ie
in
‘es
Nagendra, Harini
class spectral variances therefore provides a possible explanation for the bias in estimates of patch size, shape and
nearest neighbor distance resulting from landscapes mapped by unsupervised classification.
1 2 3 4 5 6 7 8 9 10 11 12
BAND |S {0.9 1.5 1.4 0.5 1.7 0.9 0.6 0.4 0.3 0.7 0.6 11
I U {15 +121 +156 +{L7 +!32 +|19 +117 +125 +116 +131 + 22 +124 +
BAND |S [0.7 0.8 0.9 0.5 12 0.9 0.6 0.3 0.3 0.9 0.5 1.0
2 U 109 +113 +499 +115 +127 +116 #115 +112 +117 +712 «4116 2122 +
BAND |S {6.7 4.0 4.0 22 3.9 4.0 3.0 0.6 0.6 2.0 2:0 3.9
3 U $33. —161 4198 4163 -127 +159 +|43 +136 +194 +169 +183 + 5.8 +
BAND |S {3.1 8.1 5.6 2.0 4.6 15.7..16.5 13.8 17.0 6.0 7.0 8.3
4 U.173 4172 «11354192 411701176 - 164 - 168 — [115+ {162+ H44-4 180
Table 2: For twelve landscapes, this table depicts the median of class signature variances in all four bands, for
supervised and unsupervised classifications. A “+” next to the unsupervised classification value indicates that
unsupervised class variance is greater than supervised; a “-” indicates the opposite.
4 CONCLUSIONS
In recent years, there has been an increasing interest in the quantification of landscape pattern, at various spatial scales
(Forman, 1995). Metrics of landscape pattern are believed to have significant bearing on the distribution and
maintenance of species diversity (Turner, 1989). An improved assessment of landscape pattern could not only increase
our knowledge of ecology and diversity, but also improve our efforts at preservation.
Land cover maps, created through remote sensing, enable the evaluation of such pattern. The study of landscape and
patch structure has however been mainly carried out in the temperate regions of the world (see for example, Baskent
and Jordan, 1995; Palang et al., 1998). Large parts of the tropics remain unstudied (Roy, 1993; Rey-Benayas and Pope,
1995). This paper attempts to compare the behavior of supervised and unsupervised classification techniques for
landscape pattern assessment.
Unsupervised classification provides a means of mapping where ground information is often difficult to obtain, though
it can result in increased misclassification (Nagendra and Gadgil, 1999 b). Misclassification is known to influence the
estimation of landscape pattern, but has been previously believed not to bias landscape metrics significantly (Wickham
et al, 1997). This paper however reports a significant positive bias in estimates of patch size, patch shape, nearest
neighbor distance and landscape diversity, due to misclassification by a technique of unsupervised classification.
Indices of interspersion-juxtaposition and contagion do not demonstrate this bias.
Further investigations are required to study the effect of within-class variance on patch statistics for unsupervised
classification. Simulated data with known levels of spatial auto-correlation and spectral variance, may be used for this
purpose. Whether the bias observed in landscape pattern metrics is specific to the algorithm of unsupervised
classification used here, or applies also to other algorithms of unsupervised classification, requires exploration.
Investigations along these lines are underway.
Assessing the distribution of the diversity of life forms on the earth and the efficacy of measures for their conservation
is a major scientific challenge. The Global Biodiversity Assessment (Heywood, 1995) recommends remote sensing for
country studies to identify components necessary for biodiversity conservation and maintenance. In India, as elsewhere,
there is increasing interest in the estimation of landscape metrics for this purpose (Roy, 1999). It is essential to take
cognizance of the possible effects that classification methods may have on these estimates. This exercise has strong
implications for the development of country level methodology for monitoring biodiversity in India.
ACKNOWLEDGEMENTS
I thank U. Ghate, V.V. Sivan and K.A. Subramaniam for assistance with landscape classification, and M. Gadgil and
N.V. Joshi for helpful discussions. This research was financially supported by the Ministry of Environment and Forests
and the Department of Space, Government of India. I thank the organizers of the XIXth ISPRS Congress for providing
travel and financial assistance, thereby enabling this paper to be presented at the Congress.
REFERENCES
Baskent, E.Z., Jordan, G.A., 1995. Characterizing spatial structure in forest landscapes. Canadian Journal of Forest
Research, 25, pp. 1830-1849.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 959