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.