ADVANCES IN HYPERSPECTRAL PROCESSING FOR PROVINCE- AND CONTINENTAL- WIDE MINERAL MAPPING Hewson, R.D. 1*, Cudahy, T.J. 1, Caccetta, M. 1, Rodger, A. 1, Jones, M.2 and Ong, C.1 1
[email protected],
[email protected],
[email protected],
[email protected],
[email protected] CSIRO Division of Exploration & Mining, Australia 2
[email protected] Geological Survey of Queensland, Australia ABSTRACT
Comprehensive mapping of the mineral composition within North Queensland, Australia was recently undertaken using 25,000 km2 of airborne hyperspectral imagery. High spatial resolution, web-accessible, seamless, accurate maps of mineral abundances and physicochemistries, were delivered as part of a Queensland Geological Survey initiative. This required the development of pre-processing and mineral information extraction strategies that could be applied across a large number of individual flight lines within a diverse range of environments. This study demonstrated the successful use of spectral indices to target diagnostic reflectance absorption features associated with mineral abundances, composition and variations associated with crystalline or water bonding states. A multi-level series of masks, in a logical sequence, were applied to reduce possible ambiguities within image products. A new technique was also developed to compensate for variable vegetation cover, enabling the extraction of predominantly geological information (e.g. soil, outcrop, colluvium). Index Terms— Hyperspectral, mineral mapping, spectral indices, vegetation compensation 1. INTRODUCTION Comprehensive mapping of the mineral composition of the Earth’s surface, at province to continent scales is of value for a range of geoscience applications from mineral and petroleum exploration through to soil characterization. A vision proposed by CSIRO for Australia is the generation of a complete suite of web-accessible, seamless, accurate maps of the Australian continent for mineral abundances and physicochemistries, at high spatial resolution, made available to users through the government geoscience agencies. A recent two year project between the Commonwealth Scientific Industrial Research Organization (CSIRO), Geoscience Australia (GA) and the Queensland
978-1-4244-3395-7/09/$25.00 ©2009 Crown
Geological Survey (as part of the “Smart Exploration” and Smart Mining”) has been a significant milestone towards making this vision a reality. In particular, 25000 km2 of hyperspectral mineral and compositional map products, at 4.5 m spatial resolution, have been generated for North Queensland (Figure 1) and made available via the internet (http://www.em.csiro.au/NGMM/; http://www.dme.qld.gov.au/mines/hyperspectral.cfm). A detailed description of the processing strategies, quality control issues and generated mineral maps is included in Cudahy et al. [1]. Ultimately, such data sets are designed to provide explorers and soil scientists with a window for understanding sub-surface processes over a wide variety of geological and environmental landscapes. 2. PROCESSING STRATEGIES Airborne hyperspectral data was acquired over selected sites in North Queensland during the dry seasons of 2006-2007 by HyVista Corporation (http://www.hyvista.com/) using their HyMap® scanner and processed to apparent surface reflectance [2]. Over 200 flight lines of airborne data from 17 survey areas (Figure 1) were subsequently processed using CSIRO’s C-HyperMAP software (Figure 2). CHyperMAP is based on IDL™ and imported into ENVI™ (http://www.ittvis.com) and comprising a collection of linked software steps designed to rapidly generate accurate seamless mineral maps from large volume (Gigabytes to Terabytes), of multi-flight line, hyperspectral surveys (Figure 2). It is based on a programmable feature extraction-processing pipeline flow where scripts compute diagnostic spectral parameters (absorption depths, wavelengths, widths and asymmetries). In particular, a set of routines implementing spectral pre-processing (mean normalization, continuum removal) and feature parameter extraction (depth, area, width, wavelength at minima, ratios, arithmetic and logical operators) are applied at selected wavelengths, as listed in Figure 2 (nanometers, “nm”). The rationale of extracting such features and band parameters has
IV - 701
IGARSS 2009
been previously established by several researchers using band ratios / Relative Band Depths and the wavelength positions at a absorption feature’s minimum (Figure 3), [3], and [4] respectively. The index thresholds applied to the map products, ascribed to the mineral content or composition, were established by visually checking individual pixel spectral signatures within the resulting image anomalies for each index product. For a pixel to be identified as containing a specific mineral, a number of diagnostic spectral criteria had to be satisfied, including the value representing its zero abundance. Masking was also implemented to minimize the effects of overlapping mineral/vegetation absorption features, areas of water, and deep shade. The hyperspectral data was processed on a per-flight-line basis and not as a single large mosaic by C-HyperMAP. Mosaicing was completed at the final stages of mineral map production, where georeferencing and output to standard map GIS formats were generated. At present, C-HyperMAP is utilized as a research prototype, although designed to provide the Australian geosurveys with assistance in developing national standards for processing mineral mapping data. A new technique, was also developed to compensate for variable vegetation cover, enabling the extraction of predominantly geological information (e.g. soil, outcrop, and colluvium) [5]. This approach derived a Vegetation Corrected Continuum Depth (VCCD) for AlOH group minerals based on the results of synthetic modeling of vegetation, clay (kaolinite, halloysite and dickite) and quartz reflectance signatures, sourced from the United States Geological Survey (USGS) spectral library [6]. Constrained linear mixture models were assumed of variable clay and green and dry vegetation proportions, to derive coefficients for continuum-removed absorption depths, within a multiple linear regression model of the form, NVCCD = A1D0.67 + A2D2.08 + A3D2.20 where NVCCD is the a vegetation corrected absorption depth for AlOH/Clay minerals, and, A1, A2, and A3 are coefficients for Dλ, the continuum-removed absorption depth, D, at wavelength microns, λ. Synthetic models from spectral library examples provided estimates of A1-3, from values of D0.67, D2.08 and D2.20, calculated from spectra of green vegetation, dry vegetation (cellulose) and AlOH/clay minerals, respectively [5]. These coefficients were applied to HyMap derived values of D0.67 , D2.08 and D2.20 to generate AlOH/clay maps corrected for vegetation cover, of up to approximately 60%. Sensitivity studies using various different vegetation types indicated the technique was robust and likely to be suitable for different environments [5].
3. RESULTS AND CONCLUSIONS Processing hyperspectral imagery using an approach targeting absorption and spectral features, generated consistent scene independent mineral maps for ferric and ferrous iron, opaques, kaolinite, mica, illite, smectite, MgOH mineralogy, carbonate, hydrated silica, as well as variations due to compositional, crystalline and bound water states [1]. Over a period of two years, 335 field samples were also collected for ground-truthing the HyMap® map products, which established the reliability of such an approach [1]. Additional processing of the AlOH/clay map product using the VCCD algorithm, successfully compensated for vegetation cover (Figure 4) [5]. The resulting map products were no longer dominated by anomalously high AlOH/clay associated with exposed watercourses, roads, earthworks and topographic related vegetation patterns (Figure 4). The combination of applying this hyperspectral scene independent processing approach, targeting diagnostic spectral features, and the VCCD algorithm, demonstrates the feasibility of hyperspectral mineral mapping, at province- and continental- scales over diverse landscapes and environments. 4. REFERENCES [1] T.J. Cudahy, M. Jones, M. Thomas, C. Laukamp, M. Caccetta, R. Hewson, A. Rodger, and M. Verrall, “Next generation mineral mapping: Queensland Airborne HyMap and satellite ASTER Surveys 2006-2008,” CSIRO Exploration & Mining Report P2007 / 364. {http://www.em.csiro.au/NGMM/stage_1_report.html,} 2008. [2] T. Cocks, R. Jenssen, W. I. Stewart, and T. Shields, “The HyMap airborne hyperspectral sensor: The system, calibration, and performance,” in Proc. of 1st EARSEL Workshop on Imaging Spectroscopy, Zurich, 7pp., October 1998 [3] J.K. Crowley, D.W. Brickley, and L.C. Rowan, “Airborne imaging spectrometer data of the Ruby Mountains, Montana: mineral discrimination using relative absorption band-depth images.” Rem. Sens. Environ., v. 29, pp. 121-134, 1989. [4] E.F. Duke, “Near infrared spectra of muscovite. Tschermak substitution and metamorphic reaction process. Implications for remote sensing,” Geology, v. 22, pp. 201219, 1994. [5] A. Rodger, and T. Cudahy, “Vegetation Corrected Continuum Depth at 2.2 μm: An Approach for Hyperspectral Sensors,” Rem. Sens. Environ., In press. [6] R. Clark, G. Swayze, A. Gallagher, T. King, and W. Calvin, “The USGS Digital Spectral Library: Version 1: 0.2 to 3.0 μm. US Geological Survey”, Open File Rep, 93–592, 1993.
IV - 702
Figure 1 Location map of airborne HyMap and satellite ASTER coverage across north Queensland incorporated in study. The mapping results using ASTER data are further described in [1]. HyMap radiance@sensor (BIL) Atmospheric correction: HYCORR™
surface reflectance (BSQ) C-HyperMAP
Image-based processing tools:
Base image (BSQ) products:
• statistics • scripts for spectral parameterisation • etc
• Albedo@1650 • NDVI • 1900 nm absorption depth • 2200nm absorption depth • 2200nm absorption wavelength • 2160 nm absorption depth • etc
Mosaic-based processing tools:
Base mosaic (BSQ) products:
• georeferencing • image mosaicing • threshold setting and mask generation • statistics and levelling • etc
• Albedo@1650 • NDVI • 1900 nm absorption depth • 2200nm absorption depth • 2200nm absorption wavelength • 2160 nm absorption depth • etc
Mineral mosaic product tools: • multiple scripting/masking • Z-profiler validation • etc
Mineral mosaic (BSQ) products: • kaolin content • kaolin composition • white mica content • white mica composition • etc
Final GIS product tools:
GIS mineral mosaics (JPG, TIF):
• data range/stretch • colour bars • TIF, JPG converter • etc
• kaolin content • kaolin composition • white mica content • white mica composition • etc
Figure 2 Flow diagram showing hyperspectral image processing steps [1].
IV - 703
0.8 0.75
reflectance
0.7
A
0.65
B
0.6
C
0.55 0.5 0.45 2.1
2.15
2.2
2.25
2.3
2.35
2.4
wavelength (microns)
Figure 3 Example of AlOH’s shortwave infrared signature, analyzed for spectral parameters using fitted curves to estimate absorption depths (abundance) and composition (wavelength position) [1].
a)
b) Figure 4 a) AlOH/clay mineral abundance image generated from HyMap data prior to the application of VCCD processing; b) AlOH/clay image following VCCD processing showing reduced open pit, road and river anomalous areas relative to geological exposures, particularly along strike, [5].
IV - 704