————————————————–
The difference histograms of the downscaled results between SVATARK
versus SVRB and SVRK are also shown in Figure 4. The three groups of
histograms do not indicate large differences, perhaps because the same
trend model was employed in all three downscaling methods. For the SVRB
case, the minimum differences range from -0.068
m3·m-3 to -0.033
m3·m-3 and the maximums range from
0.027 m3·m-3 to 0.036
m3·m-3, while the differences for
SVRK span from -0.063 m3·m-3 to
0.085 m3·m-3.
Although only three days of downscaled SSM predictions are presented in
Figure 4, it is evident that the three downscaling approaches generate
fine-resolution predictions with similar numerical distribution, which
is reflected in the similar performance of the cumulative distribution
functions (CDFs) derived from thirty-six-month data (Figure 5).
Moreover, as found in the visual comparison, the CDFs and density plots
of the three downscaled predictions match well with those of the 25-km
SSM product (Figure 5). When comparing the differences of the density
curves between 25-km SSM and the three downscaled SSM, SVATARK appears
to have the closest match. Over the study area, the downscaled SSM by
SVATARK is quite similar to the coarse SSM, both in spatial distribution
and values. The following validations further demonstrate its improved
performance.