The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
469
SWIR wavelength regions as found in other ecosystems. The
strong NIR spectral variation is indicative of highly varying leaf
index (LAI) at the scale of individual plant canopies (Asner et
a\., 2000). While the SWIR2 (2.1 -2.4pm) region did show some
variability in the magnitude of reflectance, but the whole
spectral shapes were highly consistent.
The NPV spectra show almost constant monotonic-increasing
reflectance in the visible-NIR region. Several spectral
absorption features in the SWIR2 (2.1-2.4pm) region are clearly
apparent. The features near 2.1pm and 2.4pm are associated
with the presence of cellulose and lignin.
The spectral characteristics of bare bedrock are also apparent,
especially the C0 3 absorption feature near 2.3pm. This is due to
that the most components of bare bedrock in karst ecosystem
are carbonate, which is mainly composed of limestone (Yuan,
1993).
Wavelength (am.)
ai n
o u
lo
Jlj
o
Ù1
Wavelength (uni)
NPV
05
<D
o 0.4
CCS
t5 o
•■i—
CD
Ltl 0
Wavelen^h ipm)
Figure2. Field spectral from the three transects: photosynthetic
vegetation (PV), non-photosynthetic vegetation (NPV), bare
bedrock (bedrock).
3.2. AutoSWIR spectral mixture analysis
range (0.4-2.5pm), only SWIR2 (2.1-2.4pm), and only tied-
SWIR2 endmember spectra. Table 1 showed the spectral
decomposition results for the study area using the four different
wavelength permutations.
Wavelength
permutation
Land-
cover
types
Study
area
Std.
Dev.
Field measurements
PV
0.436
0.010
NPV
0.272
0.010
bedrock
0.302
0.020
SUM
1.01
-
Hyperion full-range
PV
0.392
0.050
NPV
0.291
0.063
bedrock
0.377
0.046
SUM
1.06
-
Hyperion SWIR2
PV
0.157
0.102
NPV
0.312
0.041
bedrock
0.341
0.053
SUM
0.81
-
Hyperion tied-SWIR2
PV
0.412
0.060
NPV
0.269
0.033
bedrock
0.319
0.046
SUM
1.0
-
Table 1. PV, NPV, and bare bedrock cover fractions from 4
different wavelength permutations using AutoSWIR.
Spectral unmixing with the full-range (0.4-2.5pm) of Hyperion
unmixing yielded very close NPV cover fraction to the field
NPV values. It indicate that full spectral range may provide
some measure of NPV presence, but it grossly under-estimated
PV and over-estimate bare bedrock rate. This is likely duo to
the presence of very bright bare bedrock, which saturates this
wavelength region and leads to over-estimates of bedrock cover.
Spectral unmixing with the only SWIR2 (2.1-2.4pm) region of
the Hyperion data also yielded poor results (Table 1). In karst
rocky severely degradation region, especially during the
vegetation senesced period, SWIR2 spectra are dominated by
bright bare bedrock and NPV, this can lead to a substantial
under-estimate of PV cover.
Spectral unmixing with the tied-SWIR region of the Hyperion
yielded accurate estimates of all three land-cover types (Table 1).
The field measurements of PV, NPV and bare bedrock fractions
were well within the statistical uncertainty rang of the
AutoSWIR results. It was due to the tied-SWIR2 minimized the
contribution of intra-canopy structural variation to nonlinear
photon-tissue interactions. These results were consistent with
the work did by Asner and Lobell (2000), and indicate that the
tied-SWIR2 (2.1-2.4pm) spectra are a means for estimating the
dominant land-cover types (PV, NPV and bare bedrock) in karst
degradation ecosystem. Karst rocky desertification information
can be accurately extracted from EO-1 Hyperion data.
4. CONCLUSION
The Monte Carlo spectral unmixing method, AutoSWIR,
incorporated both spectral endmember variability and
uncertainty in the unmixing process. It involved generating a
large number of endmember (PV, NPV, bedrock) combinations
for each pixel by randomly selecting spectra from the database
of field spectra. The performance was evaluated using full
The research presented here indicates that SWIR2 (2.1-2.4pm)
spectral region are the main distinctive spectral characteristics
of photosynthetic vegetation (PV), non-photosynthetic
vegetation (NPV) and bare bedrock in karst degradation regions.
It has limitations in using full optical range (0.4-2.5pm) or only
SWIR2 region of Hyperion to decompose image into PV, NPV