algorithm. The Landsat MSS image of 1975
was resampled to 30 meter-pixel resolution
using nearest neighbor algorithm and then
registered to the 1993 Landsat TM image.
Two images were 18 years and 159 days apart
from each other, thus significant sun angle
differences between the two may exist.
However, at this time no effort was made to
minimize the sun angle differences. To
overcome the sensor calibration and sensitivity
differences, images were converted to
exoatmospheric reflectance using radiance
transformations suggested by (Markham and
Barker, 1986). A Digital Elevation Model
(DEM) was prepared using Survey of
Pakistan's 1:50,000 topographic maps. These
maps come with a 1000 meter grid, which
significantly enhanced their utility for such
purposes. First, all the grid intersections were
digitized. Then, elevations read from the
contour line at each grid intersection. Over
1000 elevation points were manually recorded
on a sheet of paper. Second, the elevation
points were imported to ARC/INFO in ASCII
format, and the KIRIGING procedure was used
for surface interpolation, which produced a 100
meter DEM. Finally, the 100 meter resolution
DEM was resampled to match 30 meter-pixel
resolution of TM.
6.1 Image Classification
An unsupervised classification scheme
was used for clustering purposes. This utilizes
a Iterative Self Organizing Data Analysis
Technique (ISODATA) to partition the data
into homogeneous spectral classes. The intent
of the ISODATA is to minimize the cluster
variance and to maximize the distance between
clusters. Statistical and visual analysis of the
data was then used to build meaningful clusters
of forest and non-forest vegetation classes for
both images.
120
The Landsat MSS classification results
were significantly satisfactory compared to
Landsat TM. TM image was acquired during
a time of year, that is considered a “prime
vegetation season” in this part of the
Himalayas. The presence of healthy green
cultivated fields and orchards on the Pakhli
Plain in Siran Valley introduced considerable
spectral confusion in delineation of forests on
the surrounding mountains. The MSS on the
other hand was acquired in a time of year,
when forests are the only vegetation in the
region. Most of the agricultural land on the
plain lies as fallow land, because cultivation of
winter wheat does not start until January. The
absence of agriculture greatly facilitated the
delineation process of forest classes in the
MSS image.
The problem of spectral confusion
encountered in the TM data was overcome by
using the DEM model. The knowledge of the
region’s physiography and vegetation was
incorporated in formulating logical rules to
stratify the vegetation types based on elevation
criteria. According to these rules, the
agriculture on the Pakhli Plain was restricted
to maximum elevation of 909 meters.
Agriculture in the valleys, which is mostly
terraced rice fields or corn was restricted to the
maximum elevation of 1090 meters. Any
vegetation above this elevation was considered
forest. Thus, the vegetation was classified into
three classes, a) forests b) terraced cultivation,
and c) agriculture on the plains. The 30 meter
DEM was recoded using ARC/INFO to above
mentioned three elevation classes. The
recoded DEM assisted in delineation of
argiculture on the plain from the forest pixels
on the surrounding hills. The forest-type maps
on a 1:50,000 scale, prepared by the forest
department were also used in classification
processes for both images.
determi
vegetat
class as
publish
persone
depicte
arrival :
present
Signific
observe
largest
Siran V
1992. S
the hill:
evident
efforts I
Forest I
livestoc
plantati:
data cot
capabili
assessm
visual ai
with DI
derive si
the exte
watersh
Allan, N
the ve
Hinduka
Develop
Jan. A.