Abstract:
Satellite data has long been in use to estimate the water-spread area at different
elevations of a reservoir. The delineated water-spread is used to quantify the capacity
of a reservoir. The traditional methodology involves per-pixel classification approach
to delineate the water-spread, which is the only thematic information required to
estimate the capacity of a reservoir. One of the limitations of the per-pixel approach is
that the border pixels, containing water with soil and vegetation, are also classified
entirely as water pixels, thereby giving inaccurate estimate of the water-spread area.
To accurately compute the water-spread area to the maximum possible extent, the
sub-pixel classification or linear mixture model (LMM) approach has been used in
this study for classifying the water-spread areas of Singoor reservoir, Andhra Pradesh.
IRS-1C & 1D (LISS-III) satellites were used to extract the water-spread area
for the period 2005 using per-pixel and sub-pixel classification approaches. MLC and
band threshold methods have been adopted in the per-pixel classification approach.
The estimated capacity of the reservoir from MLC, band threshold and sub-pixel
approaches are 688.48 Mm3, 727.75 Mm3 and 716.11 Mm3 respectively. The perpixel
and sub-pixel classification was validated using high resolution PAN (5m) data.
The validation shows that sub-pixel classification produced very less error (1.07%)
than the MLC (6.1%) and band threshold method (3.77%).
1997 hydrographic survey (791.22 Mm3) and the capacity estimated (2005)
using sub-pixel approach (716.11 Mm3) were used to estimate the rate of
sedimentation. Based on these results, if uniform rate of sedimentation is assumed
from 1997 to 2005, the reservoir sedimentation rate is 9.39 Mm3
Apart from this a preliminary study on the three satellite pass of 56m (IRS-P6,
AWiFS) resolution data, was carried out for the extraction of water-spread area using
sub-pixel classification methodology. The sub-pixel approach applied on the 56 m
satellite data produced an average accuracy of 97.84% when compared with the 24 m
resolution data. This indicates that the 56 m resolution data with sub-pixel
methodology can be used to get results comparable with the 24 m resolution data.
