First, the distribution (i.e., the percentage) of land cover classes in each region is computed.
Then we plot the histogram (Fig. 6) that represents the number of regions containing different
percentage range, e.g., [0, 10%), [10%, 20%) …, of each land cover class. Thus the user can
define COI by specifying the (combined) percentage of land cover classes within a region. For
example, based on the eight identified primary land cover classes, we defined four experimental
categories: 1) water/wetlands/pasture ( %); 2) grasslands/forest ( %); 3) crops/pasture
( %); 4) residential/rock/grasslands ( %). Fig. 7 shows some representative regions in
each category. An important criterion for building categories this way is to avoid significant
overlap between different categories while encompassing most of the regions in the union of all
the categories.