Accumulated confidence layer for the classification where brightness is proportional to confidence.
On the consistency scale described in the text, numerical values range from 0.0 to 10.0 in
nonoverlapping regions and 0.0 to 40.0 in regions where up to four individual scenes contribute
Relationship of Accumulated Confidence and User's Accuracy
The data set of 5300 five-by-five-pixel reference sites was used to investigate the relationship
between accumulated confidence and user's accuracy. For each pixel, the appropriate reference
class was compared to the assigned class of the final LC product.
Confusion matrices were generated for pixels grouped according to number of contributing
scenes and accumulated confidence. User's accuracies of both the forested and nonforested classes
were computed for each matrix. Figure 10.9 shows the relationship of user's accuracy vs. accumu-
User s Accuracy vs.Accumulated
Plot of user's accuracy vs. accumulated confidence for forested and nonforested reference sites
located in areas where two scene classifications were available. The results indicated that classi-
fication confidence based on consistency monotonically increases with increasing user's accuracy
and therefore is a useful indicator of the latter.