Geography Reference
In-Depth Information
overestimated by the model in the south-eastern part of
Sweden, the driest region in the country. An evaporation
correction factor was therefore introduced for this par-
ticular region as part of this initial water balance adjust-
ment. The overall water balance was continuously
checked and revised during the proceeding calibration
steps.
idea is to follow the water and nutrient fluxes through the
landscape and to fix coefficients where data are available
for a specific process or segment of processes (see
Figure 11.89
). Each step downstream in the model code
included some reconsideration about the chosen parameter
values as part of an iterative procedure. Water and nutrient
concentrations were also calibrated iteratively, which fur-
ther limited the degrees of freedom for parameter values.
Yet, as described above, the same parameter set was used
for the whole country and no river-basin specific calibra-
tion was made.
Inclusion of in-situ knowledge (see
Chapter 3
)
Lakes and dams strongly affect all downstream basins in a
river network. Simplified rating curves for 50 unregulated
lakes were estimated for use in the HYPE model, based on
observations from SMHI databases. Existing rating curves
were applied wherever available. For regulated lakes and
hydropower dams, regulation volumes and average out-
flow were taken from the SMHI Water Archive. These
were used to construct a seasonal variation for each indi-
vidual reservoir. For some 50 important reservoirs, a spe-
cific rating curve for the spillways was constructed based
on observations of discharge and water levels. For the
remaining reservoirs a generalised spillway module was
constructed based on the specific curves.
Expert judgements on dominant flow paths and response
to events
During the procedure, internal model variables (such as
flow separation in surface runoff, tile drainage, discharge
from various soils) were checked visually in a test bench of
catchments, to avoid unrealistic model behaviour due to
parameter setting (see
Chapter 10
). For instance, surface
runoff does not usually occur during summer in Sweden
and should thus not be dominant in the model during this
period. Moreover, in cold climates the snowmelt event is
the dominant characteristic of the flow hydrograph and
hence flow peaks were followed through river reaches in
order to explore whether the transit times were realistic in
the model setup.
Parameters linked to HRUs and simultaneous calibration
for a large model domain, representing different
characteristics
The first step was a definition of soil characteristics of
HRU, e.g., soil layers, soil depths, combinations of soil
type and land use. Then, calibration was performed on
groups of similar representative gauged basins, trying to
isolate key processes and characteristics. Calibration was
made regionally (see
Section 10.4.4
) on a multi-basin
level for the whole domain. It was assumed that differ-
ences in physiographical characteristics and forcing data
were sufficient to account for spatial variability, while
model coefficients were kept constant. Hence, a clay soil
in southern Sweden has the same parameter setting as in
the northern part of the country. The parameters so
obtained were then applied everywhere, also for ungauged
basins.
Multi-variable validation
To judge model credibility, other observed (orthogonal)
variables, i.e., other than river discharge (and concentra-
tions), were used (see
Section 10.4.5
). For instance, these
included observations of snowpack, groundwater fluctu-
ations and water levels in lakes (Arheimer et al.,
2011
).
Finally, the overall model was validated against independ-
ent observations from gauges that had not been included
during the model calibration, to examine model perform-
ance in ungauged basins.
Results
Assessment of the model performance was based on the
statistical criteria of NSE (daily values) and relative error
of water volume (RE). Absolute RE was on average less
than 10% and the median NSE for the whole country was
0.67. The maximum NSE was 0.94 (period: 1999
-
2008).
The S-HYPE model setup is not dependent on site-
specific calibration as previous models for Sweden were
(e.g., Arheimer and Brandt,
1998
; Arheimer,
2003
). The
purpose was to make the results in ungauged basins more
reliable. The model was therefore also checked separately
for the gauging sites that had been used in the regional
calibration procedure, and other independent sites, which
could then represent ungauged basins. This validation was
Stepwise, iterative calibration of parameter groups
Calibration of the model was carried out manually,
following a stepwise approach taking one process at a time,
to avoid equifinality and so that errors that were incurred in
some model processes are not compensated by introducing
errors in other parts of the model. Hence, groups of param-
eters responsible for certain flow paths or processes (e.g.,
soil water holding capacity) are calibrated first, after which
another group of parameters (e.g., river routing) are cali-
brated. As the model concept follows the flow paths, the
headwaters were calibrated first, then streams, lakes, rivers
and finally the overall outlet to the sea was checked. The
Search WWH ::
Custom Search