Spectral interpretation that silicic rocks are widespread on Mars implies that Earth's differentiated crust is not unique. Evaluation of observations bearing on the composition of the Martian crust (Martian meteorite petrology and a possible crustal assimilant, analysis of Mars Pathfinder rocks, composition of Martian fines, interpretation of spacecraft thermal emission spectra, and inferred crustal densities) indicates that the crust can be either basalt plus andesite or basalt plus weathering products. New calculated chemical compositions for Thermal Emission Spectrometer (TES) global surface units indicate that surface type 1 has basaltic andesite composition and surface type 2 has the composition of andesite. If these materials represent volcanic rocks, their calc-alkaline compositions on a FeO*/MgO versus silica diagram suggest formation by hydrous melting and fractional crystallization. On Earth, this petrogenesis requires subduction, and it may suggest an early period of plate tectonics on Mars. However, anorogenic production of andesite might have been possible if the primitive Martian mantle was wet. Alternatively, chemical weathering diagrams suggest that surface type 2 materials could have formed by partial weathering of surface type 1 rocks, leading to depletion in soluble cations and mobility of silica. A weathered crust model is consistent with the occurrence of surface type 2 materials as sediments in a depocenter and with the alpha proton X-ray spectrometer (APXS) analysis of excess oxygen suggesting weathering rinds on Pathfinder rocks. If surface type 1 materials are also weathered or mixed with weathered materials, this might eliminate the need for hydrous melting, consistent with a relatively dry Martian mantle without tectonics.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 A comparison of surface materials on the Earth and Moon illustrates that planetary differentiation can follow varying paths that lead to quite different crustal compositions. Until recently, the Earth's silicic continental crust (mostly a result of hydrous melting and fractionation in subduction zones) was thought to be geochemically unique [Rudnick, 1995]. However, the analysis of rocks having chemical compositions similar to andesite at the Mars Pathfinder landing site [Rieder et al., 1997; McSween et al., 1999; Waenke et al., 2001; Foley et al., 2003] suggests that the Martian crust contains silicic rocks, and the Mars Global Surveyor Thermal Emission Spectrometer (MGS-TES) mapped abundance of a global spectral unit interpreted to be andesitic [Bandfield et al., 2000; Hamilton et al., 2001] may imply that such silicic rocks are widespread. This hypothesis challenges our understanding of petrogenesis on a world seemingly without plate tectonics.
 Mars has a voluminous crust, averaging ∼50 km thickness [Zuber, 2001] and comprising >4% of the planetary volume (compared to ∼1% crust on the Earth). A crustal dichotomy separates thick, ancient (Noachian) highlands crust in the southern hemisphere from thinner, less densely cratered lowlands in the north. The lowlands are covered with younger (Hesperian) sediments of the Vastitas Borealis Formation [Scott et al., 1987; Tanaka et al., 1988, 2003], but the density of large, incompletely buried craters is comparable to that in the highlands [Frey et al., 2002], suggesting that the age of the northern lowlands basement is also Noachian. The Tharsis rise, a huge dome containing massive shield volcanoes, separates these terranes along part of their join. The volcanic surfaces on and around these volcanoes are relatively young (Amazonian), although Tharsis itself has been a locus of plume volcanism for billions of years [Phillips et al., 2001]. The lithosphere under this bulge may be >100 km thick if the load is supported isostatically [Solomon and Head, 1982]. Elysium is another, smaller plume containing large volcanoes and Amazonian flows. Figure 1, which schematically illustrates salient features of the various major subdivisions of the Martian crust, will serve as a useful guide as we consider constraints on crustal composition and origin.
 Early formation of the bulk of the Martian crust is inferred from its bombardment history [Hartmann and Neukum, 2001] and the ∼4.5 Ga measured crystallization age of ALH84001 [Nyquist et al., 2001], the only Martian meteorite that directly sampled the ancient crust. The ∼4.0 Ga 40Ar/39Ar age of this meteorite is thought to reflect the late heavy bombardment [Ash et al., 1996]. The extraction of incompatible elements during formation of this early crust depleted mantle source regions that later melted to produce younger Martian meteorites, which still carry the geochemical signature of this 4.5 Ga fractionation event in their strontium [Borg et al., 1997] and lead [Chen and Wasserburg, 1986] radiogenic isotope systems. Moreover, the former existence of short-lived radionuclides like 146Sm in the Mars mantle, as documented in Martian meteorites [Harper et al., 1995; Borg et al., 1997], demands early crustal differentiation. Some additional crust formation occurred during the Hesperian, when lavas flooded the northern plains to depths of 1–2 km [Frey et al., 2002], and significant amounts of accompanying plutonic rock must also have been added to the northern plains crust during this time.
 Mars exhibits a rich variety of volcanic landforms, leading to speculation that its crust consists mostly of igneous rocks [e.g., Greeley and Spudis, 1981]. If layering in the walls of Valles Marineris consists entirely of lava flows [McEwen et al., 1999] and such layering were globally distributed, it would suggest that virtually the entire crustal thickness is of igneous origin. However, Malin and Edgett [2000, 2001] have interpreted some Valles Marineris layers as well as thick (∼10 km) layers covering other parts of the Martian surface as sedimentary rocks, albeit likely derived from igneous precursors. There is little doubt that both volcanic and sedimentary units and landforms are common on Mars, but the relative proportions of these materials within the crust remains controversial.
 A related controversy is whether the Martian crust experienced chemical (as opposed to only mechanical) weathering processes with accompanying chemical fractionations. The preservation of igneous mineralogy on Mars is indicated by the spectral identification of pyroxenes, plagioclase, and sometimes olivine in regions not blanketed by dust [Mustard et al., 1997; Bandfield et al., 2000; Hamilton et al., 2001; Bandfield, 2002]. However, a variety of alteration phases (e.g., sheet silicates, amorphous silica, zeolites, and palagonites) in modest proportions have been suggested as analog components in thermal emission spectra of low-albedo Martian surface materials [Wyatt and McSween, 2002; Kraft et al., 2003; Ruff and Christensen, 2003; Morris et al., 2003]. Evidence for crystalline phyllosilicates has not been observed in visible/near-infrared (VISNIR) spectra, leading to speculation that any hydrous silicates must be poorly crystalline or amorphous [e.g., Bell et al., 2000]. Hydrogen in equatorial regions analyzed by the Mars Odyssey Gamma-Ray Spectrometer [Boynton et al., 2002] can be interpreted as mineral-bound OH, supporting the idea of chemical weathering. Pervasive erosional striping and widespread overland deposition, as inferred from MGS MOC images [Malin and Edgett, 2001], would have facilitated both chemical and mechanical weathering of volcanic rocks.
 Despite decades of spectral mapping of the planet's surface by orbiting spacecraft, chemical analysis of surface materials by several landers and rovers, and laboratory study of several dozen Martian meteorites, the bulk composition of the Martian crust remains undefined and its origin poorly understood. Here, we critically evaluate constraints on the chemistry and petrology of the ancient crust, derive new geochemical data from TES spectroscopy for different parts of the crust, and explore how crustal components having these compositions might have arisen on Mars.
2. Constraints on Mars Crust Composition
2.1. Petrology of Martian Meteorites
 ALH84001 offers an extremely limited and biased sampling of the ancient crust. This ∼4.5 Ga-old ultramafic (orthopyroxenite) cumulate cannot represent the bulk composition of the highlands, because partial melting of an ultramafic mantle cannot yield an ultramafic crust. Instead, it has been suggested that ALH84001 formed by fractional crystallization of a basaltic parent magma [Mittlefehldt, 1994]. The apparent failure of impacts to dislodge more meteorites from the ancient crust may stem from an inability of old crust to transmit the requisite shock waves because of scarcity of coherent rocks within these terranes [Head et al., 2002a; McSween, 2002].
 The other Martian meteorites, collectively called SNCs (an acronym for shergottite, nakhlite, chassignite), have young crystallization ages, ranging from 175 to 1300 Ma [Nyquist et al., 2001]. Shergottites are subdivided into basaltic shergottites (tholeiitic basalts, sometimes with modest amounts of cumulus pyroxenes), olivine-phyric shergottites (basalts containing olivine xenocrysts or phenocrysts), and lherzolitic shergottites (plagioclase-bearing peridotites). Nakhlites (olivine-bearing clinopyroxenites) are similar to pyroxenites in some terrestrial komatiite flows, and Chassigny (a dunite) is chronologically and geochemically linked to the nakhlites. All these meteorites are basaltic rocks or plutonic cumulates likely formed by fractional crystallization of basaltic magmas. Compositions of basaltic shergottites (including olivine-phyric samples) and nakhlites plot within the field of basalt on a chemical classification diagram for volcanic rocks (Figure 2).
 The young crystallization ages of SNCs point to their derivation from young volcanic centers, probably Tharsis or Elysium, apparently the only regions on Mars geologically recent enough to host these rocks. Attempts to locate specific launch sites for Martian meteorites using TES spectra have not been successful [Hamilton et al., 2003], because Tharsis, Elysium, and perhaps other young volcanic terranes are mantled by thick blankets of dust. Earlier studies concluded that visible/near-infrared (VISNIR) spectra of dark regions (without dust cover) on Mars were similar to basaltic shergottites [Mustard et al., 1997]. However, VISNIR spectra are primarily sensitive to the presence of ferromagnesian minerals, and the similarity in this case refers to pyroxene compositions rather than to the bulk mineralogy.
 The apparent absence of rocks other than basalts or their plutonic derivatives among young Martian meteorites suggests that the crust comprising the young volcanic centers is basaltic. However, the fact that many Martian meteorites are partial cumulates demonstrates that fractionation did occur, possibly yielding more silicic residual magmas. Attempts to constrain lava rheologies, and thereby infer lava compositions, from the eruption styles of Tharsis and Elysium volcanoes allow a wider range of magma compositions [e.g., Zimbelman, 1985; Catermole, 1987; Baloga et al., 2003], but these models suffer from ambiguities related to magma effusion rate, crystallinity, and other factors that influence flow rheology.
 Although SNC meteorites have provided critical insights into the timing of crust formation and the geochemical nature of the mantle source region after early crust extraction, they are not samples of the ancient crust and thus cannot provide direct information on its composition. Later in this paper, we will introduce evidence that the ancient crust is distinctly different from Martian plume magmatism as revealed by SNCs.
2.2. Crustal (?) Reservoir Sampled by Shergottite Magmas
 The parental magmas for basaltic and olivine-phyric shergottites sampled two distinct geochemical reservoirs. Correlations between initial radiogenic isotope ratios (87Sr/86Sr, 143Nd/144Nd, 176Hf/177Hf) and fractionations of rare earth elements (REE) could indicate incorporation of varying amounts of ancient, highly radiogenic, light REE-enriched crust by melts from a depleted mantle [Borg et al., 1997; McSween, 2002]. The assimilated material was similar in many respects to the lunar KREEP component. Alternatively, these correlations could indicate mixing of materials from complementary enriched and depleted reservoirs within the Martian mantle [Wadhwa and Grove, 2002; Borg et al., 2003]. In this case, the depleted and enriched regions must have been isolated by an early differentiation event, possibly creating a shallow, enriched mantle and a deep, depleted mantle. A correlation between the degree of geochemical contamination and magmatic oxidation state [Wadhwa, 2001; Herd et al., 2002] is consistent with crustal assimilation, although recent evidence shows that the depleted source region for nakhlites could also be derived from an oxidized mantle reservoir [Wahdwa and Grove, 2002].
 For the moment, let us assume that the assimilant was crust. Can we use the composition of this component to constrain the nature of the ancient crust through which the shergottites erupted? Unfortunately, it is easier to determine the isotopic and trace element compositions of this component than its major element abundances or petrologic identity. It is not even clear if this component was actually rock or a fluid that either metasomatized the shallow mantle or scavenged solutes from the crust. If rock was assimilated, the ancient crust probably has a basaltic composition; assimilation of andesite would have increased the silica contents of the resulting contaminated magmas. Figure 3a shows silica abundances in shergottite melts plotted versus La/Yb (a proxy for the abundance of the assumed crustal component). Although most shergottites show a positive correlation, as would be expected if the assimilant were silicic rock, the Los Angeles shergottite is inconsistent with this model. If the ancient crust can be represented by the Mars Pathfinder dust-free rock (see below), incorporation of this component should also have resulted in lower MgO/Al2O3 (Figure 3b), which is not observed. (Note: La/Yb for the Pathfinder rock is unknown but is assumed in Figure 3 to be high, as appropriate for fractionated crust.)
2.3. In Situ Chemical Analysis of Martian Crustal Rocks
 The Mars Pathfinder rover analyzed five rocks using an alpha proton X-ray spectrometer (APXS). Preliminary X-ray mode analyses of rocks at the Pathfinder site [Rieder et al., 1997] have now been revised [Waenke et al., 2001; Foley et al., 2003], owing to the inclusion of alpha mode data and to differences in conditions under which laboratory calibrations and Mars measurements were made. Element concentrations plotted versus sulfur in rocks yield straight lines, with soils clustering at the sulfur-rich ends of the rock arrays. These trends are interpreted as mixing lines between the compositions of rock and dust. Imagery indicates that rocks at the Pathfinder site are partly coated with red dust, and a correlation between sulfur content and the red/blue (750/440 nm) reflectance spectra ratio of APXS-analyzed rocks [McSween et al., 1999] reinforces the conclusion that rock analyses are contaminated by sulfur-rich dust. Extrapolation of the mixing lines to low sulfur (Waenke et al.  used 0.3 wt.% S, based on basaltic shergottite sulfur contents) yields the dust-free rock composition.
 The SiO2 concentration (57 ± 6 wt.% from Waenke et al. ; 57.7 ± 1.5 wt.% from Foley et al. ) for the Pathfinder dust-free rock plots on the low-silica boundary of the andesite field in a classification diagram (Figure 2). The preferred interpretations of this rock composition are that it is volcanic, or is a clastic sedimentary rock composed of volcanic fragments [McSween et al., 1999; Waenke et al., 2001]. However, McSween et al.  also considered the possibility that the dust-free rock composition might represent a weathered rock surface. Under terrestrial weathering conditions silica is usually leached from basalt, but some basalts exposed to semi-arid conditions develop hydrous, silica-rich coatings [Farr and Adams, 1984; Crisp et al., 1990]. Altered silicic coatings on Pathfinder rocks would be consistent with their photometric properties, which may imply varnished rinds [Johnson et al., 1999]. Bishop et al.  developed a model for the formation of Martian rock varnish, involving chemical reactions between rocks and the dust that settles on them. Foley et al.  noted that plots of alkalis versus sulfur in Pathfinder rocks show considerable scatter, perhaps suggesting mobilization of soluble elements during weathering. APXS alpha mode analyses of oxygen [Foley et al., 2003] suggest that the Pathfinder dust-free rock contains more oxygen than can be accounted for by stoichiometric combination with its cations. The excess oxygen in Shark, the rock with least dust cover, is equivalent to 3.3 ± 1.3 wt.% H2O, assuming that half the iron is ferric. This water abundance is very high for igneous rocks, and may be more consistent with the hypothesis that Pathfinder rocks have weathered, hydrous coatings.
2.4. Martian Surface Fines as Proxies for Crust Composition
 The pervasive dust that covers the Martian surface is thought to have been globally homogenized by winds [e.g., McCord et al., 1982], which may account for the compositional similarity of soil deposits at landing sites separated by thousands of kilometers [Clark et al., 1982; Waenke et al., 2001]. Fine-grained sediments are commonly used to estimate the composition of the Earth's crust [McLennan and Taylor, 1984], and surface fines might likewise provide a critical constraint on compositions of the dominant crustal rocks on Mars.
 Originally, the compositional similarity between Viking soils and basaltic shergottites was cited as evidence that the soils formed from basalts [Toulmin et al., 1977]. Following the discovery of rocks having andesitic compositions by Mars Pathfinder, a number of workers reinterpreted Pathfinder and Viking soil compositions as mixtures of basaltic (SNC) and andesitic materials in roughly equal proportions [Larsen et al., 2000; Morris et al., 2000; Waenke et al., 2001]. At face value, this would seem to be evidence for the existence of significant amounts of silicic crust. Such a model presumes that soils formed by mechanical weathering of rocks, without significant chemical modification.
 Alternatively, analyzed Martian soils could represent mixtures of a common (globally homogenized) dust component with varying amounts of local (andesitic) rock particles at the Pathfinder site and with sulfate cements at the Viking sites [McSween and Keil, 2000]. Because the Pathfinder APXS analyses of soils have recently been recalibrated [Waenke et al., 2001; Foley et al., 2003], we have replotted these compositions to see if the trends noted by McSween and Keil  persist (Figure 4). The recalibrated Pathfinder soil analyses are not as readily interpreted as dust with admixed local rock (Figure 4), but it is still possible to estimate an average soil composition from uncemented Pathfinder and Viking soils. This composition, which we assume represents the global dust, is presented in Table 1 (the calculation procedure and assumptions are also described in Table 1) and shown by Xs in Figure 4. The common dust composition itself can be modeled as basalt that has undergone a moderate degree of chemical weathering [McSween and Keil, 2000]. Various chemical weathering mechanisms (palagonitization, hydrothermal alteration, reactions of rocks with acid fog formed by volcanic exhalations) have been suggested as possible origins for Martian soils (summarized by Bell et al. ). However, the hypothesis that the global dust may have formed by simple mixing of basaltic and andesitic components does not seem to be tenable, given the mismatch for silicon, iron, and potassium in mixing calculations (Figure 5).
Table 1. Recalculated Martian Fines Composition Based on Recalibration of Mars Pathfinder APXS Analyses of Soilsa
Martian fines composition is measured in wt.% oxides. Mars Pathfinder soil averages of Waenke et al.  and Foley et al. , indicated by MP, include all reported soil analyses except A8 (cemented soil). All Pathfinder analyses were normalized to 98% to allow for unanalyzed water (estimated ∼2 wt% [Biemann et al., 1977; Yen et al., 1998]). MP(fines) is an average of the MP(Waenke) and MP(Foley) values. V(fines) is an average of all Viking 1 analyses of fines (C-1, C-6, C-7, C-8, C-9) from Clark et al. . All Viking analyses were normalized to 94.5%, to allow for unanalyzed Na2O, P2O5, Cr2O3, MnO (average Pathfinder values were assumed) and 2% water. Global dust is an average of MP(fines) and V(fines).
 The initial ambiguity in interpreting the surface type 2 lithology from deconvolved TES mineral abundances arose because volcanic siliceous glass (a major component of andesite) was shown to be spectrally similar to some alteration phases (sheet silicates, amorphous silica coatings, and K-feldspar) over the spectral ranges used in deriving the TES surface spectral end-members [Wyatt and McSween, 2002]. Absorption features between 500–550 cm−1 in laboratory spectra can be used to distinguish well-crystalline clay minerals from high-silica volcanic glass; however, this region was excluded by Bandfield et al.  while deriving the TES spectral end-members because the CO2 atmosphere of Mars is largely opaque near this spectral region [Wyatt and McSween, 2002]. Significant quantities of well-crystalline clay minerals are not indicated by near-infrared observations using Mariner 9 ISM data [Murchie et al., 2000] or telescopic observations [Blaney et al., 2003], and work by Ruff  examining the 500–550 cm−1 region in non-atmospherically corrected TES spectra points to the lack of spectral absorption features indicative of well-crystalline clays. However, these studies do suggest that poorly crystalline clays and/or other alteration phases may be permissible, and recent work [Morris et al., 2003; Ruff and Christensen, 2003] has shown palagonites and zeolites to be spectrally similar to high-silica glass. Some highly shocked feldspars (maskelynites) are also spectrally similar to high-silica glass [Johnson et al., 2002]. The original interpretation of the high-silica glass spectral end-member as a primary volcanic glass [Bandfield et al., 2000] was also shown to be too limited, as deconvolved modal abundances of the natural surface of a terrestrial flood basalt suggested it could also represent an amorphous high-silica alteration product [Wyatt and McSween, 2002]. Analyses by Kraft et al.  have further shown that the addition of high-silica alteration coatings on basalts results in a surface that is spectrally similar to andesite.
2.5.2. Distributions of Surface Compositions
 A global view of the distribution of surface type 1 and type 2 materials is shown in Figure 6. The distribution of the surface type 1 (basalt - green) unit is restricted to the southern highlands and Syrtis Major regions of Noachian or Hesperian age and to a few local occurrences in the northern plains [Bandfield et al., 2000; Rogers and Christensen, 2003]. The surface type 2 unit (andesite and/or altered basalt - red) displays the highest concentrations in the younger Amazonian-age northern lowlands regions of Acidalia Planitia and the circumpolar sand seas [Bandfield et al., 2000; Rogers and Christensen, 2003]. These materials in the northern lowlands have been mapped as the Vastitas Borealis Formation [Scott et al., 1987]. Surface type 2 compositions are also present in moderate abundances, or mixed with surface type 1, throughout the low-albedo southern highlands. Blue pixels in Figure 6 represent regions covered by a blanket of dust which prohibits spectral analysis of sand and rock compositions. The distribution of the highest concentrations and largest extents of the two surface spectral units is thus split roughly along the planetary topographic dichotomy separating the ancient, heavily cratered crust in the southern hemisphere from younger lowland plains in the north.
 It is important to note that a single interpretation of surface type 2 spectra may not be warranted everywhere on Mars. THEMIS data from Mars Odyssey show adjacent volcanic units of surface type 1 and 2 materials within the Nili Patera caldera [Ruff and Christensen, 2003]. These units have high thermal inertias and probably represent outcrops of lava and/or tuff. For these units an igneous origin, involving successive eruptions of basaltic and andesitic magmas, is a reasonable interpretation. Elsewhere, deposits of sediments with surface type 2 spectra may be more plausibly interpreted as partly weathered basalt. For example, Wyatt et al.  described deposits of surface type 1 sand dunes on the floors of large craters in Oxia Palus, adjacent to surface type 2 materials on the downwind sides of the crater walls. In this case, surface type 2 can be readily explained as a finer-grained fraction (containing some alteration materials) winnowed by winds from the coarser basaltic sediment on the crater floor.
2.6. Density of the Crust
 Using MGS MOLA data, new models of the relationship between gravity and topography [Turcotte et al., 2002] and admittance techniques [McGovern et al., 2002; McKenzie et al., 2002] suggest densities of 2.95–3.15 g/cm3 for parts of the elastic lithosphere of Mars. Turcotte et al.  have argued that the crust is volumetrically equivalent to the elastic lithosphere, although other models suggest a thinner crust [Zuber et al., 2001; Nimmo, 2002]. The inferred density is significantly higher than that estimated for the Earth's continental crust (∼2.75 g/cm3), which has an average composition of andesite. Although the Martian crustal density appears to be inconsistent with a dominantly andesitic crust, it might be consistent with hydrous magmatism, which could produce dense pyroxenitic cumulates in the lower crust [Muentener et al., 2000], perhaps resembling the ALH84001 orthopyroxenite.
 Thermal and compositional buoyancy forces in the mantle source regions of basaltic magmas cause them to ascend and erupt on planetary surfaces. On Earth, mid-ocean ridge basalts accumulate as much as several kilometers below the level of neutral buoyancy (estimated at 100–400 m), suggesting that magma density may not control ascent once the magma reaches the shallow crust [Hooft and Detrick, 1993; Ryan, 1993]. Ultimately, buoyancy is likely to control ascent at deeper levels, but compositions of the most common magmas from mid-ocean ridges indicate that eruption controls are complex [Grove et al., 1993; Michael and Cornell, 1998] and not solely a function of density contrast [Stolper and Walker, 1980]. Thus it may be unrealistic to use SNC magma densities to constrain the density (and hence composition) of the Martian crust.
3. Crust Geochemistry From Thermal Emission Spectra
3.1. Background and Method
 Volcanic rocks are commonly classified by their chemical compositions because their modal mineralogies are not always diagnostic. TES is a mineralogical tool, but it can also provide a means of estimating chemistry. Hamilton and Christensen  demonstrated that the chemical compositions of laboratory-analyzed rocks can be accurately calculated from deconvolved modal mineralogies by combining the compositions (wt.% oxides) of the spectral end-members in proportion to their relative modeled abundances. Wyatt et al.  further quantified the uncertainties in derived chemical compositions and demonstrated their use in correctly classifying volcanic rocks. Errors for most oxides, as determined from the Wyatt et al.  study of terrestrial volcanic rocks, are ± 5%. Derived chemical abundances from thermal emission spectra are thus a recasting of rock compositions into a form which complements modeled mineral abundances.
Hamilton et al.  convolved laboratory spectral data (2 cm−1 spectral sampling) of rocks from Wyatt et al.  to the lowest spectral resolution of the TES instrument (10 cm−1 spectral sampling) and showed that derived bulk rock chemistries were not significantly degraded. These results demonstrated the feasibility of using similar techniques and classification schemes for TES spectral resolution data. Hamilton et al.  also derived chemical compositions of the surface types 1 and 2 global spectral units and classified them as basalt and andesite, respectively. However, their spectral end-member set was optimized for igneous rocks.
 Here, we estimate and compare surface type 1 and 2 chemical compositions derived from three previously published modal abundances deconvolved using spectral end-member sets that include a broad range of igneous and sedimentary minerals [Bandfield et al., 2000; Hamilton et al., 2001; Wyatt and McSween, 2002]. Spectral fitting for those studies was constrained to 1280–400 cm−1, although TES data actually cover the wave number range of 1650–233 cm−1. The atmospheric correction used to derive the TES surface spectra [Bandfield et al., 2000] did not include the high wave number range of TES data due to numerous water vapor and minor CO2 features. Furthermore, there are no fundamental silicate features in the 1650–1400 cm−1 region. The wave number range of 400–233 cm−1 was not used for deconvolutions in previous studies because mineral end-member spectra in the ASU Thermal Emission Spectroscopy library [Christensen et al., 2000c] only extended to 400 cm−1. This laboratory now has a spectrometer that covers the full TES wavelength range (1650–200 cm−1) and emissivity spectra have been measured for all end-members, enabling us to expand the spectra range used for deconvolution of TES surface type 1 and 2 materials. Thus, in this study, we also estimate surface type 1 and 2 compositions from new linear deconvolutions that cover the expanded spectral range of 400–233 cm−1. Modeled spectral fits, deconvolution modal abundances, and derived chemistries from each of the previously published studies and our new work are examined to constrain Martian surface compositions.
3.2. Model Results and Classification
Figures 7a and 7b compare spectral fits for linear deconvolutions of surface types 1 and 2 spectra, respectively. Bandfield et al.  used 45 spectral end-members representing igneous, sedimentary, and metamorphic minerals, whereas Hamilton et al.  used a narrower range of 29 mineral spectra common in unweathered basalts and andesites. Wyatt and McSween  used 39 spectral end-members representing igneous and alteration minerals in partly weathered basalts. In this study, we use 46 spectral end-members representing a similar wide range of igneous and sedimentary minerals in unweathered and weathered basalts and andesites (Table 2). Overall, spectral fits produced by the linear deconvolution algorithm using the different end-member sets are very good (Figures 7a and 7b), suggesting major rock phases are well represented in the end-member libraries and that they provide acceptable fits to the rock types in this study. Spectral fits of the surface type 1 spectrum show low RMS values of 0.0018 [Bandfield et al., 2000], 0.0026 [Hamilton et al., 2001], 0.0018 [Wyatt and McSween, 2002], and 0.0022 (this study). Spectral fits of the surface type 2 spectrum show low RMS values of 0.0009 [Bandfield et al., 2000], 0.0023 [Hamilton et al., 2001], 0.0014 [Wyatt and McSween, 2002], and 0.0019 (this study). Figures 7a and 7b also demonstrate that the extended spectral range used in this study is well modeled and does not adversely affect the overall quality of modeled spectra. Deconvolved mineral abundances from each of the previous studies and new mineral abundances from this study are listed in Table 3 and shown in Figures 7a and 7b. The detection limit for mineral abundances deconvolved from TES spectra is 10–15 vol.% based on instrument uncertainties [Christensen et al., 2000a], errors associated with atmospheric corrections [Bandfield et al., 2000; Smith et al., 2000], and limits of the deconvolution technique [Ramsey and Christensen, 1998]. Surface type 1 deconvolved mineral abundances for all end-member sets are similar to within the 10–15 vol.% TES uncertainty and the 5–10 vol.% absolute uncertainty that has been associated with all modeled mineral abundances based on comparisons with petrographic point-counting modes [Feely and Christensen, 1999; Hamilton and Christensen, 2000] and high-resolution electron microbe phase mapping techniques [Wyatt et al., 2001]. Surface type 2 deconvolved mineral abundances for all end-member sets are also similar to within the 10–15 vol.% TES uncertainty, except for feldspar abundances from Hamilton et al.  and this study which represent the relative maximum and minimum modeled feldspar abundances. The low modeled feldspar abundance for surface type 2 in this study, and increase in the total of minor phases modeled well below TES detectability limits, may result from the extended wave number range used in deconvolution. The end-member sets that include a variety of alteration phases (phyllosilicates, carbonates, silica) have lower RMS errors than the Hamilton et al.  fits, which focused almost entirely on igneous phases. These results suggest that small to modest amounts of alteration minerals may be present in both surface types. Carbonate abundance is modeled well below TES detectability limits, in agreement with the conclusion of Bandfield  that carbonates are not detectable in low-albedo regions.
Table 2. Spectral End-Members Used for New Deconvolutions of MGS-TES Emissivity Dataa
Table 3. Modeled Phase Abundances for MGS-TES Surface Types 1 and 2 Materialsa
All mineral groups have detection limits of ∼10-15 vol.%. In modeled results, feldspars are dominated by plagioclase, pyroxenes by high-Ca pyroxene, glass by Si-K glass, and sheet silicates by smectite. Other category includes the sums of carbonates and sulfates individually modeled well below detection limits.
 Chemical compositions calculated from modeled mineral abundances for each of the spectral end-member sets are presented in Table 4. The derived chemical compositions plotted in Figures 2, 8, and 9 are calculated on a H2O-free and CO2-free basis (also presented in Table 4). These figures explore the likelihood that surface types 1 and 2 represent volcanic rock compositions. The normalization has the effect of increasing the other oxide components, especially silica contents. Surface type 2 deconvolutions typically contain more hydrous minerals and thus their derived compositions contain more water; consequently, their silica contents are increased more than for surface type 1. Figures 10 and 11, which explore the possibility that the derived compositions represent partly weathered volcanic rocks, are unaffected by this normalization.
Table 4. Derived Chemistry for MGS-TES Surface Types 1 and 2 Materialsa
 Surprisingly, there is relatively little difference between chemical compositions estimated using the various spectral end-member sets. Using the total alkalis versus silica classification of Le Bas et al. , all surface type 1 compositions plot within the field of basaltic andesite and are clearly distinct from Martian meteorites (Figure 2). Surface type 2 compositions plot within the andesite field (Figure 2) and near the Mars Pathfinder dust-free rock composition (in this figure, but not necessarily in other diagrams). However, surface type 2 compositions that are not recalculated as H2O-free would plot within the basaltic andesite field, closer to surface type 1 compositions. The clustering of compositions suggests that chemistry derived from TES data is relatively robust, although we caution that thermal emission spectra sample only the outer few hundred microns of grains.
4. Petrogenesis of Crustal Materials
 Martian surface compositions could reflect either igneous materials or a combination of basalt and alteration products. Below, we consider each possibility.
4.1. Origin of Andesitic Magmas
Miyashiro  introduced the FeO*/MgO versus SiO2 diagram (Figure 8) (* indicates that all iron is reported as FeO) to delineate the tholeiitic (TH) and calc-alkaline (CA) magmatic trends. In this diagram, anhydrous fractional crystallization paths (tholeiitic trends) are nearly vertical, as illustrated by arrows in the TH field showing the trends of Galapagos mid-ocean ridge lavas (kinked arrow, analyzed by Juster et al. ) and the dry liquid line of descent for basaltic shergottites (short arrow, calculated using the MELTS program by Hale et al. ). In contrast, fractional crystallization under hydrous conditions follows distinctive paths (calc-alkaline trends) that lead to higher silica contents, as illustrated by the diagonal arrows in the CA field [Sisson and Grove, 1993; Grove et al., 2003]. The effect of H2O is to expand the primary phase volumes for olivine and clinopyroxene, causing plagioclase to crystallize late. Primitive melts produced by low degrees of hydrous partial melting of mantle peridotite (olivine + orthopyroxene + clinopyroxene + spinel) have compositions indicated by the large open circle in Figure 8. These magmas would follow a fractionation trend like the diagonal arrow emanating from the circle and paralleling the TH-CA boundary. The heavy gray arrow illustrates the compositions of liquids derived from higher degrees of hydrous melting after clinopyroxene and spinel are exhausted from the source, leaving only olivine and orthopyroxene (harzburgite) in the mantle residue. The length of the heavy arrow represents the 20 to 40% melting interval at 1.2 GPa [Gaetani and Grove, 1998]. This extent of melting is encountered in the mantle wedge in terrestrial subduction zones. Liquids along this arrow are produced by incongruent melting of orthopyroxene to form olivine + liquid at high water contents [Grove et al., 2003]. These conditions are only met on Earth when the mantle wedge above the subducted slab is fluxed by an H2O-rich fluid derived by dehydration of subducted oceanic plates. Subsequent fractional crystallization of these water-rich melts at shallow crustal levels produces a family of diagonal trends, as illustrated by diagonal arrows representing various terrestrial arc lavas [Grove et al., 2003] in Figure 8.
 Our derived compositions for surface type 1 are similar to basaltic andesites found in terrestrial subduction zones. On the basis of comparison with terrestrial calc-alkaline lavas (Figure 8), surface type 1 basalts were produced by fairly extensive mantle melting under hydrous conditions and at shallow depths (∼1 GPa on Earth). Analogous terrestrial melts are in equilibrium with a refractory harzburgite source and contain high magmatic H2O contents (4–6 wt.%). Their position above the terrestrial mantle melting arrow could reflect melting of a Martian mantle with higher FeO/MgO. Derived compositions for surface type 2 are more scattered, but could represent magmas formed by higher degrees of hydrous melting of the same mantle (Figure 8). Compositions derived from this study (X-boxes in Figure 8) suggest that andesite having surface type 2 composition might have formed by fractionation of surface type 1 basaltic andesite magma under hydrous conditions in the shallow Martian crust, although that relationship is not apparent in previously published compositional pairs.
Figure 9 illustrates that other chemical trends in surface type 1 and 2 compositions more closely approximate those of terrestrial calc-alkaline lavas (in this case, lavas from Mt. Shasta [Grove et al., 2002] and hydrous lavas from Medicine Lake volcano [Kinzler et al., 2000]) than tholeiitic lavas of the Galapagos spreading center [Juster et al., 1989]. Higher Al2O3 contents are the result of delayed plagioclase crystallization in hydrous magmas. The flat TiO2 pattern in calc-alkaline rocks, reflecting early crystallization of Fe-Ti oxide, is also similar.
 The alkali abundances of surface type 1 and 2 are clearly subalkaline (Figure 2), although Neksavil et al.  proposed that some SNC magmas were alkaline. Figure 9 shows the compositions of alkaline (anorogenic) lavas from the Nandewar volcano [Abbott, 1969; Stolz, 1985]. Although the Al2O3 and FeO* abundances are similar to calc-alkaline lavas, the Mars surface compositions are clearly distinguished from alkaline rocks by their SiO2 and TiO2 contents.
 By comparison with terrestrial magmas, it may be difficult to envision how hydrous melting and fractional crystallization could have happened in the absence of plate tectonics. Subduction of hydrated slabs provides the mechanism for adding so much water to the Earth's mantle. At present, Mars is a one-plate planet that convectively translates radioactive heat in the mantle through a stagnant lid [Schubert et al., 2001]. However, some authors have proposed an early period of plate tectonics on Mars to explain various geomorphic features [Sleep, 1994; Baker et al., 2002; Fairen et al., 2002] and magnetic lineation patterns detected in parts of the ancient southern highlands [Connerney et al., 1999; Nimmo and Stevenson, 2000], although there is little or no supporting evidence [Pruis and Tanaka, 1995]. Thermal models suggest that the stagnant lid model operating during the entire Martian evolution can more readily explain the volume and timing of crust formation than can a model of early plate tectonics followed by a stagnant lid [Hauck and Phillips, 2002; Breuer and Spohn, 2003]. Also, the preservation of early-formed isotopic anomalies in the mantle source regions of SNC meteorites [Chen and Wasserburg, 1986; Borg et al., 1997; Lee and Halliday, 1997] would have been difficult during plate tectonic convection.
 In terrestrial subduction zones, the primary melts are basaltic, not andesitic, and andesites are produced primarily by fractional crystallization and assimilation of overlying silicic crust [Rudnick, 1995]. As a consequence, andesitic volcanism on Earth is mostly associated with thick, continental crust. Conversely, on Mars surface type 2 (possibly andesitic) materials overlie thin crust in the northern plains, whereas the thick southern highlands are overlain by surface type 1 (basaltic) materials. Baker et al.  suggested that early subduction with attendant hydrous mantle melting on Mars produced a thick andesitic crust in the highlands. Later floods eroded the highlands andesitic crust and delivered these sediments to the lowlands, where they covered thinner oceanic crust. The highlands were subsequently mantled by younger, plume-related basalts. There are several potential problems with this scenario. First, it seems unlikely that large impacts into the highlands would not have excavated significant amounts of andesitic crustal stratigraphy to produce a mixed TES spectral signal. And second, even if surface type 2 is interpreted as weathered basalt (see below), our derived composition of surface type 1 materials in the highlands still suggests hydrous melting, unlike the tholeiitic compositions expected for plume basalts and seen in SNC meteorites.
 It is conceivable that hydrous melting of the Martian mantle might somehow have occurred without plate tectonics. There is precedence for this idea: Rudnick  argued that the production of a part of the Earth's continental crust in the Archaean required a different mechanism than subduction, although the specific mechanism is unclear. Lowman  proposed that an ancient andesitic crust could have formed on Mars through hydrous melting without plate tectonics. In his model, mantle water was thought to be primordial, leading to an early crust of andesite with later dry, basaltic volcanism. Early melts of a wet mantle would presumably scavenge water, leading to the dry mantle source region of younger SNC meteorites, as envisioned by Waenke and Dreibus . Marsh  argued that terrestrial bimodal magmatism, leading to the production of siliceous melts from basaltic parent magmas, occurs through the formation of lenses within solidification fronts. However, it is not obvious that such a mechanism could operate on a global scale and produce volumetrically significant amounts of andesite.
4.2. Weathering of Basaltic Rocks
 Chemical weathering in the terrestrial environment almost certainly takes place under conditions different from Mars. However, Nesbitt and Wilson  indicated that leaching of major elements from weathered volcanic rocks is not greatly influenced by primary mineralogy, bulk chemical composition, or climatic conditions, so it is plausible that tools developed for terrestrial weathering but applied to the Martian regolith may provide useful insights.
 Chemical weathering of terrestrial basaltic rocks typically leads to bulk depletion of leachable CaO, Na2O, K2O, and to a lesser extent MgO, relative to FeO and Al2O3. These depletions are illustrated by arrows in molar ternary diagrams (Figure 10) devised by Nesbitt and Young  and Nesbitt and Wilson . The plotted positions of the new compositions from this study (X-boxes), as well as compositions based on previous deconvolutions, suggest that surface type 2 materials could have been produced by chemical weathering of surface type 1 basaltic andesite. Figure 10a shows that surface type 2 materials contain more Al2O3 than surface type 1, as appropriate for weathered materials, but they also contain slightly higher proportions of K2O, the component most readily leached. The relative positions of surface type 1 and 2 materials in Figure 10b are also consistent with weathering. In Figure 10c, surface type 2 compositions are displaced toward higher Al2O3 relative to surface type 1, not directly away from the easily leached apex. This displacement might occur if iron was not already oxidized to the ferric state, which is the insoluble form [Nesbitt and Wilson, 1992].
 Only a modest degree of chemical weathering is required by these diagrams (the displacement is less than the lengths of vectors representing weathered terrestrial basalts in Figure 10; weathered terrestrial basalts commonly are shifted across the feldspar-FeO or feldspar-FeO + MgO joins). Incomplete weathering is in agreement with the limited proportions of clays and other alteration phases (<50%) in the deconvolved surface type 2 spectra of Wyatt and McSween . Alternatively, surface type 2 materials could consist of basaltic sand that has been physically mixed (on TES pixel scale) with chemically weathered materials. The surface type 1 composition might also represent less weathered material, as its TES spectral deconvolutions utilize smaller amounts of clays [Bandfield et al., 2000; Hamilton et al., 2001; Wyatt and McSween, 2002].
 Weathering of basalts commonly leads to mobility of silica. In fact, the silica in basaltic rocks (usually contained in readily altered glass and plagioclase) is more easily mobilized than is the silica in quartz-bearing rocks [McLennan, 2003]. Although silica is commonly leached from basalts during chemical weathering, it ultimately must be precipitated and concentrated elsewhere. Indeed, measurements of SiO2/Al2O3 in palagonites and other clay alteration products formed by low-temperature alteration of Icelandic basaltic glasses show both depletions and enrichments of silica [McLennan, 2003]. Some Hawaiian basalts also have silica-rich weathering rinds [Farr and Adams, 1984; Crisp et al., 1990]. Thus it seems plausible that the higher silica contents of surface type 2 materials could also reflect chemical weathering of basalt.
 The Mg/Si versus Al/Si diagram (Figure 11) has been used commonly to distinguish terrestrial and Martian rocks [e.g., Rieder et al., 1997]. Compositions for surface type 1 and 2 materials clearly plot off the trend defined by Martian ultramafic and basaltic rocks. Instead, they are arrayed along a trend toward higher Al/Si (gray arrow). This arrow actually defines the chemical changes measured in palgonitized basalts and in a basalt weathering profile, as summarized by McLennan . The positions of surface type 1 and 2 compositions and the resulting slope of the trend resemble chemically weathered basalts, not Martian igneous rocks. Surface type 2 is not displaced to higher Al/Si contents as would be expected if it is more highly weathered; however, mobilization (increase) of silica might have limited Al/Si variations for surface type 2 and Mars Pathfinder rock compositions in Figure 11.
 As noted earlier, the mapped distribution of surface type 2 in the northern lowlands is concentrated in Acidalia Planitia, part of the low-albedo surface of the Vastitas Borealis Formation [Bandfield et al., 2000; Wyatt and McSween, 2002]. This formation was initially mapped on the basis of its morphologic and albedo characteristics and its occurrence below the Martian highland/lowland boundary [Scott et al., 1987]. Its various landforms have been attributed to volcanic, tectonic, glacial, periglacial, and sedimentary processes [Clifford and Parker, 2001; Head et al., 1999; Kargel et al., 1995; Kreslavsky and Head, 2001; McGill and Hills, 1992; Parker et al., 1989; Scott et al., 1987; Tanaka, 1997; Tanaka et al., 2001, 2003]. The Vastitas Borealis plains have most recently been interpreted as sedimentary deposits derived from outside the basin [Head et al., 2002b] or as altered sediments formed through local reworking of earlier deposits by permafrost processes [Tanaka et al., 2003]. The sediments are thought to be underlain by ancient ridged volcanic plains, possibly exposed through erosion as basaltic outliers [Rogers and Christensen, 2003]. Both “new-view” theories propose that these materials have undergone significant reworking during transport or by indigenous weathering, so they are consistent with alteration of basaltic sands within this depocenter. Vastitas Borealis morphology is characterized by a mixture of smooth plains, polygonal troughs and fractures, and pitted domes whose origins appear to involve interactions with water and/or ice [Head et al., 2002b; Tanaka et al., 2003]. TES spectra of the Vastitas Borealis boundary in southern Acidalia Planitia, where it is not obscured by dust, indicate that the proportion of surface type 1 material increases outside the basin [Wyatt et al., 2003].
 The hypothesis that the ancient Martian crust has experienced modest chemical weathering and concurrent chemical fractionations may obviate the need to appeal to hydrous melting, especially if both surface types 1 and 2 are partly weathered to different degrees or mixed with differing amounts of weathered materials. This model may be more consistent with the conventional view of Mars as having a relatively dry mantle and no plate tectonics.
 Because surface types 1 and 2 materials dominate the exposed Martian crust, it is logical that one or both might have been precursors to the globally homogenized dust. Figure 10 shows that the Martian dust composition (Table 1) is distinct from either surface type. Moreover, the dust cannot be rationalized as a simple mixture of the two surface types. McSween and Keil  used these diagrams to argue that the composition of Martian dust could arise by weathering of basalts having shergottite compositions, but not by weathering of the Mars Pathfinder dust-free rock (using a previous calibration of the Pathfinder rock composition). None of the rock compositions alone in Figure 10 is an obvious precursor for the recalculated global dust composition, perhaps suggesting that the process that produced the dust was not isochemical. Figure 10c also allows the possibility that dust formed by addition of iron oxides to surface types 1 or 2, possibly as a result of aeolian concentration of dense oxide grains [McLennan, 2000; McSween and Keil, 2000], which would translate the composition toward the apex containing FeO. Concentration of iron oxides is also suggested by high and variable FeO/MnO ratios in Mars Pathfinder soils [Foley et al., 2003].
 We have considered a number of diverse observations to try to resolve the question of whether the Martian crust contains significant quantities of silicic rock:
 • With only one exception, Martian meteorites have young crystallization ages, suggesting that they are samples of young (dust-covered) volcanic centers such as Tharsis or Elysium. On the basis of the compositions of SNCs, the recent crust of Mars is basaltic. ALH84001, the only meteoritic sample of the ancient Martian crust, is an ultramafic cumulate thought to have formed from basaltic magma.
 • The absence of major element variations (SiO2, MgO/Al2O3) that correlate clearly with indications of geochemical mixing (radiogenic isotopes, REE fractionations, magma redox state) in basaltic/olivine-phyric shergottites suggests that any assimilated rock component had a major element composition similar to basalt. However, it is unclear whether the admixed component represents the ancient crust through which shergottites erupted or a metasomatised mantle reservoir.
 • The composition of the APXS-analyzed dust-free rock surface at the Mars Pathfinder landing site is andesitic, but it is unclear whether this composition represents volcanic rock or a silicic weathering rind. New alpha-mode analyses of excess oxygen in Pathfinder rocks, corresponding to ∼2 wt.% H2O in even highly oxidized samples, support the interpretation the rocks have weathered coatings.
 • The analyzed chemistry of uncemented Martian soils has been used to estimate the composition of fines. The composition of globally homogenized Martian fines could plausibly mimic the average composition of the exposed crust. However, the dust composition does not conform to the compositions of either TES surface type or Martian meteorites, nor to any simple mixture of these materials. The dust is depleted in soluble oxides and appears to be chemically weathered material.
 • Previously published deconvolutions of MGS-TES spectra suggest that basalt/basaltic andesite dominates the southern highlands. Spectra for the northern lowlands can be interpreted as either andesite or partly weathered basalt. The sedimentary origin inferred for the Vastitas Borealis Formation in the northern lowlands is consistent with a weathered origin for this vast exposure of surface type 2 material.
 • Densities for the Martian elastic lithosphere estimated from global topography and gravity are higher than for the terrestrial continental crust (having the composition of andesite) and appear to be consistent with an average basaltic composition.
 This listing does not unambiguously resolve the question of whether the crust is basaltic or contains a significant andesitic component. Accordingly, we carefully considered the chemical compositions of crustal materials previously calculated from TES spectra for surface type 1 and 2, using a variety of spectral end-members. We also calculated new compositions on the basis of deconvolving TES data over a wider spectral range than used in prior studies.
 All the derived chemical compositions for surface type 1 materials correspond to basaltic andesite on the basis of the total alkalis versus silica classification diagram. On a FeO*/MgO versus silica diagram, this calc-alkaline composition resembles magmas generated by fairly high degrees of hydrous melting in terrestrial subduction zones. The chemical compositions of surface type 2 materials correspond to andesite, and could have formed by more extensive melting of hydrated mantle rocks with accompanying hydrous fractionation. This petrogenesis challenges our understanding of anorogenic magmatism, and suggests a much wetter ancient Martian mantle than is normally supposed. Early differentiation of the voluminous Martian crust would have depleted the mantle's water inventory, so that subsequent dry melting would produce SNC tholeiitic magmas.
 Alternatively, surface type 2 materials could be partly weathered surface type 1 basaltic andesite on the basis of chemical weathering diagrams. This model circumvents the problem of explaining how andesite formed in the absence of subduction, and is consistent with an interpretation of the Mars Pathfinder dust-free rock composition as a weathering rind. This hypothesis implies that chemical weathering, at least of limited extent, may be pervasive on the Martian surface, consistent with inferences of widespread sediments. It is conceivable that surface type 1 materials could also be partly weathered (to a lesser degree) or volcanic sand mixed with weathered materials (on a TES pixel scale). Chemical fractionations during weathering and/or admixture of weathered materials could eliminate the need for a hydrated Martian mantle source for volcanic rocks of the ancient crust.
 This study refines the nature of both the igneous and chemical weathering explanations for TES-derived global surface units. However, at present, we are left with multiple working hypotheses for the composition and origin of the Martian crust.
 Jeff Johnson and Hanna Nekvasil provided insightful reviews. This work was partly supported by NASA Cosmochemistry grants NAG5-12896 to HYM and NAG5-10728 to TLG.