————————————————–
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.