The rocks of Gusev Crater as viewed by the Mini-TES instrument

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 111,, doi: /2006je002747, 2006 Correction published 10 February 2007 The rocks of Gusev Crater as viewed by the Mini-TES instrument S. W. Ruff, 1 P. R. Christensen, 1 D. L. Blaney, 2 W. H. Farrand, 3 J. R. Johnson, 4 J. R. Michalski, 1 J. E. Moersch, 5 S. P. Wright, 1 and S. W. Squyres 6 Received 8 May 2006; revised 16 August 2006; accepted 15 September 2006; published 30 December [ 1 ] The Miniature Thermal Emission Spectrometer (Mini-TES) on board the Mars Exploration Rover Spirit is part of a payload designed to investigate whether a lake once existed in Gusev Crater. Mini-TES has observed hundreds of rocks along the rover s traverse into the Columbia Hills, yielding information on their distribution, bulk mineralogy, and the potential role of water at the site. Although dust in various forms produces contributions to the spectra, we have established techniques for dealing with it. All of the rocks encountered on the plains traverse from the lander to the base of the Columbia Hills share common spectral features consistent with an olivine-rich basaltic rock known as Adirondack Class. Beginning at the base of the West Spur of the Columbia Hills and across its length, the rocks are spectrally distinct from the plains but can be grouped into a common type called Clovis Class. These rocks, some of which appear as in-place outcrop, are dominated by a component whose spectral character is consistent with unaltered basaltic glass despite evidence from other rover instruments for significant alteration. The northwest flank of Husband Hill is covered in float rocks known as Wishstone Class with spectral features that can be attributed uniquely to plagioclase feldspar, a phase that represents more than half of the bulk mineralogy. Rare exceptions are three classes of basaltic exotics found scattered across Husband Hill that may represent impact ejecta and/or float derived from local intrusions within the hills. The rare outcrops observed on Husband Hill display distinctive spectral characteristics. The outcrop called Peace shows a feature attributable to molecular bound water, and the outcrop that hosts the rock called Watchtower displays a dominant basaltic glass component. Despite evidence from the rover s payload for significant alteration of some of the rocks, no unambiguous detection of crystalline phyllosilicates or other secondary silicates has been observed by Mini-TES. The mineralogical results supplied by Mini-TES provide no clear evidence that a lake once existed in Gusev Crater. Citation: Ruff, S. W., P. R. Christensen, D. L. Blaney, W. H. Farrand, J. R. Johnson, J. R. Michalski, J. E. Moersch, S. P. Wright, and S. W. Squyres (2006), The rocks of Gusev Crater as viewed by the Mini-TES instrument, J. Geophys. Res., 111,, doi: /2006je Introduction [ 2 ] Miniature Thermal Emission Spectrometer (Mini- TES) data from both Mars Exploration Rovers (MER) have contributed to a range of scientific and tactical objectives during the MER mission, including atmospheric [Smith et al., 2004, 2006] and thermophysical science [Fergason et al., 2006], mineralogical and geochemical investigations 1 School of Earth and Space Exploration, Arizona State University, Tempe, Arizona, USA. 2 Jet Propulsion Laboratory, Pasadena, California, USA. 3 Space Science Institute, Boulder, Colorado, USA. 4 U.S. Geological Survey, Flagstaff, Arizona, USA. 5 Department of Earth and Planetary Sciences, University of Tennessee, Knoxville, Tennessee, USA. 6 Department of Astronomy, Cornell University, Ithaca, New York, USA. Copyright 2006 by the American Geophysical Union /06/2006JE [Christensen et al., 2004a, 2004b; McLennan et al., 2005; Yenetal., 2005; Squyres et al., 2006; Glotch and Bandfield, 2006; Glotch et al., 2006; Clark et al., 2007], and the determination of erosional rates in Gusev crater [ Grant et al., 2006]. Here we present the thermal infrared (TIR; cm 1 )spectral observations made of the abundant rocks along the rover s traverse, limiting the scope of the report to seven classes of rocks that have been described previously: Adirondack, Clovis, Wishstone, Peace, Watchtower, Backstay, and Irvine [e.g., Squyres et al., 2006; McSween et al., 2006]. Mini-TES spectra provide the unique ability to estimate the bulk mineralogy of rocks from adistance of afew to 10s of meters away from the rover. This allows for the analysis of significantly more rocks than can be accessed by the contact instruments on the rover s mechanical arm, and also plays a major role in the selection of rocks for subsequent in-situ analysis. The diversity of rock types encountered by the Spirit rover in its traverse from the plains surrounding the lander into the 1of36

2 Figure 1. Mars Orbiter Camera image (R ) showing locations described in this work. Numbered locations are as follows: 1, Clovis outcrop; 2, Wishstone float rock; 3, Peace/Alligator outcrop; 4, Larry s Lookout; 5, Jibsheet Ridge; 6, Irvine float rock. Columbia Hills (Figure 1) has expanded significantly since the initial report of results early in the mission [ Christensen et al., 2004a]. Most of the diversity occurs in the varied terrain of the Columbia Hills. The mineralogy of the rocks holds clues to the geological history of Gusev Crater and has implications for interpreting thermal infrared observations from many other parts of Mars. [3 ] The full Athena payload [Squyres et al., 2003] has been used in the investigation of various rocks in Gusev Crater. At least one example of each of the rock classes described in this paper has been investigated with the full suite of instruments. Herein we do not attempt to incorporate detailed results from the other instruments. Instead, we present results for amuch larger set of rocks than could be measured with the contact instruments and provide mineralogical details that are unique to the Mini-TES data set. [ 4 ] The rock classes that have been described for Spirit s Gusev crater landing site and that are used in this paper are formally defined using elemental chemistry as determined by the Alpha Particle X-ray Spectrometer (APXS) measurements of brushed or abraded rock surfaces [ Squyres et al., 2006]. As shown below, however, all of the rock classes defined in this fashion also can be uniquely classified with Mini-TES spectra. The relationship between elemental chemistry and TIR spectral character has been useful in two ways during the mission. First, once we have identified a class using the APXS, and determined its TIR spectral properties with Mini-TES, we can then use Mini-TES data to determine the distribution of this class at significant distances from the rover, covering much more ground than could be done solely with in-situ measurements. Second, there have been instances where Mini-TES was the first instrument to identify arock as anew class, based on TIR spectral properties not observed previously, and then the APXS was used to determine the elemental composition that provides the formal definition for that new class. [ 5 ] Aninformal nomenclature evolved during the mission that identifies the various geographic regions encountered by the Spirit rover [Arvidson et al., 2006]. The plains include all locations from the lander to the base of the West Spur. The West Spur is the westernmost projection of Husband Hill (Figure 1), one of the tallest in the roughly north-south chain of the Columbia Hills. These names along with individual rock names can be used to relate the results presented in this work to those from the other instruments. [ 6 ] In this paper we present results for seven different rock types in roughly the chronologic order in which they were encountered, using the same nomenclature as defined by Squyres et al. [2006]. But before the Mini-TES spectra of these rocks are presented and interpreted, a brief description of the instrument and its calibration is given in section 2 along with an introduction to the issue of dust accumulation on optical components of the instrument. Spectral contributions from the atmosphere can have a significant impact on surface spectra and the approach used to derive the mineralogy of the rocks. In section 3wepresent an extensive examination of this issue. In section 4 we explain our methodology for investigating the distribution and mineralogy of rocks along the rover s traverse. Section 5and its many subsections contain the results for the various rock classes. Mini-TES observations of the Adirondack Class rocks of the Gusev plains have been reported previously [Christensen et al., 2004a], but we have improved upon the initial results as discussed in section 5.1. Section 5.2 presents the observations and spectral deconvolution results for the rocks of the West Spur known as Clovis Class. Most of Husband Hill is covered byfloat rocks known as Wishstone Class that are the subject of section 5.3. The outcropping rocks on Husband Hill are relatively rare. We present two examples known as Peace Class and Watchtower Class in sections 5.4 and 5.5 respectively. Scattered across Husband Hill are rare examples of basaltic exotics (out of place rocks) that are presented in section 5.6. Adiscussion of the mineralogical results and the clues they provide for the geologic history of Gusev Crater is presented in section 6 followed by our conclusions in section Mini-TES Instrument and Calibration [ 7 ] Details of the Mini-TES design and operation are given elsewhere [Christensen et al., 2003], but here a brief overview is appropriate. The Mini-TES is a Fourier transform infrared spectrometer operating over the spectral range from 2000 to 340 cm 1 ( 5to29 m m) with aspectral sampling of 10 cm 1.It uses an uncooled, deuterated triglycene sulfate detector with KBr internal transmissive optics. The spectrometer is mounted inside the warm electronics box (WEB) in the body of the rover to provide a more stable thermal environment and uses the periscopelike Pancam Mast Assembly (PMA) to deliver light inward [Squyres et al., 2003]. A rotating pointing mirror in the PMA head allows for arange of elevation angles ( 50 to +30 )tobeobserved and rotation of the entirehead allows a full 360 range of azimuth angles. The pointing mirror directs light to an adjacent fixed fold mirror, then down the PMA tube to a primary and secondary mirror system that acts as a Cassegrain telescope with a nominal 20 mrad instantaneous field of view. A CdTe window at the base of the PMA tube separates these four mirrors from the spectrometer inside the WEB. The 20 mrad mode yields afield 2of36

3 of view (FOV) of 10 cm in diameter for targets closest to the rover.this spot size is larger than the 4.5 cm diameter holes abraded into rocks by the rock abrasion tool (RAT) [Gorevan et al., 2003]. Consequently, no freshly abraded rock surfaces can be isolated in Mini-TES observations. However, the brushing capability of the RAT has been used to good effect to remove dust from alarge enough area to fill the Mini-TES FOV as described in subsequent sections. [ 8 ] A calibration scheme for Mini-TES spectra has been developed that relies on the single blackbody calibration target mounted within the head of the PMA. A second calibration target mounted on the rover deck is unavailable for standard calibration because of the failure of its two platinum resistance thermometers within the first few sols of the mission. The basic equation for converting raw voltage spectra to radiance is R sample ¼ V measured = F þ R inst ; where each term is afunction of wave number,r sample is the radiance of the sample (W cm 2 sr 1 /cm 1 ), which can be either asurface or sky observation, V measured is the Fouriertransformed voltage signal, Fisthe instrument response function (V/W cm 2 sr 1 /cm 1 ), and R inst is the radiance of the instrument (W cm 2 sr 1 /cm 1 ). With only one blackbody calibration target available, it is not possible to directly determine F. However, because the instrument response function was so well characterized during environmental testing in a thermal vacuum chamber on earth [ Christensen et al., 2003], aset of instrument response functions is available for use in the calibration of Martian spectra. To provide more accurate results, Fisinterpolated on the basis of the measured detector temperature acquired during each observation. An observation of the calibration target inside the PMA head allows the instrument energy term (see equation (1)) to be determined and removed from each spectrum according to the equation R inst ¼ R planck V cal target = F : Because the calibration target is viewed at least twice during every observation (beginning and end), any change in the instrument energy during the course of an observation is accounted for by interpolation. The final conversion of sample radiance spectra to unitless, temperature-independent emissivity spectra needs to account for atmospheric contributions as is described in section 3. [ 9 ] On sol 420 a wind event occurred during which dust was cleared from the solar panels [Greeley et al., 2006] but was deposited on one or more surfaces in the Mini-TES optical path. The most likely surface to be impacted is the pointing (elevation) mirror at the top of the PMA. Since sol 421, Mini-TES spectra display features due to dust that vary in intensity. It appears that the variability is related to the temperature difference (D T) between the target and the pointing mirror. The dust on the mirror is not thick enough to be opaque but it does supply either absorption or emission features, depending on the D T, that are most evident in the middle wave number range of Mini-TES ð 1 Þ ð 2 Þ data ( cm 1 ). Current efforts to remove these features appear very promising but at the time of writing, we have not yet formalized the methodology. Consequently, spectra from rocks observed after sol 420 are not used for quantitative mineralogical determinations. However, we have developed a semiquantitative correction that appears to reduce significantly the mirror-dust related artifacts allowing for a better classification of all rock spectra measured after sol 420. This correction is presented in section Atmospheric Contributions to Surface Spectra [ 10] In orbital Mars Global Surveyor Thermal Emission Spectrometer (TES) spectra, contributions from the atmosphere are well documented and understood [e.g., Smith et al., 2000; Bandfield et al., 2003]. In Mini-TES spectra of the surface, the much shorter atmospheric path lengths of the observations preclude most of the atmospheric contributions that impact TES spectra. Nevertheless, the atmosphere contributes to Mini-TES spectra with what we here call standard contributions and additional contributions Standard Atmospheric Contributions [ 11] Spectral contributions from the atmosphere are most recognizable in Mini-TES spectra in the region of the atmospheric CO 2 absorption centered at 667 cm 1 (15 m m). From orbit, the long atmospheric path length renders the center of the CO 2 band opaque, i.e., surface radiation is fully absorbed by CO 2 in this spectral region. But the surface measurements made by Mini-TES have atmospheric path lengths that are insufficient to fully absorb surface radiation in the center of the CO 2 band. Mini-TES spectra of surfaces that are warmer than the CO 2 in the path between the instrument and the surface show CO 2 in absorption. For example, the surface dust spectrum in Figure 2displays the central and side lobes of CO 2 as emissivity minima. However, asthe temperature of the measured surface falls below the temperature of the atmosphere in the beam path, the CO 2 features shallow and ultimately invert to become emissivity maxima as shown by the rock spectrum in Figure 2. In such cases, there is agreater contribution of radiance in the CO 2 region from the atmosphere than from the surface target. When the radiance spectrum is converted to emissivity, the central CO 2 feature appears as apeak with emissivity >1 because the brightness temperature used for the emissivity/temperature separation is intentionally derived from portions of the radiance spectrum that exclude the CO 2 region. The behavior of the CO 2 feature becomes a telltale indicator for the relative intensity of the thermal infrared radiation of the surface target versus the atmosphere. This behavior is readily modeled using methods presented by Smith et al. [2004], which supports the empirical observations made with Mini-TES (Figure 2). [ 12] The contribution from dust in the Martian atmosphere also can impact Mini-TES spectra. Unlike CO 2,the column density of dust between the instrument and asurface target is so small that there is no apparent absorption or emission by dust. However, thermal radiation from the sky illuminates the surface and is reflected or possibly scattered into the field of view of the instrument. Known asdown- welling radiance, its impact onmini-tes spectra is dependent 3of36

4 Figure 2. Measured and modeled spectra demonstrating the behavior of the CO 2 feature in Mini-TES spectra under different temperature conditions. The measuredmini-tes spectra are those of surface dust (red) from sol 078, sequence ID p3654 with atemperature of 273K and Humphrey rock (black), sol 056, sequence ID p3175 with atemperature of 249K. As shown by the CO 2 modeling from M. D. Smith, when the surface target is warmer than the atmosphere in the path between Mini-TES and the surface, the CO 2 feature appears in absorption (blue). When temperatures are reversed, the CO 2 feature appears in emission (magenta). on the difference inradiancebetweenthe surface targetand the atmosphere, which is a function of temperature and atmospheric opacity. Mini-TES measurements ofdownwelling radiance are made routinely using upward-looking orientations(up to 30 in elevation) ofthe pointing mirror in the PMA head.the upward-looking view means the atmosphere isseen against the cold background of space yielding features that appear in emission rather than absorption as when viewed from Figure 3. Spectral characteristics of Mini-TES sky observations. Aradiance spectrum from atypical sky observation viewed 30 above the horizon is shown (solid) with aplanck curve (short dashes) that represents the temperature at the radiance maximum of the CO 2 feature. The unitless emissivity spectrum (long dashes) is produced by dividing the Planck curve from the sky radiance spectrum. The magnitude of the features of atmospheric dust is better displayed in this form. 4of36

5 orbit (Figure 3). A broad, inverted U- or V-shaped feature attributable to atmospheric dust spans the range from 800 to 1300 cm 1.Alesser contribution occurs onthe low wave number side of the CO 2 peak. The emissivity spectrum of the atmosphere, produced by dividing the radiance spectrum by aplanck curve at the temperature of the maximum CO 2 radiance, better displays the actual shape and magnitude of the dust features (Fig ure 3). [ 13] Downwelling radiance is a recognized component of TIR emission spectra both in the field (on the ground or from airborne or orbital instruments) [e.g., Hook et al., 1992; Salisbury and D Aria, 1992; Horton et al., 1998] and in the laboratory [e.g., Ruff et al., 1997]. Different methods specific to a given measurement are used to remove the downwelling contribution from the final spectra. For Mini-TES spectra, the routinely collected, upward-looking measurements of atmospheric radiance at 30 above the horizon are used to make what we refer to below as the standard downwelling correction for the downwelling radiance that reflects off the surface target and into the spectrometer [Christensen et al., 2004b]. Although a single observation ofthe sky at asingle viewing angle does not measure the integrated downwelling radiance incident on the surface target, under most atmospheric conditions a30 upward-looking Mini-TES observation approaches the equivalent integrated radiance (M. Smith, personal communication). The conversion of sample radiance (from a surface target) to emissivity is as follows: e cor ¼ R sample R sky = R planck R sky ; ð 3 Þ where R sample is the calibrated radiance spectrum of the surface target from equation (1), R sky is the calibrated radiance of the sky at 30 above the horizon,and R planck is the calculated Planck radiance at the maximum brightness temperature of the surface target. But as is shown in the next section, there are cases where this standard downwelling correction is insufficient to remove fully the spectral contributions of the atmosphere. It should be noted that the above equation is most strictly valid for controlled laboratory conditions [ Christensen and Harrison, 1993; Ruff et al.,1997]. It requires the assumption that Kirchhoff s law relating reflectivity and emissivity ( R =1 e )isvalid, which is strictly true only for Lambertian surfaces in isothermal equilibrium with their surroundings [e.g., Badenas, 1997]. For the surface of Mars, variations due to the heterogeneous nature of the Martian atmosphere and any non-lambertian characteristics of the measured surface targets are not taken into account. These factors could be sources of error in the application of the standard downwelling correction, for which we now have evidence as described below Additional Atmospheric Contributions [ 14] Wehave identified distortions in Mini-TES spectra produced by downwelling atmospheric radiance that cannot be modeled using the standard downwelling correction of equation (3). These distortions have been seen in the spectra of disturbed soil (the dark-toned, fine particulate portion of the regolith) and the ubiquitousthin mantle of light-toned surface dust, as well as rock in spectra. A complete understanding of additional downwelling radiance will require further analysis but is likely due to some combination of non-lambertian surfaces, non-isotropic scattering by atmospheric dust, and non-isothermal conditions in the atmosphere. Because the magnitude and direction of the CO 2 feature in Mini-TES spectra serves as atelltale indicator of the contribution from additional downwelling radiance, it is reasonable to assume that the greater the depth of this feature, the smaller the impact on agiven spectrum. Conversely, spectra that show CO 2 in emission should be assumed to be distorted. We have established methods for dealing with the issue of additional downwelling radiance that are described in section 4. The following subsections detail the observations that were used to recognize this atmospheric component Additional Atmospheric Contributions Evident in Soil and Dust Spectra [ 15] Distortions in the spectra of soil and dust are easier to observe than in rocks because of the low thermal inertia and hence rapid temperature changes of fine-particulate materials. When observations of soil or dust are made during peak surface heating hours ( 11:30 13:30 local true solar time (LTST)), the CO 2 feature appears strongly in absorption (Figures 4a and 4b), an indication that the surface temperature is much greater than the atmosphere (>20 ). Conversely, when the same targets are observed in the mid- to late afternoon after substantial cooling has taken place, they yield spectra that are slightly distorted, mostly in the middle wave number range ( cm 1 )(Figures 4a and 4b). The marked reduction in contrast of the CO 2 feature demonstrates the decreasing temperature difference between the lower atmosphere and the surface. A more dramatic example is found in the case of alight-toned (dusty) soil patch in front of the rock called Bonneville Beacon, encountered en route to Bonneville crater [ Arvidson et al., 2006]. This surface was included in multiple-time-of-day observations that spanned a range of morning and afternoon hours such that the soil was in sunlight around noon and then was fully shadowed by the rock at 15:50 LTST. The temperature dropped from 271K to 243K during this time. The resulting spectra (Figure 4c) show CO 2 strongly in absorption for the warmest temperature and strongly in emission for the coolest temperature. The changes in the middle wave number range ( cm 1 ) are especially dramatic. The distinct Christiansen feature at 1250 cm 1 and the weakly concave-up shape of the warmest spectrum give way to astrongly convex-up shape and ashift of the apparent Christiansenfeature in the coolest spectrum. The new emissivity maximum is shifted to lower wave numbers, an indication of the addition of radiance in this location, presumably from the atmosphere. [ 16] In all of the above cases the standard downwelling correction does not restore the distorted spectra to their less distorted shape. The distorted mid-afternoon spectra of the disturbed soil and adjacent dust described above are little changed following the standard downwelling correction (Figure 5a). Similarly, the extreme example of the Bonneville Beacon dusty soil convincingly demonstrates that the spectrum of the coolest surface cannot be made to resemble that of the same surface in the warmest state using the standard downwelling correction (Figure 5b). In this example, we substituted the standard 30 elevation sky 5of36

6 target transformation of Mini-TES spectra from the Opportunity rover in Meridiani Planum Additional Atmospheric Contributions Evident in Rock Spectra [ 17] Although additional downwelling radiance can affect the spectrum of any rock, we have chosen to demonstrate its impact on the dark-toned plains rocks known as Adirondack Class. Figure 7 shows the prominent local emissivity peak at 430 cm 1 in the spectrum of these rocks that is good evidence for an olivine component [Christensen et al., 2004a]. Spectra from the rover s Mössbauer (MB) instrument clearly support the olivine-rich characteristic of Adirondack Class rocks [ Morris et al., 2004]. The 430 cm 1 feature in Mini-TES spectra is similar to afeature attributedto olivine in TES spectra of olivine-rich surfaces elsewhere on the planet [Hoefen et al., 2003; Hamilton and Christensen, 2005] (blue spectrum in Figure 7) although it is shifted due to adifferent Fe/Mg ratio [e.g., Hoefen et al.,2003].but the Figure 4. The spectra of soil surfaces showing the temperature-dependent effects of downwelling atmospheric radiance. The spectra (sol 048, p3628) of (a) dark-toned soil in the rover wheel track and (b) light-toned dusty soil outside of the track observed at two different times of day become distorted as the surface cools. (c) This effect is displayed more dramatically for another dusty soil target in front of the rock Bonneville Beacon observed at multiple times of day (sol 047, p3154). radiance used in equation (3) with sky observations from closer to the horizon (10 and 20 above the horizon) where the higher temperatures and opacity of the atmosphere provide even greater radiance, but which ultimately have little effect on the final correction (Figure 5b). However, a ratio of the coolest spectrum to the warmest spectrum provides clues to the cause of the distortion. It reveals a spectrum that looks remarkably similar to the emissivity of the atmosphere when scaled and offset from its original magnitude (Figure 6). From this analysis comes an indication of an additional downwelling radiance component that is not accounted for by the standard downwelling correction. Glotch and Bandfield [2006] demonstrated a similar result using the very different approach of factor analysis/ Figure 5. The standard correction for downwelling radiance does little to remove the distortion observed in the spectra from Figure 4. (a) The cooler soil spectra of Figures 4a and 4b are shown with and without downwelling correction. (b) The major spectral distortion of the coldest Bonneville Beacon soil in Figure 4c is little changed regardless of whether the increasing atmospheric radiance from 30,20,or10 above the horizon is used in the downwelling correction. 6of36

7 Figure 6. Aratio (solid) of the coldest to warmest spectra in Figure 4c reveals ashape similar to the emissivity of the sky (dashes). This suggests that the spectral distortions shown in Figure 4are the result of an additional downwelling radiance component that is not accounted for using the standard downwelling correction. Figure 7. Mini-TES spectra of two Adirondack Class rocks compared with ates spectrum of Nili Fossae, an olivine-rich region on Mars (blue), and alaboratory spectrum of olivine (green). The feature at 430 cm 1 is related to olivine. The Mini-TES spectra compare favorably to the TES spectrum only in the low wave numberregion (<600) because of the impact of atmospheric downwelling radiance at higher wave numbers. 7of36

8 Figure 8. Mini-TES spectra of two Adirondack Class rocks, Sarah and Humphrey, show adifferent degree of distortion due to downwelling radiancethat is reflected in the orientation of the CO 2 feature. (a) In the middle wave number range, Sarah looks more like the TES spectrum of the olivine-rich Nili Fossae region and aspectrum derived from CIPW norms of Humphrey than does the spectrum of Humphrey itself. This is due to a smaller fraction of downwelling radiance in the Sarah spectrum. (b) The distortion from cm 1 in the spectrum of Humphrey cannot be eliminated regardless of whether the increasing atmospheric radiance from 30,20,or10 above the horizon is used in the downwelling correction. This is evidence for an additional downwelling component that is not accounted for in the standard downwelling correction. similarity between the Mini-TES and TES spectra diminishes in the middle wave number range ( ) where abroad emissivity minimum between cm 1 and aconvex-up shape from cm 1 departs from the TES spectrum. Initial efforts to model Adirondack Class spectra by Christensen et al. [2004a] using linear deconvolution demonstrated that surface dust supplies a feature in this region and along with olivine, oxides, and minor clinopyroxene yields a satisfactory fit tothe broad minimum. But the spectra were not well modeled in the high wave number range (>1250 cm 1 )where dust shows astrong absorption and dark-toned rocks do not. Also, the paradox of a large dust component (>50%) in the modeled results of dark-toned, minimally dusty rocks was recognized by Christensen et al [2004a]. We now understand that these rock spectra include contributions from additional downwelling radiance, as shown below, that produce spurious deconvolution results. [ 18] Because of the relatively high thermal inertia of the Adirondack Class rocks [Fergason et al., 2006], they commonly had temperatures <245 K even during the period of peak diurnal heating. By comparison, atmospheric temperatures 100 m above the surface during the seasons in which most of these rocks were encountered (L s = ) routinely reached 240 K to 250 K[Smith et al., 2004]. Spectra of cold rocks typically show CO 2 in emission, an indication that the atmosphere may supply a significant fraction of the radiance. But there are examples where CO 2 is seen as a weak absorption feature, indicating a smaller contribution from the atmosphere as in the spectrum of the Adirondack Class rock called Sarah (Figure 8a). In these cases, there are spectral changes in the region between 1000 to 1200 cm 1 where aconcave-up shape replaces the more common convex-up shape like that seen in the Adirondack Class rock Humphrey. The result is a spectrum that more closely resembles the TES spectrum of olivinerich material seen elsewhere on the planet (Figure 8a). Although the spectra of Adirondack Class rocks may not necessarily match the TES spectrum in all details, the increased similarity in certain cases suggests that when these rocks have a larger temperature difference (D T) with the atmosphere, their spectra are less distorted. [ 19] Additional evidence that Adirondack Class spectra should look more like the TES spectrum comes from a synthetic spectrum we produced using the CIPW normative mineralogy determined by McSween et al. [2004] from APXS measurements of the RAT-abraded surface of Humphrey (no APXS measurements were acquired of Sarah). By linearly combining laboratory mineral spectra in the proportions and compositions given by the normative mineralogy, it is possible to create a synthetic version of the Humphrey spectrum. The result shows important general similarities to the spectrum of the rock Sarah including a concave-up, very asymmetric shape in the middle wave number range ( ) and a strong peak centered at 430 cm 1 (Figure 8a). The convex-up shape seen in the spectrum of Humphrey and many other examples of Adirondack Class rocks is a distortion from the true shape consistent with the effects of additional downwelling radiance described in section 3.2.1, whereas the spectrum of Sarah is less affected. The standard downwelling correction does not remove the distortion. For example, if the spectrum of the rock Sarah represents a less distorted version of the Humphrey spectrum, then the standard downwelling correction should make the Humphrey spectrum look more similar to the Sarah spectrum. This is not the case regardless of which sky observation (30,20,or 10 elevation) is used in the correction (Figure 8b). It appears that deconvolution of Mini-TES spectra encumbered by additional downwelling radiance requires an alternative strategy to yield correct results, which we describe in the next section. 4. Methodology [ 20] Wehave used Mini-TES spectra of rocks to support two general objectives: (1) to examine the diversity and 8of36

9 distribution of rocks along the rover s traverse and (2) to characterize the mineralogy of the rocks. The first objective can be met in most cases without concern for spectral contributions from atmospheric or mirror dust. The second objective requires methods that account for the spectral features due to dust and temperature errors Spectral Classification of Rocks [ 21] Mini-TES spectra are impacted to varying degreesby contributions from dust in the atmosphere, on rock surfaces, and after sol 420, on the instrument s pointing mirror. The low wave number range of Mini-TES spectra (< 600 cm 1 ) is the least impacted by dust, making it possible in most cases to compare and contrast rock spectra to one another without any corrections using the features of this spectral range. As shown in Figure 3, atmospheric dust has weaker features in the low wave number range than in the middle range. Consequently, the spectrum of a surface target is much less affected by atmospheric contributions in this range under most temperature and opacity conditions. Surface dust also has weak features below 600 cm 1 compared to rocks (Figure 2). As shown in section 4.3, dust on the mirror has only minor effects in the low wave number range in most cases. On the basis of these empirical observations, we group into a single class any rocks with common spectral features in the low wave number range. The strength of these features and their distinctive characteristics are the result of the fundamental bendingmodes of the silicate phases that dominate the rocks. As shown in subsequent sections, classifying rocks by their low wave number (< 600) spectral features is, in most cases, a robust strategy even in the absence of any dust corrections Determination of the Mineralogy of Rocks [ 22] The mineralogy of rocks measured by Mini-TES is determined through deconvolution of their emissivity spectra. Using alinear least squares algorithm combined with asuite of spectral end-members, the individual phases and their abundance in the rock (the areal fraction) can be determined [e.g., Ramsey and Christensen, 1998; Feely and Christensen, 1999; Hamilton and Christensen, 2000]. Table 1shows the various end-member sets used in this work that are further described in subsequent sections. Multiple endmember sets and multiple deconvolution runs were used because of (1) the numerical limit on end-members imposed by the matrix algebra employedinthe linear deconvolution and (2) our goal of characterizing the uniqueness of various end-member solutions. The spectra were fit over the range cm 1 (Mini-TES channels 5:102) which excludes the ends of the spectra where the signal-to-noise is the lowest. Truncation of the spectra above 1350 cm 1 retains enough of the high wave number range to allow for the investigation of potential roll-off in emissivity due to surface dust (section 4.2.2). The CO 2 range from cm 1 (channels 24 45) also was excluded as described in the next section. The remaining numberofspectral data points is 76, which limits the number of possible endmembers to less than or equal to this value. In practice, we limited the number of end-members to 60 in most cases to avoid spurious results. [ 23] Mini-TES spectra include varying contributions from the sky, surface dust, dust on the mirror, and as is described below, spectral slope due to temperature error. We present a semiquantitative correction for mirror dust in section 4.3, but until it can be fully validated, we exclude spectra acquired after the sol-420 dust contamination event from any mineralogical determinations. The methods used to deal with the other three contributions are described below. The mineralogical results obtained by the deconvolution technique are affected by the accuracy for which these additional contributions can be accounted. A full assessment of this accuracy is beyond the scope of the current work. Instead, for each of the three contributions, we present a rationale for the assumption that they do not compromise our ability to obtain results with comparable accuracy and precision as achieved during Mini-TES testing [ Christensen et al., 2003]. The raw results ofeach deconvolution run shown in accompanying figures are presented with 1% precision and with ungrouped phases to emphasize the subtle variations from one run to the next. These are not the final results. Table 2 contains the final results, for which we (1) take into account the results from best-fit cases with comparable root-mean-squared (RMS) error values; (2) group together spectrally and mineralogically similar phases; and (3) average the abundance values, rounding them to the nearest 5%. This approach is similar to that done for previous Mini-TES results [ Christensen et al., 2004a, 2004b; Glotch and Bandfield, 2006; Glotch et al., 2006]. The detection limits typically quoted for TES results (10 15%) [e.g., Bandfield et al., 2000; Christensen et al., 2000b] likely are higher than can be achieved with Mini-TES spectra because of the much diminished effect of the atmosphere on Mini-TES spectra. Until a more detailed assessment is undertaken, we assume that the Mini-TES detection limits are comparable to those suggested by Christensen et al. [2003] based upon laboratory spectra, or about 5 10% depending on mineral phase Sky Component [ 24] Insection 3wedemonstrated that the downwelling radiance from atmospheric dust could significantly impact the spectra of surface targets under certain conditions. Because the standard downwelling correction (equation (3)) does not fully account for this sky component, we have implemented the following strategy to address it. If we assume that the sky component is linearly additive, as is done in the deconvolution of TES spectra [ Bandfield et al., 2000; Smith et al., 2000], then it can betreated asan additional end-member spectrum during deconvolution. Because spectral deconvolution is done using emissivity spectra, this necessitates the use of an emissivity spectrum of the sky. Although such a spectrum is not strictly valid given the heterogeneity of atmospheric temperatures, it is still possible to assign a single temperature to a given sky observation based on the brightness temperature at the center of the 15 m mco 2 feature. Dividing the sky radiance spectrum by a Planck curve at the maximum brightness temperature yields an emissivity spectrum (Figure 3) that can be used in deconvolution. Given the variability in atmospheric dust opacity over the course of the mission and the variability of downwelling radiance on a diurnal time scale, it is most appropriate to use a spectrum of the sky that is closest to the time-of-day when agiven surface spectrum was measured and within afew sols. Here we use Mini-TES observations of the sky measured at 30 above 9of36

10 Table 1. End-Member Phases Used in Various Combinations for the Deconvolution of Mini-TES Spectra of Gusev Crater Rocks a Category Name Category Name Primary igneous minerals Albite (Cleavelandite) Glassy phases K-rich glass b WAR Albite WAR Silica glass b Oligoclase WAR Quenched Basalt b Oligoclase BUR Obsidian 8 b Andesine BUR Mars glass c Andesine WAR Basaltic glass HWMK124D, Rind, Spot B c Labradorite WAR Basaltic glass HWKV340A, Matte Uneven Surface c Labradorite BUR-3080A 176 Basaltic glass HWKV340A, Glassy Black Flat Surface c Labradorite WAR-RGAND Maskelynite (chunk) ASU-7591 Bytownite WAR Shocked An 22.6 GPa d Anorthite BUR Shocked An 37.5 GPa d Anorthite WAR Augite HS-119.4B 56 Sulfates Gypsum var. Alabaster ML-S11 Augite BUR Anhydrite ML-S9 Augite NMNH Celestite ML-S13 Augite NMNH Kieserite KIEDE1 <1mm e Augite DSM-AUG Glauberite GBYAZ1-R1 e Augite NMHN Epsomite 2 e Diopside WAR Fe-oxides Black Hematite Coating (Swansea AZ) f Diopside NMNH Synthetic Packed Magnetite Powder MTS5 f Diopside HS-15.4B 16 Synthetic Packed Goethite Powder GTS2 f Diopside HS-317.4B 32 Diopside DSM-DIO Secondary silicates Serpentine HS-8.4B 14 (alteration phases) Diopside WAR Serpentine BUR Hedenbergite, manganoan DSM-HED Kaolinite KGa-1b granular 185 Hedenbergite, manganoan NMNH-R Halloysite WAR-5102 solid 189 Hedenbergite NMNH Saponite ASU-SAP01 granular 194 Pigeonite b Ca-montmorillonite STx-1 solid 197 Enstatite NMNH Na-montmorillonite SWy-2 granular 200 Enstatite NMNH-R Nontronite WAR-5108 granular 203 Enstatite WAR Fe-smectite SWa-1 solid 207 Bronzite BUR Illite IMt-2 granular 211 Bronzite NMNH Crystalline heulandite g Bronzite NMNH Crystalline stilbite g Bronzite NMNH-C Beidellite Sbdl-1 h Hypersthene NMNH-B Nontronite Nau-1 h Pyroxmangite HS-325.4B 33 Nontronite Nau-2 h Wollastonite BUR Hectorite Shca-1 h Forsterite BUR-3720A 8 Montmorillonite Swy-1 h Forsterite AZ Phosphates Wavellite ML-P7 73 Olivine KI 3115 Fo68 b Meta-variscite ML-P4 95 Olivine KI 3362 Fo60 b Pyromorphite ML-P3 77 Olivine KI 3373 Fo35 b Apatite ML-P1 86 Olivine KI 3008 Fo10 b Fayalite WAR-RGFAY a Unless otherwise noted, spectra are from the ASU TES library [ Christensen et al., 2000a]. b Hamilton et al. [2001]. c R. V. Morris. d A. Baldridge. e Johnson et al. [2002b]. f Glotch et al. [ Glotch et al., 2004]. g S. W. Ruff. h Michalski et al. [2006]. the horizon because they are the ones most commonly acquired. The region around the core of the CO 2 band centered at 667 cm 1 must be excluded during deconvolution as is done with TES spectra because it cannot be modeled as a simple linear combination of surface and sky components. In the current work, this region extends from 560 cm 1 to 780 cm 1 (Mini-TES channels 24 45) Surface Dust Component [ 25] The rocks and soils observed along the rover s traverse are mantled to varying degrees by Fe-oxide-rich dust [Bell et al., 2004] whose Mini-TES spectrum is similar to the global average dust spectrum measured by TES [ Christensen et al., 2004a]. We have assumed that the TIR spectral features of the dust are homogenous throughout the region and only vary in contrast according to the amount of dust present on agiven surface. The darkest rocks are the least dusty [ Bell et al., 2004], so we limit our mineralogical determinations to these rocks in an effort to minimize the impact of surface dust on our results. We further assume that the spectrum of the dust component 10 of 36

11 Table 2. Best-Estimate Mineralogy of Four Rock Types in Gusev Crater Based on Combining the Output From the Lowest RMS Mini-TES Spectral Deconvolutions a Mineral Adirondack Class Na-plagioclase 20 Clinopyroxene 30 Olivine (Fo 10&60 ) 40 High-silica glass 10 Value Clovis Class Na-plagioclase 5 10 Pyroxene 5 10 High-silica glass 5 Basaltic glass Sulfate Goethite 5 10 Secondary silicates 0 20 Wishstone Class Ca-plagioclase 55 Olivine (Fo 10&60 ) 15 Basaltic glass 10 Phosphate 10 Sulfate 10 Watchtower Class Na-plagioclase Orthopyroxene 5 Olivine ( Fo 60 ) 0 10 High-silica glass 0 15 Basaltic glass 35 50* Phosphate 5 Sulfate 5 15 Secondary silicates <5 a Values are in percent rounded to the nearest 5. See section 4.2 for a discussion of the uncertainties. combines linearly with the spectral components of the substrate, which appears generally validonthe basis of work by Johnson et al. [2002a] and Graff et al. [2001]. The dust spectrum used as an end-member (Figure 2) comes from amini-tes observation of adust-covered hollow [e.g., Grant et al., 2004] on the rim of Bonneville crater. Of all the dusty surfaces observed by Mini-TES throughout the mission, its spectrum shows the features of dust with the greatest spectral contrast, which we assume equates to the purest example Spectral Slope Component [ 26] Anadditional end-member spectrum has been added to address the apparent sloping continuum present in some rock spectra. In the case of some of the Adirondack Class rock spectra for example, this slope is evident as an overall decrease inemissivity from higher to lower wave numbers that does not conform to the appearance of the TES spectrum of an olivine-rich surface or the CIPW normderived spectrum of the Adirondack Class rock Humphrey (Figure 8). Such sloping can arise during the conversion of a spectrum from radiance to emissivity. Ifthe radiance of the sample is divided by aplanck curve where the temperature used for the Planck curve is too low,anegative slope will be superimposed on the emissivity spectrum. This is possible if the maximum emissivity of the target is less than one or if temperature errors are present in the initial calibration of the data or when the Mini-TES FOV contains surfaces with *The values are correct here. The article as originally published is online. 11 of 36 different temperatures, including different facets of the same rock. The latter situation is similar to that encountered during the testing-phase of Mini-TES in which a set of rock targets was measured, each of which was smaller than the Mini-TES FOV [ Christensen et al., 2003]. In this situation, a slope end-member served to account for the variable brightness temperatures of the rocks and mounting hardware within the Mini-TES FOV, producing more accurate results. [ 27] Wehave produced aslope end-member by dividing a Planck curve at atemperature of 250 Kbyone at 247 Kand normalized to an emissivity of 1.0 at 1300 cm 1 (the approximate location of the Christiansen feature of most silicate rocks). These temperatures are intended to simulate those that commonly were encountered during measurement of the plains rocks. We treat slope as an end-member spectrum rather than attempting an a priori correction because there is no way of knowing what temperature error may exist in agiven spectrum. The angle of the slope is adjusted via the deconvolution algorithm where a fractional amount of the spectrum <1 decreases slope and >1 increases slope. Any fractional contribution of the slope end-member is analogous to the use of a blackbody end-member, the utility of which was first recognized by Lyon [1965] and now is widely used in deconvolution of thermal infrared emissivity spectra. The utility of the slope end-member in deriving more accurate mineralogical results is demonstrated in section Mirror Dust Correction [ 28] The mirror-dusting event of sol 420 occurred during a time when the rover was encountering Wishstone Class rocks. Consequently, there are spectra of these rocks before and after the event that help in validating our correction. Figure 9 demonstrates the spectral variations of Wishstone Class rocks after sol 420, documenting the variations according to the D T between the rock and the mirror. Most of the variation can be attributed to dust on the Mini-TES pointing mirror. Those spectra that have a deep, generally U-shaped feature in the middle wave number range ( cm 1 )show the effect of dust when the rock target is warmer than the mirror dust. This shape results from the broad, deep absorption feature in the dust superimposed on the spectrum of the rock. The dust feature is observed as an emission peak in the same location when the mirror dust is warmer than the rock target. This contribution is similar to that of the additional downwelling radiance from atmospheric dust described in section 3.2. The spectra in Figure 9 that show a weak convex-up shape in the middle wave number range but lacking a clear peak also are impacted by dust in emission. Distinguishing between contributions from warm dust on the mirror and warm dust in the atmosphere requires additional analysis, but in general, rocks measured in the late afternoon (>16:00 LTST) are much less affected by atmospheric dust. It is notable that the low wave number range of these spectra shows little variability due to dust, an additional demonstration of the importance of this spectral region for spectral classification. [ 29] A preliminary scheme has been devised to allow for a semiquantitative correction for mirror dust. It makes use of the Mini-TES external calibration target that, although unused for standard calibration of spectra due to the failed

12 Figure 9. Wishstone Class spectra showing systematic variations due to the temperature difference ( D T) between the rock and the dusty Mini-TES pointing mirror following asol 420 wind event. All of these spectra were measured after 16:00 local true solar time such that the effect of downwelling radiance is small compared tothe effect of mirror dust. The temperature values shown indicate the D TinKelvin between the rock and mirror. When this D Tislarge and positive, absorption by the mirror dust is prominent. As this value gets smaller and becomes negative, emission by the mirror dust becomes prominent. Note that the low wave number range is mostly unaffected. PRT (see section 2), serves as a known spectral reference. The spectral character of the external calibration target (a V-grooved, black-painted aluminum plate) at the beginning of the mission was nearly unit emissivity,but over time the accumulation of dust on its surface resulted in a departure from its initial blackbody character (Figure 10). The change occurred sometime between sol 13 and sol 102, with the large uncertainty due to the absence of external calibration target observations during this time. From sol 102 until the mirror-dusting event on sol 420, the spectrum of the external calibration target remained relatively constant (Figure 10). After sol 420 the spectrum changed dramatically, revealing adeep V-shaped feature centered at 1100 cm 1. The spectral contrast of this feature is dependent on the temperature difference (D T) between the target and mirror (the target temperature is derived from its measured spectral radiance; the mirror temperature is measured with an embedded PRT). The greater the temperature of the target relative to the mirror, the deeper the feature is. As the D Tbecomes negative, i.e., when the target becomes cooler than the mirror,aninverted V-shape emerges. Prior to sol 420, the spectral contrast of the external calibration target was independent of the D T, an indication that dust had not yet accumulated on the mirror. [ 30] Although the solar panels were cleared of a measurable amount of dust during the wind event [ Greeley et al., 2006], Pancam and Navcam images of the external calibration target before and after the event show no obvious change in the accumulated dust on the target, although this needs to be quantified. If we assume for now that the spectral shape of the external calibration target is that which was observed between sols 102 and 420, then any departure from this shape is due to the mirror dust. To correct the spectrum of arock or soil target we use acorrection based on the spectrum of the external calibration target with a similar D Tasthe surface target,selected from the many 10s of such spectra measured since sol 420. For example, the spectrum of a Wishstone Class rock called Rousseau (Figure 11) with a D Tof 12 Kcan be corrected using the spectrum of the external calibration target with a comparable D T according to the equation e cor ¼ e rock e cal e cal pre dust ; ð 4 Þ where e cal is the emissivity of the external calibration target with a comparable D T and e cal pre-dust is the average emissivity of the external calibration target prior to sol 420. The result is shown by the purple spectrum in Figure 11. The corrected Rousseau spectrum is much more similar to the average Wishstone Class spectrum obtained from rocks measured before sol 420. Although there is some mismatch in the middle wave number range, it may be due to actual variations between this rock and the average Wishstone spectrum. At this point, such a correction serves to better assess the spectral character of rocks that have no pre-sol 420 counterpart, which is the case for two of the basaltic exotics described in section 5.6 as well as avariety 12 of 36

13 Figure 10. Spectra of the Mini-TES external calibration target over time, all of which have a temperature difference between the target and pointing mirror of 30 K ± 3. The dramatic change between sol 414 and sol 422 is the result of dust blown onto the Mini-TES pointing mirror by awind event on sol 420. The smaller change between sol 013 and 166 is due to the accumulation of dust on the external calibration target on the rover deck. Figure 11. The spectrum of the Wishstone Class rock called Rousseau before and after correction for mirror dust compared with the uncontaminated average Wishstone Class spectrum. Rousseau (sol 561, sequence ID p3095) shows the effect of alarge positive temperature difference between the rock and the Mini-TES pointing mirror ( 12 K). The large difference between the average Wishstone Class spectrum (long dashes) and Rousseau (solid) inthe middle wave number range is due to absorption by mirror dust. We have developed asemiquantitative correction that relies on measurements of the Mini-TES external calibration target that removes most of the mirror-dust artifact (short dashes). 13 of 36

14 Figure 12. Mini-TES spectra of rocks from the Gusev plains show acommon set of features best represented by the emissivity peak at 430 cm 1 that is due mostly to olivine. Aspectrum of Gusev surface dust (red) is shown for comparison. of other rock types not discussed in the current work. Future improvements to the correction likely will need to account for the different radiance terms associated with the target and mirror dust to minimize the potential for introducing slope into the corrected spectrum. The validity of the assumption of a known spectral appearance for the external calibration target also will have to be assessed more rigorously through the analysis of Pancam color images of the target. If it can be demonstrated that the dust on the external calibration target has remained unchanged, then the assumption is robust. 5. Results 5.1. Adirondack Class Rocks [ 31] Hundreds of spectra of rocks on the plains from the lander to the base of the West Spur (Figure 1) were acquired both from targeted Mini-TES observations of individual rocks and with multiple-pointing raster observations that include many rocks. Immediately apparent in images from the rover cameras is the fact that the rocks display arange of apparent albedo (hereafter referred to as tone to indicate relative versus quantitative albedo). Pancam spectra indicate that the range from light-toned to dark-toned appears to represent arange of dust coating or mantling as opposed to an intrinsic characteristic of lithology [ Bell et al., 2004; Johnson et al., 2006]. We validate this observation with Mini-TES spectra in the following section along with a description of the spectral features used to classify Adirondack rocks Spectral Characteristics of Adirondack Class Rocks [ 32] Examples of Mini-TES spectra of dark-toned plains rocks are shown in Figure 12 among which is the spectrum of Adirondack, the type example for the olivine-rich class of plains rocks [Squyres et al., 2006]. These spectra and all Mini-TES spectra are identified with a sol number, a sequence identifier, and a name. A Mini-TES spectrum of surface dust also is included for comparison. Several important features are apparent in Figure 12. The rock spectra show lower emissivity at low wave numbers (<600 cm 1 ) and higher emissivity at high wave numbers (>1250 cm 1 ) compared to the dust spectrum. These characteristics are consistent with those used to identify dust-free surfaces at the global scale [Ruff and Christensen, 2002]. Most significant is the lack of a roll-off or decrease in emissivity at high wave numbers that is diagnostic of the presence offine particulate silicates (< 60 m m). Despite the increased noise in Mini-TES spectra at high wave numbers due to the decrease in Planck radiance, the signal is sufficient to identify the presence or absence of the roll-off. Although dark-toned rocks are not dust-free, as evidenced by beforeand-after images of surfaces brushed by the RAT[Bell et al., 2004], the dust that is present does not mask the features of the substrate. The TIR wavelengths measured by Mini-TES essentially are insensitive to a few microns of surface dust [Graff et al., 2001; Johnson et al., 2002a]. [ 33] Using the low wave number spectral characteristics of dark-toned rocks as adiscriminator, Mini-TES spectra (Figure 12) reveal acommon rock type from the lander to Bonneville crater, to the base of the Columbia Hills. With the exception of the apparent coating on the brushed surface of the rock Mazatzal [Christensen et al., 2004a], dark-toned Adirondack Class rocks show a characteristic emissivity peak at 430 cm 1 that persists with varying spectral contrast regardless of the appearance of the middle wave number range or CO 2 feature. This emissivity peak and its location uniquely identify the Adirondack Class rocks 14 of 36

15 Figure 13. Comparison of light-toned rocks before and after brushing with the Rock Abrasion Tool (RAT). (a) The spectra of the light-toned, natural surfaces of two plains rocks (Mazatzal, sol 086, p3218 and Route66, sol 091, p3223) and two rocks from the Columbia Hills (Clovis, sol 233, p3339 and Scale, sol 382, 3862) look very similar to dust from acold surface where the spectrum has been distorted by downwelling radiance (magenta; alsosee Figure 4c). The same surface when warm produces adust spectrum that is less distorted (green) and does not match the cold, dusty rocks. (b) Multiple placements of the RAT have been used to clear dust from alarge enough area to fill the Mini-TES field of view as approximated by the white circle in this approximate true color Pancam image of the rock Route 66 courtesy of marswatch.astro.cornell.edu/pancam_instrument/true_color.html. (c) The spectra of brushed surfaces from the same rocks as in Figure 13a show very different features compared to their unbrushed version. compared with other rock types observed in the Columbia Hills. [ 34] The light-toned rocks observed by Mini-TES share a common set of spectral features (Figure 13a) that results from surface dust that is thick enough to fully obscure the spectral character of the substrate at TIR wavelengths (i.e., >100 m m[graff et al.,2001; Johnson et al.,2002a]). This is displayed convincingly where such rocks have been brushed using the RAT. Multiple placements of the RAT on agiven rock have been used to clear dust from alarge enough area to fill the FOV of Mini-TES (Figure 13b). The resulting spectra (Figure 13c) are dramatically different from their unbrushed version, which also is true for lighttoned rocks in the Columbia Hills. [ 35] Because we now recognize the spectral distortions caused by additional downwelling radiance, it is clear that the spectra of dust-covered rocks most closely resemble that of surface dust when it has asmall ornegative temperature difference ( D T) relative to the atmosphere. The dusty soil in front of the rock Bonneville Beacon serves to make this point. The spectra from both acold and warm version of this surface are shown in Figure 13a. The unbrushed rock spectra better match the cold dust than the warm dust spectrum demonstrating that additional downwelling radiance has the same impact on the spectrum of dust whether the dust is covering soil or rocks. Given that dust on arock raises its albedo and thus diminishes the efficiency of solar heating, dusty rocks typically have an unfavorable D T with the atmosphere leading to the distorted dust spectral character. The distorted dust spectrum of light-toned rocks serves an indicator that the actual spectral character is masked by surface dust. We have observed no light-toned rocks with a spectrum other than that of distorted surface dust. We conclude then that all light-toned rocks on the plains are dust-covered Adirondack Class rocks rather than a compositionally distinct population Mineralogy of Adirondack Class Rocks [ 36] The combined data from all of the rover s analytical instruments make a strong case for the identification of olivine-bearing basaltic rocks on the plains [e.g., Squyres et al., 2004]. The initial Mini-TES results demonstrated that the spectra of these rocks could be modeled using olivine of intermediate composition ( Fo ) along with other components [ Christensen et al., 2004a]. However, there were notable inconsistencies between the modeled results and other observables. Most striking was the fact that the spectra of the darkest, least dusty rocks were modeled using an inordinate amount of a surface dust spectral end-member (>50%). Also problematic was the absence of any feldspar component in the modeled results. We now recognize that the spectra of many of the Adirondack Class rocks have features that arise from additional downwelling radiance. The initial deconvolution did not address this problem and therefore yielded spurious results. [ 37] The best examples of plains rocks for use in deconvolution are those that are the least dusty and show the least impact from downwelling radiance. This applies to the darktoned rocks where the spectra show CO 2 in absorption and a concave-up shape in the cm 1 range. These two spectral parameters are telltale features for diminished contribution of downwelling radiance as described in section However, there are very few Adirondack Class spectra that display this combination of features. The best example is the rock named Sarah, yet even this spectrum is not free of contributions from downwelling 15 of 36

16 radiance that persist following the standard downwelling correction as described in section [ 38] To investigate the mineralogy of the rock Sarah, its spectrum was deconvolved using a suite of mineral spectra (end-members) along with a sky spectrum. In this case, a sky spectrum (30 elevation angle) was available from a comparable time of day as the Sarah spectrum ( 13:00) but from five sols after the Sarah measurement. The standard downwelling correction (equation (3)) was applied to account for the small but non-negligible radiance component from the sky that is well understood. An end-member set was assembled for this work that contained spectra of components expected in olivine-rich basalt along with some alteration phases (Table 1). [ 39] Three cases were used to investigate the role of the slope and sky end-members: one in which the slope and sky were excluded, asecond where slope was included, but without the sky spectrum, and a third that included both slope and sky spectra. The results are shown in Figure 14. In the absence of slope and sky end-members, the Sarah spectrum is poorly modeled (Figure 14a), which is evident both in the quality of the fit and in the modeled components. Most notable is the lack of any plagioclase component, which is highly unlikely on the basis of the normative analysis from the chemistry of Adirondack Class rocks [ McSween et al., 2004]. Through the use of the slope end-member, the second case shows a factor of two improvement in the spectral goodness of fit (indicated by root-mean-squared (RMS) error) and more realistic components (Figure 14b). Although absolute RMS error values are ill suited for comparisons of fit with other rock spectra, they are well suited to evaluating improvements to the fit of a single spectrum. A plagioclase component is modeled in this case, although at <10% abundance, it is inconsistent with the 40% in the published CIPW norms for Humphrey [ McSween et al.,2004]. Finally, the incorporation of the sky end-member yields only aslightly better fit (Figure 14c) but with alarger fraction of plagioclase, likely an indication of a more accurate result. By subtracting out and normalizing for the surface dust, slope, and sky components, a new spectrum of Sarah emerges (Figure 15) that compares more favorably to the TES spectrum from olivine-rich material in Nili Fossae and to the spectrum we produced from the published CIPW norms for Humphrey [ McSween et al., 2004]. Although the match is imperfect, we interpret the similarity as support for the strategy of using slope and sky end-members in the deconvolution. The results suggest that the Adirondack Class rocks are consistent with the following mineralogy: plagioclase with a sodic composition that may be zoned on the basis of the presence of oligoclase and albite in the results; clinopyroxene with no clear indication of any orthopyroxene; and olivine with a composition that includes both Fo 10 and Fo 60,perhaps indicative of zoning. Ahigh-silica glass component is modeled at <10% and may not be a robust identification. Table 2 includes a best estimate of the mineralogy of Adirondack Class rocks based on the deconvolution results of Sarah shown in Figure 14c Clovis Class Rocks [ 40] Upon reaching the Columbia Hills the rover first encountered the West Spur of Husband Hill (Figure 1). Beginning with a collection of variably dusty rocks named Viera Cairns on the west side of the West Spur and across to its eastern contact with the plains, all of the rocks observed by the Mini-TES display adistinctive spectral feature that readily separates them from the Adirondack Class rocks and allows them to be grouped into one spectral class. The major-element geochemistry and Mö ssbauer spectra of these rocks also supports their grouping into a single class named Clovis after the type outcrop [ Squyres et al., 2006] Spectral Characteristics of Clovis Class Rocks [ 41] Figure 16 shows examples of the range of spectral characteristics of the rocks from west to east along the traverse of the West Spur. Although there are significant variations in the cm 1 range due to downwelling radiance, the low wave number range (< 600 cm 1 ) shows much less variation. As demonstrated in section 3.2, the middle wave number range is very sensitive to contributions from atmospheric dust whereas the low wave number range is much less so. The spectra that are least encumbered by atmospheric contributions are those that show the CO 2 feature strongly in absorption. Unlike the plains rocks, many of the rocks on the West Spur have spectra that display such aco 2 feature. This is due mostly to the fact that atmospheric dust opacity had dropped significantly by the late fall season [ Lemmon et al., 2004] and with it, atmospheric temperatures [Smith et al., 2004]. The combination of a cooler, less dusty atmosphere and rocks with lower thermal inertia [Fergason et al., 2006] and hence slightly warmer afternoon temperatures led to more favorable conditions for reduced impact from atmospheric downwelling radiance than with the plains rocks. The West Spur rocks have a broad, U-shaped feature in the middle wave number range ( cm 1 )unlike anything encountered in the spectra of the Adirondack Class rocks. But it is the deep, relatively narrow absorption at 460 cm 1 that is so distinctive. This feature persists with varying spectral contrast in all dark-toned rocks observed by Mini- TES on the West Spur regardless of the appearance of the middle wave number range or CO 2 feature. It thus serves as adiscriminator for these rocks comparedwith the other rock types encountered in Gusev crater Mineralogy of Clovis Class Rocks [ 42] All of the rover s instruments indicate that the Clovis Class rocks are distinct from Adirondack Class rocks. Observations made with the Microscopic Imager (MI) show that these rocks are clastic and poorly sorted compared to the basaltic Adirondack Class rocks that display probable olivine megacrysts in an aphanitic groundmass [ McSween et al., 2004; Squyres et al., 2006]. APXS and MB results indicate asignificant level of alteration [ Gellert etal., 2006; Ming et al., 2006; Morris et al., 2006]. Pancam multispectral measurements reveal a relative reflectance maximum at shorter visible wavelengths and anear infrared absorption that is either absent or shallower than in Adirondack Class rocks [Farrand et al., 2006]. Their dramatically different TIR spectral character compared to Adirondack Class also indicates mineralogy quite different from olivine-rich basalt. Because of the strong spectral similarity among all of the rocks of the West Spur, we have produced an average spectrum for deconvolution with the intent of improving spectral signal-to-noise and diminishing subtle variations between the spectra that may not be significant. Although five rocks on the West Spur were subjected to full IDD 16 of 36

17 Figure 14. Spectraldeconvolution of the Adirondack Class rock Sarah using the end-member spectra listed in Table 1. The raw results with values in % abundance are shown (see section 4.2 for caveats). The second column of values represents the modeled components normalized to exclude surface dust, slope, sky, and blackbody components. (a) With the slope and sky end-members excluded from the first deconvolution run, the modeled fit is poor, and the modeled end-member components are unrealistic. (b) Including aslope end-member to account for any temperature errors significantly improves the modeled fit. (c) Including asky end-member yields aslight improvement in fit as quantified by the spectral root-mean-squared (RMS) error, but with changes in the modeled end-member components that better match the CIPW norms derived from the APXS chemistry of the nearbyhumphrey rock. 17 of 36

18 Figure 15. Upon removal and normalization for the surface dust, slope, and sky components obtained from the deconvolution of Sarah shown in Figure 14c, the spectrum (black) is notably similar to that produced from the CIPW norms derived from the APXS chemistry of the nearby Humphrey rock (purple) as well as ates spectrum of the olivine-rich Nili-Fossae region (blue). Figure16. Spectra of Clovis Class rocks from the West Spur. Like Adirondack Class rock spectra, there is variation in the middle wave number range of Clovis Class rock spectra due to distortion by downwelling radiance that coincides with the magnitude and direction of the CO 2 feature. The relatively deep and narrow absorption at 460 cm 1 is common to all Clovis Class rocks and is mostly unaffected by downwelling radiance. 18 of 36

19 campaigns (Wooly Patch, Clovis, Ebenezer, Tetl, and Uchben), none are included in the Mini-TES spectral average because they were either too cold or too dusty to yield good quality spectra. However, Clovis and Ebenezer display the feature at 460 cm 1 with sufficient clarity (Figure 16) that we are confident in the assignment of all rocks with this feature to Clovis Class. [ 43] An average spectrum of Clovis Class rocks was produced from six of the spectra shown in Figure 16 (Bramsen, Toltecs, Plate, Pico, Palenque, and Cocomama). Each was first subjected to the standard downwelling radiance correction (equation (3)) resulting in only minor changes in spectral contrast. Six deconvolution runs were undertaken using the end-member sets shown in Table 1. These multiple iterations using different combinations of end-members were intended to investigate the potential alteration phases in the Clovis Class rocks as suggested by results from the other rover instruments [e.g., Ming et al., 2006]. The sky spectrum used in each case came from an average of the six different sky observations used in the standard downwelling correction of the individual rock spectra. [ 44] The first deconvolution run used an end-member set consisting only of primary igneous minerals found in basaltic rocks (Table 1). The poor fit of the modeled spectrum (Figure 17a) makes it clear that the Clovis Class rocks must have components beyond just these primary igneous minerals. It is instructive though to see that the dominant component in this deconvolution is albite. This is due to the fact that the Christiansen feature (the emissivity maximum) of the Clovis Class average spectrum occurs at a relatively high wave number position ( 1250) and the only available end-member spectrum that can fit this feature is albite. As is shown below, however, some sulfates also have their Christiansen feature in this location and are more likely components than albite. [ 45] The incorporation into the end-member set of spectra from glassy materials ranging from experimentally shocked plagioclase [ Johnson et al., 2002b] to basaltic glass, improves the fit of the deconvolution results (Figure 17b), most notably in the low wave number range. The absorption feature at 460 cm 1 that is so diagnostic of the Clovis Class rocks is better fit by glass components than by crystalline igneous phases alone. This deconvolution marks the first appearance of a spectral component that dominates all subsequent runs. It comes from a hand sample of naturally quenched basalt collected by R.V. Morris (HWKV340A) from a recent aa flow on Hawaii s Kilauea volcano that is spectrally similar to an artificially (water) quenched Kilauea basalt described by Wyatt et al. [2001]. For convenience we will refer to it as abasaltic glass to emphasize its amorphous spectral character although the state of crystallinity has not been established via XRD. Its spectrum supplies afeature at 460 cm 1 that best fits that of the Clovis Class average spectrum (Figure 18). It also better fits the notable asymmetry of band depth between the middle and low wave number spectral ranges. [ 46] The APXS measurements of Clovis Class rocks revealed an elevated sulfur contentcompared to Adirondack Class rocks that is assumed to be due to sulfate minerals [Gellert et al., 2006; Ming et al., 2006], which supports the inclusion of sulfates in the deconvolution end-member set. A set of six sulfate spectra was added to the glass spectra and redundant primary igneous end-members were extracted (Table 1) to maintain a total number of end-members of 60. The addition of sulfates lowered by nearly 50% the RMS error of the modeled fit (Figure 17c). Asmall amount of gypsum ( 10%) improves the spectral fit in the range from 1000 to 1250 cm 1, the location of the most prominent absorption feature of sulfates. The basaltic glass component still dominates, with lesser plagioclase and pyroxene remaining as permissible components (i.e., they improve the fit). [ 47] Mössbauer measurements of Clovis Class rocks give clear indication of goethite with lesser hematite [ Morris et al., 2006], justifying the inclusion of these additional spectral end-members in the deconvolution analysis. Magnetite was also added to better evaluate the possible contributions from Fe-oxide minerals (Table 1). Deconvolution using these additional end-members resulted in a 25% reduction in RMS error with goethite emerging at 10% abundance (Figure 17d). Goethite has a strong absorption at low wave numbers and alesser doublet in the middle range that serve to improve the fit (Figure 18). The combination of basaltic glass, sulfate, goethite, and minor pyroxene and plagioclase produces the lowest RMS error of any other combination of end-members. Although the quality of the modeled fit and low RMS value suggest that this combination of phases is present in Clovis Class rocks, it may not represent aunique solution. The additional deconvolution runs described below explore the uniqueness of this result. [48] Phyllosilicates [Ming et al., 2006] and nonspecific secondary aluminosilicates [Wang et al.,2006] have been suggested as significant components of Clovis Class rocks on the basis of normative analyses of APXS data. Consequently, we have investigated this possibility using an endmember set that includes many of the candidate minerals suggested by these authors (herein referred to as secondary silicates) along with the previous end-members (Table 1). The deconvolution result (Figure 17e) shows a slight improvement in the fit below 1100 cm 1 that is not reflected in the overall RMS error, which essentially remains constant relative to the previous run (0.141 versus 0.144). This result demonstrates that minor amounts ( 20%) of secondary silicates are permissible components of the Clovis Class average spectrum but their inclusion neither improves nor degrades the modeled fit over that obtained using no secondary silicates. [ 49] Given that some secondary silicates are spectrally similar to some amorphous materials [e.g., Wyatt and McSween, 2002; Kraft et al., 2003; Michalski et al., 2006], the predominance of a basaltic glass component in all of the deconvolution runs in which it was included merits additional scrutiny. We successively excluded glass end-members beginning with the HWKV340A basaltic glass spectrum that had been the persistently dominant component. With each successive run, different examples of basaltic glass spectra dominated the results although with increasing RMS error, until in the absence of any basaltic glass end-members, maskelynite was incorporated as the dominant component. In nature, maskelynite or diaplectic glass is produced by shock pressure applied to feldspar during meteorite impact. The sample used in this 19 of 36

20 Figure of 36

21 Figure 18. The average Clovis Class spectrum versus laboratory spectra of various candidate phases. The feature at 460 cm 1 is best fit by basaltic glass. The prominent spectral doublet of dioctahedral clay minerals (bottom four spectra) in the low wave number range is not agood match to this feature. work is of labradorite composition from the Manicouagan impact structure [Hamilton et al., 1997] and is grossly similar spectrally to basaltic glass (Figure 18). We concluded this series of deconvolution runs by excluding the spectra of all glassy materials from the end-member set including shocked feldspars (Table 1) to establish how well crystalline end-members model the Clovis Class spectrum. The result is apoor fit with nearly double the RMS error of the best-fit case involving basaltic glass (Figure 17f), demonstrating that crystalline secondary silicate minerals cannot substitute for the basaltic glass component. It is noteworthy that in this and previous deconvolution runs where secondary silicates were included, serpentine, heulandite (a zeolite mineral), and saponite (a trioctahedral phyllosilicate) are present in one or more of the modeled results. No dioctahedral clay minerals are among the results, which likely is due to the fact that they have astrong doublet that does not fit the 460 cm 1 feature of the Clovis Class spectrum (Figure 18). Trioctahedral clays, serpentines, and some zeolites have a single strong absorption in this spectral region, making them better candidates. [ 50] The library of spectral end-members available to us at the time of writing did not include a suite of amorphous or poorly crystalline secondary silicates such as allophane, the presence of which has been advocated by Ming et al. [2006] for Clovis Class rocks. If such materials have a spectrum similar to that of the basaltic glass that so dominates the spectrum of Clovis Class rocks, then it is possible that what now appears to be asignificant primary glass component is actually the result of aqueous alteration. Although all amorphous materials are broadly similar, there certainly are important characteristics that differentiate them spectrally, i.e., they are not interchangeable, a fact demonstrated in the current work. Therefore future work will be needed to resolve the question of whether the amorphous component in Clovis Class rocks is primary or secondary in origin. Additional discussion of this issue is given in section 6. [ 51] The Mini-TES result indicating the presence of a sulfate component is consistent with the elevated sulfur in Clovis Class rocks as determined by the APXS [ Gellert et al., 2006]. The exact mineralogy of the sulfate is somewhat less certain. Three of the four deconvolution runs shown here that included sulfate end-members resulted in gypsum as the sole sulfate component. The fourth run resulted in Figure 17. Spectral deconvolution of the average Clovis Class spectrum using various combinations of end-member spectra. The raw results with values in % abundance are shown (see section 4.2 for caveats). The secondcolumn of values (in %) represents the modeled components normalized to exclude surface dust, slope, sky, and blackbody components. (a) Using only the crystalline, primary igneous minerals shown in Table 1, a poor fit is obtained. (b) Combining the amorphous phases shown in Table 1substantially improves the modeled fit. The feature at 460 cm 1 is better fit by volcanic glasses than by crystalline igneous minerals. (c) An enrichment in sulfur in Clovis Class rocks measured by the APXS prompts the use of sulfate end-members in the deconvolution (Table 1). The resulting fit has a nearly 50% lower RMS error over the previous iteration and clearly visible improvements to the fit throughout the spectrum. (d) By incorporating the Fe-oxide minerals detected with the Mössbauer instrument into the end-member set (Table 1), amodest improvement in fit is obtained,reaching the lowest RMS value of any of the iterations. (e) The inclusion of the secondary silicates shown in Table 1has no significant impact on the modeled fit but demonstratesthat minor abundances ofsome phyllosilicates and perhaps zeolites are possible. Dioctahedral clay minerals are notably absent in the results. (f) The spectra of all amorphous end-members were removed in the final iteration (Table 1) to evaluate the potential for crystalline secondary silicates to substitute for them. The result is anear doubling of the RMS error and clear misfit throughout the spectrum. 21 of 36

22 kieserite and epsomite in place of gypsum. This latter case is one of the two with the lowest RMS error, soitappears that there is not just one type of sulfate capable of supplying a good fit. [52] Spectral contributions from surface dust, slope, and sky are persistent components in the results of all of the various deconvolution runs. The most meaningful assessment of their contributions comes from the best-fit model results, which are the two that include glass, sulfates, and oxides (Figures 17d and 17e). In both cases, surface dust is present at 10% abundance, which could mean a checkerboard mix of clear and dust-covered surfaces, a continuous cover of dust with athickness that produces 10% of the spectral contrast of a spectrally opaque layer of dust [Crisp and Bartholomew, 1992; Graff et al., 2001; Johnson et al., 2002a], or some combination of these. The slope component is 10%, which translates into an equivalent temperature determination error of only a few tenths of a degree. The sky component is incorporated at 5% abundance which means that it is reduced in contrast by 95%. Given the already very low contrast of the sky spectrum for the low dust opacity period during which the Clovis Class rocks were measured, 5%is a modest correction. [ 53] The above set of deconvolution runs yields a picture of the bulk mineralogy of Clovis Class rocks that is necessarily restricted to only the phases that we used. Basaltic glass accounts for 40% of the total components after normalization for surface dust, slope, sky, and blackbody components. Combined with another 5% of an obsidian-like glass, these rocks may be composed of nearly half primary glass material. Olivine is notably absent from the best-fit deconvolution results, which is consistent with only a minor amount detected by the MB [ Morris et al., 2006]. Plagioclase and pyroxene components are not uniquely identified but 5 15% each of sodic plagioclase and either ortho- or clino- pyroxene are permissible components. Sulfates are required to produce the best fits to the Clovis Class average spectrum with gypsum at 10% abundance the most likely candidate but with epsomite and perhaps kieserite as permissible components at up to 25% total abundance. Goethite is the only Fe-oxide detected and is required in the amount of 5 10% to achieve the best fits. Finally, secondary silicates such as serpentine, heulandite, and saponite with acombined total of 20% are permissible components of Clovis Class rocks but are not uniquely identified. Dioctahedral clay minerals like montmorillonite are unlikely constituents of this rock class due to their low wave number doublet absorption feature that is a poor fit to the pronounced singlet of the Clovis Class spectrum (Figure 18). Table 2 includes a best estimate of the mineralogy of Clovis Class rocks based on the results from the two deconvolution runs with the lowest RMS values Wishstone Class Rocks [ 54] With the completion of the traverse from west to east across the West Spur (Figure 1), the rover passed an apparent lithologic contact manifested by the absence of rocks with Clovis Class spectral character and the presence of rocks with Adirondack Class spectral character. But within 40 m of this contact, a single float rock was observed with adistinctly different spectrum than either Adirondack or Clovis Class (Plymouth, sol 321, p3808). Another 150 mdrive brought the rover to the lower flank of Husband Hill upon which were 10s of rocks with similar spectral character. Pancam mult ispectral observations of these rocks also are distinct from both the Adirondack and Clovis Class rocks with the presence of a relative reflectance maximum at 754 nm and NIR band minimum at 934 nm with asteep slope from 754 to 864 nm (indicative of a more pronounced NIR absorption than was observed in the Clovis class rocks). An IDD campaign on arock dubbed Wishstone provided the name for this new class of rocks [ Squyres et al., 2006] Spectral Characteristics of Wishstone Class Rocks [ 55] Figure 19a presents 27 spectra that reveal the similarities and differences among the Wishstone Class observations. As with the Clovis Class rocks of the West Spur, significant spectral variations due to atmospheric dust are present in Wishstone Class rocks in the middle wave number range depending on the D T between rock and sky. An additional contribution from dust appears in the spectra following sol 420 due to the accumulation of dust on the Mini-TES optics (sections 2 and 4.3). Atmospheric dust opacity remained low during the first weeks of the initial encounter with the new rock type, but the winter season produced low surface temperatures such that typical rock temperatures were 250 K. Consequently, the late afternoon observations in which the D T was the largest produced the best spectra, examples of which are shown in Figure 19b. The low wave number range (< 600 cm 1 ) again yields the most diagnostic spectral features. Wishstone Class rocks all display aremarkable set of features at low wave numbers unlike any other rocks observed to date: anarrow emissivity peak at 560 cm 1 followedbyasharp absorption trough at 540 cm 1 and anotably flat-topped or perhaps double-peaked, relatively wide emissivity peak centered at 470 cm 1. A more subtle but distinctive emissivity peak is evident at 1070 cm 1 that represents adeparture from Clovis Class rocks in the middle wave number range. All of these features are present in the spectra of intermediate plagioclase feldspars (Figure 19b). We have observed at least 95 rocks with these spectra l features scattered across the north-northwest flank of Husband Hill Mineralogy of Wishstone Class Rocks [56] For the purpose of deconvolution, spectra from four different rocks were averaged together to improve the SNR and diminish spectral characteristics that may not be representative of the class (Figure 19b). The spectrum of the rock Wishstone is not included in this average because the two different observations of it yielded inferior spectra. However, the quality of one of the two is sufficient to demonstrate a clear linkage with the other rocks whose spectra represent the Wishstone Class average, i.e., the diagnostic low wave number features are in evidence (Figure 19a). Most Wishstone Class rocks were observed after sol 420, but their spectra are not included in the average because of the contribution from mirror-dust. As with the Clovis Class average spectrum, an atmospherically corrected version of the Wishstone Class average spectrum has been used for deconvolution runs. This uses the standard downwelling correction (equation (3)) applied to each individual spectrum in the average. The CO 2 feature appears 22 of 36

23 Figure 19. Spectra of Wishstone Class rocks from the north flank of Husband Hill. (a) Spectral variations in the middle wave number range are due to distortions from downwelling radiance and, following awind event on sol 420, dust on the Mini-TES pointing mirror. (b) Four Mini-TES spectra were averaged for use in spectral deconvolution. As shown by the laboratory spectra of three plagioclase feldspars, Wishstone Class rocks have clear spectral features arising from plagioclase. strongly in absorption for each of the four spectra used in the average (Figure 19b) so we have assumed that any additional downwelling component is small. However, an average sky spectrum is included in the end-member set in an effort to address any non-negligible additional downwelling component. The same spectra of slope and surface dust as used in the previous deconvolution runs are used here. [57] The initial deconvolution run (Figure 20a) used only spectra from primary igneous minerals (Table 1) demonstrating that the diagnostic features that identify Wishstone Class spectra can be fit using just these components. Most dominant is a plagioclase component as expected on the basis of the clear presence of spectral features attributable to plagioclase. However, there remains obvious misfit from 800 to 1000 cm 1 demonstrating the absence of one or more components from the modeled results. [ 58] Measurements from APXS and MB warrant the addition of other end-member spectra. Data from the APXS indicate a pronounced phosphorous enhancement in Wishstone, likely due to a phosphate component [ Gellert et al., 2006; Ming et al., 2006]. MB spectra show the presence of ilmenite [Morris et al., 2006]. The end-member set was expanded accordingly with available phosphates and oxides (Table 1) and the deconvolution was performed again. The results are essentially unchanged with no improvement in RMS error or any significant change in the modeled components (Figure 20b). This result does not yet rule out a phosphate or ilmenite detection because it is possible that other components are still missing that would balance spectral contributions from these minerals. [ 59] The addition of glasses and shocked feldspars to the end-member set (Table 1) yields amodest decrease in RMS 23 of 36

24 Figure 20. Spectral deconvolution of the average Wishstone Class spectrum using various combinations of end-member spectra. The raw results with values in % abundance are shown (see section 4.2 for caveats). The second column of values (in %) represents the modeled components normalized to exclude surface dust, slope, sky, and blackbody components. (a) Arelatively good fit is obtained using only the primary igneous minerals listed in Table 1. (b) The addition of phosphate and Fe-oxide end-members in response to observations made with the APXS and MB instruments does nothing to improve the fit. (c) Including the amorphous phases shown in Table 1produces only aminor decrease in RMS error. (d) The final iteration incorporates the sulfate and secondary silicate phases shown in Table 1, producing amarked reduction in RMS error mostly due to asulfate component. error coming from an improved fit in the low wave number range (Figure 20c). The inclusion of 40% of a shocked plagioclase (An 75 [ Johnson et al., 2002b]) is anoteworthy change in the modeled results. However, the spectrum of this component shows only minor changes compared with the unshocked version, consistent with the relatively low shock pressure (22.6 GPa) applied in the creation of the sample [Johnson et al., 2002b]. As described below, this shocked plagioclase ultimately is not retained in the best-fit deconvolution results, which is an indication that it is not a robust detection. [ 60] Although evidence from the other instruments indicates that Wishstone Class rocks are only weakly altered [e.g., Ming et al., 2006], the potential for sulfate components is suggested by slightly elevated sulfur in APXS measurements [Gellert et al., 2006]. In the final deconvolution run, asuite of sulfate spectra was included along with spectra of secondary silicates for completeness in assessing the degree of alteration (Table 1). The results show anearly 40% reduction in RMS error (Figure 20d). Amajor reshuffling of the modeled components accompanied this deconvolution run. Plagioclase is still the dominant component at nearly 60% abundance, but the pyroxene component that was present at significant levels in the previous runs is absent in this final run. Olivine is retained at about the same abundance as in other runs ( 15%). A sulfate component (anhydrite) has been modeled at anon-negligible abundance ( 10%) and for the first time a phosphate component (wavellite) is present at 10 % abundance. Wavellite is a hydrated aluminophosphate of secondary origin. Additional deconvolution runs with an expanded set of phosphates is needed before wavellite can be considered a robust detection. Table 2 includes a best estimate of the mineralogy of Wishstone Class rocks based on the results from this run. 24 of 36

25 Figure 21. AMini-TES spectrum of the fine particulates produced from the RAT grind into Peace rock compared with other fine particulate materials measured by Mini-TES. (a) APancam approximate true color image showing the Peace RAT hole and abundant tailings with the approximate Mini-TES field of view shown by the white circle. (b) The PeaceRAT tailings show aprominent emissivity peak near 6 m m (black spectrum) that likely is due to acomponent with molecular boundwater. Although Gusev surface dust (purple) displays asimilar feature, the rest of its spectrum is notably dissimilar. AMini-TES spectrum of the RAT tailings on Bounce rock in Meridiani Planum (blue) shares similar spectral characteristics to Peace but without the bound water feature. Adominant pyroxene component in both rocks is likely responsible for the feature at 800 cm 1. [ 61] A small surface dust component is present in all of the deconvolution runs described above. The best fit case uses 10%, an indication that the rocks used in the average spectrum were relatively dust-free. The modest emissivity roll-off at high wave numbers supports this conclusion. None of the runs required asky component to achieve the best fit. Aslope component is present in the best-fit case at 15%, which translates to an equivalent temperature determination error of 0.5 C Peace Class Rocks [ 62] The first outcropping of rock encountered by the rover on Husband Hill is called Peace. Asecond outcrop called Alligator 20 m away has similar characteristics that combine to distinguish these rocks from others in the Columbia Hills. Their chemistry measured by APXS serves to group them into asingle class called Peace; additional observations suggest they are mafic sandstones cemented by sulfates [ Squyres et al., 2006]. Although both outcropswere measured prior to the mirror-dusting event on sol 420, they were poor targets for Mini-TES due to the abundance of dust on their surfaces. Mini-TES spectra from their undisturbed surfaces reveal the same features associated with other cold, dusty surfaces observed throughout the mission: the spectrum of dust distorted by downwelling atmospheric radiance (Figure 13a). However, the fine material generated from the RAT grinding operation on Peace yielded important spectral features whereas the brushed surface of Alligator provided a clearer view past the dust. [63] The RAT grind on Peace (Figure 21a) produced one of the deepest holes of the mission (9.7 mm), an indication of its relative softness compared with other rocks along the rover traverse [ Arvidson et al.,2006]. The tailings generated by the RAT accumulated in a substantial pile at the front of the rock. The Mini-TES observation of the RAT hole included much of this fine-particulate material. Fine particles (< 60 m m) produce spectral scattering effects that reduce or eliminate restrahlen features and introduce transparency features [e.g., Salisbury and Eastes, 1985]. The Mini-TES spectrum displays these characteristics (Figure 21b). A pronounced decrease in emissivity is present on the high wave number side of the Christiansen feature with relatively high emissivity throughout the rest of the spectrum except for adeeper feature at 800 cm 1 that we interpret as a transparency feature. Although these spectral characteristics generally resemble those of Gusev surface dust, there is a clear distinction between the Peace tailings and dust (Figure 21b). This also is apparent in the Pancam approximate true color image (Figure 21a), assuming the color of the tailings is not due to a photometric effect. However, an important overlap between dust and the tailings is present in the form of apronounced emissivity peak at 1630 cm 1 that typically is attributed to the fundamental bending mode of molecular bound water [e.g., Salisbury et al., 1991; Bishop and Pieters, 1995]. In the spectrum of Peace, it is unlikely that this feature is simply due to dust given the clear mismatch of the rest of the high wave number range (Figure 21b). Instead it is more likely that the 1630 cm 1 feature is attributable to one or more phases in the Peace rock that contain bound water. On the basis of the high sulfur content observed by the APXS [ Gellert et al., 2006], there could be a hydrated sulfate present that is responsible for the bound water feature. Also, Pancam spectra of the Peace RAT tailings exhibit visible-wavelength spectral 25 of 36

26 Figure 22. Mini-TES spectra of the target Scale on the outcrop called Alligator. (a) The spectrum of the natural surface (black) is dominatedbythe features of dust distorted by downwelling radiance, whereas the brushed surface reveals more distinctive spectral character (purple). (b) Although the spectrum of the brushed surface (black) is distorted by downwelling radiance, there is more similarity between Scale and Adirondack Class rocks than to Clovis and Wishstone Classes. features similar to those exhibited by some hydrous ironbearing sulfates (lack of a 535 nm band and maximum reflectance at 673 nm) [Lane et al., 2004]. [ 64] The apparent transparency feature at 800 cm 1 in the Peace tailings spectrum clearly is shifted to lower wave numbers than that of the dust spectrum (Figure 21b). The position of this feature is remarkably similar to one observed in the RAT tailings from Bounce rock at the Opportunity rover landing site in Meridiani Planum (Figure 21b). Mini-TES spectra of that rock prior to grinding, along with MB spectra of it, clearly indicate a dominant pyroxene component [Christensen et al., 2004b; Klingelhofer et al., 2004]. We infer then that the 800 cm 1 feature in the Peace rock also is due to an abundance of pyroxene, aresult supported by the MB measurements on this rock [ Morris et al., 2006]. Note, however, that the Bounce rock spectrum does not show a 1630 cm 1 feature, an indication of the absence of any hydrated phases which is consistent with its APXS-derived normative chemistry [ Rieder et al., 2004]. [ 65] The Alligator outcrop was not abraded with the RAT. Instead it received multiple brushings in an effort to remove dust from a large enough area for a Mini-TES observation (called Scale). On the basis of the spectra before and after brushing (Figure 22a) it is clear that enough dust was removed from a large enough area to improve the observation. But the target was still cold relative to the atmosphere such that there was asignificant spectral contribution from downwelling radiance. The low wave number range reveals a spectral shape clearly distinct from that of the Wishstone and Clovis Class rocks with the most similarity to Adirondack Class, yet still not adirect match to it (Figure 22b). Given the magnitude of the contribution from downwelling radiance and the current uncertainty in quantifying it, we will defer the deconvolution of this spectrum to future work Watchtower Class Rocks [ 66] A large outcropping of rock called Larry s Lookout on the crest of the Cumberland Ridge of Husband Hill (Figure 1) is the location of the rock Watchtower whose spectral and chemical characteristics define yet another distinctive rock type. Mini-TES observations of rocks at this location and some on the adjacent Jibsheet Ridge outcrop have a common spectral character that allows them to be grouped together. As with the Wishstone Class, Mini- TES spectra of these rocks were acquired before and after the sol 420 wind event that deposited dust on the pointing mirror. We will limit the analysis to only those observations made prior to sol 420, which excludes all of the Jibsheet Ridge and adjacent Methuselah outcrops that are grouped into the Watchtower Class [Squyres et al., 2006]. The full implementation of a mirror dust correction will allow a more complete investigation of these rocks Spectral Characteristics of Watchtower Class Rocks [ 67] Figure 23a shows the best set of spectra of rocks from Larry s Lookout made prior to sol 420. The features of the low wave number range (< 600 cm 1 )clearly distinguish these rocks from the dominant Husband Hill rock-type Wishstone Class (Figure 23b). Gone is the distinctive set of plagioclase-related features. Instead there is asingle, deep absorption with an apparent minimum near 450 cm 1.The absence of the plagioclase features as well as any olivine feature gives these spectra an appearance similar to but distinct from that of the Clovis Class rocks (Figure 23b). The spectrum of Watchtower provides the best example of the spectral character of these rocks when they are least encumbered by downwelling radiance, as evidenced by the relatively deep CO 2 feature and greater contrast compared with the other examples shown in Figure 23a. Because the Watchtower rock spectrum is the best example and because the rock was the subject of afull IDD campaign, it alone was used for deconvolution. Its SNR is sufficient for this purpose such that no averaging together of other spectra was necessary Mineralogy of Watchtower Class Rocks [ 68] The same set of primary igneous mineral endmembers used in the previous deconvolutions serves as the starting point for the deconvolution of the Watchtower spectrum (Table 1). A poor fit is obtained when only 26 of 36

27 Figure 23. Spectra of Watchtower Class rocks from Larry s Lookout. (a) The varying impact of downwelling radiance is most evident in the middle wave number range in this set of spectra. (b) The low wave number range displays spectral characteristics distinct from the previous rock classes but most similar to Clovis Class. primary igneous minerals are used (Figure 24a). Because of the spectral similarity between Watchtower and Clovis Class, it is reasonable to assume that Watchtower also may contain a glass component. With the inclusion of the same glasses and shocked plagioclasesasused for the Clovis deconvolution (Table 1), the fit in the low wave number range is improved (Figure 24b). Once again, the Hawaiian glassy basalt end-member HWKV340A supplies aspectral shape that fits the low wave number range. The middle wave number range ( cm 1 )remains poorly fit until the addition of sulfate end-members (Table 1), which lowers the RMS error by 50% (Figure 24c). There is no comparable improvement in fit beyond this combination of endmembers. For completeness, phosphates and oxides were added to the end-member set (Table 1), with inclusion of the former group motivated by the elevated phosphorous abundance detected by APXS [ Gellert et al., 2006]. A few percent of the phosphate wavellite emerges in the modeled results of this deconvolution run, but with no apparent improvement in fit or RMS value (Figure 24d). Finally, secondary silicates were added (Table 1) to assess their potential contribution to the Watchtower spectrum. A slight improvement was obtained that is evident in a small reduction in RMS error ( 10%) and the slightly improved fit between cm 1 (Figure 24e). However, the glass components remain as a substantial part of the modeled results with only a negligible contribution ( 3%) from secondary silicates (stilbite). [ 69] To test the robustness of the results with regard to the glasscomponents, they were removed insuccessive stagesas was done with the Clovis Class deconvolution (section 5.2), producing very similar results. With the successive removal of each basaltic glass end-member, a different one was incorporated into the modeled results but always withincreasing RMS error. When all of the basaltic glass end-members were excluded, a shocked plagioclase end-member (37.5 Gpa) emerged as the dominant component ( 55% abundance) but again producing a poorer fit. Finally, all of the glassy endmember spectra were excluded (Table 1). The poor fit of the modeled spectrum demonstrates that the crystalline secondary silicates used in this work cannot substitute for the glass components (Figure 24f). [70] Like Clovis Class rocks, Watchtower Class appears to contain a significant abundance of basaltic glass (>35%) with little to no secondary silicates. Along with the basaltic glass component, silica-rich obsidian glass occurs in the best-fit results. In all of the deconvolution runs wher e glasses were included, they represent at least 45% of the modeled components. Sulfates at 5 10% abundance are required to achieve agood fit with gypsum as the favored component. Plagioclase feldspar in the form of oligoclase is the dominant primary igneous mineral that is apersistent component in all of the runs, ranging from 10 25% abundance. The results for pyroxene and olivine components are not as clear, but some combination of these minerals totaling 20% is permissible. Table 2 includes a best estimate of the mineralogy of Watchtower Class rocks based on the results from the three deconvolution runs with the lowest RMS values. [ 71] A surface dust component is present in all of the deconvolution runs at 10% abundance, an indication that the Watchtower rock used for deconvolution was relatively dust-free. The modest emissivity roll-off at high wave numbers supports this conclusion. All of the runs required a sky component at 5% abundance to achieve a good fit. This suggests that there is a small downwelling component in the spectrum that is not accounted for in the standard downwelling correction. A slope component is present in the best-fit case at 15%, which translates to an equivalent temperature determination error of 0.5 C Basaltic Exotics [ 72] Dozens of small (<50 cm), dark-toned, smoothtextured float rocks have been observed since the rover arrived at the base of Husband Hill. We describe these rocks as exotics because they typically are found as isolated occurrences surrounded by rocks with a different lithology. Their texture, dark-tone, and in the case of one of the classes, spectral similarity to the rocks of the Gusev plains support the use of the term basaltic. Another feature of these rocks that is worth noting is their morphology, which commonly has the appearance of a ventifact, with facets 27 of 36

28 Figure 24. Spectraldeconvolution of the Watchtower spectrum using various combinations of endmember spectra. The raw results with values in %abundance are shown (see section 4.2 for caveats). The second column of values (in %) represents the modeled components normalized to exclude surface dust, slope, sky, and blackbody components. (a) Using only the primary igneous end-members shown in Table 1, apoor fit is obtained. (b) The fit is substantially improved, especially in the low wave number range, with the inclusion of the amorphous phases shown in Table 1. (c) The RMS error is reduced by 50%, and the fit in the middle wave number range is clearly improved with the introduction of the sulfate end-members shown in Table 1. (d) No improvement is achieved by adding the phosphate and Fe-oxide end-members of Table 1. (e) Asmall reduction in RMS error accompanies the incorporation of the secondary silicates shown in Table 1 but with only a negligible amount of a zeolite component incorporated into the modeled result. (f) The final iteration excluded all amorphous phases (Table 1) to test the robustness of the apparent large abundance of glass components. This resulted in more than double the RMS error of the best-fit case. 28 of 36

29 Figure 25. Adirondack Class basaltic exotics. (a) Afloat rock called Maids Milking (dashed) on Husband Hill is a good spectral match to the Adirondack Class rocks (solid) on the plains. (b) Fifteen additional examples have been found scattered across Husband Hill. The spectra show variations in the middle wave number range due mostly to mirror dust, but all have a common emissivity peak at 430 cm 1 and minimum at 510 cm 1 (vertical lines) that links them to each other and to Adirondack Class rocks. In chronologic order,they are as follows: Maids Milking sol 342 p3822, Fortress sol 417 p3883, Ham Shank sol 458 p3961, Cricket sol 466 p3974, Grasshopper sol 471 p3978, Nautilus sol 483 p3425, Flotation sol 483 p3427, Kohinoor sol 509 p3465, Pongal sol 510 p3466, Cannon sol 511 p3469, Rudder sol 512 p3467, Kerouac sol 518 p3474, Howl sol 522 p3482, Fountainsol 530 p3495, Liberte sol 553 p3067, and Lorraine sol 558 p3081. and edges that suggest aeolian abrasion. Three distinct classes were first identified with Mini-TES spectra, two of which have been confirmed using the IDD instruments. These are described in detail below. Wehave chosen to report on the Irvine Class exotics while excluding several distinctive rock types encountered prior to them in an effort to present the basaltic rocks of Husband Hill as aset Adirondack Class Exotics [ 73] With the rover parked next to Wishstone on lower Husband Hill, Mini-TES observed dozens of rocks with the same plagioclase-rich spectral character as Wishstone. But one rock stood out on the basis of its texture, dark tone, and spectral character. Dubbed Maids Milking (part of aseries of Christmas-themed rock names), it clearly displays the spectral features of Adirondack Class rocks from the plains (Figure 25a). As the rover made its way up to the summit of Husband Hill, 15 additional examples of this spectral class were observed both before and after the sol 420 wind event leading to variable contributions from mirror-dust (Figure 25b). In all cases, the prominent peak at 430 cm 1 is the diagnostic feature that relates these rocks to the plains basaltic rocks. It is this feature that was attributed most directly to oli vine in Adirondack Class rocks Backstay Class Exotics [ 74] On sol 477 Mini-TES observed a small float rock called Backstay adjacent to the Jibsheet Ridge outcrop (Figure 1) that had atexture, tone, and morphology similar to previous basaltic exotics. However, its spectrum is different than the Adirondack Class exotics (Figure 26). Although it still has a prominent peak at low wave numbers similar to that of Adirondack Class, the peak is shifted to higher wave numbers ( 460 cm 1 )and asecondary peak at 530 cm 1 is evident (see also Figure 27). Backstay also shows some similarity to Adirondack Class in the middle wave number range with an absorption minimum at 885 cm 1.However, there is additional fine structure between cm 1 in the Backstay spectrum that is absent in Adirondack Class spectra (see also Figure 27). On the basis of the apparent spectral distinction between Backstay and the Adirondack Class rocks, it was deemed worthy of a full IDD campaign, the results of which confirmed the Mini-TES discovery of a new rock class [ Squyres et al., 2006]. [ 75] At least 8 other examples of Backstay-like rocks have been observed by Mini-TES between the location of Backstay and the summit of Husband Hill, all sharing similar low wave number features but with varying dust contributions in the middle wave number range (Figure 27a). Figure 27b shows the results of applying the semiquantitative correction for mirror dust described in section 4.3, demonstrating that the correction yields results that are at least selfconsistent. The mirror-dust correction produces spectra that look quite similar to each other, including the fine structure between cm 1,despite the significant variations in their uncorrected form. [ 76] Because the accuracy of the mirror-dust correction has not been evaluated fully, we will defer the deconvolution of the Backstay spectrum to future work. McSween et al. [2006] presented a CIPW norm-derived spectrum for Backstay using the same strategy as we have done for an Adirondack Class spectrum (section 3.2.2) in an effort to evaluate the normative mineralogy of Backstay. The result was a poor fit to the measured spectrum. Given that the APXS and MB results indicate that Backstay has minimal alteration [Gellert et al., 2006; Morris et al., 2006], it is unlikely that the poor fit is due to spectral components representing alteration products. CIPW normative calculations do not include glass components, so the presence of glass in the rock could produce aspectrum that is poorly fit by igneous minerals alone. However, as indicated by McSween et al. [2006], MB spectra do not show asignificant glass component. 29 of 36

30 Figure 26. Another class of basaltic exotics was first identified by Mini-TES with the float rock called Backstay near Larry s Lookout/Jibsheet Ridge on Husband Hill. The vertical lines show the shift inthe low wave number features that among other differences, spectrally distinguishes Backstay Class (solid) from Adirondack Class rocks, represented here by the basaltic exotic called Fountain (dashed). [ 77] Some of the features in the Backstay spectrum are consistent with spectral features in the SNC meteoriteeeta as shown by Squyresetal. [2006] that could be due to pigeonite. Hamilton et al. [1997] demonstrated that the dominance of a pigeonite component in the spectrum of EETA led to apoor fit when it was deconvolvedwith an end-member set in which pigeonite was absent. Similarly, if Backstay has asignificant pigeonite component, modeling it using a set of pyroxenes without the proper pigeonite likely would yield poor results. Likewise, CIPW normative calculations do not model pigeonite, such that the modeled spectrum of McSween et al. [2006] may be flawed by its absence. Ideally,afull set of pigeonite spectral end-members will be included in deconvolution modeling of Backstay Class rocks Irvine Class Exotics [ 78] By sol 590 the rover had crested the summit of Husband Hill and driven afew 10s of meters southward (Figure 1). A collection of smooth, dark-toned rocks each no bigger than 15 cm was observed to have a vague alignment suggestive of an eroded volcanic dike 2.5 min length (Figure 28). A Mini-TES observation of a tight cluster of four of these rocks dubbed Irvine revealed a spectrum different from the basaltic rocks observed previously on Husband Hill. The rover departed the scene to conduct an imaging campaign 20m to the east but then returned to the putative dike to conduct an IDD and remote sensing campaign. Although Mini-TES observations of additional rocks along the trend of the alignment showed a common spectral character, no observations from any of the other instruments provided definitive evidence of adike. However, the measurements from APXS and MB instruments did support the Mini-TES discovery of anew class of basaltic exotic [ McSween et al., 2006]. [ 79] Figure 29a shows the uncorrected spectrum of Irvine along with four other spectra with similar low wave number spectral features. Three of these are from rocks along the trend of the alignment (Compagnoni, Unsoeld and Whillans) and one is from a rock well away from the alignment in a nearly perpendicular direction (Adze is 20 m to the north). Figure 29b shows these same spectra after correcting for mirror dust using equation (4). The corrected versions show a common set of features in the middle wave number range as well as those in the low wave number range that were used to group the rocks into asingle class. As shown in Figure 30, there is some overlap between Irvine and Backstay Class spectra, most notably in the shape and position of an emissivity minimum at 885 cm 1. Note, however, that the Christiansen feature of Irvine Class spectra is shifted slightly toward lower wave numbers compared to Backstay Class as is the emissivity peak in the low wave number range. Such differences are sufficient to distinguish the two basaltic compositions. [ 80] As with Backstay, the deconvolution of Irvine must await a more thorough evaluation of the mirror-dust correction in order to ensure the most accurate results. McSween et al. [2006] demonstrated that the CIPW norm-derived spectrum of Irvine produces apoor fittothe measured spectrum. Like Backstay, it is possible that if the pyroxene component in Irvine is dominated by pigeonite, the spectrum will not be modeled correctly. 6. Discussion [ 81] Mini-TES has served a principal role in characterizing the mineralogy of the rocks encountered by the rover, the impact of which we describe below. Perhaps less apparent is the impact of Mini-TES on the tactical operations of the rover and the exploration of the landing site. As one of 30 of 36

31 Figure 27. Nine examples of Backstay Class basaltic exotics have been found by Mini-TES scattered across Husband Hill. (a) The middle wave number range shows pronounced variations among these spectra due to mirror dust, but the similarity of the low wave number features allows the spectra to be grouped together. (b) Following correction for mirrordust, the spectral similarity applies throughout the full spectral range. three remote sensing instruments including Pancam and Navcam, Mini-TES observations have helped to identify targets of interest and eliminate the need to drive to targets that did not warrant further scrutiny by the IDD instruments. Given the effort of approaching targets for IDD work and the multiple sols required for such work [Arvidson et al., 2006], the ability to retire this effort with a degree of confidence has been one of the valuable contributions from Mini-TES. The few 10s of basaltic exotics discovered with Mini-TES spectra demonstrate the obvious utility of spectral remote sensing. The much larger populations of rocks that have been identified with Mini-TES spectra further extend this utility by putting into context the rocks measured by the IDD instruments. For example, only three members of the Wishstone Class were measured by the IDD instruments in just one location, yet 95 Mini-TES spectra demonstrate that they are the most ubiquitous rock type on the north side of Husband Hill. Much work remains to be done in refining the details contained in the Mini-TES observations, but as discussed below, sufficient information has emerged from them to help us better interpret the results from other data sets Primary Mineralogy [ 82] Mini-TES spectra have revealed a range of igneous minerals and glasses that represent a set of primary components forming the bulk of the rocks on the plains and hills visited by the Spirit rover. The unambiguous identification of olivine and plagioclase provides ground truth for orbital TES observations, fulfilling one of the goals of the Mini- TES experiment. Olivine of intermediate Fe content has been identified elsewhere on the planet in TES spectra [Hamilton et al., 2003; Hoefen et al., 2003; Christensen et al., 2005; Hamilton and Christensen, 2005] and the spectral features used in those identifications have been corroborated at the Gusev crater site. Plagioclase feldspar has been detected across the globe through deconvolution of TES spectra [Bandfield, 2002] but now we have clearly identified features in Mini-TES spectra of Wishstone Class rocks that support detections from orbital data. We have not observed any evidence for quartz, nor have we found any rocks with the spectra of the TES type 2 putative basaltic andesite [ Bandfield et al., 2000]. [ 83] The apparent identification of a basaltic- or lowsilica glass component in substantial abundance in some of the rocks of the Columbia Hills has not been reported previously for other locations across the planet. Although other materials of secondary or authigenic origin might supply similar spectral characteristics, none have been identified using the spectral libraries available to us. As with the identification of high-silica glass in the TES type 2 spectrum, the basaltic glass component observed in Mini- TES spectra will be the subject of future inquiry and debate, some of which is included below Secondary Mineralogy [ 84] Mini-TES observations of the seven Gusev Crater rock classes reported here reveal no clear signature of wellcrystalline secondary silicates. This result is consistent with data from the other rover instruments for Adirondack, Backstay, Irvine, and Wishstone Class. The first three appear to be igneous rocks with little evidence for aqueous alteration [ Ming et al., 2006; Squyres et al., 2006; McSween et al., 2006]. The clastic rocks of Wishstone Class show evidence for only modest alteration [ Ming et al., 2006; Squyres et al., 2006]. However, Clovis and Watchtower Classes, which also appear to be clastic rocks, are significantly altered on the basis of APXS and MB results [ Gellert et al., 2006; Ming et al., 2006; Morris et al., 2006]. They are enriched in S, Cl, and Br [ Gellert et al., 2006], are highly oxidized, and contain the mineral goethite [Morris et al., 2006]. These and other chemical indicators along with the presence of only minor amounts of primary igneous minerals suggest alteration of the rocks by aqueous fluids that should have produced abundant secondary aluminosilicates [Ming et al., 2006]. Yet Mini-TES spectra show that these rocks are dominated by abasaltic glass component with only minor phyllosilicate or other secondary silicate 31 of 36

32 Figure 28. Avague alignment of rocks seen in this Navcam mosaic from sol 599 contains the cluster of rocks called Irvine (inset Pancam approximate true color image) that represents athird class of basaltic exotics. Mini-TES observations of the other named rocks along the alignment yielded similar spectra as shown in Figure 29. components possible. It is reasonable to expect that clastic rocks that may have formed from impact or volcanic processes [Squyres et al., 2006] could be rich in basaltic glass. But the presence of this glass in apparently altered rocks presents somewhat of a paradox. [ 85] Ming et al. [2006] modeled the mineralogy of Clovis and Watchtower rocks using chemical data from the APXS supplemented by data from the MB and Mini-TES in an effort to constrain the possible range of alteration. Two cases were modeled using a combination of CIPW normative analyses, apparent elemental associations, and mass balance calculations: case 1 investigated the potential for pervasive alteration and assumed that secondary phases are present; case 2 was for minimal alteration with the assumption that primary phases are present. Case 1revealed >60% secondary aluminosilicates in Clovis and Watchtower. Recast as primary phases in case 2, Clovis would have 33% pyroxene and 34% feldspar and Watchtower would have 22% and 30% of these phases, respectively. Although Ming et al. [2006] favored case 1 for these rocks (i.e., pervasive alteration), Mini-TES results are more in line with the case 2 results if basaltic glass is included among the primary phases. [ 86] Itisreasonable to assume that basaltic glass would have the bulk composition of feldspar +pyroxene(±olivine) phases. But the description of a primary basaltic glass in the context of rocks that clearly show some degree of alteration needs to be considered, especially with regard to their very oxidized character. The average ferric to total iron ratio of Clovis and Watchtower is 0.84 and 0.83, respectively [ Ming et al., 2006]. Such highly oxidized rocks would seem to preclude the idea of aprimary basaltic glass component. Clearly most of the Fe 2+ within the putative glass would have to be altered to Fe 3+.Yet spectrally, inthe TIR wavelengths, this oxidation may not be apparent. With the exception of work by Minitti et al. [2002], little has been done to document the TIR spectral changes with oxidation of basaltic glass. The work of Minitti et al. [2002] demonstrates that dry oxidation of synthetic basalt glass samples produces spectral changes compared to the unoxidized precursors. But these changes appear to result mostly from a surface coating of nanophase hematite that is easily removed, leaving behind a darkened, oxidized glass. The spectral character in the absence of the coating was not measured. [ 87] Spectra of two basaltic glass samples collected by R. V. Morris and measured at ASU according to the procedures of Ruff et al. [1997] provide some clues about the possible spectral variation with oxidation (Figure 31). Sample HWKV201 is beach sand (<1 mm particles) produced from flows of the Kilauea volcano, Hawaii that contains mostly Fe 2+.PN-64 is aparticulate sample (<1 mm particles) of a thermally oxidized, visibly red basaltic tephra from the interior of the Pu u Nene cinder cone on Kilauea volcano and contains mostly Fe 3+ (R. V. Morris, personal communication, 2006). The spectral similarities between these two samples are greater than their differences. Both have a broad, U-shaped emissivity minimum centered at 1000 cm 1 and a narrower one at 450 cm 1 (Figure 31). Their differences are most evident at high wave numbers and in the region between cm 1.Although the mineralogy of these two samples has not been determined, it may be that ahydrous component in the oxidized sample (PN-64) is responsible for the spectral differences. Evidence for this comes from the weak bound water feature at 1630 cm 1 in the spectrum of PN-64. Regardless of the differences, it is clear from this simple comparison that fresh and oxidized basaltic glass can appear very similar in the TIR wavelengths raising the possibility that the apparent identification of primary basaltic glass in Mini-TES spectra may in fact represent oxidized, i.e., altered, basaltic glass. [ 88] Palagonitization is a process that can yield oxidized basaltic glass and typically is accompanied by the formation 32 of 36

33 provides an approximate match to the low wave number region of the Watchtower spectrum (Figure 32). However, the spectra of the terrestrial rocks are much more V-shaped in the middle wave number range compared with the U-shape of the Martian rock spectra. More work is required to assess the significance of the similarities and differences between these spectra, but palagonitization should be considered a candidate process to explain in part the alteration of Clovis and Watchtower Class rocks. [ 89] If Clovis and Watchtower Class rocks were formed from material produced by impacts into basaltic materials, as favored by Squyres et al. [2006] and Ming et al. [2006], then we suggest that oxidized basaltic glass could be the major constituent of these rocks. In this view, their elevated oxidation state and lack of abundant primary igneous minerals is not due to pervasive aqueous alteration. Instead, these characteristics represent the initial condition of the clasts that make up the rocks. Perhaps we are seeing an accumulation of impact material that is dominated by basaltic glass that underwent only modest aqueous alteration leading to enrichments in S, Cl, and Br and the production of small amounts of goethite but not asignificant amount of secondary silicates. Palagonitization may have played a part in the alteration process. Figure 29. Irvine Class basaltic exotics. (a) Rocks that are on an alignment with Irvine as shown in Figure 28, as well as one other that is not, display similar low wave number characteristics. (b) Upon correction for mirror dust, all five spectra show similarities throughout the full spectral range. of amorphous or crystalline hydrated products [e.g., Hay and Iijima, 1968; Bell et al., 1993; Morris et al., 1993]. Various workers have measured the TIR spectra of palagonitic materials but typically in the form of particulate ash or soil samples [e.g., Crisp and Bartholomew, 1992; Roush and Bell, 1995; Graff et al., 2001] that show the concomitant particle-size effects. If Clovis and Watchtower Class rocks are palagonitized, their spectra need to be compared to those of palagonitized rocks or equivalent coarse-particulate material. Two such examples are shown in Figure 32 compared to the average Clovis Class and Watchtower spectra following removal and normalization for contributions from surface dust, slope, and sky end-members as determined by the deconvolutions shown in Figures 17e and 24e, respectively. The spectrum of a hand sample of moderately palagonitized hyaloclastite tephra from the Laki, Iceland eruption collected by J. R. Michalski compares favorably in the low wave number region to the Clovis Class spectrum (Figure 32). Similarly, the spectrum of a hand sample of moderately palagonitized tuff from a tuff ring near Fort Rock, Oregon collected by W. HFarrand 7. Conclusions [ 90] The Mini-TES instrument on the Spirit rover has successfully fulfilled its intended role of providing remotely sensed mineralogical information for use in directing the rover to rocks of interest and extending the reach of the rover s IDD instruments. From its observations during the first 600 sols of the mission, the following conclusions can be drawn: [ 91] 1.No phyllosilicate-rich rocks have been observed by Mini-TES as might be expected if Gusev crater once held alake. Although this does not rule out its existence, we have observed no mineralogical indicators that unambiguously support its former presence. [ 92] 2.The rocks on the plains between the lander and the Columbia Hills are dominated by a single population of olivine-rich basalts called Adirondack Class. Given the hundreds of rocks on the plains observed by Mini-TES to have the same spectral character, it is unlikely that another population of rocks is present but overlooked. [ 93] 3.The West Spur of Husband Hill is dominated by a single population of rocks known as Clovis Class with a very different mineralogy than the plains rocks. Primary igneous minerals are only minor components of these rocks. An amorphous phase best modeled by quenched basalt is the dominant component. This amorphous phase may represent oxidized or palagonitized basaltic glass. Wellcrystalline secondary silicate phases are not uniquely identified although minor sulfate and goethite components are present. [ 94] 4.The dominant rock type on the northern flank of Husband Hill is aplagioclase-rich rock called Wishstone Class observed only as float. These rocks contain up to 60% plagioclase of intermediate composition. Phosphate and sulfate components at or near the detection limit improve the modeled fit of the average Wishstone Class spectrum 33 of 36

34 Figure 30. Irvine Class rocks (red), although spectrally similar to Backstay Class (black), are sufficiently different to merit their own spectral classification. but are not uniquely identified. No secondary silicates are evident. [ 95] 5.The outcrop known as Peace on Husband Hill displays clear evidence for one or more hydrous phases along with a dominant pyroxene component. [ 96] 6.Another outcrop on Husband Hill called Larry s Lookout is the type location for Watchtower Class rocks that resemble Clovis Class rocks in their abundance of amorphous phases, which may represent oxidized or palagonitized basaltic glass. No crystalline secondary silicates are uniquely identified. [ 97] 7. Although Wishstone Class rocks dominate the lithology on the north flank of Husband Hill, float rocks with three distinctive basaltic compositions are found scattered across the hill. One type clearly resembles Adirondack Class rocks from the plains. The second type, called Backstay Class, shows no obvious source region. The third type, called Irvine Class, shows some evidence for alocal Figure 31. Laboratory spectra of fresh and oxidized basaltic glass samples from Kilauea volcano, Hawaii, provided by R. V. Morris. Their spectral similarity suggests the possibility for fresh (dashed) and oxidized (solid) basaltic glass to substitute for one another in spectral deconvolution of material that contains such components. 34 of 36