Elsevier

Remote Sensing of Environment

Volume 129, 15 February 2013, Pages 17-31
Remote Sensing of Environment

Unmixing the effects of vegetation in airborne hyperspectral mineral maps over the Rocklea Dome iron-rich palaeochannel system (Western Australia)

https://doi.org/10.1016/j.rse.2012.10.011Get rights and content

Abstract

Quantitative iron (oxyhydr-)oxide (Fe-Ox), AlOH-clay and carbonate abundance maps of the Rocklea Dome in Western Australia have been derived from hyperspectral visible-near to shortwave infrared (VNIR–SWIR) airborne data that were compensated for the influence of vegetation cover. The quantitative mineral maps were validated against field data, including ~ 5500 VNIR–SWIR spectra and ~ 300 portable X-ray fluorescence measurements. The error on these airborne mineral abundance estimates averages 13.4 wt.% Fe for the Fe-Ox abundance, 4.0 wt.% Al2O3 for the AlOH-clay abundance and 0.025 for the 2320D parameter used to quantify the carbonate abundance. The unmixed quantitative mineral abundance maps improve geological mapping, including characterisation and exploration for channel iron deposits (CID). In particular, some areas with outcropping CID, which appear subeconomic from the airborne Fe-Ox abundance map without vegetation removal, show as potentially economic CID resources when the influence of vegetation cover is unmixed from the airborne hyperspectral data. These results show that seamless maps of mineral contents can be achieved using data collected from both proximal (drill core and field) and remote (airborne and satellite) hyperspectral sensing systems.

Highlights

► Generic approach to compensate quantitative mineral maps for vegetation cover ► Validation of unmixed airborne mineral abundance maps against ground data ► Application of quantitative mineral abundance maps to geological mapping ► Application of quantitative mineral abundance maps to channel iron ore exploration

Introduction

Airborne or satellite based spectroscopic sensing at visible to infrared wavelengths has proven a valuable tool, not only for mapping surface soils and geology, but also for recognising alteration haloes around mineralised systems (Cudahy and Barry, 2002, Hewson et al., 2005, Kruse et al., 1990, van der Meer et al., 2012, van Ruitenbeek et al., 2006 and references therein). The maps produced in these studies provided detailed information on relative mineral abundances and compositions.

Various workers have shown that quantitative mineralogy (and geochemistry) is achievable from proximal reflectance data collected at visible-near to shortwave infrared (VNIR–SWIR) wavelengths (Clark and Roush, 1984, Fredericks et al., 1985, Haest et al., 2011, Haest et al., 2012a, Zhang et al., 2001; VNIR–SWIR reflectance spectroscopic data will from here on be referred to as infrared spectroscopic data (VIRS data)). It should be possible to produce quantitative mineral maps from a remote sensing platform, if the effect of vegetation cover is compensated.

Several authors have successfully quantified the fractions of bare soil and green vegetation and dry vegetation on a per pixel basis in remotely sensed hyperspectral images, by modelling the remotely measured spectra as linear combinations of end member spectra of bare soil and green vegetation and dry vegetation (Asner and Heidebrecht, 2002, Roberts et al., 1998). Rodger and Cudahy (2009) demonstrated that it is not only possible to calculate the fraction of bare soil in the pixel of interest, but also its components, like for example the AlOH-clay abundance. Once the AlOH-clay abundance in the bare soil is known, its quantity can be adjusted to more accurately measure the AlOH-clay abundance of a given pixel through compensation of the diluting (mixing) effects of green vegetation and dry vegetation. Rodger and Cudahy (2009) first estimated the contributions of clay, green vegetation and dry vegetation at the pixel level using continuum-removed absorption depths, namely: (1) combination of Al-OH bending and OH stretching vibrations at ~ 2200 nm to estimate the AlOH-clay abundance; (2) chlorophyll absorption at 670 nm related to green vegetation; and (3) cellulose–lignin absorption at ~ 2100 nm associated with dry vegetation. From these estimates and associated modelled mixing relationships, vegetation-free AlOH-clay abundances were retrieved for pixels with up to 60% vegetation cover. The resulting vegetation unmixed AlOH-clay abundance map was validated: (1) by comparing ground and airborne AlOH-clay abundance measurements (n = 30) and (2) by visually confirming that areas with and without vegetation did not correspond to breaks in mineral abundances, except if they corresponded to geological contacts.

The procedure presented by Rodger and Cudahy (2009) differs significantly from the approach taken in previous studies that unmixed the effect of vegetation. Rodger and Cudahy (2009) did not use spectral end members to model the fractions of soil, green vegetation and dry vegetation on a per pixel basis (e.g. Asner and Heidebrecht, 2002, Roberts et al., 1998). They rather quantified these fractions using the continuum-removed depths of the diagnostic absorption features. However, they quantified only one mineral component of the soil, namely the AlOH-clay abundance, instead of calculating the more general ‘fraction of soil not covered by vegetation’.

The mineralogy exposed in a given pixel typically comprises more than just AlOH-clays, such as kaolinite, illite and montmorillonite. Also Fe-(oxyhydr-)oxides (Fe-Ox), carbonates, chlorites, amphiboles, etc., can commonly be observed at the surface. Thus accurate mapping of mineral contents would also improve geological mapping and mineral exploration, especially if quantitative measures are required to model specific geological, metamorphic and regolith units. For example, Anand and Butt (2010) developed models based on specific mineral percentages that characterise various lateritic horizons developed over different parent rocks. To use these types of quantitative models for geological mapping requires the extraction of mineral percentages from the spectral imagery, which can only be obtained when the obscuring effects of both green vegetation and dry vegetation are removed. Armed with both spatially comprehensive mineral percentages and appropriate quantitative mineral–geological models, geoscientists will be able to not just recognise the footprints to new mineral deposits, but also a 1st pass assessment of the size and quality of any ore resource exposed at the surface.

In this study, we build on the methodology of Rodger and Cudahy (2009), by first establishing a generic procedure for estimating the per-pixel contribution of both green vegetation and dry vegetation that is then used to unmix their contribution from a range of other spectrally-derived mineral products. This includes for example the contents of iron (oxyhydr-)oxides, carbonates, as well as clays, which are the major mineralogical components of soils and rocks exposed in the “target” palaeochannels, hosting channel iron ore deposits (CID), in the Rocklea Dome area in the Hamersley Basin of Western Australia. Airborne IRS data were collected from over this test site with the derived vegetation-unmixed mineral contents validated against ground traverses/samples of field spectra and portable X-Ray Fluorescence (PXRF) measurements.

Section snippets

Geological setting

The Rocklea Dome area comprises a wide range of geological and regolith units, variably covered by green vegetation and dry vegetation (mostly Spinifex grass and bushes). The bedrock geology comprises an exposed double, shallow plunging anticline with a hinge zone trending approximately WNW-ESE. The exposed core of the Rocklea Dome consists of an Archaean age monzogranite pluton and minor cross-cutting mafic and ultramafic intrusives that form part of the Pilbara Craton, underlying the

Methods

The methodology comprised several steps, namely: (1) reduction of the airborne AMS data to apparent reflectance through calibration and radiative transfer correction; (2) processing of the AMS apparent reflectance data to retrieve the vegetation unmixed mineral products; (3) ground data collection of both PXRF and IRS data; (4) processing of the ground IRS data to spectrally simulate the AMS data and then using the same mineral content extraction methods, i.e. continuum depths of diagnostic

Vegetation removal from remote mineral abundance estimates

The Fe-Ox abundance, AlOH-clay abundance and carbonate abundance estimates for all vegetation free ASD spectra collected along the transects are plotted as a function of the green and dry vegetation cover parameters (Fig. 3A–F). The NDVI and 2100D parameters for the ASD data average 0.047 and 0.0034, respectively, and these values are chosen to represent 0% green or dry vegetation cover.

Green vegetation spectra have been selected from an unpublished green vegetation spectral library compiled by

Applications

The vegetation unmixed mineral abundance images improve mapping of basement units (granite, chert and schist in AlOH-clay abundance image; Fig. 10) and the palaeochannel, if CID is outcropping at the surface (Fe-Ox abundance image; Fig. 11), which has applications for geological mapping and CID characterisation/exploration.

The AlOH-clay abundance map without vegetation unmixing (Fig. 10A) shows increased but variable AlOH-clay abundances. The granite is highlighted with pink polygons and

Summary and conclusion

Maps of Fe-(oxyhydr-)oxide, AlOH-clay and carbonate contents were generated by unmixing the effects of both green and dry vegetation cover at the pixel-level in airborne hyperspectral VNIR–SWIR imagery collected over the Rocklea Dome in Western Australia. The accuracy of these mineral content maps was validated using ~ 5500 field ASD VNIR–SWIR spectra and 300 field PXRF geochemical measurements. Estimates (0–100%) of the green and dry vegetation components at the pixel-level were gauged using

Acknowledgements

We would like to express our sincere gratitude to the Western Australian Department of Commerce and the CSIRO Minerals Down Under Flagship for financial support to this research project. Murchison Metals Limited is acknowledged for providing the raw AMS data. Constructive comments from an anonymous reviewer of RSE have improved this paper and are appreciated.

References (30)

  • G.P. Asner et al.

    Spectral unmixing of vegetation, soil and dry carbon cover in arid regions: comparing multispectral and hyperspectral observations

    International Journal of Remote Sensing

    (2002)
  • A. Berk et al.

    MODTRAN (TM) 5: 2006 update — art. no. 62331F

  • A. Berk et al.

    MODTRAN5: a reformulated atmospheric band model with auxiliary species and practical multiple scattering options

  • R.N. Clark et al.

    High spectral resolution reflectance spectroscopy of minerals

    Journal of Geophysical Research — Solid Earth and Planets

    (1990)
  • R.N. Clark et al.

    Reflectance spectroscopy — Quantitative-analysis techniques for remote-sensing applications

    Journal of Geophysical Research

    (1984)
  • Cited by (28)

    • Characterization of iron ore by visible and infrared reflectance and Raman spectroscopies

      2021, Iron Ore: Mineralogy, Processing and Environmental Sustainability
    • Proximal and remote spectroscopic characterisation of regolith in the Albany-Fraser Orogen (Western Australia)

      2016, Ore Geology Reviews
      Citation Excerpt :

      Green and dry vegetation have overlapping absorption features with major mineral groups and areas with more than ca. 60% green vegetation were masked out in the ASTER Geoscience Products used. Rodger and Cudahy (2009) and Haest et al. (2013) showed the potential for unmixing green vegetation from hyperspectral remote sensing data and a similar approach could also be applied to multispectral ASTER data. Fire scars were frequently observed in ASTER imagery in the northeastern Albany–Fraser Orogen and affected selected mineral maps, such as the AlOH Group Composition, where a fire scar is displayed to the northeast of NLD210 (Fig. 7a and b).

    • Evaluating simple proxy measures for estimating depth of the ~1900nm water absorption feature from hyperspectral data acquired under natural illumination

      2015, Remote Sensing of Environment
      Citation Excerpt :

      There have been many studies which have used hyperspectral imagery to map minerals using their unique spectral curves in the shortwave infrared between 2000–2500 nm (e.g., Ellis & Scott, 2004; Haest et al., 2013; Kruse, 1988; Lagacherie, Baret, Feret, Madeira Netto & Robbez-Masson, 2008; Murphy, Schneider & Monteiro, 2014; Viscarra Rossel & Chen, 2011).

    • Characterization of iron ore by visible and infrared reflectance and, Raman spectroscopies

      2015, Iron Ore: Mineralogy, Processing and Environmental Sustainability
    • Spectroscopic analysis of green, desiccated and dead tamarisk canopies

      2015, Photogrammetric Engineering and Remote Sensing
      Citation Excerpt :

      Stressed tree mapping had a kappa coefficient of around 0.70, indicating that asymptomatic trees are likely to have significantly higher reflectance in the red-NlR regions, senescent trees are likely to have significantly lower reflectance in the NIR regions, and dead trees are likely to have significantly higher reflectance in the SWIR regions. Our analysis applied wavelength selection techniques on canopy spectra collected in situ before classification by RF, resulting in an increased classification accuracy of tamarisk canopy types compared to using the full wavelength range (Asner et al., 1998; Somers et al., 2010; Somers and Asner, 2013). Using selected wavelengths (bands) that display high between-class variability and low within-class variability, not only was higher classification accuracy achieved, but the computational time was greatly reduced (Table 2 and Table 3).

    View all citing articles on Scopus
    View full text