We studied a dark, intracrater feature in Amazonis Planitia using visible, thermal, and spectral data from the Mars Global Surveyor (MGS) Thermal Emission Spectrometer (TES), MGS Mars Orbiter Camera (MOC), and 2001 Mars Odyssey Thermal Emission Imaging System (ODY THEMIS) instruments. Visible and thermal data indicate that there are heterogeneities within the dark feature at meter to kilometer scales and suggest that it represents eroded terrain containing inactive ripple-like bed forms (possibly armored or indurated) as well as loose, possibly actively saltating, sediment. In addition, part of the dark intracrater material may be covered with dust, either as an optically thin (few microns thick) layer or as thicker discontinuous patches. The thermal properties of the dark material are consistent with a mixture of sand, rock, and bedrock, whereas the rest of the crater floor has thermal characteristics of fine (∼35 μm) particulates. Spectral data indicate that the crater floor is covered by a layer of dust that is spectrally and compositionally similar to the globally homogenized surface dust in other high-albedo regions on Mars. The mineralogy of the dark materials is mafic (dominated modally by pyroxene and olivine); however, unlike other lithologies common in low-albedo regions (i.e., surface types I and II), its derived bulk chemistry indicates that it is an ultramafic lithology with the lowest silica content detected on Mars to date. Sheet silicates and glasses are not identified above detection limits, and as such, the materials do not display evidence for significant chemical weathering. The geomorphology and visible distribution of the dark, intracrater materials combined with the lack of an obvious source in the region outside the crater suggest they are the product of erosion of a local source.
 First imaged by the Mariner 6 and 7 spacecraft, low-albedo features on the floors of Martian craters have been acknowledged as a widespread and significant surface feature [Arvidson, 1974; Sagan et al., 1973]. Prior work has shown that these intracrater features are found globally in both low- and high-albedo regions, are characterized by a lower albedo and higher thermal inertia than their surroundings, and commonly are interpreted to be accumulations of sand-sized particles [Aben, 2003; Christensen, 1983; Edgett and Christensen, 1994; Fenton et al., 2003]. Studies of these intracrater materials therefore can yield important information about a variety of geological processes. Dune forms and other aeolian features can be used to determine local and regional wind patterns. The composition, grain size, and volume of the sand comprising the intracrater materials can yield information on the source and movement of the sand, local and/or regional geology, as well as rates of weathering. Examination of dark materials isolated within high-albedo (i.e., dust-covered) terrains is especially important as it can provide a rare opportunity to determine the composition of an area that is otherwise covered by dust.
 During a survey of intracrater features in the Amazonis Planitia region, we identified one feature, located on the floor of an unnamed, 60 km-diameter crater at 17.92°N, 190.26°E, that exhibited unusual day to nighttime variations in the size of the feature in THEMIS brightness temperature images. Figure 1 is an MGS MOC wide-angle image of the crater and its associated intracrater dark feature (note that in this paper we describe features observed in visible wavelength data in terms of albedo, which has been measured and confirmed by TES observations; also, the studies cited above have documented the low albedo of similar features relative to their surroundings). The dark feature lies near the western edge of the crater floor, and is roughly 8 km in diameter. Extending to the northeast of the darkest materials is an area intermediate in albedo between the darkest area and the brighter crater floor materials. The main purpose of this study was to determine the geologic setting and composition of the dark-toned feature through analysis of its visible, thermophysical, and spectral properties.
2.1. Instrument Description and Data Selection
 The Thermal Emission Imaging System (THEMIS) has visible (VIS) and thermal infrared (IR) subsystems. The IR subsystem is a multispectral imager that provides mineralogical and atmospheric information from ten-band images (nine wavelengths between ∼1475–667 cm−1; 6.78–15 μm) [Christensen et al., 2003]. It collects swaths ∼32 km wide at a spatial resolution of ∼100 m/pixel. The visible/near infrared imaging subsystem acquires images with ∼19 m/pixel spatial resolution at five wavelengths between 23,474–11,628 cm−1 (0.426–0.860 μm) [Christensen et al., 2003]. The calibration approach for the VIS and IR subsystems is described by Christensen et al.  and calibration specifics are provided by Christensen .
 We selected THEMIS IR and VIS images for the study area on the basis of location, and the IR image with the highest maximum brightness temperature was used for spectral analysis to maximize the signal-to-noise ratio (SNR). We applied a constant radiance offset correction, which removes the effects of atmospheric emission and systematic calibration radiance offsets [Bandfield et al., 2004b], to the daytime IR image. To completely remove the effects of atmospheric components from the IR spectral data, we applied the correction of Bandfield et al. [2004b]. Briefly, this approach consists of (1) obtaining a TES surface spectrum of a spectrally homogeneous “training” area in the vicinity of the feature of interest in the THEMIS image (and at approximately the same elevation; in this case, an area in the northeast portion of the crater floor), (2) convolving the TES surface spectrum to THEMIS spectral resolution, (3) assuming that the training area in the THEMIS image has a surface signature matching that of the spectrally convolved TES surface-only spectrum and the remaining contribution to the THEMIS spectral signature of the training region represents the atmospheric spectral signature, and (4) subtracting the atmospheric spectral signature derived from the training area from the entire THEMIS image.
 The Thermal Emission Spectrometer (TES) is a hyperspectral, interferometric spectrometer with selectable ∼5 or ∼10 cm−1 sampling between ∼1650–200 cm−1 (∼6–50 μm) and a spatial resolution of about 3 × 6 km [Christensen et al., 1992, 2001]. Additionally, broadband visible (∼33,333–3448 cm−1; 0.3–2.9 μm) and thermal (∼1960–67 cm−1; 5.1–150 μm) bolometers are used to derive albedo and thermophysical properties of the Martian surface [Christensen et al., 2001]. The calibration of the spectrometer and bolometers, as well as quality fields assigned to the data that aid in the selection of high-quality data for analysis, are described by Christensen et al. .
 We selected TES spectra over the intracrater feature on the basis of location and data quality. To restrict our analysis to the highest quality data available, we selected spectra with high target temperatures (>270 K), low atmospheric dust opacity (<0.1), no solar panel movement, image motion compensation (IMC) off, no major phase inversions, low risk of phase inversions, and little to no high gain antenna motion. Spectra collected during the global dust storm of 2000 (Ls = 190°–270°) [Smith, 2004] were avoided, as were spectra collected during orbits with unacceptable amounts of microphonic noise [Bandfield, 2002; Hamilton et al., 2003].
 Data from two TES OCKs (Orbit Counter Keeper; this is the orbit number used by the TES team, which differs from the MGS Project orbit number by +1683), 1908 (Ls = 116°) and 2587 (Ls = 142°), met the above criteria, and were associated with albedos that were significantly less (∼0.15) than the average albedo in the area (∼0.27). Because the intracrater feature is distinguished primarily by its relatively lower albedo (Figure 1), TES spectra for the feature were chosen on the basis of location and albedo (<0.18). Due to the small size of the feature, only three spectra from each OCK met these criteria (OCK 1908: Incremental Counter Keeper (ICK) 1843, detectors 4 and 5 and ICK 1844, detector 1; OCK 2587: ICK 1921, detector 5 and ICK 1922, detectors 1 and 2). Because of significant (∼20 K) temperature differences between the dark feature and the surrounding terrain, there is additional minor noise in these data that commonly is referred to as “ringing”. Ringing is an artifact introduced into the data because the TES electronics do not have time to recenter the base level of the interferogram between observations acquired under conditions of rapid target temperature change [Christensen, 2001]. The result is a high-frequency (point-to-point) variation that is superimposed on the data, and is most apparent where the signal is low in the raw uncalibrated data (at the limits, or ends, of the spectral range). To minimize the effects of all noise in the data, these six spectra (three from each OCK) were averaged together for analysis. We also selected six TES spectra from the same OCKs and detectors over the brighter crater floor a few kilometers north (OCK 1908: ICK 1845, detectors 4 and 5 and ICK 1846, detector 1; OCK 2587: ICK 1923, detector 5 and ICK 1924, detectors 1 and 2). These spectra have high albedos (>0.23), and are at approximately the same elevation as the low-albedo material. The six bright floor spectra from these OCKs also were averaged to improve SNR. Thus, in total we analyzed two spectra, one average spectrum for the dark material and another for the nearby bright crater floor material.
2.2. Linear Deconvolution
 The thermal infrared is useful for determining the abundances of phases in a geologic material because in this wavelength region the spectra of individual component phases combine linearly in proportion to their areal abundance to produce a composite spectrum [Lyon, 1964; Thomson and Salisbury, 1993]. Therefore a linear deconvolution approach [Ramsey and Christensen, 1998] can be applied to our average TES spectra in order to determine the modal mineralogies of the two surfaces. Specifically, a linear deconvolution algorithm utilizes a set of end-member spectra (e.g., a subset of a library of spectra) to perform a least squares fit to an unknown spectrum. The output of the algorithm includes a modeled spectrum, a list of the end-members used in the best fit, the fraction of each end-member used, and a root mean square (RMS) error.
 Initial deconvolution runs included a wide array of mineral phases from the ASU spectral library [Christensen et al., 2000b] and previous studies [Christensen et al., 2000a; Hamilton, 2000; Hamilton and Schneider, 2005; Wyatt et al., 2001] in the end-member set. However, several minerals (quartz, garnet, micas, phosphates, sulfates, amphiboles, as well as some carbonates and phyllosilicates) were not used in modeling either of the spectra and so were removed from the end-member list in order to limit the total number of input end-members thereby improving the accuracy of the deconvolution algorithm [Ramsey and Christensen, 1998]. The final deconvolution set contained 32 mineral end-members, listed in Table 1, which included feldspars, pyroxenes, olivines, phyllosilicates, oxides, carbonates, and glass [Christensen et al., 2000b]. The deconvolution algorithm was limited to the wave number range 1290–825 cm−1 (∼7.8–12.1 μm) and 508–250 cm−1 (∼19.7–40 μm), which includes all the major silicate features as well as two major carbonate features (the gap between 825–508 cm−1 (∼12–20 μm) is the region of atmospheric CO2 absorption). Compared to other mineral groups (especially silicates), carbonates have a relatively featureless spectrum across the deconvolved range with only two major features, which may increase their likelihood of being used as a graybody spectrum to compensate for variations in spectral contrast (e.g., as a result of particle size differences). To minimize this effect, and to provide a general means of compensating for differences in spectral contrast, a blackbody spectrum was included as an end-member in the deconvolution [Hamilton et al., 1997].
Table 1. Mineral End-Members Used for Deconvolution of TES Emissivity Spectra
Maskelynite (labradorite) ASU-7951
Albite WAR-0235 174
Labradorite BUR-3080A 176
Bytownite WAR-1384 177
Anorthite BUR-340 178
Anorthite WAR-5759 221
Labradorite WAR-RGAND01 222
Pigeonite Wo10En36Fs54 33,34
Augite NMNH-9780 157
Enstatite (average of HS-9.4B 30 and NMNH-34669 160)
Hypersthene DSM-FER01 152
Bronzite NMNH-93527 168
Olivine Fo68 KI 3115
Olivine Fo60 KI 3362
Olivine Fo35 KI 3373
Olivine Fo52.5 KI 3372
Forsterite AZ-01 38
Fayalite WAR-RGFAY01 167
Calcite ML-C10 99
Dolomite C28 128
Fe66Mg34CO3 C56 134
Kaolinite KGa-1b solid 186
Nontronite WAR-5108 solid 204
Fe-smectite Swa-1 solid 207
Illite IMt-2 granular 211
 Atmospheric components in the TES spectra are modeled (and subsequently removed) by including six atmospheric end-members representing H2O ice clouds, atmospheric dust, pure CO2 gas, and pure H2O water vapor [Bandfield, 2002; Smith et al., 2000] in the deconvolution end-member set along with geologic phases. Work by Smith et al.  has shown that modeling TES spectra with these typical Martian atmospheric spectral shapes allows for separation of the surface-only spectral shape, which is modeled by the mineral spectral end-members as well as the average high-albedo surface (dust) shape described by Bandfield and Smith  (note that although we consistently refer to this spectrum as “high-albedo surface dust” this surface dust shape is actually representative of both high- and moderate- (>0.20) albedo surfaces on Mars).
 The quality of the fit of the modeled spectrum to the TES spectrum is assessed through visual inspection, which consists of plotting the original and modeled spectra together and producing a residual error, or difference, spectrum to verify that the modeled spectrum closely matches the spectral shape of the original spectrum across all wavelengths. Although RMS errors are useful for comparing multiple fits of the same spectrum, or for comparing the model fits of spectra with comparable noise and spectral contrast levels, RMS values can be affected by factors other than the goodness of fit and absolute accuracy (e.g., spectral contrast). We do not report RMS errors associated with the model fits in this work because we observed that the two surfaces in the crater exhibit differing spectral contrasts. On the basis of previous studies using TES data and this deconvolution algorithm, the general detection limit for a mineral or rock component on the Martian surface is ∼10–15% [Bandfield, 2002; Bandfield et al., 2000; Christensen et al., 2000c]. Therefore modeled mineral abundances less than this should be interpreted with caution. However, it has been demonstrated that some phases, notably olivine, may have lower detection limits due to strong, narrow spectral features that are readily distinguishable from other mineralogies [e.g., Hamilton et al., 2003].
3.1. Thermophysical Properties
 We first examined the thermophysical properties of the dark intracrater feature with THEMIS data. Figure 2 includes daytime and nighttime THEMIS IR images of the crater floor, which show that the dark feature is thermally distinct and warmer than the crater floor both during the day and at night. The size of the warmer area is different between the day and nighttime images, however, and it appears markedly larger in the nighttime image. When comparing these images to Figure 1, it is apparent that a large part of the area intermediate in albedo between the darkest area of the feature and the brighter crater floor is thermally indistinct from the crater floor during the day, but warmer at night. The three colored boxes in Figure 4 correspond to the three units distinguished by their relative band 9 (794 cm−1; 12.6 μm) brightness temperatures: warm materials, intermediate-temperature materials, and cold crater floor materials. The unique size and shape of each box results from our effort to cover as large of an area in each unit as possible to provide a significant number of pixels; the red, green, and cyan boxes contain 480, 660, and 2068 pixels respectively. The average band 9 daytime and nighttime brightness temperatures within these boxes were calculated and are plotted in Figure 3 against the minimum TES-derived albedos for each unit. This graph shows that the warm materials have a significantly lower minimum albedo than the intermediate-temperature materials and cold crater floor materials, and are always warmer than both of these areas. This figure also confirms that the intermediate-temperature materials, when compared to the cold crater floor materials, have a slightly lower minimum albedo, and are very close in average temperature during the day but significantly warmer at night.
 Next, we obtained the TES-derived thermal inertias [Mellon et al., 2000] for each of the three units from TES footprints that fell near the center of the box defining the unit (Figure 2). The warm materials have the highest thermal inertia, ranging in value from ∼370–410 (all thermal inertia values are in SI units of J/m2 K/sec). Assuming that these thermal inertia values represent a homogenous material with a uniform particle size, and using the thermal inertia relationship of Presley and Christensen , these values correspond to equivalent particle diameters of ∼1.9–2.9 mm. Thermal inertia values for the intermediate-temperature materials fall in a narrow range around 370, and are consistent with an equivalent particle diameter of ∼2.0 mm. Finally, the cold crater floor unit exhibits thermal inertia values consistently around 150, which corresponds to an equivalent particle diameter of about 35 μm. The validity of the assumptions involved in calculating equivalent particle diameters for the materials in this study is addressed below.
3.2. Geomorphic Characteristics
 High-resolution visible images of the dark, intracrater feature reveal a complex structure. In THEMIS visible images (Figure 4), the feature appears to be composed of two albedo sub-units: a very low albedo, discontinuous layer of what appears to be sediment, and an intermediate-albedo region that is adjacent to the darker unit. In addition, the surface in this intermediate-albedo area appears eroded on the northeast corner of the intracrater feature, displaying ridges, mesas, and craters. The white dash-dot line in Figure 4 shows the approximate boundary between the two thermal units, warm and intermediate-temperature materials, which comprise the dark, intracrater feature. Although this thermal boundary is not associated with an obvious visible boundary, it seems that the lowest albedo material in the area is only associated with the warm materials unit, and is not present in the intermediate-temperature materials unit. Figure 5, portions of higher-resolution (∼1.5–3 m/pixel) MOC narrow angle images, show that the warm materials unit contains abundant dark sediment and brighter ripple-like forms. The darker sediment appears to be superposing these brighter bed forms and accumulating in topographic lows and around barriers such as the mesa. These sediment accumulations are not visible in the intermediate-temperature materials in the highest resolution MOC images available (Figures 5d and 5e); however, the intermediate-temperature unit does contain a few bright rippled bed forms.
 The greater accumulations of darker sediment on the southwest sides of the mesas in Figure 5a suggest a general wind direction from the northeast. The general orientation of darker streaks over the brighter rippled bed forms around the mesas supports this, as does the overall trend of the darkest sediment surfaces being located in the southwest portion of the area, highlighted in the THEMIS false color image in Figure 6. The THEMIS false color image also emphasizes that the darker sediment appears to emanate from two point-like sources located on northeastern edge of the dark sediment surfaces. Also emphasized in Figure 6 are dark-toned slope streaks on the crater wall (visible to a lesser degree in Figure 4). Dark-toned slope streaks are common in high-albedo (i.e., dusty) regions on Mars, and result from dust avalanches caused by oversteepening of dust particle air fall deposits [Sullivan et al., 2001].
3.3. Spectral Characteristics
 To evaluate the composition of the dark, intracrater feature and the rest of the crater floor materials, we examined the spectral variations across the area using THEMIS data. We produced a series of decorrelation stretched (DCS) images [Gillespie et al., 1986], an example of which is shown in Figure 7. A decorrelation stretch of spectral radiance data highlights spectral differences among materials in a scene by suppressing correlated variations due to topography and temperature. All of the band combinations that we examined produced similar results: the dark feature appeared as a distinct, homogeneous spectral anomaly relative to the surrounding region (Figure 7). Figure 8 shows the average THEMIS emissivity spectra within the boxes for the warm materials unit (green box) and the cold crater floor materials (cyan box). The shaded region of Figure 8 represents the wave number region containing THEMIS bands 1 and 2 (both bands are located at ∼1475 cm−1, or 6.78 μm). The cold crater floor materials exhibit lower emissivities than the warm materials in this wavelength region, and this is indicative of scattering by fine (<∼65 μm) particulates [e.g., Aronson and Emslie, 1973; Bandfield and Smith, 2003; Hunt and Logan, 1972; Hunt and Vincent, 1968; Ruff and Christensen, 2002; Salisbury and Eastes, 1985]. The higher emissivities associated with the warm materials indicate that scattering by fine particulates is not occurring and that the material has an effective particle size of >∼65 μm (note that emissivity values of coarse particulates exhibit values greater than unity because band 3 (∼1280 cm−1; 7.8 μm), which is used for the temperature determination in the THEMIS atmospheric correction algorithm, contains slightly more atmospheric dust absorption than bands 1 and 2 [Bandfield et al., 2004a, 2004b]). The unshaded portion of the graph shows the wave number region (∼1280–800 cm−1, or 7.8–12.5 μm) where the primary spectral features of both surfaces occur, and highlights spectral shape dissimilarities between the warm materials spectrum and the cold crater floor spectrum. The emissivity minimum in the warm materials spectrum is at a higher wave number (∼900 cm−1; 11 μm) than in the cold crater floor spectrum (∼850 cm−1; 11.8 μm), and it also contains a downward kink near ∼1070 cm−1 (9.3 μm) that is not present in the crater floor spectrum. Compared to the warm materials, the cold crater floor materials have increased emissivity overall in this wave number region, further supporting their fine-grained nature as silicates typically show increased emissivity in this wave number range as particle size decreases [e.g., Aronson and Emslie, 1973; Hunt and Logan, 1972; Hunt and Vincent, 1968; Ruff and Christensen, 2002; Salisbury and Eastes, 1985]. For coarser-grained materials, Salisbury et al.  demonstrated that low emissivity at lower wave numbers (<∼1000 cm−1; >∼10 μm) within the major silicate absorption range (∼1250–833 cm−1; ∼8–12 μm) generally is indicative of a lithology dominated by mafic minerals. Therefore the 900 cm−1 (11 μm) minimum in the THEMIS warm materials spectrum suggests that these materials are mafic-rich. We also extracted an average THEMIS emissivity spectrum for the intermediate-temperature materials unit (red box in Figure 2) and verified that this spectrum was a close approximation of the spectral average of the warm materials unit and the cold crater floor.
 Although THEMIS provides a way to examine the TIR spectral properties of the Martian surface at high spatial resolution, TES provides hyperspectral data that can be used to extract more detailed compositional information. TES spectra were selected on the basis of co-location with the albedo feature, and given that the warm materials and intermediate-temperature materials units are too small individually to fill the field of view of a single TES detector, spectra over their combined area were selected (Figure 7b). Ruff and Christensen  described the use of spectral ratioing for performing a first-order atmospheric removal that permits observation of the spectral differences between two surfaces. Because there are only a small number of TES pixels over the dark intracrater feature, we used spectral ratioing to verify that the TES spectral character of the low-albedo feature is different than the rest of the crater floor. We produced an average TES spectral ratio by dividing the average spectrum of the low-albedo material by the average spectrum of the nearby bright crater floor for each OCK and then averaging these two ratios to reduce the effects of noise. Figure 9 shows that there is a spectral difference between the dark materials and the bright crater floor, and it is also interesting to note that the Amazonis Planitia ratio has a distinctly different spectral character than the Arabia Terra intracrater feature ratio shown by Ruff and Christensen .
 Next, we deconvolved the average TES spectra of the crater floor and low-albedo feature with the suite of geologic and atmospheric end-members described in section 2.2. The average crater floor spectrum was modeled well using only the surface dust shape of Bandfield and Smith , thereby confirming that the bright crater floor is covered in a globally homogeneous dust that is spectrally, and thus compositionally, similar to other high-albedo regions on Mars. The measured and modeled surface spectra of the low-albedo feature are shown in the upper half of Figure 10, and the mineral end-member contributions, normalized to exclude surface dust, are listed in Table 2. We utilized the method of Hamilton and Christensen  to calculate weighted average solid solution compositions of pyroxene, feldspar, and olivine. The composition of the low-albedo material is dominated by pyroxene (primarily clinopyroxene with a composition of Wo46En46Fs8) with lesser amounts of olivine (Fo50) and plagioclase (An83). Carbonates, oxides, and sheet silicates were all modeled at 10%, below the generalized detection limit. We convolved the TES surface spectrum to THEMIS resolution for comparison with the THEMIS warm materials spectrum; this is shown in Figure 8. The TES spectrum for the dark materials shows a strong similarity to the THEMIS spectrum, sharing both its emissivity minimum at ∼900 cm−1 (11 μm) and the downward kink at ∼1070 cm−1 (9.3 μm). The slightly lower spectral contrast of the TES spectrum compared to the THEMIS spectrum is caused by inclusion of high-albedo surface dust (modeled at 55%) in the TES surface spectrum of the low-albedo materials. Because the surface dust spectrum is relatively featureless across the wavelength region used in our deconvolutions, it approximates a graybody spectrum, reducing the overall contrast of the TES spectrum. The surface dust modeled in the TES spectrum of the low-albedo materials could represent spatial mixing at sub-pixel scales with this crater floor dust. Because image motion compensation is not used, TES pixels are elongated in the along-track direction (the TES footprint locations in Figure 7 are approximate and do not show this along-track elongation), and one or more of the TES footprints included in our average spectrum could intersect part of the dusty crater floor. In fact, the values of the Dust Cover Index (DCI, defined and described by Ruff and Christensen ) for the low-albedo material TES pixels fall as low as 0.960. DCI values above 0.962 generally indicate dust-free surfaces, whereas values below 0.940 designate dust-covered surfaces. Values that fall in between these cut-offs are interpreted to represent surfaces partially covered by dust [Ruff and Christensen, 2002].
 Because the ratio spectrum of Ruff and Christensen  differs from our ratio spectrum, implying a different composition, we also deconvolved the low-albedo spectrum for the intracrater feature in Arabia Terra to demonstrate the mineralogical differences that are associated with different dark/bright ratios. (This analysis is not intended to imply or evaluate any geologic relationship between materials in Amazonis Planitia and Arabia Terra.) The deconvolved spectrum was an average of the same eight TES spectra used by Ruff and Christensen  (OCK 3461, ICK 1851, detectors 1–3, 5,6 and ICK 1852, detectors 2, 3, and 6). The measured and modeled surface spectra for the Arabia Terra materials are shown in the lower half of Figure 10, and Table 2 lists the modeled phase abundances. As expected, the composition of the Arabia Terra intracrater feature is different from that of the Amazonis Planitia intracrater feature, containing more plagioclase and glass, and less olivine and pyroxene. This modeled composition generally supports the basaltic interpretation of the Arabia Terra ratio spectrum proposed by Ruff and Christensen .
4.1. Thermophysical Properties of the Low-Albedo Intracrater Feature
 Visible and thermal data indicate that there are significant differences not only between the low-albedo intracrater feature and the brighter crater floor, but also between the two thermal sub-units (warm and intermediate-temperature materials) that comprise the low-albedo feature itself (Figures 2–5). TES thermal inertias suggest that both the warm and intermediate-temperature units are composed of relatively larger effective particle size materials than the cooler crater floor materials, which are effectively dust-sized particles. THEMIS and MOC visible data confirm this general interpretation, but the greater detail of the visible imaging shows that the warm materials include accumulations of dark sediment as well as more coherent bedrock and some brighter surfaces. Thus the assumption of homogeneous warm materials is not correct, and the calculated particle sizes presented in section 3.1 do not represent the full range of particle sizes present on the surface. Although the intermediate-temperature materials appear to be free of dark sediment accumulations, it is visibly darker than the crater floor and contains what appears to be coherent bedrock with sharp relief. During the daytime (when temperature is governed primarily by the albedo of a material) the intermediate-temperature unit is closer in temperature to the bright crater floor unit, whereas at night (when the temperature is dominated by the thermal inertia of the upper few centimeters of the surface) larger, indurated, or more compacted particles cause the unit to be significantly warmer than the crater floor. Because the daytime temperature, as well as the albedo, of the intermediate-temperature unit is much closer to that of the crater floor, these observations can be interpreted to indicate that a thin layer of dust overlies the intermediate-temperature unit, moderating its thermal inertia. Because the intermediate-temperature unit is darker than the crater floor at visible wavelengths (Figures 1, 4, and 5), this layer of dust must be optically thin at visible wavelengths, or less than about 10 microns. Alternatively, these thermal and visible properties could be explained by a patchy distribution of optically thick dust within the intermediate-temperature unit, causing checkerboard, subpixel mixing in the thermal and visible data. Yet another possible scenario is that the entire intracrater dark-toned feature is covered in dust (either as an optically thin coating or as thicker discontinuous patches), and the warmer daytime temperature of the warm materials unit is due to the darker sediment contained within it. Considering the high albedo of the Amazonis region in general, we interpret these darkest materials to be relatively free of dust, in which case, they probably indicate geologically recent saltation in the area. The brighter rippled bed forms in Figure 5 are interpreted to be inactive and possibly indurated or armored, as the darker sediment appears to be superposing them.
 In summary, the observed thermal inertia values most certainly represent averages of a range of thermal inertias within the temperature units, and result from a complicated mixture of dust, sediment, and more coherent bedrock. Additionally, if the brighter bed forms are indurated or armored, this would further complicate the interpretation of thermal inertia values. Despite these complications, we interpret the darkest sediment (located in the warm materials unit) to be recently or currently saltating. In addition, even though are there no dark sand accumulations in the intermediate-temperature unit, the thermal inertia of this unit is twice as high as that of the crater floor and is at the lower end of the range of inertias for the warm unit. Thus the intermediate unit's visible albedo and inertia suggest that sand is present, probably trapped in microtopography such as small pits, crevasses/cracks, etc. We therefore interpret the majority of the dark, intracrater feature to be composed of a mixture of sand, rock, and bedrock, with a small amount of dust present in or on the intermediate-temperature unit.
4.2. Compositional Nature of the Low-Albedo Materials
 Previous global-scale studies of the Martian surface using TES data have determined that low-albedo regions are dominated by two lithologies: a basaltic composition (surface type I) and an andesitic [Bandfield et al., 2000; Christensen et al., 2000c], altered basalt [Wyatt and McSween, 2002], or glass-rich basalt [Noble and Pieters, 2001] composition (surface type II). Previous compositional studies of other dark, intracrater features [Aben, 2003; Fenton et al., 2003; Rogers and Christensen, 2003; Wyatt et al., 2003] have found that the spectral and compositional properties of these features are similar to typical surface type I or II lithologies. The spectra of these two surface types are shown in Figure 11 and their modeled compositions are given in Table 2. The spectrum of the low-albedo materials from the intracrater feature in Amazonis Planitia shows little similarity to either the surface type I or surface type II spectra, instead displaying distinct spectral characteristics. The most prominent differences between the Amazonis Planitia low-albedo spectrum and both surface type spectra are the 1200–950 cm−1 (8.33–10.5 μm) slope and the strong emissivity minimum centered at ∼900 cm−1 (11 μm) in the low-albedo spectrum, which is not present in either of the global surface type spectra. In addition, the slope at 525–450 cm−1 (19–22 μm) is not present in the spectra of the two surface types, and the features at lower wave numbers for surface type I and the low-albedo material have different spectral shapes. Surface type I has two local minima in the 400–350 cm−1 (25–28.6 μm) range, whereas the low-albedo spectrum has a broader minimum centered around 350 cm−1 (28.6 μm). The modeled mineral abundances of our dark, intracrater feature represents a mafic composition; whereas surface types I and II are dominated modally by plagioclase, the low-albedo material studied here is dominated by pyroxene and olivine (Table 2).
 The compositions of Martian lithologies (including meteorites) may be compared using bulk chemistry data [Hamilton et al., 2001; McSween et al., 2003, 2004, 2006]. Details on the derivation of bulk chemistry from TES modeled mineral modes and the errors associated with this technique are discussed by Hamilton and Christensen , Hamilton et al. , and Wyatt et al. , and the full derived bulk chemistry for the low-albedo material is listed in Table 3. (Note that for the bulk chemistry calculation, we first normalized the mineralogy to exclude carbonates, as they should not be considered in a comparative chemical analysis of igneous rocks.) Figure 12 compares the bulk chemistry (alkalis versus silica [Le Bas et al., 1986]) calculated from TES modeled phase abundances for the low-albedo feature in Amazonis Planitia to other Martian lithologies. The low-albedo material comprising the intracrater feature studied in this work contains much lower silica than most other known Martian lithologies and, on the basis of this silica content, is classified as an ultramafic composition rather than the mafic composition indicated by its modeled mineralogy.
See text for explanation of derived bulk oxide calculation. All reported values are normalized to exclude H2O and CO2 abundance and rounded to nearest 1%. Uncertainties for each oxide are minimum standard deviations expected for bulk oxide chemistry derived from laboratory data convolved to TES spectral resolution [Hamilton et al., 2001].
 Oxides make up 15% of the modeled mineralogy (normalized to exclude carbonates), and this undoubtedly moderates the calculated silica content. Although both of the oxides (magnetite and ilmenite) used to model the TES spectrum are common in terrestrial mafic to ultramafic rocks and therefore geologically plausible in the modeled mafic mineralogy, there may be spectral reasons to exclude them. For example, errors in temperature determination when converting radiance data to emissivity result in a negative spectral slope [Ruff et al., 1997]. Because the infrared spectrum of magnetite has a negative slope with only one small feature in the TES wavelength region available for deconvolution, it is possible, although not demonstrable, that magnetite might be erroneously modeled if there were slight errors in the temperature determination used for conversion of TES data to emissivity. One possible source of temperature error is the inclusion of colder, dusty crater floor materials in the TES pixels' field of view for the low-albedo intracrater materials (resulting from down-track smear of the pixels). Inclusion of such materials means that the TES spectrum represents a broad range of temperatures, and a single temperature used to convert to emissivity may introduce a minor slope. The 55% high-albedo surface dust included in the model of the low-albedo spectrum supports the idea that the “low-albedo” pixels actually observed some higher-albedo surfaces. In the event that the oxides in the model might be erroneously identified, we recalculated the bulk chemistry to exclude the modeled oxide minerals (Table 3 and Figure 12). These data show that regardless of whether or not oxides are included in the bulk chemistry derivation, the calculated SiO2 content falls below the upper bound of 45% for terrestrial ultramafic rocks and below the SiO2 content of most known Martian lithologies.
 Although limited expanses of lithologies other than surface type I or II have been found in low-albedo regions on Mars, never before has such a low bulk silica lithology been identified on Mars. Olivine-rich basalts have been identified in a number of areas, most notably in Nili Fossae [Hamilton et al., 2003; Hoefen et al., 2003]. TES data also have revealed orthopyroxene-rich materials in Eos Chasma [Hamilton et al., 2003] and dacite in the Nili Patera caldera of Syrtis Major [Christensen et al., 2005]. Even a quartzofeldspathic lithology has been identified in northern Syrtis Major [Bandfield et al., 2004a]. All of these lithologies however, have bulk silica contents distinctly higher than the Amazonis Planitia lithology described here, and none of them contain as much modal clinopyroxene. In addition, none of the surface spectral shapes of these lithologies show a strong spectral match to the surface spectrum of the intracrater materials from Amazonis Planitia. The olivine-rich lithology in Nili Fossae has the most similar overall spectral signature (Figure 11), however its exact spectral shape in the 1250–833 cm−1 (∼8–12 μm) region has clear differences with the Amazonis Planitia intracrater materials spectrum. We note that Rogers et al.  did use TES data to identify a pyroxene-rich material in Ares Vallis, although they generally referred to the lithology as an olivine-rich unit. In contrast to the present work, the modeled mineralogy for the Ares Vallis unit suggested an ultramafic lithology whereas the derived bulk chemistry produced an SiO2 content of at least 48 wt.%, suggesting a mafic lithology. Although Rogers et al.  did not explicitly state whether the carbonate and sulfate in the modeled mineralogy were included in the calculation of the derived bulk chemistry, excluding these mineral groups would only increase the bulk silica content of the Ares Vallis lithology. So, although the material identified by Rogers et al.  is pyroxene-rich and modally dominated by mafics, its bulk silica content is significantly higher than the intracrater low-albedo material in Amazonis Planitia identified by this work.
 The significance of our identification of such a low bulk silica lithology is two fold. First, it extends the known igneous diversity of Mars to an ultramafic composition. Rather than supporting the original view of a planet composed of only two generally basaltic lithologies, basalt and basaltic andesite [Bandfield et al., 2000], it supports the more recent proposition that the composition and geology of Mars is far more diverse [Christensen et al., 2005]. Second, it may provide a link between the compositions identified on Mars by remote sensing data and the Martian meteorites. Although none of the modal mineralogies or bulk chemistries of any of the Martian meteorites match that of the low-albedo material in Amazonis Planitia, all of the Martian meteorites are modally dominated by mafics such as pyroxene and olivine rather than plagioclase. In particular, two meteorites (Chassigny and ALH 77005) have ultramafic bulk silica contents, although their mineralogies do not match well with that of the Amazonis Planitia intracrater material.
4.3. Origin of Low-Albedo Intracrater Material
 Previous studies have attributed both local [Aben, 2003; Christensen, 1983; Edgett and Christensen, 1994; Rogers and Christensen, 2003] and regional [Christensen, 1983; Edgett and Christensen, 1994; Fenton et al., 2003] sources to the sediment comprising dark, intracrater features. In the case of this intracrater feature, visible images strongly suggest that the dark sediment in the feature currently is being (or recently has been) blown out by a dominantly southwesterly wind from two point-like sources on the northeastern end of the feature. This wind direction is consistent with the orientation of dark-toned wind streaks in the region, most notably the large wind streak associated with nearby (∼430 km) Pettit crater. Many investigators have studied sand transport distances on Mars [Edgett and Christensen, 1991; Greeley et al., 1980; Iversen et al., 1976; Rogers and Christensen, 2003; Sagan et al., 1977], and most recently Rogers and Christensen  estimated that fine sand can travel up to ∼1000 km on Mars. However, there is no obvious source outside the crater for the dark sediment in the surrounding region within this range (particularly to the northeast). Although it is possible that the source could be obscured by dust in this rather dusty region of Mars, the evidence for recent or active saltation of dark sediment in the intracrater feature suggests that the source would be visible as well, due to similar movement and saltation of the source sediment. In addition, the nearest dark, intracrater feature, in Pettit crater at a distance of ∼430 km, has a basaltic to basaltic andesitic composition similar to surface type I [Aben, 2003; Rogers and Christensen, 2003], suggesting that our mafic-rich lithology might be limited to this particular intracrater feature. Consequently, we believe that the origin of the dark sediment within the intracrater feature is from local erosion and deflation of the crater floor.
 The presence of significant amounts of pyroxene and olivine combined with the lack of strong evidence for abundant alteration phases in the intracrater feature suggest that the material in the feature has undergone little weathering. Although it is possible that weathering phases have been removed preferentially, there is no obvious evidence of this such as a deposit of alteration phase materials in the surrounding area. There are two possibilities that might explain the mafic-rich composition and low bulk silica nature of this feature: (1) it is the result of sorting or segregation by an unknown process or (2) it represents a previously unidentified local or regional lithology. Although previous work has suggested that sorting processes have operated in other intracrater features on Mars, these processes produced deposits which are composed of two compositionally distinct units [Wyatt et al., 2003], which is not observed in the feature we have studied. Additionally, there are no exposed deposits of differing composition in the immediate surrounding area (within ∼400 km). For these reasons we favor the interpretation that the intracrater material represents a primary lithology with little or no removal or segregation of alteration phases or other materials.
 Assuming that the modeled mineralogy of the intracrater feature represents a primary, relatively unaltered lithology, the combination of mineral compositions present is quite unusual. The average An number (83) of plagioclase is significantly higher than previously identified by TES, and generally is indicative of an unevolved, ultramafic lithology. The Mg # of the clinopyroxene (which is all augite) is 85, and is therefore also in agreement with an ultramafic source. The average olivine composition, however, is Fo50, which is more iron-rich than would be generally expected in a low-silica, ultramafic lithology associated with such high An and Mg numbers. Previous work has established that average modeled compositions of plagioclase and pyroxene match measured compositions with errors of ∼10–15 Mg# or An# [Hamilton and Christensen, 2000; Wyatt et al., 2001], and a similar accuracy is expected for olivine compositions [Hamilton et al., 1997], so uncertainties in the derived solid solution composition cannot explain the low Mg# of the modeled olivine. There are two possible explanations for detection of a unique mineralogical assemblage such as this, which is not in equilibrium. First, the detected lithology could be a combination of one or more separate lithologies. Although the intracrater feature is heterogeneous in terms of relative albedo (and likely mean particle size) on scales much smaller than the size of the TES footprint, the THEMIS IR spectra and decorrelation stretch do not exhibit significant spectral variation within the feature. Second, the lithology could represent an igneous cumulate rock. Cumulates can produce mineral assemblages with unusual compositional variations, and there are terrestrial cumulates, associated with layered mafic intrusions, that produce rocks in which plagioclase with a high An number coexists with ferro-silicates with a lower than expected Mg # [Hess, 1989]. A cumulate origin also would explain the general mafic mineralogy as well as the ultramafic bulk silica content of the lithology. In fact, several of the Martian meteorites (including Chassigny and ALH 77005) possibly represent plutonic or volcanic cumulate rocks [e.g., Nyquist et al., 2001].
 Although we favor the interpretation of an igneous cumulate rock, the unusual mineralogical assemblage, ambiguous origin, and obvious morphologic heterogeneity at the limit of the remote sensing instruments make it difficult to place this low-albedo intracrater material in a well-defined geologic context. The depth to diameter ratio of the crater is as expected for typical Martian craters [Strom et al., 1992], so although post-formation crater fill cannot be discounted there is no reason to invoke it. Alternatively, the intracrater material may have been excavated from depth during the time of crater formation. Regardless of the exact origin of the materials, previous workers have interpreted and mapped the crater and its surrounding vicinity as either volcanic plains material of Hesperian age or ancient undivided material (of either Hesperian or Noachian age) [Scott and Tanaka, 1986]. Therefore the mafic-rich and seemingly alteration-free material exposed on the floor of this crater seems to suggest that, at least in this area, the climate at the time of deposition and since subsequent exposure of these materials has not been conducive to extensive chemical weathering. The possibility exists however, that these materials represent crater fill that post dates the formation of the crater and its surrounding terrain, and is now being exposed by erosion and deflation.
 Through our analyses of this low-albedo, intracrater feature with MGS TES, MOC, and ODY THEMIS data sets we have found the following:
 1. The thermal and spectral properties of the low-albedo material are consistent with a mixture of sand, rock, and bedrock and the composition of the feature generally is mafic to ultramafic. Relative to surface types I and II, the material comprising the low-albedo feature appears to have significantly greater amounts of pyroxene and olivine and lesser amounts of plagioclase. Derived bulk chemistry indicates that this material is ultramafic (∼40 wt.% SiO2) with the lowest bulk silica content observed on Mars to date.
 2. The rest of the crater floor is covered with a layer of dust that is thick at thermal wavelengths (>50 μm) and is spectrally (and thus compositionally) indistinguishable from the surface dust that is present in other high-albedo regions on Mars.
 3. The low-albedo feature is heterogeneous in terms of relative albedo and morphology on scales of meters to kilometers, and part of the dark intracrater materials may be covered with dust, either as an optically thin (few microns thick) layer, or as thicker discontinuous patches. Although some of the dark feature appears to be composed of immobile and perhaps indurated bed forms, it is possible that the darkest areas of the feature contain actively saltating sediment that prevents dust accumulation in this rather dusty region of Mars.
 4. The distribution of the dark sediment within the feature as well as the lack of an obvious source for the dark materials in the surrounding region suggests that the sediment is locally derived. The absence of evidence for sorting or segregation processes suggests that the mafic-rich composition and low bulk silica nature of this intracrater feature likely represents a previously unidentified local or regional lithology.
 5. Finally, the presence of significant amounts of pyroxene and olivine combined with the lack of sheet silicates does not provide evidence for extensive weathering; if weathering and/or alteration has occurred, there is no apparent evidence in the area for preferential removal of alteration phases, or it is too low in abundance to be detected by TES.
 We are in the process of conducting similar analyses for other low-albedo intracrater features in Amazonis Planitia to determine if the distinctive composition we observed in this feature is present in other craters in the area and can be established as a larger more regional unit [Schneider and Hamilton, 2005, 2006]. Initial results indicate that while a few occurrences of mafic to ultramafic intracrater materials are present in the Amazonis Planitia region, they are isolated and likely do not represent a regionally pervasive unit. Examination of other spectral data sets (e.g., OMEGA, CRISM) may aid in helping to establish the presence or absence of alteration phases that may be present in low abundances in the low-albedo intracrater feature.
 We would like to thank Melissa Lane for providing the spectra of samples z194, z101, and z043. We would also like to thank Lori Fenton and Ken Edgett for helpful discussions regarding this work and Josh Bandfield and Steve Ruff for thorough and insightful reviews that improved the quality of this manuscript. This work was supported by NASA's Mars Data Analysis Program (NAG5-13421). This is HIGP publication 1439 and SOEST publication 6766.