The SVVIMVC product developed and utilized as a part of this study
played an important role in the classification of Natural Vegetation and
Agriculture classes. The index is particularly effective in enabling MVC
compositing from a large number of images with substantial data
gaps, and providing coherent images having high informational value.
Further, the SVVI MVC utilizes all cloud-free pixels available during
the compositing period in order to detect the presence of agriculture.
Spectral and texture values from this product enabled the discrimination
of Forest, Secondary Forest, Savanna, and Agriculture classes.
While the Hansen et al. (2013) global forest cover product is found to
be substantially affected by SLC-off data gaps within our study area,
the SVVI MVC brightness and texture products are less influenced by
SLC-off data gaps and yielded map products with little SLC-off patterning
inmost areas. However, clouds have high SVVI values, so it is necessary
to ensure that all cloud-masking procedures are completed
successfully before creating MVC products from SVVI images. We observed
a few instances where no clouds were indicated in the Fmask
product but substantial cloud cover was present in the image. Further,
the Fmask procedure often failed to detect very small clouds (e.g.,
20 ha in size or smaller) and this initially caused misclassification of Forest
as Agriculture, among other errors. Image scenes containing undetected
clouds in the Fmask products were identified by an analyst and
removed from the MVC process.