 |
| Titel |
A priori discretization quality metrics for distributed hydrologic modeling applications |
| VerfasserIn |
Hongli Liu, Bryan Tolson, James Craig, Mahyar Shafii, Nandita Basu |
| Konferenz |
EGU General Assembly 2016
|
| Medientyp |
Artikel
|
| Sprache |
en
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 18 (2016) |
| Datensatznummer |
250130874
|
| Publikation (Nr.) |
 EGU/EGU2016-11197.pdf |
|
|
|
|
|
| Zusammenfassung |
| In distributed hydrologic modelling, a watershed is treated as a set of small homogeneous
units that address the spatial heterogeneity of the watershed being simulated. The ability of
models to reproduce observed spatial patterns firstly depends on the spatial discretization,
which is the process of defining homogeneous units in the form of grid cells, subwatersheds,
or hydrologic response units etc. It is common for hydrologic modelling studies to
simply adopt a nominal or default discretization strategy without formally assessing
alternative discretization levels. This approach lacks formal justifications and is
thus problematic. More formalized discretization strategies are either a priori or a
posteriori with respect to building and running a hydrologic simulation model.
A posteriori approaches tend to be ad-hoc and compare model calibration and/or
validation performance under various watershed discretizations. The construction and
calibration of multiple versions of a distributed model can become a seriously limiting
computational burden. Current a priori approaches are more formalized and compare overall
heterogeneity statistics of dominant variables between candidate discretization schemes and
input data or reference zones. While a priori approaches are efficient and do not
require running a hydrologic model, they do not fully investigate the internal spatial
pattern changes of variables of interest. Furthermore, the existing a priori approaches
focus on landscape and soil data and do not assess impacts of discretization on
stream channel definition even though its significance has been noted by numerous
studies.
The primary goals of this study are to (1) introduce new a priori discretization quality
metrics considering the spatial pattern changes of model input data; (2) introduce a
two-step discretization decision-making approach to compress extreme errors and meet
user-specified discretization expectations through non-uniform discretization threshold
modification. The metrics for the first time provides quantification of the routing
relevant information loss due to discretization according to the relationship between
in-channel routing length and flow velocity. Moreover, it identifies and counts the
spatial pattern changes of dominant hydrological variables by overlaying candidate
discretization schemes upon input data and accumulating variable changes in area-weighted
way. The metrics are straightforward and applicable to any semi-distributed or
fully distributed hydrological model with grid scales are greater than input data
resolutions.
The discretization metrics and decision-making approach are applied to the Grand River
watershed located in southwestern Ontario, Canada where discretization decisions are
required for a semi-distributed modelling application. Results show that discretization
induced information loss monotonically increases as discretization gets rougher. With regards
to routing information loss in subbasin discretization, multiple interesting points rather
than just the watershed outlet should be considered. Moreover, subbasin and HRU
discretization decisions should not be considered independently since subbasin input
significantly influences the complexity of HRU discretization result. Finally, results
show that the common and convenient approach of making uniform discretization
decisions across the watershed domain performs worse compared to a metric informed
non-uniform discretization approach as the later since is able to conserve more watershed
heterogeneity under the same model complexity (number of computational units). |
|
|
|
|
|
|