513
ort No. 116,
.1 Development
:erns in Azare
1 experimental
systematic low
»y. Agricul-
n and Planning
se in Busia
Planning and
ort No. 118.
(1984). Land
y of National
:hnical report
:periments in
ising airborne
985-1 pp 9-13.
0). Land use
sts. In, New
land use and
Lications Ltd.
(1985). The
a disaster.
987-994.
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
Tropical forest cover classification using Landsat data
in north-eastern India
Ashbindu Singh
Indian Forest Service, Forest Department, Government of Manipur, Imphal
ABSTRACT i Landsat MSS data were used for mapping tropical forest ctver classes in two site
conditions of northeastern India. These are areas of rugged mountainous terrain and spec
trally complex forest cover types. A supervised classification procedure based on a minimum
distance to means algorithm was used in the analysis. Classified images were reclassified
using a 3x3 majority filter. A quantitative evaluation of the results were carried out to
determine the accuracy of computer-aided forest classification.
The computer-aided analysis of Landsat MSS data has shown that 'forest* and 'non-forest*
could be classified with a high degree of accuracy, but further break-up of the classes did
not yield satisfactory classification accuracy. However, it does provide definitive infor
mation about the location of 'closed forest'. The reclassification scheme, modestly improved
the classification accuracy in areas of homogeneous cover types but the results were not
encouraging for the aras of heterogeneous cover types.
1 INTRODUCTION
The tropical forest biome is, biologically
and ecologically speaking, the most complex
and diverse biome on earth (Richards 1973).
However, the location and extent of the area
under tropical forests are poorly known.
The published statistics are presented with
a spuriously high degree of precision but
possess a very low level of accuracy and
reliability (Persson 1977). Due to inacces
sibility of such areas and ruggedness of
the terrain in which tropical forests are
located the application of satellite data
offers only possibility of mapping forest
cover types in such regions with any degree
of regularity. Since Landsat MSS data are
available at regular intervals they may be
used to derive a data base for monitoring
tropical forest resources. The objective of
this study is to assess the suitability of
digital Landsat MSS data for tropical forest
cover classification and determine the le
vels of accuracy of computer aided forest
classification.
2 THE STUDY METHODOLOGY
2.1 The study area
The study was conducted in the northeastern
part of India. Two areas of 256 x 256
Landsat pixels-each labelled here as Area-1
and Area-2 were choosen for investigation.
Floristic composition of forests and topo
graphic features were different for both
the areas. Area-1 has been undergoing rapid
change due t© shifting cultivation, wereas,
Area-2 is ©nly marginally affected by this
practice. Area-1 is situated in the eastern
hills of Himalayas; the terrain is extremely
rugged. Dense subtropical evergreen forest
occur on high hills and moist deciduous fo
rest are found in the valleys and at lower
altitudes. The majority of moderately steep
to gentle slopes (10-30%) have been severely
affected by practice of slash and bum agri
culture.
Area-2 is located in outer parts of the Hi
malayas. This is an area with subdued relief.
The forests can be classified into tropical
semi-evergreen forest and secondary moist
bamboo brakes. The typical scrub formation
in this area is degraded stages ®f climax
high forests resulting from excessive human
interference (Champion and Seth 1968).
2.2 Definition of cover classes
A suitable definition of cover classes is
required in order to assign the ground sur
face conditions to specific classes. In thi^
study land cover rather than land use formed
the basis ®f classification. Stand density
and stand height were considered to be the
most important parameters' for defining the
classes. The major cover classes occuring
in the study areas are given in table 1.
2.3 Reference data set
For Area-1 and Area-2 forest types maps pre
pared with the help of black and white ae
rial photographs after field checking at
the scale ©f 1:50,000 and 1: 63,360, res
pectively, were used for generation of sam
ple sites. A 3 x 3 pixel observation was
chosen as a sampling unit. All the pixels
in the window were utilised for the sample.
30 sample sites for each category were lo
cated using random spatial coordinates. In
the classification analysis half of these
were used as a training set and the remai
ning half as a testing set.
2.4 Landsat data
The Landsat-2 CCT of path 145 ROW 042 dated
6-12-1981 collected by the National Remote
Sensing Agency ef India at the receiving
station in Secunderabad, India was used in
the analysis.
