the likelihood. Full resolution time variability was not applied since even the
original model was able to simulate short sections (typically a few weeks) of the
measured discharge series with almost perfect fit (Nash-Sutcliffe index above 0.95).
Instead, parameters were varied from one storm event to the other. Parameters were
varied individually, keeping the others fixed. This way their individual capability for
increasing the likelihood could be analysed. The presented approach helped to avoid
introducing unnecessary detail in the model structure wherever it was not absolutely
essential.
The simulation and prediction uncertainty of the modified model structure with
time-invariant parameters will be assessed finally with a stochastic model bias description
technique. We will present the results of this final step with realistic uncertainty bands for
predictions. |
|
|
|
|
|
|