 |
| Titel |
Searching for a robust parameter estimation strategy for large river basins |
| VerfasserIn |
Luis Samaniego, Oldrich Rakovec, Rohini Kumar, Juliane Mai, Sabine Attinger, Matthias Cuntz, Martin Schrön, Stephan Thober, Matthias Zink |
| Konferenz |
EGU General Assembly 2015
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 17 (2015) |
| Datensatznummer |
250107679
|
| Publikation (Nr.) |
 EGU/EGU2015-7391.pdf |
|
|
|
|
|
| Zusammenfassung |
| Large scale hydrologic models as well as land surface models require a large number of
fine-tuned effective parameters per grid cell to be able to accurately predict variables
of interest (e.g., streamflow, soil moisture) across locations and scales. Finding
those sets of parameters has been an active area of research in hydrological sciences
during the last decades. Up to date, many approaches exist but none is entirely
satisfactory. This problem is drastically enhanced in large scale river basins due to the
non-linear computational costs associated with increasing resolution and basin
area.
In this study we demonstrate that the Multiscale Parameter Regionalization technique
(Samaniego et al. 2010, WRR) applied to the mesoscale hydrologic model mHM 5.2
(www.ufz.de/mhm) is an effective method to find quasi scale invariant parameter sets (i.e.,
global regionalization or regularization multipliers) over 250 Pan-European river basins
varying from 100 km2 to 500Â000 km2. Two different parameter estimation strategies, single
vs. multi-basin, are tested. In both cases, the Shuffled Complex Evolution algorithm is used
to estimate parameters using the Kling-Gupta efficiency metric as an objective
function.
Both single and multi-basin calibration strategies are tested with a number of
performance metrics against observed streamflow, remotely sensed soil moisture (SM) and
total water storage (TWS). The streamflow records are obtained from the GRDC
repository (www.bafg.de/GRDC). The SM and TWS products are ESA-CCI with a
spatial resolution of (0.25 x 0.25)° (www.esa-soilmoisture-cci.org) and GRACE
(Landerer and Swenson 2012, WRR; www.nasa.org) with a spatial resolution of (1 x
1)°.
In most cases, the single-basin optimisation strategy is the best alternative for a given
basin but its transferability can not be guaranteed. The multi-basin technique is at least as
good as the best cross-validated results obtained for the single-basin calibration using only
streamflow. In general, 50% of the basins exhibit a NSE larger than 0.5 with the multi-basin
strategy. An advantage of the multi-basin optimization technique is that it generates
physically plausible fields of distributed variables such as soil moisture, which exhibit spatial
continuity rather than a patchy distribution generated by numerical artifacts during the
single-basin optimization. The multiscale parameterization technique allows to assimilate
disparate information of available data sets (e.g., TWS, SM) on their native resolution. Cross
validation experiments show that TWS in addition to streamflow in the objective
function lead to the best results in terms of predicting observed streamflow. On the
contrary, SM used with streamflow can not reproduce observed runoff dynamics
adequately. From these findings it can be concluded that the available soil moisture
product has limited information for inferring model parameters in large river basins. |
|
|
|
|
|
|