We used spectral indexing and linear deconvolution to compare thermal infrared emission spectra of Fo91, Fo68, Fo53, Fo39, Fo18, and Fo1 olivine samples to Mars Global Surveyor Thermal Emission Spectrometer (TES) data over low-albedo regions of Mars. The Fo91, Fo68, Fo53, and Fo39 spectral end-members were confidently identified on Mars, a range of compositions wider than inferred from Martian meteorites. Small (less than hundreds of square kilometers) occurrences of the Fo91 spectral end-member are present in the rims of the Argyre and Hellas impact basins and may represent Martian mantle materials. The Fo68 spectral end-member is common throughout the highlands, chasmata, outflow channels, and Nili Fossae region. The Fo53 spectral end-member occurs in eastern Syrtis Major, the Nili Fossae region, and smooth-floored craters of the highlands. Although less abundant than Fo68 and Fo53, the distribution of the Fo39 spectral end-member suggests that some olivine on Mars is more Fe-rich than olivine in Martian meteorites. Global maps of olivine show that (1) materials containing 10–20% of olivine are common in the southern highlands of Mars, (2) olivine is most common near the topographic dichotomy boundary, and (3) olivine becomes uncommon near the poles suggesting that it may be influenced by topography and/or latitude (climate). Olivine is found in early Noachian to Amazonian terrains, some of which may be coeval with phyllosilicate and sulfate deposits detected by OMEGA implying that any early Noachian wet period of Mars' climate history may have been globally inhomogeneous or insufficient to weather the olivine that remains today.
 Olivine is an important rock-forming mineral on Earth, and its presence and composition can be used to place constraints on the origin, evolution, and erosion of its host rock. It is observed on the Moon [e.g., Pieters, 1982; Tompkins and Pieters, 1999] and many asteroids [e.g., Binzel et al., 1997; Lucey et al., 2002] using remote sensing techniques, and has been measured in lunar meteorites [e.g., Papike et al., 1998] and chondrites [e.g., Brearley and Jones, 1998]. Olivine is a component of many Martian meteorites including the Shergottite EETA79001 and most Nakhlites, and it comprises up to ∼60% of some lherzolitic Shergottites (e.g., ALH A77005) and 90% of Chassigny [Mason, 1981; McSween and Treiman, 1998; McSween et al., 2006, and references therein]. The crystallization of olivine plays a critical role in the chemical differentiation of magma chambers, and its composition and associated minerals record the chemistry of its parent magma, as well as its subsequent evolution due to differentiation, reequilibration, melting, or alteration.
 Olivine is a nesosilicate mineral found in mafic and ultramafic igneous rocks, and has strong spectral features in the thermal infrared (∼2000–200 cm−1). The molecular structure of olivine contains isolated silica tetrahedra separated by two distinct metal ion sites, which can be filled by Mg, Fe, Ca, and Mn [Klein and Hurlbut, 1985]. The most common solid-solution series of olivine ranges from Mg-rich forsterite (Mg2SiO4) to Fe-rich fayalite (Fe2SiO4). The Mg/Fe content of olivine within this series is denoted by its forsterite number (Fo #), which is defined as molar Mg/(Mg + Fe) × 100. Spectral variation in thermal infrared observations of olivine has been documented by several authors and correlates with its Mg/Fe content [e.g., Duke and Stephens, 1964; Burns and Huggins, 1972; Salisbury et al., 1991; Fabian et al., 2001; Hoefen et al., 2003; Koike et al., 2003; V. E. Hamilton et al., The thermal infrared spectra of the olivine series: Forsterite to fayalite, manuscript in preparation, 2008], a fact that can be exploited to determine the Mg/Fe content of olivine using astronomical measurements and remote sensing data of planetary surfaces.
 The presence of olivine on Mars is inferred from analyses of thermal infrared emission data collected by the Mars Global Surveyor Thermal Emission Spectrometer (TES). Hoefen and Clark  searched the TES data set for spectral features characteristic of different compositions of olivine, and presented a preliminary global map of olivine by composition. Their work culminated by showing the compositional variation of olivine in eastern Syrtis Major and a large regional occurrence of olivine in the region surrounding the Nili Fossae [Hoefen et al., 2003]. They identified the olivine in these areas as compositions of Fo∼40–70. Bandfield  used a linear least squares fitting technique to map the distribution of minerals, including olivine, on Mars. He modeled the TES data set using a library of spectra containing two olivine compositions, Fo91 and Fo1, and found geographic concentrations of olivine spectral shapes around the Nili Fossae and north of Argyre Basin. Hamilton et al.  modeled the TES data set with the global spectral surface types 1 and 2 of Bandfield et al. [2000b] and spectra of Martian meteorites, including Chassigny and ALH A77005, which are dominantly composed of Fo68 olivine [Prinz et al., 1974; McSween et al., 1979]. Hamilton et al.  found spatial concentrations of spectral shapes like those of the olivine-rich meteorites in many locations across the midlatitudes and most extensively near the Nili Fossae.
 Olivine on the surface of Mars is also inferred from other spacecraft data. Visible near infrared (VNIR) data from the Mars Express Observatoire pour la Minéralogie, l'Eau, les Glaces, et l'Activité (OMEGA) instrument was used to map the presence and location of surficial olivine-bearing deposits [Bibring et al., 2005; Mustard et al., 2005; Poulet et al., 2007]. Direct observations made by the Spirit and Opportunity rovers give strong evidence for olivine in basalts measured by the Mössbauer spectrometer and MiniTES instruments [Christensen et al., 2004b; Morris et al., 2004], and visual inspection of crystallographic shapes in rock surfaces imaged by the Microscopic Imager (MI) suggests that olivine exists as megacrysts [McSween et al., 2004].
 The occurrence of olivine has been used to argue for water-poor histories for large portions of Mars [e.g., Christensen et al., 2003; Hoefen et al., 2003]. Compared to other silicates, olivine has the least polymerized structure and can break down easily in the presence of weathering agents [Goldich, 1938; Deer et al., 1992]. Within the olivine Mg-Fe solid solution series, it has been shown that increasing the Mg content bolsters its resistance to chemical weathering and lowers its kinetic dissolution rate by nearly one order of magnitude over the range from fayalite to forsterite [e.g., Wogelius and Walther, 1992; Pokrovsky and Schott, 2000; Stopar et al., 2006]. Therefore, the Mg content of olivine plays an important role in the calculated timescales during which water may have been present on the Martian surface and influences its preservation [Stopar et al., 2006].
 The global distribution, composition, and abundance of olivine on Mars is important information that can be used to synthesize data from its meteorites, ground observations, and models of mantle evolution. For example, estimates based only on Martian meteorites constrain the bulk of olivine on Mars to Fo∼60–90, but ground measurements from the Mars Exploration Rovers (MERs) have normative compositions as low as Fo51 [McSween et al., 2004], suggesting a larger range may exist.
 This work examines the global distribution and composition of olivine on Mars inferred from TES data using spectral indexing and deconvolution. We use these techniques together in order to increase the area on Mars covered by TES data and obtain quantitative mineral abundance information from TES spectra. We verify the results with Thermal Emission Imaging System (THEMIS) images, compare and contrast our results with previous analyses, and make inferences about the origin and evolution of olivine-bearing materials on Mars.
 The TES instrument is a Fourier transform Michelson interferometer that collects data from ∼1700 to 200 cm−1 (∼6–50 μm) in either 143 or 246 channels [Christensen et al., 1992]. The TES consists of an array of six detectors, each of which has a spatial footprint of 3 × ∼8 km on the surface of Mars [Bandfield, 2002]. For this work we utilize single-scan (143 channels, 10 cm−1 spectral sampling) apparent emissivity spectra derived from the calibrated radiance spectra calculated for each detector and observation, except where explicitly noted. For a complete description of the TES instrument and radiometric calibration, see Christensen et al. [2001a].
 We selected TES spectra for this study using specific quality criteria including derived surface temperatures >250 K, emission angles <10°, albedo values <0.2 (relatively dust free surfaces), atmospheric dust opacities <0.2 (measured at 9 μm), and water ice opacities <0.15 [Bandfield et al., 2000b; Pearl et al., 2001]. Water ice opacities >0.10 are higher than used by previous studies [e.g., Bandfield et al., 2000b; Bandfield, 2002], but including these spectra allowed us to increase spatial coverage over the Martian surface. We single out and discuss the results of our analyses from the high water ice opacity spectra in section 4.1. We did not include any data that showed risks of major or minor phase inversions, or data collected at times when the spacecraft's solar panels or high gain antenna were moving quickly. In order to avoid spectral regions dominated by atmospheric CO2, as well as decreased signal-to-noise at the ends of the measured range, we constrained the data to 73 points per spectrum distributed between 1302–825 and 508–240 cm−1 (∼8–12 and 20–42 μm) [Bandfield et al., 2000b; Christensen et al., 2000c; Hamilton et al., 2001]. In this paper, we refer to data from 1302 to 825 cm−1 as the high-wave number region and data from 508 to 240 cm−1 as the low-wave number region.
 A noise anomaly, possibly associated with spacecraft vibrations, affects the high-wave number portions of TES data above approximately Orbit Counter Keeper (OCK) 7000 and worsens with increasing OCK [Bandfield, 2002; Hamilton et al., 2003]. This anomaly is difficult to detect automatically, and it can result in erroneous mineral detections in full-spectrum mixture modeling of the data [Hamilton et al., 2003]. For our spectral modeling analyses we only included spectra acquired prior to OCK 8000, ∼2.3 × 106 spectra, to increase spatial coverage of Mars but avoid the most anomaly prone data. For maps made with spectral indices, we included all the spectra in the TES database up to OCK 22,000, ∼9.3 × 106 spectra, as our indices use only the low-wave number regions of TES spectra, which appear to be unaffected by the anomaly [Bandfield et al., 2004].
2.1. Olivine Samples
 To select olivine spectra for our analyses, we examined thermal emission spectra of ten olivine samples ranging from Fo91 to Fo1 (Table 1). The Fo91 and Fo1 spectra were obtained from the Arizona State University (ASU) spectral library [Christensen et al., 2000b] and represent samples with a particle size fraction of 710–1000 μm measured with ASU's Nicolet 670 FTIR spectrometer. Eight intermediate compositions of olivine from the Kiglapait layered mafic intrusion in north coastal Labrador were provided to us by S. A. Morse, and their compositions are characterized by Morse . Particulate samples with grain sizes >100 μm are appropriate for the dark regions of Mars [e.g., Ruff and Christensen, 2002], and we sieved each of the Kiglapait samples into discrete particle size fractions between 106 and 246 μm and selected the largest size fraction in each sample for analysis (Table 1). We measured the olivine samples using the University of Hawaii's Nicolet 470 FTIR spectrometer [Hamilton and Lucey, 2005], and used the method described by Ruff et al.  to derive emission spectra from the measured voltages. We used a spectrum of atmospheric water vapor to subtract out many of the spectral features due to humidity in the sample chamber from each measured spectrum. Laboratory data of all ten olivine samples were acquired at 2 cm−1 spectral sampling but were convolved to the 10 cm−1 spectral sampling of the TES data to determine appropriate band minima and maxima resolvable by the TES instrument (Figure 1). Similar to previous studies, the thermal emission spectra of the olivines in our suite have recognizable minima that linearly decrease in wave number with increasing Fo# (Figure 1) [Salisbury et al., 1991; V. E. Hamilton et al., manuscript in preparation, 2008].
Table 1. Sample, Fo Number, Particle Size, and Source of Olivine Samples Used in This Studya
Particle Size, μm
Although the WAR-RGFAY01 in the ASU spectral library is listed as fayalite, the Fo number (Fo #) here is based on our own microprobe analyses of the sample.
 A spectral index is an efficient way to identify spectral shapes in a large data set by mathematically reducing spectral features of interest to a single number [e.g., Kauth and Thomas, 1976; Richardson and Wiegand, 1977; Jackson, 1983; Christensen et al., 2000a]. Typical indices use a few spectral bands over features of interest and a simple mathematical formula to identify the possible presence of those features in each spectrum. Critical absorptions are spectral features consistently present in all of the spectra of a mineral type [Hamilton, 2000], and we created spectral indices to search for a low-wave number critical absorption of olivine that shifts in position as a function of Mg/Fe, thereby distinguishing between its compositions (V. E. Hamilton et al., manuscript in preparation, 2008). Similar to Christensen et al. [2000a] and Bandfield et al. , our spectral indices follow the form of
The index value (Ifo#) represents the relative depth of the spectral feature defined by comparing its emission minimum (ɛλ2) to its flanking emission maxima (ɛλ1 and ɛλ3). Ifo# is designed to be >1.0 when the feature is present and ≤1.0 if the feature is absent or inverted. Ideally, each ɛλ is the average of measured apparent emissivity values for a pair of adjacent TES channels in order to reduce point-to-point noise present in some data.
 We selected spectral bands in the 500–300 cm−1 region of TES data because it is relatively free of features associated with suspended dust and atmospheric CO2 gas, and it serves as a spectral window through which surface emissivity features are observed, even in data that are not atmospherically corrected [Ruff and Christensen, 2002; Bandfield et al., 2004]. Olivine has relatively narrow spectral features in this wave number region that shift with Fo# and allow us to distinguish between multiple compositions (Figures 1 and 2) . However, the narrowness of the olivine's low-wave number spectral features prohibits the use of two adjacent emissivity values at every ɛλ. Where available, we used averages of two adjacent emissivity values, but it was commonly necessary to use only one value at each ɛλ. Although using a single critical absorption characteristic of olivine for all of the indices was preferable, the spectral feature characteristic of olivine in the low-wave number region contains two critical absorptions (Figure 2) (V. E. Hamilton et al., manuscript in preparation, 2008). The higher-wave number critical absorption becomes indistinguishable in the spectra of olivine samples with Fo<25, and we had to target the lower-wave number critical absorption for those indices.
 We created six spectral indices to identify the full range of Mg/Fe in olivine, and the band positions used for each index are listed in Table 2. The compositional accuracy of our spectral indices is limited by the spectral resolution of the data, because the difference in band minima between similar compositions of olivine is not large and the minima of similar compositions of olivine directly overlap at the 10 cm−1 spectral sampling of the TES instrument. Specifically, the low-wave number emissivity features of Fo68 and Fo60; Fo39, Fo35, and Fo25; and Fo10 and Fo1 are indistinguishable using spectral indices (Figure 2). We used the emission minima of olivine's low-wave number critical absorptions measured under laboratory conditions to determine a linear function of wave number position versus Fo# for each critical absorption. We then calculated the range of olivine compositions predicted to have a minimum at the ɛλ2 TES channel(s) of each index, and titled the index by the full range of Fo content that it potentially identifies.
Table 2. Wave Numbers and Channels for Each Index Valuea
Values listed in parentheses are the TES channels corresponding to the wave number listed. If one value for ɛ is listed, the emissivity value at that band was used in the index calculation. If two values are listed, the average emissivity value for the two bands was used. Wave numbers are in units of cm−1.
370.3 (22)/380.9 (23)
370.3 (22)/380.9 (23)
359.8 (21)/370.3 (22)
253.9 (11)/264.5 (12)
 The low-wave number regions of thermal infrared spectra exhibit features of many minerals other than olivine, and we determined the minerals to which our indices may also be sensitive by calculating the olivine index values for all of the spectra in the ASU mineral library. This analysis identified nonolivine mineral spectral shapes that might return a high olivine index value. Minerals that produced index values of at least half that of the associated olivine are listed in Table 3 and include mostly carbonates along with a few silicates. Carbonates have not been confidently identified in abundances >5% on Mars [Bandfield et al., 2003]; however, clinopyroxene (diopside) is common in TES deconvolution results [e.g., Bandfield et al., 2000b; Christensen et al., 2000c] and may produce false identifications in the Fo58–74, Fo42–57, and Fo25–41 indices. On the other hand, no mineral listed in Table 3 produced an index value greater than that calculated for the indexed olivine, and false identifications of pyroxene in the spectral index analyses should be recognized by comparing its spatial distributions to the deconvolution analyses of this work and that of Bandfield  and the maps of pyroxene-dominated meteorite spectra by Hamilton et al. . Quartz was shown to be ambiguous with the Fo25–41 index and has been detected on Mars but in only a few well-known locations that are easily identifiable [Bandfield et al., 2004]. We did not test the index calculations against combinations of two or more mineral spectra, which generally have less well defined individual spectral features because of interference between the component spectra but may also add linearly to produce features that return high index values.
Table 3. Minerals That Give High Index Values for Each Olivine Index by Compositiona
Multiple samples of these phases produced index values higher than the threshold.
 Unfortunately, the lower-wave number critical absorption we needed to use for the Fo11–24 and Fo0–10 indices encroaches into the <300 cm−1 region, which can have highly variable emission features attributed to atmospheric water vapor and ice [e.g., Smith et al., 2000; Smith, 2002]. We calculated the index values for these spectral indices in order to test whether spectral indices could be used to locate occurrences of these compositions even with atmospheric interference. However, we expect most variations in these indices to be dominated by atmospheric components.
 For TES data, the wavelength-dependent radiance measured at the spacecraft is the sum of the radiance of the atmosphere and the radiance of the surface, which is attenuated by atmosphere. Mixing of the atmospheric and surface spectral components of thermal infrared data is generally nonlinear; however, their respective contributions to TES data can be closely derived by a linear approximation if the Martian surface temperature is relatively warm (>250 K) compared to the atmosphere during the observation [Bandfield et al., 2000a; Smith et al., 2000]. These conditions are met by all of the TES data selected for our analyses. The atmospheric dust and water ice spectral shapes derived from TES data by target transformation are relatively invariant except for some features attributed to CO2 and H2O gas [Bandfield et al., 2000a], and they can be used as spectral end-members in the deconvolution algorithm to simultaneously derive the fractional contributions of the atmospheric and surface spectral end-members to each TES spectrum [Bandfield et al., 2000a; Smith et al., 2000].
 We calculated the surface mineral abundances associated with each TES spectrum by normalizing their individual fractional contributions so that the surface components sum to unity [Smith et al., 2000]. Similarly, the atmospheric spectral components can be multiplied by their derived fractional contributions and subtracted from the observed apparent emissivity spectrum to retrieve an atmosphere-removed (surface) emissivity spectrum [Smith et al., 2000], which we use illustrate the spectral features in some olivine-bearing areas. We made maps both of fractional contribution and surface abundance for individual spectral end-members and combinations of spectral end-members by binning the irregularly spaced TES observations according to latitude and longitude. In general, we report the mean value in each bin.
 It was not necessary to include all ten of the olivine samples measured here as spectral end-members in the deconvolution analyses because some compositions (e.g., Fo68 and Fo60) have very similar band shapes and minima at the 10 cm−1 spectral sampling of TES (Figures 1 and 2). The linear independence of such similar end-members is very weak, and including them in the spectral library can result in variations to their derived fractional contributions that track minor natural differences, noise, and systematic errors rather than significant spectral differences. The corollary is that, like the spectral indices, each olivine spectral shape represents a range of similar compositions dependent on the spectral sampling of TES and the differences in spectral features correlated with Mg/Fe content. However, the compositional range represented by each spectral end-member is difficult to estimate using linear deconvolution because it is dependent on spectral features in the entire range of the spectral end-member. We used only six of the ten olivine spectra (Fo91, Fo68, Fo53, Fo39, Fo18, and Fo1) as spectral end-members in deconvolution analyses because they have relatively unique spectral shapes among the olivine series, and we list them by their specific Fo# recognizing that they do not represent exact compositions found on Mars.
 Other spectral end-members used in the deconvolution analyses included spectra of pure minerals and glasses, one nearly monomineralic Martian meteorite, atmospheric components, and a blackbody spectrum (unit emissivity at all wave numbers) (Table 4). The mineral spectral end-members in our set were measured in laboratory conditions [Christensen et al., 2000b] and selected to identify a broad range of olivine compositions as well as commonly associated minerals and weathering products [Michalski et al., 2005; Rogers and Christensen, 2007]. We adjusted the spectral contrast of two potential weathering products, saponite and illite, to match other “solid” clays in the ASU spectral library because those samples had smaller and larger apparent grain sizes than the rest of our clay spectral end-members. A full description of the sample differences and spectral adjustment for the saponite and illite samples is given by Rogers and Christensen . We included a spectrum of Martian meteorite ALH 84001 to better approximate orthopyroxene shapes known to be present on Mars [Hamilton et al., 2003]. We utilized atmospheric end-member spectra derived from TES data representing water vapor, water ice clouds, suspended dust, and atmospheric CO2 [Bandfield et al., 2000a; Smith et al., 2000]. Variations in the total contrast of thermal infrared spectra can result from particle size effects [e.g., Aronson et al., 1966; Salisbury and Wald, 1992; Moersch and Christensen, 1995] as well as lava vesicularity [Ramsey and Fink, 1999], and the fractional contributions of the end-member spectra in the best fit model may not sum to 1.0. To account for this possibility, we included a blackbody spectral end-member, which allows the deconvolution algorithm to better match mixture spectra of different total contrasts [Hamilton et al., 1997].
Table 4. End-Members Used for Deconvolution Analyses
 The error in determining most mineral end-member abundances by deconvolution of thermal emissivity data collected in laboratory conditions is estimated to be ±5–10% [Feely and Christensen, 1999; Christensen et al., 2000c; Wyatt et al., 2001]. This estimate includes errors accrued from spectral variability in the precision of the thermal infrared instruments [Ramsey and Christensen, 1998], normalizing out the blackbody spectral end-member [Hamilton et al., 1997; Feely and Christensen, 1999], and determination of the modal mineralogy of laboratory rock samples using a microprobe [Wyatt et al., 2001]. An additional ∼5% uncertainty is added when laboratory data are resampled to TES spectral sampling [Hamilton et al., 2001], and uncertainties may be even higher when deriving mineral abundances from their fractional contributions in TES spectra that have large atmospheric contributions. However, the uncertainty for each mineral is different depending on the depth, shape, and location of the spectral bands [Christensen et al., 2000c], and minerals with narrow and distinct band minima may be detected at relatively low fractional contributions [Bandfield, 2002].
 Compared with the rest of the ASU spectral library, olivine has particularly strong spectral features and its detection limit on Mars has been suggested to be as low as 5% fractional contribution [Hamilton et al., 2003]. As a qualitative example, in Figure 3 we show a representative TES spectrum, its atmosphere-removed spectrum, and models generated by using different fractional contributions of olivine to show that differences of 5% can be visually identified as erroneous. The best fit model to the TES spectrum derived by deconvolution contains a Fo91 fractional contribution of ∼10%. We added or removed 5% increments of the Fo91 spectral shape to the best fit model, renormalized the other end-member contributions and recreated the model with the new fractional contributions. A visually poor model fit is identifiable at 900 cm−1 in the TES emissivity spectrum and can be clearly distinguished at 1100, 900, and 325 cm−1 in the atmosphere-removed surface spectrum.
 We generated two more empirical examples to estimate the detection limit of olivine and place lower bounds on the believable abundances in our deconvolution analyses. In the first example, we compared the spectral contrast of the olivine features found in observed spectra to those of pure olivine spectra in our samples by using index values. In theory, the fractional contribution of the olivine spectral shape should be linearly correlated with the index values calculated for each spectrum, i.e., as the olivine concentration increases, the feature deepens and the index value increases. This correlation provides an estimate of the fractional contribution of olivine, assuming that the depth of the feature is based only on fractional contribution of olivine without blackbody/particle size effects or the presence of other phases that might increase or decrease the spectral contrast. For our index maps, the minimum index value empirically found to correlate with olivine-like spectral shapes, 1.015, corresponds to pure olivine spectra with contrasts of 8%, 12%, and 13% for the Fo91, Fo68, and Fo53 spectral end-members, respectively. In practice there is only a weak relationship between Fo58–74 and Fo42–57 spectral indices and deconvolved fractions of the Fo68 and Fo53 spectral end-members (Figure 4). In the second example, we deconvolved several TES spectra identified by index values within the range we empirically estimated as valid. The resulting fractional contributions had a mean olivine abundance of 7.8 ± 3.7% for the Fo91, Fo68, and Fo53 compositions. The minimum value obtained in this calculation, ∼4%, represents the lowest limit at which olivine spectral shape was detected independently by the spectral index and deconvolution. Both examples bolster our confidence in detection of olivine at abundances as low as 5%.
2.5. Validation of Isolated Detections
 We applied decorrelation stretches (DCS) to data from the Mars Odyssey Thermal Emission Imaging System (THEMIS) to validate potential olivine-bearing materials identified in small numbers of TES spectra. The THEMIS instrument consists of infrared and visible multispectral imagers that have spatial resolutions of 100 m and 18 m, respectively (see Christensen et al. [2004a] for a complete description). The infrared imager of THEMIS records data in 10 bands distributed between 1500 and 650 cm−1 and complements the TES because it collects data in the same wave number range at higher spatial resolution, albeit with decreased spectral resolution. The DCS enhances the spectral variation in a multiband image [Gillespie et al., 1986; Gillespie, 1992], and, using THEMIS bands 8, 7, and 5 (centered at 850, 905, and 1070 cm−1) mapped as red, green, and blue, olivine appears purple in the processed image [Hamilton and Christensen, 2005; Rogers et al., 2005]. It is not yet clear if different compositions of olivine can be distinguished in THEMIS spectra, but DCS images are a quick way to visually scan an area and verify the presence of olivine suggested by a high index value or deconvolution analysis of TES data.
 Additionally, we visually inspected many distinct occurrences of olivine identified by TES spectra to verify the presence of the olivine spectral shape in the apparent emissivity and atmosphere-removed emissivity TES spectra. In some cases we made spectral ratios of olivine-bearing materials to nonolivine bearing materials in order to confirm that the band minima in the ratio matched those of the olivine composition identified by indices or deconvolution.
3.1. Spectral Index Analyses
 We calculated the six spectral indices shown in Table 2 for ∼9.3 × 106 TES spectra (OCKs 1583 to 22,000) meeting our quality constraints, as described above. We performed an extensive manual evaluation of individual spectra with index values >1.0 and determined that values between ∼1.015 and 1.060 represent what we believe to be valid detections of the olivine-like shape for most cases. Although any index value >1.0 should indicate the presence of a feature in that spectrum, small point-to-point emissivity differences, which may be due to noise or minor spectral components other than olivine, also produce values slightly >1.0. Unfortunately, this range also excludes from identification some spectral mixtures that contain olivine, but whose olivine spectral features are diminished because of other spectral components in the mixture. For example, the surface type 1 spectral shape derived by Bandfield et al. [2000b], which likely contains 5–10% olivine [Rogers and Christensen, 2007], produces olivine index values >1.0, but <1.015. In a few cases, some index values between 1.015 and 1.060 may also be erroneous where they are correlated with the same detector number(s) in the same OCK and sequential Incremental Counter Keepers (ICKs). Values >∼1.060 are strongly correlated with orbit track-dependent noise and are not reliable.
 We removed from the data set any index values outside of the valid range as well as those suggesting detector-correlated noise, and we mapped the remaining values into a spatial context (Figure 5). Strictly interpreted, index values do not represent abundance, but merely identify the presence of olivine-like spectral contributions to the total signal measured by the TES. Therefore, we mapped only the locations of values that fell within the acceptable range. Because we calculated six indices for every spectrum, it was possible for a single spectrum to be identified by more than one index within the valid range. The index values for different compositional indices are not directly comparable as they use different band positions, and the mean index value of the spectra changes with each index (Figure 4); for example, most of the Fo42–57 index values are >1.0 while most of the Fo75–100 values are <1.0. In order to present the spatial distributions of valid olivine index identifications and eschew the bias of higher index values associated with a particular index we show only the highest Fo# identified for each spectrum in this map.
 The Fo75–100, Fo58–74, and Fo42–57 spectral indices identify spectra in spatially coherent patterns and are shown in Figure 5 as red, yellow, and blue points, respectively. The Fo75–100 index is the least ambiguous of the six indices with respect to other minerals in the ASU spectral library (Table 3) and identifies 161 spectra in the region surrounding the Nili Fossae, 76 spectra in the northern portion of Argyre rim, and 22 spectra in the northern portion of Hellas rim. Spatial groupings containing fewer spectra are located in two craters in Aurorae Planum, and in Ganges Chasma, Simud Vallis, and Oxia Palus. Possible olivine-bearing materials identified by only one or two spectra, but validated using THEMIS DCS images (described in section 3.2), are located in Chryse Planitia, northern Terra Sirenum, and Terra Tyrrhena. The Fo58–74 index identifies a large grouping of spectra in the area surrounding the Nili Fossae, similar to previous work using spectral feature identification [Hoefen et al., 2003; Martínez-Alonso et al., 2006]. The Fo58–74 index also identifies many spectra inside the southwest rim of Isidis Basin, in two craters in Aurorae Planum, in Ganges and Eos Chasmata, southern Xanthe Terra, and on the northern portions of the Hellas and Argyre Basin rims. Small groupings of one to five TES spectra are located in Chryse Planitia and throughout the southern highlands of Terrae Tyrrhena and Cimmeria. The Fo42–57 index identifies many more total spectra than the Fo75–100 and Fo58–74 indices combined; however, it also shows less spatial coherence suggesting that it is influenced by noise in the spectra. Spatial groupings of spectra identified by the Fo42–57 index are found in the region surrounding the Nili Fossae and two craters in Aurorae Planum. Groupings of fewer spectra are found distributed across Syrtis Major, Xanthe and Tyrrhena Terrae, and inside craters in northern Noachis and Cimmeria Terrae.
 The Fo25–41, Fo11–24, and Fo0–10 indices also produce values between 1.015 and 1.060, but the spatial distributions of spectra they identify suggest that they are inaccurate, and we do not show them. Specifically, the distribution of spectra identified by the Fo25–41 index lacks spatial coherence, perhaps because it uses only one emissivity value at each ɛλ making it prone to spectral noise. The Fo11–24 index also lacks spatial coherence, and appears to be dominated by highly variable emission features associated with water vapor and ice in the <300 cm−1 region [e.g., Smith et al., 2000; Smith, 2002]. The Fo0–10 index produces values in the valid range that exhibit spatial coherence in Terra Meridiani, Aram Chaos, and the Valles Marineris. However, those regions and the spatial outlines of the distributions of spectra identified correspond directly to regions previously identified as hematite-rich [Christensen et al., 2001b]. A similar correlation was pointed out by Bandfield , who noted the lack of independent evidence for olivine in those areas. Hematite was not identified a priori as a potential source of ambiguity in Table 3, because the hematite sample in the TES spectral library (BUR-2600), which we tested, has minima at slightly higher wave numbers than the spectral shape derived for hematite on Mars by Christensen et al. [2000a].
 As an additional component to the index analyses, we explored the possibilities of using the double-scan (286 channels, 5 cm−1 spectral sampling) apparent emissivity TES spectra to confirm and improve the Fo# determination. The increased spectral resolution would be particularly useful for further refining the Mg-rich (Fo75–100) olivine identification. Some double-scan TES spectra do exist over the occurrences of olivine identified by the Fo75–100 and Fo58–74 olivine indices in northern Argyre and Terra Tyrrhena, respectively, but the spectra are too few (<10) and too noisy to be useful. As a broader test case, we attempted to find and average a number of double-scan TES spectra over Nili Fossae and generated one of the most promising spectral shapes by averaging over 20 spectra (covering a total of nearly 500 km2) in the area. The overall shape of olivine is obvious in the resulting average spectrum, and, consistent with the single scan TES data, it does appear qualitatively to be more like Fo58–74 than Fo75–100 or Fo42–57. However, the exact band minimum, which is necessary to refine the Fo# determination beyond what is possible from the single-scan TES data, was still obscured by the increased noise inherent to double-scan TES data, especially in the low-wave number features. Averaging even more spectra may allow these data to be used to refine Fo# determinations, but it also increases the danger of averaging locations that have multiple compositions of olivine and defeating its purpose.
3.2. Deconvolution Analyses
 We compiled fractional contribution maps of surface mineral classes (feldspar, high-calcium pyroxene, low-calcium pyroxene, olivine, silica/phyllosilicate, and sulfate) by summing the contributions of their individual end-members to each TES spectrum, binning the data at 2 pixels per degree, and reporting the mean fraction for each bin/pixel (Figure 6). These data represent the mean fractional contributions of each mineral group to the apparent emissivity spectra, which include contributions from atmospheric and blackbody (spectrally neutral) components. Data are mapped wherever they meet the criteria for this study, e.g., no data with albedo values <0.2 were available over large portions of the Tharsis region, Arabia Terra, Utopia and Elysium Planitae, Vastitas Borealis, or the central Hellas Basin. Mean fractional contributions higher than the ±10–15% detection threshold of TES are present in all of the maps. The feldspar, pyroxene, olivine and sulfate maps broadly correlate with the surface type 1 maps of Bandfield et al. [2000b], whereas the silica/phyllosilicate map correlates with both surface type 1 and surface type 2 maps. In general, high fractional contributions for all mapped phases are found over Syrtis Major, Xanthe and Tyrrhena Terrae, and portions of Cimmeria and Sirenum Terrae. Fractional contributions >15% are common in the feldspar and silica/phyllosilicate maps. An RMS error map stretched from 0.00 to 0.01 emissivity exhibits RMS errors that are lowest near the equator and increase toward the higher latitudes, but nearly all areas have RMS values <0.01 suggesting that they are relatively well modeled using the end-member set chosen for this study. RMS error values >0.01 emissivity are associated with orbit tracks implying a spacecraft or atmospheric anomaly for those data rather than poorly modeled surface mineralogies. Detailed analyses of the nonolivine maps are ongoing and will be useful in examining refinements to the mineralogy of the surface enabled by including a wider compositional range of olivine spectral end-members.
 In order to map the distribution of olivine in terms of surface abundance, we normalized the fractional contributions of the olivine spectral end-members to exclude the atmospheric and blackbody contributions to the spectra [Hamilton et al., 1997; Feely and Christensen, 1999; Smith et al., 2000]. We summed the abundances of all the olivine spectral end-members, binned the resulting total olivine abundance data at 2 pixels per degree and present maps of the mean value and standard deviation within each bin/pixel (Figure 7). The total mean olivine abundance map shows that materials containing 10–20% olivine are relatively common in the low-albedo regions of Mars and have clear regional variations (Figure 7a). For example, Syrtis Major shows less total olivine than Terra Tyrrhena directly to the south, the western third of Syrtis Major has mean olivine abundances almost ∼10% less than central and eastern Syrtis Major, and the region northeast of Syrtis that contains the Nili Fossae has olivine abundances >20%. There are notably low standard deviations over most of Syrtis Major (Figure 7b) suggesting that olivine is modeled with similar abundances for the TES spectra within each bin. Broad regions of elevated olivine abundance with values >10% are observable in Tyrrhena, Cimmeria, Sirenum, and Xanthe Terrae. Within these regions, the map shows discrete occurrences (1–5 pixels across) with up to 20% total olivine abundance. The northern and eastern rims of the Argyre and Hellas Basins also have localized enhancements of olivine; however, they also have elevated standard deviations, suggesting that both high and low abundances of olivine are modeled for spectra present in those bins and exposures of olivine may be smaller than the bin size in these regions. The data also identify regions of low olivine abundance, e.g., the olivine abundances <5% across much of central and eastern Acidalia and the region of southern highlands west of Hellas basin.
 Approximately 6% of the pixels mapped from our linear deconvolution results show fractional contributions >0.05 for more than one composition of olivine suggesting that, at least in some locations, multiple compositions may be spatially coexistent. One possible explanation is simply that olivine of different compositions are spatially distinct but adjacent in the same TES footprint (∼3 × 8 km); e.g., the boundary between two abutting lava flows. Another possibility is that olivine of different compositions is spatially mixed and interspersed; e.g., if olivine grains of different compositions are collecting together in a sedimentary basin. A third possibility is that different compositions of olivine are present in the same grains, e.g., chemical zoning, such as the thin Fe-rich rims over Mg-rich cores found in some Martian meteorites [McSween and Treiman, 1998]. Alternatively, the olivine detected may be of an intermediate composition relative to the end-members used and require more than one spectral end-member to properly model. These scenarios cannot be distinguished with remote sensing data alone and linear deconvolution requires an assumption of “checkerboard” mixing, which may be violated by zoned mineral grains. For simplicity in this work, we interpret multiple olivine compositions modeled in the same TES spectrum as intergrain compositional variations.
 We created abundance maps for the six olivine compositions in our end-member set (Fo91, Fo68, Fo53, Fo39, Fo18, and Fo1) to illustrate their individual distributions on Mars (Figure 8). Similar to the total olivine abundance map, we mapped the mean value in each 2 pixels per degree bin after normalizing the fractional contributions to remove the atmosphere and blackbody end-members. These maps illustrate not only the abundance of the olivine spectral end-members as a function of location, but also the wide range of Mg/Fe content in olivine interpreted from TES data. Globally, the two intermediate compositions, Fo68 and Fo53 dominate these maps at abundances >5%. However, the Fo39 map also shows spatially coherent occurrences >5% when binned at this resolution. (Abundance maps binned at 16 pixels per degree for the six olivine end-members used in this study showing more detail are provided online at http://www.higp.hawaii.edu/datasets/koeppen/JGR_TES_olivine.)
 Inspection of the 16 pixels per degree Fo91 spectral end-member abundance map reveals that only a few small locations on Mars show spectral signatures matching the most forsteritic sample in our olivine suite, and most occurrences are not visible in the 2 pixels per degree maps. For example, the largest area modeled with Fo91 olivine is <900 km2 but exhibits Fo91 abundances >10% in some spectra. THEMIS DCS images of the area show olivine-like signatures throughout the hills, massifs, and rim of a small impact crater (32.3°W, 44.2°S) that excavated into the Argyre rim materials (Figure 9). Other occurrences of the Fo91 spectral end-member appear as groupings of 4–8 pixels located almost exclusively in and around the northern rims of Argyre and Hellas basins.
 The Fo68 spectral end-member is the most common olivine spectral shape identified on Mars and has a distribution similar to the results of Hamilton et al. , who used spectra of the Martian meteorites, including olivine-rich ALH A77005 and Chassigny, as spectral end-members. In both 2 and 16 pixel per degree maps, many spatially coherent occurrences of Fo68 correlate with surface features at all scales including craters, channel floor materials, and volcanic plains. As in previous studies, the Fo68 spectral end-member is the dominant composition of olivine within the materials in the region surrounding the Nili Fossae, and many of the pixels in that area show >10% Fo68 content even at 2 pixels per degree (Figure 8) [Hamilton et al., 2003; Hoefen et al., 2003]. Fo68 also occurs on the Isidis basin rim south of Syrtis Major [Hamilton et al., 2003; Hoefen et al., 2003], in knobs, mesas, and channel floor materials within many of the chasmata of the Valles Marineris, in the floor of Ares Vallis [Rogers et al., 2005], north of Argyre basin, and in the floor materials of two craters in Aurorae Planum [Hamilton et al., 2003]. The Fo68 spectral end-member map shows broad spatially coherent occurrences north of Argyre and in Noachis and Tyrrhena Terrae (though notably less over Tyrrhena Patera). It is commonly observed in discrete occurrences between −60 and 30°N, and small groupings of 5–10 pixels occur in isolated mesas, crater floor materials of Sirenum, Aonia, Tyrrhena, and Cimmeria Terrae, Noctis Labyrinthus, Amazonis, north and northwest Hellas, east Argyre, Chryse Planitae, and in the ridged plains southeast of Elysium Planitia. The Fo68 spectral end-member is also modeled globally in many isolated pixels, but we did not manually validate each occurrence using THEMIS data.
 The Fo53 olivine end-member map shows occurrences in numerous areas throughout the equatorial and southern highlands (Figure 8). The Fo53 spectral shape is abundant in the region surrounding the Nili Fossae, though pixels modeled with Fo53 olivine >10% are spatially confined to the northeastern portion of the area similar to the distributions of Hoefen et al. . Small occurrences <2 pixels across are located in the northern rims of Argyre and Hellas Planitiae; however, Fo53 is not as pervasive as Fo68 in these regions, and only rarely do both compositions have high values in the same pixel. Similarly, we observe the Fo53 spectral end-member in the Valles Marineris floor materials and in the channel floors of Chryse Planitia coincident with Fo68, but the abundances of Fo53 are typically less than the abundances of Fo68 in these regions. Finally, the Fo53 spectral shape is present in numerous craters throughout the Sirenum, Noachis, Tyrrhena, and Cimmeria Terrae, many of which contain smooth interior materials at THEMIS scales (e.g., Figure 10).
 The Fo39 spectral end-member is less common than the Fo68 and Fo53 end-members in terms of areal distribution and abundance (Figure 8), and the Fo39 spectral shape is identifiable in many of the atmosphere-removed spectra over these occurrences (e.g., Figure 11). The deconvolution results show broad occurrences of Fo39 over Syrtis Major and much of the equatorial and southern highlands. Aonia, Xanthe, Noachis, Tyrrhena, and Cimmeria Terrae show Fo39 abundances >5% and some locations have abundances >10%. Almost all discrete enhancements of Fo39 occur in the smooth-floored interiors of craters.
 In general, derived abundances for the Fo18 spectral end-member in TES data are less spatially coherent than those calculated for Fo68, Fo53, and Fo39, and in fact appear somewhat randomly distributed at both the 2 and 16 pixel per degree resolutions. Many values >10% follow orbit tracks suggesting that atmospheric effects or noise may complicate identification of Fo18 (Figure 8). Finally, a number of Fo18 abundances >5% are commonly associated with identifications of other olivine end-members, making us less confident in the compositional precision of those results. We examined THEMIS DCS images over a number of small groupings of pixels that identified Fo18, but none of the locations we checked showed the olivine spectral shapes in both TES and THEMIS data. Fo18 is one of the few compositions found in TES data over high northern latitudes, and olivine was observed in OMEGA data of lobate crater ejecta in those regions [Bibring et al., 2005]. However, even those locations identified by OMEGA data did not produce evidence of unambiguous olivine detections of any composition at the scale of TES observations. The lack of validated Fo18 spectral end-member detections from TES does not preclude the presence of similar compositions on Mars, but our manual inspection of many individual TES spectra did not reveal any unambiguous occurrences of this spectral end-member.
 Fayalite, represented by the Fo1 spectral end-member, was modeled only at very low abundances except for a few isolated pixels (Figure 8). Although every case was not checked, most abundance values >5%, e.g., a group of pixels between Terra Tyrrhena and Hellas Planitia, did not correlate with olivine-like spectral shapes in THEMIS data. However, ten TES spectra from OCK 6918 over an intercrater plain in the northern rim of Hellas Basin (62°E, 28°S) show very weak (∼3%) fractional contributions of olivine that expand to >10% after conversion to abundance (Figure 12). These spectra appear noisy and show contributions less than our typical threshold of confidence, but the derived surface temperatures for the spectra are ∼275 K and a Fo1-like shape appears observable in the surface emissivity spectra (Figure 12b). Spectral variations identified in THEMIS DCS images over the potential occurrence indicate the presence of olivine in knobs and plains. This detection requires more analyses to ensure a confident detection of fayalite, but it is compelling enough to state that the presence of fayalite on the surface of Mars deserves further study.
3.3. Distribution of Olivine With Latitude, Elevation, and Age
 We made histograms of the pixels mapped with >5% fractional contribution of each olivine spectral end-member to more clearly visualize the distribution of olivine by composition. First, we binned the data as a function of elevation to look for any trends in global stratigraphy. To avoid areal biases, we normalized each 100 m elevation bin by the total number of pixels containing usable TES data at that elevation, and the resulting data represent the percentages of pixels modeled with a specific olivine spectral end-member greater than our empirically determined detection limit (Figure 13). In general, the Fo68, Fo53, Fo39, and Fo18 spectral end-member histograms crest at −1000 m and taper off toward higher and lower elevations (the nonnormalized elevation histogram of all pixels mapped in this study crests at 2000 m suggesting that areal biases have been successfully removed). These histograms show that olivine is relatively common between −2500 and 2500 m, a range that includes most of the southern highlands and excludes the northern lowlands and most of Hellas basin. The only discrete stratigraphic enhancement in olivine fractional contribution is most visible in a 500 m thick peak at −2300 m to −1800 m in the Fo53 histogram, but it is also visible in the Fo68, Fo39, and Fo18 histograms. This elevation primarily corresponds to Xanthe Terra, a region the spectral end-member maps show to have elevated olivine of all compositions in both discrete (crater and channel floors) and regional occurrences. The histogram of the Fo91 spectral end-member map shows that it is the least commonly modeled olivine composition at any elevation. The Fo1 spectral end-member is the second least common olivine modeled over the detection limit, less than 2% of pixels were mapped with >0.05 Fo1, and its histogram does not vary with elevation indicating that the pixels mapped as Fo1 are randomly distributed.
 Second, we binned the data as a function of latitude to look for potential climatic influences on the surface mineralogy and normalized each one-degree latitude bin by the total number of pixels containing usable TES data at that latitude (Figure 14). The histograms of the Fo68, Fo53, Fo39, and Fo18 spectral end-member maps all show that olivine is most common (4–12% of the pixels mapped) between 60°S and 30°N, which correspond primarily to highland regions. The Fo68 spectral end-member is the most common olivine mapped and contributes >0.05 to over 12% of the spectra at 10°S, tapering off toward the higher and lower latitudes. The peak in the Fo68 histogram at ∼22°N corresponds to the latitudes containing the Nili Fossae region. The Fo18 histogram shows the highest percentages of spectra containing olivine at latitudes between 60°S and 45°S, and >30°N; however, we reiterate that none of the Fo18 olivine identified by linear deconvolution or high spectral index values could be manually verified. Like the latitudinal histograms, the Fo91 and Fo1 spectral end-member maps show very few fractional contributions of olivine >0.05.
 We plotted the locations where olivine was modeled over geologic maps of Mars to roughly estimate the age of the materials in which they reside [Scott and Tanaka, 1986; Greeley and Guest, 1987]. The Fo91 spectral end-member, present in the rims of the ancient basins, lies almost exclusively within the Nplh and Nm units, the oldest geologic units on Mars. In the northern Hellas basin the Fo91 signature also appears to occur on younger Hesperian plains (Hr) encroaching from Tyrrhena Patera, but inspection of DCS images show that the olivine signatures are commonly present in kipukas, older hills and massifs that are surrounded by smooth, younger looking lava flows. The Fo68 spectral end-member was modeled in materials with a wide range of mapped ages, but it was identified many times in Noachian cratered highlands and Hesperian channel floor units (e.g., Nplh, Nple, Npl1, Npl2, Hch, and Hcht). The age of the channel floor materials are particularly ambiguous because they may have been emplaced earlier in Martian history and exposed to impact cratering only in the Hesperian. The Fo53 and Fo39 spectral end-members are found throughout the southern highlands in a variety of Noachian and Hesperian aged terrains; however, both compositions also appear at least locally concentrated within craters lying in Noachian terrains. These craters commonly have smooth floors and appear to be filled in by lavas or sediment that are younger than the Noachian Fo68-dominated intercrater plains by superposition. Although quantitative crater counts have not been done within the smooth-floored craters, typical examples (e.g., 10 and 11) have only a few small (<2 km diameter) craters on their surfaces suggesting that their fill is Amazonian in age [Hartmann and Neukum, 2001; Hartmann, 2005], which coincides with the timing of volcanism implied by most of the Martian meteorites excluding Chassigny (180 Ma to 1.4 Ga) [McSween and Treiman, 1998].
4.1. Comparisons to Previous Work
 A few authors have explored the distribution of olivine on Mars using spectral index approaches. Hoefen et al.  conducted a thermal infrared study of the composition of olivine in the region containing the Nili Fossae using spectral feature identification, which is essentially a complex spectral index. They found slightly more Fe-rich olivine compositions (Fo∼70–40) around the Nili Fossae than our index analyses (Fo∼100–45), which was expected because we mapped the highest Fo # implied by valid index values. However, the olivine compositions found by Hoefen et al.  in the Nili Fossae region are virtually identical to those found by our deconvolution analyses (Fo∼75–50), and the spatial distributions they mapped match up quite well with our deconvolution maps. Searching for smectite clays, Ruff and Christensen  mapped the distribution of materials on Mars that have an emission minimum at 530 cm−1, and one such material is olivine. The position of the 530 cm−1 feature is outside the TES spectral range that we analyzed by deconvolution because it is influenced by atmospheric CO2, and the 530 index map gives a somewhat independent look at the distribution of materials that include olivine-rich locales [Ruff and Christensen, 2007]. However, although many of the small occurrences seen in their 530 index map are also observed in our 16 pixels per degree total olivine fractional contribution and abundance maps, some areas highlighted by the 530 index are mapped as relatively olivine poor (e.g., Syrtis Major) suggesting that methods that discriminate multiple spectral features (e.g., spectral feature fitting or linear deconvolution) are necessary to confidently identify olivine-rich areas.
 The fractional contribution maps of Figure 6 are directly comparable to the mineral maps of Bandfield . In general, the results correlate rather well, however some differences exist. First, our results indicate fractional contributions of total olivine >20%, whereas the olivine map of Bandfield  shows enhancements only slightly >10%. This difference is almost certainly due to the increased number of olivine spectra in our end-member set versus those of Bandfield , who only had access to spectra of Fo91 and Fo1 olivine compositions. Second, we modeled fractional contributions of 10–20% low-calcium pyroxene end-members throughout the southern highlands (Figure 6) compared to ∼5% fractional contributions modeled by Bandfield . This may be due to our inclusion of pigeonite and ALH 84001 spectral end-members, which were not used by Bandfield . High abundances of low-calcium pyroxene have not been a common result of previous studies using TES data; however, recent work by Rogers and Christensen  also modeled low-calcium pyroxene as 5–15% and it may be more common than previously thought. Additionally, studies using OMEGA data have concluded that many regions in the Noachian highlands contain low-calcium pyroxene [Bibring et al., 2005; Kanner et al., 2005; Mustard et al., 2005], and many areas that contain high-calcium pyroxene are best modeled by a combination of both high- and low-calcium pyroxene [Mustard et al., 2005].
 We included TES data that exhibited a larger range of water ice opacities (<0.15) than many previous studies including those of Bandfield  (<0.10). Additionally, Rogers et al.  suggest that water ice opacities as low as 0.06 can affect the surface emissivity spectra. Less than 2.5% of the spectra selected for our analyses had water ice opacities >0.10, but we verified that including these data would not adversely affect our mapped results. To do this, we created global fractional contribution and abundance maps of olivine using only spectra with water ice opacities from 0.10 to 0.15, and we compared them to maps made using spectra with water ice opacities <0.10. There were very few (and very small) differences in the resulting values and spatial relationships of olivine making us confident that including those spectra do not significantly distort the global-scale distribution of olivine on Mars.
 Our mean total olivine abundance map (Figure 7a) is similar to the olivine abundance map of McSween et al. . However, our map shows occurrences of olivine in the northern lowlands (e.g., Kasei Valles), across Terra Tyrrhena, and in central Noachis Terra that do not appear in their study. Many differences likely stem from our inclusion of the Fo53 olivine spectral end-member, which was not available to McSween et al. , and our inclusion of a larger number of TES spectra.
Rogers et al.  mapped the distributions of fine-scale differences in TES-derived global surface types, and Rogers and Christensen  modeled the mineralogy of these global surface types using both linear deconvolution [Ramsey and Christensen, 1998] and least squares fitting with nonnegativity constraints [Lawson and Hanson, 1974]. Rogers and Christensen  used a group of olivine spectral end-members similar to McSween et al. , but they too only reported only the total derived olivine content. Similar to the results of our work, these studies mapped surface types modeled with high olivine contents (10–15%) in Tyrrhena, Cimmeria-Iapygia, Hesperia, and Meridiani surface types modeled with low olivine contents (<5%) in the high latitudes such as Solis Planum and Northern Acidalia.
Poulet et al.  mapped the distribution of forsterite and fayalite on Mars using spectral parameters (i.e., spectral indices) that utilized a 1-μm band absorption in OMEGA data to infer the presence of olivine. They used OMEGA data to identify a number of olivine occurrences between −50 and 50° at a spatial resolution of 2–5 km/pixel (∼12–32 pixels per degree) [Poulet et al., 2007], and most of the olivine occurrences they identified appear to be corroborated by our study. They did not identify regional enhancements of olivine; however, by using spectral indices alone, olivine contained in mixtures may not be accurately identified [Poulet et al., 2007] and modeling the abundance of olivine requires radiative transfer modeling [Poulet and Erard, 2004]. Poulet et al.  appears to infer more abundant Fe-rich olivine on Mars than our study; however, the near infrared spectra of forsterite and fayalite are very similar, especially if the grain size of fayalite is smaller than the grain size of forsterite [Poulet et al., 2007]. On the other hand, the distribution of Mg-rich olivine inferred from OMEGA data in the northeastern rim of Argyre [Poulet et al., 2007] matches very well with the distribution of the Fo91 spectral end-member that we infer from spectral indices and linear deconvolution modeling in that area.
Mustard and Cooper  searched the Syrtis Major region for spectral variations in TES data which might correspond to slope differences observed in NIR data from the Imaging Spectrometer for Mars (ISM). ISM observations from the eastern portion of Syrtis Major have a relatively strong spectral slope compared to the western portion that have been interpreted as the presence of ferric cement and fine coatings of dust [Mustard et al., 1993; Murchie et al., 2000], or penetrative oxidation [Mustard and Cooper, 2005]. Mustard and Cooper  interpreted virtually no differences in the TES data between the two parts of Syrtis Major leading them to suggest that dust may have been redistributed between the two instruments' scientific campaigns. However, our maps show up to 10% more total olivine in the eastern portion of Syrtis Major than in the western portion implying that at least some TIR spectral differences are evident in TES data. Furthermore, if relatively Fe-rich olivine (a mix of the Fo68, Fo53, and Fo39 spectral end-members) is more abundant in the east, perhaps it provides a mechanism to fuel penetrative oxidation in that region, giving rise to the NIR spectral slopes observed in ISM data.
 There is not a clear correlation between the total olivine abundance nor Fe-content in olivine from our maps and the distribution of Fe on the surface from Gamma Ray Spectrometer (GRS) data [Taylor et al., 2006]. For example, GRS data show that Syrtis Major contains ∼5% more Fe than Terra Tyrrhena directly to the south, a trend mirrored by higher Fo39 contents over Syrtis and higher Fo68 and Fo53 contents over Terra Tyrrhena. On the other hand, Tyrrhena and Noachis Terrae show variations in Fe from GRS data that do not spatially correlate with maps of the total olivine or its individual compositions. Similarly, the fractional contributions of other Fe-bearing minerals such as pyroxene do not match up well with the distribution of Fe observed by GRS. The lack of correlation suggests that the top 100–500 μm of the surface measured by the TES may not be representative of the meter or so depth measured by GRS, or that the Mg/Fe content in olivine (and perhaps other minerals) is independent of the bulk composition of the near surface. One possibility that might explain this discrepancy is that oxides or weathering products enriched in Fe are more abundant in some regions, thereby increasing the total Fe content independent of the observed olivine content or its compositions.
4.2. Implications of Latitude, Elevation, and Age Correlations
 Our analysis of olivine as a function of elevation reveals that, regardless of the data biases based on our albedo constraints, olivine is much more common in the midelevations of the southern highlands than in the northern lowlands, Hellas Basin, or Tharsis bulge. Additionally, olivine is most common near the hemispherical dichotomy boundary at −1000 m, an elevation that includes the floors of many outflow channels and related features, suggesting that olivine is more common in the once buried materials that have been exposed there. Xanthe Terra, which has even more channels and outflow features at a slightly lower elevation, has a strong peak in the Fo53 elevation histogram (Figure 13) that may also stem from materials exposed in channels, or olivine-rich materials eroding from higher elevations. In either case, it is difficult to tell if these enhancements represent a semiglobal olivine-rich layer or merely the region where olivine is most commonly exposed resulting from relatively recent geologic events. Interestingly, many of the regions at these elevations are subjected to relatively strong surface winds [Joshi et al., 1997; Fenton and Richardson, 2001], especially during periods of high obliquity [Armstrong and Leovy, 2005], which could work both to expose olivine-rich surfaces and mechanically weather olivine grains in situ.
 The latitude histograms of olivine (Figure 14) strongly suggest that there is at least some latitudinal control on the distribution of olivine. Olivine of any composition is virtually nonexistent at latitudes poleward of 30°N, which corresponds to the highland-lowland dichotomy boundary. If the bedrock materials of the relatively young northern lowlands are inherently less mafic than those of the southern highlands [Bandfield et al., 2000b], olivine may have never been present at all in the north. Alternatively, if olivine was present in the northern lowlands, it may now be heavily altered or covered in dust [Bibring et al., 2006]. Olivine is also extremely uncommon at latitudes poleward of 60°S suggesting that olivine is not merely correlated with southern highland materials. If we disregard the histogram of the Fo18 spectral end-member (and the peak in its latitude histogram at 50°S, because this composition could not be manually verified), then olivine becomes significantly less common poleward of 45°S, the latitude at which the porous Martian regolith is modeled to contain steady state ground ice [Mellon and Jakosky, 1995]. This region is characterized the constant (and balanced) process of sublimation of ground ice to the surface and condensation of water vapor from below [Mellon and Jakosky, 1995], which could aid in the alteration of olivine grains in the southern latitudes. Distinguishing between these scenarios will require lithologic observations and mapping of northern lowland bedrock to determine if olivine was ever present, or is present and hidden, and/or kinetic modeling of olivine dissolution by water in an icy environment that includes sublimation and condensation.
 The age of the olivine deposits can be used to place constraints on the origin and evolution of olivine as well as the climatic history of Mars. Olivine has been used to argue for a long-lived dry climate for much of Martian history [e.g., Christensen et al., 2003; Hoefen et al., 2003]; however, the presence of hematite, phyllosilicates, and sulfates indicates that at least some regions on Mars may have harbored episodic surface and/or groundwater [Christensen et al., 2001b; Squyres et al., 2004; Bibring et al., 2005]. A key discovery of the OMEGA instrument is the presence of phyllosilicate deposits in the ancient intercrater plains near the Nili Fossae that are overlain by olivine-bearing flows from Nili Patera [Bibring et al., 2006; Mustard et al., 2007]. Bibring et al.  interpret this stratigraphy as evidence of a global, long-lived “wet” climate during the early Noachian followed by a globally dry climate from the late Noachian until present-day. However, we found olivine exposed in a number of locations mapped as early Noachian (e.g., Argyre Planitia, Hellas Planitia, and many intercrater plains) suggesting that some olivine-bearing materials may have been coeval with the phyllosilicate deposits mapped by OMEGA. If the olivine-bearing materials and phyllosilicates were coincident, it implies that the early Noachian, wet climatic conditions proposed by Bibring et al.  were not globally homogenous, or at least were not so wet as to alter all of the early Noachian olivine. More detailed geologic mapping of olivine-bearing materials and phyllosilicate deposits is necessary to more clearly reveal the geologic distribution and chronology of the two climate regimes.
4.3. Possible Origins of Olivine on Mars
 Our maps can be used to identify relatively small occurrences of olivine, constrain their composition, and distinguish broad geochemical provinces. In turn, the distributions and compositions can give clues to the provenance and evolution of olivine and its host rock. In this section we explore possible origins of different olivine compositions on Mars.
 The most forsteritic spectral end-member used in this study, Fo91, was found in small occurrences within the rims of the Argyre, Hellas, and Isidis basins using both spectral indexing and deconvolution. Although this spectral end-member represents a range of Fo# compositions (see section 2.3), it is clear that it indicates a more Mg-rich composition of olivine than is found in most Martian meteorites and nearly all of the Martian surface in general as observed by remote sensing. The nature of forsteritic olivine suggests that it may have come from a very early, undifferentiated magma source with a Mg/Fe ratio higher than we might expect given the high Fe content estimated for the Martian mantle [Wänke et al., 1994; Lodders and Fegley, 1997]. However, if majoritic garnet was a significant early crystallizing phase in the Martian magma ocean, the residual liquid would have been enhanced in Mg [Borg and Draper, 2003]. The presence of forsteritic olivine in the rims of the two largest impact basins on Mars as well as around the Nili Fossae, an area at the intersection of the Utopia and Isidis basins, suggests a relationship between that composition and the formation of the impact basins; the impacts that formed the Argyre, Hellas, and Isidis basins may have excavated deeply enough to expose mantle materials. However, the abundance of the Fo91 spectral end-member in these occurrences (∼10%) is lower than expected for rocks purely of mantle origin, and the olivine-bearing materials there may be partially eroded or mixed with crustal materials as impact breccia.
 The areas in which Fo91 spectral end-member is identified are ancient and geologically complex. For example the olivine-rich materials near the Nili Fossae have been suggested to be the result of igneous intrusives [Hoefen et al., 2003], subaerial flows [Hamilton and Christensen, 2005], and Isidis-related impact melt [Mustard et al., 2006]. The diversity of olivine compositions identified near the Nili Fossae suggests that perhaps all these processes have occurred in the region. Forsterite is up to ten times more resistant to low-temperature weathering by water than fayalite [Wogelius and Walther, 1992; Stopar et al., 2006], which may have played a part in its long-term preservation in the proximity of more Fe-rich, and potentially younger, olivines. Detailed studies of these locations may reveal clues to the nature and composition of the mantle at the time of basin formation.
 The compositions of the Fo68 and Fo53 spectral end-members closely resemble the composition of olivine in most Martian meteorites [e.g., McSween and Treiman, 1998] as well as observations from MER [McSween et al., 2006], and they are the most widely distributed compositional range on Mars. If the materials that bear the Fo68 and Fo53-like spectral signatures are petrogenetically related to the Martian meteorites as their olivine composition suggests, it points to a single process that created most of the olivine-bearing materials on Mars. Borg and Draper  have shown that the chemistry of the Martian shergottites can be modeled by melting a depleted component of the Martian mantle and adding trapped late stage liquid interacting with the cumulate stack, and perhaps this process formed the parent magmas of all the materials bearing the spectral signatures of Fo68 and Fo53. The Fo68, Fo53, and Fo39 spectral end-members are found in broad occurrences throughout the highlands (Figure 8), and discrete occurrences are found in channels, chasmata, intercrater plains, and smooth crater floors. Although there are exceptions, Fo68 is the most common spectral end-member found in the chasmata, channels, and intercrater plains, whereas Fo53 is more common in the smooth crater floors. Discrete occurrences of the Fo39 spectral end-member are found exclusively within smooth crater floors. This may suggest that the Fo68-like olivine is a more dominant constituent of intrusive or buried extrusive materials being exposed in the chasmata and channels, and of crustal surface materials as a whole. Slightly more Fe-rich olivine (represented by the Fo53 and Fo39 spectral end-members) is more common in lava or sediment filled craters.
 The relatively Fe-rich Fo39 spectral end-member is less common in our deconvolution results than Fo68 or Fo53, but it still has significant fractional contributions to TES data in spatially coherent occurrences (Figure 8). The presence of olivine this Fe-rich is enigmatic compared to the more Mg-rich compositions typical of olivines found in the lherzolitic shergottites, Fo76–60, [McSween and Treiman, 1998] or compositions typical of olivines in ultrabasic terrestrial rocks, Fo91–86 [Deer et al., 1992] such as those commonly erupted from Mauna Loa [e.g., Garcia et al., 1995]. However, Fe enrichment of olivine is not entirely unexpected as the Martian mantle is thought to contain more FeO than that of Earth by at least a factor of two based on measurements of the density and moment of inertia of Mars [Wood et al., 1981; Bertka and Fei, 1998a, 1998b] and estimates of its bulk composition [Wänke and Dreibus, 1988; Longhi et al., 1992; Lodders and Fegley, 1997; Taylor et al., 2006]. Additionally, the MER Spirit measured some average olivine compositions as Fe-rich as Fo40 [McSween et al., 2006]. The disconnect between the compositions of olivines measured in Martian meteorites and those measured here by remote sensing echoes the conclusions of Hamilton et al. , that the Martian meteorites likely are not representative of the surface as a whole.
 We did not confidently identify the Fo18 and Fo1 spectral end-members on the Martian surface at the spatial scale of the TES observations above our detection limits. In terrestrial settings, Fe-rich olivines are constituents of alkaline and acidic plutonic rocks but only occur in small amounts in volcanic rocks [Deer et al., 1992], and they do not form at all if there is water present in the magma. The formation of magnetite in a magma, which is controlled by the oxygen fugacity in the melt, can also remove Fe from the liquid and prevents the formation of high-Fe olivines. Fayalite is found in banded iron formations, but occurs as the result of high-grade metamorphism [Klein, 1983, 2005], a process not yet identified on Mars. In Martian rocks, olivine in some Martian Nakhlites and has compositions as Fe-rich as Fo30 to Fo17 composing up to 15% of the rock [Friedman Lentz et al., 1999]; however, the Fe-rich (Fo17) compositions are only found in the rims of more magnesian olivines [McSween and Treiman, 1998]. A more detailed search through those spectra identified as containing the Fo18 or Fo1 spectral shapes may reveal confident identifications of these compositions, but if these Fe-rich olivines are present on Mars, they are relatively rare on its surface.
4.4. Comparison to a Model of the Martian Mantle
 The olivine compositions we have observed on Mars can be compared to the compositions predicted by modeling of the mantle's formation. Recent models by Elkins-Tanton et al.  propose a scenario in which Mg-rich olivine forms in a magma ocean and sinks to the core-mantle boundary, thereby enriching the residual magma in Fe. Subsequently forming (and sinking) olivine crystals are increasingly Fe-rich as well as increasingly dense. The resulting density profile is gravitationally unstable and will reorganize to achieve a gravitationally stable interior placing Mg-rich olivine near the surface and Fe-rich olivine at depth [Elkins-Tanton et al., 2005]. In this scenario, the Mg-rich olivine we observe in the rims of Argyre and Hellas Basins may represent olivine-bearing materials excavated from the top of the mantle after overturn was completed. However, the ubiquity of relatively Fe-rich olivine in the current Martian crust requires a more complex scenario. According to the postoverturn model, olivines as Fe-rich as the Fo53 and Fo39 spectral end-members must have originated from very deep within the Martian mantle, but their widespread distribution on the surface would seem to require significant volcanism tapping the same layer on a global scale. Alternatively, perhaps overturn was not complete or the entire mantle and crust did not participate in overturn; however, these scenarios are argued against using dynamical models [Elkins-Tanton et al., 2005].
 We suggest that overturn of the cumulate pile took place and all layers participated, but the resulting solid layers remained overlain by a leftover magma, which was relatively Fe-rich. After reorganization, the mantle from bottom to top would have consisted of a Fe-rich crystalline olivine layer, an Mg-rich crystalline olivine layer, a relatively Fe-rich liquid, and a secondary crust solidified from the Fe-rich liquid. In this scenario, the upper layer of the adolescent mantle, whether it overturned again or not, may have enriched the surface olivine in Fe to the degree we observe with TES. Observed occurrences of modeled basalt in the southern highlands that include the most Fe-rich olivines may then record the last vestiges of this Fe-enriched magma layer as it froze. Even if this scenario was not exactly reproduced, it is clear that magma ocean crystallization and overturn may have produced a complex, heterogeneous mantle with the variety of Mg/Fe contents we observe in the olivine of Mars.
 Our deconvolutions of TES data used an expanded suite of intermediate composition olivine spectra and support previous work, which found that materials containing 10–20% olivine are relatively common throughout most of the southern highland surfaces of Mars. We identified a wide range of spectral end-members, Fo91 to Fo39, on Mars that exist at the spatial scales of TES observations and larger.
 The total compositional range includes occurrences of olivine that are both more Mg- and Fe-rich than previously inferred from Martian meteorites and remote sensing. Occurrences of Mg-rich olivine occur almost exclusively in the ancient Noachian rims of large impact basins (Argyre and Hellas) and may represent primitive or rehomogenized mantle material; detailed studies of these locations could provide insights into the composition of the Martian mantle. Fo68 and Fo53 are the most commonly occurring spectral end-members, consistent with observations from MER and of Martian meteorites. These compositions are found in a variety of terrains, but are most common in outflow channel floors and smooth-floored craters. We identified olivine as Fe-rich as Fo39 on Mars in both regional enhancements and discrete occurrences. The occurrence of this Fe-rich olivine spectral end-member at all suggests that some olivine-bearing materials on the surface of Mars may be more Fe-rich than indicated by the Martian meteorites. It also may suggest the existence of an early, Fe-rich magma layer that overlaid the overturned cumulate stack in an early Martian magma ocean and may have become a significant component of the Martian crust.
 Olivine is most common in the southern highlands near the topographic dichotomy boundary, an elevation that contains many outflow channels exposing subsurface materials. Xanthe Terra, an area on Mars riddled with outflow channels shows a definitive increase in the pervasiveness of the Fo53 spectral shape. These may represent semiglobal, relatively olivine-rich layers, or be a result of the actions of outflows and strong surface winds that erode and abrade the areas.
 Olivine is uncommon in the northern lowlands implying either that it was never present in the younger terrains of Mars, or it has been altered or covered by dust. Olivine also decreases in prevalence in the poleward of 45°S suggesting that latitudinal climate variations (e.g., the presence of steady state ground ice) may have played a role in weathering olivine out of the southern highland materials in those regions.
 We correlated the olivine spectral end-members found in this study with their mapped geologic ages and found that olivine appears to have been brought to the surface throughout Martian history (early Noachian to Amazonian). If early Noachian olivine was coeval with the deposition of phyllosilicates and sulfates identified by OMEGA, it implies that the early climate of Mars was either (1) inhomogeneous, with both wet and dry regions existing on the surface simultaneously or (2) the wet conditions required to produce the phyllosilicates and sulfates were not wet enough or long-lived enough to completely alter the olivine present on the surface of Mars today.
 The authors wish to thank S. A. Morse for providing the olivine-rich mafic concentrates from which the pure olivine samples were extracted. J. Taylor, H. McSween, and M. Garcia added useful discussions and petrological insights to our results and interpretations. J. Bandfield and T. Hoefen produced thoughtful reviews that helped clarify and improve the manuscript. This research was supported by NASA grant NAG5-13421, and it is HIGP publication 1493 and SOEST publication 7165.