The lunar rock and mineral characterization consortium: Deconstruction and integrated mineralogical, petrologic, and spectroscopic analyses of mare basalts


  • Peter J. ISAACSON,

    Corresponding author
    1. Department of Geological Sciences, Brown University, Box 1846, Providence, Rhode Island 02912, USA
    Search for more papers by this author

    1. Planetary Geosciences Institute, Department of Earth & Planetary Sciences, University of Tennessee, Knoxville, Tennessee 37996, USA
    2. The Pheasant Memorial Laboratory for Geochemistry and Cosmochemistry, Institute for Study of the Earth’s Interior, Okayama University at Misasa, Tottori 682-0193, Japan
    Search for more papers by this author
  • Carlé M. PIETERS,

    1. Department of Geological Sciences, Brown University, Box 1846, Providence, Rhode Island 02912, USA
    Search for more papers by this author
  • Rachel L. KLIMA,

    1. Department of Geological Sciences, Brown University, Box 1846, Providence, Rhode Island 02912, USA
    2. Planetary Exploration Group, Space Department, Johns Hopkins University Applied Physics Lab, Laurel, Maryland 20723, USA
    Search for more papers by this author
  • Takahiro HIROI,

    1. Department of Geological Sciences, Brown University, Box 1846, Providence, Rhode Island 02912, USA
    Search for more papers by this author
  • Yang LIU,

    1. Planetary Geosciences Institute, Department of Earth & Planetary Sciences, University of Tennessee, Knoxville, Tennessee 37996, USA
    Search for more papers by this author
  • Lawrence A. TAYLOR

    1. Planetary Geosciences Institute, Department of Earth & Planetary Sciences, University of Tennessee, Knoxville, Tennessee 37996, USA
    Search for more papers by this author



The lunar rock and mineral characterization consortium (LRMCC) has conducted coordinated mineralogy/petrography/spectroscopy analyses of a suite of pristine lunar basalts. Four basalt slabs (two low-Ti, two high-Ti) and paired thin sections were analyzed. Thin sections were analyzed for mineralogy/petrography, while the slabs were used to prepare particulate separates of major mineral phases and bulk samples. Mineral separates and particulate bulk samples were crushed to controlled grain sizes and their reflectance spectra measured in the NASA RELAB at Brown University. The resulting data set provides an essential foundation for spectral mixing models, offers valuable endmember constraints for space weathering analyses, and represents critical new ground truth results for lunar science and exploration efforts.


The current generation of lunar exploration is rich, with a number of orbital missions in operation and future surface missions in the planning phases to address critical, outstanding questions in lunar science. The present suite of lunar orbiters has returned and is returning global data from a range of sophisticated instruments, which will allow the lunar science community to address questions that have been poorly constrained due to limitations in the available data.

Optical instruments offer the highest spatial resolution for analyzing and mapping the Moon’s surface mineralogy. A number of high-quality instruments on the current suite of orbiters provide the high spatial resolution, broad spectral coverage, and high spectral resolution data needed for accurate interpretation of lunar surface mineralogy (Ohtake et al. 2007; Matsunaga et al. 2008; Kiran Kumar et al. 2009; Mall et al. 2009; Pieters et al. 2009; Tschimmel et al. 2009). Interpretation of such data, however, relies in large part on links to returned samples analyzed in terrestrial laboratories (e.g., Pieters 1986, 1999; Lucey et al. 2000; Pieters et al. 2000; Gillis et al. 2003). As such, careful measurement of samples returned by the Apollo and Luna programs is essential for the current era of lunar exploration.

Optical instruments that measure reflected visible to near-infrared (VNIR) radiation are sensitive to surface mineralogy because highly diagnostic absorption features occur in this wavelength region. The basic reflectance properties of the major lunar minerals are known and based on mineral physics. In the VNIR, the major lunar minerals have diagnostic absorption features caused by crystal field transitions in transition metal (largely Fe2+) cations in distorted crystal lattice sites. Pyroxenes have diagnostic absorptions at 1 and 2 μm, while olivine has a combination absorption near 1 μm (Burns 1970, 1993; Adams 1974). Both minerals’ absorption features shift in generally understood ways based on mineral composition (e.g., Burns 1974; Cloutis and Gaffey 1991; Sunshine and Pieters 1998; Klima et al. 2007). Iron-bearing plagioclase has a weaker but still diagnostic absorption feature near 1.3 μm, although it is often overwhelmed by the more optically active pyroxene or olivine features in spectra of bulk rocks (Bell and Mao 1973; Adams and Goullaud 1978; Burns 1993). Example reflectance spectra of common lunar minerals are presented in the Supporting Information (Fig. S4). While the major crystal field absorptions are generally well understood, minor element substitutions are sensitive to the local environment and can have substantial effects on the spectral properties of lunar minerals. Because lunar conditions are often distinct in subtle ways, it is important to use real lunar materials as ground truth rather than relying on synthetic or terrestrial analogs. The lunar rock and mineral characterization consortium (LRMCC) has conducted careful, integrated measurements of lunar materials with the goal of extending the library of well-characterized lunar materials available as ground truth.

The LRMCC builds on the work of the lunar soil characterization consortium (LSCC), which conduc-ted coordinated mineralogy/petrography/spectroscopy analyses of a suite of lunar soils of different composition and maturity, but did not analyze the unweathered component of the lunar sample suite (Taylor et al. 2001, 2010). The LRMCC is conducting similar coordinated analyses, and has initially focused on pristine, unweathered mare basalt samples and mineral separates prepared from the bulk rock samples. The measurement and compositional characterization of both mineral separates and the rocks from which they were prepared are of particular value for spectral unmixing and space weathering models. In the case of spectral unmixing, the LRMCC samples represent exceptional constraints for models, because spectra of both bulk rock samples and their exact constituent (endmember) minerals were obtained. Spectral unmixing is discussed further in the Mixing Models section. In the case of space weathering models, these measurements provide accurate constraints on the starting materials for the space weathering process, which, when combined with the results of the LSCC, should provide a more complete assessment of the space weathering of lunar materials (e.g., Pieters et al. 2000; Hapke 2001; Noble et al. 2001, 2007; Taylor et al. 2001, 2010; Sasaki et al. 2003). In total, the combined results from the LSCC and LRMCC provide a suite of well-characterized ground truth lunar materials that span a range of compositions and maturities. These ground truth data provide a strong foundation for lunar science and exploration with these new high spectral resolution data sets.

Methods, Samples, and Sample Preparation

Samples, Petrography, Modal Mineralogy

The LRMCC was allocated slabs and associated thin sections of two low-Ti basalts (15058 and 15555) and two high-Ti basalts (70017 and 70035). The thin sections and slabs were created specifically for the coordinated analyses of this study. The thin sections and slab samples were prepared from adjacent portions of the parent samples to ensure similar mineralogy. The thin sections (15058,248, 15555,971, 70017,541, 70035,193) were used for detailed mineralogy and petrography analyses. The slabs (15058,276, 15555,965, 70017,535, 70035,188) were used to prepare particulate bulk samples, a portion of which were used for the separation of mineral separates. Mineral separation was done by hand picking under an optical microscope at the University of Tennessee. Mineral separates were prepared from a 125–250 μm size fraction and visually checked for purity. Grains rejected for purity from these separations (i.e., grains with intergrown phases) were crushed to less than 125 μm to remove impurities. Ethanol was used in the crushing process for 15058 and 70035, while 15555 and 70017 were crushed without the use of any liquids. No liquids were used to sieve the samples. Representative subsamples of the mineral separates were used to prepare grain mounts for chemical analyses. Because minerals were separated on the basis of color, the separations do not follow strict compositional divisions (e.g., pigeonite from augite). Thus, it is important to analyze directly the compositional ranges of the prepared separates through analysis of these grain mounts, both to verify that the separates are representative of the range of mineral compositions found in the thin section and to understand precisely the composition of the minerals to be analyzed with reflectance spectroscopy.

The paired thin sections were used to analyze the petrography, modal mineralogy, and mineral composition of the basalt samples. Sample petrography was evaluated using a combination of optical microscopy and backscattered electron imagery. Chemical analyses of minerals were performed with the Cameca SX-50 electron microprobe (EMP) at the University of Tennessee. Operating conditions were as follows: 15 keV accelerating voltage, 20 nA beam current, and a 2 μm beam diameter, except for plagioclase analyses, in which case a 5 μm diameter beam and 10 nA current were used to minimize loss of volatile elements K and Na. Modal mineralogy was determined using the Oxford FeatureScan software package of an Oxford instrument energy dispersive spectrometer (EDS) interfaced to the microprobe. Threshold values of cation abundance (based on EDS measurements) were specified and used to classify the various phases in the thin section. EDS spectra are qualitative (absolute cation abundances cannot be determined). Cation abundance thresholds were thus defined as “corrected window percent” (CWPC). These values are defined as percentage of the counts in an element’s energy window (where its “peak” in EDS spectra occurs) relative to the total count over all windows after a background removal. However, these CWPC are generally comparable to the given element’s oxide weight percentage (Taylor et al. 1996). Modal mineralogy was then determined by an automated point counting procedure on the classified thin section. A detailed description of this procedure is provided by Taylor et al. (1996). The CWPC values used in this study are reported in Table 1.

Table 1.   Corrected window percent ranges used to classify phases for modal mineralogy analyses.
Mineral phaseCationInclusive range
K-rich glassSi40–90

The chemical compositions of the prepared mineral separates were also determined by EMP analysis. A representative subset of grains from each separate was prepared as grain mounts for EMP. Compositions were compared to those obtained from thin section analyses to ensure that the prepared mineral separates were representative of the range of compositions found in the bulk sample.

While the mineral separates are expected to be compositionally similar to the bulk samples when considered inclusively, it is necessary to check their composition, as there is a possibility that the preparation of mineral separates by hand picking may bias the composition of the separates. Furthermore, the thin sections were classified modally by mineral chemistry, while the modal abundance of the mineral separates in the bulk sample cannot be estimated. Thus, accurate compositional assessments of the thin sections are important in that they allow one to estimate the degree to which the various pyroxene compositions are represented in the separates and the degree to which the separates are representative of the overall mineral compositions in the thin sections. In other words, the composition of the mineral separates is critical for placing the mineral separates into the context of the measured modal abundances from the thin sections.

Reflectance Spectroscopy

The particulate samples were used for spectroscopic measurements in the NASA Reflectance Experiment Laboratory (RELAB) at Brown University. Particulate samples were first received at Brown with a grain size range of about 125–250 μm. Initial measurements of these samples were made across a range of wavelengths, but this grain size range is too coarse for typical lunar applications. The samples were crushed and sieved to a series of finer grain sizes, first less than 125 μm and then less than 45 μm, and measurements were collected at each grain size prior to preparation of the next. Samples were crushed under dry conditions, with no liquids used either in the crushing or sieving process. A rough estimate of grain size distribution in the less than 125 μm splits was obtained prior to crushing the samples to less than 45 μm. Prior to crushing, the less than 125 μm splits were passed through a 45 μm sieve, and the masses of the less than 45 μm and 45 μm to 125 μm splits were measured by weighing. These results are presented graphically in the Supporting Information (Fig. S5). While not a true estimate of grain size distribution, this approach provides a rough estimate of the relative abundances of coarse versus fine particles in the less than 125 μm splits. The less than 125 μm separates are largely dominated by coarse particles, though most contain a non-negligible component of fine particles. These results indicate that the mean particle size of the less than 125 μm separates tends to be larger than might be assumed. Several separates, most notably plagioclase, tended to have slightly more fine particles in the less than 125 μm split, perhaps due to lower resistance to the crushing process.

Particle size should not be mistaken for the grain size of the samples, particularly for the bulk samples (e.g., Bohren and Huffman 1983; Poulet and Erard 2004). When the samples are crushed, a mixture of single-grain and multigrain particles is produced. In the case of multigrain particles (either single-phase or multiple-phase), photons interact with multiple grains/phases while “traversing” a single particle. This situation has optical properties very different from that in which mineral grains are physically separated from one another. Even in typical nonlinear mixture models designed to deal with intimate mixtures, particles are assumed to be fundamentally mono-minerallic, with void spaces that affect the material’s scattering properties present between grains (e.g., Hapke 1993). Multimineral particles represent a more complex system to model accurately, and the reader is cautioned that the samples discussed here often exhibit these additional complexities.

Spectra were acquired as bidirectional reflectance (BDR) at 30° incidence and 0° emergence from 0.28 μm to 2.6 μm, and as biconical reflectance by FT-IR to 50 μm. BDR spectra are the most applicable as ground truth and to quantitative models, because they match the observation geometries of orbital sensors. A more detailed description of the RELAB facilities is provided by Pieters and Hiroi (2004). BDR spectra were acquired with 0.005 μm (5 nm) spectral resolution. The spectral resolution of FT-IR measurements is variable with wavelength in nanometers, as spectra were acquired with a spectral resolution of 4 cm−1. We spliced the acquired FT-IR spectra to the BDR spectra through multiplicative scaling, which provides a better constraint of absolute reflectance and band strength beyond 2.6 μm. Biconical measurements are not directly comparable to BDR measurements for quantitative assessments of band strengths and albedos.

Spectral reflectance results are presented by wavelength range: visible/NIR (300–2600 nm), NIR/short wavelength mid-IR (2.5–8 μm), and mid-IR (7–9.5 μm). The midinfrared wavelength region contains a number of diagnostic mineral absorption features, although the process responsible for most features in this wavelength region is fundamentally different from that in the visible/NIR for the lunar minerals of interest to this study. In the visible/NIR, the absorption features are caused by crystal field transitions largely in Fe2+ ions. In the midinfrared, absorptions are linked to vibrations, stretches, and bends in crystal lattice structures tied to Si-O bonds with fundamental absorptions occurring at longer wavelengths (near 10 μm). Specific features are caused by overtones and combinations of these fundamental molecular vibrations, stretches, and bends (e.g., Salisbury and Walter 1989; Clark 1999; Cooper et al. 2002).


Mineralogy and Petrography

We discuss the major aspects of the mineralogy/petrology analyses of these samples, particularly those aspects most relevant to analysis of the reflectance spectra. Detailed mineralogy/petrography data are presented in the Supporting Information to this article. Modal abundances determined from analyses of the thin section samples are presented in Table 2. Note that these modal abundance data for the thin sections are not equivalent to the modal abundance of the mineral separates, as separations were performed manually on the basis of color rather than composition. This distinction is discussed further below and in the Mixing Models section. Overall sample texture is illustrated briefly in Fig. 1. Sample 15058 is generally subophitic, whereas 15555 and the high-Ti samples are generally poikilitic. Pyroxenes, the dominant mineral in these samples, were found to have compositions similar to those reported by previous studies (Bence and Papike 1972; Papike et al. 1974; Ryder 1985). Pyroxenes are strongly zoned for all samples analyzed by the LRMCC. The low-Ti sample pyroxenes are zoned from core to rim, becoming significantly more iron-rich toward the rim. The high-Ti sample pyroxenes exhibit complex sector zoning, making a common core to rim zoning profile difficult to establish, although the pyroxenes generally become more Fe-rich in their rims.

Table 2.   Modal abundances from thin section analysis.
Mineral phase15058,24815555,97170017,54170035,193
Olivine0.1 (fayalite)
K-rich glass0.
Figure 1.

 Photomicrographs of LRMCC thin sections illustrating sample texture. a) and e) were collected with reflected light, and the other images with transmitted light (plane-polarized for [c] and [h], cross-polarized for [b], [d], [f], [g]). a) demonstrates the subophitic texture characteristic of 15058. b) exhibits zoning from pigeonite core to augite rim in a pyroxene grain from 15058; a similar trend is observed in pyroxenes from 15555 (d). c) 15555 thin section showing euhedral chromite grain within pyroxene and poikilitic enclosure of olivine in plagioclase. The sample exhibits poikilitic enclosure of pyroxene and olivine by plagioclase. e) exhibits the typical poikilitic texture of plagioclase in 70017, with elongated pyroxenes intergrown with ilmenite and plagioclase. g) shows the poikilitic texture of plagioclase in 70035. f) and h) show the complex sector zoning observed in pyroxenes from 70017 (f) and 70035 (h). Inclusions of ilmenite within pyroxenes are apparent in both cases.

The mineral separates are all quite pure, being estimated at 98–99% purity for all prepared separates based on visual inspection. However, in many cases, when prepared for reflectance measurements, the separates show minor amounts of visible contamination by other phases, so caution should be taken when interpreting the reflectance data. The composition of the prepared mineral separates (determined from analysis of the grain mounts) and of the mineral phases (determined from thin section analyses) are presented graphically in Fig. 2. The prepared mineral separates were found to have compositional ranges consistent with those observed in the thin section analyses. Average geochemical compositions of the mineral phases analyzed in the thin sections and of the prepared mineral separates are presented in the Supporting Information (Tables S1–S8).

Figure 2.

 Compositions of major silicate mineral phases found in LRMCC samples. The left column shows the compositions of the prepared mineral separates, while the right column illustrates the compositions found in the thin section analyses. The compositions of the prepared separates generally reflect those found in thin section. Note that an olivine separate was obtained only from 15555,965. While olivine is present in the other samples, its modal abundance is too low to allow the preparation of a separate.

Mineral separates from slab sample 15058 are identified as green pyroxene, brown pyroxene, and plagioclase. Green pyroxenes are orthopyroxenes and pigeonites, containing primarily Mg-rich compositions. Brown pyroxenes are primarily augites, containing both Mg-rich and Fe-rich clinopyroxene compositions. Plagioclases are very Ca-rich (typical of plagioclases from mare basalts) and relatively Fe-rich (about 0.6 wt% FeO), and are compositionally indistinguishable from the plagioclases in the thin section.

Sample 15555 separates consist of greenish olivine, two varieties of pyroxene (reddish brown and light brown), and plagioclase. Pyroxene compositions are consistent with those observed in the polished thin sections; the reddish-brown pyroxenes are generally Mg-poor augite, and the light brown pyroxenes are generally Mg-rich subcalcic augite. Olivine and plagioclase have very similar compositional ranges to those observed in thin section, though the olivine separate lacks the fayalitic olivines (Fo5–10) seen in thin section. The olivine contains small inclusions of Cr-rich spinel (chromite). These are volumetrically minor, but are quite significant optically. These chromites can be seen by visual inspection under a binocular microscope, and occur as inclusions (possibly nucleation sites) within olivine grains. Due to their presence within olivine grains, their strong absorption features, and their small size, they will have a disproportionally large optical effect. The plagioclase has slightly lower FeO than 15058 (about 0.5 wt%)

Mineral separates from 70017 include pyroxene, plagioclase, and opaque oxides dominated by ilmenite. The pyroxenes were separated into two classes: dark brown pyroxenes are generally augites, while light brown ones are mostly pigeonites. The plagioclase separate has compositions consistent with those found in analysis of the polished thin section, and has lower FeO than found in the Apollo 15 plagioclase separates (about 0.35 wt%). The high-Ca pyroxenes (augites) were found to be rich in TiO2 (up to about 3.5 wt%).

Sample 70035 separates are light brown pyroxene, dark brown pyroxene, plagioclase, and opaque oxides dominated by ilmenite. As in 70017, the dark brown pyroxenes are mainly augites, while the light brown pyroxenes are generally pigeonites. Plagioclase compositions are essentially identical to those observed in the thin section sample, and FeO content is similar to the 70017 plagioclase (approximately 0.35 wt%). The high-Ca pyroxenes were found to contain abundances of TiO2 comparable to those observed in the 70017 high-Ca pyroxenes (approximately 3.5 wt%). Some of the lower-Ca augites do not show enrichment in TiO2, suggesting that the system was enriched in TiO2 at some point during the augite crystallization sequence, but not while pigeonites and subcalcic augites formed later in the sequence. This is discussed in the Supporting Information (see the sections on Mineral Composition and Pyroxene Zoning and Crystallization History, Figs. S1–S3).

Mineral separates often span a range of compositions. For simplicity, pyroxenes are labeled as “augite” or “pigeonite” in the reflectance spectra figures (Figs. 3–8), although these labels do not always correspond exactly to the composition of the separate. The key between actual separate names and the separate labels as plotted is given in Table 3, which lists the separate names as presented above, the separates’ general composition as given above, and the separates’ labeled composition in the reflectance spectra figures.

Figure 3.

 Visible and near-infrared spectra of coarse particle samples. Pyroxenes are labeled as pigeonites and augites for labeling simplicity, but some compositional overlap exists, and the “pigeonite” separates do contain some minor orthopyroxene.

Figure 4.

 Visible and near-infrared spectra of fine particle samples.

Figure 5.

 Midinfrared spectra of coarse particle samples in the volume-scattering wavelength region.

Figure 6.

 Midinfrared spectra of fine particle samples in the volume-scattering wavelength region.

Figure 7.

 Midinfrared spectra of coarse particle samples in the wavelength region of transition from volume to surface scattering. The approximate wavelength ranges where Christiansen features occur are noted by arrows.

Figure 8.

 Midinfrared spectra of fine particle samples in the wavelength region of transition from volume to surface scattering. The approximate wavelength ranges where Christiansen features occur are noted by arrows.

Table 3.   Mineral separate identification key.
Basalt sampleSeparate nameSeparate compositionSeparate label (Figs. 6–11)
15058,276Brown pyroxeneAugiteAugite
Green pyroxeneOPX/PigeonitePigeonite
15555,965Light brown pyroxeneMg-rich subcalcic augitePigeonite
Reddish-brown pyroxeneMg-poor augiteAugite
70017,535Light brown pyroxenePigeonitePigeonite
Dark brown pyroxeneAugiteAugite
70035,188Light brown pyroxenePigeonitePigeonite
Dark brown pyroxeneAugiteAugite

Reflectance Spectroscopy

The results of the spectral reflectance analyses are described in this section grouped by wavelength range. Labels given to spectra shown in the figures (Figs. 3–8) are linked to specific mineral species in Table 3.

Visible/NIR Bidirectional Reflectance

The spectra of coarse and fine particle separates are shown in Figs. 3 and 4, respectively. The fine-grained sample spectra are generally consistent with the coarse-grained sample spectra, though with increased albedos and decreased band strengths.

Each basalt sample contains a range of pyroxene compositions, as well as abundant plagioclase. Only 15555 contains sufficient modal olivine to prepare a mineral separate (the other samples exhibit about 2% modal olivine or less). Pyroxenes were selected and separated into two groups on the basis of color. These pyroxenes vary in composition as indicated by Fig. 2, which leads to different overall wavelength positions of the 1 and 2 μm absorption features. The separates containing more Ca-rich pyroxene (generally greater than 15–20% Ca) exhibit spectra with longer wavelength absorption features. The pyroxene separate spectra appear to exhibit 1.2 μm absorption features, which may vary in strength between compositions. Quantitative deconvolutions (e.g., see the Quantitative Absorption Modeling section) of the spectra are required to confirm this observation, however. The Apollo 17 samples are high-Ti basalts, and the augite separates from those rocks contain relatively high abundances of TiO2 (up to 3.8 wt%, as discussed in the Mixing Models section). The Apollo 17 augite spectra show a weak but perceptible feature in the visible (0.4–0.75 μm), which is essentially absent in the Apollo 17 pigeonite separate spectra. Where present, this feature is more pronounced in the coarse-grained sample pyroxene spectra.

The plagioclase spectra show relatively strong absorption features near 1.3 μm. The plagioclase spectra show marked drop-offs in reflectance at short wavelength (less than 0.4 μm), and often shallow absorption features in the visible (0.4–0.8 μm). They also exhibit a broad absorption or altered continuum slope beyond the principal 1.3 μm feature. The Apollo 17 plagioclase spectra differ from the Apollo 15 plagioclase spectra in strength of the 1.3 μm feature and in long wavelength continuum slope. Specifically, the Apollo 17 plagioclase spectra exhibit weaker 1.3 μm features and steeper long wavelength continuum slopes, reaching higher reflectance values at 2.6 μm. The 70035 plagioclase spectra have a weak feature near 1 μm, which is notably stronger in the fine-grained sample spectrum (Fig. 4).

The olivine separate spectra for 15555 exhibit the characteristic composite absorption feature near 1 μm (Burns 1974; Sunshine and Pieters 1998; Dyar et al. 2009), as well as an asymmetric long wavelength absorption feature. The ilmenite spectra are generally flat and featureless through the visible, with a pronounced upturn around 1.7 μm and a roll-off at long wavelengths (2.4 μm to 2.6 μm). They do exhibit a minor reflectance minimum at about 0.5 μm, and a peak in reflectance at about 1 μm which becomes stronger at finer particle size.

The bulk basalt sample spectra generally resemble the pyroxene spectra, as pyroxene is the dominant optical phase in these samples. The 1 and 2 μm pyroxene absorptions in the bulk sample are broadened relative to the 1 and 2 μm features observed in spectra of the pyroxene separates. The 15555 bulk sample spectrum has an especially broad 1 μm absorption. The broad visible wavelength feature seen in the Ti-rich pyroxene spectra is also subtly apparent in the high-Ti bulk sample spectra. The high-Ti sample spectra exhibit notable differences from spectra of the pyroxenes separated from those samples and from the low-Ti bulk sample spectra. The high-Ti bulk sample spectra have reduced albedos, have weakened 1 and 2 μm pyroxene absorptions (especially at the smaller particle size), and show pronounced upturns in reflectance at wavelengths beyond about 1.5 μm.

NIR/Short Wavelength Mid-IR 2.5–8 μm (Volume Scattering)

The spectra of coarse and fine particle samples for the long wavelength NIR region are shown in Figs. 5 and 6, respectively. Spectra are plotted in reflectance, though approximate “emissivity” spectra can be obtained by the simple Kirchhoff’s law relationship, which holds that emissivity is approximately equal to 1––reflectance (ε = 1−r) (e.g., Nicodemus 1965; Clark 1999). This approximation has been shown generally to be correct to first order (Salisbury 1993). For the purposes of this discussion, the NIR/short wavelength mid-IR wavelength region extends from 2.5 μm to 8 μm, stopping just short of the Christiansen feature for silicates. The pyroxene separate spectra exhibit several absorption features near 5 and 6 μm. Plagioclase spectra exhibit weak absorption features near approximately 4.4 μm. The Apollo 17 plagioclase separate spectra are somewhat brighter in the 3 to 4 μm range than the Apollo 15 plagioclase separate spectra. Olivine has some weak but detectable features between 5 and 6 μm that are often very diagnostic (Pieters et al. 2008). Spectra of all separates exhibit sharp absorption features near 3.4 μm. These features are notably stronger in the 15058 and 70035 separate spectra, and are quite weak in the bulk sample spectra. Ilmenite lacks major features in this wavelength region, but does show evidence for the same 3.4 μm absorption seen in spectra of the other separates. Ilmenite (FeTiO3) is not a silicate mineral and thus does not exhibit Si-O vibrational absorptions. It is darker than the other separates in the 3–4 μm region. As in the NIR, the bulk sample spectra generally follow the behavior of the pyroxene separates, though features caused by other minerals can be detected as well.

Spectra of all fine particle separates and some of the coarse particle separates exhibit a broad absorption feature near 3 μm. Among coarse particle samples, only plagioclase consistently exhibits a 3 μm feature. All fine particle samples (separates and bulk) exhibit a 3 μm feature, though the feature is the strongest in the plagioclase spectra.

Mid-IR 7–9.5 μm (Volume–Surface Scattering)

This wavelength region generally contains the transition from volume to surface scattering in silicate minerals. As in the long wavelength NIR wavelength region, spectra are plotted in reflectance. Spectra for the LRMCC coarse and fine particle samples are presented in Figs. 7 and 8, respectively. This wavelength region contains spectral features that are useful compositional indicators, most notably the Christiansen feature, a reflectance minimum and an emission maximum (e.g., Salisbury 1993). The position of the Christiansen feature is controlled by the position of the fundamental vibrational absorption feature, which is in turn controlled by a material’s composition. While the wavelength is physically controlled by the degree of depolymerization of the silicate material, the wavelength of the Christiansen feature has been shown to correlate with index measurements of material chemistry such as SCFM (Si/[Si + Ca + Fe + Mg]). The feature generally moves to longer wavelengths as materials become more mafic (SiO2 content decreases) (e.g., Salisbury and Walter 1989; Cooper et al. 2002).

The lunar mineral separate spectra behave as expected. The Christiansen features for the plagioclase separate spectra occur at shorter wavelengths, near 8 μm in all cases at both grain sizes. The pyroxene Christiansen features occur at longer wavelengths, at slightly greater than 8.5 μm. The olivine separate spectra have a Christiansen feature at a very slightly longer wavelength than the pyroxene spectra, as expected from olivine’s more mafic composition. They have composite features, with a second feature near 9.2 μm. The fine particle olivine separate spectrum (Fig. 8) differs from the coarse particle spectrum (Fig. 7) in that the longer wavelength “component” (approximately 9.2 μm) of the Christiansen feature becomes stronger, and the reflectance minimum of the spectrum shifts to this component of the feature. Otherwise, the fine particle sample spectra are similar to their coarse-grained counterparts, albeit with increased spectral contrast. They also have steeper slopes from about 7 μm to the Christiansen feature due to their higher reflectance values at shorter wavelengths. The ilmenite spectra lack a distinctive Christiansen feature in this wavelength region, and have higher reflectance throughout, particularly in the fine-grained sample spectra. Ilmenite does not have a Christiansen feature in the same wavelength region as the silicate minerals because as a Fe-Ti oxide, it does not have a Si-O vibrational absorption feature. A feature for ilmenite occurs at significantly longer wavelengths (approximately 12.5 μm). The bulk sample spectra contain composite features, due to the presence of multiple mineral species in the sample that have Christiansen features at different wavelengths.



This paper focuses on integrated analyses of petrography and reflectance spectroscopy. The mineralogy and petrography results are presented here in the context of the insight they provide for interpreting the reflectance spectroscopy results. More specific mineralogy and petrography results are discussed in the Supporting Information. The mineral separate compositions are presented explicitly, as they are directly relevant to the interpretation of the reflectance spectra. Mineral separate compositions were measured prior to crushing the samples to controlled particle size ranges for reflectance spectroscopy measurements. The process of crushing the mineral separates may impart a compositional bias, as certain phases, compositions or textures may be more or less resistant to crushing and thus may be concentrated into a particular size separate (e.g., Lin and Somasundaran 1972; Jari 1995). While we do not expect such a bias to be a substantial source of error, particularly for the particle sizes analyzed by the LRMCC, it must be considered in interpreting the results of this study.

Visible/NIR Bidirectional Reflectance

The reflectance spectroscopy results for this wavelength region are presented in Fig. 3 (coarse particles) and Fig. 4 (fine particles). The positions of the 1 and 2 μm pyroxene absorption features are consistent with pyroxene compositions for the 2-pyroxene separates prepared from each sample, as the higher-Ca pyroxenes have absorption features at longer wavelengths (e.g., Adams 1974; Cloutis and Gaffey 1991). The 15058 augite spectra appear to have slightly stronger 1.2 μm absorption features relative to the Apollo 15 pigeonite spectra, which could be suggestive of a faster cooling history (Klima et al. 2008), although a stronger 1.2 μm absorption could also be caused by the higher-Ca content of the augites (e.g., Burns 1993; Klima et al. 2007, 2008). However, such conclusions cannot be drawn without quantitative deconvolutions of individual pyroxene bands in the spectra (e.g., Sunshine et al. 1990; Klima et al. 2008) such as those presented below.

The pyroxene separates from both Apollo 17 samples are more compositionally distinct than those prepared from the Apollo 15 samples (see Fig. 2). Their reflectance spectra reflect this compositional diversity, as they exhibit clear differences in the position of the 1 and 2 μm absorptions. The Apollo 17 pyroxene spectra appear to have weaker 1.2 μm absorptions than those observed in the Apollo 15 sample spectra. On the whole, the Apollo 17 pyroxene separates have higher abundances of CaO than do the Apollo 15 separates, suggesting that in this case, the comparatively weak 1.2 μm bands are due to a slightly slower cooling history. We make this interpretation because the weak 1.2 μm absorptions cannot be attributed to less iron partitioning into the M1 site, as the high-Ca content would tend to cause more iron to partition into the M1 site, not less (Klima et al. 2008). The slow cooling interpretation is consistent with the results of petrographic analyses of the paired thin section samples and previous analyses of these samples, which found them to have cooled relatively slowly (Brown et al. 1975). However, confirmation of the relative strengths of the 1.2 μm absorption features in these spectra requires quantitative deconvolutions.

Several interpretations are possible for the broad visible wavelength absorption observed in the Apollo 17 augite separate spectra (observed most prominently in Fig. 3). These pyroxenes have relatively high TiO2 contents, roughly twice those observed for pyroxenes in the Apollo 15 samples. It has been suggested that the presence of Ti3+ cations in the pyroxene structure can cause such absorptions in the visible region (Burns et al. 1976; Burns 1993), although a Fe2+-Ti4+ metal–metal charge transfer is also a possible explanation (Loeffler et al. 1975; Burns et al. 1976). Ti in these pyroxenes is likely to exist in both +3 and +4 valence states, so multiple transitions could contribute to the observed absorption feature. The feature is slightly stronger in the augite separate spectra relative to the pigeonite spectra, and augites in the Apollo 17 high-Ti samples have elevated Ti contents because they appear to have coprecipitated with ilmenite found in a more Ti-rich environment than pigeonites (this is discussed in the Supporting Information). These pyroxenes exhibit sector zoning, and the augites typically have small inclusions of ilmenite (illustrated in Figs. 1f and 1h). An alternative interpretation is that the ilmenite inclusions produce the broad feature across the visible in the Ti-rich pyroxene separate spectra. Similar features have been observed in analyses of microcrystalline ilmenite-bearing glassy phases, and have been attributed to the microcrystalline ilmenite (Pieters and Taylor 1989; Tompkins and Pieters 2010). If ilmenite were responsible for this feature, the transition responsible would likely be the Fe2+-Ti4+ metal–metal charge transfer in the ilmenite structure, in which these are the two dominant cations (Loeffler et al. 1975). We favor the structural Ti interpretation due to the apparent high abundance of TiO2 in the pyroxene and lack of direct visual evidence for microcrystalline ilmenite, although resolution of the structural Ti in pyroxene versus microcrystalline ilmenite interpretation is beyond the scope of this article and likely requires additional analytical techniques.

The broad multicomponent absorption centered near 1.05 μm in the olivine separate spectra (Figs. 3 and 4) is consistent with the olivine’s composition (approximately Fo50) based on the methods of Sunshine and Pieters (1998) and Isaacson and Pieters (2008). As is commonly observed in lunar olivine spectra, the spectra also have slightly asymmetric absorption features near 2 μm. Such features are not typically observed in spectra of synthetic or terrestrial olivines. We attribute this feature primarily to the presence of small chromite inclusions in the olivine (Pieters et al. 1990; Isaacson and Pieters 2010), as spinels have characteristic, strong absorption features near 2 μm (Cloutis et al. 2004). The chromite is observed as small crystalline growths within olivine crystals, a common feature of lunar olivine (Dymek et al. 1975; Papike et al. 1998). Photons interacting with olivine also interact with the chromite grains without having to pass through void spaces between mineral grains. Common nonlinear models for radiative transfer in particulate media (e.g., Hapke 1993, 2008) do not account for this situation quantitatively, but can be used to indicate that such a situation has major effects on the overall reflectance properties of a material. Because olivine in 15555 typically intergrows with plagioclase and pyroxene, as illustrated in Fig. 1, it is possible that trace amounts of pyroxene and plagioclase exist in our olivine separate. While visual inspection of the olivine separate shows it to be relatively pure, olivine is not optically active at wavelengths around 2 μm, and it is possible that minor quantities of pyroxene contamination could contribute to such features. However, the character of the absorption and the lack of any pyroxene features near 1 μm, confirmed by the quantitative deconvolutions performed by Isaacson and Pieters (2010), point to chromite being responsible for the 2 μm feature in the olivine spectra.

The Apollo 15 plagioclase spectra (Figs. 3 and 4) have strong absorption features relative to other pure plagioclase separate spectra, likely due to the high Fe content (Bell and Mao 1973) of these separates (about 0.5 wt% FeO). The Apollo 17 plagioclase separate spectra have slightly weaker 1.3 μm features, likely due to their lower Fe content (about 0.3 wt%). All plagioclase spectra presented here exhibit additional absorptions at longer wavelengths. The long wavelength edge of the 1.3 μm absorption feature has lower reflectance than its counterpart on the short wavelength side for the plagioclase spectra presented here (most apparent in the coarse particle spectra presented in Fig. 3). Although the 1.3 μm feature clearly dominates, the plagioclase absorption feature is not simply a single crystal field transition in Fe2+. While most absorption properties of plagioclase are understood based on mineral physics (e.g., Bell and Mao 1973; Adams and Goullaud 1978; Pieters 1996), the long wavelength behavior may be related to Fe content and a separate absorption feature (Hofmeister and Rossman 1984; Cheek et al. 2009). Lastly, the 70035 plagioclase is a somewhat less pure separate than the 70017 plagioclase, as evidenced by a weak absorption at 1 μm superimposed on the wing of the plagioclase 1.3 μm absorption (most apparent in the fine particle spectrum in Fig. 4). We interpret this feature as a pyroxene absorption likely caused by minor pyroxene contamination of the plagioclase separate. The feature is stronger in the finer particle separate spectrum, likely because the strength of the plagioclase absorption decreases, and the more optically active pyroxene absorption becomes proportionally stronger.

Ilmenite acts as an opaque across visible wavelength when the grain size is significantly greater than the wavelength of the light being measured (i.e., grain size much greater than 1 μm). Thus, the ilmenite separate spectra are dark and relatively featureless through the visible and part of the NIR before showing increased reflectance beyond approximately 1.7 μm (seen more prominently in Fig. 4, but also apparent in Fig. 3). The magnitude of features in the ilmenite spectra is strongly dependent on particle size (they become more prominent with decreasing particle size, as evidenced by the greater spectral contrast in the ilmenite spectra in Fig. 4 relative to those in Fig. 3), but the overall spectral properties of these lunar ilmenites are consistent with measurements of synthetic ilmenite (Riner et al. 2009).

The bulk sample spectra exhibit largely mirrors the pyroxene spectra. Pyroxene is the dominant optical phase in these samples, so this is unsurprising. The effect of the range of pyroxene compositions on the bulk sample is observed in the broader 1 and 2 μm absorptions in the bulk sample spectra as compared to the individual pyroxene separate spectra. The prominent inflection near 1.2 μm seen in all bulk basalt spectra (Figs. 3 and 4) has many potential contributions. Pyroxene, olivine, and plagioclase all have absorption features in this region, although not all play equal roles in bulk sample spectra. Plagioclase has an absorption feature in this region, although it is quickly overwhelmed in mixtures with more strongly absorbing phases like pyroxene and olivine. Careful quantitative analysis (i.e., spectral deconvolution and quantification of the properties of individual absorption features) is required before drawing conclusions about the detectability of plagioclase in such spectra (Klima et al. 2008). The pyroxene absorption is the most dominant component, as the inflection is observed in olivine-rich and olivine-poor samples alike. The effect of olivine is apparent in a comparison of the 15555 bulk sample spectra with other bulk sample spectra; the proportionally stronger 1.2 μm feature in 15555 relative to the 1.2 μm features exhibited by the other bulk sample spectra is attributed to the contribution of the olivine absorption.

The Apollo 17 bulk sample spectra differ markedly from the Apollo 15 bulk sample spectra. The most prominent differences are decreased albedo, significantly weakened absorption features at 1 and 2 μm, and pronounced long wavelength (greater than 1.7 μm) reflectance upturns in the Apollo 17 bulk sample spectra. The curvature of the bulk sample spectra in the 2 μm region is caused by the reflectance upturn, and mimics an absorption feature. These properties are attributed to the abundant ilmenite, which exhibits a reflectance maximum near 1 μm, is generally opaque, and exhibits an upturn in reflectance beyond approximately 1.7 μm. These effects are observed in both coarse particle and fine particle spectra, although they are exhibited more clearly by the fine particle spectra in Fig. 4. The difference between spectra of the coarse and fine particle splits of the Apollo 17 bulk samples is one of the most prominent differences between particle size separates, so we have illustrated this difference explicitly in a separate figure. Figure 9 shows visible/NIR reflectance spectra of coarse and fine particle sizes for the 15058 bulk sample, 70017 bulk sample, and 70017 ilmenite separate. These spectra are also plotted in Figs. 3 and 4, but are replotted here to illustrate clearly some prominent effects of particle size differences. This figure illustrates the dramatic effect of ilmenite on reflectance spectra of basalts, as the principal difference between the low-Ti (15058) and high-Ti (70017) samples is the abundant ilmenite in 70017. The spectral effects of ilmenite described above are exhibited quite clearly in the 70017 bulk sample spectra, as is the much more pronounced effect of ilmenite at finer particle size (compare the heavy black lines; the only difference between the two is particle size, assuming no preferential sorting during the crushing process). The effects of ilmenite on reflectance spectra are similar to some of the optical effects of space weathering, which include decreased albedo, increased spectral slope, and weakened absorption features. All of these effects are observed when comparing the low-Ti (Apollo 15) to the high-Ti (Apollo 17) bulk sample spectra. These samples are optically immature, so these results hint at the complexity of addressing space weathering and maturity in ilmenite-rich regions of the Moon, and generally indicate that ilmenite is a major complication in interpretation of reflectance spectra (e.g., Hiroi et al. 2009).

Figure 9.

 Selected coarse and fine particle visible to near-infrared reflectance spectra. Sample spectra plotted include 15058 bulk (low-Ti, ilmenite-poor), 70017 bulk (high-Ti, ilmenite-rich), and 70017 ilmenite. The abundant ilmenite in the 70017 bulk sample causes suppressed absorption features, reduced overall albedo, and a pronounced “red” spectral slope at longer wavelengths. These effects are far more prominent at the fine particle size, suggesting that the effect of ilmenite on reflectance spectra of basalts is strongly dependent on texture and particle size, as discussed in the text.

NIR/Short Wavelength Mid-IR 2.5–8 μm (Volume Scattering)

Many spectra presented in Figs. 5 and 6 have broad absorptions at 3 μm. The feature near 3 μm is characteristic of an O-H vibrational absorption (Farmer 1974). We attribute this feature to adsorbed terrestrial water, which persists despite measurement in a dry air purged environment (Gibson and Moore 1972; Epstein and Taylor 1973; Dyar et al. 2010; Hibbitts et al. 2010). This feature is apparently the strongest in the plagioclase separate spectra. The 3 μm feature also is generally more apparent in coarse particle sample spectra (Fig. 5). A number of explanations for this observation are possible. First, plagioclase may be more susceptible to adsorption of water, perhaps due to a surface chemistry process. Additionally, although plagioclase is a nominally anhydrous mineral (it has no OH or H2O in its chemical formula), it has been shown to contain structural water and OH (e.g., Bell and Rossman 1992; Seaman et al. 2006), in some cases up to approximately 1000 ppm (e.g., Johnson and Rossman 2003, 2004), so the 3 μm absorption may be caused partially or entirely by internal water in the plagioclase grains. Alternatively, the apparent stronger absorption feature in plagioclase may be simply an effect of the relative transparency of plagioclase relative to the other phases analyzed. The mean optical path length is greater in a transparent phase, which increases the number of times light interacts with grain surfaces. If water is adsorbed on grain surfaces, additional surface interactions due to this increased mean optical path length would produce a proportionally stronger 3 μm absorption feature. The strength of the O-H absorption may also be anticorrelated with the depolymerization state of the Si-O bond in silicates (strongest absorption in plagioclase). The 3 μm absorption is the weakest for ilmenite, although this observation would be consistent with both the mean optical path length explanation and the Si-O depolymerization explanations.

Spectra of all separates show sharp absorption features near 3.4 μm, although the strength of these features is variable between spectra. These features are attributed to the C-H fundamental stretch, which is diagnostic of organic hydrocarbons (e.g., Cruikshank and Brown 1987; Gaffey et al. 1993; Clark 1999). The features are much stronger in spectra of separates from 15058 and 70035, likely because these samples were subjected to a preparation step in which they were processed in ethanol (CH3CH2OH) which caused some ethanol to remain in the separates as a trace contaminant. The weaker features in the spectra of separates from 15555 and 70017 likely are due to small amounts of contamination during sample preparation, but these samples were not processed in ethanol, explaining the weaker 3.4 μm absorption. The features are also weaker in the bulk samples, which were not subjected to the preparation step involving ethanol.

The pyroxene separate spectra show similar absorption properties to one another at these wavelengths. The position and strength of their absorption features do not show significant variation with mineral composition.

Absorption features in the olivine spectra between 5 and 6 μm are weak (slightly more apparent in Fig. 6) but highly diagnostic. These features have potential value for remote detection, as they provide an independent indication of the presence of olivine beyond the 1 μm crystal field absorptions (which can be complicated by the presence of other absorbing phases).

Plagioclase spectra show a weak but important absorption feature near approximately 4.4 μm. Because the 4.4 μm feature occurs in a wavelength region unaffected by the mafic crystal field absorptions, the presence of this feature may allow plagioclase to be detected unambiguously. The Apollo 17 plagioclase separate spectra are somewhat brighter than their Apollo 15 counterparts in the 3 to 4 μm region. The Apollo 17 plagioclase separates have slightly lower iron contents than the Apollo 15 plagioclase separates, although we cannot attribute unequivocally this brightness difference to variable iron contents.

The bulk sample spectra generally follow the behavior of the pyroxene spectra. The presence of olivine is apparent in the fine particle olivine-bearing bulk sample spectrum (15555), as evidenced most notably by the feature near 5.7 μm. The plagioclase absorption at 4.4 μm is weak-to-absent in the bulk sample spectra, suggesting that the modal abundance of plagioclase in these basaltic samples may be below the detection limit for this feature.

Mid-IR 7–9.5 μm (Volume–Surface Scattering)

As predicted by theory, plagioclase separates exhibit short wavelength Christiansen features, pyroxenes and olivine have features at longer wavelengths, and bulk samples show composite absorption features dominated by a variety of phases. These spectra, illustrated in Figs. 7 and 8, are consistent with previous work suggesting that this wavelength range is very useful for remote detection of plagioclase (e.g., Salisbury and Walter 1989); the plagioclase “feature” is readily apparent in the bulk sample spectra as well as in plagioclase separate spectra. The 15555 olivine separate spectra exhibit an unusual property, in that the reflectance minimum appears to shift between particle size separates. This behavior is not apparent in the bulk sample spectra, suggesting that the other phases dominate the bulk sample spectra. This apparent shift in the Christiansen feature position may be explained by a change in mineralogy due to preferential crushing. Regardless, it is not clear which phase is responsible for this behavior, as the Christiansen feature for spinels typically occurs at wavelengths beyond this wavelength region (Cloutis et al. 2004).


Mixing Models

One of the principal goals of reflectance spectroscopy is remote determination of planetary surface mineralogy. Visible to near-infrared reflectance spectra contain diagnostic mineral absorption features, as discussed above. Mixtures of minerals, such as those found in rocks and rock powders, produce reflectance spectra that are effectively a composite of the individual component mineral reflectance spectra. The quantitative relationship between the mixed spectrum and the individual endmember spectra is complex, however, and is an area of ongoing research (e.g., Mustard and Pieters 1987; Hapke 1993; Hiroi and Pieters 1994; Clark 1995; Shkuratov et al. 1999; Poulet and Erard 2004).

Mixture models can be classified into linear and nonlinear varieties. Linear models are the simplest, and assume that the endmembers are physically separated by more than the wavelength of the incident radiation and that photons can interact with only one endmember prior to being reflected. Linear models can be considered spatially akin to a checkerboard. Nonlinear models treat the complexities that arise when components are intimately mixed and photons can interact with multiple endmembers prior to being scattered back to a sensor.

Mixture models rely on accurate knowledge of the endmember spectra used. Most endmember spectra, regardless of their quality, are approximations, because the mineral composition and general character cannot match exactly those of the minerals found in the mixture being studied. The LRMCC samples and mineral separates presented here offer the ability to overcome this limitation, because the endmember minerals were extracted from the exact bulk sample being analyzed. The thin sections are directly linked to the slab samples, and portions of the slabs were used to prepare separates while other portions were retained as “bulk samples.” The detailed characterization of the samples, their component mineral composition, and their modal abundances represent a very accurate understanding of the “ground truth” which a mixture model seeks to estimate. These samples thus represent the “ideal” case to test mixture models: realistic mixtures with realistic mineralogical properties (as opposed to synthetic mixtures created from laboratory mixtures of pure endmembers), and precise knowledge of the endmember mineral spectra. Models that cannot reproduce the “truth” for the bulk samples need refinement. The number of unconstrained variables is significantly reduced (although not zero) compared to most other modeling situations. Spectral mixture modeling is complex, and full treatment of this subject for the LRMCC results is beyond the scope of the present work. However, we present an example set of analyses for one of the four samples presented here, 15058.

In the case of pyroxene separates for the LRMCC samples, modal abundances as reported from thin section analyses should not be equated with modal abundance of the prepared mineral separates. Modal abundances from the thin section were determined for mineral compositions, and separations were performed visually, so assumptions must be made to convert measured abundance with mineral separate abundance. The modal abundances of the specific mineral separates from 15058 were estimated by classifying the mineral separates (e.g., Brown/Green Pyroxene) as combinations of mineral compositions (e.g., orthopyroxene), for which abundances were measured. This classification was performed by comparing the composition of the pyroxene separates (Fig. 2, left) to the range of compositions observed in analysis of the thin sections (Fig. 2, right). Our assumptions are that Green Pyroxene = orthopyroxene + pigeonite, and that Brown Pyroxene = augite + Fe-pyroxene. This is not strictly true based on the mineral composition results (Fig. 2), but is sufficient for approximation (although this classification is a potential source of error for the mixture model results).

Initial example nonlinear fits calculated for one of the samples and mineral separates are illustrated in Fig. 10. These fits were calculated using the methods of Hiroi and Pieters (1994), and are discussed by Hiroi et al. (2009). Briefly, volume abundance of endmembers and surface roughness of the bulk sample were allowed to vary. Isotropic scattering was assumed. The fit in Fig. 10a allows effective grain size of the endmembers to vary (“unconstrained”), while the fit in Fig. 10b constrains the grain sizes to a constant value of 72 μm (“constrained”). More detail regarding the model parameters and assumptions is provided by Hiroi and Pieters (1994). The line and symbol styles in Fig. 10 are largely the same as used in the reflectance spectroscopy figures (Figs. 3–8). The fit is plotted as “x” symbols. The fit in Fig. 10a has a calculated root mean square deviation (RMSD) of 0.55% over the full wavelength range used for the fit (0.4–2.5 μm), and the fit in Fig. 10b has a RMSD of 1.68% over the same wavelength range.

Figure 10.

 Examples of preliminary nonlinear spectral unmixing models using the coarse particle 15058,276 bulk sample and mineral separate spectra. Line styles and symbols follow the approach employed for the reflectance spectroscopy figures (Figs. 3–8), with some minor changes. The bulk sample spectrum is no longer plotted as a heavier line, and its cross symbols have been hidden. The calculated fit is plotted as “x” symbols. The unmixing method follows the approach of Hiroi and Pieters (1994). a) Fit with effective grain size for the endmembers (mineral separate spectra) allowed to vary (“unconstrained”). This fit has a calculated root mean square deviation (RMSD) of 0.55% over the full wavelength range used for the fit (0.4–2.5 μm). b) Fit with effective grain size for the endmembers constrained to a constant value of 72 μm. This fit has a RMSD of 1.68% over the wavelength range used for fitting (0.4–2.5 μm).

The resulting modeled abundances for the fits in Fig. 10 are listed in Table 4, along with the modeled effective grain sizes for unconstrained fit (Fig. 10a). Also presented in Table 4 are the actual modal abundances as determined from analysis of the thin section sample, based on the groupings described above. Because only three endmembers were used to perform the unmixing, the sum of the three modeled abundances is 100%. In determining the modal mineralogy of the thin section, more phases were reported than two pyroxenes and plagioclase. To facilitate comparison between modeled and measured abundances, we also report the thin section abundances of the two pyroxene groupings after normalizing the two pyroxene plus plagioclase components to a total of 100% (analogous to the spectral unmixing results). The quality of the fits is generally good, and fitting generally is able to reproduce the modal abundances determined from the thin section, especially considering the assumptions described above. The abundances of Brown Pyroxene and Plagioclase differ significantly from the thin section values in both fits, indicating that despite the quality of these samples and spectra, questions about the mixture model remain. The deviations in effective grain size for the unconstrained fit merit further investigation, as the samples were crushed and prepared with near-identical methods. For example, plagioclase is modeled to have a much lower grain size than the pyroxene separates. It is possible that differing physical properties for the various separates lead to varying particle sizes in the crushed bulk sample and mineral separates, although it is also possible that the optical properties of plagioclase are not fully accounted for by these models.

Table 4.   Modal mineral abundances from thin section analysis and spectral fitting for 15058.
 Green pyroxeneBrown pyroxenePlagioclase
  1. aSee discussion in the Mixing Models section.

  2. bFree refers to the fit shown in Fig. 10a, in which effective grain size is allowed to vary. Constrained refers to the fit shown in Fig. 10b, in which effective grain size was constrained to a constant value of 72 μm.

Thin sectiona29.333.830.1
Norm. thin sectiona31.436.332.3
Spectral fitting, freeb32.152.315.6
Modeled effective grain size (μm)55.558.530.5
Spectral fitting, constrainedb295120

Quantitative Absorption Modeling

To characterize fully the specific crystal field absorption features in the spectra acquired in this study, quantitative analyses of the absorption band properties are required. Band properties such as band position, band strength, and band width are diagnostic of mineral/rock composition and are the basis of remote mineralogical analyses with reflectance spectroscopy. To use these spectra in mixture models and in interpretation of remotely acquired data, quantification of these parameters is essential.

Currently, one of the best methods for quantitative assessment of absorption features is the Modified Gaussian Model (MGM) (Sunshine et al. 1990). The MGM models the shape of a spectrum by applying a series of Gaussian curves to fit the shapes of individual absorption features. The properties of the fitted Gaussian curves can be interpreted as the properties of the individual, deconvolved component absorption features in the modeled spectrum. This approach removes ambiguity in interpretation of absorption features in reflectance spectra; the positions and strengths of the absorption features are clear, and interpretations of mineralogy based on those spectra are more reliable.

Example fits to one of the LRMCC bulk samples are presented in Figs. 11 and 12. The 15058 bulk sample spectra are the simplest example for deconvolution because the sample lacks abundant olivine or ilmenite, which complicate the other bulk sample spectra. While this sample is the simplest of the four in mineralogy, the deconvolutions are not simple. We performed fits for both the coarse particle (Fig. 11) and fine particle (Fig. 12) separate spectra. We performed two fits for each spectrum, one using single pyroxene absorptions for the 1 and 2 μm features and one using two pyroxene absorptions for each feature. Using two absorption features is reasonable based on the range of pyroxenes found in the sample, but when fitting an “unknown” spectrum, one should always use the least number of bands required for a numerically and physically reasonable fit. In other words, additional bands should not be added to reduce the “error” of the fit unless there is a sound reason to do so based on mineralogy and mineral physics. As can be seen in the RMS error of our fits, the single pyroxene fit is significantly worse than the two pyroxene fits for both size separates. Additionally, the single pyroxene fits required the use of a linear continuum slope to produce a stable fit. The two pyroxene fits show improved RMS error in both cases. The fine particle separate spectra showed improved RMS error for both single and two pyroxene fits compared to their coarse particle counterparts, suggesting that the bands may be saturated in the coarse-grained separate spectra. When two absorptions are used at 1 and 2 μm, the shorter-wavelength absorption is modeled to be stronger in both grain size separates, consistent with the mineralogy of the sample (Sunshine and Pieters 1993; Klima et al. 2007, 2008). These MGM fits provide quantification of the specific absorption band properties observed in the spectra: band width, band strength, and band position. Such quantification is critical for reliable application of these spectra as ground truth, and will be performed for the bulk sample and selected mineral separate spectra acquired in this study.

Figure 11.

 Modified Gaussian model deconvolutions of the 15058,276 bulk sample coarse particle separate spectrum. Deconvolution examples are shown using one (a) and two (b) absorption features to fit the 1 and 2 μm pyroxene absorption features. The principal pyroxene absorptions are plotted in black and dashed (in the case of the two pyroxene fit) lines, and the 1.2 μm pyroxene absorption in dotted lines. Other absorption features are plotted in gray.

Figure 12.

 Modified Gaussian model deconvolutions of the 15058,276 bulk sample fine particle separate spectrum. Deconvolution examples are shown using one (a) and two (b) absorption features to fit the 1 and 2 μm pyroxene absorption features. The principal pyroxene absorptions are plotted in black and dashed (in the case of the two pyroxene fit) lines, and the 1.2 μm pyroxene absorption in dotted lines. Other absorption features are plotted in gray. RMS error is plotted offset from 0 by 0.2 for clarity.


The lunar rock and mineral characterization consortium (LRMCC) has conducted coordinated mineralogy/petrography/spectroscopy analyses of a suite of lunar basalts. Slab samples were used to prepare particulate separates of bulk samples and mineral separates, and paired thin sections were used for mineralogy/petrography analyses. The LRMCC data are important for lunar science in many contexts. The spectra represent controlled ground truth for optical remote-sensing instruments, which rely on such ground truth measurements to conduct mapping. The samples analyzed by the LRMCC are unweathered, and provide an excellent complement to the results of the lunar soil characterization consortium (LSCC), which analyzed the weathered component of lunar surface materials in a suite of lunar soils. The spectra also represent some of the best constraints on spectral mixture models yet produced, as the mineral endmembers are produced directly from the rock sample being investigated (“unmixed”), and their compositions are well known. Quantitative analyses of spectral mixing and absorption features are ongoing. The combined data sets from the LSCC and LRMCC represent an invaluable resource for current and future lunar science and exploration efforts.


Acknowledgments— The authors gratefully acknowledge the support of NASA NLSI contract number NNA09DB34A as well as Cosmochemistry grants NNG05GG15G (C. M. P.) and NNG05GG03G (L. A. T.). The NASA RELAB is supported as a multiuser facility under grant NNG06GJ31G. Constructive and thorough reviews by Timothy Glotch and M. Darby Dyar improved the quality of this manuscript substantially.

Editorial Handling

Dr. A. J. Timothy Jull