4. DISCUSSION
In this study, only local variations of three parameters – initial soil
moisture, hydraulic conductivity, and the exponent of the VIC model –
are tested independently. However, the interactions among these
parameters are not explored herein. Global sensitivity analysis, such as
the Morris method used in the previous study (Li et al., 2021b), can
measure the variation of each parameter relative to other parameters, so
it provides a clearer picture of the parameter interactions. Needless to
say, initial soil saturation state is the dominant controller for the
differences between the simulations with and without re-infiltration
process. When the soils are fully saturated, the with and without
re-infiltration scenarios are almost identical if other parameters are
the same. Combined with our previous study that underlies the importance
of infiltration and initial soil moisture for flood inundation modeling,
we highly recommend taking into consideration the initial soil moisture
state, as it has not been well-recognized in the hydraulic model
community. This can be achieved via three ways: 1) warm up the model for
a relatively long period prior to the simulation period (Chen et al.,
2020); 2) parameterize the initial soil moisture and calibrate it,
similar to the way we treat initial in-channel water depth (Xue et al.,
2013); 3) approximate it using observations or other model simulations,
like what has been done in the real case study in Section 3.2 (Flamig,
Vergara, & Gourley, 2020). The first approach is ideal because it
eliminates uncertainties in parameterization (such as equifinality) or
error propagation from observations/simulations to models; it is,
however, the most computationally expensive approach for hydraulic
modeling compared to the other two. Approach two and three are more
pragmatic, while both inherit uncertainties or errors. We prefer the
third approach if the data source is found to be trustworthy. For
instance, in our case study, we used the simulated soil moisture product
from the operational CREST/EF5 model which shares the same land surface
processes as the CREST-iMAP.
The results relating to different rainfall events are not factored in
this study, yet different event characteristics interact with soil
infiltration dynamics, thereby causing differences between the two
schemes (Zhang et al., 2020). One could envision the differences
increase with rainfall rates if they did not exceed the maximum
infiltration capacity and the soils are not saturated. However, once
reaching the maximum infiltration rates or soil saturation, the
differences are unlikely to further enlarge. Other parameters regarding
terrain characteristics are also effective in re-infiltration. For
instance, an increase in slope will leave less room for surface water to
re-infiltrate, which explains why re-infiltration compromises its
importance in hillslope hydrology (Corradini et al., 2002; Zhang et al.,
2020).