XIX-B8, 2012
avipanah, 2003].
gression analysis
es existed in the
" image analysis.
ed.
ig bhattacharrya
ased on training
etic bands with
to mean (MD)
primary results
density classes,
sifications were
viating from the
35mx35m) was
assessment of
the use of error
ccuracy and an
)].
ayers of digital
mage indicated
classes (0-596,
jared for about
asses showed
ent of 50% and
:ctral similarity
forest density
th 3 classes (0-
ppa coefficient
sure 4 presents
TRO
ters
ses resulted
84)
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B8, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
Table 1. Error matrix of the best classification with three
density classes
= >
a Dn S
an 8 + eiu g
ST | IT) ES | 5
=
2 =
g
Dnm =
2° >
E co o oo Ô
SE D = D
S 9 c = e =
È 9
Be os 2
2
=
> SI
s ON
5s oo — — o c
o 5 en co Mo c
= 5 c e e e
8 S
M
= No] © uv pr
3 «&| uni SS o
© + | N| ES e
EM — U- = xt
e e ©
ee = en -— x
S NISI C
—
s
= mn | 2 |5 &
£z eu = oo > en
= v — oo
g
© ON en oo e
— <t uv e =
v — ©
— CQ en
8 —
= s
c ©
= >
"
un
c3
w—
=
5. CONCLUSION
Based on this investigation, fuzzy and maximum likelihood
classifiers indicated almost the same and the highest overall
accuracy and kappa coefficient. According to the best band sets,
the distance-based vegetation indices could improve the
classification results slightly.
Considering the low kappa coefficient (0.44), even if reaching
to pretty good overall accuracy (7096), the result of
classification was not desirable because forest canopy is a
continuous variable and the decision boundaries do overlap
[Skidmore ef al., 1988]. Remotely sensed data and classification
methods applied in this research could not well classify forest
density as a continuous variable in this low density forested
area. Similar research confirms this too [Joshi et al., 2006].
Overall, it could be concluded that this approach is not
appropriate for operational mapping of vast Pistachio forests.
Higher spectral and spatial satellite data or multispectral digital
aerial photos such as UltraCam images and object-based
classification should be investigated as an alternative approach.
In this case, the tree crowns could be detected and sum of areas
per ha can be used for density estimation.
6. REFERENCES
Alavipanah, S. K., 2003, Remote sensing application in the
earth science. University of Tehran Publication, 478 pp.
Congalton, K.G. and Green, K., 1999. Assessing the accuracy
of remotely sensed data: Principles and practices. New York,
NY: Lewis Publishers, 137 pp.
Dorren, L.K., Maier, A.B. and Seijmonsbergen, A.C., 2003.
Improved Landsat-based forest mapping in steep mountainous
terrain using object-based classification. Forest Ecology and
Management, 183, 31-46 (2003).
Eastman, J. R., 2006. IDRISI Andes Guide to GIS and Image
Processing. CLARK University, Version 15.00, 327 pp.
Joshi, C., Leeuw, J. D., Skidmore, A. K., Duren, I. C. V. and
Oosten, H. V., 2006. Remotely sensed estimation of forest
canopy density: a comparison of the performance of four
methods. International Journal of Applied Earth Observation
and Geoinformation, 8, 84-95.
Revised Plan of Pistachio Forest of Khajehkalat Studies, 2009.
Natural Resources General Office of Razavi Khorasan
Province, 139 pp.
Skidmore, A. K., Forbes, G.W. and Carpender, D.J., 1988. Non-
parametric test of overlap in multispectral classification.
International Journal of Remote sensing, 9, 777-785.
