Accounting for dependencies in regionalized signatures for predictions in ungauged catchments
A recurrent problem in hydrology is the absence of streamflow data to calibrate rainfall–runoff models. A commonly used approach in such circumstances conditions model parameters on regionalized response signatures. While several different signatures are often available to be included in this proces...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Copernicus Publications
2016-02-01
|
| Series: | Hydrology and Earth System Sciences |
| Online Access: | http://www.hydrol-earth-syst-sci.net/20/887/2016/hess-20-887-2016.pdf |
| Summary: | A recurrent problem in hydrology is the absence of streamflow data to
calibrate rainfall–runoff models. A commonly used approach in such
circumstances conditions model parameters on regionalized response
signatures. While several different signatures are often available to be
included in this process, an outstanding challenge is the selection of
signatures that provide useful and complementary information. Different
signatures do not necessarily provide independent information and this has
led to signatures being omitted or included on a subjective basis. This paper
presents a method that accounts for the inter-signature error correlation
structure so that regional information is neither neglected nor
double-counted when multiple signatures are included. Using 84 catchments
from the MOPEX database, observed signatures are regressed against physical
and climatic catchment attributes. The derived relationships are then
utilized to assess the joint probability distribution of the signature
regionalization errors that is subsequently used in a Bayesian procedure to
condition a rainfall–runoff model. The results show that the consideration
of the inter-signature error structure may improve predictions when the error
correlations are strong. However, other uncertainties such as model structure
and observational error may outweigh the importance of these correlations.
Further, these other uncertainties cause some signatures to appear repeatedly
to be misinformative. |
|---|---|
| ISSN: | 1027-5606 1607-7938 |