Table 4 The Effect of correctly classifying
all misclassified urban pixels.
Class
Correct
(before
smoothing)
Total %
Correct
(after
smoothing)
Total %
Oak
1845/1915
252
93
261
97
•I
1947/9
97
82
104
87
SW CH
1961
86
98
85
97
E L
1933
139
66
154
73
II
1949
53
54
54
55
EL/HL
1981
31
50
30
48
H L
1971
97
96
100
99
D F
1966
128
71
144
80
N S
1940
1
1
3
3
"
1971/2
145
83
159
91
NS/SP
1966
67
31
49
23
S P
1928
77
93
74
89
C P
1965/6
92
79
90
78
URBAN
149
96
151
97
AGRIC
390
85
425
93
One
of the aims of
this
research
is to
build up a 'database' system to explore not
only the relationships outlined above but
also the influence of additonal variables
such as slope, aspect and soil type. Jones
(1972) demonstrated that production levels
for forest plots were shown to correlate with
aspect, average elevation and slope giving a
multiple correlation value of r=0.957.
One aspect to emerge so far from this study
is the very high correlation between SPOT
bands 1 and 2. One of these bands becomes
virtually redundant for the purposes of
spectral separation. Nelson et al (1984)
found that when using TM Simulator data for
forest cover-type mapping, the most useful
waveband combinations used at least one band
from the visible near infra-red and mid
infra-red spectral regions. Thus the
addition of a new mid infra-red sensor
proposed for the SPOT 3 satellite (expected
launch 1991) may greatly enhance spectral
discrimination. There is also evidence to
suggest that multitemporal classification
techniques improve classification accuracy.
4 CONCLUSION
The SPOT satellite will have the potential
for forest-cover mapping at species level,
and in some cases even different ages of the
same species. In this study the major
handicap to this was the presence of urban
areas within the study area. The high
spectral variance of the urban class resulted
in gross misclassification errors. Applying
a textural classifier or using a
multi-layered classification approach may
overcome this problem.
There would appear to be a relationship
between stocking and band 2 reflectance, and
size and band 3 reflectance. It is intended
to explore this further as well as
identifying the influence of additonal
variables - such as slope, aspect and soil
types, by the merging of data sets to create
a database system.
Satellite remote sensing has greatest
potential with regards to commercial
forestry. The main limitation in mapping
diversely specied, varied-aged natural forest
is the pixel resolution; too many 'mixed'
pixels would invariably result, rendering any
subsequent classification meaningless.
However the commercial forester's requirement
is often restricted to straight, even-grained
timber of selected species, resulting in
even-aged stands of single species which are
consequently easier to map.
REFERENCES
Anderson, J.R; Hardy, E.E; Roach J.T and
Witmer, R.E 1976. A land use and land
cover classification system for use with
remote sensor data.Geol. Survey
Professional Paper 964.
Buchheim, M.P;Maclean, A.L and Lillesand, T.M
1984. Forest cover type mapping and Spruce
Budworm defoliation detection using
simulated Spot imagery. Spot Simulation
Applications Handbook.Proceedings of the
1984 Spot Symposium. May 20-23 Scottsdale
Arizona.
Cross,A & Mason D.C.1985. Segmentation of
remotely sensed images by a split and
merge process.
Dury,S.J; Collins,W.G.and Hedges,P.D.1986.The
developement of a methodology for forestry
management using Spot simulation data.
Proceedings of the Int. Symp. on
Photogrammetry and Remote Sensing for the
developing countries,New Delhi,India.
Grainger, A. 1980. The state of the world's
tropical forests. The Ecologist Vol.10
Nos.1/2 p.6-54.
Holmes, G.D 1980. The ecology of even-aged
plantations : An introduction to forestry
in upland Britain. Quarterly Journal of
Forestry Vol 74 (2) 73-81.
Isako, J. 1979. Texture analysis by space
filter and application to forest type
classification. Proceedings of the
Fifth Int.Symp.on Machine Processing
of Remotely Sensed Data.Purdue Univ.,
LARS, Lafayette, Indiana, p 392-393.
Kachhwaha, T.S. 1983. Spectral signatures
obtained from Landsat digital data for
forest vegetation and land-use mapping in
India. Photogrammetric Engr. and Remote
Sensing 49 : 5 pp 685-689
Mayer, K.E and Fox III, L. 1981.
Identificationof conifer species groupings
from Landsat Digital Classifications.
Photogrammetric Engr. and Remote Sensing
47 : 11 pp 1607 - 1614.
Nelson, R.F; Latty, R.S and Mott, G. 1984.
Classifying northern forests using
Thematic Mapper simulation data.
Photogrammetric Engr.and Remote Sensing 50
: 5 pp 607-617.
Schary, R.W. 1972. Plants for Man. Prentice
Hall,Inc. Eaglewood Cliffs.
428