View
13
Download
4
Category
Preview:
Citation preview
Chapter 6
- 206 -
Chapter 6 Remote Sensing of the Regolith
The following Chapter documents the methods and results of the information
extraction performed on the remotely sensed datasets. The Chapter opens with the
processing and results from the lowest resolution imaging data, Landsat TM. Detailed
analysis of ASTER data was not performed over the White Dam area no results are
presented here. Detailed descriptions of the processing steps of ASTER are discussed in
the Appendix VI. The second portion of the Chapter documents the processing and
extraction of information from hyperspectral imagery. Before meaningful results could
be obtained from the HyMap data, pre-processing was required. Atmospheric
correction experiments are discussed in detail, followed by the process of information
extraction, which involved the use of indices and un-mixing techniques. This was
performed with the intention of producing seamless distribution maps of surface
materials and to aid regolith interpretations. Results from the hyperspectral processing
methods are followed by interpretations of the gamma-ray data, which was used for
regolith mapping of materials, along with ortho-photography and DEMs. The
interpretations of the regolith-landforms, identified using the information extracted from
the remote sensing datasets are discussed in the subsequent Chapter on regolith
mapping.
Incomplete coverage ASTER data was obtained for the study area. The four
tiles that consisted of the HyMap swath area were made up of one scene that was not yet
acquired and one that was significantly influenced by cloud cover. Pre-processing was
performed as described in Appendix VII and spectral processing was attempted on the
SWIR dataset using similar techniques to those used on the HyMap data (described
latter in this Chapter). The results from the ASTER data did not show any clear
indication that there were identifiable minerals, as the spectra of the pseudo-reflectance
data did not correspond to resampled ASD spectra of materials collected from the area.
The endmember analysis failed to find any significant results.
The process of more in depth analysis, such as the use of level 3 reflectance data
and the on-demand ordering of the two required scenes was not followed up, as
mineralogical information could be extracted from the higher resolution (spectrally and
spatially) HyMap data. By late 2001 Research by Dr. Robert Hewson (CSIRO Division
of Exploration and Mining/ CRC LEME) had also commenced on the analysis of the
I C Lau Remote Sensing Results
- 207 -
Curnamona region using ASTER data, which included the study area, and the projects
would have been considered to have been overlapping if both used ASTER data.
Landsat Dataset analysis In this research the information extracted from Landsat TM data served as a
benchmark to compare with the abilities of higher resolution targeted datasets. Landsat
TM is widely used in regional mapping by a wide-range of disciplines and has become
the standard remote sensing tool due to its extensive coverage (spatially and temporally)
and the widespread knowledge of the methods for extracting information. The
limitation of Landsat TM is the small number of broad bands and the large pixel size.
The characteristics of Landsat data were discussed in Chapter 4.
Landsat Pre-processing
Remote sensing data can be supplied in a array of processing levels, ranging
from raw data acquired at sensor, reflectance calibrated or processed end-products,
ready for interpretation. Some data suppliers offer processing services that allow the
user to quickly interpret features in the imagery rather than having to spend time
correcting for instrument and atmospheric effects.
Raw data consists of digital numbers (DN), which requires the application of the
sensor coefficients. Once these coefficients are applied, the data can be corrected to
radiance at sensor by the use of gains and offsets. This data can then be atmospherically
corrected to produce apparent or relative reflectance. Absolute reflectance can only be
determined if the incident EMR is known.
Raw Landsat 5 TM data were acquired georectified to AGD 66/AMG 54. The
data were map-to-map projected in ER Mapper (version 6.4) to GDA 94/MGA 54.
Dark pixel subtraction was performed to remove atmospheric backscattering effects and
published gains and offsets were used to convert the digital numbers to radiance.
Examination of the output bands suggested that the data were dominated by
albedo and lacked a great deal of variance (Figure 6.1a). The radiance data were
imported into ENVI and log residuals were calculated using CSIRO Exploration and
Mining Mineral Mapping Technologies Group (MMTG) Kwik Log Residual plug-in.
The resulting images displayed more variation than the raw imagery, reflected by lower
correlation statistics between the bands. The process also generated an albedo image
(Figure 6.1b), which was useful for creating masks of the bright and dark targets. The
high contrast of the LR results can be seen by comparing the corresponding LR bands in
Figure 6.1a, with the raw data in Figure 6.1c.
Chapter 6
- 208 -
Landsat Processing and Information Extraction
The results from the Log Residual corrections (Figure 6.1c) were examined to
determine the features highlighted by each spectral band and the results summarised in
Table 6.1.
Band TM1
In band TM1 swampy regions and the walls of dams had high DN (Figure 6.1c
i). The dams consisted of highly weathered kaolinitic saprolite or red-brown clayey
sands. The high values for this band were attributed to the presence of water in these
regions. Areas with a lack of vegetation, such as alluvium and soil-rich areas, had very
low DN. Some of the basement exposures and the road and railway track had high to
medium DN. Variation in the values for band TM1 for these materials was thought to
be related to their heterogeneity. The bituminised Barrier Highway contained a
multitude of repaired sections, observed during ground-truthing, and displayed a darker
appearance to the older portions (discussed in atmospheric correction of the
hyperspectral imagery and shown in Figure 6.9). These variations in the road material
were seen in the HyMap imagery and in ASD measurements of bitumen samples
collected along the highway.
Band TM2
Band TM2 was similar in appearance to band TM1, which was expressed in the
scene statistics, with a 0.829 correlation coefficient between the two bands. Dams, the
highway and swampy regions had high DN in band TM2 (Figure 6.1c ii). Areas of
badlands in the southern region of the swath had a high DN, where Adelaidean rocks
were exposed. Areas of soil and alluvium had a medium-low DN.
Band TM3
Band TM3 highlighted areas of creeks and bare soil, with high DN (Figure 6.1c
iii). The badlands region in the southern portion of the scene had high DN, whereas
swampy vegetated regions and dams had low DN. Areas of outcrop were contrasted by
the surrounding alluvial material, which had medium DN.
Band TM4
Band TM4 highlighted the drainage channels extremely well, with high DN
(Figure 6.1c iv). This was attributed to the abundance of vegetation along creeks and
I C Lau Remote Sensing Results
- 209 -
erosional depressions. Bedrock exposures appeared dark in band TM4 due to the low
reflection of EM radiation by rock forming minerals and clays.
Landsat Band
Feature Description
Swamps High values due to the presence of water Dams Water in dams Dam walls High reflectance materials such as quartz and no
vegetation Paved roads and railway
Dark materials with blue colour
TM1
Bare soil regions Absorption due to hematite and Fe-oxides Dam walls High DN TM2 Paved roads High DN Badlands High reflectance due to minimal vegetation and
abundance of exposed soil and RCA material Bare soil and creeks High DN
TM3
Swamps and vegetation
Low DN due to absorptions in the ‘red’ region of the VIS
Drainage channels Ephemeral creeks lined with green vegetation were highlighted
TM4
Saprolite Low DN due to low of reflectance in this region. Drainage and roads Low reflectance for these materials in this region
and appeared dark. TM5
Saprolite Adelaidean contrasts Willyama Supergroup basement due to low reflectance compared to the older rocks. This may relate to the ferruginous and darker Adelaidean rocks in comparison to the silicious and felspathic Willyama Supergroup lithologies.
TM6 Topography Northeast trending slopes had a higher DN than southwest slopes, especially in the MacDonald Corridor region
Alluvial areas Clays and vegetation cause low reflectance. Vegetation Absorption feature
TM7
Clays Absorption feature
Table 6.1 Summary of the individual band characteristics of Landsat TM Olary scene.
Chapter 6
- 210 -
Figure 6.1 Comparison of raw and Log Residual (LR) corrected Landsat TM data over the White Dam area. A greater variation between bands can be seen in the LR image, which had a lower amount of correlation between bands than the raw data.
I C Lau Remote Sensing Results
- 211 -
Band TM5
Band TM5 showed a low value for drainage features containing vegetation and
the highway, which displayed as a dark linear feature across the area (Figure 6.1c v).
The badlands region also displayed low DN, which contrasted the high DN of the
Willyama Supergroup basement exposures. Areas of alluvium and soils displayed
medium DN.
Band TM7
Band TM7 appeared very similar to band TM5, and also possessed a correlation
coefficient of 0.7 with TM5 and TM2. Drainage lines were highlighted with a low DN,
while Willyama Supergroup basement had a high DN. Soils had a moderate DN, with
similar values as TM5, as shown in Figure 6.1c vi.
Thermal Infrared Band TM6
The TIR band was examined to determine the amount of useful information that
could be extracted. The data indicated that topographic effects dominated the imagery
and showed that the northeast trending slopes had a higher albedo than the southwest
slopes (Figure 6.2). This may have been related to the presence of Adelaidean rocks on
the southern sides of ridges in the MacDonald Corridor region. The ridges in the
northern regions of the study area did not display variation in the TIR value with
different aspect, which was attributed the lower slope angle and single lithology of the
topographic features. Water bodies and alluvial areas had lower values than the
adjacent sheet flow dominated regions. This was attributed to the greater abundance of
vegetation colonising these features related to drainage.
False Colour Composite
The 741 RGB false colour composite (FCC) showed areas that appeared red-
magenta as representing saprolite exposures (Figure 6.3). Blue areas had a low
abundance of Fe-oxides and vegetation, as well as high clay contents. The blue regions
in the imagery also corresponded with dark coloured materials, such as the highway and
railway. Areas of healthy vegetation were highlighted in green. Creeks and alluvium
were highlighted as yellow areas due to high clay and vegetation content, while areas of
prominent quartz lag were displayed as white, representing a high reflectance in all
bands. Pale purple areas highlighted graded roads and quarries.
Chapter 6
- 212 -
Figure 6.2 (i) High-oblique and (ii) low-angle images of the 60 m resolution thermal band of Landsat TM, draped over a DEM of the White Dam area. The white rectangle represents the area of the HyMap coverage. Image is presented without illumination effects and displays higher values on slopes with a northeastern aspect, reflecting the orientation of the sun at the time of acquisition (mid-morning). Densely vegetated regions display a lower thermal value then areas of exposed basement and lag dominated terrains. Note the sub-horizontal line striping in (i) from instrument noise.
I C Lau Remote Sensing Results
- 213 -
Figure 6.3 Landsat TM RGB 741 false colour composite (FCC) of the southern portion of the White Dam area, showing the ability of a simple band combination to highlight different materials and features.
Band Ratios
Band rations were successful at further discriminating features from the White
Dam Landsat TM scene. The results of the ratios found to be useful are shown in Table
6.2. An RGB ratio composite of TM5/TM4, TM4/TM3, TM5/TM7 (Figure 6.4) was
used to highlight Fe-oxides, vegetation and clays, respectively. This combination was
shown to work well in the bedrock-dominated regions to the west of the White Dam
area. However, the study area was heavily influenced by vegetation and alluvial
materials, making the interpretation and discrimination of lithologies difficult.
Although the combination was found to be less than ideal for lithological information
extraction in the study area, band ratios were found to be useful for the discrimination of
regolith-landform units. The surficial features highlighted by the ratios were used as
mapping surrogates, such as areas of saprolite, alluvial plains and sheetflow-dominated
areas. Individual band ratio images corresponding to Table 6.2 are shown in Figure 6.5.
Chapter 6
- 214 -
Figure 6.4 Landsat TM FCC of ratios TM5/TM4 TM4/TM3 TM5/TM7 (RGB) demonstrating the ability to differentiate in situ regolith (Fe-rich) from transported materials (vegetated and clayey regions). Saprolite is displayed as bright red-pink areas. Vegetation had a blue-green colour and areas of bare soil or clays, with minor vegetation cover, were dark blue.
Directed Principal Component Analysis (DPCA)
Directed principal component analysis involves the use of selected or
constructed bands of data in a PCA to enhance correlated and uncorrelated information
(Fraser & Green 1987; Chavez & Kwarteng 1989). This process is useful for removing
specific features from a dataset that dominate or mask subtle features (Fraser & Green
1987). Fraser (1991) and Dickson et al. (1996) demonstrated the use of DPCA for
mapping of regolith materials, using Landsat TM ratios TM4/TM3 and TM5/TM7, to
discriminate clay and vegetated regions. Glikson & Creasey (1995) used DPCA and
band ratios to highlight Fe-oxides (TM5/TM4), clays (PC2 of DPCA) and vegetation
(TM4/TM3) for lithological and regolith mapping.
An RGB image of DPCA band 2 (of TM5/TM7 and TM4/TM3), TM5/TM4 and
TM7+TM1 was created to highlight regolith features (Figure 6.6). Significant
similarities can be seen in images of TM4/TM3 and DPCA band 1 Figure 6.5, which
were assigned to identify areas of vegetation.
I C Lau Remote Sensing Results
- 215 -
Ratio Feature Description TM5 TM5/TM3
TM3 (Iron-oxides)
Showed drainage channels, which contain more lithic fragments and less vegetation due to weathering of basement rocks, as having a dark appearance. The Adelaidean badlands were dark. Alluvial areas around White Dam water-feature were dark. Some areas of Willyama were bright and had high values.
TM3 (Iron-oxides) TM3/TM1 TM1
Vegetated drainage appeared dark, as does the Willyama Supergroup basement. The areas of bare soil on the depositional alluvial plains appeared bright.
TM5 Band TM5 had an absorption feature due to water for vegetation
TM5/TM4
TM4 Band TM4 had a high reflectance due to leaf cell structures
Showed some regions of green vegetation but mostly shows Willyama saprolite-rich areas as dull highs. The Adelaidean saprolite and colluvial areas were displayed as lows. Structural trends were seen in the northwestern region of the image, outside of the field area. (Fe-oxide) highlights areas of bedrock as being areas with a high DN and green vegetation in the alluvial channels as areas with low DN. Much of the Adelaidean had a dark tone, which may be due to vegetation or the lack of ferruginous material.
TM4 (Green vegetation) TM4/TM3 TM3 (Iron-oxides)
Enhanced green vegetation, channels and creeks. Clay-rich areas were dark as were most of the regions of outcrop and colluvial regions. Swampy regions and vegetated drainage channels were bright while the vegetated paddock is considerably brighter than the two neighbouring paddocks to the north and south. Greatly affected by vegetation, shown by paddock boundaries.
TM7 TM7/TM4
TM4
Highlights all drainage in dark tones, with the Willyama Supergroup rocks highlighted as bright tones. Alluvial and soils between the channels appeared medium)
TM5 (Dry vegetation) TM5/TM7 TM7 (Clay)
Highlighted creeks where swampy and green vegetation occurs. This ratio highlights regions that had absorption features in band TM7 such as clays and micas Swampy areas in channels were high as were channels containing vegetation. Adelaidean rocks were dark. The paddock where vegetation is thicker appeared lighter than the sparsely vegetated paddock.
TM7 TM7+TM1 TM1
Shows vegetated regions as lows while areas of high albedo such as bare soil and quartz were displayed as highs. Colluvial regions in the Adelaidean badlands also occur as highs
Table 6.2 Description of the results from selected Landsat TM band ratio combinations that were useful for highlighting features in imagery of White Dam area.
Chapter 6
- 216 -
Figure 6.5 Band ratio results of selected band ratios of Landsat TM imagery over the White Dam area.
I C Lau Remote Sensing Results
- 217 -
Figure 6.6 Directed Principal Components Analysis and ratio composite image (RGB: DPCA2 TM5/TM4, TM7+TM1) of Landsat TM data, highlighting regolith materials. Red were clay-rich areas, bright greens were Willyama Supergroup-derived saprolite, blues correspond to the Adelaidean metasediments, the badlands/erosional region and areas containing abundant lithic and quartz gravel lags.
The composite image showed clay-rich areas adjacent to creeks and overbank
deposits as red areas. Saprolite exposures were highlighted as bright green and blue
areas, with the later corresponding to Adelaidean lithologies. Purple-blue hues
corresponded to regions of soil quartz lag and colluvial lag.
Summary of Multispectral Imagery for Remote Sensing of the
White Dam Area
Landsat TM was found to be useful at mapping broad scale mineral and
vegetation associations, which related to regolith-landforms. Alluvial landforms were
prominently found to be highly colonised by photosynthetic vegetation making band
TM4 useful for mapping these units. Soils and bare areas typically displayed higher
values in band TM5. Sheetflow dominated landforms were more difficult to identify
due to their high variability, relating to regolith materials, surface lags and vegetation
communities. Areas colonised by chenopod shrublands appeared different from areas
containing prominent quartz lag. The use of band ratios and other enhancement
techniques were useful for further discriminating the regolith-landform types. Directed
PCA results offered a slight improvement in the tones of features.
Chapter 6
- 218 -
The processing of ASTER imagery, although higher resolution and containing
more bands, was problematic. This was attributed to the minor technical problems
persisting in the relatively new data. The amount of pre-processing of ASTER data
were similar to what is required of hyperspectral imagery, although the final product is
inferior in spectral and spatial resolution. ASTER offers a reasonable compromise over
airborne hyperspectral data (or space-borne data such as Hyperion) as a first pass tool or
for targeting the acquisition of higher resolution data.
Hyperspectral I7mage Processing Advancements in the spectral and spatial resolution of sensors, as demonstrated
by hyperspectral imagery, has allowed the improvement of geological mapping due to
the greater capacity of these sensors to differentiate different landscape features,
regolith materials and minerals over air photographs and early multispectral satellite-
born sensors. Hyperspectral remotely sensed data allows the operator to perform
traditional processing techniques, similar to those used on multispectral data, as well as
detailed mineralogical information extraction. To generate a satisfactory interpretation
of the data, an adequate understanding of the spectral properties of materials that
constitute the regolith-landform units is necessary.
Increased spatial resolution of satellite imagery has allowed users to not only
discriminate cover types, but also start to identify materials in broad groups (such as
clays, Fe-oxides, vegetation). A major problem with low spatial resolution is the
mixing of the spectral signal from the materials within the instruments ground pixel
area. Sensors mounted on aerial platforms allow the data to be acquired at much lower
altitude, and as a benefit, produce imagery with a much higher spatial resolution. The
use of higher spatial resolution sensors that cover a larger amount of the EM spectrum
(with higher spectral resolution) has resulted in the greater ability to identify the
individual minerals and constituents of surficial materials. Mixing is still apparent in
hyperspectral imagery but the greater number of bands improves the likelihood of
resolving the individual spectral features of the components relating to the spectral
response.
Pre-processing
Before meaningful information could be extracted from the White Dam
hyperspectral imagery the data required correction for atmospheric effects and
acquisition errors. This pre-processing involved a series of steps that were required to
be performed in order. Deviation from the sequence can cause differences in the
I C Lau Remote Sensing Results
- 219 -
processing outcomes, as many of the steps involve data statistics. A flowchart of the
pre-processing method used on the HyMap imagery is outlined in Figure 6.7, which also
shows a variety of atmospheric correction methods that are not discussed in this
Chapter. More information on the alternative methods of correcting for the atmosphere
is documented in Appendix VII.
Figure 6.7 Flowchart of pre-processing steps
Atmospheric Correction-Theory
Radiative Transfer (RT) Modelling
Radiative transfer models such as the ATmosphere REMoval (ATREM)
program (Gao & Goetz 1990; Gao et al. 1992; Gao et al. 1993) use scene information
and spectral features of the water vapour absorptions to model the atmosphere at the
time of flight. This method attempts to correct for the solar irradiance atmospheric
scattering and absorption as determined for the general conditions at the time of data
acquisition.
Each pixel is examined for water vapour by analysing the bands at 0.94 µm and
1.14 µm. Other gases (O2, O3, CH4, CO, CO2 and N2O, see Chapter 4), which absorb
incident radiation causing noise in the data, were modelled using the user-entered scene
Chapter 6
- 220 -
information. The transmittance spectra of the atmospheric gases were derived using the
Malkmus narrow band model (Malkmus 1967) based on the HITRAN 92 (Rothman et
al. 1992) database and the 5S (Simulation of the Satellite Signal in the Solar Spectrum)
code (Tanré et al. 1986), which is used to model scattering effects due to atmospheric
molecules (Rayleigh scattering) and aerosols.
User-entered information consisting of flight height, time, date, visibility and
location (latitude and longitude) were used to calculate the solar zenith angle. The
modelling uses the solar zenith angle to calculate information on the solar irradiance
above the atmosphere, which can then be used to divide the radiance at sensor values to
obtain apparent reflectance. Information on the dominant surface materials (green
vegetation, Fe-oxides etc) was used to make small adjustments the model.
Relative depths of the water absorption bands were used to calculate the water
vapour content of each pixel. This is used along with the transmittance spectra, derived
from atmospheric gases, to calculate scaled surface reflectance (Teliet 1989; Gao &
Goetz 1990; CSES 1992). The advantage of RT techniques is the correction of
hyperspectral data without a priori knowledge (Dwyer et al. 1995).
There were a range of software programs available to model the atmosphere
including ATREM (Gao & Goetz 1990; Gao et al. 1992; CSES 1999) ACORN
(Atmospheric Correction Now) (Analytical Imaging and Geophysics LLC 2002; Miller
2002) HyCorr (Hyperspectral Correction), HATCH (High-Accuracy Atmosphere
Correction for Hyperspectral Data) (Qu et al. 2000; Goetz 2002) and FLAASH (Fast
Line-of-sight Atmospheric Analysis of Spectral Hypercubes) (Goetz 2002). HyCorr
was developed by CSIRO Division of Exploration and Mining Mineral Mapping
Technologies Group (MMTG) based on ATREM and is the main atmospheric
correction software used in this research. Other RT models not mentioned above
include MODTRAN (Berk et al. 1998; Berk et al. 1999) and LOWTRAN (Kneizys et
al. 1988), which operate in a similar manner to ATREM, but use additional ground
based measurements to characterise the thermal structure and water content of the
atmosphere (Aspinall et al. 2002).
The ATREM (and HyCorr) software use a three band ratioing technique in the
region of the water absorption features to calculate the parameters of the atmospheric
gases, which occur as high frequency features. This interpolation can induce errors in
the radiative transfer model, resulting in the output scaled reflectance image possessing
systematic errors and spikes (Berk et al. 1989) (Figure 6.8). The ATREM model does
I C Lau Remote Sensing Results
- 221 -
not account for the coupling effect between aerosol scattering and gaseous absorption,
which predominantly occurs when viewing dark surfaces (Berk et al. 1989). Problems
also arise in the ATREM correction when the surface reflectance spectrum is not linear
over the wavelengths where the three band ratioing technique is situated. Non-linear
reflectance occurs in Fe-oxide-rich soils and for wet green vegetation (Qu et al. 2000).
An attempt to counter this problem was integrated into the HyCorr software, which
allowed the option of using only the 1.14 µm region when the scene environment is-rich
in Fe-oxides.
Improvements to the Malkmus band model had been performed in HITRAN 96,
99 and 2000 (Rothman et al. 2003), which were incorporated into newer atmospheric
correction software such as HATCH and FLAASH. Unfortunately these programs were
not available to the author at the time of this research to compare their performance.
Spectral Smoothing
A spectral smoothing program called Empirical Flat Field Optimal Reflectance
Transformation (EFFORT) (Boardman & Huntington 1996; Boardman 1998) had been
developed to remove the noisy features remaining after atmospheric correction. The
EFFORT process examines the data to extract an average featureless spectrum by
applying a Legendre polynomial fit to each spectra to determine how well it is
modelled. A linear regression technique is used to calculate gains, which were applied
to the featureless spectra with the intension of making them have the same appearance
as the modelled spectra. These gains were applied to the remaining data to remove the
noise and spikes. Spectral features of key minerals and other components that may have
been removed by the smoothing process spectral features are reinstated using the
‘Reality Boost’ option. The option was designed to improve the appearance by
emphasising absorption features of target minerals that may have been removed as part
of the noise. Hence, in theory, making it easier to identify endmembers in the processed
spectra.
Problems had been encountered with the ATREM-EFFORT correction
technique, regarding the comparison of corrected data with reference and field derived
spectra (Bierwirth et al. 2002; Lau et al. 2003). This is due to EFFORT smoothing or
completely removing diagnostic mineral absorption features during the process. The
Reflectance-Mean Normalisation (RMN) technique was used by Bierwirth (2000) and
Bierwirth et al. (2002) to correct this problem. The RMN involves the correction of the
Chapter 6
- 222 -
band means of the scene to fit an estimated overall mean of reflectance calculated from
field spectra or known endmembers.
Empirical Line (EL) Method
The use of spectral measurements of materials from within the scene using field
or laboratory spectrometers to correct the airborne imagery has been well documented
(Clark et al. 2002). The term Empirical Line (EL) Calibration (Roberts et al. 1985;
Conel et al. 1987; Kruse et al. 1990) has been used for the technique where at least two
materials with contrasting albedo (bright and dark) were used to correct spectral data.
The technique requires a uniform area within the swath where field measurements were
collected and deconvoluted to the imagery spatial bandwidths.
These averaged and deconvoluted field spectra were used to convert the data
from at-sensor radiance values (microwatts per square centimetre per nanometre per
steradian - µW cm-2 nm-1 sr-1) to percent reflectance using a linear regression technique.
A series of gains and offsets were calculated for each band based on the difference
between the actual spectrum and the calculated spectrum, (Ben-Dor & Kruse 1994;
Dwyer et al. 1995 Perry 2000). The model assumes that the entire scene is
topographically flat, with no scan angle or view angle effects (Perry 2000). This basic
assumption causes problems when large variations in elevation occur within a scene
(Ferrier 1995). Complications may occur with locating the sample pixel on the ground
or conversely, finding the field site corresponding to the spectral sample collected from
the imagery.
Comparison of Atmospheric Correction Techniques
Perry et al. (2000) found the empirical line and model based methods produce
comparable results on Hyperspectral Digital Imagery Collection Experiment (HYDICE)
data, with the ATREM model producing a better correction for the VIS/NIR and SWIR1
(0.4-1.9 µm) regions and the empirical line better for the SWIR2 (greater than 1.9 µm).
Analysis of hyperspectral imagery for silicate minerals and alteration products requires
technique with the greatest accuracy in the >2.0 µm region, therefore the empirical line
correction is the ideal calibration technique. Correction of imagery acquired over
inaccessible or in extremely rugged terrains using a model-based technique would to
produce sufficient results to produce identifiable spectra.
Dwyer et al. (1995) used EL and ATREM models to correct high altitude
AVIRIS data and performed automated extraction techniques to prove that the model-
based correction produced comparable results. ATREM tended to weaken the size of
I C Lau Remote Sensing Results
- 223 -
the mineral absorptions, reducing the effectiveness of automated feature extraction. It
was noted that the overcompensation of the 0.9 µm water absorption by version 2 of
ATREM, masked Fe features in this region. HyCorr and latter versions of ATREM (i.e.
version 3) incorporated the use of only the 1.14 µm water band for the water vapour
calculation if the scene consists of abundant Fe-minerals (CSES 1997). Farrand (1997)
used a modified flat field (MFF) correction (Mustard & Pieters 1987) on the ATREM
data to remove systematic features proximal to the wavelengths of the atmospheric
features.
Spectral Subsetting
Subsetting of the hyperspectral data (Kaufmann et al. 1991) has been shown to
improved the results of mineral mapping by allowing a more focused analysis of the
spectral regions where diagnostic absorptions occur (Taylor et al. 1994). The process of
endmember extraction on the SWIR region alone is performed on hyperspectral data in
the attempt to remove the influence of green vegetation on the processing techniques.
Conversely, mapping of Fe-oxides requires the VNIR wavelength region, which can be
influenced by vegetation. Spectrally subsetting the data can led to problems with the
interpretation of the SWIR region, where vegetation and shadow influences create
irregular absorption features that interfere with mineral identification. The
identification of the mixed spectral signatures may require the reversion back to the
original data, containing the full spectral range.
Pre processing of the HyMap Imagery Part1-Atmospheric
correction
The HyMap data used in this analysis consisted of 5 runs acquired in November
1998. The data had a spectral resolution of 5 m and an average overlap of ~500 m
between the sub-parallel swaths, which were orientated in an northeasterly direction.
HyCorr + EFFORT and Reality Boosts
The raw HyMap data (Figure 6.8a) were received from MIM Exploration Pty Ltd
in an unprocessed state. Each swath consisted of two portions, which were required to
be mosaiced into full lines before any further processing was performed. A calibration
file was obtained from HyVista Pty Ltd and the data were atmospherically corrected
using HyCorr at PIRSA by Mauger, A.J. 2001 (pers. comm.). EFFORT polishing was
applied as a subroutine of HyCorr and a full suite of ‘Reality Boosts’ were used in this
Chapter 6
- 224 -
preliminary correction attempt. HyCorr was found to overestimate the abundance of
water vapour, causing problems in the 0.940 - 1.130 µm and 1.380 - 1.870 µm regions
(Figure 6.8b). The 0.760 µm O2 and 2.005 µm and 2.055 µm CO2 absorption features
were also found to be problematic, resulting in negative spikes around these
wavelengths, as documented by Kneizys et al. (1988) and Staenz et al. (2002).
The boosted EFFORT algorithm was found to introduce green vegetation and
kaolinite features into the spectra of each pixel of the data (Figure 6.8c). The absorption
features were found to be more difficult to resolve and identification of the constituent
materials were hindered by the persistent presence of the 2.160 µm absorption, red edge
feature at 0.76 µm and a reflectance peak at 2.330 µm. Absorption features in the 2.250
- 2.400 µm region were consisted of two sharp absorptions at 2.304 µm and 2.401 µm,
due to the induced reflectance high at 2.330 µm. The positions of the Al-OH absorption
features in the 2.160 - 2.200 µm region were found to be broadened or shifted to
different wavelengths by the EFFORT and reality boost processes.
Figure 6.8 Comparisons of (a) Raw HyMap and (b) scaled reflectance results for HyCorr-only processing kaolinititic material from a dam wall (dotted spectra) and a pixel of the canopy of an eucalyptus camaldulensis (black spectra). For reference ASD spectra of green vegetation (red spectra) and kaolinised-goethitic saporolite (green) are shown. Notice large positive spikes in the 1.1 µm and 1.3 µm regions and noise in the NIR regions. A CO2 feature had not been corrected sufficiently at 2.05 µm. The lower spectrum is displayed with the continuum removed. (c) HyCorr + EFFORT scaled reflectance spectrum of kaolinititic material from a dam wall. The lower spectrum is displayed with the continuum removed. The wavelengths of both spectra were in micrometres (µm). Notice the flat shape between 0.9 µm and 2.1 µm, the vegetation feature at 0.7 µm and the exaggerated kaolinite doublet at 2.2 µm.
I C Lau Remote Sensing Results
- 225 -
Figure 6.8 (d) ACORN corrected scaled reflectance spectrum of kaolinititic
material from a dam wall. The lower spectrum is displayed with the continuum removed and wavelengths were in nanometres (nm). Several noticeable problems exist with the ACORN corrected dataset, which include negative near-UV/VIS reflectance, negative spikes and poor correction for atmospheric gases. Large positive spikes at 1.1 µm occur in some pixels. (e) HyCorr and normalisation offsets applied of a kaolinite material of a dam wall. Data is displayed as wavelengths (micrometres- µm) with the lower spectrum having its continuum removed to accentuate the absorption features. The spectrum is smooth in the 2.0-2.5 µm SWIR region but still had numerous artefacts in the NIR, making it unsuitable for Fe-oxide and vegetation discrimination. The omission of reality boosts from the HyCorr corrected processing (f) shows a close response for a pixel representing the typical soil from the region. The ASD reference spectra, shown in blue, has a high correlation to the HyMap soil pixel in the VNIR and 2.2 µm regions, however does not correspond above 2.3 µm. (g) Empirical line calibration of HyCorred data showing a scaled reflectance spectrum of kaolinititic material from a dam wall. The lower spectrum is displayed with the continuum removed. The 2.2 µm kaolinite absorption appeared left asymmetric rather than a doublet, as seen in the EFFORT corrected data. The VNIR features were smoother and had a closer resemblance to reference spectra and ASD measurements of field samples. The VNIR region displays a mixed Fe-oxide signature of hematite and goethite minerals. The final correction method using a mixture of HyCorr and gains calculated from ASD measurements of collected field samples (h) was found to produce the closest comparison to geological materials.
Chapter 6
- 226 -
HyCorr + EFFORT (No Reality Boosts)
A detailed analysis was carried out on the different settings of the HyCorr
software, involving the trailing of a diverse combination of the programs parameters.
The use of EFFORT without any Reality Boost parameters resulted in a featureless
spectrum devoid of recognisable absorptions in the SWIR region (Figure 6.8f). The
data in the SWIR region were found to be almost unusable for processing, as most of
the absorption features were not resolvable or identifiable. However, the VNIR
produced highly comparable results for Fe-rich soils and other ferruginous materials
(Figure 6.8f).
HyCorr Without EFFORT + Offset Calculations
HyCorr corrected data, without EFFORT polishing (Figure 6.8b), were found to
contain numerous high amplitude positive and negative spikes that were inferred to be
errors in the radiative transfer model. The larger features appeared to be additive and
were reduced by the use offset coefficients (Figure 6.8e), which were calculated using
the average spectrum of the swath. The average spectral response and the standard
deviation were normalised resulting in a value for each band. The generated values
were added back to the spectra of each pixel within the swath using ENVI. This method
produced a ‘cleaner’ spectrum that appeared to correspond better to reference spectra
(USGS Digital Spectral Library (Clark 1993) and field spectrometer measurements,
although a number of unwanted absorption features were still present.
The overall shape of the spectra were ‘flat’ throughout the 1.0 - 2.2 µm region,
with the exception of the 1.4 and 1.9 µm water absorption features. This spectral
response contrasted the ASD FieldSpec measurements of soils and rocks collected from
within the swath region. Spectral measurements of soils showed a large contrast
between the VNIR and the SWIR regions, with a greater reflectance in the 2.0 µm
region. The high reflectance was attributed to quartz, feldspars and clays, the
predominant constituent minerals of soil.
Empirical Line Calibration
From the observation of the difference between the ASD FieldSpec
measurements of field samples and the airborne data processed by the above methods, it
was concluded that an EL approach would produce the most comparable spectra to field
and reference data. A quarry, excavated for road materials, was used as a bright target
I C Lau Remote Sensing Results
- 227 -
and the Barrier Highway was used as the dark target. Samples collected from the
locations in November 2001, November 2002 and March 2003 were measured using an
ASD FieldSpec FR in April 2003. Small pieces of bitumen were chipped off of the
edge of the highway for the dark target (Figure 6.9). Light-coloured, rounded nodules
of regolith carbonate and weathered saprolite lag material were collected from a
homogeneous region in the quarry for the light target. The ASD measurements were
resampled to HyMap wavelengths and the EL calibration performed within ENVI
(version 3.5).
Figure 6.9 Sample collection site of the dark-target material for empirical line calibration of the HyMap imagery.
The resulting EL calibration for the HyMap swath contained mixed results. The
laboratory spectra collected with the ASD FieldSpec were found to be comparable to
the EL calibrated HyMap data and were especially coherent in the NIR and SWIR
regions (Figure 6.8f). The data were found to contain negative values in the shorter
wavelengths, which was attributed to poorly chosen regions of interest for the dark
pixels within the imagery or the asphalt sample not representing the bitumen road of the
HyMap imagery. On examination of the field area, the bitumen road was found to be a
less than an ideal dark target, as the worn asphalt in the centre of the road appeared a
shade of light grey. This contrasted the dark, freshly tarred edges, where the samples
were collected. Without a sufficient dark target for correction, a modified empirical
method was required to be implemented.
Chapter 6
- 228 -
HyCorr + Modified EL Calibration
The modified EL technique used spectra of pixels where samples were collected
and measured with the ASD FieldSpec to further improve the calibration of the HyCorr
corrected data, in a similar manner as the offset method. Average HyMap spectra for
the ROIs corresponding to the sample sites were collected and the ASD FieldSpec
measurements of the sample, collected from within the swath area, corresponding to the
ROI, were used to calculate gains. The ASD FieldSpec (>2000 bands) measurements
were re-sampled to the HyMap wavelengths (128 bands) prior to the normalisation
procedure. The normalisation process involved dividing the HyMap data by 10000 to
reduce it to % reflectance values, to match the units of the ASD FieldSpec
measurements, and dividing the results by the re-sampled ASD FieldSpec
measurements. The normalised data were then compared to the other regions to
determine which were the flattest and possessed values closest to zero. The normalised
spectra with the closest match to the reference library were chosen for the calculation of
the gains.
These gains were applied back to each pixel using the Band Math function in
ENVI. The results produced spectra with a clean, smooth appearance and a greater
variability in shape of the hull (Figure 6.8h), unlike the HyCorr corrected data (Figure
6.8b, c, and f). The spectra from the modified EL corrected data were found to be
comparable to ASD FieldSpec measurements of samples from within the swath. The
technique described above was applied to address the problems encountered when
comparing reference and field-derived spectra with the ATREM+EFFORT corrected
HyMap data.
The calibration gains were used to correct the four remaining swaths. Spectra
for the swath overlap regions were examined for consistency and are shown in Figure
6.10. Differences between the swaths were minimal, with minor variation in the
reflectance levels between the HyMap runs. Absorption features between swaths had
similar wavelengths for all the regions examined (Figure 6.10). A strong similarity was
found with ASD measurements of collected samples from the corresponding locations.
I C Lau Remote Sensing Results
- 229 -
Figure 6.10 Comparison of co–registered pixels from adjacent HyMap swaths corrected by the HyCorr + ASD Gains method. The dashed spectrum is an ASD FieldSpec measurement of samples collected from the corresponding pixel area.
Alternative Radiative Transfer Models
A second radiative transfer atmospheric correction program was briefly trialled
as a comparison to the HyCorr software. The results from the ACORN program found
the SWIR region to be highly comparable to the HyCorr + EFFORT result, although the
left shoulder of the kaolinite feature was shifted to shorter wavelengths, causing the Al-
OH-related feature to broaden (Figure 6.8d). The ACORN correction displayed a small
absorption feature at 2.0 µm, related to CO2 gas. Bright materials were found to have
negative values for the near UV region (<0.5 µm), and dark materials were negative
below 0.7 µm. Difficulties in correcting for atmospheric effects were shown by an
unexpected deep absorption feature at 0.6 µm, a shoulder feature at 0.7 µm and high
frequency, low amplitude absorptions in the NIR. It is the author’s opinion that these
errors can be removed with careful use of input parameters in the radiative transfer
software and the application of correctional offsets during post-atmospheric processing.
Chapter 6
- 230 -
Atmospheric Correction Conclusions
Analysis of hyperspectral imagery for phyllosilicate minerals and alteration
products requires a technique with the greatest accuracy in the >2.0 µm region,
therefore an EL correction was considered the most ideal calibration technique. If the
location of the imagery was in an inaccessible area, the model-based technique should
be able to produce sufficient results to generate identifiable spectra. However, the
calculation of offsets to correct for the inefficiencies of the radiative transfer model
would be required to visually improve the spectra. Consideration of the effects that the
input parameters will have on the outcome, and care should be taken before applying an
atmospheric correction program.
A modified RT-EL technique was found to produce the best results for the
White Dam HyMap data (Figure 6.8h), using a combination of field derived and within
scene data to correct the HyCorr processing. The results contained the smoothest
spectra with the least amount of noise and the most identifiable absorption features,
especially in the VNIR region.
Multi-Swath Correction
The collection of targets on each of the ground areas for a multi swath dataset
may be impractical due to access restrictions or unachievable due to the lack of
homogeneous sites (Stow et al. 1996). An alternative technique is the matching of the
mean and standard deviation of the overlap regions of adjacent swaths with an apparent
reflectance corrected swath (Perry et al. 2000; Taylor 2001). This approach can yield a
higher continuity between swaths and hence a less disjointed endmember abundance
mosaic map. This process was used to correct the remaining four swaths of the HyMap
imagery that were not corrected by the modified RT-EL technique.
Comparison of HyMap data with ASD data.
As part of the validation of the airborne hyperspectral data with regard to its
ability to correctly map minerals, ASD spectra of materials collected from within the
swath area were resampled and compared to the corresponding HyMap pixels. Efforts
were made to collect the best representation of the soil or rocks that occurred at the
locality, as the field samples were representing an area approximately 0.1 m2, whereas
the HyMap pixels were 5 m2, and were susceptible to mixing with undesirable
materials, such as vegetation, man-made materials and structures.
I C Lau Remote Sensing Results
- 231 -
Analysis of homogeneous areas of soil, where vegetation cover was minimal,
displayed highly similar spectra for the HyMap and ASD FieldSpec. Examination of
the variation in the ASD soil spectra showed changes in the depth of the 0.9, 1.4, 1.9
and 2.2 µm absorption features and shifts in the shoulders related to the Fe-oxide
features. The 1.4 and 1.9 µm features are highly affected by noise in the HyMap data
due to atmospheric water and are not comparable to the ASD measurements.
Exposures of basement were found to produce less comparable HyMap spectra,
as many of the outcrops were colonised by small chenopods, lichen and low trees. Field
validation found that many of the exposures were under 5 m2 in size, making mixing
with soils and vegetation a significant problem. Some basement exposures consisted of
dark rocks, which were expected to have poor spectra responses in the SWIR region.
Goethite was identified in numerous samples of saprolite (Figure 6.11a & Figure
6.12c), however, was not as obvious in the HyMap pixels corresponding to outcrop.
Samples measured with the ASD containing hematitic spectral features displayed a high
correlation the HyMap pixel of where the material was collected (Figure 6.11c, d, f &
Figure 6.12d, f). The 2.16 µm kaolinite spectral feature was subtle in the HyMap
imagery, making interpretation of kaolinite crystallinity difficult (Figure 6.11d). The
lack of definition of this feature may be due to the atmospheric correction procedure or
the mixing of pixels dominated by soil materials. The presence of kaolinite (or
halloysite) was inferred in the HyMap imagery, where the 2.2 µm absorption displayed
a feature in the 2.16 µm region, causing an asymmetric shape to the absorption (Figure
6.11d & Figure 6.12f).
The presence of regolith carbonate was noted at numerous locations and found
in samples collected from sites within the HyMap swaths. The asymmetric calcite
absorption features were found in samples in the 2.3 µm region in the ASD samples and
was observed in some of the corresponding HyMap pixels (Figure 6.11e). The depth of
the 2.3 µm absorption feature was found to vary with intensity and may not have been
related to the presence of the mineral calcite and may have either been due to cellulose
in dry vegetation or secondary features related to the minerals kaolinite, illite or
muscovite.
In most cases the overall spectral shape of the HyMap pixel displayed a close
resemblance to the corresponding ASD sample with minor differences in reflectance.
The variation was often seen to affect the albedo of the VNIR or SWIR regions
separately, which was attributed to the differences between the illumination conditions
of the laboratory and the field (Figure 6.11d & Figure 6.12b, e). Similar spectral
Chapter 6
- 232 -
variations were found during experiments on the materials analysed with ASD
instrument if the sample was stirred or the orientation of the light source was altered.
The mixing of pixels with aspectral materials and small abundances of other materials
which were not collected during field sampling may be related to the differences in the
overall shape of the HyMap spectra.
Figure 6.11 HyMap spectral responses and corresponding ASD measurements of samples collected from around the White Dam swaths, used to validate the atmospheric correction. The ASD spectra have been resampled to HyMap wavelengths (from 2151 to 128 bands).
The ASD spectra of soil samples were found to resemble the corresponding
HyMap pixels better than the rock samples. This may be due to the more homogeneous
nature of soil samples than the variable rocks in the eastern OlD.
I C Lau Remote Sensing Results
- 233 -
Figure 6.12 HyMap spectral responses and corresponding ASD measurements of samples collected from around the White Dam swaths, used to validate the atmospheric correction. The ASD spectra have been resampled to HyMap wavelengths (from 2151 to 128 bands).
Generally soil and muscovite-rich rock ASD spectra displayed a close
resemblance to the corresponding HyMap pixels from where the sample was collected.
This was an encouraging result and proved that the atmospheric correction was
successful. The 5 m2 pixel size of the HyMap induced some mixing of rock-soil and
vegetation. The spectrum for kaolinite-rich materials was less obvious in the HyMap
imagery making it difficult to detect using visual analysis of the reflectance data. This
could be solved by the use of empirical techniques, such as ratios and indices, to extract
areas of higher kaolinite abundance, as discussed later in this Chapter).
Chapter 6
- 234 -
Pre processing of the HyMap Imagery Part 2-Geocorrection and
Image Cleanup
Obvious contrast differences can often be seen in corrected airborne scanner
images. These contrasts that can occur along the image or across the swath were due to
variations in the sun angle and sensor geometry which cause illumination differences
while acquiring the data. Careful flight planning enables the minimisation of these
effects on airborne imaging data (Taylor 2001; Mauger A.J. 2001, pers. comm.). Lack
of correction can drastically affect endmember abundance maps by creating false
gradients or the reduction of continuity across multiple swaths.
Orientation of the flight paths to acquire the data parallel to the direction of solar
illumination is one method for reducing the effects that occur across-track, but may
introduce “hot-spots” along the swath (Taylor 2001). The image processing software
ENVI has a routine that attempts to correct for both types of illumination variations.
The average brightness of each band is calculated for each line of the image and a
polynomial is applied to the across-track average brightness. An offset is calculated for
each pixel and applied to the data to remove the illumination variation.
Orientation of the White Dam HyMap dataset was chosen to collect information
over the most prospective areas of the MacDonald Corridor and White Dam Prospect
from the smallest number of runs. This has caused a large amount of across-track
brightness variation, which was required to be corrected for the production of seamless
mineral abundance maps.
Artefact Removal and Data Preparation
Due to the nature of how the HyMap sensor records information, the data
required transposing (flipped horizontally and rotated 90°). Transposing can be
performed at any point in the processing but is often early in the pre-processing within
ENVI to aid feature recognition. Runs are often collected in alternating directions and
require additional rotating for correct orientation.
Once the HyMap imagery were corrected sufficiently to allow comparison with
field samples and reference library measurements, ‘cross-track correction’ was
performed to remove brightness variations. Variations across the swath, which
stemmed from the angle of the sun at the time of acquisition of the data, were corrected
using an RSI ENVI version 3.5 module. The process involved the analysis of the
across-swath rows of data for a systematic variation in the digital numbers and the
I C Lau Remote Sensing Results
- 235 -
calculation of a polynomial fit to remove the variation. Comparisons of the uncorrected
and corrected results are shown in Figure 6.13.
Once the data were cross-track corrected, they were analysed band-by-band to
find channels with large amounts of noise or spurious features. The bad bands were
omitted as the noisy pixels often influenced the processing steps that rely on the
statistical properties of the data. Band 1 and 128 of the HyMap data had poor signal to
noise ratios due to low amounts of EM energy in these regions, and therefore, these
bands were not used in the latter processing steps. The lower wavelength regions of the
HyMap imagery were typically more affected by Rayleigh scattering, which is the
source of noise in the blue and near UV regions of remotely sensed data.
Figure 6.13 WD001 swath showing (upper) corrected and (lower) uncorrected cross-
track data.
Bands near the atmospheric windows were often spatially incoherent, displaying
large amounts of ‘salt and pepper’ noise, and are not used in processing. A remnant
feature of the model-based atmospheric correction was the creation of positive spikes in
the 1.4 µm and 1.9 µm wavelength regions due to poor correction for water. Bands in
these regions were also omitted from further processing.
The final pre-processing step involved the masking of dark pixels and clouds. A
threshold is defined for dark pixels and ROIs were defined around pixels affected by
cloud and cloud shadow. Once the above processing steps had been completed the data
were ready for processing.
Geometric Correction (Post-Processing)
Geocorrection is typically performed in the final stages of processing, prior to
map production and data presentation, as shown in Figure 6.15. Airborne whiskbroom
scanners, such as the HyMap instrument are inherently susceptible to geometric
distortions, as each pixel is collected at a separate unit of time. Distortions between
neighbouring pixels within the imagery arise due to differences in the platform velocity
and orientation during the acquisition. These properties make geometric correction of
whiskbroom scanner data particularly challenging. Polynomial (“rubber-sheeting”)
geometric correction techniques generally do not perform well (Clark et al. 1998), as
Chapter 6
- 236 -
they assume linear or curvilinear variations of pixel locations (Aspinall et al. 2002).
Aspinall et al. (2002) recommended a triangulation technique, using ground control
points (GCP), as the method for correcting whiskbroom airborne datasets. High spatial
resolution hyperspectral data were particularly sensitive to local topographic variations
at the scale of metres (Conese et al. 1993), making ortho-rectification using DEM
information highly favourable. Topography can influence the spatial and spectral
characteristics of a dataset, especially in extremely rugged terrain (Feng et al. 2003).
Since the 1998 acquisitions, HyVista Pty Ltd have installed the HyMap sensor on a
gyroscopic platform with accompanying sensors for roll and pitch angles, as well a
DGPS. This allows a more precise geometric correction to be performed.
Georectification of the White Dam HyMap dataset was performed using ground
control points (GCP) selected from 1.25 m ortho-rectified digital air photographs.
Between 50 and 200 GCPs were used for each swath. A triangulation warp-method was
found to produce the least amount of distortion in the central regions of the swath. The
edges of the swath displayed artefacts where a low abundance of GCPs were used.
Polynomial and linear methods were found to have a poorer fit, with a greater distance
between corresponding features on the ortho-image, although they did not show such
edge-effects. The large overlap of the swaths allowed the triangulation edge errors to be
disregarded, as these regions were cropped in the final presentation stage. All
classification products were corrected to produce 5 m pixels.
Information Extraction and Spectral Un-mixing
Two types of processing were performed on the HyMap data. The first involved
the processing steps of noise removal, data reduction and feature extraction, often
referred to as ‘hourglass’ processing, due the removal of redundant data Figure 6.14.
The second method involved the ratioing of bands and masking of features using
spectral indices. A flowchart of the processing steps performed in the extraction and
presentation of information from the HyMap imagery is shown in Figure 6.15.
Figure 6.14 Diagrammatic representation of the ‘hourglass’ processing technique of mineral map production, using un-mixing techniques. The term comes from the decrease in data, and corresponding file size, through the processing steps.
I C Lau Remote Sensing Results
- 237 -
Figure 6.15 Information extraction flowchart, detailing the un-mixing techniques used on the HyMap Imagery.
The ‘hourglass’ processing routine involves the reduction of a large dataset, with
numerous bands and pixels, into a cloud of data that contains a condensed
representation of the features of the scene. These steps are extensively described in
Appendix VIII. The first step involves the sorting of the data by noise using a dual
principal component function called minimum noise transform. The bands with a
higher degree of noise produced images with almost no coherent features and were not
used in the subsequent processing steps.
The data were further reduced by the extraction of ‘pure pixels’. These were
projected in an n-dimensional space with reference back to the original corrected
dataset. The pixels on the fringes of the cloud represent outlier spectra that represent a
pure material or mineral. These outlier pixels are termed endmembers and can be used
together in different combinations to model the mixtures of materials within a scene.
Chapter 6
- 238 -
These endmembers were the initial primary product and a representation of the
materials that existed within the study area and that were able to be extracted from the
data. Identification of endmembers can be performed using a combination of spatial
distribution, spectral feature recognition, and ground control field samples.
Recognition of features becomes intuitive after extended examination of high-
resolution imagery. Ground truthing of features and the collection of referenced
samples is helpful for identification of spectral signatures. Field samples can be
analysed by spectrometers and compared to the airborne imagery. Endmembers can
consist of materials or minerals, for example, a clayey soil of montmorillonite and
kaolinite or green vegetation signature of a eucalyptus tree. Mixing can occur between
endmembers in non-linear abundances, which can cause great difficulty when
attempting to interpret and identify the constituents of a pixel.
The use of SWIR regions alone for un-mixing is still undecided, as problems
arise when endmembers possess similar SWIR features but completely different VNIR
spectral features, relating to differing vegetation concentrations or Fe-oxide abundances.
This causes problems when performing endmember extraction steps (for example, n-
dimensional visualisation) because endmembers with the same SWIR spectra but
different VNIR exist.
To overcome this problem it is recommended by the author that only the SWIR
spectra (to 1.2 µm +) is used, which will still show vegetation and Fe-oxide presence in
the materials but would not be affected by CFA, CTS and other effects. The exact cut-
off region would need some testing to find the best compromise for different scenes.
The total removal of the 1.2-2µm region may be favourable for regolith dominated
terrains, as minerals in this region were usually obscured by the 1.4 and 1.9 water
features. The possibility of missing minerals such as alunite, gypsum and epidote is
possible, although these minerals were quite rare in weathered materials and regolith
dominated terrains.
A commonly used method for the mapping of alteration minerals involves the
selective use of the SWIR region of the spectra. This technique requires additional
processing to extract information on Fe-oxides and vegetation, which have their
diagnostic features in the VNIR and only have subtle features in the SWIR region.
Use of the whole spectral range of the White Dam HyMap dataset often found
the spectra to consist of combinations of different features in the SWIR and VNIR. For
example a SWIR kaolinite feature may be seen in three different endmembers, one with
goethite, one with hematite and one with vegetation VNIR features.
I C Lau Remote Sensing Results
- 239 -
Endmember Extraction from the HyMap Imagery
The HyMap datasets were individually examined and processed to determine the
endmembers for each swath. The combined HyMap dataset was found to have
corresponding endmembers that could be identified in the other swaths. There were
also a small number of unique individual endmembers that were not found to be
prevalent within the whole dataset. The endmembers were split in the VNIR and SWIR
subsets for the process of performing un-mixing on selected ranges of wavelengths.
The endmembers of the five swaths were examined for unique and representative
spectra, which were used to build the combined SWIR and VNIR libraries. The spectra
of the combined endmembers are shown in spectral plots of Figure 6.16 to Figure 6.19.
A total of eighty-five endmembers were extracted from the five swaths during
the individual processing. Many of the endmembers were replicated in the other runs,
showing a homogeneity of materials across the runs and indicating that the pre-
processing had been successful. Duplicates and spectra with similar shapes but lower
albedo were removed from the subsequent processing steps after re-evaluation of the
extracted endmembers, reducing the number to sixteen. Some of the spectra that
displayed goethite mineralogy in the SWIR with deep CFA and CTS features and Al-
OH features in the SWIR. Others displayed the same SWIR features but different
VNIR features. There was a group of spectra that displayed vegetation influences in
the VNIR and some in the 1.7µm region of the SWIR. Nine spectra were selected to
represent the VNIR region (Table 6.4). Absorption features at 2.35µm, which may
have been related to RCA or dry vegetation, were identified in a number of spectra. A
few of the endmembers displayed mixed or un-identifiable features and were not
named.
One spectrum displayed an almost exact match to illite (USGS Spectral Library-
Clark 1993) as well as a good match for muscovite (USGS Spectral Library- Clark
1993). The identified spectra consisted of: kaolinite, poorly ordered kaolinite, kaolinite
smectite, muscovite, illite, goethite, hematite, dry vegetation, green vegetation, road tar
(asphalt), regolith carbonate and soil spectra resembling ASD measurements.
Chapter 6
- 240 -
Figure 6.16 Combined SWIR endmember spectra, extracted from the HyMap imagery.
I C Lau Remote Sensing Results
- 241 -
Figure 6.17 Combined SWIR endmember spectra, extracted from the HyMap imagery.
Chapter 6
- 242 -
Figure 6.18 Combined SWIR and VNIR endmember spectra, extracted from the HyMap imagery.
I C Lau Remote Sensing Results
- 243 -
Figure 6.19 Combined VNIR endmember spectra, extracted from the HyMap imagery.
The combined endmembers were selected to represent the main components of
the five swaths with the intension of classifying the most prominent materials within the
Chapter 6
- 244 -
area. The endmembers were identified as consisting of different forms of kaolinite,
muscovite and regolith carbonate materials, as well as soils, Fe-oxides, vegetation and
man-made materials. Table 6.3 and Table 6.4 document the swath from which the
endmembers were extracted and briefly describe the constituents. The two sets of
combined endmembers were used to perform un-mixing on the individual swaths. The
generated endmembers were used in their origin swath for Spectral Angle Mapper
(SAM) and Mixture Tuned Match Filtering (MTMF) un-mixing processes to determine
their distribution.
Swath Endmember Description 1 1 Highly Crystalline Kaolinite 1 6 Dry-Vegetation 1 11 RCA-Goethite. Quarry material 1 17 Fe-/Al-OH Goethite/Hematite 1 21 ?Chlorite/Calcite- Al-OH Hematite 2 2 RCA 2 8 Shadow/Road/Water/Cloud (Dark materials) 2 9 Green Vegetation 3 21 Muscovite-Phengite 3 23 Very Low Crystalline Kaolinite (Regolith materials) 3 26 Muscovite (Fe-rich) 4 1 Muscovite-rich Soil 4 5 Kaolins 5 1 Low Crystalline Kaolinite 5 5 Soil 5 6 RCA ?Mafics/Amphibolite
Table 6.3 Combined SWIR endmembers from the White Dam HyMap dataset.
Swath Endmember Description 1 6 Dry Vegetation and Soil 1 12 RCA- Muscovite and Goethite 1 16 Tin Roof 2 8 Road/Shadow/Water (Dark Materials) 2 9 Green Vegetation 3 20 Hematite CFA 4 8 Average Soil 4 16 Soil and Vegetation 4 19 Road
Table 6.4 Combined VNIR endmembers from the White Dam HyMap dataset.
The SAM processing was performed using a 0.1 radian classification cut-off.
The SAM classification was found to be dominated by the WD003#20 (Hematite CFA)
and WD004#1 (muscovite-rich soil) endmembers, displaying little variation in the
I C Lau Remote Sensing Results
- 245 -
VNIR and SWIR classifications. The results of SAM, along with the other types of un-
mixing techniques that were less successful than MTMF, are not presented.
Mixture Tuned Match Filtering using Combined Endmembers
Processing Parameters and Cut-offs
Regions of interest (ROI), with thresholds of >0.1 for the matched filter (MF)
score and <100 for infeasibility, were created from the results of the MTMF. The
intersection regions of the MF and infeasibility ROIs were used to generate a class
image, which was georectified and converted to a vector overlay. The overlays were
exported to ArcView and mosaiced in the GIS. A total of 16 SWIR and 9 VNIR MTMF
endmember distribution maps were created for the whole swath.
The threshold values for the creation of ROIs were determined to be an
acceptable and realistic representative of the minerals occurring in the highlighted
pixels. Percentage cut offs were trialled using 95% and 99% abundances for the MF
score and 30% - 40%, or greater for the infeasibility results. The resulting quantity of
classified pixels for each endmember showed that the use of percentages did not
produce sufficient numbers of a material that was abundant in the swath, whereas
materials that were not common, such as the corrugated galvanised roof of the CDMA
tower building at MacDonald Hill, were being over classified. The cut-off of 0.1 MF
score and 100 and 150 infeasibility score for the SWIR and VNIR respectively, were
determined to classify the spectral shape of the endmember without appearing like
another material. These values were found by calculating the average spectra of the
intersection ROIs and comparing it to the original endmember spectra.
VNIR Results from Un-mixing using Mixture Tuned Match Filtering
The results of the VNIR MTMF analysis found that much of the regolith-
dominated area to the north of the MacDonald and Kalabity Shearzone Ranges were
mapped as WD004#8 (Average Soil), as shown in Figure 6.21. This was attributed to
the presence of PSA mantling most regolith-landforms, especially where the slope
gradient was lower. The areas of exposed basement and areas where active erosion was
occurring, such as erosional sheetflow-dominated plains and channels (Figure 6.20), had
a weak association with WD002#6, demonstrating the relationship of PSA and
depositional landforms.
WD003#20 (Hematite CFA) was the second most abundantly classified
endmember and was related to the badlands and areas of the TCC that appeared orange-
Chapter 6
- 246 -
pink to red. These areas were identified as having higher abundance of Fe-oxides, in
particular hematite. Channels that were not colonised by vegetation, such as the
northward trending features that drained from the MacDonald Ranges, also were
classified as WD003#20 (Figure 6.21).
The river red gums that occurred along Bulloo Creek, near the intersection with
the Bulloo Creek Road, were mostly not classified by any of the MTMF results, with
some pixels identified as WD002#9 (green vegetation), as seen in Figure 6.23. Most
of the channels and creeks in the WD001 swath were classified as WD003#20 (hematite
CFA), whereas the swampy regions that contained vegetation were classified as
WD001#6 (dry vegetation-soil). In general, the VNIR vegetation-orientated
endmembers were found to classify a small abundance of pixels throughout the swath,
even though field observations found there to be large amount of vegetation in some
locations. This was attributed to the vegetation removal and masking procedures
performed in the pre-processing, which removed the obvious vegetation influenced
pixels.
WD001#12 (RCA/Goethite) was found around the old dam in the northern
regions of the White dam flood plain (Figure 6.20). This endmember occurred in a
restricted zone at the contact of the Adelaidean and Willyama Supergroup basement
rocks along the MacDonald Ranges (Figure 6.21). There was also a high correlation
with RCA/Goethite with the SWIR kaolinite endmembers, which occurred in the highly
weathered, regolith-dominated regions of the swaths (Figure 6.20). North of the Green
and Gold Prospect exposures of basement related to the Kalabity Shearzone Ranges
displayed a linear east-west occurrence of RCA/Goethite (Figure 6.21). This was
related to the weathered saprolite outcrop in this region.
The endmember WD001#16 (Tin roof), corresponding to the galvanised roof of
the building at the CDMA tower was found to highlight scattered dark pixels that
occurred along the vegetated margins of the Barrier highway (Figure 6.21) and other
pixels that appeared dark in the TCC. Ground truthing found a number of wrecked
vehicles and scrap metal discarded in the regions of the highlighted pixels. The exposed
ironstone at the Wilkins Prospect and the CDMA tower were also successfully classified
in the MTMF process by this endmember (Figure 6.21).
Drill spoils at the White Dam Prospect were classified by WD004#19 (Road)
(Figure 6.20). This endmember also highlighted swampy depressions and Adelaidean
rocks in the MacDonald Ranges (Figure 6.21), which displayed low albedo in the VNIR
region. The Barrier Highway was largely not classified by this endmember, with only a
I C Lau Remote Sensing Results
- 247 -
few pixels highlighted (Figure 6.21). This could be due to the variable nature of the
materials used in the construction.
SWIR Results from Un-mixing using Mixture Tuned Match Filtering
The SWIR endmembers attempted to classify regolith minerals, materials
associated with basement and vegetation. The regolith materials identified included
highly weathered saprolite subcrop in sheetflow-dominated areas, PSA mantle, RCAs,
and soils-rich associated with bedrock exposures.
Four different kaolinite endmembers were extracted, and were categorised on
their crystallinity. WD004#5 (Kaolin) displayed the least defined 2.16 µm feature and
was related to very poorly crystalline kaolinite mixed with other clay minerals. The
endmember displayed a close association with WD003#23 (Very Low Crystalline
Kaolinite). These endmembers were found to classify areas identified as highly
weathered saprolite exposures in sheetflow-dominated areas, in the northeastern part of
the area (Figure 6.26). These areas appeared a dark grey or a similar colour to the
surrounding materials in the TCC.
WD0032#23 was found to correspond to some of the drill spoils at the White
Dam Prospect (Figure 6.26) and the materials flanking exposures of saprolite in the
MacDonald and Kalabity Shearzone Ranges (Figure 6.27). This distribution was
attributed to the weathering of colluvial material and the subsequent downslope
dispersion. These endmembers were found to be useful for identifying highly
weathered material and areas of shallow soils overlying Willyama Supergroup
basement.
The kaolinite endmembers WD005#1 (Low Crystalline Kaolinite) and
WD001#1 (High Crystalline Kaolinite) classified features relating to highly weathered
saprolite that were largely in situ. This included dam materials (for example Lucky
Dam, Cartwrights Dam, Little White Dam and White Dam were highlighted by these
endmembers), Willyama Supergroup-derived saprolite exposures in the MacDonald and
Kalabity Shearzone Ranges (Figure 6.27 & Figure 6.25) and small areas of highly
weathered in situ materials in the regolith-dominated areas (Figure 6.26 & Figure 6.24).
Three endmembers were identified as possessing carbonate-like features in the
2.3 µm wavelength region and were examined for their abilities to map RCAs. Cross-
referencing of the spectra used to define the endmember and the distribution of their
corresponding pixels was found to correlate to areas where RCAs were abundant. From
Chapter 6
- 248 -
these observations it was hoped that less prominent occurrences of regolith carbonate
would be able to mapped by the HyMap data.
Figure 6.20 Northern White Dam HyMap MTMF results of the VNIR endmembers showing the dominant features consisting of materials identified as hematite and soil. Areas of dark outcrop were mapped under the road and shadow endmembers (green and brown pixels). Highly weathered saprolite corresponded to goethite.
I C Lau Remote Sensing Results
- 249 -
Figure 6.21 Southern White Dam HyMap MTMF results of the VNIR endmembers showing the dominant features, consisting of hematitic and soil materials. Areas of dark outcrop were mapped under the road and shadow endmembers (green and brown pixels). Highly weathered saprolite corresponded to goethite.
Chapter 6
- 250 -
Figure 6.22 Northern White Dam HyMap MTMF results of the vegetation endmembers of the VNIR and SWIR. The Highlighted with blue arrows are the direction of watercourses. The red classification represents areas identified to possess absorption features in the 2.25-2.40 µm region, possibly related to chlorite or carbonate minerals.
I C Lau Remote Sensing Results
- 251 -
Figure 6.23 Southern White Dam HyMap MTMF results of the vegetation endmembers of the VNIR and SWIR. The red classification represents areas identified to possess absorption features in the 2.25-2.40 µm region, possibly related to chlorite or carbonate minerals. The pixellated nature of the results reflects the mixed nature of the HyMap data.
Chapter 6
- 252 -
Figure 6.24 Northern White Dam HyMap MTMF results of the SWIR endmembers showing the dominant regolith materials. Highly weathered saprolite corresponded to high crystalline kaolinite (blue pixels). Regolith carbonate associated with weathered saprolite was classified by the cyan areas, while a second endmember displays a more restricted distribution for RCAs (red pixels). Soil areas reflect alluvial areas (brown), whereas muscovite-rich soils (apple green) classified sheetflow dominated areas with shallow soils. Dark, aspectral outcrops were mapped as shadow (in dark green).
I C Lau Remote Sensing Results
- 253 -
Figure 6.25 Southern White Dam HyMap MTMF results of the SWIR endmembers showing the materials in the MacDonald Ranges are less-weathered than the northern regolith-dominated area. RCAs (cyan) were associated with the Willyama Supergroup rocks, whereas the Adelaidean rocks were mapped as shadow (dark green). The shear zone in the Kalabity Shearzone Ranges region was mapped as possessing Fe-Al-OH absorption features. Badlands were mapped as soil (brown), whereas the soils mantling Willyama Supergroup rocks were mapped as muscovite-rich (apple green).
Chapter 6
- 254 -
WD001#11 (RCA/Goethite) displayed a distinctive distribution that correlated
with graded roads, quarries and areas of shallow soils overlying Willyama Supergroup
basement. The roads to the Bulloo Creek homestead and the CDMA tower were easily
identified and displayed similar spectral features as the near-by quarries (Figure 6.25).
WD001#11 classified material related to the Willyama Supergroup along the length of
the MacDonald Ranges, as well as the flanking sheetflow-dominated areas to the north.
These areas displayed shallow soils and often were characterised by large abundances of
fragmented hardpan RCA near the surface (Figure 6.24).
In comparison, the other two endmembers with absorption features in the 2.3 µm
displayed a limited distribution of sparely scattered pixels. WD002#2 (RCA) was
associated the material on the edges of the Barrier Highway and rabbit warrens, which
occurred as small, rounded groups of bright pixels in the TCC. The WD005#6
(RCA/Amphibolite) endmember displayed absorption features that were attributed to
Mg-OH minerals and was used to identify possible occurrences of mafic outcrops. The
distribution was widely scattered, although a number of dark coloured, rounded features
were highlighted and interpreted as mafic outcrops (Figure 6.26).
The endmembers WD004#1 (Muscovite-rich soil) (Figure 6.24 & Figure 6.25)
and WD005#5 (Soil) (Figure 6.26 & Figure 6.27) were chosen to classify areas that
were mantled by PSA and other soil materials. WD004#1 displayed a wide distribution
and was abundant throughout the scene. The class identified sheetflow areas flanking
basement exposures and shallow soil areas of both the Willyama Supergroup and
Adelaidean rocks (Figure 6.25)). These typically occurred adjacent to areas classified
as WD003#21 (Muscovite-Phengite). The distribution of pixels classified by this
endmember was similar to that of the dark red coloured areas of the TCC that
represented bare soils. WD005#5 was highly correlated to the badlands and Adelaidean
areas to the south of the MacDonald Ranges (Figure 6.27), as well as areas that were
mapped as sheetflow dominated depositional plains (Figure 6.26) (CHpd) on the 1:30
000 regolith-landform map.
The endmembers used to map bedrock were WD003#21 (Muscovite-Phengite),
WD003#26 (Muscovite- Fe-rich), WD001#17 (Fe-/Al-OH Goethite/Hematite) and
WD001#21 (Chlorite /Calcite- Al-OH Hematite). WD001#21 displayed the largest
distribution of these endmembers and highlighted the Willyama Supergroup rocks in the
MacDonald Hill region, emphasising the contact with the Adelaidean metasediments
(Figure 6.23). The crest of ridges were characterised by this endmember throughout the
I C Lau Remote Sensing Results
- 255 -
MacDonald Ranges, with only scattered occurrences in the northern regions of the
swath (Figure 6.22). This was attributed to the more calc-silicate Ethiudna Subgroup
lithologies to the north, whereas the southern areas are dominantly granites and rocks of
the Wiperaminga Subgroup.
The muscovite mineral endmembers displayed contrasting distributions to each
other but were similar to the abundant WD001#21. WD003#26 (Muscovite Fe-rich)
predominantly occurred in areas mapped as bedrock in the Kalabity Shearzone Ranges
region. These areas were related to shear zones that hosted abundant pegmatitic
material, rich in muscovite. The outcrop at the Wilkins Prospect, which occurred at the
eastern margin of the Kalabity Shearzone Ranges, was not classified by this
endmember, instead highlighted by WD003#21 (Muscovite-Phengite).
The Muscovite-Phengite endmember displayed a wider distribution than
WD003#26, highlighting the channels draining to the north from the MacDonald
Ranges (Figure 6.27), as well as outcrops to the north of the White Dam Prospect, in the
far northern part of the swaths (Figure 6.26). This endmember was also associated with
the quarries along the road to the Bulloo Creek homestead (Figure 6.27).
WD001#17 (Fe-/Al-OH Goethite/Hematite) mapped features corresponding to
dark coloured areas identified in the TCC as outcrop. These occurred locally in the
MacDonald Ranges and to the north of the Green and Gold Prospect, in the Kalabity
Shearzone Ranges (Figure 6.25). Trees growing in channels and on alluvial plains were
also highlighted by WD001#17.
Vegetation endmembers used in the SWIR MTMF were more successful at
identifying areas than the VNIR MTMF classification. The Dry Vegetation
endmember (WD001#6) highlighted a large area that appeared a grey tone with a
smooth texture. This was identified as an area colonised by a closed chenopod
scrubland. Alluvial channels and areas mapped as overbank deposits adjacent to creeks
were also highlighted. Green Vegetation (WD002#9) was found to have a similar
distribution as Dry Vegetation, highlighting alluvial-related landforms, such as arid-
swamps, channels an overbank deposits (Figure 6.20 & Figure 6.21).
The endmember WD002#8 was used in the MTMF classification process to
highlight features representing low albedo spectra. Shadows, trees and channels were
classified by this endmember, as well as dark coloured outcrops, which did not possess
strong spectral features. This included large areas of Adelaidean-derived saprolite and
erosional depressions draining from these exposed rocks (Figure 6.25).
Chapter 6
- 256 -
Figure 6.26 Northern White Dam HyMap MTMF results of the SWIR endmembers showing the dominant regolith materials. Highly weathered saprolite corresponded to the endmember identified as low crystalline kaolinite (orange pixels). The Kaolin endmember (purple) mapped channels draining from weathered bedrock areas, mapped as Muscovite-Phengite (brown pixels). Weathered materials around the muscovite-phengite outcrops were mapped as very low crystalline kaolinite (cyan). Soils around these regions were also identified (muscovite Fe-rich) as red areas.
I C Lau Remote Sensing Results
- 257 -
Figure 6.27 Southern White Dam HyMap MTMF results of the SWIR endmembers showing the bedrock mineralogy of the MacDonald Ranges, showing the less weathered saprolite (Muscovite-Phengite, brown areas) and low crystalline kaolinite (cyan pixels) of the more highly weathered areas. Channels containing lithic materials were mapped as muscovite-phengite (linear features trending north from the MacDonald Ranges).
Chapter 6
- 258 -
Information Extraction of the HyMap Imagery using Spectral
Indices
Spectral indices involve the use of ratios of selective wavelength features on
high-resolution spectra. An example of an index is the NDVI (Normalised Difference
Vegetation Index), which uses information in the VNIR region to interpret green
vegetation. Similar indices exist for Fe-oxides, clay minerals and other combinations.
The use of spectral indices should be employed with masks that remove the effects of
unwanted components, such as shadow. A NDVI can be constructed to determine
pixels that contain vegetation, which can be used to create a vegetation mask. The mask
can be inverted to display the pixels that were not heavily influenced by vegetation,
which may be of geological interest. Similar masks can be generated to aid the
discrimination of white mica minerals from kaolin minerals. A flowchart of the spectral
indices performed on the HyMap imagery is shown in Figure 6.28, with a detailed
description of the parameters of the indices found in Table 6.5.
Figure 6.28 Spectral Indices flowchart showing the sequence of processing required to extract mineral information from the HyMap data.
I C Lau Remote Sensing Results
- 259 -
Prior to the use of spectral indices masks were generated for each swath of
pixels for each swath with very high and low albedo, which corresponded to cloud,
vegetation, shadow and cloud-shadow. Cloud-shadow was easily identified within the
imagery by visual inspection, with the affected pixels displaying a reduced albedo,
while retaining the overall spectral shape of the materials. It was this characteristic that
made it critical that these pixels were masked out, as these spectra would have adverse
effects on the results of MTMF and the index processes if left in the processing.
Name Ratio Masks Threshold Leaf/Surface Index Bands 48+59/51+53= (1.119 µm+1.279 µm)/
(1.164 µm+1.193 µm) 1.03 or 1.05
Cellulose Index Bands 100+110/104+105 =(2.205 µm+2.184 µm)/ (2.078 µm+2.096 µm)
0.95 or 0.97
Fe-oxide Abundance Bands 23+46/31+35 = (0.761 µm+1.190 µm)/ (0.886 µm+0.919 µm)
A1&A2 1.05 and 1.055
Fe-oxide Abundance Bands 23/31 A1&A2 0.98
Hematite:Goethite Bands 33/35 = 0.886 µm/0.919µm B1 orB2
Goethite:Hematite Bands 35/29 = 0.919 µm/0.868 µm B1 or B2
Hydrated Fe-oxide 1.265 µm*1.289 µm/1.334 µm*1.347 µm Bands 58+59/63+64 = 1.265 µm +1.279 µm /1.334 µm +1.347 µm
C1 or C2
Al-OH Abundance Bands 106+113/110+111 A1&A2
Kaolinite Abundance Bands 109/108 = 2.167 µm /2.150 µm A1&A2 E1 0.952
White Mica Al-rich Band 110/111 = 2.184 µm /2.201 µm A1&A2 F1
White Mica Al-poor Band 113/113 = 2.219 µm /2.236 µm A1&A2 F1
MgOH/CO3 Index Band 115+122/118+119 A1&A2 E1 1.015+ or 1.02+ (95%)
Table 6.5 Description of the Spectral Indices used in the extraction of information from the HyMap imagery.
Name Bands Wavelengths Leaf/Surface water features Bands 48+59/51+53 (1.119 µm+1.279 µm)/(1.164 µm+1.193
µm) NDVI Merton/geochemical index
Bands 13 and 24 0.685µm and 0.854µm
NDVI photochemical reflectance index
Bands 3 and 5 0.533 µm and 0.563 µm
Red edge inflection Variable 0.7 µm and 0.74 µm Cellulose mask Bands 100+110/104+105 (2.205 µm+2.184 µm)/(2.078 µm+2.096
µm) NDVI traditional Rouse Bands 12 and 23 0.671µm and 0.837 µm
Table 6.6 Vegetation indices trialled on the HyMap imagery for the creation of vegetation masks.
Chapter 6
- 260 -
Vegetation Indices
Five types of vegetation indices were trialled to evaluate their effectiveness at
masking vegetation in semi-arid imagery. The indices were performed on
atmospherically corrected HyMap data of swath 1, which had been cross-track corrected
and had the bad bands removed as described earlier in the Chapter. The parameters of
the vegetation indices are presented in Table 6.6.
The Leaf/Surface Index (Figure 6.29) showed high vales for the dark areas of the
TCC imagery, which were presumed to be vegetation, such as the area around
Cartwrights Dam, the swampy region south of the Barrier Highway and the area in the
bottom corner of the swath corresponding to an alluvial swampy region in the
Adelaidean. Linear features perpendicular to topographic contours were highlighted,
which were interpreted as vegetation in erosional depressions. Banded vegetation and
the patchy Maireana s., (bluebushes) regions were highlighted as bright linear features
and patches respectively. Areas of Atriplex vesicaria occurred as smooth, mid- to
bright-tones. Areas of outcrop, un-vegetated soil, tracks, road and the railway did not
display significant values for this index and were classified as vegetation-free areas.
Thresholding the Leaf/Surface Index with a value of 1.03 produced a similar ROI mask
as the dark pixel generated mask using a SWIR band. The cloud features would still
need to be manually drawn and the masks integrated.
The Geochemical NDVI appeared to have a very similar distribution as the
Leaf/Surface Index, although it displayed a stronger contrast between very bright (dense
vegetation) region and low to pale regions (sparse vegetation). This ratio was found to
discriminating green vegetation better than the leaf index due to the higher contrast with
the background features (non-green vegetation).
The Photochemical, Red Inflection and the Traditional NDVI highlighted the
green vegetation features as well as the roads and tracks. There was a poor distinction
between the highway, vegetation and the railway track. These indices masked
important Fe-oxide features, making it less effective in regolith-dominated terrains.
These indices would be more suited to environments containing temperate vegetation
and vegetation dominated scenes.
I C Lau Remote Sensing Results
- 261 -
Figure 6.29 Results of the two vegetation indices showing the distribution of green, photosynthetic vegetation (green areas, corresponding to the Leaf/Surface Index) and areas of dominantly dry or woody tissue (brown areas, corresponding to the Cellulose Index, and representing arid vegetation and heavily grazed plants). Notice the north-northwest linear edges across the swaths representing fence lines.
Chapter 6
- 262 -
The Cellulose Index (Figure 6.29) had a distribution covering the areas adjacent
to creeks in low-lying areas. The material in erosional depressions, emphasized by the
green vegetation-dominated leaf/surface index, was not highlighted. The Cellulose
Index was more effective at picking out semi-arid vegetation due to the greater
abundance of non-green vegetation in the scene. This index was more likely to
highlight woody vegetation and dry/arid, non-leafy vegetation, such as chenopods and
heavily grazed plants that consisted of twigs and stems. Vegetation, such as Atriplex
vesicaria and Maireana sp., were identified using this method, whereas the leaf/surface
index identified trees (eucalyptus camaldulensis) small bushes (Emu bush, Hop bush)
and ephemeral grasses.
In the lower-central regions of the image a fence line boundary was clearly
delineated by the cellulose index due to the differing abundance of vegetation growing
between the paddocks. The southern paddock had a higher abundance of vegetation and
subsequently had a ‘bluer’ appearance in the TCC using HyMap bands 16-10-3 RGB.
The western edge of the paddock displayed a similar trend to the paddock further to the
west, which was more heavily grazed. A number of regions in the middle-upper portion
of the swath, which occurred to the north east of the White Dam floodplain, showed a
high abundance of cellulose. A north trending fence line occurred in this region, which
produced a contrast between the more heavily grazed eastern paddock and the vegetated
western paddock. The results show that cellulose had a wider distribution and a greater
abundance of pixels. This depicted sheetflow and alluvial plain regolith settings,
whereas the leaf/surface index was more representative of alluvial channels and
erosional depressions.
Field observations of the basement exposures from the White Dam area had
shown the prominent colonisation of fractures by trees and shrubs, as well as the growth
of cryptograms (lichen) on the surfaces of rocks. These features may dominate the
spectral signal and led to the masking of basement rocks in the early steps of the
processing.
The Leaf/Surface and Cellulose Indices were chosen for the creation of masks.
The two indices were thresholded using 0.97 and 1.05 cut offs for Cellulose and
Leaf/Surface respectively. These results were combined to construct a vegetation mask,
which was applied to the data to remove the pixels affected by vegetation from the
further processing steps. It should be noted that the threshold values used were selected
to allow material that may be partially vegetated to not be masked-out, as these regions
may of contained features of interest.
I C Lau Remote Sensing Results
- 263 -
Fe-Oxide Abundance
Fe-oxide Abundance Indices involved the use of similar ratios ((0.761 µm +
1.190 µm) / (0.886 µm + 0.919 µm and 0.761 µm/0.886 µm), which were applied to a
subset of vegetation masked VNIR bands (Figure 6.31). The former ratio appeared to
define a selective distribution of materials, which highlighted creeks and watercourses
as lows and areas of outcrop and bare-soil as highs. The later ratio displayed a
distribution was similar distribution of materials highlighted, with the only observed
difference of the graded road displaying as a high.
The Fe-oxide Index was found to correspond well to areas that had a high red
band DN in the TCC. This was attributed to the high Fe-content of the red coloured
soils, shown in Figure 6.30. Some of the white areas of the TCC, that were classified as
quartz-lag-dominated regions displayed a high Fe-oxide Abundance. This was
attributed to a high abundance of ferruginous material, such as ironstone lags, on the
surface. The area of the depositional plain that consisted of large areas of soil had a
high response.
Pale blue areas, which would have been mapped at saprolite exposures, had a
lower result. These outcrops consisted of grey-dark coloured rocks with minor
ferruginisation at the surface (Figure 6.30). The Adelaidean saprolite areas had a very
low response with a moderate response in the areas of alluvium and colluvium where
weathering is probably more abundant (Figure 6.30 i).
Figure 6.30 Fe-oxide Abundance image (left) and HyMap TCC (bands 16 10 3 RGB-right) from an area near MacDonald Hill. Saprolite (i) and the dark areas adjacent to creeks (iv) in the Abundance image are masked-out vegetation. The areas flanking the saprolite exposures (ii) (iii) display the highest Fe-oxide Abundances. Erosional depressions draining from exposures and creeks display moderate abundances.
Chapter 6
- 264 -
Figure 6.31 Fe-oxide Abundance and Mg-OH/RCA Abundance distributions from indices. Note the association of Fe-oxides with drainage features and alluvial regions, whereas the Mg-OH/RCA distribution is correlated with low rises and shallow soil areas flanking saprolite exposures. The Fe-oxide Abundance shows considerable differences in vales for swaths 1 and 4, which was due to poor levelling. Swath 1 contained minor patches of cloud and was slightly darker in albedo. Swath 4 had small noise features in the Fe-oxide regions that were not obvious until after processing was preformed.
I C Lau Remote Sensing Results
- 265 -
The interpretation of the results from the Fe-oxide related indices was that the
saprolite in the southern areas consisted of bleached kaolinite or the fresh rock had not
been exposed to extensive weathering and ferruginisation. This resulted in the lack of
formation of a ferruginous duricrust. The other possibility was that the rocks were
bleached saprolite, where the ferruginous material had been eroded and deposited
elsewhere, for example, in the erosional depressions and lower lying neighbouring
landforms. This is possible as the channels in the region and alluvial depositional areas
had a high Fe-content. The saprolite may have possessed a low Fe-content or was
covered by a weathering rind or cryptograms, which would result in a lack of response.
Saprolite in the northeastern regions of the study area displayed a high Fe-oxide
response. These basement exposures belong to the calc-silicate Ethiudna Subgroup and
coincided with the displayed highs for areas of the calc-silicate areas that form the
erosional rise to the northeast of the White Dam reservoir. The deep red/orange region
of the TCC to the northwest of White Dam reservoir also had a high Fe-oxide
Abundance. Areas surrounding outcrop appeared to have a high abundance. This may
related to weathering and colluvial material surrounding saprolite exposures. The
red/orange areas of the TCC to the north of MacDonald Hill and the region around the
Bulloo Creek road had a high abundance. The ration appeared to have a high
correlation to areas that appeared red in the visible and correlate with the band 16, at
0.65 µm. The alluvial depositional plain area of swampy clays also appeared to have
high Fe-oxide Abundances. Areas of exposed saprolite appeared to have a low Fe-oxide
Abundance, as seen in the area to the north of the White Dam Prospect. The rocks in
this region were gneisses and granites of the Wiperaminga Subgroup, which may not be
as Fe-rich as the calc-silicate rocks to the east and south.
A second Fe-oxide Abundance Index was trialled to determine if a similar result
would be obtained. The ratio used the parameters of 0.761/0.883 µm, and produced
similar distribution maps for Fe-oxide Abundances, with a very low value for regions of
saprolite outcrop to the north of the White Dam Prospect. Creek and streams had a very
high abundance. The depositional plain of banded vegetation to the northwest of the
White Dam reservoir and the alluvial depositional plain with swampy regions, displayed
high abundances of Fe-oxides.
The MacDonald Corridor region and the exposed saprolite/subcrop to the north
and the Wilkins Prospect had a high Fe-oxide Abundance. The White Dam floodplain
and regions around Cartwrights Dam (that were not vegetated) had a high response.
Chapter 6
- 266 -
The areas mapped as subcrop that had a red colour in the TCC displayed a medium to
high response. The swampy clay regions corresponding to depositional plains adjacent
to alluvial channels had high responses, as did the northern most dam of the swath and
the smooth textured area around the dam. The grey smooth areas that were thought to
be calc-silicate subcrop had low to no responses. The vegetated areas of the
depositional plain north of the Wilkins Prospect had a low response, which may be due
to the vegetation obscuring the Fe-oxide response. Some of the TCC maroon areas of
the Adelaidean depositional alluvial areas had a moderate to low Fe-oxide response.
Figure 6.32 Fe-oxide Abundance image (left) and HyMap TCC (bands 16 10 3 RGB-right) from an area east of Bulloo Creek Homestead, in an alluvial dominated region. Dark areas (i) adjacent to creeks, which were masked out in the Abundance image. Saprolite exposures (ii) and alluvial landforms display the moderate-high Fe-oxide Abundances.
Hematite: Goethite Ratio
The Hematite:Goethite Ratio was calculated using the ratio of 0.886/0.919µm.
The resulting data appeared to contain a minor vertical striping related to noise, possibly
due to the proximity of the water absorption band at 0.92. A second ratio, which used
the wavelengths 0.919 µm/0.868 µm, was trialled to discriminate goethite and hematite.
The data also contained vertical striping on the alluvial plain region to the northwest of
the White Dam reservoir, which was attributed to noise in the bands used in the ratio.
The two ratios show similar results with the distribution of goethite
predominantly occurring in regions related to creeks and channels, quarries, dams and
some areas of subcrop of weathered basement. Hematite was found to be associated
with the clay-rich/soil areas of the alluvial plain, as well as the dark brown swampy
regions in the central-northern areas of the imagery had a strong hematitic mineralogy
(Figure 6.33).
I C Lau Remote Sensing Results
- 267 -
Figure 6.33 Hematite:Goethite Ratio of a sheetflow and subcropping basement area of calc-silicate derived saprolite, which is dissected by an alluvial channel. (ii) The clayey material adjacent to the channel displayed a higher hematite proportion than the areas where the (i) colluvial cover was thinnest and goethite was more prominent.
Areas interpreted as quartz lag were found to be associated with higher
abundances of goethite than the areas distally flanking saprolite exposures. This could
be related to weathering of the calc-silicate rocks or the greater abundance of water in
these regions.
The calc-silicate erosional rise landforms to the northeast of the White Dam
reservoir exhibited a contrasting goethite abundance compared to the depositional plain
around the White Dam reservoir, which displayed more hematitic spectral
characteristics (Figure 6.34). Goethite was associated with the calc-silicates on the
southern side of the creek running south from White Dam reservoir. Areas of milky
grey colour in TCC corresponding to the northeast part of the White Dam floodplain did
not have Fe-oxide Abundance within the threshold requirements and therefore were
considered to had a low Fe-abundance.
Figure 6.34 Hematite:Goethite Ratio of the materials in the northern part of the White Dam alluvial plain, showing the higher proportion of goethite in the swampy regions and associated with the sheetflow-dominated slopes flanking basement subcrop in the left side of the image. The sediments of the alluvial plain displayed a higher proportion of hematite, shown in blue.
Chapter 6
- 268 -
Hematitic mineralogy was found to be prevalent in regions of fresh basement
outcrop where weathering was less prominent, areas of known muscovite mineralogy
and the swampy clay depositional plains associated with flooding areas (brown spotty
regions in the TCC). Most of the alluvial channels, plains and soils had a mixed
abundance that appeared to be slightly more goethitic. Areas with a high albedo in the
TCC, such as those interpreted as having bare soil or quartz lag, had goethitic
mineralogy.
Goethite was found to occur in the channels carrying material from the
MacDonald ranges and the badlands regions of the Adelaidean. A similar pattern was
found for the hematitic exposures of Willyama Supergroup. The outcrops were
commonly dissected by goethitic channels. The badlands of the Adelaidean region had
a higher goethite content where as the Adelaidean alluvial areas displayed a hematite
dominated mineralogy, as shown in Figure 6.35.
There were a few pixels in the southern and eastern edges of Cartwrights Dam
that correspond to goethite, whereas the rest of the dam did not have a significant Fe-
oxide Abundance. The material of the northern dam was goethitic, whereas the material
on the depositional plains was a mixture of the two. The area around the Wilkins
ironstone tended towards a mixed response, although there were a small number of
pixels immediately adjacent to the ironstone that corresponded to the mine workings
showing a high goethite response.
Figure 6.35 Hematite:Goethite Ratio showing the contrasting difference in the mineralogy of (i) the badlands, (ii) the colluvial regions flanking the Adelaidean exposures and alluvial-dominated regions in the far left of the images, which have a intermediate mineralogy (green areas).
Hydrated Fe-oxide Abundance
The Hydrated Fe-oxide Abundance was calculated using the formula 1.265
µm*1.289 µm/1.334 µm*1.347 µm, which corresponded to bands 58+59/63+64. The
Hydrated Fe-oxide Abundance was applied using un-masked data, which appeared to
I C Lau Remote Sensing Results
- 269 -
produce similar results to the first Hematite:Goethite Ratio, with much of the White
Dam floodplain displaying high vales, as well as the White Dam reservoir and the some
of the channels. A mask using the first Fe-oxide Abundance was applied to the data
with a threshold of 0.99. The result showed that the dam had a high abundance of
hydrated Fe3+. The soil and alluvial plain material around white dam tended to have a
higher ferric abundance away from the dam and the creek system and appeared to have
less near the creek. The material from upslope could be more goethitic than the material
near the creek. This goethite-rich material was interpreted as originating form
weathering of the saprolite. The area near the creek could be a mixture of overbank and
flood material, which may have a more hematitic or “normal” Fe-oxide Abundance than
the weathered saprolite material. The trend displayed in the Fe-oxide Abundance may
be related to dispersion of minerals down slope or the abundance of colluvial materials.
The Fe-oxides in the drainage channels displays lower values than outcrops and
areas near outcrop. The results were unclear for much of the swath, although the
badlands area displayed a high score, which contrasted strongly to the area to the south
corresponding to alluvial and exposures of Adelaidean rocks. The alluvial and
subcropping region appeared a maroon colour in the TCC, whereas the badlands region
had a high albedo with a colour to pale blue to white.
Aluminium Hydroxide Abundance
The Aluminium Hydroxide Abundance image (Figure 6.36) used the index of
2.1148 µm+2.2535 µm /2.1845 µm +2.2015 µm on a SWIR subset of data that had been
masked for cloud, shadow and vegetation. The resulting image was very similar to the
Fe-oxide Abundance image and also corresponded to the red coloured regions on the
ortho-image and the HyMap TCC red band (band 16/ 0.65µm). The similar distribution
of the Fe-oxides to the Al-OH Abundance was inferred to represent the presence of
saprolite and ferruginous regolith materials.
Drainage areas and Willyama Supergroup outcrops exhibited high Al-OH
Abundances, whereas the Adelaidean exposures had a low abundance. Some of the
exposed and shallowly covered saprolite areas did not display a high Al-OH Abundance
and were interpreted as quartz-rich, or calc-silicates that did not possess high amounts
of Fe-oxides or Al-OH within their constituent materials.
Chapter 6
- 270 -
Figure 6.36 Distribution map of Al-OH Abundance (apple green) with overlays
of Kaolinite (dark green) and White Mica (blue and red). The White Micas are
differentiated by their mineralogy, with Al-poor micas having a more phengitic
mineralogy.
I C Lau Remote Sensing Results
- 271 -
Lucky Dam was highlighted as having a high abundance of Al-OH as did the
drill spoils at the White Dam Prospect. The Al-OH Abundance was useful for
identifying outcrop in the alluvial/sheetflow-dominated region between Lucky and
Cartwright dams. This region contained an abundance of regolith carbonate and
sheetflow-related regolith materials. Areas of saprolite occurred at the higher elevation
points as inliers with abundant vegetated regions in between. The vegetation varied
from Atriplex vesicaria to Maireana Sp. (both pearl and black bluebush). The area to
the north, between the track and the dams, consisted of vegetation-dominated sheetflow
regions with a low Al-OH Abundance.
The maroon area of the alluvium in the TCC and subcropping Adelaidean had a
high Al-OH Abundance, whereas the badlands had a low abundance. The Adelaidean
material upslope displayed a high abundance, which decreased down slope to the south.
This was thought to be related to the transport of weathered materials by colluvial
processes and overland flow. Different regions that would be grouped as saprolite
displayed variations in the abundance of Al-OH. Primary mineralogical variations and
associated weathering products were attributed to the differences seen in the Al-OH
Abundances.
Some outcrops were predicted to be Al-OH poor, such as dolomites and
Adelaidean glacial lithologies, whereas granitoids should have a high Al-OH
Abundance, due to the abundance of micas and feldspars, which weather to form
kaolinite. The maroon regions of the Adelaidean alluvial and subcropping region
displayed a moderate to high response, which was contrasted by the low of the quartz
lag regions and badlands to the north. The channels of the badlands displayed a
moderate to high response, which was attributed to the erosion of materials from the
MacDonald Ranges, upslope to the north and west. The red to orange areas in the
northern side of the MacDonald ranges in the TCC, displayed very high Al-OH
Abundances, with dispersion trails into the erosional depressions and creeks. The
outcrop in the region displayed a moderate Al-OH Abundances.
Some of the northward draining creeks of the MacDonald Ranges had a high Al-
OH Abundance, and were spatially associated with high Al-OH saprolite, displaying
direct pathways to the outcrop, as shown in Figure 6.37. The low Al-OH creeks did not
had a direct pathway to the exposed saprolite, which occurred as ‘inliers’, surrounded
by vegetated slopes and colluvial and sheetflow-dominated landforms.
Chapter 6
- 272 -
Figure 6.37 Al-OH Abundance image from shallow soil areas to the north of the MacDonald Ranges. Isolated inliers high abundances relating to basement subcrop occur in the lower portion of the images are not flanked by erosional depressions as seen in the left and central areas of the images, which drain into the upper of the two channels.
Material draining northward from the MacDonald corridor had a high abundance
for considerable distances, which reflected the greater relief of the MacDonald regions
over the regolith-dominated plains and the higher energy of streams that carry eroded
material to the north.
In the region to the north of the MacDonald Ranges leading to the Wilkins
Prospect, there were a number of inliers of outcrop that have high to moderate Al-OH
Abundances. These exposures have erosional depressions and channels heading away
towards the east and north. The low Al-OH Abundance areas were associated with high
concentrations of vegetation and subsequently appeared greyish in the TCC (Figure
6.38). The material eroded from the MacDonald ranges was deposited as large, wedge-
shaped fans adjacent to the dam. This alluvial region joins to the eastward flowing
Cutana creek that traverses the low area to the south of the Wilkins Prospect and then
flows north into Cartwrights Dam. Isolated pods of high Al-OH Abundance, attributed
to higher standing regions, occurred on the alluvial plain.
I C Lau Remote Sensing Results
- 273 -
Figure 6.38 Al-OH Abundance image of an area to the west of the Wilkins Prospect (high area in the lower right part of the images), showing the masked vegetated areas and sheetflow-dominated areas that were colonised by chenopods (grey areas in TCC), which displayed low abundances. The basement exposures in the upper and right parts of the images, relating to shear zones, have high abundances.
The region containing Kalabity Shearzone Ranges had a high Al-OH
Abundance, related to the presence of muscovite- and white mica-rich exposures related
to the shear zones. The cores had high abundances that were fringed by halos of low
abundances. Narrow, linear high abundances, occurred perpendicular to the circular
halos and related to the dispersion of weathered materials down flanking erosional
depressions. The dispersion appeared to be wider spread than the Fe-oxide Abundance.
The creeks displayed a lower Al-OH Abundance with distance from the outcrop. Areas
of outcrop that displayed a moderate Fe-oxide Abundance were seen to have high Al-
OH Abundances. The Adelaidean exposures had a lower abundance than the Willyama
Supergroup exposures especially compared to the shear zone regions, adjacent to the
Barrier Highway.
The outcrops to the north of the Barrier highway had a high Al-OH Abundance.
This high Al-OH region stretched across to the south side of the Barrier Highway and
led to the Wilkins Prospect area. The area around the Wilkins prospect had a very high
abundance, which saturated the upper limits of the data. The creek to the east of the
exposures at the Wilkins Prospect had a high response, which flowed into Cartwrights
Dam. The area directly to the south and west of the Wilkins Prospect consisted of a
flood plain dominated by swampy regions and vegetation. The vegetation was
Chapter 6
- 274 -
highlighted by the cellulose and leaf index and masked-out in the subsequent processing
steps. The non-masked expanses had low Al-OH Abundances, which were related to
the increased quantity of clays. This contrasted the Kalabity Shearzone Ranges that
consisted of white mica-rich and bleached, kaolinised saprolite.
The region around Cartwrights Dam had a high response, which was correlated
to the orange smooth regions in the TCC, which were probably bare soils.
The swampy brown orange regions of the White Dam floodplain had a high Al-
OH Abundance, as did the neighbouring area that appeared to host basement subcrop.
Other areas of subcrop that had a less maroon appearance and a more grey look were
attributed to the calc-silicates and had a low to abundance of Al-OH. The creeks in the
region and the erosional depressions that drain off of the grey smooth calc-silicate areas
had a low to medium response that were related to weathering materials from the
lithology of the exposure or the movement of top soil (PSA) into creeks. The creeks in
the region had a moderate to high response.
White Dam flood plain consisted of low Al-OH Abundance regions, which were
mapped, using the ortho-photographs, as subcrop of calc-silicate saprolite and alluvial
materials, which had a high Al-OH Abundance (Figure 6.39). Much of the sheetflow-
dominated region to the northeast of the floodplain appeared to have a low Al-OH
Abundance except for specific channels and erosional depressions, which carried
material eroded from patches of high Al-OH. These represented areas where the
lithology had a more Al-OH-rich mineralogy or had been intensely weathered to form
kaolinite.
Widespread regions of Al-OH corresponding to red/orange regions in the TCC
occurred in the northeastern part of the swath. In general, the quartz lag-dominated
regions corresponded to calc-silicate lithologies and had a low Al-OH Abundance,
whereas saprolite derived from granitoid and quartzo-felspathic gneiss had a higher
abundance of Al-OH and displayed a red appearance in the TCC, possibly due to
ferruginous weathering. The swampy brown orange regions of the TCC showed a low
to medium abundance of Al-OH. These regions correlated to areas with high
abundances of hematite. There were high to medium concentrations within the brown
regions, which were isolated outcrop or bare soil patches.
I C Lau Remote Sensing Results
- 275 -
Figure 6.39 Al-OH Abundance of the alluvial dominated region on the southern area of the White Dam floodplain, showing the clayey areas having high abundances. (iii) Areas of shallow soils, vegetation and sheetflow-dominated landforms adjacent to the channel and alluvial regions display low abundances. The low abundances reflect the underlying lithology of this region and mantling PSA materials.
Al-OH Abundance calculated on the vegetation masked data showed highs for
creeks and channels, material within the White Dam reservoir and the old dam to the
south west, on the depositional and erosional plain region, and regions of exposed rocks
in the northern regions of the swath and the channels, the drill spoils and the erosional
rise of saprolite to the south of the main White Dam Prospect.
A threshold of the results of 1.1-1.5 was used to create a Al-OH mask, which
was used to identify kaolinite and white mica minerals. The mask was applied to a
reduced subset of wavelengths between 2.1 and 2.25 µm.
Kaolinite Abundance
The Kaolinite Abundance Index used the ratio of 2.167/2.150 µm. The results
showed high Kaolinite Abundance occurred at the fringes of the areas mapped by the
Al-OH Abundance Index (Figure 6.40). Outcrop regions appeared to have a zonation of
low kaolinite in the centres with high Kaolinite Abundance at the fringes. This was
attributed to the extensive weathering of saprolite and transportation of kaolinite
materials downslope, with white mica minerals occurring in the centres of the Al-OH
regions. These regions could represent less weathered saprolite that still possessed
muscovite and other primary minerals related to fresh bedrock.
Chapter 6
- 276 -
Figure 6.40 Kaolinite Abundance demonstrating the high abundance haloes around bedrock exposures, related to weathering of the eroded materials.
Creeks displayed a low Kaolinite Abundance, which may be related to high
amounts of montmorillonite and other clays, or the large abundance of lithic fragments
derived from micaceous rocks and feldspars, which eroded from the exposures, but had
not yet weathered to kaolinite. The latter may be true in regions flanking erosional hills,
as seen to the north of the MacDonald Hill Ranges, as shown in Figure 6.41. The creeks
in this region were bare of vegetation and contain coarse gravels, with abundant lithic
fragments of feldspars, mica ± heavy mineral sands. To the south of MacDonald Hill
the creeks flowing south from the Adelaidean basement rocks had a higher abundance
of kaolinite than the exposures. The regions that corresponded to high Al-OH
Abundances in this region. This may be attributed to the richer kaolinite mineralogy of
these rocks.
Figure 6.41 Dispersion of high Kaolinite Abundances (iii) down slope and in (iv) erosional depressions draining from weathered Adelaidean bedrock exposures in the topographically elevated MacDonald Hill Region.
I C Lau Remote Sensing Results
- 277 -
The channel originating from the central part of the Kalibity inlier, which cuts
by Bulloo Creek road, had a low Kaolinite Abundance, possibly due to a large
abundance of feldspars and micas derived from the Willyama Supergroup lithologies
(Figure 6.42). The materials weathering from the Adelaidean rocks in the MacDonald
Corridor displayed smaller sized halos, but possessed intense occurrences of kaolinite.
These rocks may weather quickly and discretely, as suggested by the confined
distribution of flanking weathered materials. Most creeks and alluvial areas away from
the exposures displayed moderate to low values, which were related to low crystalline
kaolinite or the presence of smectite.
Figure 6.42 Low Kaolinite Abundances in the channels eroding from the central portion of the Kalibity Inlier. Figure shows the low abundances at the central area of saprolite exposures and high abundances at the fringes of the dispersion of eroded materials. Vegetation masking has removed information from prominent regolith-landform features, such as (iv) and alluvial fan and (iii) a drainage depression adjacent to bedrock exposures.
The calc-silicate regions appeared to have a higher abundance of kaolinite than
the regions to the south that were interpreted as gneisses and granites, and had a more
white mica-rich mineralogy. In the upper northern regions of the swath, the areas with
an orange appearance in the TCC had a low Kaolinite Abundance. Around the
exposures adjacent to the Barrier Highway kaolinite occurred as highs surrounding
inliers of low abundance, which could be related to weathering of outcrop.
The northern part of the White Dam floodplain had a low Kaolinite Abundance,
related to a large abundance of smectite in this region. The area directly east of the
creek that runs south of White Dam reservoir displayed a high abundance of kaolinite,
which was attributed to the weathering of saprolite exposures to the east.
Kaolinite was found to occur in the Adelaidean alluvial and flood plain areas,
which had low cores with the highest values at the fringes of the classified areas.
Chapter 6
- 278 -
Occurrences of high Kaolinite Abundances also occurred in the drainage depressions
that carried material eroding from the low hills of Adelaidean basement into the
badlands region. The material in the MacDonald Ranges appeared to have cores of low
Kaolinite Abundance with halos of increasing Kaolinite Abundance. The Adelaidean
basement had a higher Kaolinite Abundance than the Willyama Supergroup basement,
which displayed predominantly as lows.
Channels appeared to have a low abundance of kaolinite, which may be due to
the presence of clays, rather than kaolinite. The northern dam showed that areas of bare
soil had low abundances of kaolinite in the central portion with an increase at the
fringes.
The results of the Kaolinite Abundance Index were used to build a mask using
the threshold of 0.01-0.95, which was used to mask out the regions of kaolinite. The
mask was applied to the SWIR vegetation masked Al-OH Abundance results and a
White Mica Index was applied to determine the mineralogical composition of the white
mica. The ratio used the bands 110/111 and the results theoretically show that the Al-
poor mica having high values.
White Mica Mineralogy Index
The first ratio of White Mica Abundance used bands 110/111, which coincided
with the wavelengths 2.184 µm/2.201 µm. Theoretically Al-poor mica (phengite)
would have a high value for this index and Al-rich mica (paragonite and muscovite)
would have a low value. The results showed a number of saprolite exposures as having
high values, which may indicate the presence of phengite. Linear zones of alteration,
which when reconstructed run roughly east-west, are shown in Figure 6.43. The
material in the creek bed displayed Al-poor mineralogy.
Alluvial regions adjacent to the shallowly covered Adelaidean basement had a
low to moderate response, representing a high to moderate Al abundance. The material
from the MacDonald corridor had a low Al abundance in isolated locations, which had
weathering trails down the creeks and drainage depressions that extended to the north.
These anomalies were associated with Willyama Supergroup exposures and weathering
products.
I C Lau Remote Sensing Results
- 279 -
Figure 6.43 White Mica Abundance image showing the Al-poor characteristics of the channel and different exposures of saprolite.
The outcrop on the northern side of the MacDonald Ranges had isolated
occurrences of highs, which had dispersion pathways from the outcrops in the erosional
depressions and the larger alluvial channels (Figure 6.44). The drainage features
trended to the north and ended abruptly at a fan, which joined the eastward flowing
Cutana Creek. The fan represented a change in the energy of the stream due to the slope
angle changes, resulting in a drop in energy and led to the deposition of material. This
related to a change of sediment load from material carried as lithic fragments to the
transport of silt sized particles. The change in material being transported was reflected
in this ratio by the change in Al-OH parameters. The materials in the streams proximal
to outcrop consisted of micas and lithic fragments, whereas the distal streams carried
smectite and weathered minerals.
In the northeastern regions of the MacDonald Ranges scattered exposures of
basement displayed a moderately high response (Al-poor) that trended to Al-rich (low
response) at the fringes, which may have been related to the increase in kaolinite
downslope. The outcrop tended to have a lower response (became Al-richer) to the
north, away from the MacDonald Ranges, which was related to increased weathering of
the saprolite.
Chapter 6
- 280 -
Figure 6.44 White Mica Abundance images of an area of shallow soil cover in the vicinity of a fence line, showing the effects of vegetation on the ratio. Note the (ii) dispersion of intermediate materials in the drainage feature. The area in the left part of the image shows an intermediate to Al-poor mineralogy, whereas the more vegetated area has abundances reflecting a more Al-rich mineralogy.
The creek trending north from the MacDonald Ranges had a high value,
indicating an Al-poor mineralogy, which was attributed to the erosion of relatively fresh
materials from the saprolite in the hills of the MacDonald Corridor or the presence of
alluvial minerals that were Mg-rich. The clayey areas of the alluvial depositional plain,
displayed as brown mottled appearance on the TCC, had a low response (Al-rich). Near
Little White Dam, between the Wilkins Prospect and the MacDonald Ranges, a fan
marked the drop in carrying capacity of the stream, due change in slope gradient. A fan
had formed where the overbank deposits had broken the banks of the stream, which was
highlighted by the White Mica Index by a moderate to low response (Al-rich).
Minerals in the area to the south of the Wilkins Prospect had a low Al
abundance, related to the presence of muscovite and Mg-rich micas associated with
alteration of the Wiperaminga Subgroup basement rocks. The area contained
leucocratic pegmatitic sweats throughout the migmatised basement rocks, which caused
the high White Mica Abundances. Immediately adjacent to the ironstone associated
with the mineralisation at the Wilkins Prospect there were no data recorded from the
indices on the Al concentration. This was attributed to the low Al-OH Absorption
recorded for the ironstone, the presence of vegetation and sheetwash sediments in
localised region around the exposed ironstone body. The material in the creek draining
from the Wilkins region had a high response (similar to outcrop).
White Mica Abundance displayed a high abundance for the outcrop region near
the Barrier Highway, which coincided with retrograde shear zones. This was related to
the higher than normal abundance of muscovite in these regions and the occurrence of
pegmatitic melts, observed in the field during ground truthing.
I C Lau Remote Sensing Results
- 281 -
To the southwest of Cartwrights Dam an area was found with a high White Mica
response (Al-poor). A number of regions to the west of Cartwrights Dam had a Al-poor
response, which appeared to be related to subcropping material. Some of the
subcropping grey material displayed low Al White Mica Abundances, whereas the
adjacent regions did not display any noteworthy values for the Al-OH Abundance.
Much of the brown clayey alluvial depositional plain of the WD flood plain had a low
score (Al-rich).
In the northern regions of the swaths there were patches of the quartz lag
material that had a high Al White Mica Abundance, with no real clear reason for the
patterning. Neighbouring regions with a similar TCC appearance had a completely
different Al content or no Al abundance at all. These examples highlighted the
variability that is not obvious from air-photos. The northern dam had a high response
(Al-poor), as did the smooth orange -soil area in the TCC to the southeast of the dam.
The clayey areas of the alluvial depositional plain, which displayed as orange-
brown regions in the TCC, had high White Mica Abundances. Tracks running through
the area in a northerly direction can easily be seen. An interesting point is the fact that
the pale blue regions to the east had a variable composition with some regions showing
an Al response and others having none.
The White Mica Abundance showed the mica in the Barrier highway region had
a high value from the ratio and therefore in Al-poor, possible phengitic in mineralogy.
The basement exposures to the north of Bulloo Creek had a high White Mica
Abundance, while the creek also displayed a relatively high abundance. Much of the
other material had a low abundance including the creeks that drain the regions. These
features had a high Al content and were probably kaolinite- or muscovite-rich. The clay
material in the dam floor of Lucky dam had a different chemistry than the dam walls,
which may be more muscovite-rich. An area directly to the east of the White Dam
Prospect that displayed a lower abundance of Al, which is represented by a bluish black
rounded spot on the TCC.
A similar ratio to the one above was applies using a longer wavelength (bands
110/112). The ratio of 110/112 was also processed to determine the effectiveness of
this ratio for determining white mica abundance. The results were almost identical.
Magnesium Hydroxide and Carbonate Index
A Mg-OH/CO3 index (Figure 6.31) was created by the author in an attempt to
map regolith carbonate by the calcite features. Calcite possessed a deep, asymmetrical
Chapter 6
- 282 -
absorption in the 2.335 µm region, with inflection features on the short wavelength side
of the absorption feature. Due to the dominance of Al-OH absorption features in the
scene, the feature at 2.2 µm is predominantly deeper than the 2.3 µm absorption
throughout the swaths. Kaolinite, muscovite and illite all had secondary features in the
2.3 µm region in the hyperspectral data. These phenomena caused the 2.3 µm region to
have absorption features associated with the Al-OH minerals, as well as Mg-OH and
CO3 minerals. Most regolith carbonate-related absorption features occurred within soils
or mixtures of other materials (including vegetation), which caused the carbonate
features to be obscured and undetected. Most of the regolith carbonate material in the
region occurred in association with soil.
The areas of known RCAs, such as rabbit warrens, can be seen in the TCC
imagery as roundish spots throughout the landscape. They predominantly occurred in
alluvial landforms or in shallow-soil regions on low rises that fringe weathered saprolite
(Figure 6.45).
Figure 6.45 Mg-OH/RCA Abundance Index showing an interpreted distribution of regolith carbonate in the MacDonald Ranges region. Ground truthing of the region found abundant hardpan RCAs in the sheetflow dominated landforms that flanked the basement exposures.
The bands 115+122/118+119 were used in an attempt to pick out features with a
broad absorption in the 2.3369 µm region, with shoulders at ~2.27 µm and 2.385 µm.
Results from the ratio highlighted in the HyMap imagery were identified as rabbit
warrens in the TCC, which displayed as white or cyan coloured circles (Figure 6.46).
I C Lau Remote Sensing Results
- 283 -
Some areas were identified as outcrop displayed high values for the Mg-OH/Carbonate
Index and may have been either saprolite indurated by RCAs, or saprolite consisting of
Mg-OH minerals derived from lithologies that were Mg-OH-rich, such as amphibolite.
All of these features had 2.2 µm absorption features, attributed to soil mantles or
weathering products.
Figure 6.46 HyMap spectra of a regolith carbonate-rich soil from a rabbit warren and a calcite spectra from the USGS Reference Library, showing a slight absorption in the 2.34 µm region. This feature was used in the Mg-OH/Carbonate Index to map the distribution of regolith-carbonate and mafic materials.
Summary of Indices Performed on the HyMap Imagery
The pre-processed data were evaluated for areas of green and woody vegetation
using (1) and (2) indices. These regions were used to mask the dataset for subsequent
processing. The vegetation masks were retained as a classification product. Areas of
minimal vegetation, identified from the use of vegetation masks were examined for Fe-
oxide, Mg-OH/CO3 and Al-OH Abundances. The Fe-oxide Abundance was used to
highlight the areas for further processing. A mask was generated of the higher
abundance areas and the Hematite:Goethite Ratio was applied. The areas representing
goethite were used in a mask to determine areas of hydrated Fe-oxides. The AL-PH
Abundance Index was used on vegetation-masked data, in turn is used to determine
areas of Kaolinite and White Mica. The Mg-OH/RCA Index was designed to determine
the areas where RCA occurred. Examining the HyMap spectral response of regions
where RCA were observed in the field, found a strong Al-OH absorption in the 2.2 µm
region that was accompanied by a broad, asymmetric 2.3µm absorption.
Chapter 6
- 284 -
The use of indices appeared to be a better option for mapping Fe-oxides than the
use of un-mixing techniques.
Indices for other materials such as regolith carbonate could be attempted using
CO3 feature at 2.3µm
The use of indices for separating regolith clays and smectite from outcrop white
micas (such as muscovite) as been shown to be successful with this dataset and could be
used to map outcrop.
Vegetation indices proved to work extremely well, with different types of
vegetation and vegetation communities being able to be discriminated. Although the
masking of vegetation can prove to be a problem as vegetation occurs on outcrop in
many of the saprolite exposures in the region. This was due to colonisation of fractures
in silicified/ Na-Ca altered rocks and the growth of vegetation on well-drained soils
overlying intensely weathered saprolite.
Analysis of Airborne Gamma-Ray Survey Sata over the White Dam Region.
Pre-processing
The radiometric data were supplied as separate datasets that required levelling
and mosaicing. Data over the White Dam Prospect were acquired by MIM with closer
line spacing, and on average, had a higher abundance eU and displayed a cleaner eU
channel for this radioelement. The larger dataset was part of the BHEI regional dataset
of the southern Curnamona Province. No further pre-processing was performed on the
data, except for registration to GDA94/MGA 54.
Image Presentation and Information Extraction
Four processing methods were used on the radiometric data. The initial analysis
consisted of examination of the individual channels as pseudocolour images with an
artificial hill-shade to highlight trends and variation. The radioelements were displayed
as ternary images (K eTh eU-RGB) to highlight associations between the elements.
Landform associations were interpreted from the hill-shaded ternary image draped over
a DEM. The final method of method involved the use of traditional supervised
classification techniques on the ternary data.
I C Lau Remote Sensing Results
- 285 -
Classification of the Radiometric Data
The radiometric data were subset to an area the same area as the HyMap data
coverage in ENVI. The subset was split into two areas due to differing properties of the
gamma ray surveys, which were processed separately. The subsets were classified
using a maximum likelihood technique with supervised class selection. Classes were
selected based on: potassium highs, thorium highs, uranium highs, potassium and
thorium highs; thorium and uranium highs, potassium and uranium highs, potassium
lows, potassium mediums, thorium lows, thorium mediums, uranium lows, uranium
mediums, Adelaidean rocks and drainage channels (Table 6.7).
The smaller White Dam radiometric dataset did not contain Adelaidian rocks
and the large drainage channel so they were not included in the classification. The
supervised classification was found to reliably delineate areas of outcropping Willyama
Supergroup rocks as K, Th and K+Th areas. Adelaidean rocks were poorly
differentiated from the drainage sediments, which appeared magenta in colour (Figure
6.47).
Number Colour Description 1 Red U high railway 2 Green K-Th high 3 Blue Drainage plain 4 Yellow K high 5 Cyan K high MIM 6 Magenta K-Th-U High MIM 7 Sea green Adelaidean 8 Maroon Adelaidean 2 9 Purple Th high 10 Coral K drainage 11 Aquamarine Th moderate values 12 Orchid MIM cyan coloured highs 13 Sienna MIM dull areas 14 Chanteuse MIM pinky K moderate
Table 6.7 Radiometric classification descriptions and class colours.
Chapter 6
- 286 -
Figure 6.47 Supervised classification results performed on the ternary radiometric data, displayed in colours corresponding to the RGB image.
Total Count
The total count band showed the radioactive materials that were contributing to
the signals in the other channels. There was a distinct lack of signal from the Adelaidean
units adjacent to the MacDonald Ranges, corresponding to the badlands area, with some
signal from an area further to the south. The regions dominated by alluvial material
displayed low responses and corresponded to the areas of lower topography.
Uranium
The eU channel contained a large amount of noise and appeared speckled, even
after filtering, using techniques typically applied to radar data, which attempted to
I C Lau Remote Sensing Results
- 287 -
remove background noise. In the northern portion of the flight area the data appeared
much more coherent with a number of eU anomalies. The eU channel appeared
attenuated in the areas of alluvial material and in the Adelaidean lithologies. This was
attributed to the abundance of transported material and the increased abundance of
moisture in the alluvial landforms.
The railway line was prominent due to elevated eU, appearing blue in the ternary
image (Figure 6.48). This anomaly was attributed to the presence of materials with
radiogenic isotopes dropped from the transport of ore from the Radium Hill and Broken
Hill deposits to Port Pirie.
Thorium
There were numerous areas where the eTh channel had a coinciding spatial
distribution to the K% channel. The eTh anomalies in these areas were more widely
distributed than the concentrations of K values. This was attributed to dispersion of
colluvial and lag material from the basement outcrop travelling downslope, whereas K
corresponded to the exposure at the top of the rise. When draped over a DEM, the
down slope areas appeared to have low K and decreasing eTh values (Figure 6.50).
High anomalies of eTh were related to the presence of residual ferruginous
materials, such as ironstone lags. These were seen in the northeastern part of the study
area, where sheetflow processes dominated much of the landscape over subdued
topographic relief landforms (Figure 6.49).
Potassium
In areas where there is a K% anomaly and no eTh response, the saprolite was
interpreted as a K-rich rock such as a granitoid. The K% channel displayed a more
localised halo than the eTh channel around basement outcrop. There were a number of
linear anomalies of higher than surrounding values in areas of uniform low response,
which can be attributed to freshly weathered materials in alluvial channels, such as K-
feldspar-rich lithic fragments. The K channel appeared to have the least amount of
noise and produces the most coherent image (Figure 6.49).
Alluvial
The alluvial regions defined on the ortho-image and the Landsat TM imagery
had the appearance of large, uniform expanses of magenta in the RGB K-eTh-eU image
(Figure 6.48). The region around the White Dam reservoir and the areas to the west and
south occur as broad, elongate features. These areas were defining regions of potassium
and uranium from soils and vegetation. Moisture, held in clays, would attenuate the
Chapter 6
- 288 -
signal of these regions. Regions of dense vegetation along alluvial courses had a high
eU concentration, which backs up observations by Dickson et al. (1996).
The Adelaidean units had a dark brown to magenta colour, reflecting their low
radiogenic response (Figure 6.48).
Green areas representing eTh enriched areas had mottled yellow patches
associated with them where there were high concentrations of both K and eTh. These
regions were defining outcrops or shallowly buried Willyama Supergroup basement.
The mineralisation regions of Wilkins and Luxemburg displayed elevated eTh and K
(yellow areas). Cyan coloured areas were thought to represent areas of weathered mafic
rocks or regions of ferruginous materials (duricrusts) where there had been a
concentration of eU and eTh
Figure 6.48 Ternary K eTh eU RGB image of the combined radiometric dataset of the White Dam area. The upper section of the image was from a MIM airborne survey
I C Lau Remote Sensing Results
- 289 -
whereas the lower portion was part of the BHEI regional survey of the southern Curnamona Province. Alluvial-dominated areas have a dark green-dark brown colour.
Figure 6.49 Individual pseudocolour radiometric channels, hill-shaded and draped over a DEM. K and eTh display similar distributions, whereas eU contains a higher level of noise and is less coherent.
Three dimensional draping of radiometric data
The radiometric imagery was draped over the 25 m DEM as pseudocolour
images. Radiometric highs correlated with topographic highs when the gamma-ray data
Chapter 6
- 290 -
were draped over a DEM (Figure 6.50). This was attributed to the exposures of
Willyama Supergroup rocks, in particular the granitoids and sheared rocks, which
exhibited high radiogenic responses.
The uranium channel was found to display highs in the northeastern region of
the dataset, where granitic rocks are reportedly outcropping (Figure 6.49). This region
was visited in November 2002 where a series of hills covered with colluvial forming
granitic rocks outcropped with Aeolian, alluvial and colluvial soils with abundant
chenopods between the hills.
The areas of prominent drainage contained uranium lows. Examples include the
alluvial plain in the centre of the area. The Willyama Supergroup and Adelaidean rocks
in the McDonald Corridor region all display low values. The western central part of the
image displays a slight anomaly. This has been interpreted to the granitic outcrops
along the road to the Bulloo creek homestead.
The thorium image displays a similar trend to the uranium channel in that the
alluvial and Adelaidean regions are distinguishable by lows (Figure 6.49). This was
particularly pronounced in the central region of the flight lines and the northern draining
slopes of the McDonald Corridor. There is a anomalous region in the western region of
the data in an area of pronounced lag and sheet-flow material. Thorium anomalies
display a strong association with topography. Areas of higher topography generally
represent areas of basement exposures. The Wilkins prospect displayed a very high
anomaly (red), which could be an indication of the thickness of the regolith or the
radiogenic properties of the rocks in the region or an alteration association due to the
mineralisation. There is no railway anomaly, as seen in the uranium data. The northern
granitic rocks that are anomalous in the uranium channel are also high in the thorium.
The potassium image displays similar anomalies to the thorium channel (Figure
6.49). Areas of topographic highs display high anomalies, where as low and
topographically flat regions have low anomalies. These regions are indicative of
drainage regions and areas of thicker regolith cover. The potassium channel displays
elongate features interpreted to be due to erosion of potassic material into drainage
features.
I C Lau Remote Sensing Results
- 291 -
Figure 6.50 Oblique perspective view of a ternary K eTh eU (RGB) image of the White Dam area draped over a DEM. Draped imagery was useful for showing the relationship between regolith materials and units to landforms. Low hills and erosional rises were found to be related to saprolite exposures or subcrop with shallow soil mantles.
Recommended