development and mangroves had receded approximately 1 km to
the east. By 2000s (ETMþ),
Only a ring of mangrove at the shoreline remained. The evidence
of development is apparent with the building of diked areas and
canals as the forest was removed. This area falls outside of the
managed forest reserves and contrasts sharply with the mangrove
forested areas to the south and east, which remained generally
unchanged during the same period.
Again, the net mangrove loss over thewhole of the Sundarbans is
about 1% as the numerous areas of loss are counter-balanced by
areas of gain. Most of this gain is found in areas where new land
formed through deposition has become vegetated. One of the exceptions
is an area of afforestation located in the Jilla forest block on
the northern forest boundary of the India side. This area of
approximately 400 ha was completely degraded in 1975, but had
been re-vegetated by 1989 andwas generally indistinguishable from
surrounding forested areas in a remote-sensing image by 2000s.
Overall accuracy of 86%, 85%, and 79% were achieved for
2000s, 1990s, and 1970s classification with the Tau coefficient of
0.85, 0.83, and 0.76, respectively. The tau coefficient for the year
2000, for example, indicates that our classification systems produce
a map on which 85% more pixels were classified correctly
than would be expected by random assignment. This means that
for this classification, we were correct 85% of the time. Confusion
arose in discriminating flooded and water bodies, and nonmangrove
and barren lands classes. Mangrove class was relatively
well classified.
The canopy closure layers derived from NDVI measurements for
the three mosaics showchanging patterns of forest condition in the
Sundarbans. The pattern of healthy upper-story vegetation is
different in the different era classification results. Therefore, the
least healthy areas in 2000s are different from the least healthy
areas of 1990s. Furthermore, the pattern of relatively unhealthy