UNIVERSITY OF COPENHAGEN DEPARTMENT OF GEOSCIENCES AND NATURAL RESOURCE MANAGEMENT
Potentials and challenges for hyperspectral mineral mapping in the Arctic Developing innovative strategies for data acquisition and integration
PhD Thesis − Sara Salehi Supervisor: Prof. Rasmus Fensholt Co-supervisor: Dr. Bjørn Henning Heincke Submitted on: 3 rd of July, 2018
GEOLOGICAL SURVEY OF DENMARK AND GREENLAND DEPARTMENT OF PETROLOGY AND ECONOMIC GEOLOGY
In memory of my father … To my mother With love and eternal appreciation
PREFACE This thesis has been submitted in compliance with the requirements for the Ph.D. degree at the Faculty of Science, University of Copenhagen, Denmark. The work described in this Ph.D. dissertation was undertaken at the Department of Geosciences and Natural Resource Management (IGN), University of Copenhagen and the Geological Survey of Denmark and Greenland (GEUS), Department of Petrology and Economic Geology. The work was carried out from January 2015 to July 2018 and included a total 6 months stay at three different research institutes and organizations: The Faculty of Geo-Information Science and Earth Observation (ITC) at the University of Twente in Enschede (The Netherlands); the German Remote Sensing Data Center (DFD), an institute of the German Aerospace Center (DLR) in Oberpfaffenhofen and the Helmholtz Institute for Resource Technology (HIF) in Freiberg (Germany). Prof. Rasmus Fensholt was principle supervisor from the Department for Geosciences and Natural Resource Management. From the Department of Petrology and Economic Geology (GEUS), late Tapani Tukiainen was principle supervisor from January 2015 to January 2016 and Dr. Bjørn Henning Heincke was co-supervisor throughout the whole project. The research was funded as part of the “Geological Remote Sensing” project by the Geocenter Denmark (no. 2-2014), which aims to stimulate and coordinate joint projects within geoscientific research domain in order to improve the quality of geological and geographical studies. While the aims of the project were broad ranging, I primarily focussed on the challenges for hyperspectral mineral mapping in the arctic environment of Greenland, to identify potential solutions and approaches to address them. The outcomes of these investigations are presented and summarized in five subsequent research papers. All papers except for Paper III (ready to be submitted) are already published in peer-review journals. Paper II is a conference paper. Two additional papers published in Geological Survey of Denmark and Greenland bulletin are added in the appendix section to help the reader get a better background from the work that has been carried out for this research. The introduction chapter starts with describing the motivation behind the research, providing some brief background to the main data and methods used and outlining the research objectives. This is followed by a summary of each research paper and a section including the main conclusions and outlooks. The published/submitted research papers are presented in the last section.
2
ACKNOWLEDGMENTS The completion of this thesis would not have been possible without the support and encouragement of several special people. I would like to take this opportunity to show my gratitude to those who have assisted me in a myriad of ways. First and foremost, I would like to thank my supervisors Prof. Rasmus Fensholt and Dr. Bjørn Henning Heincke, who have both dedicated a great deal of time and effort into guiding me through my Ph.D. Not only did they allow me the freedom to express myself academically, but also gave me all the support I could wish in terms of valued discussions and sound advice. I also had the fortune to work with late Tapani Tukiainen and to learn a great deal from him in the first year of my Ph.D. My thanks go to Leif Thorning for his excellent guidance and endless encouragement. His depth of knowledge and experience has helped me immensely throughout my work, for which I am very grateful. I would also like to express my deep appreciation to Prof. Benoit Rivard and Dr. Derek Rogge for insightful discussions, valuable advices and expert guidance aside from working with me on our paper. Thanks to fellow co-authors from Institute Freiberg for hosting me and for their contributions to data collection in the summer of 2016. I also had the privilege of working with Dr. Christian Rogaß in the last months of my PhD. His opinions inspired me and gave me a different perspective to the world of science. Some special words of gratitude go to my friends and colleagues, former and current, who have always been a major support when things became a bit discouraging. Mojgan, Amu Hoji, Katrine, Erika, Anouk, Anaïs, Mojtaba and Milad - I could never have done it without you. I would also like to thank Dr. Karen Hanghøj and Dr. Stefan Bernstein for the opportunities they have given me to sharpen my skills. Finally, I would like to thank my family to whom I owe a great deal. Great appreciation and enormous thanks to my mother, who has been a constant source of support and encouragement. I am especially grateful for her confidence in me and for the sacrifices she made, so that I am here today. To my late father, who encouraged me to be the best person I can be and to do the best I can in everything, I wish to achieve. He has given me all the freedom to pursue my career. I will always remember how patiently he would listen to my daily complaints, and how interested he was in learning the details about my field of study. This is for you Papa, as I promised.
3
SUMMARY Most of the studies using hyperspectral data for geological applications have addressed areas in arid to semi-arid climates. This Ph.D. thesis presents research examining how well geological mapping works under the arctic, high relief conditions of Greenland, using hyperspectral data acquired from different platforms and at various scales. Building upon the results derived from regional airborne hyperspectral data, it is demonstrated that one of the main sources for potential misclassification of pixels in the Arctic is the subpixel spectral mixing of lichens and their rock substrate. To address the gaps in current knowledge about the effect of lichens with respect to geological applications, I investigate here a) how lichens affect spectral recognition of common rock-forming minerals (paper I), b) how to estimate lichens abundance by developing generic indices that can be used regardless of the mineralogy of underlying rocks and the species of lichens (paper II) and c) how to tackle this obstacle by using suitable mapping approaches (paper III). The second major challenge that has been addressed is the lack of feasible approaches to capture the hyperspectral data as part of a large-scale operation (i.e., areas of hundreds or thousands of square kilometres) in a time- and cost-effective manner, in particular for areas of difficult access. Coastal cliffs are examples of major well-exposed outcrops in the Arctic that are mostly inaccessible and not observable by air/spaceborne nadir remote sensing sensors due to steep topography. The application of close range sub-horizontal hyperspectral imaging using terrestrial platforms and integration of spectral data with accurate terrain models has recently gained attention for mapping of steep outcrops. However, observing a geological target at close range (i.e. within a range of one to several hundred metres) is not always feasible. In other words, larger targets such as steep mountain slopes, sea- or lake-faced cliffs are often only fully visible from an opposing location such as a neighbouring mountain or shoreline. The distance between the sensor and the target of interest can then easily extend to several kilometres. For this reason, I proposed a new data acquisition strategy, namely longrange terrestrial outcrop sensing, which is tested for the first time during a field campaign I led in the summer of 2016 in South West Greenland in the region between the fjords Ikertoq and Søndre Strømfjord. Karrat region in West Greenland was selected as the second test site for demonstrating the applicability of this new data acquisition approach. The results of both surveys are discussed in more details in (Rosa et al. 2017; Salehi and Thaarup 2018). Despite the promising results achieved by using this approach, the rugged topography and difficult terrain accessibility in the Arctic often hinder the instrumentation setup and limit the employment of such a data acquisition strategy. To overcome these limitations, I have investigated the potential of 4
using a platform in motion (such as a boat/ship) to continuously acquire the hyperspectral data while sailing along the fjords (paper IV). In addition, the two-dimensional maps generated from hyperspectral imaging are transformed to three-dimensional hyperclouds and integrated with terrain models generated from oblique photogrammetry. The high spatial resolution of terrain models allows investigating e.g. faults and the general morphology of lithologies whereas spectral data provides information regarding the mineralogy and chemical composition of the rocks. My observations suggest that regardless of using terrestrial or moving platforms, performing the required preprocessing for data captured from distant targets is not straight-forward. Firstly, the logistical setup of “visible” reference targets for radiometric correction with the same orientation and distance as the distant target outcrop is not possible. Secondly, large distances between the sensor and the outcrop lead to major atmospheric distortions. Thirdly, owing to the large scale of the observed surface and the sensor viewing perspective, pixels within one scene can represent a range of different distances and orientations, leading to highly variable radiometric distortions. For those reasons, correction methods established for nadir acquisitions need to be adapted to account for the special conditions of long-range sub-horizontal sensing of outcrops (paper V).
5
RESUMÉ De fleste studier af brugen af hyperspektrale data til geologiske formål har været fokuseret på aride og semi-aride områder og klimaer. I denne phd thesis fremlægges forskning, der er gennemført for at undersøge, hvor velegnede hyperspektrale data indsamlet fra forskellige platforme og i forskellige skalaer er til geologiske kortlægningsopgaver i arktiske, høj-relief områder af Grønland. Baseret på resultater fra regionale flybårne hyperspektrale data påvises det (artikel I), at en vigtig grund til at pixels i arktiske områder undertiden klassificeres fejlagtigt er den spektrale sammenblanding af signaler fra lichener og de underliggende bjergarter. Det påvises, hvordan mængden af lichener kan estimeres ved at udvikle generelle indekser, der kan bruges uafhængigt af den underliggende mineralogi og hvilke lichen-arter, der er tale om (artikel II), samt hvordan anvendelsen af passende fremgangsmåder for kortlægningen kan løse problemer relateret til tilstedeværelsen af lichener (artikel III). En anden stor udfordring som adresseres, er vanskeligheden ved at opmåle og registrere de hyperspektrale data i forbindelse med eller som en del af store logistiske operationer dækkende store og varierende geografiske områder med en udstrækning på hundreder eller tusindvis af kvadratkilometre på en tids- og omkostningsmæssig effektiv måde, især når det drejer sig om vanskeligt tilgængelige områder. I de arktiske egne er kystvendte bjergskråninger eksempler på store, vel-blottede områder som for det meste ikke så let kan opmåles fra remote sensing instrumenter i satellitter eller fly lodret over stedet. Anvendelse af sub-horisontale hyperspektrale billeder (imaging) og integrering af de spektrale data med nøjagtige terræn-modeller har for nyligt tiltrukket sig opmærksomhed i forbindelse med kortlægning af næsten vertikale blotninger vha. hyperspektral skanning fra landbaserede platforme. Det er imidlertid ikke altid muligt at komme til at observere et sådant geologisk målområde tæt på, dvs. fra en afstand af et til to hundrede meter. Med andre ord, større målområder, som for eksempel stejle bjergsider og klinter vendt mod havet eller søer, er ofte kun fuldt synlige fra modvendte placeringer på bjergtoppe eller kystforlandet. I sådanne tilfælde kan afstanden mellem sensorer og målområdet let blive mange kilometer. Af denne årsag foreslog jeg anvendelse af en ny strategi for dataopsamling, nemlig jordbaseret opmåling på lang afstand (terrestial outcrop sensing) af større blotninger. Jeg ledte det feltarbejde i sommeren 2018 i Sydvestgrønland mellem fjordene Ikertoq og Kangerlussuaq, som for første gang afprøvede metoden; resultaterne fremlægges her. Til trods for de lovende resultater opnået ved denne landbaserede fremgangsmåde, lægger den alpine topografi og de svære adgangsforhold ofte hindringer i vejen for en registrering af de ønskede data. 6
For at overvinde disse begrænsninger, har jeg undersøgt muligheder for at bruge en platform i bevægelse (skib eller helikopter) til kontinuert registrering af de hyperspektrale data under sejlads i en fjord (artikel IV). Desuden kan de todimensionale kort, som dannes ud fra de hyperspektrale data, transformeres til tredimensionale ”hyperskyer” og integreres med tredimensionale terrænmodeller skabt fotogrammetrisk ud fra specielle skråfoto. Den store spatielle opløsning af terrænmodellerne kan udnyttes til at undersøge forkastninger, strukturer og generel morfologi, hvortil de spektrale data kan bibringe yderligere information om mineralogi og kemisk sammensætning af bjergarterne. Mine observationer påpeger, at uanset om der bruges en landbaseret eller en bevægelig platform, er det ikke en simpel sag at præ-behandle data opfanget fra fjerne mål. For det første er det rent logistisk ikke muligt at opsætte de nødvendige synlige referenceplader for den radiometriske korrektion parallel med og i samme afstand, som de langt fjernere geologiske mål befinder sig. For det andet, medfører de store afstande mellem sensor og blotning et behov for store atmosfæriske korrektioner. For det tredje, pga. den store udstrækning af den observerede overflade og sensorens perspektiv i landskabet, kan pixler inden for en scene repræsentere forskellige afstande og orienteringer, hvilket medfører stærkt variable radiometriske forvrængninger i data. Af disse årsager er det i høj grad nødvendigt at modificere korrektionsmetoder for lodrette opmålinger før disse metoder anvendes under de specielle forhold som vedrører subhorisontal hyperspektral fjern-skanning af overflader. (artikel V).
7
LIST OF PAPERS Paper I. Salehi, S., Rogge, D., Rivard, B., Heincke, B. H., & Fensholt, R. (2017). Modeling and assessment of wavelength displacements of characteristic absorption features of common rock forming minerals encrusted by lichens. Remote Sensing of Environment, 199, 78-92. doi: 10.1016/j.rse.2017.06.044
Paper II. Salehi, S., Karami, M., & Fensholt, R. (2016). Identification of a robust lichen index for the deconvolution of lichen and rock mixtures using pattern search algorithm (case study: Greenland). International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences, XLI-B7, 973-979. XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic. doi: 10.5194/isprs-archives-XLI-B7-973-2016
Paper III.
Salehi, S., Mielke, C., & Rogass, C. (2018). Mapping of arctic resources in
presence of abundant lichen using airborne imaging spectroscopy and spaceborne Sentinel-2 and Landsat-8 OLI data (Case Study: the Niaqornarssuit complex in South West Greenland). To be submitted in European journal of remote sensing.
Paper IV. Salehi, S., Lorenz, S., Vest Sørensen, E., Zimmermann, R., Fensholt, R., Henning Heincke, B., Kirsch, M., & Gloaguen, R. (2018). Integration of Vessel-Based Hyperspectral Scanning and 3D-Photogrammetry for Mobile Mapping of Steep Coastal Cliffs in the Arctic. Remote Sensing, 10(2), 175. doi: 10.3390/rs10020175
Paper V. Lorenz, S., Salehi, S., Kirsch, M., Zimmermann, R., Unger, G., Vest Sørensen, E., & Gloaguen, R. (2018). Radiometric correction and 3d integration of long-range ground-based hyperspectral imagery for mineral exploration of vertical outcrops. Remote Sensing, 10(2), 176. doi: 10.3390/rs10020176
8
LIST OF PAPERS PUBLISHED IN GEOLOGICAL SURVEY OF DENMARK AND GREENLAND BULLETIN Geological Survey of Denmark and Greenland Bulletin is an Open Access journal that publishes comprehensive research papers with focus on Greenland, Denmark, and the North Sea regions. At least two, typically non-Danish scientists externally referee the papers.
Paper I. Salehi, S., & Thaarup, S. (2018). Mineral mapping by hyperspectral remote sensing in West Greenland using airborne, ship-based and terrestrial platforms. Geological Survey of Denmark and Greenland Bulletin, 41 (in press).
Paper II.
Salehi, S. (2018). Hyperspectral analysis of lithologies in the Arctic in areas with
abundant lichen cover. Geological Survey of Denmark and Greenland Bulletin, 41(in press).
9
ABBREVIATIONS ASD
Analytical Spectral Devices FieldSpec®3 HiRes Spectroradiometer
ATCOR4
Airborne Atmospheric and Topographic Correction Model
BRDF
Bidirectional Reflectance Distribution Function
EnGeoMAP
EnMAP Geological Mapper
FOV
Field Of View
GEUS
Geological Survey of Denmark and Greenland
HSI
hyperspectral imaging
IFD
Iron Feature Depth
ISMA
Iterative Spectral Mixture Analysis
PARGE
Parametric Geocoding and Orthorectification
PS
Pattern Search algorithm
SMA
Spectral Mixture Analysis
SSIM
Structural Similarity Index Measure
SWIR
shortwave infrared
10
Table of contents 1 MOTIVATION, BACKGROUND AND OBJECTIVES .................................................... 12 1.1 INTRODUCTION ...................................................................................................................................................... 12 1.1.1 Hyperspectral remote sensing for geological mapping and exploration ............................................................. 12 1.1.2 Challenges towards analysis of hyperspectral data under arctic conditions ........................................................ 12 1.2 THESIS OBJECTIVES .............................................................................................................................................. 20
2 STUDY SITES ................................................................................................................. 21 2.1 KARRAT.................................................................................................................................................................... 22 2.2 SISIMIUT-KANGERLUSSUAQ REGION............................................................................................................... 23 2.3 LIVERPOOL LAND.................................................................................................................................................. 23
3 HYPERSPECTRAL DATASET ....................................................................................... 24 3.1 REGIONAL AIRBORNE HYPERSPECTRAL DATA (HyMAP)............................................................................ 24 3.2 FIELD MEASUREMENTS ....................................................................................................................................... 26 3.2.1 Long-range terrestrial hyperspectral scanning in the southern Nagssugtoqidian orogeny (South West Greenland) and Karrat region (Central West Greenland) ............................................................................................ 28 3.2.2 Integration of ship-based hyperspectral scanning and 3D-photogrammetry ....................................................... 29
4 DATA PROCESSING AND ANALYSIS OF THE SURFACE MINERALOGY ................ 31 4.1 ENDMEMBER EXTRACTION AND SPECTRAL UNMIXING ............................................................................ 31 4.2 ABSORPTION WAVELENGTH MAPPING............................................................................................................ 32
5 INTRODUCTION TO REASEARCH PAPERS ............................................................... 33 5.1 INTRODUCTION TO PAPER I ................................................................................................................................ 35 5.2 INTRODUCTION TO PAPER II ............................................................................................................................... 36 5.3 INTRODUCTION TO PAPER III.............................................................................................................................. 38 5.4 INTRODUCTION TO PAPER IV ............................................................................................................................. 39 5.5 INTRODUCTION TO PAPER V............................................................................................................................... 41
6 SYNTHESIS, CONCLUSIONS AND OUTLOOK ............................................................ 42 7 REFERENCES ................................................................................................................ 44 8 RESEARCH PAPERS ..................................................................................................... 51 9 GEUS BULLETIN RESEARCH PAPERS ....................................................................... 80 11
1 MOTIVATION, BACKGROUND AND OBJECTIVES 1.1 INTRODUCTION 1.1.1 Hyperspectral remote sensing for geological mapping and exploration
While multispectral images have been in regular use since the 1970s, the widespread use of hyperspectral images is a relatively recent trend. This technology comprises the remote measurement of specific chemical and physical properties of surface materials through imaging spectroscopy. Regional geological mapping and mineral exploration are among the main applications that may benefit from hyperspectral technology. Minerals and rocks exhibit diagnostic spectral features throughout the electromagnetic spectrum that allow their chemical composition and relative abundance to be mapped. Reflectance spectra in the 350-2500 nm wavelength region contains mineralogical information due to electronic absorption features in transitional metals, especially iron (Clark 1999; Hunt 1977) and vibrational absorption features in carbonate, hydrate and hydroxide minerals (Clark 1999; Hunt 1977). Some of these minerals such as iron oxides and hydroxides, clays and sulfates are often pathfinders of mineral deposits. Variations in chemical composition often cause shifts in the position and shape of absorption bands in the spectrum. Thus, tracing the exact absorption wavelength position is a key factor for mineral identification. The ability of hyperspectral sensors for measuring hundreds of contiguous spectral bands in comparison to multispectral image data provides sufficient spatial and spectral resolution for analysis of surface mineralogy and for describing diagnostic mineral absorption signatures unavailable from other sources. Airborne hyperspectral sensors (e.g., HyMap, AVIRIS, HySpex) and upcoming spaceborne satellites (e.g., EnMap) offer the opportunity to estimate the mineralogical composition of the surface under examination without having a direct contact with the targets of investigation. This capability is important, not only for mapping the presence of specific minerals over large areas, but also for discriminating among minerals that are not readily identified through inspection in the field. 1.1.2 Challenges towards analysis of hyperspectral data under arctic conditions
Considerable research has been devoted to the use of hyperspectral remote sensing technology for geological applications in arid and semi-arid environments (Harris et al. 2005; Sabine 1999); however, less research has been devoted to the efficacy of hyperspectral methods for mineral exploration purposes in the Arctic, and therefore, this is still an open field for research. Greenland is ideally suited for geological remote sensing given the relatively scarce vegetation cover and good 12
bedrock exposures. The use of hyperspectral remote sensing technology to extract geological information under complex inaccessible mountain terrains of this region certainly has a very low cost/benefit ratio, in comparison to on-ground geological fieldwork. However, spectral mixture of different surface materials within a pixel (such as multiple lithologies or lichen and vegetationcovered outcrops; see Fig. 1), rugged topography and steep inaccessible slopes, low sun angle and rapidly changing extreme weather conditions are among the challenges that must be tackled in order to build a successful workflow for analysis of remote sensing data in the arctic environment of Greenland.
Figure 1. Illustration of the problem of mixed pixels. When the spatial resolution is not fine enough, various surface materials lie in the same pixel. In this case, a hard classification process cannot give an accurate information about the pixel coverage.
Airborne hyperspectral sensors are characterized by high spectral resolution and a spatial resolution, which can differ from few to tens of meters depending on the angular distribution of radiation emitted from the surface, the angular response of the spectrometer, and the sensor altitude. The major problem caused by this relatively low spatial resolution is the mixture of signatures from various components occupying individual pixels. This can mask the diagnostic mineral features and potentially complicate the subsequent mineral identification and mapping. A pervasive example of such mixing of signatures in high latitude and subarctic environments is the presence of abundant lichen growing on rock outcrops. Lichens impede the identification of rocks/minerals by preventing transmission of light to the rock substrate and thereby masking the mineral features (Ager and Milton 13
1987; Bechtel et al. 2002). The spectrum of a lichen-covered rock can be significantly different from the bare rock spectrum. On the other hand, the lichens reflectance peaks near 2220 nm overlaps with the OH absorption band at 2200 nm in hydroxyl-bearing minerals. This may interfere with locating hydrothermal alterations and mapping these potentially mineral rich areas. The presence of lichens in mixed spectra can also induce spectral shifts in the wavelength position of the diagnostic mineral absorption features within the shortwave infrared (SWIR) range (Fig. 2), which may lead to misclassification and false positives, if spectral feature fitting and wavelength feature mapping methodologies are used for mapping or for inferring mineral chemistry. This can be of significant importance for mineral and deposit vectoring as the presence of abundant lichen coverage causing slightly shifted features for a given spectrum can be erroneously identified as a path to a deposit. A number of studies have investigated the spectral mixing of lichens and their rock substrate and the associated challenges in identification of mineralogy based on interpretation of mixed spectra (Ager and Milton 1987; Bechtel et al. 2002; Laakso et al. 2015; Laakso et al. 2016; Rivard and Arvidson 1992; Satterwhite et al. 1985). Despite many analyses focusing on spectral properties of lichens and rock-encrusting lichens, little attention has been paid to modeling of the effects of varying lichen abundances on rock signatures. Therefore, a rigorous examination of the effect and scale of wavelength displacements of mineral features as a function of mixing with different lichen abundances is conducted in Paper I.
Figure 2. Variability in the spectra of pure rock as compared to rock encrusted by 60 percent lichen cover in (a) full spectral range and (b) SWIR range. The spectra are offset for clarity. Middle and lowermost curves are indicative of pure lichen and rock, respectively. The uppermost curve is a computer-derived spectrum showing linear mixture of 60 percent lichen and 40 percent rock. Broad absorption features around 2.1 and 2.3 μm are characteristic features related to lichens and are highlighted by dash line in “a” and “b”. (c) The corresponding hull quotient of pure rock, lichen and the mixed spectra. An arrow denotes the wavelength displacement of rock absorption feature from shorter to longer wavelength.
14
Earlier observations suggest that the presence and abundance of lichens in a mixture can be estimated using absorption features at 680, 1730, 2100, and 2300 nm (Laakso et al. 2016; Théau et al. 2005; Zhang et al. 2004; Zhang et al. 2005). Numerous spectral indices for deconvolution of rock and lichen spectra have been established in previous research using these characteristic features and based on a limited number of rock/lichen samples (Bechtel et al. 2002; Morison et al. 2014; Petzold and Goward 1988; Rees et al. 2004; Rollin et al. 1994; Satterwhite et al. 1985). However, there are thought to be around 15000 species of lichens worldwide and not all lichen species are likely to be distinguished using these approaches. In other words, a much larger number of samples, from various study sites, should be measured and compared to evaluate the applicability of these indices. For this reason, defining a set of robust indices that can be used as a proxy for estimating the abundance of lichens regardless of the lichen species and the mineral composition of the rock substrate is vital to the interpretation of remote sensing data acquired in areas having abundant lichen-covered outcrops. Such an approach is presented in Paper II (Salehi et al. 2016). In addition, better techniques are required to increase the accuracy of the information derived from remote sensing imagery and extract the key spectral signatures more accurately in areas covered by abundant lichens (Paper III). Aside from the complications with processing mixture signatures and extracting reliable information for performing mineral mapping, a second major challenge in the Arctic is the accessibility of the outcrops and the lack of feasible approaches to capture the data as part of a large-scale operation in a time- and cost-effective manner. Near-vertical cliff sections particularly offer excellent rock exposures for investigating and characterizing mineral deposits. The spatial extent of these outcrops can range over kilometers, resulting in costly and time-consuming data acquisition, mapping, and interpretation. Due to the inaccessible nature of alpine, near-vertical topography, detailed mapping of lithologies and the spatial variation of mineral–chemical content is extremely challenging (Sørensen 2011; Svennevig et al. 2015). In addition, air/spaceborne nadir remote sensing sensors offer inadequate viewing direction for mapping of minerals in steep mountain cliffs. Oblique photogrammetry using handheld digital cameras, developed at the Geological Survey of Denmark and Greenland (GEUS), has been used extensively in areas of difficult access in Greenland for regional scale spatial information data acquisition and mapping. Using this technique, geological outcrops are digitally captured with high spatial resolution and accuracy from a helicopter or a sailing ship (Sørensen 2011; Sørensen, Bjerager, et al. 2015; Sørensen, Pedersen, et al. 2015; Svennevig 2014; Svennevig and Guarnieri 2012; Svennevig et al. 2015). With point precision and spacing of the order of a few centimetres, an enhanced quantitative element can be added to geological fieldwork and analysis, opening up new opportunities for investigations across a variety of scales in all areas of 15
field-based geology. However, being limited to three optical bands, oblique photogrammetry offers inadequate spectral resolution to allow for detection of subtle lithological differences. Integration with hyperspectral imagery (paper IV) allows detailed analysis of diagnostic absorption features of common minerals and enables the detection of subtle geochemical differences and actual quantitative analysis of outcrop composition (Bellian et al. 2007; Crowley 1986; Gaffey 1986; Goetz et al. 1985; Hunt and Salisbury 1971; Van der Meer 1995, 1996; Van der Meer and De Jong 2011). This can significantly improve collection of geological data to distinguish between lithologies and barren rock from potential economic ore deposits. In combination with photo-geological interpretations, this spectral information not only helps to evaluate the areas’ mineral potential, but also to highlight certain structures not visible in standard photographs. The fusion of terrestrial imaging spectrometry and geometrically accurate terrain/topography data, such as that derived from digital photogrammetry or laser scanning, has recently gained attention for local-scale mapping of minerals in near-vertical outcrops at distances between one to several hundred metres (Kurz et al. 2013; Kurz et al. 2012; Murphy and Monteiro 2013; Murphy et al. 2012). Within this range, the spatial resolution varies between centimetre and decimetre scale, enough to resolve even small-scale mineral compounds and fault systems. However, observing a geological target at close range is not always feasible, when it comes to the mountainous regions in Greenland. In particular, larger and vertically oriented targets such as steep mountain slopes and sea- or lake-faced cliffs are often only fully visible from an opposing location such as a neighbouring mountain (Rosa et al. 2017) or shore land (Salehi and Thaarup 2018). Long-range terrestrial hyperspectral data acquisition from remotely located steep mountain cliffs (i.e. in the range of several kilometres) is proposed and tested for the first time during a survey in summer 2016 (Rosa et al. 2017; Salehi and Thaarup 2018). In such a setup, the hyperspectral sensor is located on an opposing mountain tip (e.g. east of Qaarsukassak ridge in West Greenland (Rosa et al. 2017)) or shore land (e.g. Maligiaq, Qeqertalik fjord and White Mountain in SouthWest Greenland (Salehi and Thaarup 2018); Fig. 3) to collect the data from the target. The distance between the sensor and the target of interest can then easily exceed the close-range and extend to several kilometres.
16
Figure 3. Setup used to collect long-range hyperspectral data in the Maligiaq region (July 2016) (Salehi and Thaarup 2018).
The main drawback of this approach is that such scanning may not always be possible and difficult terrain accessibility may hinder instrumentation setup. Deployment of a platform in motion as an alternative to long-range terrestrial scanning is proposed and evaluated in Søndre Strømfjord (SouthWest) and Karrat region (West Greenland) to capture the data as part of a large-scale operation (Paper IV). Such a logistic platform provides a cheaper and time efficient solution as compared to the usage of terrestrial scanning and has the advantage that it is highly flexible and well suited for difficult and remote terrains. This simplifies the expansion of mapping in remote places (e.g. in Greenland), where a large proportion of the area is still under-explored and the lack of infrastructure reduces the capabilities to economically explore and locate mineral resources using traditional techniques.
17
Figure 4. General workflow for 3D-integration of hyperspectral data with point clouds obtained from high-oblique to near-horizontal photogrammetry.
In addition, it is demonstrated how automatic matching algorithms can be used for fusion of 2D hyperspectral imaging with oblique photogrammetry (3D pointclouds) and subsequently establishing accurate 3D surface models (hyperclouds) (Fig. 4). These models can be viewed from arbitrary directions in 3-D visualization programs and are well suited for both quantitative and qualitative analysis in large-scale geological mapping or in preparation for upcoming field operations. The 18
implemented algorithms proved to be robust towards integration of hyperspectral data and stereoimages captured from different camera locations (i.e. a helicopter or a ship). However, exploiting inaccurate viewing direction for data acquisition might complicate the subsequent matching process of the two image-sets and thereby induce distortions in the resulting georeferenced hyperspectral scans. Optimal results are generated using perpendicular view to the surface of outcrops for data acquisition. The observations from long-range sub-horizontal hyperspectral data acquisition (from both fixed (Rosa et al. 2017; Salehi and Thaarup 2018) and moving platforms (Salehi et al. 2018)) suggest that various sensor and environmental effects make the calibration and analysis of these measurements more challenging than the ones from close-range data acquisitions (Kurz et al. 2013). Firstly, in contrast to close-distance measurements, where the influence of the atmosphere is negligible, long distances lead to major atmospheric distortions. Furthermore, the logistics of setting up visible reference targets for radiometric correction and ground control points for image georeferencing is not possible. Additionally, owing to the much larger scale of the observed surface and the ground-based viewing perspective, pixels within one scene can represent a range of different distances and orientations, leading to highly variable radiometric distortions. Surrounding topography can also influence the measured at-sensor radiance by casting shadows, blocking diffuse sky irradiance or adding additional ground reflections (Kurz et al. 2013). For those reasons, the workflows established for atmospheric and topographic correction of nadir acquisitions are not applicable or need to be intensely modified to account for the special conditions of long-range spectral sensing. Accordingly, a novel approach is presented in paper V that allows the creation of fully corrected, long-range hyperspectral image data appropriate for geological applications (Fig. 5). This includes sensorinduced geometric distortion corrections as well as radiometric and topographic corrections using 3D surface data. Using the presented workflow, hyperspectral datasets from different camera locations and viewing angles at different acquisition dates and times can be integrated and merged by eliminating the effects of topography, different illumination conditions, and atmospheric absorptions. This allows using hyperspectral data in a new way, as it facilitates the evaluation of spatial relationships between hyperspectral results that are not visible from one observation point or displayable in one dataset such as opposing faces of a mountain.
19
Figure 5. Atmospheric and topographic correction workflow for long-range hyperspectral data acquisition (Maarmorilik marble cliffs, Nunngarut, West Greenland).
1.2 THESIS OBJECTIVES In an age of increasing remote sensing data availability, the potential for providing greater thematic detail relating to geology is growing rapidly. Recent advancements in hyperspectral sensors have shown great promise in this regard. However, questions remain regarding their utility in challenging environments such as the Arctic. One of the limitations of current state of the art methods concern the processing of mixture signatures (e.g. sub-pixel spectra mixture of lichens and rocks). Little is known of how the performance of current mapping methods is affected by the lichens encrusting the rocks in Arctic regions. Issues related to data quality and availability represent a further challenge. New data acquisition strategies are needed to acquire the data as part of a large-scale mapping operation in areas with rugged topography and difficult terrain accessibility. On a broader level, combining the advantages of sensors with different spatial and spectral resolutions offers great potential for the efficient use of resources through an integrated mapping system. Such systems are highly relevant for current needs related to mineral mapping in the Arctic. Greenland is one of the attractive targets for geological remote sensing because of its particular geological features and 20
processes and the vast territory, which is often under-explored due to poor accessibility and logistical challenges of the fieldwork. Such initiatives are needed to better understand the geology and expand the mapping by economically explore and locate mineral resources in this region. Hyperspectral sensors used for imaging spectroscopy can provide an information-rich representation of the surface materials and can assist the geological mapping and mineral exploration projects in Greenland from a reconnaissance to a district level of study. The overall aim of my Ph.D. research is to examine current limitations of hyperspectral mineral mapping in the Arctic (Greenland) and to establish new potentials for using such datasets by developing innovative strategies for data acquisition and integration as part of a large scale operation. For this purpose, four main objectives of my research are as following: (1) To advance the state of knowledge of the impact of varying lichen cover on the spectral recognition of common rock forming mineral features being vital for accurate interpretation of spectra (Paper I); (2) to establish robust spectral indices, insensitive to the species of the lichens or mineralogy of the rock substrate, for estimation of lichen abundance in remotely sensed imagery (Paper II); (3) to investigate the potential of different image processing methodologies for highlighting ultramafic rock units from airborne hyperspectral and spaceborne multispectral Sentinel-2 and Landsat-8 OLI data in areas of abundant lichens on rock outcrops (Paper III); (4) to develop robust, rapid and cost-efficient strategies for regional scale data acquisition in areas of difficult access in Greenland and the required image processing techniques to handle such data (Paper IV & V).
2 STUDY SITES All the research work during this PhD has been conducted for three remote and under-explored areas of Greenland that have high mineral potential, namely Karrat (Central West, Fig. 6a), SisimiutKangerlussuaq (South West, Fig. 6b) and Liverpool Land (Central East, Fig. 6c). The individual sites and case areas are presented in the relevant papers, but a short introduction to them together with an overview of which measurements are acquired and used from the respective sites are presented here and in section 3, respectively: 1) To examine lichens effect on spectral recognition of minerals (Paper I and Paper II), samples that comprise common and economically important rock types were investigated from Karrat, Sisimiut-Kangerlussuaq and Liverpool Land regions. 21
2) Mapping of arctic mineral resources in areas with abundant lichen cover using airborne imaging spectroscopy is the focus of Paper III. The regional airborne hyperspectral data from SisimiutKangerlussuaq area is used for this study. 3) Using two test cites from Karrat and Sisimiut-Kangerlussuaq regions, new strategies for multiscale hyperspectral measurements (using vessel-based or long-range terrestrial imaging systems) are presented and investigated in Papers VI and V. The spectral information is linked to the overlapping photographs acquired from a boat sailing along the fjords and a helicopter flying along the cliffs. Such subhorizontal data measurements compliment the data acquired with nadir viewing airborne and spaceborne platforms and can boost the scale of mapping in the areas of difficult access.
Figure 6. Geological map of a) Karrat region in South West, b) Sisimiut-Kangerlussuaq region in Central West and c) Liverpool Land region in Central East Greenland.
2.1 KARRAT The Karrat region in Central West Greenland (Fig. 6a) is characterized by rugged alpine terrain that in many areas is difficult to access. The morphology here is dominated by E-W trending, deeply incised fjords with 1 to 1.5 km high, steep, often nearly vertical cliffs. Rocks exposed along these cliffs are made up of Archean basement and Early Paleoproterozoic siliciclastic and carbonate 22
sequences of the Karrat Group (Grocott and McCaffrey 2017; Henderson and Pulvertaft 1987). The Archean basement comprises leucocratic orthogneiss, deformed during multiple orogenic events (Rosa et al. 2017). The Karrat Group sequence shows greenschist to amphibolite facies metamorphism (Rosa et al. 2017). The whole succession of Archean basement and Karrat Group meta-sediments was later overthrusted by Archean basement-cored nappes during the NW-SE directed compressional Rinkian event (Grocott and McCaffrey 2017). The investigated area covers large parts of the Nunngarut Peninsula at the Qaamarujuk fjord, where the former mining town of Maarmorilik is located. The nearby Black Angle Pb–Zn deposit is separated from the Nunngarut Peninsula by the smaller Affarlikassaa fjord. 2.2 SISIMIUT-KANGERLUSSUAQ REGION The Sisimiut-Kangerlussuaq region in South West Greenland (Fig. 6b) is located in the Southern part of the Palaeoproterozoic Nagssugtoqidian Orogen, approximately 500 km to the south of the Karrat area and consists of high-grade metamorphic ortho- and paragneisses and metabasic rocks (Fig. 6b and Fig. 8). The area hosts an alkaline province and a variety of ultramafic alkaline rocks, including swarms of kimberlite and lamproite dykes (Jensen et al. 2002; Larsen and Rex 1992). Kimberlites and lamproites are potential carriers of diamond, therefore several campaigns of commercial diamond exploration have been undertaken in this province since the description of the Sarfartoq carbonatite complex and the kimberlitic dykes related to this complex (Larsen and Rex 1992; Secher and Larsen 1980). 2.3 LIVERPOOL LAND Liverpool Land in Central East Greenland is underlain by Precambrian, marble-bearing metamorphic rocks and granites–quartz–monzonites of Caledonian or Neoproterozoic age belonging mainly to the Hagar Bjerg thrust sheet (Higgins et al. 2008). Some rocks on southern Liverpool Land are assigned to a lower thrust sheet. Liverpool Land forms a c. 3500 km2 horst of Caledonian crystalline rocks separated from the Jameson Land Basin to the west by a major N–S-oriented fault zone (Coe and Cheeney 1972). A northern continuation of the horst block occurs on Canning Land and Wegener Halvø where younger rocks are exposed. Previous reports confirm widespread, structurally-controlled copper mineralization along N–S-orientated lineaments in granite and migmatite in westernmost Liverpool Land (Holmes et al. 2014). The region has therefore been subject to several exploration campaigns. 23
3 HYPERSPECTRAL DATASET 3.1 REGIONAL AIRBORNE HYPERSPECTRAL DATA (HyMAP) In order to investigate the potential of using airborne hyperspectral data in the Arctic, the region between Sisimiut and Kangerlussuaq (Søndre Strømfjord) in South West Greenland was selected due to the availability of a regional airborne hyperspectral data set (i.e. HyMAP) that was acquired in 2002 (Tukiainen and Thorning 2005). Approximately 3500 line kilometers of imagery were collected using a HyMap sensor (Cocks et al. 1998) mounted on a Piper Navajo Chieftain aircraft. The survey comprises 54 flight lines covering approximately 7500 km2 of ground, out of which 19 flight lines were selected for this study. The aircraft was flown at an altitude of approximately 2500 m at which the scanner’s swath width is approximately three kilometers. Due to the rugged terrain of the survey area, the ground spatial resolution varies from three to five meters depending on the altitude from the surveyed ground. The sensor has 126 channels over the 0.45-2.5 µm wavelength range with average spectral resolution of ~10 nm. Data were delivered by the survey company (HyVista Corp., Australia) in units of radiance. The airborne data were geometrically corrected and geocoded by Tapani Tukiainen (GEUS) using the PARGE (Parametric Geocoding) software (Schläpfer and Richter 2002). This procedure reconstructs the scanning geometry for each image pixel using position, attitude and terrain elevation data. In order to conduct accurate quantitative analysis, I performed the subsequent preprocessing steps on the geocoded radiance data (i.e. atmospheric, radiometric and topographic corrections and calibration of apparent surface reflectance values using field measurements) before masking non-outcrop pixels and generating the final mosaic: The geocoded radiance data were converted to apparent surface reflectance using a radiative transfer program, Atmospheric and Topographic Correction for airborne imagery (ATCOR-4), in rugged terrain mode. The ATCOR-4 rugged terrain mode utilizes a surface elevation model to adjust illumination levels. Apparent surface reflectance values from the ATCOR4 processing were empirically adjusted using ground-based reflectance measurements from calibration
sites
measured
with
an
Analytical
Spectral
Devices
FieldSpec®3
HiRes
Spectroradiometer (ASD). ASD data were collected during the hyperspectral scanner overflights from numerous pseudo invariant features (PIFs) that were mainly homogenous, vegetation and lichen-free exposures of alluvial material.
24
Of the 126 atmospherically corrected channels of HyMAP data, I used 106 channels over the 0.47to 2.46-μm spectral range. Channels not used are primarily associated with H2O and OH absorption features near 1.4 and 1.9 μm. The images were mosaicked together using the mosaic function in ENVI (ENvironment for Visualizing Images; Harris Geospatial Solutions, Broomfield, Colorado). To improve the quality of the mosaic image, each flight line was subject to masking for clouds, cloud shadows, water, snow\ice and areas of poorly illuminated steep terrain (steep slopes facing away from the incident solar radiation). Water, snow\ice and low albedo pixels show lower reflectance values within the SWIR range as compared to rock outcrops. The SWIR range was therefore considered as appropriate to identify these pixels and consequently masking them out. For this purpose, the mean reflectance value was calculated for all the pixels of each flight line throughout the wavelength range of 1.5-2.46 μm and a single band image (here referred to as mean reflectance image) was generated. Figure 7a displays the histogram plot of the calculated statistics for the generated single band images (each color corresponds to the computed statistics for individual flight lines). The number of pixels and mean reflectance values are shown in y and x-axis, respectively. Based on the statistics of the mean reflectance at wavelengths 1.5-2.46 μm, a threshold of 100 (red dashed line in Fig. 7a) was set to mask out pixels related to water, snow\ice and deep topographic shading. A pixel containing cloud or cloud shadow was determined by comparing reflectance levels for five channels (corresponding to wavelength positions 0.5, 1.01, 1.23, 1.54, and 2.11 μm) against threshold values. Bright pixels were identified as cloud contaminated and dark pixels were identified as shadowed (Fig. 7b). The two aforementioned masks were then combined into a single mask for each flight line. In the mosaicking procedure, the masked pixels are often filled in with non-masked data from adjacent flight lines because clouds and cloud shadows shifted in the time that elapsed between aircraft passes.
25
Figure 7. a) Histogram plot of the mean reflectance values over the wavelength range of 1.5-2.46 μm for all the pixels of each flight line. The dashed red line indicates the selected masking threshold to differentiate the rock exposure from water, snow\ice and low albedo pixels; b) histogram plot of reflectance values for five channels (corresponding to wavelength positions 0.5, 1.01, 1.23, 1.54, and 2.11 μm) for one of the flight lines. The selected thresholds for extracting pixels contaminated by cloud (bright pixels) and cloud shadow (dark pixels) are indicated by red dashed lines. This threshold varies slightly between different acquisitions but is straightforward to identify as shown here. Masking is usually most successful around 150 and 3500 as shown in the plot.
Most across-track brightness gradients that appear after atmospheric correction are caused by bidirectional reflectance distribution function (BRDF) effects; because the sensors view angle and solar illumination geometry vary over a large range. The BRDF effect can be especially strong in rugged terrain with slopes facing the sun and others oriented away from the sun. No cross-track illumination or BRDF correction was applied to correct the illumination effects caused by topography and the scanning geometry (cross-track illumination change). A spectral polishing can be performed for the atmospherically and / or BRDF corrected data. 3.2 FIELD MEASUREMENTS To examine how hyperspectral data from multiple sources and scales can be combined for expanding the scale of mapping in Greenland in future; I organized a three weeks survey in July 2016 in the Sisimiut-Kangerlussuaq region, which involved a crew of seven persons. I was responsible for overall planning and coordination of the survey as well as for data acquisition and development of the processing workflows for correction of such datasets. The idea is to compliment the information
26
residing in regional airborne hyperspectral data by the data collected on the ground (see section 3.2.1) or from moving platforms (e.g. a small ship; see section 3.2.2). The Specim AisaFenix pushbroom hyperspectral scanner (Specim Spectral Imaging Ltd, Oulu, Finland) was utilized to acquire hyperspectral data in the wavelength range from 380 to 2500 nm, with a spectral resolution of 3.5 nm and 12 nm and a sampling interval of 1.5 nm and 5 nm in VNIR and SWIR, respectively (Table 1). For terrestrial data acquisition, the instrument is mounted on a rotary stage and a continuous hyperspectral image with a vertical field of view (FOV) of 32.3° and a maximum scanning angle of 130° is acquired in one measurement. In case of acquiring the data from a moving platform, the acquisition parameters were set to a rapid scanning speed resulting in a fast frame rate of 30–40 Hz, and a short integration time of 23–28 ms for VNIR and 15–20 ms for SWIR. The data were captured continuously and with sufficient overlap at a distance to the coast of 1.5 to 2.5 kilometers, which translates into images with pixel sizes on the ground (ground sampling distance) of about 2 to 4 meters. Each file has a size of approximately 600 MB, and it takes approximately one minute to collect the spectral image. During the measurements, the GPS position of the camera, acquisition time and general viewing direction of the scan were recorded. A Spectralon SRS-99 white panel was set up near the camera within the FOV and with a similar general orientation as the surface of the imaged outcrop. A field team of two persons (i.e. camera operator and an assistant) was necessary to setup the instrument and acquire the data. The data processing related to such data acquisition has been fully discussed in (Lorenz et al. 2018; Salehi et al. 2018). The target areas visited in the field were selected on the basis of preliminary interpretations of HyMap scenes and geology (Korstgård 1980). Sample locations were recorded using a handheld GPS device. Spectra of representative fresh and weathered rock surfaces were acquired in a laboratory setting using a FieldSpec®3 HiRes Spectroradiometer equipped with a contact probe (10 mm spot size) with an internal light source. This ASD has a spectral resolution of 3 nm for the VNIR, 8.5 nm for the SWIR1, and 6.5 nm for the SWIR2 and a sampling interval of 1.4 nm between 350 and 1050 nm, and 2 nm between 1000 and 2500 nm, resulting in 2151 channels in the spectral range from 350 to 2500 nm (Table 1). Each reading consists of 25 individual scans taken consecutively and averaged by the ASD, which is then converted to reflectance using a SRT-99-050 white reference panel (Bruegge et al. 1993). By the end of field campaign, the Specim AisaFenix hyperspectral scanner was transferred to Karrat region that was selected as the second test site. The new data acquisition approaches (i.e. vessel-based and long-range terrestrial scanning) established in Sisimiut-Kangerlussuaq region (Salehi et al. 2018; 27
Salehi and Thaarup 2018) were applied here again. The achieved results (Lorenz et al. 2018; Rosa et al. 2017) were promising and confirmed the applicability of such strategies. Table 1: Characteristics of the used hyperspectral instruments
HyMap
Specim AisaFenix
FieldSpec®3 HiRes Spectroradiometer
Module
Spectral Range (nm)
Spectral Resolution (nm)
Sampling Interval (nm)
VIS NIR SWIR1 SWIR2 VNIR SWIR VNIR SWIR1 SWIR2
450 – 890 890 – 1350 1400 – 1800 1950 – 2480 380 – 970 970 – 2500 350 – 1050
5 – 16 5 – 16 5 – 16 18 – 20 3.5 12 3 8.5 6.5
15 15 13 17 1.5 5 1.4
1000 – 2500
2
3.2.1 Long-range terrestrial hyperspectral scanning in the southern Nagssugtoqidian orogeny (South West Greenland) and Karrat region (Central West Greenland)
New hyperspectral datasets from relatively distant targets (>2 km distance) were acquired in the allochthonous (Fig. 8, scan 1-6) and autochtonous zone (Fig.8, scan 8) of the Nagssugtoqidian orogeny. Metasedimentary rocks in the allochthonous zone are tectonically interleaved with quartzofeldspatic gneisses between the southern shore of Qeqertalik fjord and the northern border of the Ikertoq shear zone. Parts of this zone are well exposed along the eastern shore of Maligiaq and were scanned with the Specim AisaFenix hyperspectral scanner from a distance of approximately 2–3 km (Fig. 8, scan 1-5). The metasedimentary rocks are predominantly biotite-garnet paragneisses but also comprise aluminous schists and graphitic–sulfidic varieties with up to a few percent graphite and iron sulfides, and they may include both Archaean and Palaeoproterozoic components (see legend to the regional map of Garde & Marker 2010). The targets in the Karrat region (second test site) were selected on basis of previous photogeological interpretations derived from a large number of oblique stereo-images acquired from helicopter and vessel in 2015 covering vast parts of the region (Rosa et al. 2016). Regions of interest encompass the contacts of the Archean to overlying Paleoproterozoic meta-sedimentary rocks and associated prospective horizons. To investigate the applicability and robustness of the workflow under different measuring conditions regarding climate, distance, atmospheric composition, geological diversity, and mapping objectives a third case study was examined in Corta Atalaya, an open-pit mine within the city limits of Minas de Riotinto (Spain) and under Mediterranean climate conditions. In this test site the distance to the 28
target is shorter (i.e. between 400-1100 m) compared with that at Greenland test sites (>2 km distance). 3.2.2 Integration of ship-based hyperspectral scanning and 3D-photogrammetry
Mobile mapping of steep coastal cliffs using ship-based hyperspectral scanning was tested in the outer part of the Søndre Strømfjord area (Figs 8, scan 9) and in the Karrat region along the coastline of Kirgasima peninsula while sailing from Kangerluarsuk to Inukassaat Fjord (Salehi et al. 2018). The spectral mapping products were integrated with 3D photogrammetric data to create accurate, largescale outcrop models, which are well suited for quantitative purposes in geological analysis or in preparation for field operations. The data processing related to such data acquisition has been fully discussed in Salehi et al. (2018). The steep coastal cliffs, which have been scanned in Søndre Strømfjord (Kangerlussuaq area) are located south of the borderland between the late Nagssugtoqidian Orogen and the early Kangâmiut Complex (Ramberg 1948). The Kangâmiut Complex comprises undeformed mafic dyke swarms emplaced into the high-grade gneisses of the Archaean craton of southwest Greenland and is characterized by hornblende as the dominant primary ferromagnesian mineral (Bridgwater et al. 1995; Connelly and Mengel 2000; Escher et al. 1975; Ramberg 1949).
29
Figure 8. Geological map across the southern Nagssugtoqidian Orogen simplified from Garde and Marker (2010) and locations of scanned areas mentioned in the text. Red frame outlines the area that is covered by airborne hyperspectral data selected for this study. From south to north this region comprises (K. Sørensen, personal communication 2018): an autochthonous zone in which deformation regularly increases towards the parautochthonous zone expressed in clockwise rotation of Kangâmiut basic dykes, a parautochthonous zone in which Nagssugtoqidian deformation and metamorphism are highly heterogeneous, and an allochthonous zone where juvenile metasedimentary rocks are interfolded and thrust-stacked with Archaean gneisses containing amphibolitic metadykes.
30
4 DATA PROCESSING AND ANALYSIS OF THE SURFACE MINERALOGY The choice of processing algorithms and techniques involved in analysis of hyperspectral imagery for geological applications is important to obtain valid results. One of the major issues that can affect the performance of algorithms used to process hyperspectral data is the mixed pixels, i.e. pixels containing more than one land cover type. This is especially important in case of data acquired by high altitude sensors, which cover wide areas and the recorded hyperspectral image is a combination of pure and mixed pixels. Consequently, the so-called full pixel techniques (Camps-Valls and Bruzzone 2009; Landgrebe 2005), which are based on the assumption that each pixel corresponds to the spectral signature of one predominant land cover type are not suitable for the analysis of mixed pixels. When used for scenarios with a high number of sites with mixtures of land cover classes, these approaches will inevitably lead to a high error rate. 4.1 ENDMEMBER EXTRACTION AND SPECTRAL UNMIXING The issue of mixed pixels has been the focus of numerous researches. A widely investigated approach is the use of soft classification techniques, where each pixel displays multiple and partial class membership (Nachtegael et al. 2007). These techniques are useful means of reducing the mixed pixel problem and acquiring sub-pixel scale information (Adams et al. 1993; Roberts et al. 1998). The discrimination and classification of lithological units using hyperspectral imagery primarily makes use of one common soft classification technique namely Spectral Mixture Analysis (SMA) (Singer and McCord 1979). Spectral unmixing is the procedure by which the measured spectrum of a pixel is decomposed into a collection of pure reflectance spectra referred to as endmembers. One of the main assumptions deployed in these models is that secondary reflections and/or multiple scattering effects in the data collection procedure is minimal. In contrast, nonlinear mixture models assume that the incident radiation is affected by multiple scattering effects and interacts with more than one component, so the endmembers form an intimate mixture inside the respective pixel (Keshava and Mustard 2002). Such methods require prior knowledge about the geometry and physical properties of the observed objects, which often may not be available. In the majority of applications a linear mixing model can be used without significant loss of information (Keshava and Mustard 2002). Reference endmembers can be either laboratory or field spectral measurements or can be extracted directly from the image. Image endmembers have the advantage of being collected at the same scale 31
and environmental conditions as the data, and thus can be better associated with materials in the scene. The results of spectral mixture analysis are very sensitive to the quality and number of endmembers used in the mixture modelling (Tompkins et al. 1997). In many situations, identification of the endmembers may be challenging due to insufficient spatial resolution, mixtures happening at different scales, and unavailability of completely pure spectral signatures in the scene. For a given mixed pixel, too many endmembers may overfit the data yielding an unstable solution, while too few endmembers results in large residuals and the fraction of an unmodeled component is partitioned into the fraction estimate of the selected endmembers (Roberts et al. 1998). Classic endmember extraction techniques (Boardman et al. 1995; Nascimento and Dias 2005; Neville 1999) assume that the input dataset contains at least one pure pixel for each distinct material present in the scene, and therefore, have focused on estimating the smallest simplex set containing the observed spectral vectors. However, this assumption may be difficult to guarantee if the spatial resolution of the sensor is not high enough to separate different pure signature classes at a macroscopic level. In this case, the use of image-derived endmembers may not result in accurate fractional abundance estimations since it is likely that such endmembers may not be completely pure in nature. Mixed pixels can also result when distinct materials are combined into a microscopic (intimate) mixture, independent from the spatial resolution of the sensor. Since the mixtures in this situation happen at the particle level, the use of image-derived spectral endmembers cannot accurately characterize intimate spectral mixtures. In all the investigations carried out in this thesis, the spatial–spectral endmember extraction (SSEE) method (Rogge et al. 2012; Rogge et al. 2007) yielded the best results. The reason is that this method works by analyzing a scene in parts (subsets), such that the spectral contrast of low contrast endmembers increases. This enables the assessment of subtle lithological variability across a given study area. In addition, special attention has been given to soft classification techniques such as iterative spectral mixture analysis (ISMA) (Rogge et al. 2006) and bounded value least squares (BVLS) unmixing (Mielke et al. 2016; Rogass et al. 2014). These approaches allow determining the fractional abundances of the endmembers within the scene and can be used to remove the pixels that are impacted by the lichen effects 4.2 ABSORPTION WAVELENGTH MAPPING Absorption feature parameters, i.e. wavelength position and depth, are the most essential information used in spectroscopy and can be directly linked to mineral types and their abundances, as well as to subtle changes in the chemical composition of minerals (Rodger et al. 2012; Van Der Meer 2004; 32
Van der Meer et al. 2018; Van Ruitenbeek et al. 2014). This information is particularly useful in areas where field validation is sparse and imagery contains shallow spectral absorption features. Among available absorption feature mapping/matching techniques (Bakker et al. 2011; Kopačková and Koucká 2014), the spectral feature fitting (Clark et al. 1991) and its improved version, multi-range spectral feature fitting (MRSFF), are frequently used for geological mapping applications (Kayet et al. 2018; Pan et al. 2013; Shanmugam and SrinivasaPerumal 2014). In addition, Tetracoder (Clark et al. 2003) and EnGeoMAP (Mielke et al. 2016) are the two main recently developed toolboxes that are used for integrating absorption feature matching techniques. The common principle behind these approaches is that they compare and statistically assess the fit of the image spectra to the reference spectra (endmember). If laboratory or field spectra are used, the successful definition of such endmembers usually requires prior knowledge of the material composition and its spatial distribution within the area of interest. If image endmembers are to be used, an expert with a background in spectroscopy is required to perform the image analysis for extracting the right set of endmembers.
5 INTRODUCTION TO REASEARCH PAPERS The research carried out in the course of this thesis is compiled in five papers. The first two papers deal with the challenges associated with lichen-coatings on the rock substrate that may cause misinterpretation of mixed spectra. In the first paper, the impact of lichens on the spectra of the rock substrate and the related wavelength displacements of characteristic absorption features of common rock forming minerals are modelled. The same dataset is used in the second paper, where a set of robust lichen indices are suggested for the deconvolution of lichen and rock mixtures and providing an estimation of the lichen abundance regardless of the mineral composition of the rock substrate. In Paper III, the performance of two approaches for mapping the mafic-ultramafic rocks in areas covered by abundant lichens is investigated. This study demonstrates that these approaches are suited for mapping with reasonable accuracy in the arctic environment of Greenland. The mineral information at sub-pixel level could be a valuable guide for the geological and exploration community to discover economic deposits. The main objective of the fourth and fifth papers is the development and evaluation of rapid and robust data acquisition strategies and techniques in order to provide information on mineralogy as part of large-scale mapping operations in a time and cost-efficient manner. A novel and flexible approach, i.e. using a ship/boat as a platform for hyperspectral data acquisition, is introduced in paper IV. This strategy enables the mapping of near-vertical cliff sections along fjords, coastlines and 33
valleys in remote regions, which are difficult to map by means of classic geological field campaigns considering the difficult accessibility of rugged terrain or the space-airborne remote sensing surveys due to insufficient viewing directions. Furthermore, the mapping products generated from hyperspectral datasets are brought to the third dimension (here referred to as hyperclouds) and are integrated with high-resolution three-dimensional terrain surface models (3D-pointclouds) generated from oblique photogrammetry to complement geological outcrop models with quantitative information regarding mineral variations. This helps geologists to visualize, manipulate and interact with the data in a three dimensional setting and enables them to distinguish the spatial distribution of different rock types and minerals, in a non-contact manner. This is beneficial for exploration in arctic and alpine environments and is an alternative for costly and time-consuming data acquisition, mapping, and interpretation. Based on the results from this paper (IV) and the observations from longrange ground-based hyperspectral imagery of distant targets during field campaign in summer 2016 (Rosa et al. 2017; Salehi and Thaarup 2018), the fifth paper further elaborates by presenting a workflow for correcting the extreme influence of atmospheric effects and topography-induced illumination differences remained in the dataset. These effects cannot be corrected by means of common correction tools for nadir satellite or airborne data. Therefore, the existing approaches are modified to account for the special conditions of long-range sub-horizontal outcrop sensing
34
5.1 INTRODUCTION TO PAPER I Modeling and assessment of wavelength displacements of characteristic absorption features of common rock forming minerals encrusted by lichens. Published in Remote Sensing of Environment Paper overview Although the application of hyperspectral remote sensing to lithological/mineralogical mapping is well-established, the lichens remain one of the main factors limiting its wider use. Spectral mixing of lichens and bare rock can shift the wavelength position of characteristic absorption features and complicate the spectral mapping of minerals and lithologies. The extent to which diagnostic rock features are preserved despite the presence of lichens is of major concern in remotely sensed geological studies, and estimates of the critical level of lichen coverage, below which spectral features of the mineral substrate can still be identified, are needed. This paper examines how surficial lichen cover affects the characteristics of shortwave infrared (SWIR) mineral absorption features and the efficacy of automated absorption feature extraction. For this purpose, samples comprising common and economically important rock types were investigated, originating from three tracts at different parts of Greenland, namely Liverpool Land (Central East), Karrat (Central West) and Sisimiut- Kangerlussuaq (South West). Mixed spectra were synthetically generated from laboratory spectra of common rock forming minerals and lichens. Wavelength displacements of characteristic absorption features for each mixed spectrum were then analysed as a function of percentage lichen cover. Main contributions This study highlights the importance of being cautious in any interpretation in areas characterized by abundant lichen cover on outcrops. Quantifying lichen cover effects on mineral absorption features allows for establishing strategies of how to adapt data analysis to minimize this effect. This can be of significant importance for mineral and deposit vectoring as the slightly shifted features caused by abundant lichen coverage might erroneously identified as a path to a deposit.
35
5.2 INTRODUCTION TO PAPER II Identification of a robust lichen index for the deconvolution of lichen and rock mixtures using pattern search algorithm (case study: Greenland) Published in International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences. XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic. Paper overview The ability to distinguish a lichen cover from its rock/mineral substrate and decompose a mixed pixel into a set of pure reflectance spectra, can improve the use of hyperspectral methods for mineral exploration in areas where rocks are covered by varying abundances of lichens. The objective of this study is to establish a set of robust lichen indices that directly reflects the mixture ratio of the rock and lichen in the hyperspectral data, regardless of the mineral composition of the underlying rocks and the types of lichens species. Synthetic linear mixtures of pure rock and lichen spectra have been simulated using laboratory spectroscopic data. A number of index structures (i.e. ratio, normalized ratio and subtraction) were assigned to an optimization algorithm, namely Pattern Search algorithm (PS), to find the best wavelength positions (bands) along the reflectance spectra for calculation of the indices that can be used for estimating the lichen abundance. Next, the spectra were resampled to HyMAP resolution in order to investigate the functionality of the indices for data collected from an airborne platform. Unlike conventional indices proposed for such studies, the machine learning technique used here suggests that using the average reflectance within a wavelength region is superior to using the reflectance at individual bands for calculations of the indices. At the laboratory scale, a band ratio index that is based on dividing the average reflectance at spectral region 894–1247 nm by the reflectance at 1110 nm (R894–1247/R1110) shows the best performance for lichen-substrate differentiation. While a normalized ratio index using the average reflectance at the spectral region 1106-1121 and 904-1251 nm (R1106-1121 – R904-1251/ R1106-1121+ R904-1251) yields the best results for the HyMAP resolution. Main contributions It is expected that the results of this study have significant implications for the analysis of satellite or airborne remote sensing imagery acquired over lichen-covered terrains. The proposed indices proved to be robust for variable rock types and were able to estimate the lichen coverage with acceptable error levels. The proposed methodology has the advantage of not requiring a priori knowledge about the exact effects of lichens on the reflectance of the mixtures. Instead, an automated trial and error 36
process obtains this information. Therefore, this technique can also be beneficial for identification of sensitive bands and indices for deconvolution of any other mixed spectra, whether synthetic as in this case or obtained directly from the samples.
37
5.3 INTRODUCTION TO PAPER III Mapping of arctic resources in presence of abundant lichen using airborne imaging spectroscopy and spaceborne Sentinel-2 and Landsat-8 OLI data (Case Study: the Niaqornarssuit complex in South West Greenland) To be submitted. Target Journal: European journal of remote sensing Paper overview This study investigates the use of airborne hyperspectral (HyMAP) and spaceborne multispectral (Sentinel-2 and Landsat-8 OLI) data to map the geological materials of interest, specifically maficultramafic units in areas covered by abundant lichen. The performance of two different approaches namely EnMAP Geological Mapper (EnGeoMAP) and Iterative Spectral Mixture Analysis (ISMA) are tested. In addition, the Iron Feature Depth (IFD) index is used as a potential proxy for mapping the spatial extent of mafic-ultramafic bodies. The results are quantitatively compared to the reference image using the Structural Similarity Index Measure (SSIM). Main contributions The results from this study demonstrate that EnGeoMAP and ISMA methods can be used with expert knowledge to explore large-scale airborne and satellite surveys and generate lithological maps independent of ground truth data. This is essential in arctic terranes where ground data may not exist, field access is limited and costs are prohibitive. Promising results were achieved using the IFD method for detection of mafic-ultramafic bodies, even in areas highly affected by lichen coverage. The distribution of mafic-ultramafic bodies were highlighted with more spatial details using the hyperspectral reference data. However, results also showed that mafic and ultramafic units could be still discriminated and mapped using multispectral Sentinel-2 data. The least performance is achieved for the Landsat-8 OLI data, where only the core areas of the ultramafic complex are visible. The results of this study suggest that despite extensive lichen cover in the arctic regions, valuable lithological discrimination for supporting mineral exploration, specifically mafic and ultramafic units with similar mineralogy, can be obtained at both airborne and satellite resolutions.
38
5.4 INTRODUCTION TO PAPER IV Integration of vessel-based hyperspectral scanning and 3d-photogrammetry for mobile mapping of steep coastal cliffs in the arctic. Published in Remote Sensing Paper overview A robust and cost-efficient processing chain is developed for hyperspectral data acquisition from a moving platform, i.e. a ship/boat to optimize performance and reduce the time and effort required to map large inaccessible areas in the Arctic. Hyperspectral derived mineral maps are transformed to three-dimensional “hyperclouds” and combined with terrain models (pointclouds) that are generated from oblique photogrammetry. A 3D quantification and interpretation of geology along with the topography and structural features can be carried out in the subsequent step. Main contributions Deployment of a platform in motion as opposed to terrestrial scanning has the advantage that it is highly flexible in its operation, would provide a more time-efficient solution and is well-suited for difficult and remote terrains. The proposed automatic approach for combining spectral and point cloud data has high potential for field geologists, who wish to establish accurate 3D (surface) outcrop models. These models are well-suited for visualization as well as for quantitative purposes in geological mapping or in the preparation for field operations. The geometric accuracy of final products and the access to the third dimension allow a precise mapping of the targets identified using the hyperspectral data contents and deliver not only lithological constraints but also structural control, which is an important addition to the way mapping is undertaken in the arctic. The implemented algorithms work reliably and with high accuracy even for complex geometries and with stereo images acquired at different angles and distances from different platforms. Slightly distorted (hyperspectral) data, such as images over a low-relief landscape, can be treated quickly using homographic or polynomial transformations and even data with high local distortions caused by the underlying topography can be processed. The presented workflow opens up a new range of possibilities for using hyperspectral imagery by significantly enlarging the scale of measurements. The achieved results highlight the potential of using hyperspectral data acquired from other intermediate operational mobile platforms (e.g. drones, helicopters) for regional mineral mapping in inaccessible regions. This simplifies the expansion of mapping in remote places, where a large proportion of the area is still under-explored and the lack of
39
infrastructure reduces the capabilities to economically explore and locate mineral resources using traditional techniques.
40
5.5 INTRODUCTION TO PAPER V Radiometric correction and 3D integration of long-range ground-based hyperspectral imagery for mineral exploration of vertical outcrops. Published in Remote Sensing Paper overview An adapted workflow is presented for pre-processing of long-range outcrop sensing data, including straightforward atmospheric and topographic corrections. A single atmospheric correction spectrum is used for each scene to remove atmospheric distortions. For the correction approach to be robust and independent from additional knowledge about the composition of the atmospheric layer, the atmospheric correction spectrum is derived directly and automatically from the hyperspectral image itself. The correction spectrum is neither selective nor restricted to defined components and is thus applicable for any atmospheric setting. By deploying the workflow proposed in the paper IV, each pixel of the hyperspectral image (and any generated map) has an assigned geographic position and elevation. Using this information for each pixel of the spectral raster, a geometrically and “spectrally” accurate 3D-hyperspectral datacube is created. Next, the sun incidence angle (the angle between the point normal and the sun vector) is calculated for each individual point of the point cloud and used to apply a topographic correction algorithm. Main contributions The workflow developed for 3D integration of pointclouds and hyperspectral datasets proved to be successful for two additional examples (in addition to the case studies investigated in the previous paper). The approaches used in paper IV are extended here by incorporating radiometric and topographic corrections in the processing chain, which enables the integration and merging of hyperspectral datasets acquired from different camera locations and viewing angles as well as different acquisition dates and times (i.e. varying illumination conditions). This enables reliable spectral mapping of vertical inaccessible outcrops and allows the use of hyperspectral data in a new way, as it facilitates the evaluation of spatial relationships between hyperspectral results that are not visible from one observation point or displayable in one dataset, such as opposing faces of a mountain or a mining pit. The 3D hypercloud also allows for integration with other spatial datasets such as borehole data or structural observations.
41
6 SYNTHESIS, CONCLUSIONS AND OUTLOOK Analysis of high spatial and spectral resolution hyperspectral data in high latitude regions provides spatially contiguous mineralogical and lithological information, which cannot easily be obtained in any other way. Such information is valuable in multi-disciplinary geological studies and can be combined with other types of data obtained by geophysical, geochemical and petrological techniques, to assist in traditional mapping and mineral exploration. With future high-quality hyperspectral data acquired from satellites, the availability and areal coverage of such datasets will increase, opening new possibilities for the use of hyperspectral remote sensing in geology. Despite this, hyperspectral remote sensing techniques are yet to be fully exploited, a situation that may result from multiple factors, such as a lack of readily available data processing tools and insufficient understanding about the possibilities and limitations of these techniques. In this thesis, I focus on potentials and challenges for hyperspectral mineral mapping in the Arctic and developing innovative strategies for data acquisition and integration. The results of this work have contributed to advance the use of hyperspectral remote sensing for mineral detection and exploration in general and in the Arctic in particular. The spectral mixing of lichens and rock substrates is one of the limiting factors for using hyperspectral data in the Arctic. The results of this study suggest that abundant lichen cover induces wavelength displacements in characteristic rock features and thereby can lead to misclassification. These spectral displacements are not constant and may vary for different absorption features of a given mineral. Modal abundance of the mineral in certain rock types defines the proportion of lichen needed to induce the related wavelength displacement. Background spectra related to other minerals also play a critical role in the scale of wavelength displacement. My investigations demonstrate that it is possible to establish a set of generic robust lichen indices that directly reflects the mixture ratio of the rock and lichen in the hyperspectral data. The methodology used to define these generic indices has the advantage of not requiring a priori knowledge about the exact effects of lichens on the reflectance of the mixtures and can be used regardless of the mineralogy of the rock substrate and the species of lichens. Furthermore, the performance of three different approaches is investigated to produce detailed lithological maps in an arctic region (the Niaqornarssuit complex in South West Greenland) where mafic-ultramafic rock units are exposed in the presence of lichen coatings. The results from this study indicate that despite extensive lichen cover in the arctic regions, valuable lithological discrimination for supporting mineral exploration can be obtained at both airborne and satellite resolutions. 42
Advances in airborne and spaceborne multispectral and hyperspectral sensors have been successful for mineral prospecting and the regional mapping of rock types. However, because of the nadirviewing angle of the sensor, such a configuration is of limited value for near-vertical cliff sections. Long-range terrestrial and vessel-based spectral imaging are two new approaches developed through the context of this Ph.D thesis for regional scale data acquisition and mapping. These approaches offer new possibilities for an improved understanding of outcrop composition by mapping lithology and the distribution of mineralogy in areas of difficult access. Using such a scanning setup, steep cliff sections and quarry walls can be scanned with a more appropriate viewing direction and a higher image resolution than can be detained from airborne and spaceborne platforms. This opens up a new range of possibilities for application of hyperspectral imagery by significantly enlarging the scale of measurements. This also highlights the potential of using other intermediate platforms (i.e. drones, helicopters) to complete the information acquired using airborne and spaceborne sensors. In order to do this, multi-source and multi-scale data fusion approaches for mapping surface properties must be developed. As an example, integration of hyperspectral imaging with oblique photogrammetry provides the hyperspectral information in a georeferenced framework and enables measurement at centimetre scale. The spectral images and final maps could be superimposed on the geological outcrop models, allowing a simultaneous visualization of multiple thematic results together with the conventional digital camera imagery. To extract both geometrically and spectrally correct representations of the hyperspectral data cubes, an adapted workflow is developed and integrated in the preprocessing chain for atmospheric and topographic corrections of the data captured from distant near-vertical outcrops. This enables reliable spectral mapping of these outcrops and allows the use of hyperspectral data in a new way, as it facilitates the integration of hyperspectral results that are not visible from one observation point or displayable in one dataset, such as opposing faces of a mountain or a mining pit. Encouraging results were produced for all the four case studies used (paper IV and V), which proves the applicability and robustness of the workflow in differently challenging measuring conditions regarding climate, distance, atmospheric composition, geological diversity, and mapping objectives. It is therefore concluded that integration of 3D-spectral data with other spatial datasets such as boreholes, structural observations and/or any other 3D model is feasible.
43
7 REFERENCES Adams, J. B., Smith, M., & Gillespie, A. (1993). Imaging spectroscopy: Interpretation based on spectral mixture analysis (C. Pieters & P. Englert, Trans.). New York: Cambridge University Press. Ager, C. M., & Milton, N. M. (1987). Spectral reflectance of lichens and their effects on the reflectance of rock substrates. Geophysics, 52(7), 898-906. doi: 10.1190/1.1442360 Bakker, W., van Ruitenbeek, F., & van der Werff, H. (2011). Hyperspectral image mapping by automatic color coding of absorption features Bechtel, R., Rivard, B., & Sánchez-Azofeifa, A. (2002). Spectral properties of foliose and crustose lichens based on laboratory experiments. Remote Sensing of Environment, 82(2), 389-396. doi: 10.1016/S0034-4257(02)00055-X Bedini, E. (2009). Mapping lithology of the Sarfartoq carbonatite complex, southern West Greenland, using HyMap imaging spectrometer data. Remote Sensing of Environment, 113(6), 1208-1219 Bedini, E. (2011). Mineral mapping in the Kap Simpson complex, central East Greenland, using HyMap and ASTER remote sensing data. Advances in Space Research, 47(1), 60-73 Bedini, E. (2012). Mapping alteration minerals at Malmbjerg molybdenum deposit, central East Greenland, by Kohonen self-organizing maps and matched filter analysis of HyMap data. International journal of remote sensing, 33(4), 939-961 Bellian, J. A., Beck, R., & Kerans, C. (2007). Analysis of hyperspectral and lidar data: Remote optical mineralogy and fracture identification. Geosphere, 3(6), 491-500. doi: 10.1130/GES00097.1 Boardman, J. W., Kruse, F. A., & Green, R. O. (1995). Mapping target signatures via partial unmixing of AVIRIS data Bridgwater, D., Mengel, F., Fryer, B., Wagner, P., & Hansen, S. C. (1995). Early Proterozoic mafic dykes in the North Atlantic and Baltic cratons: field setting and chemistry of distinctive dyke swarms. Geological Society, London, Special Publications, 95(1), 193-210. doi: 10.1144/GSL.SP.1995.095.01.12 Bruegge, C. J., Stiegman, A. E., Rainen, R. A., & Springsteen, A. W. (1993). Use of Spectralon as a diffuse reflectance standard for in-flight calibration of earth-orbiting sensors. Optical Engineering, 32(4), 805-814. doi: 10.1117/12.132373 Buckley, S. J., Kurz, T. H., Howell, J. A., & Schneider, D. (2013). Terrestrial lidar and hyperspectral data fusion products for geological outcrop analysis. Computers & Geosciences, 54, 249-258 Budkewitsch, P., Staenz, K., Neville, R., Rencz, A., & Sangster, D. (2000). Spectral signatures of carbonate rocks surrounding the Nanisivik MVT Zn-Pb mine and implications of hyperspectral imaging for exploration in Arctic environments Ore Deposit Workshop: New Ideas for a New Millennium, Technical Volume. Camps-Valls, G., & Bruzzone, L. (2009). Kernel methods for remote sensing data analysis: John Wiley & Sons. Clark, R., Swayze, G., Gallagher, A., Gorelick, N., & Kruse, F. (1991). Mapping with imaging spectrometer data using the complete band shape least-squares algorithm simultaneously fit to multiple spectral features from multiple materials Proceedings of the third airborne visible/infrared imaging spectrometer (AVIRIS) workshop (Vol. 42, pp. 2-3): Jet Propulsion Laboratory Publication. Clark, R. N. (1999). Spectroscopy of rocks and minerals, and principles of spectroscopy. Manual of remote sensing, 3, 3-58 44
Clark, R. N., Swayze, G. A., Livo, K. E., Kokaly, R. F., Sutley, S. J., Dalton, J. B., McDougal, R. R., & Gent, C. A. (2003). Imaging spectroscopy: Earth and planetary remote sensing with the USGS Tetracorder and expert systems. Journal of Geophysical Research: Planets, 108(E12). doi: 10.1029/2002JE001847 Cocks, T., Jenssen, R., Stewart, A., Wilson, I., & Shields, T. (1998). The HyMapTM airborne hyperspectral sensor: the system, calibration and performance Proceedings of the 1st EARSeL workshop on Imaging Spectroscopy (pp. 37-42): EARSeL. Coe, K., & Cheeney, R. (1972). Preliminary results of mapping in Liverpool Land, East Greenland. Rapport Grønlands Geologiske Undersøgelse, 48, 7-20 Connelly, J. N., & Mengel, F. C. (2000). Evolution of Archean components in the Paleoproterozoic Nagssugtoqidian orogen, West Greenland. Geological Society of America Bulletin, 112(5), 747-763. doi: 10.1130/0016-7606(2000)1122.0.CO;2 Crowley, J. K. (1986). Visible and near‐infrared spectra of carbonate rocks: Reflectance variations related to petrographic texture and impurities. Journal of Geophysical Research: Solid Earth, 91(B5), 5001-5012. doi: 10.1029/JB091iB05p05001 Escher, A., Escher, J., & Watterson, J. (1975). The reorientation of the Kangâmiut dike swarm, West Greenland. Canadian Journal of Earth Sciences, 12(2), 158-173. doi: 10.1139/e75016 Feng, J., Rivard, B., Rogge, D., & Sánchez-Azofeifa, A. (2013). The longwave infrared (3–14 μm) spectral properties of rock encrusting lichens based on laboratory spectra and airborne SEBASS imagery. Remote sensing of environment, 131, 173-181 Gaffey, S. J. (1986). Spectral reflectance of carbonate minerals in the visible and near infrared (O. 35-2.55 microns); calcite, aragonite, and dolomite. American Mineralogist, 71(1-2), 151162 Gao, L., Yao, D., Li, Q., Zhuang, L., Zhang, B., & Bioucas-Dias, J. (2017). A New Low-Rank Representation Based Hyperspectral Image Denoising Method for Mineral Mapping. Remote Sensing, 9(11), 1145 Garde, A. A., & Marker, M. (Cartographer). (2010). Geological Map of Greenland, 1:500 000, Søndre Strømfjord – Nuussuaq, Sheet 1. Second edition. Goetz, A., Vane, G., Solomon, J., & Rock, B. (1985). Imaging spectrometry for earth remote sensing. Science, 228(4704), 1147-1152. doi: 10.1126/science.228.4704.1147 Grocott, J., & McCaffrey, K. J. (2017). Basin evolution and destruction in an Early Proterozoic continental margin: the Rinkian fold–thrust belt of central West Greenland. Journal of the Geological Society, 174(3), 453-467. doi: 10.1144/jgs2016-109 Harris, J. R., Rogge, D., Hitchcock, R., Ijewliw, O., & Wright, D. (2005). Mapping lithology in Canada’s Arctic: application of hyperspectral data using the minimum noise fraction transformation and matched filtering. Canadian Journal of Earth Sciences, 42(12), 21732193. doi: 10.1139/e05-064 Henderson, G., & Pulvertaft, T. (1987). Geological map of Greenland, 1: 100 000, Mârmorilik 71 V. 2 Syd, Nûgâtsiaq 71 V. 2 Nord, Pangnertôq 72 V. 2 Syd. Descriptive text, 8 Higgins, A. K., Gilotti, J. A., & Smith, M. P. (2008). The Greenland Caledonides: Evolution of the Northeast Margin of Laurentia: Geological Society of America. Holmes, P., Thomassen, B., & Stelter, S. (2014). Technical Report on Mineral Exploration 2014, Jameson Land and Liverpool Land, Central East Greenland. Hunt, G. R. (1977). Spectral signatures of particulate minerals in the visible and near infrared. Geophysics, 42(3), 501-513. doi: 10.1190/1.1440721 Hunt, G. R., & Salisbury, J. W. (1971). Visible and near infrared spectra of minerals and rocks. II. Carbonates. Modern geology, 2, 23-30 45
Huo, H., Ni, Z., Jiang, X., Zhou, P., & Liu, L. (2014). Mineral Mapping and Ore Prospecting with HyMap Data over Eastern Tien Shan, Xinjiang Uyghur Autonomous Region. Remote Sensing, 6(12), 11829 Jensen, S. M., Hansen, H., Secher, K., Steenfelt, A., Schjøth, F., & Rasmussen, T. M. (2002). Kimberlites and other ultramafic alkaline rocks in the Sisimiut–Kangerlussuaq region, southern West Greenland. Geology of Greenland Survey Bulletin, 191, 57-66 Kayet, N., Pathak, K., Chakrabarty, A., & Sahoo, S. (2018). Mapping the distribution of iron ore minerals and spatial correlation with environmental variables in hilltop mining areas. Environmental Earth Sciences, 77(8), 308 Keshava, N., & Mustard, J. F. (2002). Spectral unmixing. IEEE signal processing magazine, 19(1), 44-57. doi: 10.1109/79.974727 Kokaly, R. F. (2012). Spectroscopic remote sensing for material identification, vegetation characterization, and mapping Algorithms and Technologies for Multispectral, Hyperspectral, and Ultraspectral Imagery XVIII (Vol. 8390, pp. 839014): International Society for Optics and Photonics. Kopačková, V., & Koucká, L. (2014). Mineral mapping based on automatic detection of multiple absorption features. EARSeL eProceedings, 13(S1), 95-99 Kopačková, V., & Koucká, L. (2017). Integration of Absorption Feature Information from Visible to Longwave Infrared Spectral Ranges for Mineral Mapping. Remote Sensing, 9(10), 1006 Korstgård, J. A. (1980). Metamorphism, deformation and chemical evolution of the Kangamiut dykes and their host gneisses across the Ikertoq Complex, the Nagssugtoqidian mobile belt, West Greenland. (PhD), Liverpool University. Kurz, T. H., Buckley, S. J., & Howell, J. A. (2013). Close-range hyperspectral imaging for geological field studies: workflow and methods. International Journal of Remote Sensing, 34(5), 1798-1822. doi: 10.1080/01431161.2012.727039 Kurz, T. H., Dewit, J., Buckley, S. J., Thurmond, J. B., Hunt, D. W., & Swennen, R. (2012). Hyperspectral image analysis of different carbonate lithologies (limestone, karst and hydrothermal dolomites): the Pozalagua Quarry case study (Cantabria, North-west Spain). Sedimentology, 59(2), 623-645. doi: 10.1111/j.1365-3091.2011.01269.x Laakso, K., Rivard, B., Peter, J., White, H., Maloley, M., Harris, J., & Rogge, D. (2015). Application of Airborne, Laboratory, and Field Hyperspectral Methods to Mineral Exploration in the Canadian Arctic: Recognition and Characterization of Volcanogenic Massive Sulfide-Associated Hydrothermal Alteration in the Izok Lake Deposit Area, Nunavut, Canada. Economic Geology, 110(4), 925-941. doi: 10.2113/econgeo.110.4.925 Laakso, K., Rivard, B., & Rogge, D. (2016). Enhanced detection of gossans using hyperspectral data: Example from the Cape Smith Belt of northern Quebec, Canada. ISPRS Journal of Photogrammetry and Remote Sensing, 114, 137-150. doi: 10.1016/j.isprsjprs.2016.02.004 Landgrebe, D. A. (2005). Signal theory methods in multispectral remote sensing (Vol. 29): John Wiley & Sons. Larsen, L. M., & Rex, D. C. (1992). A review of the 2500 Ma span of alkaline-ultramafic, potassic and carbonatitic magmatism in West Greenland. Lithos, 28(3), 367-402. doi: 10.1016/0024-4937(92)90015-Q Lorenz, S., Salehi, S., Kirsch, M., Zimmermann, R., Unger, G., Vest Sørensen, E., & Gloaguen, R. (2018). Radiometric correction and 3d integration of long-range ground-based hyperspectral imagery for mineral exploration of vertical outcrops. Remote Sensing, 10(2), 176. doi: 10.3390/rs10020176 Marion, R., & Carrère, V. (2018). Mineral Mapping Using the Automatized Gaussian Model (AGM)—Application to Two Industrial French Sites at Gardanne and Thann. Remote Sensing, 10(1), 146 46
Mielke, C., Boesche, N. K., Rogass, C., Kaufmann, H., & Gauert, C. (2015). New geometric hull continuum removal algorithm for automatic absorption band detection from spectroscopic data. Remote Sensing Letters, 6(2), 97-105. doi: 10.1080/2150704X.2015.1007246 Mielke, C., Boesche, N. K., Rogass, C., Kaufmann, H., Gauert, C., & de Wit, M. (2014). Spaceborne mine waste mineralogy monitoring in South Africa, applications for modern push-broom missions: hyperion/OLI and EnMAP/Sentinel-2. Remote Sensing, 6(8), 67906816 Mielke, C., Rogass, C., Boesche, N., Segl, K., & Altenberger, U. (2016). EnGeoMAP 2.0— Automated Hyperspectral Mineral Identification for the German EnMAP Space Mission. Remote Sensing, 8(2), 127 Morison, M., Cloutis, E., & Mann, P. (2014). Spectral unmixing of multiple lichen species and underlying substrate. International Journal of Remote Sensing, 35(2), 478-492. doi: 10.1080/01431161.2013.871085 Murphy, R. J., & Monteiro, S. T. (2013). Mapping the distribution of ferric iron minerals on a vertical mine face using derivative analysis of hyperspectral imagery (430–970nm). ISPRS journal of photogrammetry and remote sensing, 75, 29-39. doi: 10.1016/j.isprsjprs.2012.09.014 Murphy, R. J., Monteiro, S. T., & Schneider, S. (2012). Evaluating classification techniques for mapping vertical geology using field-based hyperspectral sensors. IEEE Transactions on Geoscience and Remote Sensing, 50(8), 3066-3080. doi: 10.1109/TGRS.2011.2178419 Nachtegael, M., Kerre, E. E., Philips, W., & Van der Weken, D. (2007). Soft Computing in Image Processing: Recent Advances (Vol. 210): Springer. Nascimento, J. M., & Dias, J. M. (2005). Vertex component analysis: A fast algorithm to unmix hyperspectral data. IEEE transactions on Geoscience and Remote Sensing, 43(4), 898-910. doi: 10.1109/TGRS.2005.844293 Neville, R. (1999). Automatic endmember extraction from hyperspectral data for mineral exploration International Airborne Remote Sensing Conference and Exhibition, 4 th/21 st Canadian Symposium on Remote Sensing, Ottawa, Canada. Notesco, G., Ogen, Y., & Ben-Dor, E. (2016). Integration of Hyperspectral Shortwave and Longwave Infrared Remote-Sensing Data for Mineral Mapping of Makhtesh Ramon in Israel. Remote Sensing, 8(4), 318 Pan, Z., Huang, J., & Wang, F. (2013). Multi range spectral feature fitting for hyperspectral imagery in extracting oilseed rape planting area. International Journal of Applied Earth Observation and Geoinformation, 25, 21-29. doi: 10.1016/j.jag.2013.03.002 Petzold, D. E., & Goward, S. N. (1988). Reflectance spectra of subarctic lichens. Remote Sensing of Environment, 24(3), 481-492. doi: 10.1016/0034-4257(88)90020-X Ramberg, H. (1948). Titanic iron ore formed by dissociation of silicates in granulite facies [Greenland]. Economic Geology, 43(7), 553-570. doi: 10.2113/gsecongeo.43.7.553 Ramberg, H. (1949). On the petrogenesis of the gneiss complexes between Sukkertoppen and Christianshaab, West Greenland. Meddelelser fra Dansk Geologisk Forening, 11, 312-327 Rees, W. G., Tutubalina, O. V., & Golubeva, E. I. (2004). Reflectance spectra of subarctic lichens between 400 and 2400 nm. Remote Sensing of Environment, 90(3), 281-292. doi: 10.1016/j.rse.2003.12.009 Rivard, B., & Arvidson, R. E. (1992). Utility of imaging spectrometry for lithologic mapping in Greenland. Photogrammetric engineering and remote sensing, 58(7), 945-949 Roberts, D. A., Gardner, M., Church, R., Ustin, S., Scheer, G., & Green, R. (1998). Mapping chaparral in the Santa Monica Mountains using multiple endmember spectral mixture models. Remote Sensing of Environment, 65(3), 267-279. doi: 10.1016/S00344257(98)00037-6 47
Rodger, A., Laukamp, C., Haest, M., & Cudahy, T. (2012). A simple quadratic method of absorption feature wavelength estimation in continuum removed spectra. Remote sensing of environment, 118, 273-283. doi: 10.1016/j.rse.2011.11.025 Rogass, C., Segl, K., Mielke, C., Fuchs, Y., & Kaufmann, H. (2014). EnGeoMap—A geological mapping tool applied to the EnMAP mission. EARSeL eProceedings(1), 12-17. doi: 10.12760/01-2013-2-02 Rogge, D., Bachmann, M., Rivard, B., & Feng, J. (2012). Spatial Sub-Sampling Using Local Endmembers for Adapting OSP and SSEE for Large-Scale Hyperspectral Surveys. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 5(1), 183195. doi: 10.1109/JSTARS.2011.2168513 Rogge, D., Rivard, B., Segl, K., Grant, B., & Feng, J. (2014). Mapping of NiCu–PGE ore hosting ultramafic rocks using airborne and simulated EnMAP hyperspectral imagery, Nunavik, Canada. Remote sensing of environment, 152, 302-317 Rogge, D., Rivard, B., Zhang, J., & Feng, J. (2006). Iterative Spectral Unmixing for Optimizing Per-Pixel Endmember Sets. IEEE Transactions on Geoscience and Remote Sensing, 44(12), 3725-3736. doi: 10.1109/TGRS.2006.881123 Rogge, D., Rivard, B., Zhang, J., Sanchez, A., Harris, J., & Feng, J. (2007). Integration of spatial– spectral information for the improved extraction of endmembers. Remote Sensing of Environment, 110(3), 287-303. doi: 10.1016/j.rse.2007.02.019 Rollin, E. M., Milton, E. J., & Roche, P. (1994). The influence of weathering and lichen cover on the reflectance spectra of granitic rocks. Remote Sensing of Environment, 50(2), 194-199. doi: 10.1016/0034-4257(94)90045-0 Rosa, D., Dewolfe, M., Guarnieri, P., Kolb, J., LaFlamme, C., Partin, C., Salehi, S., Vest Sørensen, E., Thaarup, S., Thrane, K., & Zimmermann, R. (2017). Architecture and mineral potential of the Paleoproterozoic Karrat Group, West Greenland. Results of the 2016 Season (Report No. 5). Geological Survey of Denmark and Greenland. Rosa, D., Guarnieri, P., Hollis, J., Kolb, J., Partin, C., Petersen, J., Vest Sørensen, E., Thomassen, B., Thomsen, L., & Thrane, K. (2016). Architecture and mineral potential of the Paleoproterozoic Karrat Group, West Greenland. Results of the 2015 season Geological Survey of Denmark and Greenland. Sabine, C. (1999). Remote sensing strategies for mineral exploration. Remote Sensing for the Earth Sciences-Manual of Remote Sensing, American Society of Photogrammetry and Remote Sensing, 3, 375-447 Salehi, S., Karami, M., & Fensholt, R. (2016). Identification of a robust lichen index for the deconvolution of lichen and rock mixtures using pattern search algorithm (case study: Greenland). International Archives of the Photogrammetry, Remote Sensing & Spatial Information Sciences, XLI-B7, 973-979. doi: 10.5194/isprs-archives-XLI-B7-973-2016 Salehi, S., Lorenz, S., Vest Sørensen, E., Zimmermann, R., Fensholt, R., Henning Heincke, B., Kirsch, M., & Gloaguen, R. (2018). Integration of Vessel-Based Hyperspectral Scanning and 3D-Photogrammetry for Mobile Mapping of Steep Coastal Cliffs in the Arctic. Remote Sensing, 10(2), 175. doi: 10.3390/rs10020175 Salehi, S., Rogge, D., Rivard, B., Heincke, B. H., & Fensholt, R. (2017). Modeling and assessment of wavelength displacements of characteristic absorption features of common rock forming minerals encrusted by lichens. Remote Sensing of Environment, 199, 78-92. doi: 10.1016/j.rse.2017.06.044 Salehi, S., & Thaarup, S. (2018). Mineral mapping by hyperspectral remote sensing in West Greenland using airborne, ship-based and terrestrial platforms. Geological Survey of Denmark and Greenland Bulletin 41(in press) 48
Satterwhite, M. B., Henley, J. P., & Carney, J. M. (1985). Effects of lichens on the reflectance spectra of granitic rock surfaces. Remote Sensing of Environment, 18(2), 105-112. doi: 10.1016/0034-4257(85)90041-0 Schläpfer, D., & Richter, R. (2002). Geo-atmospheric processing of airborne imaging spectrometry data. Part 1: parametric orthorectification. International Journal of Remote Sensing, 23(13), 2609-2630. doi: 10.1080/01431160110115825 Secher, K., & Larsen, L. M. (1980). Geology and mineralogy of the Sarfartoq carbonatite complex, southern West Greenland. Lithos, 13(2), 199-212 Shanmugam, S., & SrinivasaPerumal, P. (2014). Spectral matching approaches in hyperspectral image processing. International Journal of Remote Sensing, 35(24), 8217-8251. doi: 10.1080/01431161.2014.980922 Singer, R. B., & McCord, T. B. (1979). Mars-Large scale mixing of bright and dark surface materials and implications for analysis of spectral reflectance Lunar and Planetary Science Conference Proceedings (Vol. 10, pp. 1835-1848). Sørensen, E. V. (2011). Implementation of digital Multi-Model Photogrammetry for building of 3D-models and interpretation of the geological and tectonic evolution of the Nuussuaq Basin. PhD thesis, University of Copenhagen. Sørensen, E. V., Bjerager, M., & Citterio, M. (2015a). Digital models based on images taken with handheld cameras–examples on land, from the sea and on ice. Geological Survey of Denmark and Greenland Bulletin(33), 73-76 Sørensen, E. V., Pedersen, A. K., García-Sellés, D., & Strunck, M. N. (2015b). Point clouds from oblique stereo-imagery: Two outcrop case studies across scales and accessibility. European Journal of Remote Sensing, 48, 593-614. doi: 10.5721/EuJRS20154833 Svennevig, K. (2014). Photogrammetrical 3D-mapping of Kilen–a complex corner of eastern North Greenland. Nor. Polarinstitut Rapp, 146, 36-37 Svennevig, K., & Guarnieri, P. (2012). From 3D mapping to 3D modelling: a case study from the Skaergaard intrusion, southern East Greenland. Geological Survey of Denmark and Greenland Bulletin, 26, 57-60 Svennevig, K., Guarnieri, P., & Stemmerik, L. (2015). From oblique photogrammetry to a 3D model–Structural modeling of Kilen, eastern North Greenland. Computers & Geosciences, 83, 120-126. doi: 10.1127/1432-8364/2012/0130 Théau, J., Peddle, D., & Duguay, C. (2005). Mapping lichen in a caribou habitat of Northern Quebec, Canada, using an enhancement_classification method and spectral mixture analysis. Remote sensing of Environment, 94(2), 232-243. doi: 10.1016/j.rse.2004.10.008 Tompkins, S., Mustard, J. F., Pieters, C. M., & Forsyth, D. W. (1997). Optimization of endmembers for spectral mixture analysis. Remote Sensing of Environment, 59(3), 472489. doi: 10.1016/S0034-4257(96)00122-8 Tukiainen, T., & Thorning, L. (2005). Detection of kimberlitic rocks in West Greenland using airborne hyperspectral data: the HyperGreen 2002 project. Geological Survey of Denmark and Greenland Bulletin, 7, 69-72 Van der Meer, F. (1995). Spectral reflectance of carbonate mineral mixtures and bidirectional reflectance theory: Quantitative analysis techniques for application in remote sensing. Remote Sensing Reviews, 13(1-2), 67-94. doi: 10.1080/02757259509532297 Van der Meer, F. (1996). Spectral mixture modelling and spectral stratigraphy in carbonate lithofacies mapping. ISPRS Journal of Photogrammetry and Remote Sensing, 51(3), 150162. doi: 10.1016/0924-2716(95)00009-7 Van Der Meer, F. (2004). Analysis of spectral absorption features in hyperspectral imagery. International journal of applied earth observation and geoinformation, 5(1), 55-68. doi: 10.1016/j.jag.2003.09.001 49
Van der Meer, F., Kopačková, V., Koucká, L., Van der Werff, H. M., Van Ruitenbeek, F. J., & Bakker, W. H. (2018). Wavelength feature mapping as a proxy to mineral chemistry for investigating geologic systems: An example from the Rodalquilar epithermal system. International journal of applied earth observation and geoinformation, 64, 237-248. doi: 10.1016/j.jag.2017.09.008 Van der Meer, F. D., & De Jong, S. M. (2011). Imaging spectrometry: basic principles and prospective applications (Vol. 4): Springer Science & Business Media. Van Ruitenbeek, F. J., Bakker, W. H., van der Werff, H. M., Zegers, T. E., Oosthoek, J. H., Omer, Z. A., Marsh, S. H., & Van der Meer, F. D. (2014). Mapping the wavelength position of deepest absorption features to explore mineral diversity in hyperspectral images. Planetary and space science, 101, 108-117. doi: 10.1016/j.pss.2014.06.009 Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004). Image quality assessment: from error visibility to structural similarity. IEEE transactions on image processing, 13(4), 600612. doi: 10.1109/TIP.2003.819861 Zhang, J., Rivard, B., & Sanchez-Azofeifa, A. (2004). Derivative spectral unmixing of hyperspectral data applied to mixtures of lichen and rock. IEEE Transactions on Geoscience and Remote Sensing, 42(9), 1934-1940. doi: 10.1109/TGRS.2004.832239 Zhang, J., Rivard, B., & Sánchez-Azofeifa, A. (2005). Spectral unmixing of normalized reflectance data for the deconvolution of lichen and rock mixtures. Remote Sensing of Environment, 95(1), 57-66. doi: 10.1016/j.rse.2004.11.019
50
8 RESEARCH PAPERS
51
PAPER I
52
Remote Sensing of Environment 199 (2017) 78–92
Contents lists available at ScienceDirect
Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse
Modeling and assessment of wavelength displacements of characteristic absorption features of common rock forming minerals encrusted by lichens Sara Salehi a,b,⁎, Derek Rogge c, Benoit Rivard d, Bjørn Henning Heincke a, Rasmus Fensholt b a
Department of Petrology and Economic Geology, Geological Survey of Denmark and Greenland, Copenhagen, Denmark Department of Geosciences and Natural Resource Management, University Of Copenhagen, Copenhagen, Denmark c German Remote Sensing Data Centre, DLR, Munchnerstr. 20, D-82234, Germany d Department of Earth and Atmospheric Sciences, University of Alberta, Edmonton T6G 2E3, Canada b
a r t i c l e
i n f o
Article history: Received 17 January 2017 Received in revised form 12 June 2017 Accepted 29 June 2017 Available online xxxx Keywords: Greenland Hyperspectral remote sensing Lichen cover Mineral exploration Spectral mapping
a b s t r a c t Arctic environments provide a challenging ground for geological mapping and mineral exploration. Inaccessibility complicates ground surveys and the presence of ice, vegetation, and lichens hinders supportive remote sensing surveys. Spectral mixing of lichens and bare rock can shift the wavelength position of characteristic absorption features, thereby complicating the spectral mapping of minerals and lithologies. The extent to which diagnostic rock features are preserved despite the presence of lichens is of major concern in remotely sensed geological studies and estimates of the critical level of lichen coverage, below which spectral features of the mineral substrate can still be identified, are needed. We investigated how surficial lichen cover affects the characteristics of shortwave infrared (SWIR) mineral absorption features and the efficacy of automated absorption feature extraction. For this purpose, mixed spectra were synthetically generated from laboratory spectra of common rock forming SWIR absorbing minerals and lichens. Wavelength displacements of characteristic absorption features for each mixed spectrum were then analyzed as a function of percentage lichen cover. Distinctive trends were identified that can be used in future analysis: The strong spectral features of mica group minerals around 2200 and 2340–2350 nm maintain their integrity for up to 30% lichen cover, despite the related shift toward shorter wavelengths for higher percentage lichen cover. In contrast, very weak absorption bands around 2440 nm in (white) micas spectra are completely obscured for a lichen cover of ≥50%. Our observations suggest that the chlorite feature around 2250 nm is shifted toward longer wavelengths and the depth of this feature as well as the contrast between lichen and substrate spectra define the amount of lichen needed to mask it. Furthermore, lichens induce a spectral shift towards shorter wavelengths for the features around 2320 nm for the rocks containing amphibole, chlorite, carbonate and serpentine group minerals. In addition, no wavelength displacement is observed for chlorite, biotite and phlogopite features around 2380 nm in mixtures with lichens. By quantifying lichen cover effects on mineral absorption features, our study highlights the importance of being precautious in any interpretation in areas characterized by abundant lichen-covered outcrops. This can be of significant importance for mineral and deposit vectoring as the presence of abundant lichen coverage causing slightly shifted features for a given spectra can be erroneously identified as a path to a deposit. © 2017 Elsevier Inc. All rights reserved.
1. Introduction Hyperspectral systems are being increasingly used for geological mapping at a subpixel scale through spectral mixture analyses (SMA) (Smith et al., 1990). Large aerial coverage and relatively quick map
⁎ Corresponding author at: Department of Petrology and Economic Geology, Geological Survey of Denmark and Greenland, Copenhagen, Denmark. E-mail addresses:
[email protected] (S. Salehi),
[email protected] (D. Rogge),
[email protected] (B. Rivard),
[email protected] (B.H. Heincke),
[email protected] (R. Fensholt).
http://dx.doi.org/10.1016/j.rse.2017.06.044 0034-4257/© 2017 Elsevier Inc. All rights reserved.
production are factors, which make this type of analysis particularly attractive in areas of difficult access, such as large parts of the subarctic and arctic (Budkewitsch et al., 2000; Gou et al., 2015; Harris et al., 2001; Harris et al., 2005; Rogge et al., 2009; Rogge et al., 2014; Schetselaar and deKemp, 2000; Schetselaar and Ryan, 2009; Staenz et al., 2000). Expansion of mapping is however needed in remote places (e.g. in Greenland), where a large proportion of the area is still unexplored and the lack of infrastructure reduce the capabilities to economically explore and locate mineral resources using traditional techniques. The discrimination and classification of lithological units using hyperspectral imagery primarily makes use of SMA based on a linear
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
combination of pure reflectance spectra referred to as spectral endmembers (Singer and McCord, 1979). However, the mixture of signatures from various components within individual pixels can often mask the diagnostic spectral features thereby potentially complicating spectral unmixing and the following classification. A pervasive example of such mixing of signatures in high latitude and subarctic environments is the presence of abundant lichen growing on rock outcrops. Lichens are symbiotic organisms particularly well adapted to extreme environmental conditions, which impede the identification of rocks/minerals in two ways: (1) Lichens prevent transmission of light to the rock substrate and thereby mask the mineral features, such that no useful information can be drawn from the spectra (Ager and Milton, 1987; Bechtel et al., 2002). (2) Lichens grow on rocks in a non-random way and cause mixing of features spectrally. These mixtures can cause shifts in diagnostic absorption features of minerals and may hinder the ability to detect the spectral properties of rocks (Bechtel et al., 2002; Rivard and Arvidson, 1992). A number of studies have investigated the spectra mixing of lichens and their rock substrate and the associated challenges in identification of mineralogy based on interpretation of the spectra (Ager and Milton, 1987; Bechtel et al., 2002; Laakso et al., 2015; Laakso et al., 2016). The underlying principle is that a given spectrum is a linear mixture of the representative signatures from the constituents of each pixel (Kruse et al., 1993). This assumption is reliable for mixtures of lichens and rocks as b 3% of light is transmitted through lichens to the substrate throughout the 350–2500 nm region (Ager and Milton, 1987; Bechtel et al., 2002). Bechtel et al. (2002) examined variations in lichen spectra in relation to color, type and lichen species and proposed a set of lichen spectral indices to guide the selection of a single lichen end-member for decomposition of the rock and lichen spectral mixtures by SMA. While efficacy of hyperspectral methods for mineral exploration purposes has been the topic of various studies, less research has been devoted to studies of arctic and subarctic lichens (Harris et al., 2005; Staenz et al., 2000). A study was performed by Satterwhite et al. (1985) to determine the spectral characteristics of granitic rock encrusting lichens. However, the measurements were confined to the 400–1100 nm spectral region, which excludes the SWIR part of the electromagnetic spectrum that is relevant for the analysis of geological targets and is exploited by many hyperspectral imaging systems (Budkewitsch et al., 2000; Harris et al., 2005; Kruse et al., 2012). Rivard and Arvidson (1992) studied the variability in spectral features as based on a limited number of in situ spectral measurements (450 to 2400 nm) for a group of lichen bearing rocks, yet without identification of the species of lichens. Rollin et al. (1994) focused on the influence of weathering and lichen cover on the reflectance spectra of granitic rocks over visible and infrared wavelengths and suggested that absorption features in the SWIR region are potentially useful to generate indices applicable to detect lichens in airborne/ spaceborne hyperspectral imagery. A recent study by Li et al. (2015) revealed multiple lichen related absorption and reflection features and a set of spectral indices were developed (covering the wavelengths of 680–1320 nm, 1660–1725 nm, and 2230–2300 nm) to estimate lichen coverage in remotely sensed imagery and field spectroscopy measurements. Several gaps in current knowledge about the effect of lichens with respect to geological applications however remain. Mineral compositional variations in an alteration system appear to be strongly correlated with wavelength shifts of diagnostic absorption features (e.g. chlorites, carbonates and sericite). However, organic growth such as lichens can also induce spectral shifts in the wavelength position of the same diagnostic absorption features within the SWIR range and can cause misclassification and false positives if spectral feature fitting methodologies are used for mapping. This can be of significant importance for mineral and deposit vectoring as the presence of abundant lichen coverage causing slightly shifted features for a given spectrum can be erroneously identified as a path to a deposit. Laakso et al. (2016)
79
demonstrated that the apparent spectral shift of the iron feature toward shorter wavelengths, caused by rock-encrusting lichens, might impede classification of gossans based on the iron oxide mineralogy. However, the study did not investigate the impact of lichen cover on spectral recognition of common rock forming mineral features in the SWIR being vital for accurate interpretation of spectra. In addition, previous studies suggest that the abundance of lichens in a mixture can be estimated reasonably well and with an acceptable error level (Théau et al., 2005; Zhang et al., 2004; Zhang et al., 2005), but a rigorous examination of the effect and scale of wavelength displacements of mineral features as a result of mixing with different lichen abundances is yet to be conducted. We focus here on the SWIR region (2100–2500 nm) and investigate (a) the scale of wavelength displacement induced by different percentage of lichen cover and (b) the extent to which characteristic rock features are preserved despite rock encrusting lichens. This allows us to quantify the problem of displacement and to establish strategies of how to adapt data analysis to minimize this effect. The ASD spectral measurements of a real set of common rock types have been used as a representation of the lichen-free weathered surfaces and lichen signatures have been measured from the rocks themselves. This sample set, including mineral mixtures as well as lichen mixtures, is expected to represent a more realistic case to analyze the trends in shifts for specific minerals for real rocks as compared to using mineral spectra from the USGS library.
2. Study area and geology Arctic regions such as Greenland are ideal for remote sensing studies due to extensive areas of well-exposed rocks and low vegetation coverage. Greenland is characterized by about 20% of ice-free surface area, dominated by crystalline rocks of the Precambrian shield. This ice-free zone generally consists of very well exposed rocks that, to a variable extent, are covered by the crusts of lichens. For the purpose of this study, samples that comprise common and economically important rock types were investigated, originating from three tracts at different parts of Greenland, namely Liverpool Land (Central East), Karrat (Central West) and SisimiutKangerlussuaq (South West) (Fig. 1). The Karrat region in West Greenland comprises three main formations: the Mârmorilik (dominated by carbonates), Qeqertarssuaq (mainly siliciclastic and comprises hornblende schist, amphibolite and commonly minor carbonate), and Nukavsak (dominated by dark colored alternating pelitic and semipelitic schists) (Escher and Watt, 1976). The Sisimiut- Kangerlussuaq region is located approximately 500 km to the south of Karrat. This region is characterized by metamorphic terrains of both Archaean and Palaeoproterozoic ages with metamorphic grade varying from low amphibolite facies to granulite facies towards the north. The area hosts an alkaline province and a variety of ultramafic alkaline rocks, including swarms of kimberlite and lamproite dykes (Jensen et al., 2002; Larsen and Rex, 1992). The crystalline complex of North Liverpool Land in Central East Greenland is characterized by Precambrian, marble bearing metamorphic rocks and granitesquartz-monzonites of Caledonian or Neoproterozoic age (Coe and Cheeney, 1972). There is a rich lichen flora in all three areas, and the lichens show fairly high cover percentage (up to 95% of the ground cover) in many places (Graham, 1999). Heaths and dry cyperaceous communities, snow-patch vegetation, grassy meadows and swamps are the typical features of Liverpool Land. The Karrat region is particularly characterized by willow scrub, meadows, various heaths and steppe-like communities rich in xerophilous lichens. The Sisimiut region is characterized by alder scrub, tall luxuriant willow scrub and freshwater vegetation. The local temperature varies from 2° to 10 °C in July for Liverpool Land, Karrat and Sisimiut, respectively.
80
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
Fig. 1. Geology map of Greenland and the locations of the collected rock samples. a) Karrat region: quartzite [3 samples], b) Liverpool Land region: monzonite [3 samples], c) SisimiutKangerlussuaq region: kimberlite [9 samples], lamproite [4 samples], gneiss [3 samples], kersantite [3 samples], fenite [3 samples], granite [5 samples], carbonatite [4 samples].
3. Materials and methods
3.2. X-ray diffraction
3.1. Rock sample collection
Laboratory X-ray powder diffraction (XRD) of samples was undertaken to determine their mineralogical composition and related percentage of certain minerals. These samples were chosen because their spectra have either AlOH, FeOH, MgOH, and/or CO3 absorption features, which indicates the presence of white mica, biotite, chlorite, serpentine, carbonates, sericite and/or amphiboles. According to the results of XRD analysis, mica group minerals are present in all the samples except for carbonatite and lamproite. Amphiboles are present in lamproite and monzonite samples. Serpentine group minerals are found in kimberlite samples. Chlorite is present in our granite and gneiss samples, while carbonate group minerals are the typical mineralogy of carbonatite samples. The presence of different minerals in the rock samples derived from the XRD analysis is summarized in Table 1.
Geological Survey of Denmark and Greenland (GEUS) provided the collection of lichen bearing rock samples used in this study. The chosen hand specimen samples were selected from each of rock outcrops visited during fieldwork in summer 2005, which were then stored in closed rock storage cabinets after transportation in cloth sample bags. The samples encompass a broad range of common and important economic rock types. All samples collected have at least one weathered surface with varying degree of lichen cover to represent a combination of rock and biological material that one expects to see within any acquired imagery from such areas (Fig. 2).
Fig. 2. Examples of weathered surface of (a) gneiss and (b) monzonite samples covered by lichens, red arrows indicate examples of lichen patches while blue arrows indicate amphibole group minerals. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
Table 1 Minerals identified by XRD and the wavelength positions of the associated absorption features. Minerals
Mica
Phlogopite
Rock types Kersantite
Wavelength of diagnostic features Monzonite
Lamproite
Kimberlite
Granite
Fenite
Carbonatite
Quartzite
Gneiss
2200
2254
2320–2325
X Muscovite
X X X X X
Amphibole Serpentine Chlorite
Biotite Richterite Hornblende Antigorite Chlorite
Quartz
Dolomite Albite Albite Microcline Orthoclase Microcline + plagioclase Quartz
X X X X X
2380–2390
2438–2450 X
X
X X
X X
X
X
X
X X X
X X X X
Carbonate Feldspar
2340–2350
X
X X
X X
No feature
X
No feature
X X X X X X
X X
X X
X X X
X
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
Mineral group
X Pyroxene
Augite
X
No feature
X Olivine FeO
Forsterite Ilmenite Hematite
X X
No feature No feature
X
81
82
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
3.3. Laboratory spectral measurements
3.4. Lichen spectral characteristics
All spectral measurements were conducted in a laboratory setting using a FieldSpec®3 HiRes (ASD) Spectroradiometer equipped with a contact probe. The contact probe has an internal illumination source that provides consistent illumination and viewing conditions and ensures measuring the pure spectrum of representative lichen species and rock samples. The measured surface is approximately 1 cm determined by the diameter of the contact probe. The ASD spectrometer records 2151 bands across the 350–2500 nm spectral range with spectral resolution of 3 nm for the VNIR, 8.5 nm for the SWIR1, and 6.5 nm for the SWIR2 and a sampling interval of 1.4 nm between 350 and 1050 nm, and 2 nm between 1000 and 2500 nm. Each reading consists of 25 individual scans taken consecutively and averaged by ASD, which is then converted to reflectance using a SRT-99-050 white reference panel (Bruegge et al., 1993). Two types of datasets were collected with the ASD instrument: a) spectra from lichen-free weathered surfaces of each rock sample and b) spectra from lichens on weathered surfaces. The first dataset consists of five lichen free, weathered surfaces for each sample. The second dataset comprises lichens of different colors to account for possible inter-species spectral variations. The spectral measurements were finally averaged to create a single spectral representation of each lichen and the related rock substrate (Fig. 3). The spectral range between 2000 and 2500 nm was selected because the majority of rock forming minerals associated with the primary lithologies and various alteration styles show key absorption features within this range. These featured are mainly related to vibrational transitions associated with hydroxyl (OH) and water (H2O) and bending and stretching of the bonds between AlOH, FeOH, MgOH and carbonate (CO3) (Pontual et al., 1997). The detection and identification of these features have been the focus of hyperspectral data analysis for several geological applications and mineral exploration techniques. In addition, the identical shape of characteristic features of lichens within SWIR range acts as a background; thus, mineral features are superimposed on lichen spectra within this range (Ager and Milton, 1987; Clark, 1999; Hunt, 1977; Van der Meer and De Jong, 2011). In geological mapping or real exploration cases, the target consists of rocks rather than individual minerals. The design of analysis therefore focused on using a real set of common rock types with different key minerals, instead of mixing lichens with mineral spectra from the USGS library. Therefore, the sample set used in this study is expected to represent a more realistic case highlighting the trends of spectral shifts for specific minerals in mixtures with lichens.
Dark color lichens are characterized by low reflectance throughout the visible part of the spectrum and display a gradual increase in reflectance reaching a peak at approximately 1860 nm. This spectral signature is markedly different from the spectral response of vascular plants, being characterized by an abrupt increase in reflectance from the visible to the near infrared, followed by gradual decrease from 800 to 1300 nm. The absorption feature around 1445 nm (Fig. 4) is governed by the water content of the lichens (Bechtel et al., 2002) whereas three broad absorption features located around 1730, 2100, and 2300 nm are associated with the presence of cellulose in lichens (Ager and Milton, 1987; Bechtel et al., 2002; Rees et al., 2004). A broad feature around 2300 nm is followed by another feature near 2350 nm (Bechtel et al., 2002; Rees et al., 2004). The latter is mainly related to muscovite, chlorite or biotite bearing rock samples and therefore cannot be uniquely assigned to lichens. A similar minimum around 2345–2375 nm was reported by Bechtel et al. (2002) who suggested that it might be caused by the rock substrate (quartzite) rather than by lichens. The shape of lichens spectra can be highly variable in the visible region owing to the type of fungal components and the associated differences in pigmentations (Hale, 1967; Petzold and Goward, 1988) (Fig. 4a). However, lichen features in the SWIR region are controlled by lignin, which are similar for most foliose and crustose lichens species. Thus, aside from slight spectral differences in intensity (amplitude) of the subtle absorption features among species, there is a strong similarity in overall shape of the spectra in the SWIR (Bechtel et al., 2002; Zhang et al., 2005). These observations further support the ability of using a single lichen end-member for deconvolution of foliose and crustose lichen/ rock mixtures within this spectral range (Bechtel et al., 2002; Zhang et al., 2005). In this study, all lichen species under consideration were of the crustose type. Slight differences observed in our measurements between lichens in the SWIR (Fig. 4b) are largely explained by uncertainties about whether or not the entire field of view of the sensor has been covered with lichen thallus. For example, the spectra for crustose lichens may still show weak features that are related to light transmission to minerals underlying the lichens in microscopic scale. Micro-topographic variations can also influence the magnitude of the reflectance. 3.5. Rock samples spectral characteristics Diagnostic absorption features related to the common rock forming minerals (amphibole, carbonate, chlorite, mica and serpentine group
Fig. 3. Flowchart of the analysis design and methods applied presented successional from left to the right. Average spectra for the lichen (Śl) and the lichen-free weathered surface (Sr). Srb is the reflectance spectrum of rock type r at band b, Ślb is the reflectance spectrum of lichen type l at band b, and f is the relative proportion of lichen (see Section 3.6.1).
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
83
Fig. 4. Spectral characteristics of all lichens in the study measured with a laboratory spectrometer in (a) full spectral range and (b) SWIR range for the A = kersantite, B = monzonite, C = lamproite, D = kimberlite, E = granite, F = fenite, G = carbonatite, H = quartzite, I = gneiss samples.
minerals) were distinguished and used for further analysis (Fig. 5). The objective of this analysis is to compare the spectral features of samples of known mineralogy to spectra of related minerals selected from USGS library. The molecules AlOH, FeOH, MgOH and CO3 are identified as main components in phyllosilicates (e.g. clays, chlorite and serpentine minerals), hydroxylated silicates (e.g. amphiboles) and carbonates (e.g. calcite, dolomite). The hydroxyl anion is responsible for majority of the diagnostic absorptions in the SWIR mineral spectra related to the vibrations from the AlOH (~2160 to 2220 nm), FeOH (~2230 to 2295 nm) and MgOH (~2300 to 2360 nm) (Fig. 5a). The features in phlogopite spectrum (kersantite and kimberlite samples) are associated with OH vibrational absorptions at 2380 and 2438 nm (A and D in Fig. 5 & Table 1). The trioctahedral structure of these micas causes the domination of MgOH bending modes over AlOH. Thus, these features are displaced from the location of the absorption bands in the other dioctahedral micas. The weak features around 2200 nm could be an indication of some AlOH bending mode, although these features could be associated with more common OH-lattice combinations. Reflectance spectra of muscovite (monzonite, granite, fenite, quartzite and gneiss samples) display intense AlOH absorption feature near 2200 nm and secondary features near 2350 nm and 2450 nm (B,E,F,H and I in Fig. 5, & Table 1) (Hunt and Ashley, 1979). Important features associated with biotite in gneiss samples are the metal-OH vibrational absorptions at 2340 and 2390 nm (I in Fig. 5 & Table 1). Comparison between the wavelength positions obtained from our laboratory spectroscopic measurements on the lamproite and monzonite samples and USGS library indicates that absorption features around 2320 and 2380 nm are related to MgOH in amphiboles (B and C in Fig. 5 & Table 1). The serpentine mineral in our kimberlite sample is antigorite, which shows a MgOH feature within the range of 2320 to 2325 nm (D in Fig. 5 & Table 1). The SWIR region from chlorite spectra (granite and gneiss samples) often displays a FeOH absorption feature near 2254 nm and two MgOH features near 2320 nm and 2380 nm (E
and I in Fig. 5 & Table 1). The Mg and Fe content in chlorite composition define the strength and position of these features (King and Clark, 1989). Based on the results of XRD analysis our carbonatite sample is dolomite-rich and shows characteristic dolomite absorption features with the main CO3 absorption centered around 2324 nm (G in Fig. 5 & Table 1). The pure and unaltered feldspar and quartz do not show diagnostic absorption features within the range of 350–2500 nm. Furthermore, the characteristic spectral features of pyroxene, olivine and iron oxide group minerals are located within the visible-near infrared part of the spectrum in which lichens have different spectral characteristics caused by different colors and species. Accordingly, these five mineral groups are not assessed in this study, as it is beyond the scope of this paper.
3.6. Spectral analysis methodology 3.6.1. Synthetic linear mixtures of lichen and rock spectra Synthetic spectral mixtures of lichen-covered lithologies were analyzed in order to examine and quantify the effects of sub-pixel lichen cover on the spectra of the lithologies, and subsequently to assess the possible impacts on the mapping based on the remotely sensed data. ASD spectral measurements of the pure spectra obtained from reference lithologies and those obtained from lichen-covered surfaces were averaged separately, to create a single spectral representation for the lichen (Śl) and the lichen-free weathered surface (Sr), respectively. These averaged spectra were then used for each substrate to generate linear spectral mixtures with 10% intervals (Kruse et al., 1993; Salehi et al., 2016) (Fig. 3): Rb ¼ f Srb þ ð1− f Þ Slb
ð1Þ
84
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
where Srb = reflectance spectrum of rock type r at band b Slb ¼ reflectance spectrum of lichen type l at band b f ¼ relative proportion of lichen Subject to: f ∈f0:1; 0:2; 0:3; …; 0:9; 1g
ð2Þ
We assume equal mixing on the minerals, i.e. we do not account for preferential growth of lichens on specific minerals. 3.6.2. Processing of the spectra to extract characteristics of absorption features The hull quotient-corrected spectra are extracted for each diagnostic absorption feature to discern the spectral shifts associated with different lichen abundances. Hull quotient correction is used to reduce the effects of the background spectral slope when the absorption feature wavelength is to be accurately recorded (Clark and Roush, 1984). In practice, the hull quotient correction (continuum removal) is conducted by fitting straight-line segments (convex hull points) over the shoulders (maxima) of an absorption feature and dividing the reflectance values of the absorption feature by these convex hull points. “Shoulders” for the given mineral absorption feature are kept constant and do not change with the mixing. The resulting hull quotient values are normalized to a 0 to 1 scale to remove the effects of albedo variance in the spectrum. Here, we used an automatic peak detection approach from the MATLAB's signal processing toolbox (Ingle and Proakis, 2016) called “findpeaks” to calculate the local maxima (peaks) (Fig. 6a). Since the
focus is in extraction of the minima (absorption features), the inverted reflectance values (after Hull quotient correction) of the lichen and rock substrate are used to directly extract characteristics of absorption features, (Fig. 6a). Next, we applied a two-step procedure to measure the depth of each individual absorption feature. In the first step a horizontal line is extended from the peak (“1” in Fig. 6a) to the left and right until one of the following criteria is satisfied: The line a) intersects the signal because there is a higher peak or b) reaches the left or right end of the signal. The second step finds the minimum of the signal in each of the two intervals defined in Step 1. The higher value of the two interval minima specifies the reference level. The difference of the peak and reference level is considered as the prominence (depth) of the feature. In Fig. 6a the reference level is shown as a dashed line connecting the two interval minima (named “a” and “b”). These steps were performed for all spectral mixtures of lichen and rock until the algorithm reached the termination condition (i.e. depth b 0.0002). In other words, the peaks (in -reflectance mixtures), with depth smaller than this value (0.02%) are assumed to be completely flattened and masked away by lichens effect. All significant AlOH, FeOH, MgOH and CO3 related features present in the SWIR range (see Section 3.4) were assessed to further study the wavelength displacement and the scale of displacement of these mineral features as a function of the scale of mixing with lichens. Wavelength displacement of 20 nm is used here as a threshold to account for mineral compositional variations in an alteration system (e.g. chlorites, carbonates and sericite). This assumption is based on the findings of previous studies that considered wavelength displacements as pathfinders for deposits (Duke, 1994; McLeod et al., 1987; Post and Noble, 1993; Scott and Yang, 1997) and observed that the scales of these shifts are often below our chosen threshold. This means that mineral features with larger displacements might not be reliable assigned to the corresponding
Fig. 5. Averaged spectral reflectance curves for the collected rock types. In a) the full spectra range are shown for kersantite (A), monzonite (B), lamproite (C), kimberlite (D), granite (E), fenite (F), carbonatite (G), quartzite (H) and gneiss (I). In b) only the SWIR range are shown and characteristic features are indicated that are related to the mica group minerals (muscovite: Ms., phlogopite: Phl, biotite: Bt), amphibole group minerals (hornblende: Hbl, richterite: Rit), serpentine group minerals (antigorite: Atg), chlorite group minerals (chlorite: Chl) and carbonate group minerals (dolomite: Dol).
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
85
Fig. 6. (a) The “inverted reflectance values” for hull quotient-corrected spectra for amphibole feature of the lamproite sample. Its local maximum is indicated with “1” and the reference level is shown as a dashed line that connects the two points “a” and “b” at the shoulder of the feature, b) Different mixtures of lichen and lamproite spectra for the same feature shown in “a”, circles indicate the wavelength positions of the local minima.
minerals or mineral composition variation as the lichen effect in the spectra is dominating. 4. Results 4.1. Spectral mixtures of lichen and mica group minerals The spectral shifts of the AlOH and MgOH absorption features related to mica group minerals were investigated by plotting the absorption feature wavelength positions of muscovite, biotite and phlogopite against their encrusted lichen with different abundances. The results illustrate that features located around similar spectral regions show similar displacement trends as a function of lichen cover, albeit varying scales of displacement. At the level of 90% lichen coverage, all the mica features are obscured (feature depth b 0.0002 and/or wavelength displacement N 20 nm) and it is not possible to differentiate the different lichen-covered rock types spectrally (Fig. 7a,c,d,e,f & Fig. 8b,d & Table 2). 4.1.1. Muscovite The lichen absorption bands around 2100 and 2300 nm bound the 2200 nm muscovite band, which is opposed by a lichen reflectance peak near 2220 nm (Fig. 8a). As the percentage of lichen cover increases, the combination of the three lichen features flatten and broaden the 2200 nm band (Fig. 8b) and the lichen features increase in intensity. The spectral shifts of the features related to muscovite revealed systematic trends in the AlOH wavelength positions (band 2202 nm) with increasing the lichen abundance. This trend is manifested as a shift toward shorter AlOH feature wavelengths in granite, fenite, gneiss and monzonite (Fig. 7a & Table 2). In contrast, no spectral variability of the AlOH absorption feature wavelength was discerned for the quartzite samples (Table 2). The implication is that the 2200 nm band in quartzite samples is much deeper compared to the ones from the granite, fenite, gneiss and monzonite samples. Accordingly, the spectral signature of quartzite samples show more resistance toward the effect of lichens. In general, the depth of muscovite feature for different samples defines the scale of wavelength displacement and the amount of lichen needed to induce these displacements. The muscovite feature is much more subtle in monzonite and granite rocks, thus only a small amount of lichen, 20% and 30%, respectively, is required to hide them. This value is 60% for fenite, 80% for gneiss and 90% for quartzite. The secondary muscovite features near 2350 nm and 2450 nm vary in terms of how they are affected by lichen cover. This is due to the fact
that lichens have a feature around 2300 nm (Fig. 8c) while the shape of their spectra is convex around 2450 nm (Fig. 8d). As a result, the band at 2350 nm is shifted toward shorter wavelength, where an extra lichen related feature around 2310 nm begins to appear (Fig. 8c), while the negative slope of lichens around band 2450 nm shifts this feature toward longer wavelengths (Fig. 8d) regardless of the underlying rock. Similar to the previous muscovite feature in quartzite at 2202 nm, the 2350 nm absorption feature remains rather unaffected by the effect of lichens up to high levels of lichen cover (Fig. 7d). However, 50% lichen is enough to induce wavelength displacements above 10 nm for the 2450 nm absorption feature for these rock types (Fig. 7f) and further increasing the lichen abundance results in a displacement up to 20 nm (Fig. 7f & Table 2). Fenite samples are more sensitive toward the hampering effect of lichens. The secondary muscovite features (around 2350 nm and 2450 nm) are completely obscured at about 90% and 60% lichen cover, respectively (Fig. 8c,d). When the lichen abundance reaches 70% for band 2350 nm and 50% for band 2450 nm, differentiation of these features from lichen features becomes almost impossible. 4.1.2. Phlogopite Features from phlogopite react differently to the lichen effect. The band at 2320 nm is shifted toward shorter wavelengths, (Fig. 7c), the 2438 nm band is shifted toward longer wavelengths (Fig. 7f) due to the negative slope of lichen spectra within this range, and the 2380 nm band remains constant (Fig. 7e) with increasing abundance of lichens. The scale of displacement is almost identical for both kimberlite and kersantite rock types (7 nm for band 2350 nm and 9 nm for band 2450 nm; see Fig. 7c,f & Table 2). 4.1.3. Biotite The hydroxyl band around 2340 nm is shifted toward shorter wavelengths with an average of 8 nm (Fig. 7d & Table 2), while a secondary band around 2400 nm remains constant (Fig. 7e & Table 2). Increasing the lichen cover makes both biotite features shallower and above 60% lichen coverage the secondary feature is barely observable in the spectra. 4.2. Spectral mixtures of lichen and amphibole group minerals The broad 2300 nm lichen band is paired with a similarly positioned amphibole feature around 2324 nm for monzonite and 2315 nm for lamproite samples, respectively (Fig. 9a). Since there is little contrast
86
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
Fig. 7. Wavelength displacements of features related to common rock forming minerals via mixing with different lichen percentage. The plots are split and arranged based on ascending wavelength within the range of 2000–2500 nm, and hence multiple absorption features of the same minerals are shown in different plots depending on the related wavelength region.
between the reflectance of the rocks and lichens within this region, the changes in absorption band characteristics are generally minor with increasing lichen cover (Fig. 9b & Table 2). Nevertheless, increasing the lichen percentage flattens the amphibole band, causing a total shift of 2 nm and 1 nm toward the shorter wavelength, for monzonite and lamproite respectively, where the lichen 2300 nm band appears (Fig. 7c, 9b & Table 2).The amphibole feature within this range is stronger for lamproite as compared with monzonite samples. This means that the position of this band is consistent until the lichen cover reaches 80% for lamproite, while 40% lichen is enough to displace this feature for monzonite (Table 2). The second amphibole band around 2380 nm occurs in an area where the lichen curve is slightly convex (Fig. 9a). The characteristics of this feature are rather stable and no spectral displacement is observable up to high levels of lichen coverage (Fig. 7e). An average lichen cover of 90% obscures the amphibole feature and introduces features belonging to the lichens (Fig. 9c).
4.3. Spectral mixtures of lichen and serpentine group minerals The band near 2300 nm in the lichen is opposed by a steeply dropping absorption feature in the kimberlite samples around 2325 nm, which is related to antigorite (Fig. 10a). This absorption feature moves toward shorter wavelengths by increasing the lichen cover up to 90% (with an average of ~ 10 nm); while it becomes shallower and broader (Fig. 10b & Table 2).
4.4. Synthetic mixtures of lichen and chlorite group minerals There is a negative slope in the lichen spectrum around 2254 nm, where the strong FeOH chlorite absorption feature is situated. It shifts the chlorite feature toward longer wavelengths by an average of 9 nm for gneiss with increasing lichen coverage up to 90% (Fig. 11b & Table 2).
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
87
Fig. 8. (a) The averaged spectra of pure rock and lichen for fenite substrate in SWIR and (b,c,d) the corresponding hull quotient and band centers of mixed spectra associated with muscovite features. The spectral intervals (10%) that are used to investigate the three main absorption features are highlighted. An x denotes the wavelength positions of the local minima.
A similar trend toward longer wavelengths is observed in the granite samples with an average displacement of 8 nm for 40% lichen coverage (Fig. 7b). However, increased lichen coverage causes the chlorite feature gradually to smooth out for granite. This means that perfect spectral matching of chlorite features becomes challenging in presence of 45% lichen coverage for granite samples. Additionally, a lichen feature begins to appear which causes a shift of up to 16 nm until it reaches pure lichen spectra (Fig. 7b & Table 2). The secondary chlorite features around 2320 and 2386 nm exist only in our granite samples. Since lichen has a broad feature around 2300 nm but has a convex shape around 2380 nm, chlorite features within these
ranges show different characteristics (Fig. 11d,e). The 2320 nm band shifts toward shorter wavelength with an average of 6 nm, while the feature around 2386 nm remains stable but becomes shallower and gradually disappears as the lichen abundance increases up to 90% (Fig. 11d,e & Table 2). 4.5. Spectral mixtures of lichen and carbonate minerals Lichens have a feature around 2300 nm hence, very close to a feature associated with carbonates (2320 nm). It is therefore possible that a lichen feature is misinterpreted as a dolomite feature. Adding
Table 2 Summary of the wavelength of diagnostic features and the total shifts induced by lichens per mineral group. Mineral group
Minerals
Rock type
Mica
Phlogopite
Kimberlite Kersantite Granite Quartzite Fenite Gneiss Monzonite Gneiss Lamproite Monzonite Kimberlite Gneiss Granite Carbonatite
Wavelength of diagnostic features and the related shifts (nm) 2200
Muscovite
Serpentine Chlorite
Biotite Richterite Hornblende Antigorite Chlorite
Carbonate
Dolomite
Amphibole
2254
2320–2325
2340–2350
−7 −3 Constant −4 −1 −8
2380–2390
2438–2450
Constant Constant
+9
−7 −6
−8 −1 −2 −10 +9 +16
−6 −8
+20 +12
Constant Constant Constant
Constant
88
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
Fig. 9. (a) The averaged spectra of pure rock and lichen for lamproite substrate in SWIR range and (b,c) the corresponding hull quotient and band centers of mixed spectra associated with amphibole (richterite) features. The spectral intervals (10%) that are used to investigate the main absorption features are highlighted. An x denotes the wavelength positions of the local minima.
up the percentage of lichen cover (up to 90%) causes the dolomite absorption feature to move toward shorter wavelength with an average of 8 nm (Fig. 7c & Table 2). The width of the dolomite feature remains almost unchanged, but its depth decreases with increasing amount of lichens. Fifty percent of lichen coverage induces a spectral displacement of 3 nm and at 90% lichen coverage the dolomite feature is completely obscured (Fig. 12b).
5. Discussion Spectral mixing between different rock substrates and lichens influences the spectral shape and wavelength positions of target minerals in the SWIR wavelength region. These effects can thus impede the classification of mineralogy interpreted from spectra. We examined the extent to which characteristic features related to common rock forming
Fig. 10. (a) The averaged spectra of pure rock and lichen for kimberlite substrate in SWIR range and (b) the corresponding hull quotient and band centers of mixed spectra associated with antigorite feature. The spectral intervals (10%) that are used to investigate the main absorption features are highlighted. An x denotes the wavelength positions of the local minima.
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
89
Fig. 11. (a,c) The averaged spectra of pure rock and lichen for granite and gneiss substrate in SWIR range and (b,d,e) the corresponding hull quotient and band centers of mixed spectra associated with chlorite feature. The spectral intervals (10%) that are used to investigate the main absorption features are highlighted. An x denotes the wavelength positions of the local minima.
minerals (amphibole, mica, serpentine, chlorite and carbonate group) are preserved in the presence of varying lichen abundance and further investigated the scale of related wavelength displacements within the SWIR region. Considering the challenge of lithological mapping in areas of multiple lichen species, spectral unmixing in this spectral range is preferred for determining lichen and substrate abundances, when one is not concerned with lichen species composition (Morison et al., 2014; Zhang et al., 2005). In the latter case the spectral region between 800 and 1300 nm performs best at lichen-substrate differentiation and interspecial lichen differentiation (Morison et al., 2014; Salehi et al., 2016), however the shape of lichen spectra can be highly variable in
this region owing to the type of fungal components and the associated differences in pigmentation (Hale, 1967; Petzold and Goward, 1988). The results of this study imply that convex, concave or flat shape (local slope) of lichen spectra around the absorption features related to each of the aforementioned minerals is one of the factors, which determines the trend and the scale of displacement of features within mixture spectra. Accordingly, mineral features that are located within an area where the lichen curve is slightly convex withstand the damping effect of lichens and no spectral displacement is observable for these features. However, further increasing the lichen abundance reduces the depth of these features and at certain coverage masks them away. In addition, the negative slope of lichen spectra around mineral
90
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
Fig. 12. (a) The averaged spectra of pure rock and lichen for carbonatite substrate in SWIR range and (b) the corresponding hull quotient and band centers of mixed spectra associated with dolomite feature. The spectral intervals (10%) that are used to investigate the main absorption features are highlighted. An x denotes the wavelength positions of the local minima.
absorption features, shifts features toward longer wavelength while the opposite shift, toward shorter wavelength, is observable for the positive slope. This can further confuse identification of a mineral, when it is characterized by multiple absorption features (such as chlorite, amphibole and mica group minerals). According to the results for the white mica features, which are located within the range of 2200 nm to 2350 nm, lichens induce a shift toward shorter wavelength regardless of the underlying rock type (Fig. 7a,c,d & Table 2). The findings of previous studies indicate that wavelengths of the AlOH feature in Na-bearing white micas are in the range 2190 to 2195 nm, those of muscovite (normal potassic micas) are between 2200 and 2208 nm, and those of phengite are between 2216 and 2228 nm (Post and Noble, 1993). On the other hand, the wavelength position of the AlOH feature, present in white micas, shifts systematically toward shorter wavelengths as the Al content of the octahedral sites increases. The synthesized spectral mixtures of white mica and lichens show similar trends, which may cause problems in identification of white mica features in the presence of lichens and cause misinterpretation of these mixtures as Al-rich white micas. Muscovite and phlogopite have another feature around 2438–2450 nm upon which lichen induces a shift toward longer wavelength (Fig. 7f & Table 2). The mica feature within this wavelength region is generally shallow and for 40 to 50% of lichen cover these features disappears. The depth of the feature and the contrast between lichen and substrate defines the amount of lichen needed to mask this feature. Therefore, differentiating this feature as either a biological or a mineralogical material becomes problematic. Similar to white micas, the FeOH absorption feature that is present in chlorite and biotite (Scott and Yang, 1997) is shifted toward shorter wavelengths with increasing Mg content and toward longer wavelengths with increasing Fe content (McLeod et al., 1987). According to the results, chlorite FeOH features in mixtures with lichen shifts toward longer wavelength, which can then be misinterpreted as increasing the Fe content (Fig. 7b & Table 2). Biotite features, within the range of 2380 nm to 2390 nm, can withstand the hampering effect of lichens (Fig. 7e & Table 2) while a trend toward shorter wavelength is observed for the FeOH features which can falsely be interpreted as escalation of Mg content (Fig. 7d & Table 2). The mixing of carbonates group minerals with lichens induces a shift toward shorter wavelengths (Fig. 7c & Table 2). According to the literature, dolomites are distinguishable from calcite spectra based on a small shift in the carbonate absorption feature (Gaffey, 1987). This must be considered for spectral interpretation of carbonate group minerals to avoid confusing the mixture of calcite and lichens with dolomite rich carbonates.
Diagnostic absorption feature around 2320–2325 nm can be referred to amphiboles and serpentine group minerals (Fig. 7c & Table 2). In general, mixing of the aforementioned mineral groups and lichens induces a shift toward shorter wavelengths. This effect is less pronounced for amphiboles (~ 2 nm regardless of the rock type) and almost identical for serpentine group minerals (~7–10 nm). Since these absorption features are sharper and stronger than those of the lichens, the rocks maintain their characteristic features within this range and the deep 2320 nm absorption band can still be detected as a weak feature at 80% lichen cover (Fig. 7c). Amphibole group minerals have another absorption feature around 2380 to 2390 nm (Fig. 7e) for which no changes in the wavelength positions are observed as the result of mixing of these groups of minerals and lichens. Moreover, compositional variations in minerals are typically characterized by shifts in the wavelength positions of the absorption features, with the overall shape of the spectra remaining generally unchanged (Pontual et al., 1997). This is while the induced shifts by lichens are not constant and may vary for different absorption features of a given mineral i.e. each feature may be affected differently depending on the shape of lichen spectra within that range. For example, diagnostic absorption features related to amphiboles are located both around 2320 and 2380 nm. Increasing the lichen percentage induces a shift towards shorter wavelengths for the first feature, while no spectra displacement is observed for the second amphibole band around 2380 nm. It is important to note that shifts are not just related to rock/lichen percent mixture, but also to the modal abundance of minerals in certain rock types. Considering the variability between absorption depths of minerals in different rocks, not all rocks can be treated equal. At lower mineral percentages, the proportion of the target mineral may be too low to change the spectral characteristic in a way that the associated absorption features become significant. This lower limit of detection varies for different minerals and combinations of minerals. However, minerals that have more pronounced absorption bands are generally detectable at lower percentages than weakly absorbing minerals. Moreover, background minerals (the mixtures with other minerals) and overlapping features will have effect on related absorption depth and play a critical role in the scale of wavelength displacement. Additional shoulders on the dominant mineral spectra and broadening of absorption features are other characteristics of a mixed spectrum, which complicate any interpretation (Pontual et al., 1997). As an example, the main problem in the interpretation of the spectra of chlorite alteration and in the evaluation of trends in spectral shifts in chlorite is that it may be difficult to determine chlorite composition in mixtures with sericite. This is because the sericite 2344 nm (ALOH) absorption overlaps the chlorite MgOH absorption and can mask its wavelength. The solution then
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
depends on the dominant mineral within the mixture. If the chlorite absorptions are stronger than the sericite then the wavelength of the absorption near 2344 nm should primarily be influenced by the chlorite MgOH absorption and should reflect the chlorite composition. However, if the sericite is spectrally dominant and the chlorite is only present as a weak feature then it would not be possible to determine the chlorite composition using the wavelength of the MgOH absorption (Pontual et al., 1997). The results of this study facilitate a more accurate interpretation and analysis of hyperspectral data, which might ultimately lead to a more effective lithological mapping in lichen-rich environments. However, several complications must be considered in connection with quantitative analysis. One of the points that merit consideration is that lichens do not grow randomly on rock surfaces, but in many cases grow preferentially on certain minerals. Thus, the mixing characteristics between lichens and minerals are not necessarily equal which complicates quantitative analysis. Additionally, our experimental design implies that we mix a bulk rock signature with lichens. Therefore, we do not account for preferential growth of lichens on specific minerals and we assume equal mixing of lichens with all minerals. This will have an impact on the results, such that shifts are not always identical because each rock has different modal percentage of minerals. It may be possible to use more than one spectral characteristic in order to extend the range of mineral percentages over which the qualitative relationship can be applied. Using more than one spectral characteristic also allows for cross checking between spectral values for unknown samples. This helps to ensure that the correlation between mineral percentage and spectral data is holding. The relatively small proportion of minerals (i.e. with shallow features) in the study area may exacerbate detection of the spectral properties of rock outcrops in the airborne/spaceborne spectrometry. The detection of shifts in mineral absorption features will also be subject to the given sensor, specifically the spectral sampling interval. Lower sampling intervals of airborne/spaceborne sensors may be manifested as a shift toward shorter or longer wavelengths for some absorptions features depending on the minima positions. The sampling interval will also impact the ability to detect shifts, such that chemical compositional changes that result in broad and continuous wavelength displacement (such as within biotite, chlorite, and muscovite) may only be possible with high spectral resolution datasets with narrow bandpass sampling intervals. Given the above problems with quantifying potential spectral shifts, this work has highlighted distinctive trends that can be used to define the extent to which changes in mineral composition relevant to exploration activities may be captured despite of a certain percentage of the lichen coverage in future analysis. 6. Conclusions Our study highlights the importance of careful selection of wavelengths for image analysis in environments where the spectral signatures are strongly influenced by spectral mixing with lichens. These findings indicate that applications such as mapping hydrothermal alteration zones by means of hyperspectral remote sensing require either filtering out lichens before further analysis or being precautious in any interpretation as abundant lichen coverage induces wavelength displacement in characteristic rock features. It may be useful to extend the analysis to more than one spectral characteristic as this would allow for cross checking between spectral values for unknown samples and helps to ensure that the correlation between mineral percentage and spectral data is holding. Given the above challenges in quantifying potential shifts, this work has identified a number of distinctive trends in spectral shifts that can be used in future analysis: 1. Generally, mineral features located within a spectral domain where the lichen curve is slightly convex, withstand the obscuring effect of lichens and no spectral displacement is observed for these features
91
(such as mica features between the range of 2380 to 2390 nm). The negative slope of lichen spectra around an absorption feature shifts mineral features toward longer wavelengths (such as mica features above 2400 nm) while the opposite shift, toward shorter wavelength, is observed if the lichen curve has a positive slope (such as the features around 2320 nm in carbonate, amphibole and serpentine group minerals and mica features within the range of 2200 to 2350 nm). 2. Spectral shifts are not constant and may vary for absorption features of a given mineral i.e. each feature may be affected differently depending on the shape of lichen within a given spectral range. For example, the feature related to the chlorite group minerals around 2254 nm is shifted toward longer wavelength, while the one around 2320 nm shifts toward shorter wavelength and the 2380 nm band maintain its spectral characteristics. 3. Spectral shifts are not only related to rock/lichen percent mixture, but also to the modal abundance of minerals in certain rock types. Background minerals and associated overlapping features will have effect on related absorption depth and play a critical role in scale of wavelength displacement. 4. Lichens do not grow randomly on rock surfaces, but in many cases grow preferentially on certain minerals. Thus, the mixing characteristics between lichens and minerals are not necessarily equal. We do not account for preferential growth of lichens on specific minerals and we assume equal mixing of lichens with all minerals. This will have an impact on the results, such that shifts are not always identical because each rock has different modal percentage of minerals. Acknowledgements The authors gratefully acknowledge use of the facilities and rock samples of the Geological Survey of Denmark and Greenland (GEUS). Thanks to T. Balic Zunic (University of Copenhagen) for his support in XRD analysis, to L. Thorning and P. Guarnieri (GEUS) for geological discussions. Thanks are also due to T. Tukiainen and T.F.D. Nielsen (GEUS) for assistance with sample preparation. The authors acknowledge the financial support of Geocenter Denmark (2-2014). References Ager, C.M., Milton, N.M., 1987. Spectral reflectance of lichens and their effects on the reflectance of rock substrates. Geophysics 52, 898–906. Bechtel, R., Rivard, B., Sánchez-Azofeifa, A., 2002. Spectral properties of foliose and crustose lichens based on laboratory experiments. Remote Sens. Environ. 82, 389–396. Bruegge, C.J., Stiegman, A.E., Rainen, R.A., Springsteen, A.W., 1993. Use of Spectralon as a diffuse reflectance standard for in-flight calibration of earth-orbiting sensors. Opt. Eng. 32, 805–814. Budkewitsch, P., Staenz, K., Neville, R.A., Rencz, A.N., Sangster, D., 2000. Spectral signatures of carbonate rocks surrounding the Nanisivik MVT Zn-Pb mine and implications of hyperspectral imaging for exploration in Arctic environments. Ore Deposit Workshop: New Ideas for a New Millennium, Technical Volume. Clark, R.N., 1999. Spectroscopy of rocks and minerals, and principles of spectroscopy. Manual of Remote Sensing. 3, pp. 3–58. Clark, R.N., Roush, T.L., 1984. Reflectance spectroscopy: quantitative analysis techniques for remote sensing applications. J. Geophys. Res. Solid Earth 89, 6329–6340. Coe, K., Cheeney, R., 1972. Preliminary results of mapping in Liverpool Land, East Greenland. Rapp. Grønl. Geol. Unders. 48, 7–20. Duke, E.F., 1994. Near infrared spectra of muscovite, Tschermak substitution, and metamorphic reaction progress: implications for remote sensing. Geology 22, 621–624. Escher, A., Watt, W.S., 1976. Geology of Greenland. Geological Survey of Greenland. Gaffey, S.J., 1987. Spectral reflectance of carbonate minerals in the visible and near infrared (0.35–2.55 μm): anhydrous carbonate minerals. J. Geophys. Res. Solid Earth 92, 1429–1440. Gou, S., Yue, Z., Di, K., Wang, J., 2015. Mineral abundances and different levels of alteration around Mawrth Vallis, Mars. Geosci. Front. 6, 741–758. Graham, A., 1999. Late Cretaceous and Cenozoic History of North American Vegetation: North of Mexico. Oxford University Press on Demand. Hale, M.E., 1967. In: Arnold, Edward (Ed.), The Biology of Lichens London. Harris, J., Eddy, B., Rencz, A., de Kemp, E., Budkewitsch, P., Peshko, M., 2001. Remote Sensing as a Geological Mapping Tool in the Arctic: Preliminary Results from Baffin Island, Nunavut. Natural Resources Canada, Geological Survey of Canada. Harris, J.R., Rogge, D., Hitchcock, R., Ijewliw, O., Wright, D., 2005. Mapping lithology in Canada's Arctic: application of hyperspectral data using the minimum noise fraction transformation and matched filtering. Can. J. Earth Sci. 42, 2173–2193.
92
S. Salehi et al. / Remote Sensing of Environment 199 (2017) 78–92
Hunt, G.R., 1977. Spectral signatures of particulate minerals in the visible and near infrared. Geophysics 42, 501–513. Hunt, G.R., Ashley, R.P., 1979. Spectra of altered rocks in the visible and near infrared. Econ. Geol. 74, 1613–1629. Ingle, V.K., Proakis, J.G., 2016. Digital signal processing using MATLAB. Cengage Learning. Jensen, S.M., Hansen, H., Secher, K., Steenfelt, A., Schjøth, F., Rasmussen, T.M., 2002. Kimberlites and other ultramafic alkaline rocks in the Sisimiut–Kangerlussuaq region, southern West Greenland. Geol. Greenl. Surv. Bull. 191, 57–66. King, T.V., Clark, R.N., 1989. Spectral characteristics of chlorites and Mg-serpentines using high-resolution reflectance spectroscopy. J. Geophys. Res. Solid Earth 94, 13997–14008. Kruse, F., Lefkoff, A., Boardman, J., Heidebrecht, K., Shapiro, A., Barloon, P., Goetz, A., 1993. The spectral image processing system (SIPS)—interactive visualization and analysis of imaging spectrometer data. Remote Sens. Environ. 44, 145–163. Kruse, F.A., Bedell, R.L., Taranik, J.V., Peppin, W.A., Weatherbee, O., Calvin, W.M., 2012. Mapping alteration minerals at prospect, outcrop and drill core scales using imaging spectrometry. Int. J. Remote Sens. 33, 1780–1798. Laakso, K., Rivard, B., Peter, J., White, H., Maloley, M., Harris, J., Rogge, D., 2015. Application of airborne, laboratory, and field hyperspectral methods to mineral exploration in the Canadian Arctic: recognition and characterization of volcanogenic massive sulfide-associated hydrothermal alteration in the Izok Lake deposit area, Nunavut, Canada. Econ. Geol. 110, 925–941. Laakso, K., Rivard, B., Rogge, D., 2016. Enhanced detection of gossans using hyperspectral data: example from the cape Smith Belt of northern Quebec, Canada. ISPRS J. Photogramm. Remote Sens. 114, 137–150. Larsen, L.M., Rex, D.C., 1992. A review of the 2500 Ma span of alkaline-ultramafic, potassic and carbonatitic magmatism in West Greenland. Lithos 28, 367–402. Li, L., Solana, C., Canters, F., Chan, J., Kervyn, M., 2015. Impact of environmental factors on the spectral characteristics of lava surfaces: field spectrometry of basaltic lava flows on Tenerife, Canary Islands, Spain. Remote Sens. 7, 15864. McLeod, R., Gabell, A., Green, A., Gardavsky, V., 1987. Chlorite infrared spectral data as proximity indicators of volcanogenic massive sulphide mineralization. Pacific Rim Congress, pp. 26–29. Morison, M., Cloutis, E., Mann, P., 2014. Spectral unmixing of multiple lichen species and underlying substrate. Int. J. Remote Sens. 35, 478–492. Petzold, D.E., Goward, S.N., 1988. Reflectance spectra of subarctic lichens. Remote Sens. Environ. 24, 481–492. Pontual, S., Merry, N., Gamson, P., 1997. G-Mex Vol. 1, Spectral Interpretation Field Manual: Kew, Victoria 3101, Australia, Ausspec International Pty. Ltd.. Post, J.L., Noble, P.N., 1993. The near-infrared combination band frequencies of dioctahedral smectites, micas, and illites. Clay Clay Miner. 41, 639. Rees, W.G., Tutubalina, O.V., Golubeva, E.I., 2004. Reflectance spectra of subarctic lichens between 400 and 2400 nm. Remote Sens. Environ. 90, 281–292.
Rivard, B., Arvidson, R.E., 1992. Utility of imaging spectrometry for lithologic mapping in Greenland. Photogramm. Eng. Remote. Sens. 58, 945–949. Rogge, D., Rivard, B., Harris, J., Zhang, J., 2009. Application of hyperspectral data for remote predictive mapping, Baffin Island, Canada. Rev. Econ. Geol. 16, 209–222. Rogge, D., Rivard, B., Segl, K., Grant, B., Feng, J., 2014. Mapping of NiCu–PGE ore hosting ultramafic rocks using airborne and simulated EnMAP hyperspectral imagery, Nunavik, Canada. Remote Sens. Environ. 152, 302–317. Rollin, E.M., Milton, E.J., Roche, P., 1994. The influence of weathering and lichen cover on the reflectance spectra of granitic rocks. Remote Sens. Environ. 50, 194–199. Salehi, S., Karami, M., Fensholt, R., 2016. Identification of a robust lichen index for the deconvolution of lichen and rock mixtures using pattern search algorithm (case study: Greenland). Int. Arch. Photogramm. Remote. Sens. Spat. Inf. Sci. XLI-B7, 973–979. Satterwhite, M.B., Henley, J.P., Carney, J.M., 1985. Effects of lichens on the reflectance spectra of granitic rock surfaces. Remote Sens. Environ. 18, 105–112. Schetselaar, E., deKemp, E., 2000. Image classification from Landsat TM, airborne magnetics and DEM data for mapping Paleoproterozoic bedrock units, Baffin Island, Nunavut, Canada. Int. Arch. Photogramm. Remote. Sens. Spat. Inf. Sci. 33, 1325–1332. Schetselaar, E.M., Ryan, J.J., 2009. Remote predictive mapping of the Boothia mainland area, Nunavut, Canada: an iterative approach using Landsat ETM, aeromagnetic, and geological field data. Can. J. Remote. Sens. 35, S72–S94. Scott, K., Yang, K., 1997. Spectral reflectance studies of white micas. Australian Mineral Industries Research Association Ltd. Report. 439, p. 35. Singer, R.B., McCord, T.B., 1979. Mars-large scale mixing of bright and dark surface materials and implications for analysis of spectral reflectance. Lunar and Planetary Science Conference Proceedings, pp. 1835–1848. Smith, M.O., Ustin, S.L., Adams, J.B., Gillespie, A.R., 1990. Vegetation in deserts: I. A regional measure of abundance from multispectral images. Remote Sens. Environ. 31, 1–26. Staenz, K., Nadeau, C., Secker, J., Budkewitsch, P., 2000. Spectral unimixing applied to vegetated environments in the Canadian arctic for mineral mapping. Int. Arch. Photogramm. Remote. Sens. 33, 1464–1471. Théau, J., Peddle, D., Duguay, C., 2005. Mapping lichen in a caribou habitat of northern Quebec, Canada, using an enhancement_classification method and spectral mixture analysis. Remote Sens. Environ. 94, 232–243. Van der Meer, F.D., De Jong, S.M., 2011. Imaging Spectrometry: Basic Principles and Prospective Applications. Springer Science & Business Media. Zhang, J., Rivard, B., Sanchez-Azofeifa, A., 2004. Derivative spectral unmixing of hyperspectral data applied to mixtures of lichen and rock. IEEE Trans. Geosci. Remote Sens. 42, 1934–1940. Zhang, J., Rivard, B., Sánchez-Azofeifa, A., 2005. Spectral unmixing of normalized reflectance data for the deconvolution of lichen and rock mixtures. Remote Sens. Environ. 95, 57–66.
PAPER II
53
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016 XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
IDENTIFICATION OF A ROBUST LICHEN INDEX FOR THE DECONVOLUTION OF LICHEN AND ROCK MIXTURES USING PATTERN SEARCH ALGORITHM (CASE STUDY: GREENLAND) S. Salehi a, b *, M. Karami a, R. Fensholt a a
b
Department of Geosciences and Natural Resource Management, University Of Copenhagen - (sara.salehi, moka, rf)@ign.ku.dk Department of Petrology and Economic Geology, Geological Survey of Denmark and Greenland, Copenhagen, Denmark
[email protected] Special Sessions, SpS16 - EARSeL
KEY WORDS: Hyperspectral Remote Sensing, Mineral Exploration, Geological Mapping, Spectral Mixture Analysis, Spectrometry, Lichen ABSTRACT: Lichens are the dominant autotrophs of polar and subpolar ecosystems commonly encrust the rock outcrops. Spectral mixing of lichens and bare rock can shift diagnostic spectral features of materials of interest thus leading to misinterpretation and false positives if mapping is done based on perfect spectral matching methodologies. Therefore, the ability to distinguish the lichen coverage from rock and decomposing a mixed pixel into a collection of pure reflectance spectra, can improve the applicability of hyperspectral methods for mineral exploration. The objective of this study is to propose a robust lichen index that can be used to estimate lichen coverage, regardless of the mineral composition of the underlying rocks. The performance of three index structures of ratio, normalized ratio and subtraction have been investigated using synthetic linear mixtures of pure rock and lichen spectra with prescribed mixing ratios. Laboratory spectroscopic data are obtained from lichen covered samples collected from Karrat, Liverpool Land, and Sisimiut regions in Greenland. The spectra are then resampled to Hyperspectral Mapper (HyMAP) resolution, in order to further investigate the functionality of the indices for the airborne platform. In both resolutions, a Pattern Search (PS) algorithm is used to identify the optimal band wavelengths and bandwidths for the lichen index. The results of our band optimization procedure revealed that the ratio between R894-1246 and R1110 explains most of the variability in the hyperspectral data at the original laboratory resolution (R2=0.769). However, the normalized index incorporating R1106-1121 and R904-1251 yields the best results for the HyMAP resolution (R2=0.765). 1. INTRODUCTION Lithological mapping using remote sensing depends, in part, on the identification of rock types by their spectral characteristics. The potential for rock type and mineral identification benefits from an understanding of the way in which surface processes modify those spectral characteristics. One of the important factors which merit consideration is the presence of organic growth, such as lichens on the rock surface. Lichens can be found from arctic and subarctic to tropical regions and are particularly well adapted to extreme environmental conditions. In arctic regions of the world, lichen cover can be so extensive that only a small portion of the rock's surface is exposed. Consequently, any acquired imagery and the related brightness values represent some unknown combination of rock and biologic material. More specifically, the spectrum of a lichen covered rock surface can be significantly different from the spectrum of bare rock, depending on the spectral contrast between the lichen and the rock substrate. The presence of lichen can increase, decrease or have little effect on the spectral reflectance of the rock surface (Satterwhite et al., 1985). Much of the existing knowledge on the reflectance spectra of lichens corresponds to the spectral resolution of multispectral airborne and spaceborne imaging systems such as Landsat MSS and TM sensors (Ager and Milton, 1987; Leverington and Moon, 2012) and Airborne Thematic Mapper (ATM) (Grebby et al., 2014). Evidently, more information is potentially
available from high-resolution spectra of hyperspectral imagery. Hyperspectral remote sensing systems are becoming more readily available, increasing the ability to map different landcover units through end-member and spectral unmixing analyses (Rogge et al., 2006; Rogge et al., 2007; Sheng et al., 2015). Such high resolution dataset provides more flexibility for optimal selection of wavebands for discriminating between different land-cover types and lichens (Laakso et al., 2015). However, rather limited hyperspectral data are currently available for studies of arctic and subarctic lichens. The first major data set was published by Rivard and Arvidson (1992), who conducted field observations over the range 450 to 2400 nm to obtain in situ spectra of different exposed lithologic units. The survey included measurements of gneiss, granite, anorthosite, and amphibolite rocks on the west coast of Greenland. Their research illustrated that the spectra of bare amphibolite and tonalite substrates were significantly altered by lichen cover as they were rather flat and featureless. Rollin et al. (1994) investigated the influence of weathering and lichen cover on the reflectance spectra of granitic rocks over visible and infrared wavelengths. They concluded that all the lichen affected spectra showed identical diagnostic absorption features in the shortwave infrared region (SWIR). They further suggested that these features, which were found to occur only in the spectra of lichen covered surfaces, were potentially useful for lichen identification by spectral measurements from satellite or airborne sensors. Despite analyses focusing on spectral properties of lichens and rock-encrusting lichens, little attention has been paid to model
This contribution has been peer-reviewed. doi:10.5194/isprsarchives-XLI-B7-973-2016
973
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016 XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
lichen effects and to minimize their effect by mapping varying lichen abundances on rock exposures. Ager and Milton (1987) suggested band ratios of TM3:TM4 and TM5:TM2 from laboratory modelling to minimize lichen effects in TM images. Bechtel et al. (2002) suggested a set of spectral indices to discriminate lichens and a set of spectral indices to guide the selection of a single lichen end-member for use in the Spectral Mixture Analysis (SMA) of rock and lichen. Discrimination of different lichen species was conducted by using ratios of reflectance at 400/685 and 773/685 nm. An index using the band ratios 2132/2198 and 2232/2198 nm showed the similarity of lichen spectra in the SWIR and a distinguishing feature between rocks with hydroxide bearing minerals and lichens. However, it was emphasized that not all lichen species is likely to be distinguished using this approach as the study was based on a limited number of lichen samples (three grey – black crustose species). It was concluded that a much larger number of lichen samples, from various study sites, should be measured and compared using the ratio of reflectance at 400/685 nm against 773/685 nm to evaluate the ability to separate lichens by species at these wavelengths. A recent study carried out by Li et al. (2015) showed that rock substrates covered by lichens are characterized by multiple lichen related absorption and reflection features. They demonstrated that lichen growth is one of the major factors controlling spectra of Tenerife lava surfaces and therefore tested several spectral indices to estimate lichen coverage at both site and spot scales. Their focus was in particular on the increase in reflectance from 680 to 1320 nm and the decreases from 1660 to 1725 nm and from 2230 to 2300 nm and their results showed that indices using 1660 and 1725 nm generally performed better. They further discussed that 2230 and 2300 nm have a very strong correlation with the lichen coverage but the use of wavelengths in the SWIR region from remotely sensed platforms is overshadowed by the high sensitivity to varying atmospheric conditions. As normalization helps to minimize the effects of variable illumination conditions and no differences between the ratio and normalized difference indices using 1660 and 1725 nm was observed in their study, they concluded that a normalized difference index using 1660 and 1725 nm is better in characterizing the relationship between spectral signatures of lava surfaces and surface lichen coverage. The lichen index was then applied to Hyperion imagery of their study area for mapping lichen coverage on lava surfaces. This study was motivated by the fact that the presence of lichens may affect the interpretation of mineralogy. Deconvolution of the effect of spectral mixing of rocks and lichens through spectral unmixing methodologies is thus an essential step for improving the applicability of hyperspectral methods for mineral exploration. Our objective is to identify a robust lichen index for the deconvolution of lichen and rock mixtures using a pattern search algorithm (PS). We have selected the 350–2500 nm spectral range because of its relevance to many current hyperspectral remote sensing systems and that much of the analysis of hyperspectral data for geological application is based upon the detection and identification of important OH features that occur in minerals within this range. The results of this study are expected to have a significant implication for the analysis of satellite or airborne remote sensing imagery. The proposed infrared ratios applied to the analysis of hyperspectral data can provide a simple means to reduce the effect of varying lichen abundances covering rock exposures. Such information is vital to the interpretation of remote sensing data acquired in areas having abundant lichen-
covered outcrops. In this context, changes induced by different percentage of lichen cover in the spectra shape of common rock forming minerals have been simulated using laboratory spectroscopic data obtained from lichen covered samples collected from Karrat, Liverpool Land and Sisimiut areas in Greenland. Synthetic linear mixtures of pure rock and lichen spectra with prescribed mixing ratios are then examined in order to investigate the performance of three index structures of ratio, normalized ratio and subtraction. Location and width of absorption features are parameters that were examined during this study for both laboratory spectroscopic resolution and HyMAP resolution. 2. STUDY AREA Greenland is the largest island in the world, surface of which is covered by about 80 per cent of the Inland Ice. The largest part of the ice-free area is made up of crystalline rocks of the Precambrian shield. The land surface of Greenland is a glaciated terrain, often with pronounced topography in places displaying alpine landforms with partial ice and snow cover. The action of glaciers and ice sheets during and since the Quaternary glaciations resulted in extensive areas of well exposed rocks. However, the well exposed geology is to a variable extent covered by the crusts of lichens which complicate the spectral mapping of the minerals and lithologies. For the purpose of this study lichen bearing rock samples were collected from Liverpool Land (Central East), Karrat (Central West), and Sisimiut (South West) areas in Greenland. The geology of central East Greenland is dominated by the N–S orientated Caledonian Fold Belt formed by the collision between Laurentia and Baltica 465–400 million years ago (Higgins et al., 2004). Liverpool Land forms a c. 3500 km2 horst of Caledonian crystalline rocks separated from the postCaledonian Jameson Land sedimentary basin to the west by a major N–S-oriented fault zone. The crystalline complex of North Liverpool Land is composed of Precambrian, marble bearing metamorphic rocks and granites–quartz-monzonites of Caledonian or Neoproterozoic age belonging mainly to the Hagar Bjerg thrust sheet (Coe and Cheeney, 1972). The Monzonite sample studied in this paper has been collected from this region. The Karrat group in west Greenland comprises three main formations i.e. the Mârmorilik, Qeqertarssuaq, and Nukavsak formations (St-Onge et al., 2009). The Qeqertarssuaq Formation is a metasedimentary sequence consisting of semipelitic to pelitic schist, quartzite, and quartzitic schist (Escher and Pulvertaft, 1976). This area has been selected to collect the quartzite sample (Figure 1). Southern West Greenland hosts an alkaline province with a variety of ultramafic alkaline rocks, including swarms of dykes traditionally described as kimberlites and lamproites (Larsen and Rex, 1992; Jensen et al., 2002). The region hosts several clusters of kimberlitic dykes and sills (more than 200 outcrops), which appear to be controlled by preexisting joint systems or concordant with the enclosing gneiss. A large number of dykes are located in the vicinity of the Sarfartoq carbonatite complex (Larsen and Rex, 1992; Jensen et al., 2003). Our Kimberlite, Lamproite, Carbonatite, Fenite, Granite and Gneiss samples were collected from this region (Figure 1).
This contribution has been peer-reviewed. doi:10.5194/isprsarchives-XLI-B7-973-2016
974
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016 XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
different colours to account for possible inter-species spectral variation for each sample. The measured surface in each case is approximately 10 mm determined by the diameter of the contact probe. These measurements were used to generate synthetic mixtures (Section 3.2), and subsequently identify the optimal band wavelengths and bandwidths for the lichen index (Section 3.3). 3.2 Modelling the Impact of Lichen on the Spectra of the Rock Substrate Using Synthetic Linear Mixture Analysis Due to the presence of lichen cover at length scales of a meter or greater, the majority of pixels in the remotely sensed imagery will encompass mixtures of bare rock/soil and lichen. To quantify the effects of sub-pixel lichen cover on the spectra of the lithologies, and assess its impact on remotely sensed mapping, synthetic reflectance spectra were generated by mixing the representative spectra of the lithologies with the spectra of common lichen types using a spectral mixing technique. The principle of spectral (un)mixing is that the resulting spectrum for an image pixel containing a mixture of several distinct materials will be a linear combination of their signatures (Kruse et al., 1993). The assumption of linear mixture of lichens and rock substrate is reliable as lichens prevent transmission of light to the underlying rock substrate (Ager and Milton, 1987; Bechtel et al., 2002).
Figure 1. Geology Map of Greenland at scale 1:500.000 and the Locations of the Collected Samples
ASD spectral measurements of the lichen-covered surfaces were averaged separately for each rock type, to create a single spectral representation of lichen for specific rock substrates (Śr), (Figure 2). The same procedure was repeated for each rock type to produce a representation of the lichen-free weathered surfaces (Sr). These averaged spectra were then used to generate linear spectral mixtures of lichen and rock for each substrate with 1% intervals: Rb = f Srb + (1- f) Śrb
3. METHODOLOGY
Where
3.1 Laboratory Spectral Measurements All spectral measurements were made with a FieldSpec®3 HiRes (hereafter referred to ASD) Spectroradiometer that records 2151 channels within the 350-2500 nm wavelength range with a spectral resolution of 3 nm @ 700 nm, 8.5 nm @ 1400 nm, 6.5 nm @ 2100 nm and a sampling interval of 1.4 nm @ 350-1050 nm and 2 nm @ 1000-2500 nm. All measurements were conducted using a contact probe device which contains its own light source. Use of a contact probe provides consistent illumination conditions during data acquisition and ensures measuring the spectra of pure patches of lichen species. Radiance values were converted to reflectance values by calculating the ratio of the radiance of the sample to the radiance of a 99% reflective reference panel (Spectralon, SRT-99-050, Labsphere, North Sutton, NH, USA) under the same illumination and viewing conditions (Bruegge et al., 1993). Dark current and white reference measurements were repeated for each rock sample. Each spectrum acquired in laboratory consists of 25 individual measurements recorded consecutively and averaged by the ASD instrument. Two sets of measurements were collected from lichen-free weathered surfaces of each rock sample and from lichens on weathered surfaces. The first set of measurements was collected from 5 lichen-free, weathered surfaces of each sample. The second set of measurements was acquired from lichens of
(1)
Srb = reflectance spectrum of rock type r at band b Śrb = reflectance spectrum of lichen type r at band b f = relative proportion of lichen
Subject to: f ∈ {0.01, 0.02, 0.03, …, 0.99, 1}
(2)
3.3 Lichen Spectra Characteristic Lichen reflectance increases steadily in the range 700-1400 nm and remains high at longer wavelengths (Figure 2). This steady increase is distinctly different from the reflectance curve of vascular plants, characterized by an abrupt increase from the visible to the near infrared and decreases slowly from 800 to 1300 nm (Ager and Milton, 1987). Dark colour lichens have low reflectance in the visible part of the spectrum and show a gradual increase in reflectance at longer wavelengths, reaching a maximum around 1860 nm. The absorption feature near 1445 nm is caused by water in the lichens (Bechtel et al., 2002). Three broad absorption features near 1730, 2100, and 2300 nm, are attributable to the presence of cellulose in lichen (Ager and Milton, 1987; Bechtel et al., 2002; Rees et al., 2004). A distinct feature near 2350 nm also exists in some rock samples in this study and cannot be uniquely associated to lichens. The absorption in 1730 nm occurs with another absorption feature at approximately 2080 nm (Bechtel et al., 2002).
This contribution has been peer-reviewed. doi:10.5194/isprsarchives-XLI-B7-973-2016
975
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLI-B7, 2016 XXIII ISPRS Congress, 12–19 July 2016, Prague, Czech Republic
1445 1730 2100 2300
Figure 3. Schematic representation of the averaging concept used for index calculations
Our aim is to find band wavelengths [b1, b2, b3, b4] in a way that INDX can be used for estimation of the corresponding f for the spectra. However, since we know that the values of INDX might also need rescaling to directly estimate f— which always has a value between 0 and 1— we introduce rescaling coefficients p1 and p2 as well: 𝐼𝑁𝐷𝑋'()* = 𝑝- . 𝑅)-.)/ − 𝑅)1.)2 + 𝑝/
Figure 2. The averaged ASD spectral measurements used as representation of pure lichen spectra
3.3.1 Optimization of Lichen Index: In order to estimate the lichen coverage, the performance of three different index formulations is analysed: 𝐼𝑁𝐷𝑋'()* = 𝑅)-.)/ − 𝑅)1.)2 (3) 𝐼𝑁𝐷𝑋34*56 = 𝐼𝑁𝐷𝑋>63?
789:8; 78
