516
array set out in rows and columms which ex
presses the number of observations assigned
ta a particular land caver type (Landsat)
relative ta the actual land caver (reference
data). The values alang the diaganal repre
sent the percentage af carrectly classified
pixels fer each class and aff diaganal values
represent errars af cammissians and amissians.
Mapping accuracy M af class I (Kalensky and
Wightman 1976)
«1 '*>' N^7 x 10054
Where - mapping accuracy af class I
N ■ number af carrectly classified
1 pixels in class I
E_ = number af erraneaus pixels in
1 class I
averall Landsat classification accuracy:
Tatal carrect pixels
Tatal pixels
The tables 2-5 summarize the result af the
classification far bath areas and the aver
all summary af the results abtained by app
lying twa classification schemes is given
in table 6.
Table 6. Summary af the accuracy af results
far the twa classification schemes
Area-1
Area-2
6 classes = 44.29%
5
classes =53.39%
Class!-
fication
2 classes = 85.92%
2
classes =79.60%
6 classes = 43.70% 5 classes =62.85%
Recla
ssifi
cation 2 classes = 84.29% 2 classes =85.73%
Landsat MSS system has its limitations. How
ever, it does provide definitive information
about the location af ‘closed forest*. It is
expected that finer resolution satellite data
such as that from the Thematic Mapper and
SPOT would provide mare specific information.
ACKNOWLEDGEMENTS
This w©rk was carried out during postgraduate
studies in the Department af Geography Uni
versity af Reading, England; the author is
grateful ta Dr.J.R.G.Tawnshend, Director,
NERC unit far Thematic Information Systems
at University of Reading for his valuable
suggestions. The author is also thankful to
the Commonwealth Scholarship Commission in
the U.K and the Forest Department, Government
af Manipur, India, respectively, for their
financial support and sponsorship.
REFERENCES
Champion, H.G. & S.K.Seth 1968. Revised
Survey ©f Forest Types af India, New Delhi,
Govt.of India Press.
Justice, C. & J.R.G.Tawnshend 1982. A compa
rison af unsupervised classification pro
cedures of Landsat MSS data for an area af
complex surface conditions in Basilicata,
Southern Italy, Remote Sensing af Environ,
12:407-420.
Kalensky, Z. & J.M.Wightman 1976. Automatic
forest mapping using remotely sensed data.
Prec. 16th IUFRO world congress, Division
6, Norway: 115-135.
Perssan, R. 1977. Scape and approach to world
forest resource appraisals. Res. Rep. No.
23, Royal College of Forestry, Stockholm.
Richards, P.W. 1973. The tropical rain forest.
Scientific American 229; 58-68.
Schawengerdt, R.A. 1983. Techniques for image
processing and classification in remote
sensing. New York, Academic press.
5, RESULTS AND DISCUSSION
The application af computer aided analysis
af Landsat MSS indicated that use af super
vised technique did not yield acceptable
accuracy as far as tropical forest caver cla
ssification is concerned. However, closed
forests which are among the mast productive
in terms of primary production and source
af ‘gene pool' can be identified with a high
degree af accuracy (upta 90%) even in areas
af rugged mountainous terrain and spectrally
complex forest caver tapes. This capability
itself is very useful to forest planners con
cerned with the large area inventories in a
region with a large inaccessible forests.
These are the areas about which very little
information is available. Contextual consi
derations i.e. reclassification scheme, mo
destly improved (6% to 9%) the classifica
tion accuracy in areas of homogeneous cover
types (Area-2) but the results were not en
couraging for the areas of heterogeneous
cover types (Area-1)• Also in mountainous
terrain, shadow is one of the major sources
of misclassification as in Area-1 roughly
15% of the pixels were affected by shadow.
In Area-2 due to subdued relief only 2% of
the pixels were affected by shadow. In con
clusion, in areas affected by shifting cul
tivation due to complex spatial distribu
tion and intermixing of cover types, the
