Analysis of a spectrally unique deposit in the dissected Noachian terrain of Mars



[1] We have analyzed a spectrally anomalous dark intracrater deposit and an adjacent lighter-toned fan-shaped deposit in the dissected Noachian terrain of Mars using a combination of high spatial and spectral resolution remote sensing data sets. The spatial proximity of these two deposits and their dust-free spectral character make these prime targets for (1) determining whether the dark-toned deposit is derived from the fan-shaped deposit and (2) constraining the nature and mineralogy of the fan-shaped deposit. Stereo observations and derived digital elevation models of the of the fan-shaped deposit show a set of three flows lobes. Calculations of the yield strength, based on morphometry of the lobes, indicate yield strengths consistent with those of other volcanic flows on Mars, although an aqueous origin cannot be discounted. The dark deposit consists of a mantle that forms dunes at its thickest portions and extends as a thinner veneer. We find no obvious morphologic association between the dark deposit and adjacent fan. Multispectral imaging from the THEMIS VIS and IR instruments show that the two deposits have different spectral characteristics, probably caused by differences in their mineralogy rather than their physical properties. Finally, modeling of MGS/TES spectra of the combined deposits indicates a feldspar-rich composition that is possibly complemented by smaller quantities of alteration products such as carbonates, and possibly sulfates and clays. These results indicate that (1) there is no association between these deposits and (2) the combined deposits have a primarily igneous mineralogy, although some aqueous alteration may have occurred.

1. Introduction

[2] The heavily cratered terrains of the southern highlands are the oldest geologic units on Mars. They were formed during and immediately after the period of heavy bombardment, and although heavily cratered, they have been extensively dissected by channels and valley networks that were presumably formed contemporaneously by liquid water [e.g., Carr and Clow, 1981; Baker and Partridge, 1986]. Because the Noachian terrains formed during (and immediately after) the period of heavy bombardment, studies of these landforms' morphology and composition may yield information about the planet's past climate. Not only do the valleys that dissect the region raise the possibility of more clement atmospheric conditions on early Mars, but they also provide an excellent probe into the composition of the cratered Noachian highlands, if the eroded materials that were deposited at the mouths of these valleys are still present. Unfortunately, dust masks the spectral signature of most of the Noachian highlands' bedrock materials [Ruff, 2001] and although there are regions where lower albedo, coarser grained material is present (e.g., in intracrater deposits), its source is usually not known. It is therefore both rare and of high interest to find Noachian materials with spectral signatures that are not obscured by mantling dust and whose source is well established. Recently, deposits of such materials were identified by K. S. Edgett (personal communication, 2003) in the Mars Global Surveyor's Mars Orbiter Wide-angle Camera data (MGS/MOC-WA) data set [Bell et al., 2003a]. The materials are within an unnamed 17 km diameter crater in the Aeolis quadrangle of Mars (near latitude 12.8°S, longitude 157.5°E) and are characterized by an 8 × 11 km intracrater deposit of anomalously high reflectance in the blue relative to the surrounding terrain and an adjacent, even bluer, 1.5-km-diameter fan-shaped deposit located at the mouth of a small source valley (Figure 1). The spectral conspicuousness of these two deposits, along with their interesting associated morphology, makes this region a prime target for an in-depth analysis of ancient Martian deposits and processes. In this paper, we are interested in (1) determining whether there is any association between the dark intracrater deposit and the fan-shaped deposit, and (2) constraining the mineralogy of both deposits in an effort to understand the physical processes involved in their genesis. In order to do so, we have used MOC stereo imaging to constrain the morphology and rheologic properties of these deposits, and we have every visible and infrared spectroscopic dataset that is available to constrain the relationship and mineralogy of the deposits.

Figure 1.

Multispectral views of the intracrater deposit using (a) MOC/WA Blue filter (bandpass at 400–450 nm, E2100582); (b) MOC/WA Red filter (bandpass at 575–625, E2100581); (c) THEMIS/VIS (V04277003) bands 1 (420 nm), 2 (540 nm), and 3 (650 nm); and (d) MOC/NA (bandpass at 500–900 nm, E2100216). North is up in all the images. Arrows in Figures 1a and 1b indicate the crater containing the deposits of interest. The white box in Figure 1c shows location of MOC image in Figure 1d.

2. Data Sets and Analysis Methods

[3] High-resolution images and spectra of this region have been acquired by the MGS/MOC Narrow Angle (NA) Camera, the MGS Thermal Emission Spectrometer (TES), and by the Mars Odyssey Thermal Emission Imaging System's Visible and Infrared cameras (THEMIS VIS and IR). In this work, we used a combination of these data sets to characterize the morphology and constrain the mineralogy of the deposits in order to identify the processes that have been active in this region.

[4] Our morphological analysis is based on stereo MOC/NA ∼3 m/pixel images. Multispectral analyses were performed using high spatial resolution THEMIS images (five channels at 36 m/pixel in VIS and 10 channels at 100 m/pixel in IR) to characterize the spectral variability and its association to morphological features, and higher spectral resolution TES spectra (143 channels in the ∼6–50 μm region using 3 × 8 km pixels) to constrain the mineralogy of the larger-scale features.

[5] The spectral variability of the region was characterized using simple ratios and other spectral parameterizations as well as more complex end-member-identification methods such as minimum noise fraction analysis [e.g., Green et al., 1988]. Analysis of the TES spectra involved atmospheric correction using the method of Bandfield et al. [2000], and spectral unmixing using singular value decomposition [e.g., Heath, 2002] and statistical unmixing methods [Noe Dobrea, 2004].

3. Context and Morphology

[6] The deposits of interest are found inside one of the highly degraded (rimless and flat-floored) craters of the heavily cratered and dissected ancient Noachian highlands (Figures 1a and 1b) [Tanaka, 1986]. The larger deposit extends from the middle of the crater southwards, draping over the southern rim and 5 km into the surrounding plains (Figure 1a). It abuts, but does not mantle the fan-shaped deposit. High-resolution (2.85 m/pixel) MOC/NA images show that, near the crater wall, this deposit consists of a mantle of low-albedo material thick enough to form barchanoid dunes and obscure the underlying geology of the more ancient and heavily cratered terrain (Figure 1d). The deposit appears to thin toward the center of the crater, as the underlying morphology becomes more readily apparent. Outside the crater, the deposit also appears too thin to mask the underlying plains' morphology.

[7] The lighter-toned, fan-shaped deposit is located at the north end of a short (∼2 km) valley, that incises the crater's south wall. It consists of three discernable sets of superposed fan-shaped segments (Figure 1d), possibly the result of multiple depositional episodes. Its surface has a crater density similar to that of the unmantled crater floor, and is thus assumed to be of roughly the same age as the crater floor. The morphology of this deposit and its source valley suggest either a mass-movement emplacement facilitated by a lubricating agent such as liquid water (debris flow) or ice (gelifluction), or a volcanic flow. The spectral contrast of this deposit relative to the surrounding plains suggests that this deposit is clean of the homogenizing dust characteristic of the bright plains material. Saltating grains within the crater may be responsible for maintaining the clean surface of this unique deposit by removing dust.

[8] In order to measure the morphometric properties of the fan-shaped deposit, we have derived a topographic map of the location from a pair of MOC images taken at different times and at different emission angles. The resulting topographic data has a horizontal scale of 7.5 m/pixel and a vertical resolution of 3 m. Figures 2a and 2b show a perspective view of the deposit at a vertical exaggeration of 20× and its associated topographic profile. This view reveals three distinct lobes associated with different flow episodes. The deposit and source valley have volumes of 0.14 ± .01 and 0.08 ± 0.01 km3, respectively. Although the smaller volume measured for the valley could be caused by the infilling by mantling material (an interpretation partly supported by the presence of ripples or small dunes on the valley floor), a simpler explanation is the “bulking” caused by the disaggregation (and hence reduction in the density) of the deposit-forming material as a consequence of a mass movement. Bulking factors of 30% are common for mixed soils, and can reach upwards of 65% for rock [e.g., British Columbia Ministry of Forests, 2002; U.S. Department of Transportation, 1996].

Figure 2.

Perspective view of (a) the fan shaped deposit and (b) its associate topographic profile. The view has been vertically exaggerated 20x to accentuate the three layers that are interpreted to represent three separate flow episodes. The lobes in Figure 2b have been identified by numbered arrows. The crosses correspond to the minimum and maximum points used for the measurement of lobes' heights.

[9] In order to estimate the rheological properties of the flow, we have also measured the heights and sizes of the largest (bottom-most) and intermediate lobes. Both have heights of 63 ± 3 m and radii of approximately 750 m. Measuring the dimensions of the flow features allows us to estimate their yield strengths, which can then be compared to those of other well-known flows to constrain their properties [e.g., Squyres, 1978]. Because the lobes appear morphologically similar to some viscous flows, we model each lobe as a Bingham plastic [e.g., Fink and Griffiths, 1998]. A Bingham plastic is defined by its shear stress (τ) as

equation image

where τ0 is the yield strength, η is the plastic viscosity, and ɛ′ is the shear rate. Equation (1a) implies that the shear rate will be 0 when the shear stress is less than the yield strength. When the shear stress is greater than the yield strength, the plastic will deform at a rate inversely proportional to the viscosity. The yield strength can be estimated from the morphometry of the flow [Blake, 1990],

equation image

where C ∼ 0.5 [Blake, 1990; Squyres, 1978], ρ is the material's density, g is the surface gravity, H is the thickness of the lobe, and R is the radius of the lobe.

[10] We assume a density of 2.5 ± 1 g/cm3, where the variance is intentionally large in order to encompass all reasonable densities into our calculations. For this density, each of the two larger lobes of the fan-shaped deposit would have a yield strength of ∼0.23 ± 0.12 bars (∼2.5 ± 1.3 × 104 Pa). The deposit sits on a slope of no more than 2° (as measured from the surrounding region), and shows no evidence of having evolved into a coulee. The maximum radius that such a deposit can reach on a 2° slope is ∼5 km [Blake, 1990, equation (37)], which is consistent with our observations.

[11] Table 1 depicts the yield strengths of several volcanic and non-volcanic types of flows. The estimated yield strength of the fan-shaped deposit is lower than that measured for some volcanic flows and rock glaciers, but is comparable to that measured for volcanic flows on Olympus Mons, Ascraeus Mons, and Arsia Mons, as well as those of some basaltic flows on the Earth [Moore et al., 1978]. Although the calculated yield strength is significantly higher than that measured for experimental mud slurries, yield strength is a strong function of water/soil ratio and there is no reason to doubt that such yields strengths can exist for a given mud slurry containing a higher fraction of rock and dirt than that usually used in the lab.

Table 1. Yield Strengths for Flow Features on the Earth, Moon, and Mars
FlowYield Strength (PA)Reference
Terrestrial rhyolites, dacites, and basaltic andesites>105Blake [1990], De Silva et al. [1994], Wilson and Head [2003]
Terrestrial basaltic flows1 × 104Moore et al. [1978]
Martian volcanic flows (Olympus mons, Ascraeus Mons, Arsia Mons)1 × 104Zimbelman [1985], Warner and Gregg [2002]
Lunar domes (Gruithuisen and Mairan)3 × 105Wilson and Head [2003]
Terrestrial rock glaciers1 × 105Wahrhaftig and Cox [1959]
Martian lobate aprons1 × 105Squyres [1978]
Terrestrial lahars<3.4 × 103Manville et al. [1998]
Experimental mud slurries<102Parsons et al. [2001]

4. Spectral Characteristics


4.1.1. Calibration

[12] A 36 m/pixel THEMIS/VIS image cube of the area was acquired on 1 December 2002 (image V04277003). The data were radiometrically calibrated and reprojected to sinusoidal equal-area projection. The THEMIS-VIS radiance calibration is described by McConnochie et al. [2003] and McConnochie and Bell [2003]. THEMIS-VIS images are radiometrically calibrated via comparison to HST/WFPC2 derived radiance factor (I/F) values, and incorporate a correction for shutter smear and for a spatially variable stray light component [McConnochie et al., 2004]. However, calibrated THEMIS-VIS images we used are not corrected for the effects of a spatially uniform component of stray light, which manifests itself as a wavelength-dependent additive offset with values of 18%, 6%, 4%, 9% of the signal for the 425, 550, 650, and 750 bands, respectively [McConnochie et al., 2005]. Since radiometric calibration is accomplished by forcing the average I/F of this data set (after correcting for viewing geometry) to be equal to the average HST I/F, the presence of stray light does not bias the average I/F values of any given THEMIS-VIS frame. Instead, the stray light acts to artificially reduce the I/F contrast within each THEMIS-VIS frame. Any correction for this stray light must therefore act to increase intraframe contrasts, while leaving the frame means unchanged [McConnochie and Bell, 2003].

[13] The correction process involves two steps: (1) the derivation and correction for the additive stray light offset and (2) the renormalization of the mean value of the scene to that derived before the contrast correction [Bell et al., 2003a, 2003b]. We can derive the additive offset by comparing the radiance spectra of two closely spaced regions from within the crater: a flat illuminated region and an illuminated sun-facing slope (Figure 1c, blue and yellow selections), both of which are assumed to be Lambertian surfaces and to have the same reflectance spectrum, based on general morphologic and color similarities. Because both regions are assumed to be spectrally identical, their spectral ratio must be independent of wavelength (Figure 3a). The only difference in these values would then only be due to an additive offset (k),

equation image

where kλ is the filter-dependent linear offset in the radiance, fd(λ) is the radiance observed for the darker, less illuminated region, and fb(λ) is the radiance observed for the brighter, more illuminated region. In this case, we chose to normalize our spectra to VIS filter 3 (λ3 = 654 nm) because it exhibits the least amount of stray light offset. However, the choice of filter is not critical because the corrected data are later scaled using I/F values found from HST/WFPC2 images of the same area. The offset derived for each filter was calculated from equation (2) and added to the radiometrically calibrated data (Figure 3b), and the corrected data were subsequently renormalized so that the average scene radiance was equal to the value before the scattering correction.

Figure 3.

THEMIS/VIS spectra of (a) two uncorrected bright plains regions spectra exposed to different levels of lighting and their ratio; (b) the spectra of the same two regions and their ratio after corrections for a stray light offset; and (c) comparison of the calibrated THEMIS/VIS spectrum of bright plains material in the scene to the calibrated HST/WFPC2 spectrum of the same region.

[14] Finally, we converted the calibrated radiance data to I/F by dividing out the value of the solar irradiance spectrum convolved to each filter's bandpass. Here, I is the radiance measured on the sensor and πF is the irradiance of sunlight in the same spectral band at the top of the Martian atmosphere at the time of the observation. Comparison with spectra of this region from the Hubble Space Telescope's Wide Field Planetary Camera 2 (adjusted to the THEMIS/VIS viewing geometry) provides a validation of this calibration approach (Figure 3c).

4.1.2. Atmospheric Correction

[15] We convert the I/F values to an estimate of surface normal albedo by employing McConnochie et al.'s [2003] atmospheric correction method. A multiple-scattering radiative transfer model, DISORT [Stamnes et al., 1988], is used to generate a look-up table for I/F at the top of the atmosphere as a function of the surface normal albedo, viewing geometry, and atmospheric aerosol loading. The surface normal albedo is then estimated for each VIS pixel by linear interpolation from the look-up table.

[16] This method assumes that the surface is horizontal, nonshadowed, and has Lambertian scattering properties. Dust and water-ice absorption optical depths are determined from concurrent THEMIS-IR observations (I04277006), as described by Smith et al. [2003], and are converted to extinction optical depths at the wavelength of each VIS passband using Mars aerosol properties derived by Clancy et al. [2003] for dust and their “Type 1” ice aerosol. Our assumptions for the single-scattering albedo and single-scattering phase function of the aerosol particles are likewise taken from Clancy et al. [2003].

4.1.3. Analysis

[17] THEMIS/VIS serves both as a context imager for MOC/NA images and as a low-spectral resolution visible wavelength imaging spectrometer (Figure 1c). Features as small as the dunes and some of the larger craters on the fan-shaped deposit identified in MOC/NA data can be recognized in the VIS images (Figure 1c). It is clear from Figure 1c that both the fan-shaped deposit and the low-albedo region surrounding it have a different spectral signature from the rest of the crater's interior. The lighter tone of the fan shaped deposit (compared to the rest of the low-albedo region) also suggests that the deposit itself may have a different spectral character, caused by physical and/or mineralogical differences.

[18] Although the small number of channels provided by THEMIS/VIS precludes its extensive use as a mineralogical tool, it is used in this study to assess of the extent of spectral variability found inside the crater and to determine whether there is any compositional relationship between the dark deposit and the fan-shaped deposit. In order to do so, we first determine the degree of spectral variability and identify spectroscopic end-members in the scene and then we determine how these end-members are mixed throughout the scene.

[19] The determination of the spectral variability and identification of end-members is performed using a Minimum Noise Fraction (MNF) [Green et al., 1988] transform on the four-band THEMIS/VIS data of the crater and its interior. MNF is a variant of a Principal Components Analysis (PCA) data transform [e.g., Johnson et al., 1985], designed to segregate signal into the first several principal components and noise into the later components. Figures 4a through 4d show the four component planes derived from the MNF transform. Most of the non-noise-related spectral variability appears in planes 1–3, whereas plane 4 is noise dominated. A color composite of the first three MNF planes (Figure 4e) highlights four different spectral units associated with the underlying morphology and albedo: (1) the classical bright cratered terrain (light blue), (2) the fan-shaped deposit (light green), (3) the dune-forming dark deposit (orange), and (4) the thin coating of low-albedo material that is draped both inside and outside the crater (dark green).

Figure 4.

(a–d) Minimum Noise Fraction planes 1–4 of THEMIS/VIS filters 1–4, and (e) RGB composite of MNF planes 1–3. The yellow, red, blue, and green rectangles are the selection regions for the spectra of Figure 5.

[20] The spectroscopic relationship between these end-members is then determined by using four-band VIS spectra (Figure 5) from these four end-member regions to define three parameters that describe the spectral variability of the data set: Parameter A is the ratio of band 4 to band 1 (750 to 425 nm), and therefore represents the overall slope of the spectra; parameter B is a ratio of band 4 to band 3, which is a measure of the slope between 750 and 650 nm, and parameter C represents the continuum curvature at band 2 (540 nm), or the degree to which the band 3 to band 1 slope is “kinked.” The latter is parameterized by the depth of band 2 relative to a “continuum” defined by bands 1 and 3. In other words, we perform a linear interpolation between bands 1 and 3 to generate a “kinkless” band 2, and then define the band depth (C) by

equation image


equation image
equation image
equation image

These three parameters were calculated for all points in the region. A scatterplot of parameter A versus parameter C reveals two major classes into which most of the data plot. These correspond to the bright plains material (at the highest A and C parameter values: Figures 6b and 6c, magenta selection), and the remaining terrain (at lower A and B parameter values). The spread of the bright plains material toward higher values of A is caused by topographic slope effects and shadows, which are not accounted for in our atmospheric correction. In addition to these two major regions, we also identify two outliers at the lowest values of the A parameter (Figure 6b and 6c, red and blue selections), which correspond to the fan-shaped deposit and to the dune-forming portion of the adjacent deposit. The remaining material that has low A and C parameter values can be explained by linear mixing of these two end-members (green selection), as well as by the additional mixing of this mixture with the bright plains material (yellow and light blue selection). An additional scatter of parameter B versus C shows that they are linearly related: spectra with strong 540-nm kinks (i.e., bright plains material) also have the highest near-infrared spectral slope (Figure 6d). These results are consistent with variations in ferric mineralogy among the different terrains [e.g., Bell et al., 1993].

Figure 5.

Spectra of the regions identified in Figure 4e. Note the differences that led us to define the three parameters: A, band 1–band 4 slope; B, band 3–band 4 slope; and C, depth of band 2 relative to bands 1 and 3.

Figure 6.

(a–c) Scatterplots and map of parameters C versus A and (d) parameters B versus C, color coded to identify concentrations of points (Figures 6a and 6d), and to find correlations between the parameters and the geology (Figures 6b and 6c). Note the separation of two distinct data regions (high and low values of parameter A) and the two vertices of the region with low A parameter values (red and blue), which are associated with the different geologic units and their mixing regimes. Topographic slope effects, which are not properly accounted for in our atmospheric corrections, cause spread toward higher values of parameter A.


[21] The thermal IR (TIR) portion of THEMIS consists of an uncooled 320 by 240 element microbolometer array with nine channels centered from 6.5 to15 μm. Images are assembled in a pushbroom fashion with subsequent lines in the array co-added within each spectral channel to increase the signal-to-noise ratio. Spatial sampling is 100 m from the 420-km-altitude circular orbit of Mars Odyssey. Local times of observation range from 3 to 6 a.m./p.m. [Christensen et al., 1998].

4.2.1. Calibration and Atmospheric Radiance Correction

[22] An internal calibration flag and instrument response functions determined from prelaunch data are used to produce calibrated radiance images. Instrument temperature drift during the collection of the image is accounted for by assuming that any temperature drift in the 15-μm atmospheric CO2 band is due to the instrument and applying a correction to all spectral channels. The 1-σ noise equivalent spectral radiance for the THEMIS/IR instrument varies from 1.67 × 10−6 Wcm−2str−1μm−1 for band 5 to 4.49 × 10−6 for band 1, corresponding to a noise equivalent delta emissivity (NEΔɛ of 0.00538 and 0.0306, respectively [Christensen, 2003]. A thorough description of the calibration as well as random and systematic uncertainties is presented by Bandfield et al. [2004].

[23] An additional constant radiance correction (described by Bandfield et al. [2004]) was also applied to the data to remove radiance from atmospheric emission. This correction removes apparent emissivity effects between similar surfaces of different temperatures.

[24] Images of our study region were acquired by THEMIS/IR on 30 October and 12 December 2002 (I04277002). The calibrated radiance images were map-projected and the constant radiance correction was applied to remove the apparent emissivity temperature dependence. We selected the brightness temperature derived for the 7.93-μm band (Filter 3) to represent the value of the surface temperature because this wavelength range covers the Christiansen feature of silicates (centered at approximately 7.6 μm), and is assumed to represent a region of relatively low atmospheric or surface absorption. Brightness temperatures were derived for each pixel by interpolating a set of tabulated blackbody radiance values (computed for a wide range of temperatures and convolved to each filter's bandpass) to the measured radiance in the 7.93 μm band (e.g., Figure 7). Finally, the emissivity was calculated for each pixel of each band in the scene using

equation image

where Rν is the observed radiance at band ν and R(Tsurf)ν is the surface's calculated blackbody radiance derived for each band using the brightness temperature calculated from the 7.93-μm band.

Figure 7.

Scatterplot of THEMIS/IR measured radiance versus expected radiance for a compositionally homogeneous surface at varying temperatures.

4.2.2. Results

[25] In order to identify spectral end-members, we performed an MNF transform on the resulting multispectral cube. Figure 8 shows a THEMIS/IR brightness temperature image of the study region. Both the fan-shaped deposit and the low-albedo deposit (in this case bright because of it's relatively high surface temperature) can be identified. Figure 9 is a color composite of the first three MNF. The fan-shaped deposit (yellow) and the dune-forming low-albedo deposit (cyan) are distinct end-members that can be discerned above the systematic (“plaid”) noise pattern that is apparent primarily in the low-temperature bright plains material. On the basis of this information, we extracted spectra from these two end-members as well as a large section of the higher albedo crater floor.

Figure 8.

THEMIS/IR brightness temperature map of the region of interest. The highest temperature (245 K), found on the lit crater wall, is significantly higher than that on the crater floors bright plains terrain (215 K) or the dune-forming and fan-shaped deposits (230 and 225 K, respectively). North is up.

Figure 9.

Color composite of the first three Minimum Noise Fraction planes from the THEMIS/IR image cube of our study region. Spatial scale is the same as in Figure 8. The fan shaped deposit (yellow) and the dune forming deposit (cyan) stand out above the noise in the bright plains material due to their higher temperature and consequently higher signal to noise.

[26] Because we are primarily interested in the spectral differences between the fan-shaped deposit and the dark intracrater deposit rather than absolute emissivities, we restrict our analysis to spectral ratios (Figure 10) [see also Ruff and Christensen, 2003]. In this experiment, we took the ratio of the average emissivity spectrum of each of the deposits to that of the average emissivity spectrum of the bright plains material inside the crater (Figure 10a). The error bars displayed in Figure 10a are overly conservative because they represent a formal propagation of uncertainties in the band ratio, where the uncertainty in each band is the standard deviation of a large set of pixels representative of each unit. A ratio of the dark dunes deposit spectrum to the fan-shaped deposit spectrum shows that although both deposits have a similar spectral shape shortward of 9 μm (∼1100 cm−1), the dark dunes deposit displays additional absorption longward of 9 μm (Figure 10b). The spatially-coherent nature of this spectral difference and its correlation with specific terrains provides us with confidence in the identification of this feature (Figure 10c).

Figure 10.

(a) Ratio of the spectra of both deposits studied here to the bright plains material spectrum and (b) ratio of the dark deposit spectrum to that of the fan-shaped deposit, (c) the THEMIS/IR Band 3 to Band 7 ratio image showing the correlation between absorption longward of 9 μm and the geological terrains (the cube has been normalized to the mean spectrum of the bright plains region), and (d) the convolved spectra of minerals having the same spectral shape as in Figure 10b. We have also included the spectrum of obsidian in Figure 10d to demonstrate its implausibility in accounting for the shape of the spectral ratio. The error bars in Figures 10a and 10b are formally propagated from the variance in the spectrum of each region. Although they are overlapping and imply nondifferentiable spectra, they are conservatively defined, and Figure 10c shows that the differences we do see correlate spatially to the two deposits, and are therefore real. In order to account for the error induced by noise, the selection for the bright plains material includes most of the bright plains material on the floor of the crater. The contrast of the ratio spectrum shown in Figure 10d has been enhanced by 300% for ease of comparison.

[27] We searched the ASU spectral library [Christensen et al., 2000] (convolved to the THEMIS/IR bandpasses) for minerals that have absorptions in the same spectral region as those seen in the ratio spectra of both deposits, constraining our selections to minerals that do not have any absorptions shortward of 8.5 μm but have absorptions longward of this wavelength. Only a small suite of minerals was found to possess the requisite absorption features (Figure 10d). These minerals included Mn-rich pyroxene, garnet, and olivine, all of which are fairly common unaltered volcanic products on the Earth. Although obsidian is also a primary volcanic product common in basalts, direct visual comparisons between its spectrum and that of the ratio spectrum show that its spectrum alone cannot account for the difference in the spectra of the two regions. These results must be taken with caution, however, because there are surely many combinations of other minerals that can reproduce this spectral difference equally well.

[28] We therefore find that the two deposits show evidence for spectral variability in the midinfrared. Their spectra are responsible for most of the spectral variability encountered in the corrected THEMIS/IR data. Analysis of this spectral variability indicates that the main difference in the spectra is most likely caused by compositional differences, rather than differences in geometry or packing. It is, however, not possible to fully constrain the exact mineralogical differences in these deposits using the eight bands available. Additional constraints on the mineralogy can only be performed with an instrument of higher spectral resolution.

4.3. MGS/TES

[29] TES has a significantly lower spatial resolution than THEMIS/IR (3 × 8 km/pixel on a six-pixel detector), but a greater spectral resolution and broader spectral coverage (∼6–50 μm, 200–1700 cm−1 at 10 cm−1 per channel). Radiometrically calibrated TES spectra and derived bolometric albedo values [e.g., Christensen et al., 1998] were converted to both brightness temperature and emissivity. Brightness temperatures were calculated by inverting the Planck function for each channel, and emissivities were determined for each spectrum by dividing its calibrated radiance by that of a blackbody spectrum whose temperature equaled each TES spectrum's highest brightness temperature in the 1300–1325 cm−1 region. As with THEMIS/IR, this spectral region was selected because it contains the transparency feature in silicates and has minimal atmospheric absorption. Although either brightness temperature or emissivity can be used to remove the effects of the blackbody curve, the main advantage of using brightness temperature over emissivity is it can be calculated directly by inverting the Planck function at each wave number. On the other hand, the derivation of emissivity requires the use of an assumed surface temperature, and errors in this assumption can lead to overall tilts in the spectrum.

[30] Figure 11a shows the TES surface temperatures overlaid on a Viking Mars Digital Imaging Model (MDIM) for all the available passes where TES data were acquired of the region. The strip-to-strip (orbit-to-orbit) variability is caused by varying surface and atmospheric conditions throughout the Martian year, but same-strip variability within this limited latitude range is assumed to be caused by varying surface temperatures alone. As in the case of the THEMIS/IR data, the lower albedo regions are typically ∼10 K warmer than the surrounding higher albedo material. Because the fan-shaped deposit (∼1.5 km across) is significantly smaller than each TES footprint (3 × 8 km), it will only contribute to a small fraction of the spectrum of a single TES pixel (∼8% if all of it is in the field of view) and is therefore assumed to be undetectable. However, THEMIS/VIS results suggest that material shed off the fan and material from the dune-forming mantling deposit mix into the rest of the dark deposit. This extensive low-albedo deposit occupies more than one TES pixel, and hence dominates the spectrum in that region.

Figure 11.

(a) TES surface temperature observations overlaid on a VIKING MDIM map, and (b) TES brightness temperature spectra of two adjacent observations taken with the same detector: directly on the dark deposit (ock 5921 ick 1621 detector 1) and off the dark deposit (ock 5921 ick 1622 detector 1).

[31] Brightness temperature spectra for the dark deposit and the bright plains material are shown in Figure 11b. These two spectra come from two consecutive (same-strip) footprints of the same detector on the crater floor. Most of the similarities between the two spectra are caused by atmospheric absorptions (Figure 12), including: (1) a broad water vapor absorption centered at 1550 cm−1 and multiple small absorptions between 200 and 400 cm−1 (jagged region); (2) a broad CO2 band (centered at about 670 cm−1) along with smaller CO2 bands (970, 1075, 1250, and 1375 cm−1) and two CO2 hot bands (∼550 and 800 cm−1); (3) atmospheric dust, contributing an absorption centered at about 1050 cm−1 and a strong slope from 200 to 500 cm−1; and (4) a broad atmospheric water-ice cloud absorption centered at 825 cm−1. Major differences include a deeper 1550 cm−1 absorption in the bright terrain, a deeper 1075 cm−1 absorption in the dark terrain, additional absorption at about 830 cm−1 in the bright terrain, and a stronger spectral slope between 200 and 500 cm−1 in the dark terrain.

Figure 12.

Sample TES spectrum compared to modeled atmospheric absorptions for the most relevant gases and aerosols. These models were generated using a multiple scattering radiative transfer code (Reflex) and Martian atmospheric conditions.

4.3.1. Atmospheric Correction

[32] Atmospheric correction of the TES spectra within our study region was performed using the method of Bandfield et al. [2000]. We first derived the emissivity spectrum by dividing the observed radiance spectrum by a blackbody spectrum whose radiance matched that of the observed spectrum at 1310 cm−1. We then unmixed the atmospheric components from the emissivity spectrum by subtracting the linearly weighted spectra of a set of empirically derived Martian atmospheric end-members. These end-members were derived by Bandfield et al. [2000] for a large set of Martian surface and atmospheric conditions, and their fractional abundances in the derived emissivity spectrum were determined using singular value decomposition (SVD) [e.g., Heath, 2002] in the spectral region of 800–1300 cm−1 (Figure 13). The 250–600 cm−1 region of the spectrum was not used to derive the abundances of the atmospheric end-members because the water vapor lines in the 200–400 cm−1 region are highly variable and introduce noise into the matrix inversion process. However, the final subtraction of the atmospheric end-members does include this region, with the caveat that small-scale (<40 cm−1) variability is not necessarily interpretable.

Figure 13.

(a) Surface/atmospheric end-member model match for the derived emissivity spectrum and (b) the derived atmospherically-corrected surface emissivity spectrum of the dark deposit.

[33] The derived surface emissivity (Figure 13b) has a number of distinguishable features that allow us to constrain the mineralogy of the region. The most apparent feature is a broad 900–1300 cm−1 asymmetric absorption. Experience with library mineral spectra leads us to believe that such absorptions are typically caused by the overlap of multiple absorption features, as is evidenced primarily in the kink in the wing at 1150 cm−1 as well as an additional band at 1100 cm−1. In addition to this absorption, we also note a smaller (∼3%) absorption in the 200–450 cm−1 region.

4.3.2. Spectral Uniqueness of Deposit

[34] To assess the spectral uniqueness of the dark deposit in the midinfrared portion of the spectrum, we identified all the spectra taken within 0.5° of the deposit which exhibit similar albedo (within 0.01) and which were taken at similar dust opacities (within 0.01) as the spectrum of interest. We then applied a PCA transform on the 72 atmospherically correctable spectra that were identified by this process to identify unique spectral end-members. Figure 14a shows a scatterplot of the first three PCA eigenfunctions and reveals that the spectrum of the dark deposit is distinctly different from the others in our potentially similar sample. Comparison of the atmospherically corrected TES spectrum of the dark deposit to TES spectra of other regions in our sample (Figure 14b) shows that the dark deposit has a deeper relative absorption in the 800–1300 cm−1 region than the other regions.

Figure 14.

(a) Three-dimensional scatterplot used to identify (b) unique spectra in the 0.5° × 0.5° region around the deposit of interest. Note the separation of the magenta point with respect to the main body, indicating that the spectrum associated to this point is different from the majority of the other spectra. The magenta point represents the spectrum of the deposit of interest.

4.3.3. Mineralogic Unmixing

[35] The existence of overlapping absorptions in this spectral region makes it difficult to constrain the mineralogy from direct absorption band comparisons. Because the spectra of the contributing minerals mix linearly in the infrared [e.g., Adams et al., 1986; Johnson et al., 1992; Ramsey and Christensen, 1998], the derived surface emissivity spectrum can be linearly unmixed using a set of mineral end-members. We used a simple matrix inversion routine to unmix the TES surface emissivity spectra, using as end-members spectra from the ASU mineral library [Christensen et al., 2000], an obsidian spectrum provided by J. Bandfield, a goethite spectrum provided by T. Glotch (the two latter spectra were acquired using the same laboratory methods as Christensen et al. [2000]), and a blackbody spectrum (unit emissivity), which was introduced to account for simple multiplicative emissivity offsets. Matrix inversions require the number of channels (73 after atmospheric correction) to be greater than or equal to the number of end-members. Therefore it was necessary select a subset of the spectral library. This subset was selected by first removing all the spectra taken for samples with small particle sizes (<250 μm) or solid blocks (hand samples), and then checking every spectrum against every other spectrum of each mineral group (i.e., pyroxene, feldspar, etc.), to select only representative spectra from each general class (i.e., for a set of similar spectra, we selected only one; see Figure 15). The reason we selected sand-sized particles was that the spectrum of solid blocks in the library does not contain different absorption bands than the spectrum of sand-sized samples. It simply presents deeper absorptions. The spectra of fine powders (note that the only spectra of powders in the ASU spectral library is for clay powders) on the other hand, do not present very high contrast (and linear mixing sometimes does not apply to mixtures of powders), so they were also excluded. In the spirit of keeping things simple, we chose the samples containing sand-sized particles, where linear unmixing in the thermal infrared works best. This reduced the number of potential end-members to a total of 52 (Table 2).

Figure 15.

Sample comparison of three similar pyroxene spectra [Christensen et al., 2000].

Table 2. List of Minerals Used for Deconvolution of the Derived Surface Emissivity
  • a

    Spectrum was reduced to 55% of original.

  • b

    Spectrum was reduced to 80% of original.

2Acmite LACMNH-6800 218
3Actinolite HS-116.4B 28
4Albite WAR-0235 174
5Almandine BUR-120A 34
6Andalusite WAR-0482 182
7Andesine WAR-0024 175
8Anhydrite ML-S9 81
9Anorthite BUR-340 178
10Anthophyllite BUR-4760 24
11Apatite ML-P1 86
12Augite BUR-620 71
13Azurite C14 109
14Biotite BUR-840 25a
15Bronzite NMNH-93527 168
16Bytownite WAR-1384 177
17Ca-montmorillonite STx-1 granular 196
18Calcite C22 116
19Chalk C6 104
20Chlorite WAR-1924 40
21Diopside WAR-5780 165
22Dolomite C28 128
23Epidote BUR-1940 19
24Fayalite WAR-RGFAY01 167
25Fe66Mg34CO3 C56 134
26Ferrohornblende HS-326.4B 58
27Fluorite BUR-2080C 59
28Forsterite BUR-3720A 8
29Glaucophane WAR-0219 17
30Gypsum ML-S5 80
31Hectorite SCHa-1 granular 192
32Hematite BUR-2600 50
33Hornblende BUR-1380A 65
34Hypersthene NMNH-B18247 12
35Illite IMt-2 granular 211
36Ilmenite WAR-4119 35
37Jadeite WAR-9909 47
38Kaolinite KGa-1b granular 185
39Kyanite WAR-4482 179
40Magnesiohastingsite HS-115.4B 3
41Magnesite C60 139
42Malachite C13 108
43Muscovite WAR-5474 20b
44Nontronite WAR-5108 granular 203
45Oligoclase WAR-0234 22
46Palygorskite PF1-1 granular 209
47Quartz BUR-4120 55
48Saponite ASU-SAP01 granular 194
49Serpentine HS-8.4B 14
50Spodumene HS-210.4B 2
51Talc BUR-4640C 53
52Tremolite var. jade (Nephrite) WAR-0979 23
53Wollastonite BUR-5080 60

[36] Even with 52 end-members, linear matrix inversion is not a straightforward process. A direct unmixing using all of the end-members may yield negative fractional values for some of the end-member spectra, which would be physically unreasonable. Typically, the inversion process is iteratively run, removing the end-members with negative fractional values at the end of each iteration, until only end-members with positive fractional values remain [e.g., Bandfield, 2002]. Final fractional values for each mineral are then reported as the normalized values after removal of the blackbody's contribution.

[37] The goodness-of-fit of the model to the data is usually quantified by the root-mean square (RMS). However, RMS values may not provide the most intuitive assessment of the quality of a spectral fit. We therefore define a parameter, the error value,

equation image

where M is the model spectrum, O is the observed spectrum, σ is the 1-standard deviation error spectrum, and n is the number of points in the spectrum. The definition of this value is similar to chi-squared in statistics, but it allows us to understand how many standard deviations (on average) our model deviates from the observed spectrum. Ideally, a good fit is defined as one where the error value equals 1 or less. In reality, however, the “noise” induced by the water vapor absorptions in the 200–400 cm−1 region restricts the error parameter to a minimum of ∼1.5 (corresponding to an RMS of ∼0.05; see Table 3).

Table 3. Deconvolution Results: Using All the End-Members in Table 1, Leaving Obsidian Out, and Leaving Kaolinite Outa
End-MemberGroupNormalized Fractional Abundance
AllNo ObsidianNo Kaolinite
  • a

    Places with significant differences (several percent) in results are given in boldface for clarity.

Quartz BUR-4120 55SiO0.0187
Andesine WAR-0024 175feldspar0.2105
Augite BUR-620 71pyroxene0.07820.1214
Bronzite NMNH-93527 168pyroxene0.11730.1314
Epidote BUR-1940 19epidote0.0495
Forsterite BUR-3720A 8olivine0.07480.08060.0561
Hornblende BUR-1380A 65amphibole0.02310.0088
Magnesiohastingsite HS-115.4B 3amphibole0.00970.0986
Hematite BUR-2600 50oxide0.0453
Ilmenite WAR-4119 35oxide0.0136
Jadeite WAR-9909 47jade0.17350.19430.0995
Muscovite WAR-5474 20mica0.06120.05820.0092
Anhydrite ML-S9 81sulfate0.01750.0205
Gypsum ML-S5 80sulfate0.08560.07290.0326
Nontronite WAR-5108 granular 203clay0.0711
Ca-montmorillonite STx-1 granular 196clay0.2551
Calcite C22 116carbonate0.13730.15320.0480
Dolomite C28 128carbonate0.0799
Magnesite C60 139carbonate0.09680.0964
Serpentine HS-8.4B 14serpentine0.07550.0238
RMS 0.0049880.0050634970.0048
Mean error, ɛ (equation (5)) 1.4886631.52902921.3630

[38] This unmixing approach is very sensitive to the choice of the original end-members, and separate mixtures whose original libraries vary by only the presence of even an “insignificant” end-member can yield different final results. As an example, we performed three unmixing experiments: In the first one, we used every mineral in the reduced library, and in the second and third experiments, we removed the obsidian and kaolinite end-members, respectively. Table 3 presents the results of this experiment. Although neither obsidian nor kaolinite belong to the final mineral mix, the inclusion (or not) of these components in the library significantly affects the final derived mineral suite without significantly affecting the model's fit (Figure 16). This is particularly evident in the derived fractional abundances of jadeite, clays (e.g., Ca-montmorillonite), and feldspars (e.g., andesine). Apparently, the best-fit mathematical solution to the spectral matrix inversion problem exhibits a large number of “local minima” of comparable best-fit values, and the initial library choice acts somewhat like a random seed variable in the fitting process. Thus spectral unmixing methods relying on such an approach must be interpreted with caution.

Figure 16.

Sample fits with mineral fractional abundances described in Table 2. Note that although the final fractional abundances can be significantly different, all three fits can be considered very good (RMS around 0.005).

[39] Ideally, we would like to run the matrix inversion routine over every possible combination of all the available end-members, and then identify the recurring mineral fractions that yield the global minimum best-fit model. However, such a process requires 2n deconvolutions, where n is the number of end-member spectra, and it is therefore computationally unfeasible for large values of n. An alternative method, based in part on the method of Seelos and Arvidson [2003], runs the standard procedure described above “n” times (where n is the number of spectra in our library). In each run, one end-member spectrum is left out of the original suite, and the procedure is run with n-1 end-members. This results in n tables of calculated end-member weights, which are then compared, and all the end-members that are found to play a significant part in the mix (i.e., they have a fractional value of 8% or greater in any of the runs; we chose 8% as the threshold in order to use the maximum number of end-members while at the same time keeping computing times reasonable) are added to a new library. This resulted in a total of 27 end-members for our case study, which we then ran through the SVD routine using every possible combination. In our final fits, we only consider the models that involve 15 or fewer end-members and which fit the data within root mean square (RMS) values of 0.005 (error parameter ɛ ∼ 1.5) or less. Even with these restrictions, we end up with a cumbersome but statistically interpretable total of 184,551 candidate models.

[40] The detectability of each end-member depends on the intensity of its absorption bands or other diagnostic spectral features. We define the detectability limit of an end-member as the minimum fractional value required for it to be distinguishable over a 3-σ noise level (defined by the noise in the data) when it is linearly mixed with the spectrum of a blackbody. To do this, we generated a set of emissivity models for each mineral by linearly mixing its spectrum at different abundances with that of a blackbody. We then measured the “band depth” (emissivity outside the band minus emissivity inside the band). A band was defined to be detectable if its “band depth” was at least 3 times greater than the 1-σ of the noise. We therefore determined the detectability of each of the 27 new library end-members by linearly mixing its spectrum in ever-decreasing quantities with the spectrum of a blackbody until its most prominent features fell within 3-σ of the noise (Table 4).

Table 4. Mineral End-Member List After Final Selection
NumberMineral3-σ Detectability Limit, %
2Fayalite WAR-RGFAY01 1675
3Augite BUR-620 719
4Bronzite NMNH-93527 1689
5Diopside WAR-5780 1657
6Bytownite WAR-1384 1778
7Andesine WAR-0024 1759
8Epidote BUR-1940 196
9Fluorite BUR-2080C 592
10Apatite ML-P1 863
11Magnesiohastingsite HS-115.4B 35
12Glaucophane WAR-0219 175
13Hornblende BUR-1380A 654
14Jadeite WAR-9909 476
15Kyanite WAR-4482 1795
16Ilmenite WAR-4119 354
17Hematite BUR-2600 505
18Biotite BUR-840 253
19Muscovite WAR-5474 203
20Ca-montmorillonite STx-1 granular 1967
21Hectorite SCHa-1 granular 19211
22Nontronite WAR-5108 granular 2037
23Calcite C22 1163
24Fe66Mg34CO3 C56 1343
25Magnesite C60 1393
26Dolomite C28 1283
27Gypsum ML-S5 804

4.3.4. Results

[41] Figure 17a summarizes fractional abundances found for the new 27-end-member mineral library using our method. For each end-member (indexed in the x axis), we plot the mean of all the fractional values calculated for that end-member (asterisk), the 1-σ variance of the fractional value around the mean (red error bars), and the maximum and minimum fractional values used for this end-member (black error bars). In addition to the fractional value of each end-member, Figure 17a also presents a histogram (red) of the occurrence of each end-member in all our models.

Figure 17a.

a: Statistical fractional abundance/occurrence plot for the final selection of end-members for all models with fits better than 0.005 RMS. The mean (asterisk), variance (red error bars), and maximum/minimum (black error bars) are calculated for all the available good-fit models (184,551). The histogram shows the relative occurrence of each mineral in the models.

[42] Such a statistical approach to unmixing [Seelos and Arvidson, 2003] allows us to identify the importance of each end-member and its most likely abundance in the mixture. For example, gypsum (Figure 17a, index 27) appears in over 95% of the mineral mixtures. However, its mean fractional abundance is only 7%, just above its detectability limit. Another example is the detection of feldspar minerals such as andesine (index 7). This mineral is included in over 50% of the low RMS models, and has a relatively high mean fractional abundance (>30%). We therefore infer that feldspar is likely making a significant contribution to the derived surface spectrum. In contrast to these two cases, fayalite (index 2) only contributes to a very small fraction of the low RMS models (<10%), and its abundance in these models is usually lower than its detectability limit. We can therefore infer that fayalite is not detectable in the dark deposit spectrum. Following this approach for other library components modeled in Figures 17a and 17b, we conclude that the most significant spectral end-members in these mixtures are feldspars, jadeite (NaAlSi2O6), Ca-montmorillonite and nontronite clays, carbonates along the siderite-magnesite solid solution series, gypsum, and possibly to a lesser extent, pyroxenes.

Figure 17b.

b: Related statistical association plots (see Figure 17a). Note how andesine and bytownite do not tend to occur in the same mixtures, but the fractional abundances and relative occurrence are constant for the other minerals.

[43] Because different end-member combinations may result in models of similarly good fits, it is necessary to determine the statistically most likely combinations, and whether or not they are mutually exclusive. Figure 17b shows histograms of the relative occurrence of other end-members in models containing the end-members that we have identified as important. Among the feldspars, we find that bytownite and andesine tend to be mutually exclusive: very few models containing bytownite also contain andesine, and vice versa. On the other hand, both the occurrence and mean abundance of the other end-members remain fairly constant: pyroxenes, jadeite, iron/magnesium carbonates (specifically Fe66Mg34CO3), and gypsum appear in well over 50% of the mineral mixtures involving any of the other important end-members, with mean abundances roughly independent of the end-member we chose to focus on. This means that although andesine and bytownite tend to replace for each other in the models, the abundances and occurrence of the other end-members remain fairly constant.

[44] The spectral exchangeability of andesine with bytownite can be understood by comparing their spectra (Figure 18). Both minerals have fairly similar spectra in the 800–1300 cm−1 region, and originally we had removed labradorite from the original library because a mixture of andesine and bytownite can roughly model its spectrum. The above fits indicate that we could have also removed bytownite from the mix because the spectral shape of andesine is so similar.

Figure 18.

Comparison spectra of the three feldspars in the library. Note that although these two minerals present some spectral differences, statistical analysis indicates that they can be used interchangeably.

[45] Figure 19 presents the derived TES surface emissivity spectrum of the dark deposit along with the weighted spectra of the minerals that have mean fractional values above their detectability limit (Figures 17a and b). Most of the absorptions can be directly related to the presence of specific minerals. For example, the large wave number side of the 900–1300 cm−1 absorption, along with its kink at 1150 cm−1 can be broadly associated with either jadeite or a combination of feldspars (e.g., andesine) and gypsum. Similarly, the spectral slope at 200–400 cm−1 can be associated with either carbonates (e.g., iron-magnesium carbonates) or feldspars (e.g., andesine). To constrain further the uniqueness of each mineral group in the mixture, we combined the minerals from our library into mineral groups and compared their relative occurrences (Figure 20). Several mineral groups appear in nearly all of the fits, including pyroxenes, carbonates, and gypsum. Most other mineral groups are used in fewer than 60% of the models, so their presence cannot be considered to be as statistically unique. We therefore conclude that our most typical, low RMS model of the TES dark deposit spectrum consists of 20–40% feldspars, 15–20% jadeite, 10% carbonates, 15–20% clays, 7% gypsum, and possibly pyroxenes at their detectability limit (∼10%).

Figure 19.

Comparison of the derived TES emissivity spectrum of the dark deposit to the most relevant minerals in the library (identified from Figure 16).

Figure 20.

Fractional occurrence of each mineral group in all the good fitting models. Note that both carbonates and pyroxenes occur in 100% of the models, whereas olivines occur in very few of the models.

[46] The presence of carbonate deposits on Mars has strong implications in regards to the climatic history of the planet, and its putative identification has to be addressed with caution. Carbonates display two primary absorption features in the midinfrared caused by the bending and stretching modes of the carbonate lattice. These occur at about 315 and 1520 cm−1. Our unmixing method requires the presence of carbonates to explain the steep slope observed in the 200–300 cm−1 region of the spectrum. However, this same slope is observed in spectra of many surrounding regions (Figure 14b), bringing up the possibility that we may be dealing with some kind of not-well-modeled mixing effect or a systematic error. It is therefore important to search the high wave number region of the spectrum for the 1520 cm−1 absorption known to exist in carbonates.

[47] TES spectra that have been atmospherically corrected using the method of Bandfield et al. [2000] only extend to 1300 cm−1 and can therefore not be used to search for the absorption at 1520 cm−1. We therefore atmospherically correct the rest of the spectral range using the radiative transfer algorithm of Wolff and Clancy [2003], modified to derive surface emissivity [Noe Dobrea, 2004]. The algorithm, as described by Wolff and Clancy [2003], can be used to model the absorptive and scattering effects of the different components of the Martian atmosphere on a surface spectrum. Both the molecular opacities of CO2 and water vapor, as well as the multiple scattering effects caused by dust and water ice aerosols are considered. The optical constants of dust were obtained from M. Wolff (personal communication, 2003), and are the same as those used by Wolff and Clancy [2003]. The optical constants for hexagonal water ice were obtained from Warren [1982]. The aerosol particle sizes were assumed to follow a modified gamma distribution with reff ∼ 1.5 and veff ∼ 0.5 for dust and reff ∼ 1.5 and veff ∼ 0.1 for water ice, although variations in the reff of the water ice out to 2 μm were also considered and found to fit the spectrum if the water vapor and water ice abundances were in turn decreased. The surface emissivity spectrum derived previously using the method of Bandfield et al. [2000] was used as the model's initial surface spectrum. The abundances of the water vapor, water ice aerosol, and dust aerosol were derived by manually varying the value of the surface temperature as well as the abundances of water ice and dust aerosols, CO2, and water vapor until a minimum chi-squared was found in the 250–1300 cm−1 region. The resulting best fit is shown in Figure 21a. Once the abundances of the atmospheric components were found, they were in turn used to iteratively derive the surface emissivity spectrum for the entire TES spectral range (Figure 21b). Analysis of the derived surface emissivity spectrum shows no absorption around 1520 cm−1 above the level of the noise. On the basis of this analysis, we cannot confidently determine that the signature of carbonates is present in our spectrum.

Figure 21.

(a) Comparison of modeled spectrum to derived brightness temperature spectrum of the region of interest. The error bars for this spectrum are separated plotted along the 302 K region for clarity. (b) Comparison of the derived surface emissivity to the emissivity spectrum of a MgFe carbonate (plotted at 10% abundance). Note the strong slope in the 200–300 cm−1 that was identified by the unmixing algorithm as part of a carbonate band. Noise at large wave numbers hinders the identification of a possible second absorption at ∼1520 cm−1.

5. Discussion

[48] Our morphologic and spectroscopic analyses show no obvious association between the brighter fan-shaped deposit and the dark intracrater dune deposit. The former appears to have been emplaced through multiple flow episodes, whereas the latter has a morphology consistent with a more recent, possibly aeolian emplacement: It consists of an uncratered thick dune-forming mantle inside the crater and a much thinner veneer that extends up the crater's side and onto the surrounding plains. This morphology is similar to that of other dark intracrater deposits [e.g., Edgett, 2002]. There is no morphologic evidence that the dark material forming the dune-forming deposit is being shed from the fan-shaped deposit. The yield strength that was estimated from the fan-shaped deposit's morphometry indicates that it was emplaced either as a volcanic flow or mass movement (i.e., landslide) of relatively high yield strength. The similar volumes measured for the fan-shaped deposit and its source valley support the idea that the fan shaped material is composed of crustal material whose flow was facilitated by a lubricating agent such as water or ice.

[49] In addition to the morphological differences, THEMIS/VIS multispectral images indicate that the fan-shaped deposit and the dune-forming portion of the dark intracrater deposit have different colors, and that their mixing with bright plains material can adequately explain the observed spectral variability in the scene. Minimum noise fraction analysis of the THEMIS/IR multispectral image cube also shows that these two deposits are not spectrally associated at long wavelengths, and that the dune-forming deposit contains an additional absorption band that can be associated with some unaltered volcanic products (e.g., olivine, garnet, and/or pyroxene).

[50] TES spectroscopy of the dark deposit itself hints at a varied mineralogy involving both unaltered minerals and their alteration products. The primary constituent interpreted to statistically dominate the best-fit TES spectrum is feldspar, with smaller abundances of clays, jadeite, carbonates, clays, and possibly pyroxenes appearing in most of the best models. With the exclusion of jadeite, this mineral assemblage hints at a regolith that has undergone varying degrees of alteration: both pyroxene and feldspar are primary minerals formed from igneous processes. Carbonates are usually interpreted as aqueous alteration products derived either as surface rinds from direct interaction between the rocks and the Martian atmosphere, or as sedimentary deposits derived from dissolution and subsequent precipitation of CO2 in bodies of water [e.g., Gooding, 1978]. Clays and sulfate minerals are also typically formed as the end product of aqueous alteration by either surface-atmospheric interaction, or by direct contact of primary minerals with liquid water [e.g., Gooding and Keil, 1978; Burns, 1993].

[51] The possible presence of the mineral jadeite as a major constituent of the mineralogy of the dark deposit is difficult to explain geologically. Jadeite is a sodium-aluminum pyroxene that forms on Earth exclusively through the high-pressure (10 to 20 kbars) and relatively low temperature (600° to 1000°C) metamorphism of sodium-rich rocks [e.g., Deer et al., 1966]. On the Earth, such occurrences are found near the margins of the continental crust, such as the Alps, California, and Japan [Klein and Hurlbut, 1999]. Given the lack of compelling evidence for plate tectonics and earthlike continental crust on Mars, it is difficult to envision a scenario that facilitates the formation of a mineral like jadeite in the ancient Noachian terrains.

[52] While jadeite contributes significantly to roughly 65% of the good-fit models, there are also other good fits that do not require the presence of jadeite. For example, we find that 35% of the models that exhibit good fits to the data without using the jadeite end-member use slightly higher average fractional values of clays, feldspar, and gypsum, and slightly lower fractional values of pyroxenes to fit the data. Therefore the identification of jadeite is not necessarily unique. Indeed, in general, geologic/geochemical arguments should be considered when analyzing the results of purely statistical solutions to spectral unmixing scenarios like those described here.

[53] As a sanity check, we applied the same lines of reasoning to other mineral groups that are not used in 100% of the best-fitting mineral models. We find that models that do not contain feldspar use, on the average, a slightly higher fractional abundance of clays, and a significantly (∼5%) higher fractional abundance of aluminous silicates (e.g., kyanite) and sulfates (e.g., gypsum). Kyanite, however, is another metamorphic mineral whose presence on Mars is inconsistent with our current understanding of Martian geophysical processes, and we therefore favor the use of feldspar in our mixtures.

[54] Whereas clays exhibit a detectable fractional abundance in many of our models, their contribution to the shape of the TES dark deposit spectrum is not necessarily unique. Best-fitting models that do not use clays have slightly higher average fractional abundances of feldspar and amphiboles. The mean amphibole fractional values are still below the TES detectability limit, and the small average increase in the fractional abundance of feldspars is still consistent with our range of previous feldspar modeling results. Therefore we cannot claim a high degree of confidence in the detection of clays in the dark deposit spectrum.

[55] Both pyroxenes and sulfates (gypsum) are present in virtually every best-fit model. However, their fractional abundances in these models are very close to their detectability limits. Again, we cannot have a high degree of confidence in the detection of these minerals, although their spectra are used in virtually every model to account for small-scale features (possibly noise) in the TES spectrum (e.g., the 1150 cm−1 kink; Figure 19).

[56] Finally, carbonate minerals have a small and yet significant fractional value in our best-fit spectral models. Although the fractional values tend to be close to 10%, this is well above our calculated detectability limit, and it is clear from Figure 19 that relatively small fractional values of carbonates such as those in the siderite-magnesite solid solution series appear to contribute significantly to the shape of the spectrum at low wave numbers. In addition to this, there is not a single best-fit model among more than 105 that does not use carbonates. This gives us a high confidence level for a positive detection of carbonate minerals in the dark deposit spectrum at a fractional value of roughly 10%. However, this detectability is allayed by an increase in both multiple scattering (i.e., nonlinear effects) and noise at low wave numbers and as discussed before, a unique detection is also frustrated by our inability to detect a second absorption at 1520 cm1.

[57] We therefore conclude from this analysis that the dark deposit consists primarily of feldspars and possibly pyroxenes, perhaps mixed with some alteration products such as carbonates and possibly clays and sulfates. Neither jadeite nor metamorphic aluminous silicates such as kyanite provide a unique solution to our fit, and since they are geologically inconsistent with our current understanding of Martian history, we cannot argue strongly for their presence in our final mixture model solution.

[58] TES' large field of view makes it impossible to separate the spectrum of the fan-shaped deposit from that of the dark mantling material, so it is not possible to identify the source of each of the minerals we have identified in the TES spectrum. However, previous TES studies of the entire Aeolis region indicate that this region is predominantly basaltic [Bandfield et al., 2000] and its spectral signature is dominated by plagioclase feldspar and clinopyroxene. Because of the mobility of the mantling deposit, it is likely that it has a similar composition to the rest of the material in the region. The fan-shaped deposit, on the other hand, was clearly formed by some sort of flow process that may have involved either liquid water or ice. If so, the presence of water in the subsurface could have led to the chemical decomposition of primary minerals in the subsurface and the formation of carbonates, sulfates, and other putative alteration products, which were later exposed in the fan-shaped deposit.

6. Conclusions

[59] We have carried out a multispectral geomorphic and spectroscopic analysis of a spectrally anomalous region in the Noachian terrain of Mars. This region is characterized by a dark intracrater deposit and an adjacent fan-shaped deposit, both of which are “bluer” than the surrounding terrain. The fan shaped deposit sits at the mouth of a short (1 km) valley that incises into the southern side of an old degraded crater. Its morphology is most consistent with a flow formation. And calculations of its yield strength, based of morphometry of the lobes, indicate yield strengths consistent with those of other volcanic flows on Mars, although an aqueous origin cannot be discounted. The dark deposit, adjacent to the fan-shaped deposit, is composed of a mantling material that forms dunes at its thickest and thins out toward the center of the crater and beyond the southern rim. No morphologic association can be identified between these two deposits. Multispectral imaging using both THEMIS/VIS and IR show that the two deposits have different spectral characteristics, and THEMIS/IR spectral ratios hint that the dark deposit contains a higher concentration of primary volcanic minerals.

[60] Constraints on the mineralogy of the dark deposit were derived by a statistical unmixing analysis of MGS/TES spectra involving a large number of best-fitting spectra derived from a modest-sized, statistically derived subset of the ASU mineral spectral library. Using this method, we found that the mineralogy of the dark deposit is most consistent with a feldspar-rich material and possibly pyroxenes, perhaps complemented by smaller quantities of alteration products such as carbonates, sulfates, and clays. This type of mineral suite is consistent with an ancient Martian volcanic surface that may have been exposed to aqueous alteration processes.

[61] The formation of aqueous alteration products in the Noachian necessitates a more clement environment where liquid water can exist in relatively close proximity to the surface. Increasing amount of evidence for aqueous alteration in the Martian past has been found in the past couple of years, including the discovery of sulfate-rich light-toned outcrops in Meridiani [e.g., Squyres et al., 2004; Arvidson et al., 2005] and Valles Marineris [Gendrin et al., 2005; Langevin et al., 2005; Bibring et al., 2005], as well as the discovery of clays in Syrtis Major [e.g., Bibring et al., 2005; Poulet et al., 2005; Loizeau et al., 2005] and possibly the outcrops above Mawrth Valles. Therefore the identification of another aqueously derived process on the surface of Mars is consistent with contemporaneous findings and hints at past atmospheric conditions that were more amenable to the presence of liquid water.


[62] We thank Tim Glotch for providing us with the spectrum of goethite. This work was supported by a grant from the NASA Mars Data Analysis Program (NAG5-13586).