Table 6. TM Band 5 limits
Table 7. Thresholds for water, shore-
of 21 classes.
line, and land classes for each
Landsat scene.
TM Band 5
reflectance
Surface
Landsat
Water
Shoreline
Land
Class
Min. Max.
type
scene
classes
classes
classes
1
--
0.0
water
7 Mar 93
13
410
1121
2
0.0
0.01
31 Mar 93
15
616
1721
3
0.01
0.02
8 Apr 93
15
616
1721
4
0.02
0.03
24 Apr 93
15
616
1721
5
0.03
0.04
10 May 93
16
716
1721
6
0.04
0.05
26 May 93
15
618
1921
7
0.05
0.06
27 Jun 27 93
16
716
1721
8
0.06
0.07
29 Jul 29 93
16
716
1721
9
0.07
0.08
shoreline
30 Aug 93
15
613
1421
10
0.08
0.09
1 Oct 93
15
616
1721
11
0.09
0.10
2 Nov 93
16
716
1721
12
0.10
0.11
4 Dec 93
16
716
1721
13
0.11
0.12
14
0.12
0.13
15
0.13
0.14
Hand masking
16
0.14
0.15
All scenes were masked by hand to eliminate
17
0.15
0.16
18
0.16
0.17
terrain shadows, clouds and cloud shadows, float-
19
0.17
0.18
ing debris, stranded pools of water outside the res-
20
0.18
0.19
ervoir, and vegetated areas. A line separating the
21
0.19
0.20
land
upper end of the reservoir pool from the river was
All classes had the additional
also drawn manually, using the point where the
criterion that TM Band 4 reflec-
tance had to be ≤ 0.2029.
river becomes braided as an indicator of the up-
stream end of the reservoir pool.
even though it is very similar to Band 5, because it
Acreage tally
has a smaller data range than Band 5.
The pixels remaining in the unmasked areas
Next, the spectral band data used in the classi-
were tallied and scaled to acres. Total water sur-
fication (Bands 4 and 5) were converted from ra-
face area was then computed by summing 100% of
diance values (total amount of energy reflected)
the area of the water pixels and a prorated amount
to reflectance values (percent of incoming energy
of area from the shoreline pixels. For instance, for
reflected) using a function available in the ENVI
the 27 June 1993 Landsat scene, classes 16 were
software. This helped compensate for illumina-
tallied as 100% water, and classes 716 (10 classes)
tion variation through the year. The amount of
were prorated: class 7 at 91% water, class 8 at 82%
illumination varies with the sun elevation angle,
water, and so on, with class 16 at 9% water, and
which ranged from 29 to 62 in the Landsat data
classes 1721 as 0% water.
used (see Table 1).
The data were then classified. Any pixel with
Multitemporal water classification map
Band 4 reflectance value greater than 0.2029 was
A multitemporal water classification map was
eliminated from consideration because it was
created by combining the individual water classi-
more likely to be vegetation than shoreline or wa-
fications of the 11 scenes from 1993 in which the
ter. Remaining pixels were classified into one of
reservoir was not empty. The first class in this mul-
21 classes, based on the Band 5 reflectance value.
titemporal classification included all pixels classi-
Table 6 lists the classes with their Band 5 mini-
fied as water in the scene with the lowest water
mums and maximums.
level (2 November, 578.51 ft elevation). The second
The 21 classes were then divided into three
class included all pixels classified as water in the
surface types: water (low Band 5 reflectance),
scene with the next highest water level (1 October,
shoreline (intermediate Band 5 reflectance), and
594.65 ft), excluding those pixels already assigned
land (high Band 5 reflectance). The exact cutoff
to the first class. A similar procedure was used to
thresholds between water, shoreline, and land
create the remaining classes, up to the scene with
were selected separately for each scene by visual-
highest water level (7 March 1993, 665.86 ft). This
ly examinating the images. These thresholds are
multitemporal classification includes only the
listed in Table 7.
100% water classes from the individual scenes and
7
Back to contents
Previous Page
