Hematite-bearing deposits in the Terra Meridiani region of Mars constitute the top stratum of a partially eroded layered complex that covers dissected Noachian-aged cratered terrain. The hematite unit consists of dark plains and dunes covering a bright substrate. This substrate is fully exposed beyond the borders of the hematite-bearing deposit and consists of polygonal ground separated by ridges or valleys, together with layered deposits that have been eroded into a variety of landforms. The complex is partially covered by a regional-scale aeolian mantle that thickens toward the north. The hematite-bearing stratum exhibits low albedoes, pulse widths, and intermediate thermal inertias, whereas the underlying unit exhibits high values of these parameters. Both units have spectral emissivity signatures similar to those for the low albedo cratered terrain to the south, with the addition of hematite for the top stratum. The complex is interpreted to consist of extensive plains-forming lava flows and tephra deposits emplaced during an extensional regime and at least partially buried by an aeolian mantle. Aeolian stripping of the mantle exposed much of the complex and differentially eroded the deposits to produce the landforms existent today. Exploration of the hematite-bearing deposits by the 2003 Mars Exploration Rover, “Opportunity,” will allow testing of the hypotheses presented since this stratum has been locally reworked into dunes that only partially cover the underlying brighter portion of the complex. In particular, the rover-based measurements will allow us to test the extent to which the unusual remote-sensing properties of the units indicate aqueous alteration.
 The purpose of this paper is to summarize analyses of Mars Global Surveyor MOC [Malin and Edgett, 2001], MOLA [Smith et al., 2001], and TES [Christensen et al., 2001a] observations to identify, characterize, map, and understand the origin and evolution of the hematite-bearing deposits [Christensen et al., 2000a, 2001a] and associated units in the Terra Meridiani region of Mars (Figures 1, 2a, and 2b). The approach is to use MOC wide-angle and narrow-angle images (WA and NA, respectively), together with elevation estimates from MOLA data, to define and map units on the basis of morphology and superposition relationships. Surface properties of these units are then characterized using MOLA-based intrashot pulse widths [Garvin et al., 1999] and TES-based albedoes, thermal inertias [Mellon et al., 2001], and spectral emissivities [Smith et al., 2000]. The data indicate that: (a) the hematite-bearing unit is the remnant of the top stratum of a widespread layered complex that was deposited onto dissected cratered terrain, (b) the complex was covered at least in the northern portions of Terra Meridiani by a widespread sedimentary mantle, (c) the entire region has been subjected to differential aeolian erosion that has stripped the mantle, exposing underlying materials that in turn have been partially eroded by wind, and (d) the layered complex was emplaced as flows and tephra deposits, and the unusual albedo and spectral properties are consistent with alteration involving aqueous fluids, either during or after emplacement.
2. Definition of Geologic Units and Stratigraphy Based on Morphology, Topography, and Hematite Index
 This work builds on initial mapping of the Terra Meridiani study area (5 degrees south to 10 degrees north latitude and 20 degrees in longitude centered on the prime meridian) by Hynek et al. . It extends that mapping by incorporating all NA images within the study area acquired as part of the detailed survey of the planet [Malin and Edgett, 2001] and frames targeted on the MER landing sites located on the hematite-bearing deposits discovered by Christensen et al. [2000b, 2001b]. The study area boundaries were defined to include the hematite-bearing deposits and the borders of the underlying units within the layered units mapped by Hynek et al.  (Figure 1).
 The first step in the work was the construction of a digital base map from 32 pixel/degree MOLA elevation data processed to remove regional tilts and thus to highlight local relief. The WA red mosaic compiled by Caplinger and Malin  at 256 pixels/degree was co-registered to the MOLA map and the data sets merged by overlaying color-coded elevations onto the WA mosaic. Map units were defined from this product at a scale of 1:1,000,000, aided by examination of individual MOLA elevation profiles and NA frames located within the study area. NA frames were processed to Level 2 (i.e., in Universal Transverse Mercator format with units proportional to radiance on sensor) using the USGS Integrated Software for Imagers and Spectrometers (ISIS) software in order to place calibrated NA data in the merged WA and MOLA spatial framework. The MOLA, WA, and NA data were thus used in combination to identify and map the units on the basis of planimetric configuration, topography, brightness, and texture and to infer superposition and embayment relationships (Figures 2a and 2b). Further, the boundary of the hematite-bearing deposits was delineated using the hematite index threshold as defined by Christensen et al. [2001a] (Figures 2a and 2b).
 Regionally, the study area is within the Noachian-aged heavily cratered terrain in the northern part of Terra Meridiani (Figure 1). The lithosphere within the study area tilts downward to the northwest and has been postulated to have undergone flexure during the Noachian Epoch by the load associated with the formation of the Tharsis volcanic complex [Phillips et al., 2001]. The presence of numerous channels and topographic outliers exposed on a regional scale suggests dissection of the Noachian cratered terrain by fluvial activity immediately after tilting, perhaps due to warm and wet conditions induced by extensive degassing associated with early Tharsis-related volcanism [Hynek and Phillips, 2001]. The layered complex was then disconformably emplaced over the dissected cratered terrain, most likely during late Noachian or early Hesperian time [Hynek et al., 2002].
 The basal unit in the study area is the Noachian cratered terrain that has been extensively dissected by channel systems and is termed the dissected cratered terrain or Unit DCT (Figures 2a and 2b). The basal unit of the layered complex was termed the etched unit by Hynek et al.  and we retain that nomenclature here as Unit E. On the basis of detailed analysis of WA and NA data, the morphology of Unit E is variable and includes extensive exposures of terrain characterized by: (a) relatively flat, dark polygonal blocks separated by bright, fractured ridges and occasionally by valleys (Figure 3); (b) interconnected ridges and stepped plateaus that demarcate underlying plains into polygonal patterns (Figure 4); and (c) landforms that consist of buttes, mesas, and what appear to be yardangs, often exposing layered deposits on cliff faces and slopes (Figures 5a and 5b). Analysis of superposition relationships shows that polygonal ground can be found both stratigraphically below and above the layered deposits, depending on the region under consideration.
 The buttes, mesas, and yardang-like features evident in Unit E demonstrate that significant aeolian erosion has occurred, stripping up to several hundred meters from this deposit. For example, sections exposed on the northeastern boundary of Unit E show a transition from well-preserved polygonal ground within Unit E to partially eroded remnants of polygonal ground toward the edge of the deposits, to exposures of underlying layered deposits that have been scoured by wind at the contact with Unit DCT (Figures 5a and 5b). The eroded landforms, dunes, and streaks indicate that easterly winds were responsible for the scouring of the deposits.
 Detailed analyses of WA and NA images, MOLA gridded elevation data, and individual MOLA profiles registered to NA frames show that the hematite-bearing plains unit (Ph) is stratigraphically above the Unit E. NA data show that Unit Ph consists of smooth, dark plains that are locally reworked into dunes and that in many places only partially cover Unit E. This relationship is particularly well demonstrated at the northeastern boundary between Units Ph and E (Figures 6a and 6b). In that location the northern edge of Unit Ph corresponds to the edge of a rolling plateau. To the north of the plateau edge Unit E extends as a mappable unit for tens of kilometers, exposing several sections of layered deposits. Moving south from the Ph-E boundary it is clear that the inter-dune areas are exposures of Unit E (Figures 6a and 6b). This same pattern of dark deposits covering brighter materials is also evident within the hematite-bearing deposits located within the MER landing error ellipse (Figure 7). Finally, the northwestern edge of the Ph unit seems to transition to another plains-forming surface, herein termed Unit P (Figures 2a and 2b). Stratigraphic relationships are hard to define in detail because of a paucity of NA coverage. On the basis of available evidence we believe that Unit P is a lateral facies variation and that both deposits overlie Unit E.
 Units DCT, E, P, and Ph are covered by a mantle in the northern portion of the study area that has been partially stripped by aeolian processes to expose underlying materials. The mantled unit largely occurs on Noachian cratered terrain and is mapped as mantled cratered terrain unit or Unit MCT (Figures 2a and 2b). Examination of WA and NA images covering the boundary between the Units E and MCT shows that the characteristics associated with Unit E landforms become muted and even disappear beneath the mantle that defines Unit MCT (e.g., Figures 4 and 8). South of the contact, ridges and conical hills associated with Unit E are sharply defined, with flanks and walls that have distinct intersections with the surrounding terrain (Figure 4). Just to the north of the boundary only the ridges and tall conical hills associated with the E unit are exposed (Figure 8). The thickness of the mantle can be inferred by examination of individual MOLA elevation profiles and comparison of the exposed and mantled portions of the E Unit. For example, in MOC NA Frame M0401901 the ridges are approximately 20–30 m above the plains to the south of the boundary whereas to the north of the boundary only the tops of the ridges are exposed (Figures 4 and 8). We estimate a mantle depth of approximately 10 m near the boundary, thickening to the north.
3. Geologic Unit Albedo, Thermal Inertia, and Pulse Width Properties
 On a global basis there is a well-known negative correlation between albedo and thermal inertia, consistent with the dominance of: (a) bright, low thermal inertia dust-covered areas, (b) dark, high thermal inertia regions with exposed sands, and (c) an intermediate surface thought to contain weakly indurated materials, i.e., duricrust [Mellon et al., 2001]. On a regional scale, the signature associated with Arabia Terra (north of study area, Figure 1) exhibits low thermal inertia and high albedo values indicative of dusty surfaces, whereas the dark region, Sinus Meridiani (northern portion is in study area), has low albedo and high thermal inertia values, and the Oxia Palus region (northwest of study area) has intermediate values falling into the zone interpreted to be due to duricrust exposures. With regard to the study site, Unit DCT falls within the globally defined low albedo, high thermal inertia cluster, whereas the Unit MCT falls within the duricrust cluster (Table 1 and Figure 9). Both plains units, P and Ph fall within the dark, high inertia cluster. On the other hand, Unit E has a distinctly high inertia and intermediate albedo that makes it stand out from the clusters defined on a global and even regional basis.
Table 1. Statistics for Units
Thermal Inertia, J/(m2 K sqrt (s))
Corrected Pulse Width, m
MCT: Mantled Cratered Terrain
0.23 ± 0.01
227 ± 53
1.34 ± 0.83
Ph: Hematite Plains
0.15 ± 0.02
211 ± 27
0.68 ± 0.35
0.16 ± 0.02
232 ± 20
0.88 ± 0.60
0.19 ± 0.02
361 ± 77
1.35 ± 0.84
DCT: Dissected Cratered Terrain
0.15 ± 0.03
256 ± 40
0.81 ± 0.68
 In addition to its unusual combination of thermal inertia and albedo, Unit E is also distinct in that the unit boundaries mapped on the basis of topography and morphology also correspond well to the boundaries between relatively high and low thermal inertias areas (Figure 10). In addition, the TES-based bolometric albedo and MOLA-based pulse widths both show a high degree of spatial correlation with the distribution of Unit E, as does the spectral reflectance (0.4 to 1.0 micrometers) derived from Hubble Space Telescope data [Morris et al., 2002]. In particular, the spectral data indicate that Unit E is bright and red relative to Ph and P, although not as bright as Unit MCT and areas to the north. The high degree of spatial correlation between mapped locations of Unit E and remote-sensing properties is unusual on Mars, given the extent to which aeolian activity has reworked the surface. We suggest that Unit E color, albedo, and thermal inertia are in fact representative of the mineralogy, texture, and degree of induration of materials that comprise this unit. In particular, results are consistent with an abundance of moderately consolidated materials that are bright and red as compared to dark areas on Mars.
 MOLA data can be used to infer terrain roughness by examining the extent to which the return pulse is broadened relative to the transmitted pulse [Garvin et al., 1999]. In our case we have corrected the pulse spread for regional scale slopes, thus providing a measure of roughness at scales equal to or less than the MOLA footprint (150 m). Units P and Ph have the lowest MOLA corrected pulse width values, followed by Unit DCT (Figure 1 and Table 1), whereas the two remaining units, E and MCT, both have relatively high values. We offer two explanations for these trends. First, Unit MCT might retain a relatively rough topography associated with deposition and subsequent aeolian reworking of the accumulating materials. Second, Unit DCT may have been graded or beveled by aeolian erosion or volcanism and is thus smoother than the mantled areas. The rough signature associated with Unit E is interpreted to be due to differential aeolian erosion and consequent rough topography whereas the smooth signature associated with Units Ph and P are interpreted to be a consequence of deposition on an originally flat or rolling surface that has been only locally reworked by aeolian processes.
4. Geologic Unit Spectral Emissivity Properties
 In this section we report on analyses of TES spectral radiance data for the four major mapped units in the study area. We note that with the exception of an abundance of hematite, Units Ph and P are indistinguishable and thus Unit P results will not be presented. The intent of our analysis was to evaluate the extent to which unit mineralogy can be inferred and used to understand the origin and evolution of the layered complex. As a first step in the analysis the TES spectral database provided through NASA's Planetary Data System was searched to find single-scan, dayside, low atmospheric opacity (optical depth less than 0.2), nadir observations without significant microphonic or other noise components. A total of 127,035 individual spectra passed the constraints and were extracted from the PDS archive. Each radiance observation was then converted to an apparent emissivity spectrum by dividing the observed radiance values by the values computed for a blackbody radiator with a kinetic temperature equal to the surface temperature reported in the TES standard product tables. Visual inspection suggests that apparent emissivity spectra for Units Ph, E, and DCT are similar, although the Ph spectrum shows unique and strong evidence for hematite-related absorptions in the 300 to 500 cm−1 wave number region. On the other hand, the MCT spectrum has an overall higher apparent emissivity and a distinctly lower slope in the 800 to 1000 cm−1 spectral region as compared to the other spectra.
 The apparent emissivity spectral patterns are a combination of surface and atmospheric contributions. We have followed the method of Smith et al.  in constructing a linear least squares unmixing approach for the separation of mean spectra for each mapped unit into atmospheric and surface spectral components (Appendix A). The surface spectra were recovered as a linear combination of 48 mineral end-members judged to be important components in prior surface mineralogy retrievals using TES data [e.g., Bandfield et al., 2000; Wyatt and McSween, 2002]. The retrieved surface emissivity spectra confirm that Unit MCT is distinctly different in spectral properties as compared to the other three units (Figure 11). On the other hand, Units E, Ph, and DCT are very similar, except for the hematite-related absorptions in the Ph spectrum. These similarities have been quantified by computing the spectral angles between the various pairs of surface emissivity spectra. The spectral angle is related to the dot product of two spectra in channel space, such that an angle of zero radians indicates that the two spectra are identical and a large angle indicates significant variation from one spectrum to the other. The angle calculations demonstrate that Units DCT and E have the greatest affinity, followed by Units DCT and Ph (Table 2). Units MCT and Ph are the most dissimilar, followed by the MCT and DCT. These calculations provide quantitative verification of what can be seen through examination of the apparent emissivity spectra and the retrieved surface spectra.
 The similarity of the derived surface spectra for Units DCT, E, and Ph indicates that these units are related mineralogically. To quantify this assertion we consider the minerals that are retrieved for the four units (Figure 12). Results show that retrievals for Units DCT and E are quite similar, dominated by plagioclase feldspar and clinopyroxene. The retrieval for Unit Ph is similar to those for Units DCT and E, but with a higher proportion of hematite. Unit MCT results are intriguing in that greater proportions of sulfates and phyllosilicates are indicated relative to the other three units. Further, there is a more equant distribution of plagioclase, microcline, and clinopyroxene.
 We note that linear unmixing of emissivity spectra is a powerful way to explore the data sets, but the method comes with provisos. First, it assumes that the emissivities of the end-members combine linearly. This assumption has been proven to be valid as long as the mineral grains are optically thick [Feely and Christensen, 1999]. This is probably the case for the Units DCT, Ph, and E, but may not be the case for the high albedo, low thermal inertia and thus presumably fine-grained Unit MCT materials. Second, the set of mineral spectra used as end-members must be representative of the material that comprise the unit and associated emissivity spectrum. This may or may not be the case for the Unit MCT, in particular, since the inferred induration and aeolian origin of this unit are consistent with a wider range of minerals than might be expected for the other units. Third, the end-members must be spectrally distinct in order to recover the proper mineral components and relative abundances. As shown by Wyatt and McSween  the spectral coverage available for unmixing TES spectra leads to non-unique retrieval of volcanic glasses as opposed to phyllosilicates. Because of these issues we rank our interpretations in decreasing confidence as follows. Units DCT, E, and Ph are composed of similar materials, with the addition of hematite for Unit Ph. Unit MCT is mineralogically distinct from the other three units. Plagioclase and clinopyroxene dominate the Units DCT, E, and Ph. Each unit has some alteration products, including what appear to be sulfates and phyllosilicates. Phyllosilicates are predicted to be a dominate phase for MCT, consistent with the fine-grained nature of these deposits.
5. Hypothesis Development and Testing
 We consider in this section plausible explanations for emplacement of the layered complex (i.e., Units E, P, and Ph) and its unusual remote-sensing properties (Figure 13). To begin with, the morphologic and topographic evidence demonstrates that the layered complex was emplaced disconformably onto Unit DCT. Further, Unit E lies beneath P and Ph. At least the northern portion of the complex was covered by a widespread aeolian mantle that was cemented to become duricrust and most recently has been eroded by wind. The complex proper has also been exposed and differentially eroded by aeolian activity, providing a window into the stratigraphy of the units.
 The morphology of Unit E is complex and probably represents a combination of landforms associated with emplacement, burial, and differential aeolian erosion. We prefer a volcaniclastic origin for Unit E, and the entire complex, for several reasons. First, the deposits are widespread and not confined to a basin as they would be if the material accumulated in a lake or shallow sea. Second, the inferred mineralogy for the complex is dominated by igneous components, with the exception of hematite for the capping unit, Ph. Minerals such as pyroxene and feldspar would not survive well in a lacustrine or marine environment. Third, we interpret the regions within Unit E that expose dark polygons separated by bright ridges as exposing resistant materials that retain morphologic evidence for emplacement even after burial and exhumation (e.g., Figure 3). Specifically, we suggest the following scenario for the formation and evolution of these landforms. Lava flows were deposited onto Unit DCT and subjected to isotropic extension that fractured the flows into polygons. The pattern is remarkably similar to the portions of the fractured plains on Venus, both in terms of polygon widths and fracture patterns [Johnson and Sandwell, 1992]. Fracturing was followed by emplacement of dikes and flows that were in turn extended to form horst-graben patterns aligned with dike emplacement azimuths. These landforms were then buried by volcaniclastic deposits and subsequently exhumed by wind (Figure 14). In other regions within Unit E the fractures became conduits for eruption of volcaniclastic materials (Figure 14). Widespread tephra blanketed these areas, followed by aeolian erosion to expose the deposits. Regions above the source dikes would have accumulated the thickest deposits and perhaps become the most indurated, thereby leaving behind ridges and stepped plateaus as the region was exhumed by wind (Figure 4). Units Ph and P are interpreted to represent the last stage of the volcaniclastic activity, producing what must have been a widespread cap that covered the complex. The hematite bearing material within Ph has been locally reworked into dunes, probably as lag materials due to the high density of this mineral. In fact, the reworking would serve to disperse the material across the plains and disguise any evidence of source vents. This may help explain the unusual occurrence of hematite, pyroxene, and olivine, i.e., mechanical mixtures of an alteration phase and primary igenous minerals.
 A major remaining issue to be addressed is the explanation for the unusual remote-sensing properties associated with Units Ph and E. Unit Ph is dominated by igneous minerals and has a relatively high abundance of hematite. Hematite in this unit has been hypothesized to be present as gray, crystalline hematite grains formed in an aqueous environment [Christensen et al., 2000b, 2001a], platy outcrops of hematite produced via metamorphism in the presence of water while the deposits were buried [Lane et al., 2002], volcanic glass with embedded hematite crystals produced during high temperature devitrification [Minitti et al., 2002], and as a primary igneous mineral produced during volcanic activity [Noreen et al., 2001]. Our analyses suggest that the hematite is carried in the dark plains and dune forming materials, implying that the material is not exposed as platy outcrops. Our work does not shed light on whether the material exists as discrete grains or is contained within devitrified glass. Nor can we distinguish whether or not the hematite was produced as a primary volcanic product or via interactions with aqueous fluids.
 In the visible and reflected infrared wavelengths Unit E is brighter and redder than Unit Ph, although the spectral emissivity signature of Unit E is similar to dark regions (i.e., Unit DCT) and to the signature of Ph (sans hematite) and consistent with dominance by igneous minerals. We hypothesize that the unusual spectral characteristics associated with Unit E are due to glassy coatings that have been devitrified to cryptocrystalline clays with embedded nanocrystalline hematite crystals, i.e., similar to relatively bright, red (in visible and reflected infrared wavelengths) palagonite coatings found on Mauna Kea, Hawaii [Morris et al., 2000]. In the thermal wavelengths the small crystals would show minimal or no distinctive signatures since the Reststrahlen bands would be minimized because of the interference of Fresnel and volume-dominated emission. We note that palagonite on Earth forms in the presence of water and associated oxidizing conditions, although it appears as if the temperature of formation and length of time to form this cryptocrystalline material is poorly constrained [Schiffman et al., 2000].
 It is attractive to ascribe both the formation of the hematite in Unit Ph and the bright, red color of Unit E to the presence of devitrified glasses, although such an explanation is certainly non-unique. We are in a fortunate position to be able to test and update the hypotheses presented in this paper with the Athena Payload on the Mars Exploration Rover Mission (Table 3). The Opportunity Rover, while exploring the Ph unit, will be able to sample both the dark dune material (presumably the source of the hematite signature) and the brighter, underlying substrate that is hypothesized to be exposures of the underlying Unit E. The reason is that our examination of all NA frames within the landing error ellipse shows that this substrate is exposed in most places as patches of bright regions and subdued rolling plains and crater rims (Figures 2a, 2b, and 7). We hypothesize that that rover will find, over its expected traverse distances of hundreds of meters, exposures of the relatively bright substrate (i.e., Unit E) that underlies the hematite unit and the dark dunes and dark plains that contain hematite (i.e., Unit Ph).
Table 3. Athena Payload and Testable Hypotheses
Hypotheses to Test
Pancam: Panoramic Camera
Twelve bands (0.4 to 1.0 μm) for multispectral, stereoscopic imaging with 0.3 mrad IFOV; 9.2 deg by 18.4 deg FOV.
*Mafic dunes with hematite-bearing grains sitting on brighter substrate? *Discrete hematite grains or found as glassy coatings? *Substrate shows evidence for volcanic emplacement, including eroded flows and tephra deposits? *Substrate reflectance and emissivity are comparable to values for exposures of etched unit? Lithologic inferences?
Mini-TES: Thermal Emission Spectrometer
Emission spectra (5 to 29 μm, 10 cm−1 resolution with 8 or 20 mrad FOV)
Arm-Mounted In Situ Package
APXS: Alpha Particle X-Ray Spectrometer
244Cm alpha particle sources, and x-ray detectors, 4 cm FOV
*Measurements on same samples before and after abrading rock surfaces confirm presence of glassy coatings produced during hydrothermal alteration? *Composition and mineralogy consistent with sedimentary or volcanic origin of deposits? If igneous, extent of differentiation? *Rock and grain textures consistent with formation as flows and tephra deposits, reworked by aeolian activity?
MB: Mössbauer Spectrometer
57Fe spectrometer in backscatter mode; Co/Rh source and Si-PIN diode detectors; field of view approximately 1.5 cm2.
MI: Microscopic Imager
30 μm/pixel monochromatic imager (1024 × 1024) with 6mm depth of field
RAT: Rock Abrasion Tool
Tool capable of preparing 5 mm deep by 4.5 cm wide surface on rocks
 We expect that the mast-based remote-sensing package containing Pancam imaging and the Mini-TES emission spectrometer on the rover will be able to map the morphology, physical properties, and mineralogy of the sites visited. For example, questions that will be addressed include whether or not the hematite-bearing material appears within dunes as gray hematite grains or as glassy coatings. Or does the hematite occur as platy outcrops? Does the bright substrate show evidence for bright, glassy coatings in Pancam data and a mafic composition in spectral emissivity extracted from Mini-TES data? Further, the use of the Rock Abrasion Tool, coupled with the Microscopic Imager, Alpha Particle X-Ray and Mössbauer Spectrometers, will provide definitive tests of the hypotheses that the hematite exists as discrete grains or glassy coatings on grains within the dunes and that material comprising the bright substrate is covered with palagonite-like coatings. All of the data will be used to test the idea that the materials and landforms were emplaced volcanically as flows and tephra deposits, mantled, and then exhumed by wind action. Finally, we expect that the complete data set will provide definitive tests of the nature and extent of interaction of the emplaced materials with aqueous fluids and thus dramatically increase our knowledge about the evidence for habitability and life on Mars.
 The linear unmixing procedure for TES spectra was performed with an end-member set consisting of six atmospheric spectra provided by J. L. Bandfield (personal communication, 2002), a blackbody spectrum to account for variations in spectral contrast, and 48 pure mineral spectra (Figure A1). The primary output of the unmixing procedure is the unity-normalized parameter set assigned to the mineral end-members. These values correspond to the fractional areal coverage of each end-member within the field of view of the instrument at the time of observation. The mean normalized value for each end-member within each of the four main surface units was calculated, and the mineral end-members were grouped into the mineral classes shown in Figure 12.
 REA and FPS were supported by the NASA Planetary Geology and Geophysics Program through Grant NAG5-7830 from the Goddard Space Flight Center and by the Mars Exploration Rover Mission under contract from Cornell University. We thank the NASA Planetary Data System for access to the MGS archives used in our analyses. We also thank Philip Christensen, Joshua Bandfield, Vicky Hamilton, and Steven Ruff, Arizona State University, for help in understanding their approach to analysis of TES data.