 |
| Titel |
Examining the effects of parameter regionalization schemes on parameter transferability on large basin sampling |
| VerfasserIn |
Oldrich Rakovec, Naoki Mizukami, Andrew Newman, Stephan Thober, Rohini Kumar, Andrew Wood, Martyn P. Clark, Luis Samaniego |
| Konferenz |
EGU General Assembly 2017
|
| Medientyp |
Artikel
|
| Sprache |
en
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 19 (2017) |
| Datensatznummer |
250148172
|
| Publikation (Nr.) |
 EGU/EGU2017-12406.pdf |
|
|
|
|
|
| Zusammenfassung |
| Assessing model complexity and performing “seamless” continental-domain model
simulations (e.g., model parameters yielding good performance across entire domain) is a
challenging topic in contemporary hydrology. This study presents a large-sample
hydrologic modeling effort to examine the effects of parameter regionalization schemes.
Two hydrological models (mHM, VIC) are set up for 500 small to medium-sized
unimpaired basins over the contiguous United States for two spatial scales: lumped and
12km grid. For parameter regionalization, we use the well-established Multiscale
Parameter Regionalization (MPR) technique for both models, with the specific goal of
assessing the transferability of model parameters across different spatial scales (lumped
basin scale to distributed), time periods (from calibration to validation period), and
locations.
In terms of the scale transferability, evaluation of global model parameters at finer scale
based on calibration at coarse scale improves the KGE performance (mainly due to the
variance related term). Loss in model performance in temporal transferability is independent
from model complexity (i.e., lumped vs. distributed). Finally, we show that although the
parameter regionalization is crucial for parameter transferability to un-gauged locations,
there still remains room for improvement especially for the mean and variability
in streamflow. We present possible strategies to resolve this issue, including (1)
assessing the importance of more detailed information on the soil data (STATSGO vs.
SoilGrids), and (2) applying more advanced selection criteria for training MPR global
parameters. |
|
|
|
|
|
|