Forest was classified for 2000 and 2010 using three data products: 1)
SVVI MVC images, 2) SVVI texture MVC images, and 3) tree canopy
cover/change products by Hansen et al. (2013) (Fig. 6). The SVVI texture
MVC products from the 2000 or 2010 time series were derived by: 1)
computing 3 × 3 focal SD (i.e., texture) for each SVVI image of the
time series, 2) stacking the texture images from the time series and
computing the maximum time series value for each pixel (stack max),
and smoothing the stack max image using a 5 × 5 focal mean filter
(making texture values more consistent for classification). For 2000, a
given pixel was classified as Forest if: 1) the SVVI texture MVC was
less than or equal to 40 or the Hansen et al. (2013) tree canopy cover
product value was greater than or equal to 66% and 2) the SVVI MVC
image value was less than or equal to 150. The smoothed SVVI texture
MVC well discriminated Forest in most cases. However, some areas
characterized as Savanna had similar smooth (low) texture, and therefore
it was necessary to separate those savanna type areas using the
SVVI MVC which has moderately high values for Savanna, and low
values for Forest (≤150). For 2010, the same procedure was utilized.
However, forest gain or loss (Hansen et al., 2013) was added or
subtracted, respectively, from the 2000 tree cover product so that
2010 conditions would be represented in the updated tree cover
product.