The variation of the accuracy indices with VZA class showed that OA was hardly influenced by VZA and was greater than 90% in all VZA classes (91.9%-98.6%), while PA and UA tended to decrease as VZA increased.
(1995) is based on a static threshold method and does not account for variability in the reflectance of snow cover due to properties of the snow (grain size, moisture content, impurity, age, etc.), terrain effects, or geometric factors such as SZA and VZA, this algorithm may cause errors such as misclassification.
Most overestimation error occurs with high VZA, more than 25[degrees] from the nadir.
For this purpose, we selected two pixels, one observed with low VZA (blue circle) and the other with high VZA (red circle), as shown in Figure 10(b).
Figures 11(a) and 11(b) show a plotted reflectance profile observed by satellite and selected snow reflectance profiles from the spectral library under different SZA and DEM conditions (left) and D with W indicated by several black arrows and circle (right) at each sample pixel with different values of VZA (16.16 and 51.23[degrees], resp.).
This analysis shows that DWW is more useful than static threshold methods when the reflectance of snow cover exhibits variability due to conditions such as SZA, VZA, and the state of snow at observation time.
When evaluating accuracy based on SZA and VZA classes, OA was greater than 88% for all SZA classes and over 91% for all VZA classes.
Caption: FIGURE 9: Histogram of PA, UA, and OA in different viewing zenith angle (VZA) classes.