Contact: Dr. Torsten Prinz, IVV Geowissenschaften, Robert-Koch-Str.
26-28, D-48149 Münster, prinz@uni-muenster.de
Abstract
Remote sensing techniques offer a unique chance to analyse and to map
planetary impact craters in a relatively short time and at low cost. In
the past, studies were mainly restricted to the search for possible impact
sites (e.g. Earth) or for age determinations (crater statistics). On the
basis of Landsat-TM 5 and ERS-1 data the lithological and structural characteristics
of the complex Gosses Bluff impact crater (Australia) has been analysed
in order to obtain reasonable lithological classification approaches. The
fundamental statistical selection rule for pure colour composites of original
TM-data was the calculation of the optimum index factor (OIF), or for hybrid
colour composites (e.g. a combination of a original TM-band with a principal
component and a ratio) using the widest statistical variance for each dataset.
Additional spectral measurements were carried out for each representative
rock unit of the crater specific zones in order to estimate the quality
of supervised maximum likelihood computer classifications for geological
mapping. Complementary ERS-1 altimetric data was utilized to study the
resulting crater morphology as an expression of the displacement effects
and some structural features of the target caused by the cratering process
(e.g. diameter, fracture pattern, ejecta displacement etc.).
1 Introduction
Over the last decade satellite remote sensing has played a major role in
the search for probable impact sites on Earth (Grieve, 1987; Garvin et
al., 1992; Hodge, 1994) or for age determinations (crater statistics) of
other planetary surfaces (Shoemaker et al., 1991). In both cases the geological
approaches to the impact structures were mainly restricted to general size
considerations or the analysis of circular fracture patterns (Theilen-Willige,
1982; Grieve & Garvin, 1984). In areas where field work seems to be
impossible or not under consideration for any reason, remote sensing data
represents the only source of information. Under such circumstances the
resulting geological iterpretation becomes more subjective, because it
strongly depends on the personal experience and impression of the remote
sensing investigator. This fact has often been criticized in the past,
because it underlines the conventional descriptive nature of remote sensing
techniques applied to geology (Gravin & Schnetzler, 1988). The deeply
eroded Gosses Bluff impact structure in central Australia (N.T.) offers
a unique chance to check more objective and retraceable mathematical remote
sensing techniques for the geological interpretation of impact craters
in arid environments because the structure has been studied in many field
campaigns by several authors over the last years (Crook & Cook, 1966;
Cook, 1968; Milton et al., 1972; Brown, 1973; Barlow, 1979).
Gosses Bluff (or Tnorula, so called by the local aborigines) is situated
in the desert-like Missionary Plain of the Northern Territory, approximate
160 km WSW' of Alice Springs (fig1). The natural
vegetation is only sparsely developed and often bound to seasonal water
gullies (spinifex grass, gum trees). Therefore soils, sand, alluvial deposits
and outcroping geological formations are well exposed and readily detectable
by the TM scanner.
1.1 Geological setting
Gosses Bluff represents a complex impact crater with a central uplift feature
(in contrast to smaller simple, bowl-shaped impact craters) which was generated
145 m.a. ago by a high velocity impact of a meteorite or cometary nucleus
(Milton et al., 1972; Milton & Sutter, 1987), plunging into the plains
near the western end of the Macdonnell Range. The projectile penetrated
more than 5000 m of subhorizontal layered paleozoic sediments of the Amadeus
Basin, releasing something in the order of 1020 joules of energy during
the impact process. The cosmic projectile and great amounts of the target
rocks evaporated, while the material flow (which followed the impact induced
shock waves) then disembowelled the plain beneath the vapourisation chamber.
During the rebound of the compressed deeper strata the uppermost sequences
were fractured, uplifted from more than 3 km and partly overturned, forming
the characteristic central uplift. The resulting crater was some 23 km
across and perhaps more than 1.5 km deep from its rim (Barlow, 1979). All
this overburden, including almost all morphological trace of the original
crater has since been removed by erosion.
Today most areas of the crater are covered by sand dunes, gravel, coarse
Quaternary alluvium and calcrete (fig. 2). In some
parts relics of brecciated para-allochthonous blocks of Devonian/Carboniferous
rocks (Hermannsburg and Parke silt-/sandstones) and impact breccias (Mt.
Pyroclast) crop out. In the center of the structure the erosional remnants
of the central uplift form a semicircular peakring, in which the oldest
(innermost) exposed strata (the Ordovician Stairway and Stokes sand-/siltstones)
was denuded. The peakring itself consists out of the concentric striking
Ordovician Carmichael sandstones and the Devonian Mereenie, Harajica and
Dare silt-/sandstones. The undisturbed crater foreland is characterized
by the slightly southward dipping layers of the Devonian/Carboniferous
Undandita and Brewer conglomerates, which mark the beginning of the Macdonnell
ranges to the N. In the S, E and W wide dune fields dominate the landscape.
Seismic investigations (Brown, 1973; Barlow, 1979) and field observations
(Glikson, 1969; Milton et al., 1972) proved that the crater underlying
the sedimentary sequences are deeply disturbed, faulted and sheared into
blocks of several decameters within the structural crater boundaries.
2. Remote sensing methodology
In order to link the spectral properties of the target material in some
test areas with calculated optimized TM colour composites and with the
ERS-derived altimetric data, this study adopted three different methods:
the first one is based on laboratory spectral measurements of representative
rock samples from different crater zones and the definition of detectable
classes using ground truth. The second one is the statistical analysis
of all digital numbers (DN) within the reflective TM bands of the scene
and the determination of suitable colour composites based on the calculated
statistical parameters of each component. This step includes maximum likelihood
classifications of the predefined object classes (e.g. stratigraphic units).
In the third and final step the best fit classification results are checked
for their possible relationship to some morphological features of the crater.
2.1 Spectral measurements
In the summer of 1993/94 geological field work was carried out around the
Gosses Bluff impact structure. During this campaign several representative
rock samples were taken from each exposed stratigraphic unit of the circular
crater zones and the undisturbed crater foreland. Due to the extreme dry
outback climate, the vegetation cover was only sparsely developed and confined
to some seasonal gullies like the Undandita Creek. The spectral properties
of the target material were therefore relatively unmasked, except of the
existence of desert varnish on rock surfaces in some places (later measurements
showed that this varnish caused a decrease of the reflectance intensity
around 15 to 20 %, while the important absorption features were still expressed).
Altogether the spectral properties of 50 samples (including loose sand
and pulverized coarse alluvium) were measured over the continuous wavelength
region from 400 to 2500 nm in relation to the TM scanner sensibility range,
employing the LAMBDA-9 photospectrometer of the 'Bundesanstalt für
Geowissenschaften und Rohstoffe' (BGR) in Hannover/Germany (Perkin Elmer,
1993). Due to the fact that most solid target rocks of the Gosses Bluff
are sand- and siltstones no extreme differences in the absorption features
were detected (fig. 3). Some of the sandstones (like
the Stairway and Stokes sandstones) include kaolinized feldspatic components
which caused stronger OH-absorptions around 1410 and 1920 nm in the mid
IR. Depending upon the Fe2+,3+-, H2O- and OOH-content (hematite, limonite,
goethite) of secondary mineral phases (Mereenie sandstones) a wide and
strong albedo decrease was observed over the near IR (946 to 855 nm), reaching
a minimum in the visible blue/green (504 nm). Due to the high amount of
iron-stained clasts (e.g. hematite) almost every class (including sand
and quaternary gravels) exhibited the highest albedo in the visible spectra
(VIS) around 600 nm (typical reddish colours of the australian outback).
The spectral properties of the Quaternary calcrete showed a remarkable
albedo decrease in the mid infrared (IR) near 2350 nm. This energy absorption
is caused by the CO2-3-component of calcite which represents the main mineral
phase.
Taking all measured spectral properties into account it is obvious
that the significant albedo and spectral differences do occur in the near
IR, the mid IR and to a certain degree in the VIS spectra. This emphasizes
the importance of the TM bands 4, 5 and 7 if they are combined with one
channel from the VIS to create a colour composite (CC) in which the strongest
spectral anomalies might be visualized.
2.2 Statistical analysis of the reflective TM data
What effects have the determined spectral properties for the statistical
features of the selected TM data and how far is it possible to calculate
an optimized pure CC without having the spectral information as a control?
Table 1 shows the important statistical parameters
for all used reflective TM channels. The widest spectral variance occurs
in the IR, especially in GBTM-5 and -7. GBTM-1 and -2 exhibit more or less
the same DN standard deviations/variances (so do GBTM-3 and -4). The spectral
information of different TM bands is often strongly correlated (Schowengerdt,
1983; Grunicke, 1990), so it is necessary to evaluate their degree of correlation
in order to determine least correlated channel combinations which might
be suitable for the enhancement of some reflectance features. The strongest
correlations exists within the VIS and IR spectra (tab.
2) which can therefore be interpreted as two seperate statitistical
groups. Vice versa the lowest redundancy occurs between one dataset of
the VIS and one dataset of the IR. Judging by this statistical analysis
a combination of GBTM-1, -5 and -7 would represent the most uncorrelated
pure CC.
Although this statistical method seems to be sufficient to select suitable
TM bands for a CC it is also important to take the widest possible DN contrast
into consideration which is also a criterion for the quality of the calculated
image. Chavez et al. (1980) developed the opitmum index factor (OIF) to
evaluate the information content of any correlated dataset combinations.
Grunicke (1990), Bischoff & Prinz (1994) applied a modified OIF to
the lithological analysis of TM and MSS multispectral data and achieved
satisfying results for geological interpretation. The OIF is based on the
DN correlation (r, representative for the uncorrelated information) and
the spectral deviation (sigma, representative for the expected DN contrast):
OIF = Sum sigma (i) / Sum |r(i)|
(where i= amount of datasets/channels). The higher the OIF, the more
uncorrelated spectral information is transformed into a contrast-rich CC.
Table
3 shows the OIF-ranking of all possible three-channel combinations.
Here the CC of GBTM-1, -5 and -7 is statistically defined as the most informative
TM-dataset combination (OIF = 15.55). In this CC, most lithological classes
are expected to be distinguishable by their different contrasts and special
colour tones (note that the natural CC GBTM-3, -2, and -1 (rgb) contains
the second lowest multispectral information!). So the OIF confirms the
special spectral significance of the TM bands 5 and 7 for the lithological
interpretation of CC's (see Section 2.1). Furthermore, the OIF can be applied
to any multispectral analysis, where no ground truth is available.
2.3 Hybrid color composites
OIF-defined colour composites may suppress reflectance features if they
occur within small areas of the subscene (due to the low number of pixels).
In such cases (see Stokes sandstones (Os), fig. 4)
principal components (PC) and special ratios (R) were necessary to enhance
these small scale features (Donker & Mulder, 1976; Gillespie, 1980;
Prinz, 1995). Those single datasets can be combined to form hybrid CC's,
which highlight strong reflectance differences, no matter how spatially
limited they are. The important statistical criterion is the variance (sigma
exp.2) between each dataset (high variances are preferred for each componen)
(see
Table 4). For this study the first three PC's and the three R's 4/3,
5/1 and 5/7 were calculated and later combined to form a hybrid CC GBTM-5,
GBPC-1 and GBR-5/7.
2.4 Lithological classes and classifications
Based on calculated pure and hybrid CC's, different lithological classes
were set up for the Gosses Bluff impact crater in respect to their spectral
properties and stratigraphic position (tab. 5).
In some cases two or more classes had to be merged depending upon their
similar color signature in the CC's. The representative training fields
for each class (or group of classes) underwent a statistical evaluation
(fig.
5) before being classified by applying the Bayesian decision criteria
(=supervised maximum likelihood classifier, after Hord, 1989).
In both classifications, the main groups 'solid' and 'loose' rocks
are well defined by their spatial distribution (fig.
6a). (fig. 6b). Sandy plains are under-represented
compared to coarser gravel and alluvial deposits. Surfaces which are sealed
with calcrete show a more realistic distribution pattern, especially in
the classification based on the pure CC GBTM-751 (so does the Brewer and
Undandita conglomerate). Within the eroded crater floor even small outcrops
of para-allochthonous rocks are classified with a high accuracy (e.g. Hermannburg
sandstones and impact breccias of Mt.Pyroclast). Peakring material of the
morphological central uplift can only be classified as a merged group of
similar sandstones (Dh, Dd, Pzm). The lack of homogeneous training fields
within each narrow lithological unit combined with the shaded topography
(steep cliffs) prevented the definition of any representative pixel clusters.
In contrast the Stairway and Stokes silt- and sandstones of the crater
center were classified properly as a homogeneous surface material within
the denuded alluvial central plain.
The most significant difference between the two classification results
is the more realistic calculation of the sand/gravel distribution pattern
and the accentuated linear occurrence of alluvial deposits (plus vegetation)
along erosional gullies based on the IR sensitive hybrid CC. However, both
lithological classifications represent reasonable geological mapping approaches
for the Gosses Bluff impact crater and its environs. The statistical analysis
and interpretation of suitable multispectral remote sensing data should
therefore be considered in addition to any planned field work.
3 Altimetric data and crater morphology
The most significant morphologic feature of 'fresh' complex impact craters
is a circular system of topographic highs and lows (fig.
7) which are generated by excavation processes, ejecta displacements
and the origin of semicircular fracture zones, terraced crater rims, ring
grabens and the uplift of material in the crater center (Melosh, 1989).
In highly denuded structures such as Gosses Bluff, only a few morphologic
features are still remaining. In extremely remote areas of our planet,
accurate topographic maps are often not available. For that reason the
modelling and visualization of the topography is restricted to satellite
derived altimetric data, which is now available for almost every region
on earth.
3.1 Digital elevation model of the Gosses Bluff
In order to visualize the crater topography of Gosses Bluff, ERS-1 altimetric
data was gridded and interpolated by using the software Win-Surfer (1994)
and VistaPro (1994). Under VistaPro it is possible to generate digital
elevation models (DEM) on the basis of smoothed Surfer data grids. The
simulated virtual landscapes combine realistic textures with colours on
different IQ-levels (fuzzy logic). VistaPro functions as a single frame
generator, similiar to a camera system, which renders a new view of the
crater from a user-defined combination of heights, angles and distances
(fig.
8).
If this process is applied to the Gosses Bluff area, solid outcropping
rocks are responsible for major topographic changes within the impact crater.
The eroded inner part of the crater center forms a circular convex shaped
plain (Os) which is surrounded by the eroded remnants of highly disturbed,
sheared and partly overturned sedimentary blocks (Oc, Pzm, Dh, Dd). This
peakring reaches a maximum relative altitude of more then 300 m above ground
[a.g.] (or 930 m above sealevel [a.s.e]). Along these hills many flat-iron
structures can be observed, indicating the steep dip of single sedimentary
units. Within the denuded crater floor only para-allochthonous blocks of
brecciated material (Dr) form local hummocky anomalies with a low altimetric
e xpression. Sand, gravel or evaporites (calcrete) tend to seal the fractured
surface, generating a very smooth topography, which continues into the
undisturbed crater foreland. There is no semicircular morphologic depression
detectable which might be linked to the outer terraced craterzones or ring
grabens (tangential normal faults). According to the lithological classifications
the structural limit of the impact crater can be defined as the outermost
occurrences of the calcrete (Ql). This evaporite does not occur beyond
a mean distance of 11 km from the impact center. It is reasonable to postulate
a dramatic change in the porosity due to the highly fractured subsurface
of the crater which is a direct effect of the impact. Therefore the maximum
extension of the calcrete can be taken as an approximation of the structural
crater limits. The deduced structural diameter of approximately 22 km is
in good correspondence with a value of 23 km optained from geophysical
results (Barlow, 1979).
4 Conclusions
The statistical methods proposed in this paper allow semiautomatic, supervised
lithological interpretation and classification of stratigraphic rock units
within a complex impact crater located in a desert-like environment. The
statistical decision criteria for pure and hybrid CC's can be adapted to
other lithological classification approaches. Satellite-derived altimetric
data and its interpretation in the form of digital elevation models has
been found a useful tool for any structural consideration concerning complex
impact craters, especially in remote areas where detailed and accurate
topographic maps are often not available. Furthermore both datasets can
be integrated into the GIS environment; this enables its further analysis
or combination with other types of data e.g. geophysical sources.
Acknowledgements
The author is grateful to the staff of the Institute of Planetology (WWU-Muenster)
for useful discussions and some hardware support. Thanks are due to the
remote sensing group of the BGR in Hannover for their help during the spectral
analysis. Furthermore, the author would like to thank the Australian Geological
Survey Organisation (AGSO) for their cooperation and support. Mr. Herrman
Mabulka, the representative of the traditional owners of the Gosses Bluff,
is gratefully acknowleged for his personal cooperation. The staff of the
Aboriginal Areas Protection Authority (AAPA) and the Conservation Commission
of the Northern Territory (CCNT) are thanked for their encouragement and
support during the field campaign.
Used data
Landsat-TM 5: 10/08/92 (date), 44901 (orbit), 103/076 (path/row). ERS-1:
05/05/92 (date), 4198 (orbit), 4095 (frame).
References
BARLOW, B. C. (1979): Gravity investigations of the Gosses Bluff impact
structure, central Australia. BMR Jour. Austr. Geol. & Geophys., 4:
323-329.
BISCHOFF, L. & PRINZ, T. (1994): Der Araguainha-Krater (Brasilien):
Das geologische Bild einer großen Impaktstruktur nach Geländ„ndebefunden
und Satellitenbildanalyse. Die Geowissenschaften, 12: 5-14.
BROWN, A. R. (1973): A detailed seismic study of Gosses Bluff, Northern
Territory. BMR Austr. Rep., 163: 43p.
CHAVEZ, P. S., BERLIN, G. L. & SOWERS, L. B. (1982): Statistical
methods for selecting Landsat MSS ratios. Jour. Appl. Photogr. Eng., 8:
30-32.
COOK, P. J. (1968): The Gosses Bluff cryptoexplosion structure. Jour.
Geol., 76: 123-139.
CROOK, K. A. W. & COOK, P. J. (1966): Gosses Bluff, diapir, cryptovolcanic
structure or astrobleme. Jour. Austr. Geosc., 13: 495-516.
DONKER, N. H. W. & MULDER, N. J. (1976): Analysis of MSS digital
imagery with the aid of principal component transformation. XIII ISP Congr.
Helsinki (abstract).
GARWIN, J. B. & SCHNETZLER C. (1988): Remote signatures of recent
large impacts in the earth record: Zhamanshin and Bosumtwi. EOS Trans.
Am. Geophys. Union, 69: p.1290.
GARWIN, J. B., SCHNETZLER, C. & GRIEVE, A. F. (1992): Characteristics
of large terrestrial impact structures as revealed by remote sensing studies.
Tectonophysics, 216: 45-62.
GRIEVE, A. F. (1987): Terrestrial impact structures. Ann. Rev. Earth
Planet. Sc., 15: 245-270.
GRIEVE, A. F. & GARVIN, J. B. (1984): A geometric model for excavation
and modification at terrestrial simple craters. Jour. Geophys. Res., 12:
11561-11572.
GRUNICKE, J. M. (1990): Methodische Untersuchungen zur digitalen Bildverarbeitung
von Fernerkundungsdaten: Lithologie und Tektonik der zentralen Lechtaler
Alpen, Tirol, ™sterreich. Berl. Geow. Abhdl., 121: 115p.
HODGE, P. (1994): Meteorite craters and impact structures on Earth.
Univ. Press, 124p.
HORD, R. (1982): Digital image processing of remotely sensed data.
Acad. Press: 210p.
MELOSH, H. J. (1989): Impact cratering; a geologic process. Oxford
Monogr. Geol. & Geophys., 11: 245p.
MILTON, D. J., BARLOW, B. C., BRETT, R., BROWN, A. R., GLIKSON, Y.,
MANWARING, E. A., MOSS, F. J., SEDMIK, E. C. E., VANSON, J. & YOUNG,
G. A. (1972): Gosses Bluff impact structure. Science, 175: 1199-2007.
PERKIN ELMER (1993): Lambda-9; a photospetrometer. Perkin Elmer
PRINZ, T. (1995): Multispectral remote sensing of planetary impact
structures. EGS-Meeting Hamburg, Ann. Geophys., Space & Planet. Sc.,
13: C-741 (abstract).
SHOEMAKER, E. M, WOLFE, R. F. & SHOEMAKER, C. S. (1991): Asteroid
and impact cratering rate on Venus. Lunar Planet. Sc. Conf. XXII, 1253-12564;
THEILEN-WILLIGE, B. (1982): The Araguainha astrobleme, Central Brazil.
Geol. Rundsch., 71: 318-327.
VISTAPRO (1994): The 3-D landscape tool, Vers. 3.0. Virtual Reality
Lab., Licence of Geol. Inst., WWU.
WINSURFER (1994): Surfer for windows. Golden Softw., Licence of Geophys.
Inst., WWU.
