HOME

Ni Laterite Mineralogy and Chemistry –

A New Approach to Quantification
M A Wells1 and J Chia2

ABSTRACT
A new approach to more reliably predict the Ni (and Fe, Mg) composition of Ni laterites was
tested for RC samples from the (siliceous) oxide–style, Siberia North deposit, Kalgoorlie Nickel
Project (KNP), in the north-eastern Yilgarn (WA) using partial least squares regression (PLSR)
analysis of diffuse reflectance spectra over the 380 - 2500 nm wavelength range with a multi–
element chemistry data set. PLSR cross-validation models were developed to predict FeO, MgO
and Ni contents in composite samples with bulk composition limits of >0.08 wt per cent Ni and
<6 wt per cent MgO. Calibration and validation models showed a strong, linear trend for predicting
FeO and, to a lesser extent, Ni content, whereas PLSR models for predicting Mg content were less
reliable. This reflected the prevailing mineralogy and associated element deportment within the
oxide laterite profile. For example, Fe and Ni distribution is controlled by relatively few minerals,
predominantly goethite and, to a lesser extent, the Fe3+-bearing smectite, nontronite. Magnesium
is associated with a range of Mg-bearing minerals, some of which are compositionally complex
(eg serpentine, talc, chlorite, smectite and carbonates). Magnesium chemistry alone does not
reflect the varied mineralogy of samples with relatively high (5 - 6 wt per cent) MgO contents,
which can be spectrally distinct. Hence, samples with similar Mg contents, though comprising Mg-
bearing minerals with distinct reflectance spectra, were difficult to model (eg smectite versus Mg-
carbonates). The findings of this study demonstrate that characterisation of Ni laterite composition
may be achieved as part of routine logging of either drill chips or diamond drill core using PLSR
analysis, but that a detailed understanding of the mineral-element association is essential for
reliable predictive analysis. This has particular relevance to characterising the geometallurgy of Ni
laterites, especially where deposits may be classed as the same type, but have distinct compositional
and mineralogical differences.

INTRODUCTION
The use of infrared reflectance spectroscopy, typically covering
the wavelength range 400 to 2500 nm, has now become
an established technique for semi-quantitative mineral
analysis in exploration and mining (Ramanaidou et al, 2004;
McConachy et al, 2007). Direct mineralogical mapping and
characterisation of soils and sediments, weathering profiles
(eg laterites) and of ore systems, such as iron ore deposits is
now routine (eg Tamish, Ganz and Bowitz, 1987; Malengreau,
Weidler and Gehring, 1997; Malengreau, Muller and Calas,
1994; Crowley et al, 2003; Wells and Ramanaidou, 2004).
Recent advances in instrumentation hardware have seen
the development of automated spectral scanning systems,
such as CSIRO’s HyLogging™ systems, which enable rapid,
very high-volume spectral analysis of diverse sample media
including drill core, chips and pulps (Ramanaidou et al,
2004; Huntington, 2005; Ramanaidou, Wells and Connor,
2005), and can circumvent problems associated with small
sample size in relation to more ‘traditional’ characterisation
techniques, such as XRD and XRF analysis.
Despite the host mineralogy of oxide and smectite-type,
lateritic Ni deposits (Brand, Butt and Elias, 1998; Freyssinet
et al, 2005) such as those within Western Australia (eg Murrin
Murrin, Ravensthorpe and Cawse), comprising spectrally
‘active’ ore and gangue phases, there are few published studies
where reflectance spectroscopy was used to characterise Ni
laterite ores. This may, in part, be related to the complex
mineralogy/chemistry of these ores where Ni mineralisation
can be hosted within a range of 1:1 and 2:1 layer silicates (eg
serpentine, smectite, chlorite), and Fe/Mn oxides, such as
goethite and ‘asbolane’.
In an early study, Maquet, Cervelle and Gouet (1981)
demonstrated excellent predictive correlations, to within
two per cent (relative), of the Ni content in ‘unknown’ saprolite
ore based on a spectral model developed from known mixtures
of goethite and ‘garnierite’. At the profile and deposit-scale,
Huntington (2005) and Wells and Butt (2000) demonstrated
the effectiveness of the spectroscopic technique by mapping
variations in Ni laterite mineralogy down-hole and in block-
models, respectively.
Increasingly, partial least squares (PLS) analysis, also
known as partial least squares regression (PLSR) analysis,
has found application in modelling reflectance data to aid
characterisation and prediction of a range of soil properties,
such as swelling clay and calcium carbonate content,

1. Senior Research Scientist, CSIRO Earth Science and Resource Engineering (CESRE), 26 Dick Perry Avenue, Kensington, Perth WA 6151. Email: martin.wells@csiro.au
2. Research Scientist, CSIRO Mathematical and Information Science, Underwood Avenue, Floreat, Perth WA 6014. Email: joanne.chia@csiro.au

THE FIRST AUSIMM INTERNATIONAL GEOMETALLURGY CONFERENCE / BRISBANE, QLD, 5 - 7 SEPTEMBER 2011 187
M A WELLS AND J CHIA
organic and inorganic carbon contents (eg Goetz, Chabrillat
and Lu, 2001; Gomez, Lagacherie and Guillaume, 2008).
In one of the few studies to have used this approach for
characterising Ni laterite, chemistry, Gilman (2001), used
PLSR analysis of near-infrared (1300 - 2500 nm wavelength
range) spectroscopic measurements to predict MgO and
FeO contents with comparable accuracy to XRF analysis.
However, combined infrared–PLSR analysis was not sensitive
enough for reliable prediction of Ni content (Gilman,
2001). However, not including the VNIR wavelength range
(400 - 1100 nm) excludes important spectral information
from iron oxides, such as goethite, which can be a significant
host for Ni mineralisation. Later work by Yang, Whitbourn
and Mason (2004), using active-light-source scanning of mine
faces at the Koniambo lateritic Ni deposit, New Caledonia,
demonstrated good agreement, using PLSR analysis, between
predicted and assayed Ni, Fe, MgO and Al2O3 contents.
However, the prototype illumination scanning system tested
by Yang, Whitbourn and Mason (2004) may not meet the
operational requirements of some active mine sites.
Recently Basile et al (2010) successfully demonstrated the
application of several multivariate techniques, including
principal component analysis (PCA), regularised discriminant
analysis (RDA) and partial least squares (PLS) analysis for
quantifying the serpentine content in a range of prepared
mineral mixtures that simulated various Ni–laterite types.
Serpentine contents of 5 - 40 wt per cent, with equivalent MgO
contents of 2.18 - 17.4 wt per cent were successfully modelled
using PLS analysis (Basile et al, 2010).
This paper describes the development of reflectance
spectroscopic models using PLSR analysis of the full VNIR–
SWIR wavelength range (ie 400 - 2500 nm), for quantitative
prediction of Ni laterite ore mineralogy and composition
at low MgO contents (ie <6 wt per cent), for the (siliceous)
oxide-style, Siberia North Ni laterite deposit, Kalgoorlie
Nickel Project (KNP), in the north-eastern Yilgarn, Western
Australia.
MATERIALS AND METHODS
Geology and location
Composite (ie pulped) material from the Siberia North lateritic
Ni deposit was examined for the purpose of demonstrating
the application of PLS analysis to the problem of Ni laterite
charaterisation. The Siberia North deposit, as a representative
example of a siliceous, oxide-style Ni laterite (eg Brand, Butt
and Elias, 1998), forms part of the regional Kalgoorlie Nickel
Project, KNP, (Heron Resources) and is located ≈100 km
north of Kalgoorlie (Figure 1). A more detailed description
of the laterite profile of the Siberia North and other related
deposits of the KNP is provided by Elias (2006).
Located in the Ora Banda Domain on the western side of
the Goongarrie–Mount Pleasant anticline (Elias, 2006), the
Siberia North deposit is hosted within the mesocumulate
and adcumulate Walter Williams Formation (Hill et al, 1995;
Brand, Butt and Hellsten, 1996; Elias, 2006). Earlier work by
Brand, Butt and Hellsten (1996) and Elias (2006) provide a
more detailed description of the regional geology of the Walter
Williams Formation. As at 2008, the Siberia North deposit is
a resource of 82 Mt at 0.64 per cent Ni and 0.04 per cent Co
(www.heronresources.com.au; viewed 9th December 2009).
HyChips™ automated scanning system
Approximately 2000 composite (ie pulped) samples each
representing a 2 m interval from two traverses comprising
188 THE FIRST AUSIMM INTERNATIONAL GEOMETALLURGY CONFERENCE / BRISBANE, QLD, 5 - 7 SEPTEMBER 2011
44 drill holes from Siberia North were scanned over the
400 - 2500 nm wavelength range using the HyChips™
6 - 1 automated scanner. Samples were placed in plastic,
20 cell chip-trays and three, single ‘spot’ reflectance spectra
were collected for each sample using quartz halogen lamps
as a constant illumination source. ‘Spectralon’ was used
as the 100 per cent reflectance standard against which the
HyChips™ spectrometer was optimised and calibrated prior
to measurement. Further technical details of the system are
provided in Ramanaidou et al, (2008).
Raw reflectance data were processed using The Spectral
Geologist Core (TSG-Core™) version 7.00.030 software
(see www.thespectralgeologist.com) (Berman, Bischof and
Hungtington, 1999; Yang et al, 2005) to initially average the
three spot measurements, using equal weighting, to derive
a single spectrum for each sample pulp. Using The Spectral
Assistant (TSA™) mineral matching routine within TSG-
Core™, 10 mineral groups were identified based on the most
reliable, dominant mineral identified, viz:
 kaolin (includes kaolinite/halloysite),
 opal,
 smectite (includes montmorillonite/nontronite),
 Mg-clays (includes saponite/palygorskite),
 chlorite (includes Fe/Mg/intermediate chlorite types),
 talc,
 serpentine (as lizardite),
 primary silicates (includes hornblende/riebeckite),
 carbonates (includes dolomite/magnesite), and
 Fe-oxides (as hematite/goethite only).
Grouping was done, partly, for the purposes of simplicity
to keep the number of minerals identified to a manageable
size and by taking into account the similarities in reflectance
spectra for some of the minerals identified. For example,
the mineral class ‘Mg-clays’ includes high-Mg smectites
such as saponite, and the fibrous clay mineral palygorskite,
Si8Mg5O20(OH)2(OH2)4.4H2O, identified by XRD as a major
phase in some grab-samples.
Accessory phases including strontianite SrCO3, artinite
Mg2(CO3)(OH)2.3H2O, jarosite KFe3(SO4)2(OH)6, pyroaurite
Mg6Fe23+(CO3)(OH)16.4H2O, the TiO2 dimorphs anatase and
rutile, and Cr-bearing spinels (eg magnesiochromite) were
also identified by XRD analysis in some samples, but were
not expected to significantly influence the overall spectral
response.
X-ray diffraction analysis
Random powder, XRD analysis was conducted using an
automated Philips X’Pert MPD diffractometer. From ≈458 m
of HQ diamond core drilled in the same area as the RC chip
composites, 60 samples were selected and examined to
validate the TSA mineralogy identified from the reflectance
spectra. XRD patterns were measured using CuK a radiation
with a post-receiving slit, curved graphite m o n o -
chromator, and scanning from 5 - 65° 2i in 0.03° 2i steps,
counting for two seconds/step. Mineral identification was
facilitated using the CSIRO XPLOT® search-match software
(Raven, 1990).
Partial least squares regression (PLSR) analysis
Combined geochemical and full VNIR-SWIR, reflectance data
for each 2 m composite sample were used for PLSR analysis
using the Unscrambler™ version 6.11 software, available
from CAMO. The Unscrambler™ program is a chemometric
software package used for quantitative prediction of one
 NI LATERITE MINERALOGY AND CHEMISTRY – A NEW APPROACH TO QUANTIFICATION

FIG 1 - Location of the Siberia North lateritic nickel deposit, which is hosted within the Walter Williams Formation, situated ≈80 km NE of Kalgoorlie and forms part
of the Kalgoorlie Nickel Project (KNP, Heron Resources). Other Ni laterite deposits that make up the KNP are also shown (eg Ghost Rocks, Highway, Goongarrie Hill,
Goongarrie South, Scotia Dam). The location of the Cawse lateritic Ni deposit is also shown.

or more variables from sample reflectance spectra. PLSR
analysis is a multivariate statistical technique, which aims to
find a number of components (also known as factors or latent
variables) that explain the most variation in both predictors
and responses. The method is described in detail by Martens
and Naes (1989).
The method of ‘leave-one-out’ cross-validation was used
in this study with ≈2/3 of the data-set used to develop the
calibration model and the remaining 1/3 used as the validation
set. For example, in the calibration data set, each sample was
‘left out’ and a predictive model developed using the remaining
samples from which an estimate for the excluded sample
was calculated and compared against the ‘known’ value. The
process was repeated until every object (sample) had been
left out once and the corresponding residual variance was
then calculated. The number of components that achieved
the smallest root-mean-square error of cross-validation
between the measured and predicted values was chosen for
the calibration model. The calibration model was then tested
using the validation data set and the same ‘leave-one-out’,
cross-validation method applied.
The diagnostic root-mean square error of calibration
(RMSEC) and root-mean square error of prediction (RMSEP)
were calculated to evaluate the accuracy of the PLSR
predictions. RMSEC is calculated using the measured and
predicted values at the calibration stage, while RMSEP is
calculated from the measured and predicted values for the
validation set. Ideally, the RMSEC and RMSEP values should
be as low as possible and converge to a similar magnitude.
Log space difference (ie Hull background removed) spectra
were used for PLSR analysis. Initial testing of other smoothing
(eg Savitsky-Golay) or scattering correction (eg multiplicative)
methods did not significantly improve PLSR analysis over that
obtained using Hull background removed spectra.
RESULTS AND DISCUSSION
Geochemistry
Initial examination of the assay data was made by evaluating
histogram plots of the distribution of Ni, Mg and Fe. An
initial cut-off of six per cent MgO was applied to exclude
all samples with >6 wt per cent MgO. The six per cent MgO
cut-off represents a typical metallurgical limit to maintain
economic acid consumption, for example in the high-pressure
acid leach (HPAL) process, by minimising the occurrence of
Mg-bearing, acid-consuming phases, such as serpentine and

THE FIRST AUSIMM INTERNATIONAL GEOMETALLURGY CONFERENCE / BRISBANE, QLD, 5 - 7 SEPTEMBER 2011 189
M A WELLS AND J CHIA

Mg-carbonates (Dalvi, Bacon and Osborne, 2004; Marshall
and Buarzaiga, 2004).
The histogram of the global wt per cent Ni content
distribution shows a ‘spike’ in samples with very low Ni
contents in the range 0 - 0.04 per cent, which comprised
9.8 per cent of the total data set (Figure 2). This is considered
an artefact related to the detection limit (DL = 10 ppm or
0.001 wt per cent) for assay of Ni by XRF analysis, which for
many samples in this interval range was <<0.01 wt per cent
Ni. Also, a majority of these samples were associated with the
‘Opal’ spectral mineral class, for which these Ni contents are
too low for beneficiation by physical means (ie sieving). Hence,
a minimum Ni content of 0.08 wt per cent Ni was applied, so
that only samples with >0.08 wt per cent Ni were included.
Plots of the distribution for Ni, Fe and Mg show a definite
left-handed skewness to low contents, with a diminishing
‘tail’ towards higher Ni, FeO and MgO values even after
considering the <6 wt per cent MgO limit for plots of wt
per cent Ni and wt per cent FeO (Figure 2). The skewed
distribution can significantly affect the ability of PLSR analysis
(A)
to reliably model samples with, for example, medium–high
(0.6–1.0 wt per cent) Ni contents, as any calibration model
will be weighted too heavily towards low Ni contents. Ideally,
the data should be uniformly or normally distributed.
A number of response transformations, including a square
root and a log transform, were tested to reduce the skewness
in the assay data (Figure 3). For histogram plots of the wt
per cent Ni distribution at Siberia North, with the Ni and Mg
limits applied, the square-root transform significantly reduced
the skewness in the data (Figure 3). A further improvement
was observed by applying a log transform, with spread in the
Ni contents more closely approaching a normal distribution
(Figure 3). Hence, a log transform was applied to both the Ni
and Fe assay data, while a square-root transform was applied
to the Mg data prior to initiating PLSR analysis.
PLSR analysis
From an initial combined reflectance spectroscopic-chemical
data-set of 1668 samples, after applying the cut-off limits
for low Ni and high Mg contents, and masking out of likely
outlier samples, a total of 436 samples remained. Of these,
290 formed the calibration set, and the remaining 146 samples
were used as the validation set.

 
 (A)








(B)




 (B)

 

 




(C)


(C)

FIG 2 - Global frequency distribution plots (n = 1668) for the Siberia North
deposit for: (top) wt per cent Ni content (interval = 0.04); (middle) Fe as wt per
cent FeO (interval = 1); and (bottom) wt per cent Mg as MgO (interval = 0.6).
The dotted line in each plot represents the cumulative distribution in each case.
Only the MgO cut-off has been applied to exclude those samples containing
>6 wt per cent MgO (black or solid shading). Solid shading in some intervals for
samples with <6 wt per cent MgO are samples that have been automatically
masked out (ie NULL values).
FIG 3 - Plots of the Ni distribution for the Siberia North deposit for samples
with <6 wt per cent MgO and >0.08 wt per cent Ni for: raw assay data (top);
with application of a square-root transform (middle); and a log transform
(bottom) used to reduce the skewness in the raw data. The averaged
distribution curve is modeled and overlayed in each case (ie bell-shaped curve).
Raw data are skewed to predominantly low Ni contents with median Ni content
of ≈0.4 - 0.5 wt per cent Ni, whereas the log transform results in Ni values
better approximating a normal distribution.

190 THE FIRST AUSIMM INTERNATIONAL GEOMETALLURGY CONFERENCE / BRISBANE, QLD, 5 - 7 SEPTEMBER 2011
 NI LATERITE MINERALOGY AND CHEMISTRY – A NEW APPROACH TO QUANTIFICATION
PLSR models that included cross-validated (20 leave-one-
out segments) predictions were fitted to the Ni, Fe and Mg
data sets separately. Table 1 gives a summary of the results
achieved by PLSR analyses for the three assays. The cross-
validation analysis of the calibration set showed that the ability
of PLSR to predict Ni, Fe and Mg was good as evidenced by the
relatively low RMSEC values in each case (Table 1).
TABLE 1
Number of components and resulting RMSEC and RMSEP values used to fit the
PLSR model to the Siberia North deposit.
Element/Oxide No of RMSEC RMSEP
components
Ni 6 0.16 0.17
FeO 10 0.071 0.086
MgO 12 0.161 0.178
Plots for the cross-validated prediction for the data where
the measured value of Y (wt per cent Ni, FeO and MgO) is
plotted against its respective predicted value are shown in
Figure 4a - c. The prediction plot for the Ni calibration set
shows a general linear trend, indicating general agreement
between measured wt per cent Ni contents and predicted wt
per cent Ni values, except for those at the top right hand corner
where a ‘roll-over’ or plateau effect is observed (Figure 4a).
These are samples with high Ni contents >1 wt per cent
of which few were present in the data set; this is discussed
in more detail later. The calibration plot for Fe shows a
strong linear trend indicating very good agreement between
measured and predicted values. A generally good agreement
between the measured and predicted MgO contents in the
calibration set is shown (Figure 4c). Closer inspection of
the cross-validation (Figure 4c) and, to a lesser extent, the
prediction (Figure 4f) plots for MgO shows the presence
of two groups or ‘data clouds’ at a break of approximately
1.6 wt per cent MgO. This is explained in relation to a change
in the dominant Mg-bearing mineralogy, as is discussed later.
Results for the validation set were generally consistent with
results obtained for the cross-validation analysis of Ni, FeO
and MgO in the calibration (see Table 1). Figure 4d - f shows
plots of the validation models for Y (ie wt per cent Ni, FeO and
MgO) plotted against its respective predicted value in each
case. The plot for Ni shows a good linear trend, which indicates
general agreement between the measured and predicted
values, except for a few samples. Predictions of the Ni content
for high-Ni-bearing samples tend to be underestimated,
which is most likely a result of the small number of samples
with high Ni contents (ie >≈1 wt per cent Ni). The ‘roll-over’ or
plateau effect observed in the top-right-hand corner for both
the calibration (Figure 4A) and validation plots (Figure 4d)
may be due to the very similar Ni contents but quite different
reflectance spectra of these samples. Hence, for samples with
similar Ni assays at ≈1 wt per cent, the Ni-bearing mineralogy
may be quite varied such that these samples are not effectively
modelled by PLSR analysis. The plot of the validation model
for estimating the wt per cent FeO content based on the
calibration model (ie Figure 4b) shows a very good linear
trend over the range of FeO contents modelled in this study
(range of about 1 - 20 wt per cent FeO) (Figure 4e).
Influence of mineralogy
The occurrence of Fe at the Siberia North deposit is governed
primarily by the presence of iron oxides, predominantly
THE FIRST AUSIMM INTERNATIONAL GEOMETALLURGY CONFERENCE / BRISBANE, QLD, 5 - 7 SEPTEMBER 2011
goethite, and to a much smaller extent, by the occurrence of
the Fe3+-bearing smectite, nontronite. Goethite and nontronite
were identified spectrally and by XRD analysis as the main
Fe-bearing minerals at Siberia North. Since Fe chemistry is
essentially controlled by a single phase, this would account
for the very strong predictive trend shown in both the PSLR
calibration (Figure 4b) and validation models (Figure 4e).
The greater scatter in the calibration and validation plots for
Ni, relative to Fe, reflects the comparatively varied mineral
association of Ni in this deposit, being associated with
goethite, smectite and Mn oxides.
The validation plot for MgO shows a general linear
trend but with some scatter in the data (Figure 4f), with an
apparent separation (ie change in slope) into two groups
for MgO contents of ≈ 1.6 wt per cent. The distribution of
Mg is controlled by a range of Mg-bearing minerals, some
of variable composition, including serpentine, talc, chlorite,
smectite and carbonates (eg dolomite and magnesite). A
histogram plot of raw MgO assays (at <6 wt per cent) for the
calibration data set, coded by mineral class, shows a distinctly
bi-modal distribution with two groups broadly defined as
containing <1.6 wt per cent MgO and >1.6 wt per cent MgO
(Figure 5a). An examination of the mineralogy identified
spectrally revealed that samples with <1.6 wt per cent
MgO, which make up nearly 60 per cent of the total sample
population (raw data for MgO contents <6 wt per cent), are
dominated mainly by the mineral classes ‘Kaolin’, ‘Opal’ with
some ‘Smectite’. In contrast, the group with >1.6 wt per cent
MgO predominantly comprise the mineral classes ‘Smectite’
and ‘Opal’ with minor ‘Mg-clays’ (Figure 5a). As the Mg-
bearing mineralogy for samples with <1.6 wt per cent MgO is
due predominantly to the presence of smectite, both the cross-
validation prediction and validation curves for this group
show a good linear response (eg Figure 4c and f). In contrast,
for samples with >1.6 wt per cent MgO, the Mg-bearing
mineralogy is dominated by both smectite and Mg-clays
(Figure 5a). These have quite different reflectance spectra,
which may contribute to the grouping or separation of these
groups observed in plots of the cross-validation prediction
and validation curves.
The influence of mineralogy on the distribution of
magnesium at Siberia North is shown for greater MgO
contents, with four distinct groups evident across the range
of measured MgO values (Figure 5b). Samples with MgO
contents between ≈7 to 20 wt per cent MgO, make up about
22 per cent of the sample population (raw data) and
comprise the mineral classes ‘Mg-clay’, ‘Carbonates’ and, to
a lesser extent, ‘Talc’ (Figure 5b). Samples with MgO contents
>20 wt per cent to a maximum of 41 wt per cent MgO, which
constitute about seven per cent of the total raw sample
population, comprise mainly the ‘Serpentine’, ‘Carbonate’ and
‘Residual silicate’ mineral classes (Figure 5b).
CONCLUSIONS
Predictive reflectance spectroscopic models were developed,
using PLSR analysis of a combined mineral reflectance
spectroscopic and assay data-set for the (siliceous) oxide–
style, Siberia North Ni laterite deposit. This approach was
tested to assess the efficacy of PLSR analysis as a means of
characterising Ni-laterite ore composition. PLSR analysis
of composite RC samples was restricted to modelling Ni-
laterite composition for Mg contents <6 wt per cent MgO (and
>0.08 wt per cent Ni) to exclude the influence of high-acid-
consuming phases, such as serpentine and Mg-carbonates.
Generally, very good linear PLSR models were developed
for predicting the Fe and, to a lesser extent, Ni composition.
191
M A WELLS AND J CHIA

FIG 4 - Plots of cross-validated, calibration plots for; (A) Ni, (B), Fe and (C) Mg, and validation plots of measured versus predicted values of; (D) Ni, (E) Fe and (F) Mg, for
the Siberia North deposit. A log transform was applied to both the Ni and Fe data sets while a square-root transform was applied to the Mg data prior to PLSR analysis.
Negative values in the cross-validation (A) and validation (D) plots for Ni result from the log transform applied to Ni contents <1 wt per cent. The cross-validation
prediction (C) and validation (F) plots for MgO, show the presence of two data clouds or populations with a natural ‘break’ (ie change in slope) at ≈1.6 wt per cent MgO.
This suggests a bimodal distribution for Mg related to differences in mineralogy (refer to text and Figure 5 for further explanation). The solid, black line in each plot
represents a line of unit slope.
This success was attributed to the small number of phases some of which are compositionally complex (eg serpentine,
controlling the distribution of Fe and Ni at the Siberia North talc, chlorite, smectite and Mg-carbonates). In addition,
deposit. For example, Fe distribution is controlled by only the reliability of PLSR analysis is influenced significantly
a few Fe-bearing minerals, predominantly goethite, and by sample bias in population (data) distribution. Nickel
to a lesser extent by the Fe3+-bearing smectite, nontronite. assays for the RC composites modelled were skewed to
Similarly, Ni mineralisation at the oxide-type, Siberia North relatively low (0.5 - 0.6 wt per cent) Ni contents with few high
deposit is associated mainly with goethite, smectite and, to (>1 wt per cent) Ni bearing samples being included. Modelling
a lesser extent, Mn oxides. In contrast, predictive models for of a larger population of more closely spaced drill holes, with
characterising Mg content were less reliable. This reflected a range in element abundance that more closely approaches a
the varied mineralogical association of Mg, the distribution normal distribution (eg Ni and Mg) would greatly improve the
of which is controlled by a range of Mg-bearing minerals, robustness and utility of the PLSR approach.

192 THE FIRST AUSIMM INTERNATIONAL GEOMETALLURGY CONFERENCE / BRISBANE, QLD, 5 - 7 SEPTEMBER 2011
 NI LATERITE MINERALOGY AND CHEMISTRY – A NEW APPROACH TO QUANTIFICATION

FIG 5 - Histograms of raw (ie untransformed) MgO assay values versus mineral class for: (A) MgO contents <6 wt per cent (bin interval of 0.2), and (B) MgO contents
<42 wt per cent (ie all data, bin interval of 0.8), at the Siberia North deposit. The Ni cut-off of 0.08 per cent has been applied to MgO contents <6 wt per cent and these
samples have been masked out (ie Null or grey records). At least three, possibly four, groups or populations are evident, each with a predominant
mineralogy when all samples are considered (B). Increasing MgO content reflects the increasing occurrence of Mg-bearing minerals identified in the mineral
classes ‘Mg-clays’, ‘Mg-carbonates’, ‘Serpentine’ and ‘Talc’. The frequency of each mineral-class occurrence has been ordered to highlight the change in mineralogy
with increasing MgO content.

The results of this study highlight the need for the
development of specific PLSR models tailored to match the
inherent mineral and compositional variability of the lateritic
Ni deposit being examined, even if Ni deposits may be classed
as the same type. Indeed, the mineralogical variation within a
deposit may require development of two (or more) calibration
models, particularly for samples with similar geochemical
signatures but which are mineralogically and, therefore,
spectrally distinct.
Reliable and near-quantitative mineralogical and ore-
type characterisation of Ni-laterite deposits may require an
integrated approach using a range of methods, such as surface
geological mapping, topography and ground penetrating
radar (GPR), and careful integration of their data to more
reliably model resource estimates of Ni-laterite deposits
(Horton, 2008) and provide better prediction of their
geometallurgical characteristics. While detailed mineralogical
and compositional characterisation of Ni laterites can
be achieved at the micron scale using techniques such as
QEMScan analysis (Anderson et al, 2009), characterisation
of bulk trends in mineralogy using reflectance spectroscopy,
such as that provided by the HyChips automated scanner,
and elemental modelling through approaches such as PLSR
analysis, provide a comparatively low-cost tool to better
characterise these mineralogically complex deposits.
ACKNOWLEDGEMENTS
The authors thank Andrew Hacket for scanning and collecting
reflectance spectra using the HyChips™ scanning system.
Another of our colleagues, Erick Ramanaidou, is also thanked
for his helpful comments of an initial draft of the paper. The
authors also acknowledge Travis Naughton, of the Visual
Resources Unit (VRU, CESRE) for drafting the figures. Finally,
the authors are grateful to James Ridley (Senior Geologist,
Heron Resources) and to Heron Resources for supporting this
work and for granting permission to publish.
REFERENCES
Anderson, J C O, Rollinson, G K, Snook, B, Herrington, R and
Fairhurst, R J, 2009. Use of QEMSCAN® for the characterization
of Ni-rich and Ni-poor goethite in laterite ores, Minerals
Engineering, 22:1119-1129.
Basile, A, Hughes, J, McFarlane, A J and Bhargava, S, 2010.
Development of a model for serpentine quantification in nickel
laterite minerals by infrared spectroscopy, Minerals Engineering,
23:407-412.
Berman, M, Bischof, L and Hungtington, J, 1999. Algorithms and
software for the automated identification of minerals using field
spectra or hyperspectral imagery, in Proceedings of the Thirteenth
International Conference on Applied Geologic Remote Sensing,
1999 Vancouver, BC, Canada, Ann Arbor, Environmental
Research Institute of Michigan, pp 222-232.
Brand, N W, Butt, C R M and Elias, M, 1998. Nickel laterites:
classification and features, AGSO Journal of Australian Geology
and Geophysics, 17:81-88.
Brand, N W, Butt, C R M and Hellsten, K J, 1996. Structural and
litholgocial controls in the formation of the Cawse Nickel laterite
deposits, WA – Implications for supergene ore formation and
exploratin in deeply weathered terrains, in Proceedings Nickel
‘96, Kalgoorlie.
Crowley, J K, Williams, D E, Hammarstrom, J M, Piatak, N, Chou,
I-M and Mars, J C, 2003. Spectral reflectance properties
(0.4-2.5 n m) of secondary Fe-oxide, Fe-hydroxide, and Fe-
sulphate- hydrate minerals associated with sulphide-bearing
mine wastes, Geochemistry: Exploration, Environment,
Analysis, 3:219-228.
Dalvi, A D, Bacon, W G and Osborne, R C, 2004. The past and the
future of nickel laterites, in Proceedings PDAC 2004 International
Convention, Trade Show and Investors Exchange, March 7-10,
pp 1-27.
Elias, M, 2006. Lateritic nickel mineralization of the Yilgarn Craton,
in Nickel deposits of the Yilgarn Craton: Geology, Geochemsitry,
and Geophysics Applied to Exploration (ed: S J Barnes) (Society
of Economic Geologists: Boulder).
Freyssinet, P, Butt, C R M, Morris, R C and Piantone, P, 2005.
Ore-forming processes related to lateritic weathering, Economic
Geology, 100th Anniversary, pp 681-722.

THE FIRST AUSIMM INTERNATIONAL GEOMETALLURGY CONFERENCE / BRISBANE, QLD, 5 - 7 SEPTEMBER 2011 193
M A WELLS AND J CHIA
Gilman, P, 2001. Prediction of nickel laterite chemistry using near
infrared spectroscopy, Honours, WA School of Mines Curtin
University of Technology.
Goetz, F H, Chabrillat, S and Lu, Z, 2001. Field reflectance
spectrometry for detection of swelling clays at construction sites,
Field Analytical Chemistry and Technology 5, pp 143-155.
Gomez, C, Lagacherie, P and Guillaume, C, 2008. Continuum
removal versus PLSR method for clay and calcium carbonate
content estimation from laboratory and airborne hyperspectral
measurements, Geoderma, 148:141-148.
Hill, R E T, Barnes, S, Gole, M J and Dowling, S E, 1995. The
volcanology of komatiites as deduced from field relationships in
the Norseman-Wiluna greenstone belt, WA, Lithos, 34:159-188.
Horton, J, 2008. Integration of disparate data types for resource
estimation – A nickel laterite example, in Proceedings PACRIM
Congress 2008, pp 189-194 (The Australasian Institute of Mining
and Metallurgy: Melbourne).
Huntington, J, 2005. Hylogging analysis of two drill holes from
the Goongarrie Ni laterite deposit Kalgoorlie,WA, CSIRO: North
Ryde Sydney.
Malengreau, N, Muller, J-P and Calas, G, 1994. Fe-speciation in
kaolins: A diffuse reflectance study, Clays and Clay Minerals,
42:137-147.
Malengreau, N, Weidler, P W and Gehring, A U, 1997. Iron oxides
in laterites: A combined mineralogical, magnetic, and diffuse
reflectance study, Schweiz Mineral Petrogr Mitt, 77:13-20.
Maquet, M, Cervelle, B D and Gouet, G, 1981. Signatures of Ni2+
and Fe3+ in the optical spectra of limonitic ore from New
Caledonia: Application to the determination of the nickel content,
Mineralium Deposita, 16:357-373.
Marshall, D and Buarzaiga, M, 2004. Effect of magnesium content
on sulphuric acid consumption during high pressure acid leaching
of lateritic ores, in International Laterite Nickel Symposium –
2004 (eds: W P Imrie, M D Lane, S C C Barnett, R M Berezowsky,
E J M Jahnsen, P G Mason, J A Murray, R C Osborne, J G Reid,
J G Schofield, L E Seeley, I G Skipper & V M Torres) (The Minerals
Metals and Materials Society).
Martens, H and Naes, T, 1989. Multivariate Calibration (Wiley:
Chichester).
194 THE FIRST AUSIMM INTERNATIONAL GEOMETALLURGY CONFERENCE / BRISBANE, QLD, 5 - 7 SEPTEMBER 2011
McConachy, T F, Yang, K, Boni, M and Evans, N J, 2007. Spectral
reflectance: Preliminary data on a new technique with potential
for non-sulphide base metal exploration, Geochemistry:
Exploration, Environment, Analysis, 7:139-151.
Ramanaidou, E R, Connor, P, Wells, M, Yang, K, Whitbourn, L
and Cudahy, T, 2004. Multiscale mine imaging for enhanced ore
grade definition in mining operations, in Proceedings for the 12th
Australasian Remote Sensing and Photogrammetry Conference,
Fremantle.
Ramanaidou, E R, Wells, M, Belton, D, Verrall, M and Ryan,
C, 2008. Mineralogical and microchemical methods for the
characterization of high-grade banded iron formation-derived
iron ore, Society of Economic Geologists, 15:129-156.
Ramanaidou, E R, Wells, M A and Connor, P, 2005. Automated
core logger and mine face scanning for exploration and mining of
iron ores, in Proceedings ABM VI Brazilian Symposium on Iron
Ore, Florianopolis Brazil.
Raven, M D, 1990. XPLOT Version 3.2-user manual Manipulation of
X-ray powder diffraction data (CSIRO: Adelaide).
Tamish, M, Ganz, H and Bowitz, J, 1987. Qualitative and quantitative
determination of mineral consituents of sediments by infrared
spectroscopy, Geologie und Palaeontologie, 75:699-710.
Wells, M A and Butt, C R M, 2000. Geology, geochemistry and
mineralogy of the Murrin Murrin nickel laterite deposit (CSIRO:
Perth).
Wells, M A and Ramanaidou, E R, 2004. Spectral quantification of
clay in Pilbara Channel Iron Deposits (CID), in Proceedings 12th
Australian Remote Sensing and Photogrammetry Conference,
Fremantle, Perth, pp 86-87.
Yang, K, Lian, C, Huntington, J F, Peng, Q and Wang, Q, 2005.
Infrared spectral reflectance characterization of the hydrothermal
alteration at the Tuwu Cu-Au deposit, Xinjiang, China,
Mineralium Deposita, 40:324-336.
Yang, K, Whitbourn, L and Mason, P, 2004. Spectrally scanning
mine bench faces and predicting compositions of laterites at the
Koniambo Massif, New Caledonia: Results of the Phase-5 study
(CSIRO: Sydney).