before
|ghtness
‘fect of
IR-A and
uld be
)lom and
^. Image
ith the
| window
a set.
ition of
atively
OW.
A image
of the
nstant.
degrees
'o small
ire more
lations
ty were
graph ic
of the
ble to
IR=A or
! subtle
ws that
sensor,
gles.
on of
icy In a
le. As
| scale
larger
-A data
nation,
is used
e best
e swell
of MSS4
racy of
| tstone
| as the
ed the
creased
E TE
E RE
Blom and Daily (1982) found that the Inclusion of radar texture
information, based on variance images, Increased the rock type discrimination
accuracy by 21$. Texture information based on variance data derived from both
the SIR-A and Seasat Images was also included in the linear discriminant
analysis. The results showed that the best combination for overall
discrimination was MSS4, MSS7, a Seasat 17x17 variance image and a 15x15 SIR-A
variance image. The addition of the SIR-A data Increased classification
accuracy of . the Navajo Sandstone by 18$ to 75$, and the Salt Wash Sandstone,
an Important economic unit, by 30% to 97%. Results of discriminant analysis
based solely on SIR-A and Seasat variance images showed that 6 out of 8 of the
geologic units studied could be classified to greater than 50% accuracy based
on texture Information alone. The Salt Wash Sandstone was classified with an
accuracy of 93% based only on an 11x11 variance Image of the SIR-A data. In
the same image, there was no misclassification of the Kaibab Limestone for the
Moenkopi Siltstone, and the overall classification accuracy for the two units
was 33% and 72% respectively.
C. Interpretation of Extended Spectral Signatures
The value of coregistered Image data sets has been shown for
discriminating geologic units. However, the potential for identifying units
based on their extended spectral signatures (from gamma rays to microwaves)
has not been explored. The main reason is that it Is difficult to preserve
II thologic or compositional information throughout computer analyses that are
presently available. A technique has been developed that may make it possible
to identify as well as differentiate geologic units. This technique Involves
setting up a library of signatures of various materials at different
wavelengths based on laboratory measurements and theoretical models. Once the
library is set up, pixels in a stack of coregistered
interrogated to see If the trends in DN values in the set of Images matches
The spectral signature of a known material in the library. This technique has
been successful for visible and near-infrared images (Evans and Adams, 1981),
but has not yet been used for thermal IR and radar images. Results of the
discriminant analyses show that both radar backscatter and textural
information derived from radar images increase classification accuracy over
Landsat images alone. Other studies of radar image texture also show the
ability to relate radar Image texture to rock type (e.g. Farr, 1982, this
Images can be
volume). This Indicates that both radar texture and tone can be used as
components in a library of signatures.
CONCLUSIONS
Many techniques have been developed for analysis of multisensor image
data
With the advent of new orbital sensors and new quantitative techniques,
emphasis will be to actually identify lithologic
sensors. One technique will be to store characteristics (attributes) of
various rock types in a library of extended spectral signatures. These
signatures will contain Information about texture as well as composition.
Backscatter and texture information provided by radars with variable incidence
angles increase the accuracy in both classification and Identification of
geologic units.
sets that enable differentiation and classification of geologic units.
the
units with spaceborne
ACKNOWLEDGEMENTS
This work has benefitted from the ideas and previous work of Michael
Abrams, Ron Blom, Cathy Conrad, Mike Dally and Harry Stewart. The research
described here was carried out by the Jet Propulsion Laboratory, California
Institute of Technology, under NASA contract NAS7-100.
257
rd
EE —9—À a oe
me i, ems EEE N e RI sg t e e = eco
