Remote Sensing of Environment 99 (2005) 159 – 172

[Link]/locate/rse

Seamless geological map generation using ASTER in the

Broken Hill-Curnamona province of Australia
R.D. Hewson a,*, T.J. Cudahy a, S. Mizuhiko b, K. Ueda c, A.J. Mauger d
a
CSIRO Exploration and Mining, ARRC, Western Australia
b
ERSDAC, Tokyo, Japan
c
Sumiko Consultants Co., Tokyo, Japan
d
PIRSA Geological Survey of SA, South Australia

Received 20 September 2004; received in revised form 11 April 2005; accepted 28 April 2005

Abstract

The availability of multiple ASTER image acquisitions enables regional-scale geological mapping, though instrument, irradiance,
atmospheric and surface scattering effects can cause problems in generating seamless mosaics of geological information products. These
issues, including shortwave infrared (SWIR) crosstalk, were addressed in producing seamless ASTER geological maps over the Curnamona
Province, associated with the world class Pb – Zn – Ag Broken Hill deposit. Over 35 ASTER scenes covering an area of approximately 52,000
km2 from 14 different overpass dates were acquired. Maps of Al – OH and Mg – OH/carbonate were generated from ASTER SWIR data as
well as a map of quartz content from the thermal infrared (TIR) data. Maps of ferrous iron content were also generated from the SWIR data of
individual ASTER scenes. The SWIR bands also enabled qualitative mapping of the Al – OH composition though garnet and feldspar – rich
units were not well mapped using the TIR. Field sampling and spectral measurements, together with detailed 1 : 25,000 mapping and large-
scale HyMap surveying, constrained the accuracy of the ASTER-derived geological products.
Crown Copyright D 2005 Published by Elsevier Inc. All rights reserved.

Keywords: ASTER; Geological mapping; Broken Hill; Curnamona; Multispectral

1. Introduction (VNIR), shortwave infrared (SWIR) and thermal infrared

(TIR) wavelength regions offering the potential for discrim-
The accessibility of inexpensive, satellite-borne, multi- inating phyllosillicates and also other silicates. Several
spectral ASTER data has created new opportunities for the examples of generating mineralogical maps using single
regional mapping of geological structure and rock types ASTER scenes have proved successful (Rowan & Mars,
including alteration products, and regolith. These data have 2003; Hewson et al., 2001).
been used enthusiastically by the minerals industry around The area of study encompasses approximately 52,000
the world. The ASTER sensor was developed by Japan and km2 of the Curnamona Province from Broken Hill in
launched onboard NASA’s Terra satellite platform. ASTER western New South Wales to Olary in South Australia and
acquires imagery within a 60  60 km scene area from 14 the surrounding regolith-dominated terrain (Fig. 1). This
different spectral bands with a pixel resolution of between study examined the pre-processing issues involved with
15 to 90 m, depending on wavelength (Fujisada et al., 1998; handling over 30 ASTER scenes acquired on 14 different
Thome et al., 1998; Yamaguchi et al., 2001). Of particular dates within the Curnamona Province (Fig. 1). These pre-
interest for remote sensing geoscientists are the inclusion in processing issues included SWIR crosstalk (Iwasaki et al.,
ASTER of detectors covering the visible-near infreared 2001), which has a significant detrimental effect on SWIR
spectral signatures. Following pre-processing, a number of
* Corresponding author. Tel.: +61 8 64368 689; fax: +61 8 64368 555. quality-control issues for the ASTER-derived geological
E-mail address: [Link]@[Link] (R.D. Hewson). maps were also examined with the aid of field measure-
0034-4257/$ - see front matter. Crown Copyright D 2005 Published by Elsevier Inc. All rights reserved.
doi:10.1016/[Link].2005.04.025
160 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172

139° E 140° E 141° E 142° E 143° E

31° S
NT
Qld
WA

SA
32° S
NSW

Vic

33° S
Tas

Fig. 1. ASTER scenes acquired from fourteen different dates (each colour corresponding to different acquisitions). Blue boundary marks project study area.
(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

ments and airborne hyperspectral HyMap data (Robson et its potential economic Broken Hill-style deposits of Pb–
al., 2003). Zn base metals and possible Cu – Au systems associated
The main objective of this study was to generate new, with hydrothermal alteration. The world-class Broken Hill
accurate and seamless geological/mineralogical information Pb – Zn – Ag orebody within the eastern part of the study
from ASTER images acquired within the Curnamona area is associated with high-grade metamorphic rocks. A
Province. Specific objectives include: variety of tightly folded and high-grade metamorphic units
form well-exposed outcrops, including gneiss, schist,
1. Characterising of the SWIR crosstalk effect and assess- pelite, psammite, amphibolite and granulite lithologies
ment/development of methods that can effectively (Stevens et al., 1988). An equivalent suite of classified
remove this instrument problem; geological units belonging to the Curnamona Province is
2. Characterising of the effects of the atmosphere, espe- also found in the Olary Domain of South Australia
cially water vapor, in the SWIR; (Conor & Fanning, 2001) although this area lacks the
3. Characterising of cloud/cloud shadows and strategies for detailed 1 : 25,000 geological mapping of the Broken Hill
their masking; Domain.
4. Generating a method for generating seamless geological Of interest for exploration in the Broken Hill area is the
products; mapping of regional prograde and retrograde metamor-
5. Identifying the diagnostic spectral features that can be phism and possible metasomatic alteration, either associ-
targeted for mapping mineral groups within outcropping ated associated with the Broken Hill type syngenetic or
geological units and regolith of the Curnamona Province; epigenetic fluids. Previous studies have shown retrograde
6. Devising suitable algorithms for mapping such mineral alteration associated with the development of muscovite/
groups over multiple mosaiced ASTER scenes; sericite and chlorite (Corbet & Phillips, 1981). Rowan and
7. Validating the derived information products using field/ Mars (2003) have shown that ASTER can map AlOH
airborne data and scene-based methods; abundance and possibly changes related to AlOH chem-
8. Establishing the credibility or otherwise of the generated istry (Duke, 1994), associated with metamorphism.
maps by comparison with published geology from the
New South Wales and South Australian Geological
Surveys in conjunction with field observations and 3. Pre-processing issues and strategies for mosaicing
spectral measurements; ASTER imagery
9. Contributing more geological mapping detail to the Olary
Domain of the Curnamona Province. The generation of multi-scene ASTER image ‘‘seam-
less’’ products requires consideration of sensor character-
istics (i.e. crosstalk), atmospheric effects (scattering and
2. Geological setting transmission), soil conditions and anthropogenic effects.
Chief amongst these issues for the ASTER SWIR bands are
The study area for this project encompasses the instrumental crosstalk effects and atmospheric transmission.
Curnamona Province that continues to attract interest for In this study for the most part, level 1B (L1B) ASTER
 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172 161

radiance-at-sensor images were used for the SWIR bands 4 staggered timing of acquisition for each of the spectral
to 9, because of the possibility of inaccurate atmospheric bands for a given row/pixel. This offset of the SWIR
correction in the surface reflectance standard product, image boundaries can be of the order of 20 pixels (i.e.,
AST_07, available from the Land Processes Distributed 600 m) although there is no apparent corresponding
Active Archive Centre (LPDAAC) (Rowan & Mars, 2003). spatial offset between bands within the actual image.
Most importantly, however, at the time of this study the Such offsets between SWIR images can generate artifacts
SWIR crosstalk software correction was not routinely along the east-west boundaries of mosaicked band ratio
applied to L1B archived data obtained from either the images. Corrections were applied to each ASTER scene of
Earth Remote Sensing Data Applications Centre (ERS- SWIR images using simple conditional algorithms to test
DAC) or LPDAAC. Version 3.0 beta SWIR crosstalk for the presence of the full complement of pixel spectral
software, courtesy of ERSDAC, was applied to all L1B data.
data acquired for this study. The Level 2 (L2) ASTER ASTER’s default projection datum, WGS 84, is effec-
surface emissivity data used in this study were derived tively identical to the Australian datum, GDA94, used for
from the L1B radiance data after atmospheric correction geological mapping and results presented in this study. A
(Thome et al., 1998) and separation of the emissivity comparison of the ASTER L1B SWIR data to the 1 : 25,000
component from the kinetic temperature component (Gil- mapping available in the Broken Hill area, indicated that
lespie et al., 1998). the image data spatial accuracy was better than 100 m. An
analysis of the spatial accuracy for seven ASTER scenes
3.1. SWIR Pre-processing (SWIR bands) acquired at four different pointing angles of
8.58-, 2.87-,+ 2.88- and + 8.57-, revealed average
SWIR crosstalk is an offset or additive error in radiance residual error distances of 45.2 m, 70.8 m, 34.3 m and
due to the leakage of photons from one detector element to 98.4 m, respectively. The overall geometric accuracy from
another (Iwasaki et al., 2001). This cross-detector leakage is 25 ground control points over the seven ASTER scenes
most pronounced from band 4 to bands 5 and 9, but it was 63.7 m and regarded as sufficiently accurate for this
affects all SWIR bands. For very dark pixels adjacent to study.
bright pixels, the crosstalk effect will approach 100% of the To maximise the dynamic range of the 8-bit SWIR data
input radiance signal. A spatially dependent software and process the L1B data into calibrated radiance at the
correction for crosstalk has been developed by Iwasaki et sensor (W/m2/sr/Am), a set of gains (unit conversion
al. (2001) and has since been incorporated by the Japanese coefficients) were applied after the crosstalk correction
ASTER Ground Data System (GDS) as a part of its L1B (Abrams et al., 2002). The calibration to spectral radiance
pre-processing. GDS have made this software publicly units of the L1B ASTER data were then obtained using the
available for users to correct their existing ASTER L1B equation, Radiance = (DN – 1) * Gain.
data of crosstalk effects ([Link] Variable illumination conditions of the ASTER radiance
gds_www2002/service_e/u.tools_e/set_u.tool_ecross.html). data resulting from different solar angles can also be a
This software was applied to the Curnamona L1B data to significant seasonal effect related to the cosine of the solar
generate corrected ASTER Hierarchal Data Files (HDF) incident angle (Schowengerdt, 1997). At the latitude of the
files using input parameters listed in Table 1. These input Curnamona study area (¨32- S) and for the ASTER
parameters include the amplitude, a, of the amount of acquisition time (¨10:30 a.m.), the range of solar incidence
incident light leaked from band 4 (% units); jx , the size angles for flat ground varies from 60 to 23 degrees (from
(pixels) of the applied Gaussian filter function in the across winter to summer solstice), producing almost a twofold
track direction; and jy , the size (pixels) of the Gaussian change in spectral irradiance. However this effect is
filter function in the along track direction. cancelled if spectral normalisation (e.g., band ratios) is
An apparent east-west offset between the different employed in the information-extraction strategy (Abrams et
georeferenced SWIR bands along the scene boundaries al., 1983).
results from the inclined descending orbital path and the Another major issue involved with processing mo-
saicked ASTER SWIR data into seamless maps is the
variability of atmospheric water vapor between different
Table 1
acquisitions. The lack of ASTER spectral bands over key
Input parameters for ERSDAC Crosstalk Software (v. 3.0) applied in water absorption bands and the situation where atmospheric
Curnamona Study information from other Terra atmosphere instruments were
ASTER Band a rx ry not used routinely for ASTER corrections at the time of
5 0.15 42 39 this study, means that standard climate models have been
6 0.06 31 40 used for ASTER atmospheric corrections. Thus towards the
7 0.04 29 30 deep atmospheric absorptions at wavelengths longer than
8 0.06 31 40 2.5 Am, L2 surface reflectance data are prone to errors,
9 0.15 42 39
particularly for ASTER band 9 (e.g. 2.360 –2.430 Am).
162 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172

a seamless mosaic of overlapping imagery (Fig. 3b). By

comparison, the ratio of bands 7 and 9 show major
differences across different acquisition dates (Fig. 3c).
Although path radiance is predominantly occurring within
VNIR wavelengths, any significant additive aerosol
scattering effect within the SWIR would likely to show
up within the b4 / b7 mosaic. The importance of these
results is that differences between ASTER_s SWIR
radiance data acquired at different dates, under varying
Fig. 2. MODTRAN 4 atmospheric radiative transfer model results at
ASTER SWIR spectral resolution (bands 4 to 9) indicating effects of atmospheric and solar illumination conditions, appear
variable climates and associated water vapor on radiation. effectively multiplicative in nature. In this study ASTER
L1B radiance data were mosaicked assuming linear gain
factors to adjust for variable acquisition conditions.
Further MODTRAN modeling in the future could usefully
The variation of atmospheric transmissivity at ASTER also be undertaken to derive aerosol scattering effects and
SWIR spectral resolution was estimated, using MOD- path radiance. If ASTER SWIR responses to variable
TRAN4 (Berk et al., 1999). This atmospheric modeling atmospheric conditions are predominantly multiplicative,
indicated that radiance measurements by ASTER bands 4 path radiance and atmospheric scattering effects should be
and 7 are insensitive to changes in water vapor associated near zero.
with different modeled climatic conditions (Fig. 2). Bands The significance of the combined errors associated with
8 and 9 are the most affected by water vapor absorption crosstalk and inaccurate atmospheric correction can be
(Fig. 2). The effects of the changes in the atmospheric demonstrated by the comparison between the ASTER L2
water vapor during different ASTER L1B data acquis- surface reflectance signature and ASD (Analytical Spectral
itions from the Curnamona region become obvious for Devices Inc.) VNIR –SWIR spectral measurements, col-
mosaicked ASTER band ratios b4 / b7 and b7 / b9 using lected at the Broken Hill Airport and resampled to ASTER
L1B data within the Broken Hill region (Fig. 3a,b,c). Fig. spectral resolution (Fig. 4). The large contrast between the
3a shows the different acquisitions of ASTER at different field reflectance measurements of the gravel and bitumen
dates (i.e., different shades). A mosaic of ASTER runways is not reproduced by the ASTER L2 data (Fig.
radiance data as a simple ratio of bands 4 and 7 yields 4). In addition, the shape of the ASTER SWIR signatures,

Fig. 3. a) ASTER acquisitions for seven different dates for the eastern Broken Hill portion of the Curnamona study area (blue). b) ASTER L1b band ratio b4 /
b7. Broken Hill Domain outcrop in red. c) ASTER L1b band ratio b7 / b9. Broken Hill Domain outcrop boundary in red. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)
 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172 163

Fig. 4. Comparison between ASD field VNIR-SWIR measurements and ASTER L2 surface reflectances for Airfield 1 and Airfield 2 validation sites at
Broken Hill.

particularly at bands 5 and 9, shows a significant anomalies’’ in SWIR ratio image products. Although the
difference compared with the field measurements, due ASTER operator, GDS, uses an automatic cloud identi-
largely to the uncorrected crosstalk effect (Iwasaki et al., fication algorithm that attempts to screen all Level 1A
2001). These results emphasize the importance of correct- scenes with greater than 20% cloud cover, this process is
ing crosstalk in ASTER SWIR radiance data. still undergoing improvements and can sometimes be
Gain factors used for mosaicking Level 1B SWIR problematic in areas of limited outcrop, such as in the
radiance-at-sensor data in this study, were empirically Curnamona Province.
derived from the band means of overlapping ASTER scene An algorithm for masking clouds and their shadows
areas acquired during different satellite orbits. For most was developed in this study using thresholded ASTER
scenes, band means were derived automatically from L1B bands 10 and 3 radiance data, respectively. The low
overlapping areas using image-derived statistics. However, albedo observed from cloud shadows in VNIR band 3
areas with clouds and associated shadows required manual images, and the low radiant temperatures of thick cloud
definition of the overlapping scene areas to extract reliable tops observed from TIR band 10 images, enabled the
band-mean statistics. Gains were subsequently calculated successful mask development of this ASTER data. It was
relative to a chosen reference image (e.g., Broken Hill found, however, that manual rather than automated histo-
ASTER scene) from the ratio of band means and applied to gram thresholding was required to limit the masking to
each scene acquired from the same orbit and date. This cloud-related features instead of possible geological or
simple method effectively yielded seamless images across topographic-related effects (i.e., shadowed areas from
14 different ASTER acquisition dates for each SWIR band, sharp relief). In the example shown, band 3 and band 10
having adjusted the L1B data into ‘‘apparent’’ radiance units (Fig. 5a,b) were thresholded to produce a mask (Fig. 5c).
relative to a reference scene. The same set of SWIR gains Crosstalk-corrected L1B SWIR images were processed to
could generally be used for all ASTER scenes collected generate AlOH (including muscovite) abundance maps
along the same orbit for most cases, although an exception using band combination [(b5+b7)/b6] (Rowan & Mars,
was observed for ASTER acquisition straddling the Barrier 2003); however shadowed areas are falsely highlighting
Ranges at Broken Hill with likely localized water vapor high AlOH content (Fig. 5d). This is predominantly the
variations. result of uncorrected residual crosstalk additively contri-
Clouds represent a potential problem for seamless buting to the ASTER signal. Application of the cloud
geological mapping for several reasons. Firstly, both mask (Fig. 5c) to the AlOH abundance image produces an
clouds and their shadows can obscure the underlying improved map result, removing most of the cloud-induced
surface. Secondly, clouds can affect those mosaicking artifacts but still highlighting small outcrops in the extreme
procedures that rely on scene statistics. Hence, clouds south east of the scene (Fig. 5e).
should first be identified and then masked out. It was also
observed that residual crosstalk effects could still be 3.2. TIR Pre-processing
present in high- and low-albedo areas, including those
associated with clouds, and especially within their shadows A comparison between ASTER emissivity data and the
(Fig. 5a – e). These residual effects produced ‘‘false average of TIR spectral field measurements using Design
164 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172

Fig. 5. a) ASTER L1B band 3 highlighting cloud and associated shadow; b) ASTER L1B thermal band 10 highlighting cloud; c) cloud mask generated from
thresholded bands 3 and 10; d) ASTER L1B generated AlOH abundance anomalies. Light areas indicate interpreted high AlOH abundance; e) ASTER L1B
generated AlOH abundance imagery masked for clouds and associated shadows. Note the interpreted AlOH anomalies that were associated in Fig. 5d with
cloud covered areas.

and Prototypes’s microFTIR 101 (Hook & Kahle, 1996), apparent residual temperature. Some discontinuous line-
resampled to ASTER spectral resolution, showed similar striping, related to systematic drift in instrument response,
signatures for several validation sites including the currently was apparent; however the minor nature of this problem,
unused North Broken Hill Mine dump (Fig. 6a). The despite the relatively low signal to noise of these TIR
consistency of the ASTER-derived surface emissivity data, indicated that the final geological products were not
signatures for different acquisition dates (i.e., acquired at severely compromised by this problem.
different temperatures and/or atmospheric conditions) was
also examined using two ASTER overpasses (Fig. 6b).
Overall the ASTER derived emissivities, under different 4. Strategies for generating ASTER seamless geologic
conditions, essentially showed similar signatures though maps
small spectral variations can be observed in detail,
especially in band 14 (¨11.3 Am) (Fig. 6b). An overview of the various pre-processing steps devised
Initial attempts at generating seamless, accurate, geo- for this study dealing with the issues described above is
logical information products derived from ASTER L2 TIR illustrated in Fig. 7. These steps are time-consuming and
surface emissivity data yielded images that showed no automation of this methodology is needed, particularly
 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172 165

Fig. 6. a) Comparison between field mFTIR and ASTER L2 emissivity signatures at Broken Hill Pit Dump, b) repeatability of ASTER L2 surface emissivity
signatures for two different acquisitions (solid vs dashed lines) at Broken Hill.

involving the calculation of gain factors for the adjustment absorption features at ASTER resolution that can be
of variable illumination and atmospheric conditions between potentially confused with carbonate (Fig. 8a). Garnets
different acquisitions. (e.g., almandine, spessartine) are common high-grade
metamorphic minerals within certain units at Broken Hill
are important indicators of Broken Hill-style base metal
5. Mineral group spectroscopy at ASTER spectral mineralisation (Spry & Wonder, 1988). Garnets display
resolution broad VNIR and TIR features at ASTER spectral
resolution, especially from bands 3 to 4 and 12 to 13,
The ability of laboratory-based VNIR, SWIR and TIR though possible confusion with the spectral features of
spectroscopy to measure and enable identification of green vegetation at VNIR wavelengths and some mafic
minerals and mineral groups has already been established silicates at TIR wavelengths needs to be considered (Fig.
for several decades (Hunt & Ashley, 1979; Lyon & Burns, 8a,b). Feldspars (e.g., albite, anorthite) are common and
1963; Vincent et al., 1975; Vincent & Thomson, 1972). In important for indicating alteration associated with albitisa-
particular OH-bearing minerals and other silicates have tion in the Curnamona Province. However, the lack of a
been shown to display diagnostic spectral features within spectral band in the 9.6 Am region, because of strong
the SWIR and TIR wavelength regions, respectively (Clark atmospheric ozone absorption experienced from a satellite
et al., 1990; Grove et al., 1992; Salisbury & D’Aria, platform, renders feldspar mapping difficult at ASTER
1992). Resampling of VNIR-SWIR and TIR mineral spectral resolution.
library spectra (Clark et al., 1990; Grove et al., 1992, The modeled ASTER SWIR spectra of AlOH minerals
Salisbury & D’Aria, 1992) to ASTER spectral resolution (e.g., kaolinite, Al-poor and Al-rich mica) displayed in Fig.
provides a basis for understanding the potential limit of 8a indicate changes in the symmetry of the AlOH
extracting mineral (group) information from ASTER (Fig. absorption feature centered at 2.2 Am, or ASTER band
8a and b). In particular, MgOH minerals such as chlorite 6. Previous work by Duke (1994) has shown that white
and hornblende, have limited diagnostic SWIR spectral mica chemistry (e.g., muscovite/illite, phengite), particu-
larly its Al content, can be inferred by the wavelength of
its 2.2 Am absorption feature. Duke (1994) showed that Al
poor micas (e.g., phengite) display a AlOH absorption
feature with a longer wavelength than Al-rich micas. On
the basis of this absorption feature, observed by ASTER
bands 5, 6 and 7, an estimate of AlOH abundance was
estimated by the ASTER band combination, (b5+b7)/b6
(Rowan & Mars, 2003), and inferred white mica compo-
sition by band ratio indexes b5/b6, b7/b6 and b7/b5.
Resampling of muscovite library spectra into ASTER
equivalent spectra suggested that a high b5/b6 and low
Fig. 7. Overview of pre-processing strategies for mosaicking ASTER b7/b6 can represent a longer wavelength mica absorption
imagery at Curnamona. feature compared to the converse situation of a low b5/b6
166 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172

Fig. 8. a) Laboratory (solid lines) and ASTER equivalent resampled (dashed) VNIR – SWIR mineral library reflectance signatures; b) laboratory and ASTER
equivalent resampled TIR mineral library emissivity signatures.

and high b7/b6 result. It is also suggested by these 9a– d). In these images, brighter areas represent surface
resampled spectra, that kaolinite may be discriminated materials with deeper absorption features which are
from white mica using b7/b5. In a similar way, abundances assumed to be associated with higher abundances of AlOH,
of MgOH (e.g., chlorite, hornblende) and carbonate (e.g., MgOH-carbonate and quartz. In particular, the brightest
calcite) group minerals were estimated by the ASTER areas highlighted in Fig. 9b correlate with the mica-rich
band parameter, (b6+b9)/b8, based on their 2.33– 2.35 Am outcrops and associated colluvium within the Broken Hill
(band 8) absorption feature. The presence of ferrous iron in and Olary Domains (‘‘A’’ and ‘‘B’’, respectively) while
MgOH silicates can also display a steady rise in the SWIR quartz-rich areas tend to be associated with alluvial outwash
spectral reflectance signature approximately from 1.0 to and accumulations within the Lake Frome Basin (‘‘C’’) (Fig.
2.0 Am. This trend can be preserved at ASTER spectral 9d). The MgOH-carbonate abundance image product high-
resolution as indicated for the chlorite and hornblende lights Adelaidean carbonate-rich units south of the Olary
spectra (Fig. 8a). Estimates of the ferrous iron content Domain (‘‘D’’) and to a lesser extent, amphibolite-rich units
using the ASTER band ratio b5 / b4 were also examined in within the Broken Hill Domain (Fig. 9c).
this study. Comparisons between HyMap (Robson et al., 2002) and
Quartz has a pronounced reststrahlen TIR spectral feature ASTER data for the AlOH abundance image, [(b5+b7)/b6]
within the 8 to 9.2 Am region producing a diagnostic in the Broken Hill Domain (Area I and Area II, Fig. 9a),
emissivity signature that can be also observed by ASTER show the accuracy of the mosaicked ASTER SWIR data
bands 10 to 12 (Fig. 8b). Phyllosilicates (e.g., muscovite, product (Fig. 10a,b). The ASTER and HyMap derived
kaolinite) by comparison generally display longer TIR AlOH abundance (inverted) maps both show that the
wavelength spectral features between 8.6 to 9.6 Am or northerly and northeasterly AlOH-rich units (dark areas)
ASTER bands 11 and 12. Consequently in this study decrease in white mica abundance towards the southwest
ASTER band ratio b13 / b10 was used to map quartz-rich (Fig. 10a, b). There is also a correspondence between the
units and regolith (Fig. 8b). ASTER derived AlOH abundance image and the radio-
metric potassium concentration obtained from airborne
geophysical surveying (Fig. 10c) (Robson & Spencer,
6. ASTER seamless geologic maps for the Curnamona 1997). No explanation is available as yet for these spatial
Province patterns of K-mica abundance although it appears to be
associated with trends also associated with metamorphic
Seamless images were generated from the ASTER L1B retrograde alteration at Broken Hill (Corbet & Phillips,
SWIR radiance data to map major geological units rich in 1981). It is also interesting to note that the Broken Hill type
Al – OH, Mg – OH/carbonate and quartz abundances (Fig. deposits are located in areas relatively poor in muscovite.
 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172 167

III
C

IV
A
B
II I

D
a) b)

c) d)

Fig. 9. a) Study area for the ASTER Curnamona Project (cyan) showing the geological outcrops (grey) (AGSO, 2000) and 1 : 250,000 map sheets
encompassing the Curnamona Province; b) ASTER derived AlOH abundance imagery; c) ASTER derived MgOH-carbonate abundance imagery; d) ASTER-
derived quartz abundance imagery. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 10. a) ASTER derived AlOH abundance within the Broken Hill Domain and surrounding regolith areas (Area I—Fig. 9 a) Broken Hill Mine indicated by
‘‘X’’ ; b) HyMap derived AlOH abundance; c) Airborne geophysics derived radiometric potassium concentration.
168 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172

Further investigation is required to investigate its relevance studied in Area III (Figs. 9a and 11a) within the units of the
for the alteration history of the Broken Hill deposit. Olary Domain (Fig. 11c) containing several MgOH/
The ability of ASTER band ratio b5 / b4 to map carbonate anomalies (Fig. 11b). In particular amphibolite/
lithologies rich in ferrous iron silicates was examined and calcalbite units (396000E, 6434800N; dark green Fig. 11a)

Fig. 11. a) Olary Domain geological outcrop boundaries (Area II—Fig. 9 a) (PIRSA, 2000); b) ASTER-derived MgOH/carbonate abundance; c) ASTER-
derived ferrous iron silicate index; d) MNF bands 1, 2 and 3 of ASTER TIR surface emissivity data; e) MNF bands 2, 3, and 4 of ASTER TIR surface
emissivity data.
 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172 169

are discriminated by both the ferrous iron silicate (Fig. 11c) Fraction transformation (Green et al., 1988) applied to the
and the MgOH/carbonate image (Fig. 11b) products. Using L2 surface emissivity TIR product. However the emissivity
both of these products appears to offer the potential to data proved noisy with limited dimensionality and pro-
discriminate between MgOH group minerals and some duced no clear discrimination of the carbonate and MgOH-
carbonate minerals. In this region, the amphibolite-rich rich units (Fig. 11d,e).
units at 399500E, 6437200N are highlighted in both image A comparison between the AlOH absorption wavelength
products (Fig. 11b,c), whereas the Skillogalee and Auburn measured using the hyperspectral airborne data (Fig. 12d)
Dolomites, within the Burra Group (401800E, 6437600N; and the ASTER RGB band ratio AlOH imagery (Fig. 12c)
burgundy, Fig. 11a), are highlighted only by the MgOH/ was undertaken along the north-western margins of the
carbonate product (Fig. 11b). Note that this interpretation Broken Hill Domain (Area IV, Figs. 9a and 12a). The area
and discrimination could be complicated by ferrous-bearing straddles the Mundi Mundi fault line and associated
carbonates (e.g., ankerite, siderite and ferroan dolomite) in escarpment where an apron of alluvium and colluvium
other geological settings. Attempts at generating ASTER- flanks the uplifted Broken Hill Domain to the southeast
derived mosaics, representing ferrous iron content across (Fig. 12a). Spectral profiles of the hyperspectral HyMap
the Curnamona, proved problematic and showed significant imagery reveals that areas of blue (‘‘A’’), green (‘‘B’’) and
differences between image acquisition dates. This is likely red (‘‘C’’) shown in Fig. 12d correspond to a measurable
to be the result of residual crosstalk contributions for bands increase in the wavelength of the main AlOH spectral
4 and 5 used in the ratio product b5 / b4. In particular this feature (Fig. 12e). The ASTER RGB AlOH image on the
product would be more sensitive to crosstalk, reducing other hand shows a qualitative similarity only to the HyMap
band 4 and increasing band 5 in a non-linear manner results (Fig. 12c). Also in some areas mapped by the
related to average radiance levels present at each acquis- HyMap data as AlOH poor (e.g., ‘‘D’’, Fig. 12d), the
ition. Attempts were also made to discriminate these processed ASTER results wrongly highlighted areas of
MgOH and carbonate units using the Minimum Noise high-abundance Al-poor mica (Fig. 12b,c).

Fig. 12. a) Geology of the Broken Hill Domain – Mundi Mundi Escarpment (Area III—Fig. 10 a); b) ASTER derived AlOH abundance; c) ASTER derived
AlOH composition; d) HyMap-derived AlOH wavelength index from short (blue: 2.195 Am) to long (red: 2.207 Am) wavelength; e) HyMap SWIR spectral
signatures. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
170 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172

A series of field transects were conducted as part of the measurements (Fig. 13b) were shown to compare favour-
validation of ASTER image products at Broken Hill. These ably with the HyMap image signatures (Fig. 13c) within the
transects involved a series of closely spaced measurements SWIR wavelength region. The ASD and HyMap signatures
at 1-m intervals for approximately 150 m across the corresponding to retrograde shear schists, amphibolites, and
geological strike of contrasting Broken Hill outcropping psammite units are highlighted by grey, green and blue
units. Spectral reflectance measurements of soil and rock spectra respectively (Fig. 13b and c). Comparisons between
outcrop were measured along these transects with the ASD the New South Wales Geological Surveys published
Fieldspec Pro VNIR-SWIR spectrometer to compare with 1 : 25,000 geology (Fig. 13a) and the ASD and HyMap
ASTER spectral signatures. Along one of the transects, signatures were reasonable, showing AlOH (i.e., 2.2 Am)
ASD field measurements and samples were collected across absorption features for the mica-rich schists and psammite/
several narrow (~20 –50 m) amphibolite, retrograde schist psammopelite units (Fig. 13b and c). The ASD and HyMap
and psammite units (Fig. 13a). The ASD field spectral signatures also identified the MgOH’s spectral feature

Fig. 13. a) Darling Creek field traverse with geology (Area IV—Fig. 9 a) and ASD measurements (white squares) along traverse; b) ASD SWIR signatures
acquired from traverse; c) HyMap SWIR sample signatures corresponding to traverse; d) HyMap-derived AlOH wavelength index (blue = 2.197 Am, red = 2.202
Am); e) AlOH composition interpreted from ASTER SWIR log residuals; f) Calibrated ASTER SWIR spectral signatures (* = ASTER band centers) for
intervals along field traverse listed. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172 171

between 2.3 to 2.33 Am associated with the amphibolite cularly for bands 5 and 9. Although ERSDAC’s crosstalk
units. Discrepancies between some of the geological correction software (Version 3.0) alleviated much of the
boundaries and field/image signatures were apparent; crosstalk problem for ASTER L1B data, artifacts were still
however this may also possibly be a result of inaccuracies apparent, especially for areas of low albedo (e.g., cloud
within the 1 : 25,000 mapping or the presence of colluvial shadows) and but also for areas of high albedo. As a
float material. consequence, masking for clouds (and water) is a critical
A wavelength index image was generated from the pre-processing step.
HyMap data to represent shifts in the wavelength position of Atmospheric radiative-transfer modeling revealed that
the AlOH 2.2 Am feature (Fig. 13d). This wavelength index variable water vapor associated with different climatic
represents possible changes in the white mica chemistry as models, is a significant issue for the ASTER L1B SWIR
suggested by Duke (1994) where blue areas indicate shorter data, particularly for bands 8 and 9. The effects of variable
wavelength (i.e., 2.197 Am) Al-rich mica compared to red atmospheric conditions and from variable solar illumina-
areas representing longer wavelength Al-poor mica (i.e., tion conditions appeared to be multiplicative from band
2.202 Am) (Fig. 13d). ASD field measurements along the ratio results of different ASTER acquisitions. Successive
transect shown (Fig. 13b) indicate a change for the AlOH mosaicking of ASTER SWIR data was subsequently
absorption feature from 2.203 Am to 2.199 Am northwards derived by the application of gains upon overlapping
as suggested by the HyMap derived AlOH wavelength ASTER SWIR L1B images from 14 different acquisitions.
index image (Fig. 13d). The ASTER derived AlOH The resulting mosaicked ASTER L1B SWIR images were
composition RGB imagery was also compared with the successfully processed using band ratios to measure the
ASD signatures and geology along the Darling Creek abundance of mineral groups including AlOH and MgOH/
Traverse. However it was clear that the coarser spatial carbonate within outcropping and/or regolith units of the
nature of the SWIR data (i.e. 30 m) limited ASTER’s Broken Hill-Curnamona Province. ASTER derived AlOH
discrimination and usefulness for mapping these narrow abundance imagery compared well with large-scale air-
geological units (Fig. 13a). borne hyperspectral HyMap survey results, providing
Crosstalk-corrected ASTER L1B radiance data encom- confidence in the application of multiplicative gains to
passing this area were calibrated using ASD field spectral adjust ASTER scenes acquired on different dates. During
measurements and processed into log residuals (Green & this study, it was also found that discrimination between
Craig, 1985) for comparison with ASD and HyMap MgOH- and carbonate-rich units was possible using
spectra. The resulting ASTER data produced SWIR ASTER if ferrous iron products were generated to assist
reasonable signatures with no obvious distortion of bands the discrimination of MgOH group minerals. Partial
5 and 9 (Fig. 13f). There is the suggestion of a shift to success was also achieved in generating seamless maps
more left symmetric AlOH absorption feature (i.e., from qualitatively representing AlOH composition from ASTER
red to yellow, Fig. 13e) corresponding to possibly shorter SWIR images. However, its reliability was significantly
wavelength mica chemistry north along the traverse (Fig. reduced in areas of low albedo and in the presence of
13e). However the spectral resolution of ASTER SWIR chlorite-rich units. Despite the application of crosstalk
bands precludes accurate estimation of the AlOH 2.2 Am correction software, some residual crosstalk effects still
absorption feature’s wavelength (Fig. 13f). These ASTER proved problematic.
transect results and their comparison with field and ASTER TIR L2 surface emissivity data compared
hyperspectral data indicate that ASTER has limited favourably with field spectral measurements and also
potential to provide compositional information for small produced reasonably consistent signatures, independent of
changes in AlOH chemistry as observed in Broken Hill, acquisition conditions. As a consequence, ASTER emissiv-
even assuming well-calibrated SWIR radiance data. ity data were mosaicked and processed to generate quartz-
abundance images and were found to highlight regolith
accumulations of alluvial quartz and some units of quartzites
7. Conclusions and sandstones.

Several pre-processing steps were required to generate

seamless imagery before band combination processing was Acknowledgement
applied to target specific mineral absorption features. These
steps included correction for additive SWIR crosstalk This work was financially supported and encouraged by
effects, SWIR band image offsets and also gain adjustments the Geological Survey of South Australia within the
for variable atmospheric/illumination conditions during Department of Primary Industries (PIRSA), as part of
different acquisitions. As part of validation studies under- PMD*CRC activities. Several individuals, particularly
taken for this study, field spectral measurements revealed within PIRSA, played a key part in this project including
that ASTER’s SWIR crosstalk effect was a significant Paul Heithersay, and Stuart Robertson and others within the
factor for the ASTER L2 surface reflectance data, parti- Curnamona Team. Japan’s ERSDAC provided ASTER data,
172 R.D. Hewson et al. / Remote Sensing of Environment 99 (2005) 159 – 172

crosstalk-correction software, and support during field Grove, C. I., Hook, S. J., & Paylor II, E. D. (1992). Laboratory
validation activities in Broken Hill. LPDAAC of the United reflectance spectra of 160 minerals, 0.4 to 2.5 micrometers. JPL
Publication, 92 – 102.
States Geological Survey also provided ASTER data. Hewson, R. D., Cudahy, T. J., & Huntington, J. F. (2001). Geologic and
Support was also gratefully received from CSIRO’s Glass alteration mapping at Mt Fitton, South Australia, using ASTER
Earth Project, its coordinator, Joan Esterle and co-worker, satellite-borne data. IEEE 2001 International geoscience and remote
Joanna Parr. US and Japanese members of the ASTER sensing symposium, 9 – 13 July.
Hook, S. J., & Kahle, A. B. (1996). The Micro Fourier Transform
Science Team provided technical feedback during this
Interferometer (AFTIR)—A new field spectrometer for acquisition of
research. Andrew Rogers modelled the effects of variable infrared data of natural surfaces. Remote Sensing of Environment, 56,
atmospheric conditions upon ASTER data. New South 172 – 181.
Wales Department of Mineral Resources and Geoscience Hunt, G. R., & Ashley, R. P. (1979). Spectra of altered rocks in the visible
Australia granted permission to publish radiometric data and near infrared. Economic Geology, 74, 1613 – 1629.
from the Discovery 2000 geophysical database. Processed Iwasaki, A., Fujisada, H., Akao, H., Shindou, O., & Akagi, S. (2001).
Enhancement of spectral separation performance for ASTER/SWIR.
HyMap data of Broken Hill were gratefully received from Proceedings of SPIE, The International Society for Optical Engi-
Peter Hausknecht of HyVista. neering, 4486, 42 – 50.
Lyon, R. J. P., & Burns, E. A. (1963). Analysis of rocks and minerals by
reflected infrared radiation. Economic Geology, 58, 274 – 284.
References PIRSA (2000). South Australian Geological Survey. Geoscientific GIS
Olary Region. Adelaide’ Primary Industries and Resources of South
Australia.
Abrams, M. J., Brown, D., Lepley, L., & Sadowski, R. (1983). Remote Robson, D., Cocks, P., & Taylor, G. (2003). Calibration, processing and
sensing for porphyry copper deposits in southern Arizona. Economic interpretation of hyperspectral data over the Broken Hill region. In
Geology, 78, 591 – 604.
proceedings of sixteenth Australian Society of Exploration Geophysics
Abrams, M., Hook, S., & Ramachandran, B. (2002). ASTER user
Conference, 16 – 19 February 2003.
handbook. JPL Publication, 2, 25. Robson, D. F., & Spencer, R. (1997). The New South Wales government’s
AGSO (2000). Broken Hill geoscientific GIS dataset. Canberra’ Australian Discovery 2000—Geophysical surveys and their effect on exploration.
Geological Survey Organisation.
Exploration Geophysics, 28, 296 – 298.
Berk, A., Anderson, G. P., Acharya, P. K., Chetwyn, J. H., Bernstein, L. S., Rowan, L. C., & Mars, J. C. (2003). Lithologic mapping in the Mountain
& Shettle, E. P., et al. (1999). MODTRAN4 users manual. Hanscom, Pass, California area using Advanced Spaceborne Thermal Emission
Mass’ Air Force Research Laboratory. and Reflection Radiometer (ASTER) data. Remote Sensing of Environ-
Clark, R. N., King, T. V. V., Klejwa, M., & Swayze, G. A. (1990). High
ment, 84, 350 – 366.
spectral resolution reflectance spectroscopy of minerals. Journal Geo- Salisbury, J. W., & D’Aria, D. M. (1992). Emissivity of terrestrial materials
physical Research, 95(B8), 12653 – 12680. in the 8 – 14 Am atmospheric window. Remote Sensing of Environment,
Conor, C. H. H., & Fanning, C. M. (2001). Definition of major igneous and 42, 83 – 106.
sedimentary units of the Olary Domain, Curnamona Province. MESA
Schowengerdt, R. A. (1997). Remote sensing—models and methods for
Journal of Primary Industries and Resources of S.A., 19, 51 – 56. image processing (2nd edR). San Diego’ Academic Press.
Corbet, G. J., & Phillips, G. N. (1981). Regional retrograde metamorphism Spry, P. G., & Wonder, J. D. (1988). Manganese-rich garnet rocks
of a high-grade terrain: The Willyama Complex, Broken Hill, Australia.
associated with the Broken Hill lead – zinc – silver deposit, New South
Lithos, 14, 59 – 73. Wales, Australia. Canadian Mineralogist, 27, 275 – 292.
Duke, E. F. (1994). Near infrared spectra of muscovite. Tschermak Stevens, B. P. J., Barnes, R. G., Brown, R. E., Stroud, W. J., & Willis, I. L.
substitution and metamorphic reaction process: Implications for remote (1988). The Willyama supergroup in the Broken Hill and Euriowie
sensing. Geology, 22, 621 – 624.
Blocks, New South Wales. Precambrian Research, 40, 297 – 327.
Fujisada, H., Sakuma, F., Ono, A., & Kudoh, M. (1998). Design and pre- Thome, K., Palluconi, F., Takashima, T., & Masuda, K. (1998). Atmos-
flight performance of ASTER instrument protoflight model. IEEE pheric correction of ASTER. IEEE Transactions on Geoscience and
Transactions on Geoscience and Remote Sensing, 36(4), 1152 – 1160. Remote Sensing, 36(4), 1199 – 1211.
Gillespie, A., Cothern, J. S., Rokugawa, S., Matsunaga, T., Hook, S. J., &
Vincent, R. K., Rowan, L. C., Gillespie, R. E., & Knapp, C. (1975).
Kahle, A. B. (1998). A temperature and emissivity separation algorithm Thermal-infrared spectra and chemical analyses of twenty-six igneous
for Advanced Spaceborne Thermal Emission and Reflection Radio- rock samples. Remote Sensing of Environment, 4, 199 – 209.
meter (ASTER) images. IEEE Transactions on Geoscience and Remote
Vincent, R. K., & Thomson, F. (1972). Spectral compositional imaging of
Sensing, 36(4), 1113 – 1126.
silicate rocks. Journal of Geophysical Research, 77(14), 2465 – 2472.
Green, A. A., Berman, M., Switzer, P., & Craig, M. D. (1988). A Yamaguchi, Y., Fujisada, H., Kahle, A., Tsu, H., Kato, M., Watanabe, H.,
transformation for ordering multispectral data in terms of image quality Sato, I., & Kudoh, M. (2001). ASTER Instrument performance,
with implications for noise removal. IEEE Transactions on Geoscience
operation status, and application to earth sciences. IEEE 2001 Interna-
and Remote Sensing, 26(1), 65 – 74. tional geoscience and remote sensing symposium, 9_13 July.
Green, A. A., & Craig, M. D. (1985). Analysis of aircraft spectrometer data
with logarithmic residuals. JPL Publication, 85-41, 111 – 119.