those in the final image (Table 1). Because the plots 
used for this determination are the same as were used 
in some steps of the classification process this is not 
a true test of classification error. 
The classification did a good job of 
discriminating tree cover from other cover types, our 
most basic objective. The exposed soil classification 
included all the herbaceous crops and most pastures, 
while the various forest classifications (dense forest, 
restinga, and mangrove) included nearly all forested 
plots except the earlier stages of regrowth and one 
plantation plot of undetermined tree density. Urban 
areas and some of the miscellaneous soil bank and 
related plots were not reliably categorized. 
Recognition of these areas may have been 
Table 1. Classification at field plots versus in the image 
complicated by scattered trees, but field observations 
indicate such extremely disturbed areas do not 
represent a large part of the region and such 
misclassifications are not particularly problematic for 
our purposes. Misclassification at two of the twelve 
pasture plots may have arisen from remnant trees 
although this has not yet been confirmed from plot 
records. 
The classification did not reliably distinguish 
between dense forest, restinga, and mangrove, 
although limited plots were available for the latter 
two types. Distinguishing between the different 
forest community types was a lower priority for the 
project, but improving discrimination of important 
vegetative communities such as salt marsh and 
mangrove would help both 
conservation and ecotourism 
planning efforts. 
  
  
  
classification in image The most troubling 
misclassification ^ concerned 
classification at bare soil dense restinga man- beach outside @ the woody crops of manioc 
field plots forest grove area and particularly banana. A 
ae 3 number of plots -------------- short shrub, manioc is often 
grown in small plots. 
urban 2 2 3 Difficulty in its classification 
pasture 10 2 may arise from mixed pixel 
crops conditions. As a subsistence 
corn 2 crop it should usually be 
ginger 6 cultivated near other crops 
manioc 1 2 and thus may not seriously 
banana 1 5 confound indications of 
transition forest femoteness. «ii 
: Misclassifying banana 
capoerinha S 2 - 
trees as forest is understand- 
capoeira 3 ; 
rit ii | 2 able for relatively mature 
capoeirdo (late) plantations given their high 
forests ; canopy height and cover. 
plantations l 3 However, it is an important 
secondary dense 4 1 | limitation of the current 
primary dense 5 coverage. Banana is an 
restinga 3 important crop in the region, 
mangrove 2 1 often cultivated in large 
vérzea (swamp) 2 plantations near the ends of 
litoral arenosa 2 roads. Their proximity to 
sait marsh 2 existing roads may help in 
other interpretation of conservation 
; implications and remoteness 
uncategorized 3 1 ; . . 
: but will require further 
soil banks 3 4 1 ; oer : 
wm | | attention in future reclassi- 
femoved sol fication efforts. 
recently ] 
Note: Several categories in the image had no plots and are not shown. 4.3 Coverages and Maps 
“Outside area” refers to plots in bay or otherwise outside the classified 
area. Coverages for each 
  
76 
cover type 
unified co 
steps. The 
cover type 
forest as t 
and overla 
cover type 
Rasterizati 
be perfor 
environme 
Th 
from the | 
about land 
The combi: 
overlain \ 
restinga, h 
identificati 
minimal o 
beaches an 
shown in s 
and topog 
provided 1 
annotated [ 
A 
composite 
broader adj 
both of th 
color limits 
Simpler 
distribution 
As 
a case stuc 
project’s s 
utility of tl 
GIS datab: 
capacities 
evaluated | 
work might 
In| 
study migh 
field seaso 
would hav: 
concurrent 
plots would 
diversity wi 
A more int 
seems pre 
preliminary 
done first, 
then supen 
truthing, an