|
Titel |
A test for the transferability of regionalized model parameters across calibration scales and locations |
VerfasserIn |
R. Kumar, L. Samaniego |
Konferenz |
EGU General Assembly 2012
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 14 (2012) |
Datensatznummer |
250066833
|
|
|
|
Zusammenfassung |
Several recent studies have advocated on improving evaluation procedures for testing the
reliability of distributed hydrologic models. The traditional approach of model evaluation in
calibrated conditions against daily stream flow has proven to be inadequate in identifying a
robust hydrologic model or a parametrization scheme among several competing ones, as
noticed in several model inter-comparison studies (see e.g. DMIP-I & II; Reed et al. 2004,
Smith et al., 2011).
In this study, we propose a model evaluation procedure aiming at testing the reliability of
a distributed hydrologic model for predictions beyond the calibrated conditions. Specifically,
our goal was to assess the effectiveness of the transferability of model parameters to scales
and locations other than those used during calibration. For this case study, we used a grid
based distributed mesoscale hydrologic model (mHM, Samaniego et al. 2010, WRR) and
show the effectiveness of two parameterization methods: one based on hydrological response
units (HRU) and the other based on the multiscale parameter regionalization (MPR)
technique.
mHM with both parameterization methods was set-up in 45 southern German river basins
covering a wide range of physiographical characteristics and drainage area (100Â km2 to
12Â 700Â km2). Model simulations were carried out at four spatial resolutions: (2, 4, 8,
16)Â km in the period from 1980 to 2008. At each scale, a set of free parameters of
both methods were estimated using a dynamically dimensioned search algorithm
(DDS).
Both parametrization methods provided nearly similar acceptable performances for daily
discharge simulations at all four modeling scales as long as their free parameters were
calibrated at each scale. A significant deterioration (up to 40%) in performance of the HRU
method was, however, noticed when its free parameters calibrated at a given modeling
scale were shifted to another scale. The MPR method, on the contrary, exhibited a
quasi scale-invariant performance, meaning that the loss in the model performance
while transferring the free parameter across scales was almost negligible (less than
2%).
The performance of both parametrization methods for daily discharge simulations in all
investigated basins was nearly similar as long as their free parameters were calibrated for
each basin at a given scale. The Nash-Sutcliffe efficiency during the evaluation period
(1988-2008) varied between 0.62 to 0.92 across basins. However, the results of the cross
validation test at interior locations and non-nested basins (i.e. both assumed ungauged for
testing, also known as proxy basin test) corroborated the superiority of the MPR over HRU
method.
These results, therefore, supported the research hypothesis that the traditional way of
evaluating model performance at calibrated conditions will fail to identify a robust
parametrization technique. Whereas the proposed evaluation method helped to identify the
most reliable parametrization technique for mHM. |
|
|
|
|
|