For the sume error margin. the number of pixels decreases to
97.5 *5 (band 69) and 36.3 % (205) for a compression ratio of
40:1. X 6. 5-75 error margin in band 205 includes 95.8 9o (10-0)
294, (20:1), and 89.786 (30: 1} of the pixels
d th
Mas y
VIN Term rt ete rt ——
16:1
Pd:
Band 6% (1011 nm)
250000
—— ali: d
i
200000 i |
d Siu |
À |
1000008
i
Sigg T
a :
- i
i T T ¥ 1 % 1
*
Frequency (Pixels)
i T Y T v Y -v T"
i I 3 à 5 & 7 K o 3 ] EE 13
PRAD {%u)
1040040 —— A — m —
Band 205 (2319 nm) o del
ded ddd Fd | |
FTTTT : —
— —— 40:1] }
— |
= i
= |
mo 5605204 1 d
-— |
* }
+ i
Zo aeu i
= i
= i
= |
20890% + :
]
i
s po a acwqu-napeuspomepr eet
2 K. 15 46 189 314 38 3%. 44 dà 43; dd 52
PRAD (*4)
Figure 2. Frequency {pixel} distribution of PRAD for band 69
(1011 nm) and 205 (2319 nmi
The spectral difference (PRAD) between original and de-
compressed data for individual pixels behaves similarly to these
two fidelity assessment cases. An example derived for an
"alunite" pixel for the compression ratios 10:1, 20:1, and 40:1
is shown in Figure 3. It indicates that the relative differences for
the 10:1 and 20:1 compression ratios are very similar with
errors below 0.3 % in most bands. These errors increase up to
1% above 2398 nm. As expected, errors of up to 6.3 % occur in
the strong atmospheric water absorption regions at 1380 nm
and 1870 nm. The results, retrieved from the comparison of the
original spectrum and the spectrum compressed at 40:1,
revealed larger errors, up to 10.2 %, with the strong water
absorption regions excluded.
5.2 Data Products
The results of the endmember selection using the IEA approach
are summarized in Table 3 for the original and de-compressed
data cubes. The endmembers found were identified using the
USGS spectral library and mineral map, which was derived
from AVIRIS data (USGS, 2002). Compared to the original
data, the same endmembers were retrieved from data at
compression ratios 10:1 and 20:1, although the IEA procedure
selected some of the endmembers in a different order, resulting
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002
PRAD (a)
Ana Sul 13uu 1800 sun
Wavelength (nm)
Figure 3. PRAD calonlated from original and de-compressed
radiance spectra for an “alunite” pixel
in a different endmember number for a given endmember.
Accordingly, no endmembers were lost. This is not the case for
the 40:1 compression ratio where the endmember 14, dickite
(mixed with kaolinite), was not retrieved.
Lrg Endmember
EM #
CON
Calcıte (mxed with
unknown
Adunite
A aelimite
Montmoriionite
{mixed with
unknown
Pca
Raelimte
Muscovite {maxed
with unknown)
Aaolimnite (mix
with Alunite)
Alunite (mixed wit
Kaclinite)
unite (mixed with
EK aochlinite)
Buddinztonite
Montmornlionite
Dickite (mixed with
Kaolinite?
m Montenoritionite ER |
ahie 3. Endmembers extracted from the different cubes with
the automatic [EA procedure
Differences between endmember spectra, extracted from
original data and decompressed data, generally increase with
increasing compression ratio. This trend is reflected by the
SAM fidelity measure with average values of 0.004
(compression ration of 10:1), 0.008 (20:1) and 0.010 (40:1) for
all the endmembers. However, deviations from this trend occur
for muscovite (endmember 8) and for calcite (2) where the
SAM values for the 20:1 compression ratio are higher than
those at the 40:1 ratio. It should be noted in this context that
endmember spectra, extracted from the original data and de-
comp!
pixels
for a
up to
absor]
In or
endmé
buddi
2500-
identi
The fi
for th
(comp
centre
Reflectance (Ya)
+
Reflectance (%a)
ta
Du
The R
data
compl
averag
ratio c
(relati