4 Results and discussion
4.1 Downscaled 1-km SSM
Figure 4 displays the 25-km SSM images in comparison with 1-km
downscaled SSM predictions by three different models (i.e., SVATARK,
SVRK and SVRB) for May 1 of 2011, July 20 of 2013 and September 22 of
2015. It can be inferred that the 1-km downscaled results provide more
detailed information and variations of the SSM spatial distribution
within each 25 km × 25 km grid. The SSM data at fine spatial resolution
can improve the characterization of the spatial variability of the SSM,
which are useful for filling the gap between low-spatial-resolution SSM
satellite observations and the needs of catchment-based or regional
hydroecological studies. In the downscaled SSM images, the maximum and
minimum values of SSM are shown in blue and red, respectively. The blue
areas are near the water bodies and in areas with low elevation. Besides
surface water, the negative correlation with elevation is another
primary factor affecting the spatial distribution of SSM under a
sub-frigid zone. The results from the proposed SVATARK method showed
spatial patterns that were similar to those of the 25-km SSM, while SVRK
and SVRB produced smoother downscaled results. The coherence of the ATAK
predictions ensures that the average of the disaggregated predictions is
equal to the original areal data, and confers the downscaled SSM of
SVATARK a continuous pattern. Visual comparison of the downscaled SSM
products confirmed that the failure of SVRK to predict extreme SSM and
the failure of SVRB to properly capture high SSM were influenced by the
kriging and bilinear interpolations. SVATARK exhibits better results in
modeling extreme SSM (both maximum and minimum) than the other two
downscaling methods.