near-infrared spectral variations of martian surface materials from ism imaging spectrometer data

Icarus147, 444–471 (2000)

doi:10.1006/icar.2000.6446, available online at http://www.idealibrary.com on

Near-Infrared Spectral Variations of Martian Surface Materialsfrom ISM Imaging Spectrometer Data

Scott Murchie

The Johns Hopkins University Applied Physics Laboratory, Laurel, MarylandE-mail: [email protected]

Laurel Kirkland

Lunar and Planetary Institute, Houston, Texas

Stephane Erard

University of Paris, Paris, France

John Mustard

Department of Geological Sciences, Brown University, Providence, Rhode Island

and

Mark Robinson

Northwestern University, Evanston, Illinois

Received February 3, 2000; revised May 3, 2000

Imaging spectrometer data from the ISM instrument on Phobos2 were used to characterize spatial variations in near-infrared spec-tral properties of the martian surface, to determine the correspon-dence between near-infrared and visible-wavelength spectral vari-ations, and to assess lithologic variations in the surface materials.All data were radiometrically calibrated and corrected for effectsof atmospheric gases using previously described methods. The datawere also corrected photometrically to a standard geometry, andthe estimated contribution of light backscattered by atmosphericaerosols was removed to isolate the reflectance properties of surfacematerials. At shorter near-infrared wavelengths, the surface variesbetween three major spectral types which correspond to known vis-ible color units. Dark gray materials have 1- and 2-µm absorp-tions consistent with a pyroxene-containing lithology, and brightred dust has a shallow 0.9-µm absorption consistent with a poorlycrystalline ferric mineralogy. Dark red soils are spectrally similarto dust although lower in albedo. In some cases their 0.9-µm fer-ric iron absorption is deeper and offset toward longer wavelengthsthan in dust. These attributes agree well with those determinedin situ for comparable materials at the Mars Pathfinder landingsite. At longer wavelengths, significant regional heterogeneities areobserved in the slope of the spectral continuum and the depth ofthe 3-µm H2O absorption. The 3-µm band is stronger in bright redsoils than in most dark gray soils, but the strongest absorptionsare found in intermediate-albedo dark red soils. Observed spectralvariations suggest the presence of at least four surface components,

dust, pyroxene-containing rock and sand, one or more crystallineferric minerals, and a water-bearing phase. These are broadly con-sistent with four surface components that have been inferred fromground-based, orbital, and landed spectral studies and from in situcompositional measurements. We also conclude from our analysisthat most albedo and spectral variations result from the coating ofdark mafic rock materials by bright ferric dust. Dark red regions,however, are inferred to have dust-like compositions but lower albe-dos, due in part to intermixture of a dark, crystalline ferric mineral.Both of these major conclusions are strongly supported by landedinvestigations by Mars Pathfinder. The layered materials in VallesMarineris are the only geologic formation with distinctive spectralproperties, including an enhanced 3-µm H2O band and pyroxeneabsorptions which imply a mineralogy distinct from materials inthe surrounding highlands. These properties provide important ev-idence for the layered materials’ origins and are most consistentwith mechanisms that involve volcanism restricted to the interiorsof the chasmata. c© 2000 Academic Press

Key Words: Mars; surfaces; spectroscopy, water; Phobos space-craft.

yernet’s

44

0019-1035/00 $35.00Copyright c© 2000 by Academic PressAll rights of reproduction in any form reserved.

1. INTRODUCTION

The mineralogic composition of the martian surface lais key to understanding three important aspects of the pla

4

oo

o

iao

t

r

-s

d

a

tifd

a6

u

e9

po-neore of

me-6).ereg-

h-

sorp-hiene,

im-rk-like

thelike

ark

ger

oilskerst

bylticrveree

e-

to be981,darkin-nce

ar-ge-r toem.nsver-areofec-sts

NIR SPECTRAL VARIATIONS

evolution. First, mineralogy of crustal materials provides infmation on the history of crustal formation and planetary vcanism. Second, mineralogy of alteration products providesformation on past weathering environments and past climaThird, mineralogic differences between different geologic fmations allow insights into past geologic processes. Informtion on the mineralogic mineralogic composition of the martsurface layer comes from two primary sources, spectroscmeasurements andin situelemental abundance measuremen

Spectroscopic measurements acquired from above the asphere were reviewed by Soderblom (1992), Roushet al.(1993),and Bell (1996). They include three-color visible measuremefrom the Viking Orbiters, near- to mid-infrared data from thMariner 6 and 7 Infrared Spectrometers (IRS), visible, neinfrared, and mid-infrared data from ground-based, airboand Earth-orbiting telescopes, near-infrared data from the Iming Spectrometer for Mars (ISM) on Phobos 2, and mid- afar-infrared data from the Infrared Interferometer Spectrome(IRIS) on Mariner 9 (Hanelet al.1972) and the Thermal Emission Spectrometer (TES) on Mars Global Surveyor (Christenet al. 2000a, 2000b). Unique, extremely high spatial resotion visible and near-infrared multispectral images were aacquired from the surface by the Imager for Mars Pathfin(IMP) (Smith et al. 1997, McSweenet al. 1999, Bell et al.2000).

These diverse spectral data sets have led to broad agreeabout some major attributes of the surface materials, butoverall weakness and subtlety of mineralogic absorptionsleft key details unresolved. At visible wavelengths, the surfconsists of three major color units, low-albedo dark gray sointermediate-albedo dark red soils, and higher-albedo brighsoils. The bright red soil has low thermal inertia, suggestthat much of it consists of thick accumulations of dust (Kiefet al. 1977, Christensen 1986). Its strong UV absorption eand weak absorptions near 0.53, 0.66, and 0.86µm are wellfit by ferric oxides, occurring mostly in nanophase form withsmall amount of a crystalline ferric phase. The crystalline phis widely thought to be hematite (Adams and McCord 19McCordet al.1977, 1982, Singeret al.1979, Morriset al.1989,1990, Morris and Lauer 1990, Bellet al. 1990, Bell 1992). Atnear-infrared (NIR) wavelengths, a strong absorption at 3µmshows that some molecular water is present, probably boas mineral hydrates (Moroz 1964, Houcket al.1973, Pimentalet al.1974). Absorptions due to silicates in the dust are weadefined at best, suggesting poorly crystalline materials. Featpurportedly identified in different data sets are located at∼2.25,2.35, and 2.40µm, and have been attributed to clays of mixcomposition including Al-bearing phases (Bell and Crisp 19Murchie et al. 1993, Beinroth and Arnold 1996), Mg-bearinclay (Singeret al.1985), scapolite (Clarket al.1990), or hydrouscarbonates (Calvinet al.1994).

Dark gray soils are somewhat more well-understood. N
telescopic and ISM data have shown that dark gray regionshibit 1- and 2-µm absorptions diagnostic of pyroxene (Sing
OF THE MARTIAN SURFACE 445

r-l-in-tes.r-a-npic

ts.mo-

ntsear-ne,ag-ndter

enlu-lsoer

mentthehasceils,redngerge

ase9,

und

klyres

d3,

g

IR

et al. 1979, Erardet al. 1991, Mustardet al. 1993, Mustardand Sunshine 1995, 1996). Systematic differences in thesitions of these two absorptions indicate diversity in pyroxecomposition (cf. Adams 1974, Cloutis and Gaffey 1991). Fmost dark gray regions, the band positions suggest a rang2-pyroxene basaltic composition comparable to that of SNCteorites (Mustardet al.1993, Mustard and Sunshine 1995, 199At the Mars Pathfinder landing site, IMP spectra show that this a similar diversity of mafic rock types. Most rocks have a lonwavelength 1-µm band suggestive of a composition rich in higFe clinopyroxene (McSweenet al.1999). However, a minorityof the rocks at the site have a deep, shorter-wavelength abtion consistent with a more magnesian clinopyroxene (Murcet al.2000). TES data indicate the presence of single-pyroxeclinopyroxene-rich basalts in other regions including Terra Cmeria (Christensenet al.2000b). The spectral properties of dagray regions may thus be summarized as evidencing basaltrock with some variations in composition.

Dark red soils, though intermediate in albedo, appear fromlimited spectral measurements available to be ferric in naturebright red dust. ISM spectra of the two largest “classical” dred regions (Oxia and Lunae Planum) reveal 0.9-µm absorp-tions comparable to those in bright red dust but with a lonwavelength center (Murchieet al. 1993, Murchie and Mustard1994, Murchieet al. 1996, Mustard 1995). IMP spectra fromthe Mars Pathfinder landing site also show that dark red sare grossly similar spectrally to bright red dust, although darand with a 0.9-µm absorption distinctly deeper than in the du(Smithet al.1997, McSweenet al.1999).

In situ measurements of surface materials were acquiredthe Viking Landers and Mars Pathfinder. In addition, the basaSNC meteorites are thought to originate from Mars and to seas proxies for dark gray rocks. The dust and soils at all thsites are similar in their elemental abundances (Toulminet al.1977, Bellet al.2000), consistent with the prevailing hypothsis that martian dust is well-mixed by winds (e.g., Greeleyet al.1992). The major elemental abundance variation appearsthe relative abundance of sulfate salts (Clark and VanHart 1McSween 1999). Elemental abundances in the bright andred soils are indistinguishable, indicating that only a slight meralogic difference may create the observable spectral differe(Riederet al.1997, Bellet al.2000).

In summary, the current evidence for mineralogy of the mtian surface layer has clear implications for the three keyologic questions identified. (1) Dark gray materials appeabe mafic crustal materials and sediment derived from th(2) Bright and dark red materials have similar compositioconsistent with weathered basalt, perhaps reflective of aaged source-region compositions. Minor spectral differencescaused by slight differences in the mineralogy or crystallinityferric minerals. (3) There is little or no correspondence of sptral variations with individual geologic deposits. This sugge
ex-
erthat most rock units are similar in composition, hidden by ho-mogenized surface materials, or some combination of both.

IE

csdp

a

ow

tIo

lu

f

irt

essa

a

entspec-eightsur-ric

is,cor-ours

a-arkck

red

ve-orchpsehed

isallestaal-itiesing

thetioneol-ge-ol-

nakahethe), in

.a).rageits

M

446 MURCH

In this paper we present a detailed analysis of ISM speto determine how they constrain the second and third issuegarding the surface layer, the composition of altered soils anlithologic differences among geologic deposits. The geograregions covered in this study are shown in Fig. 1. Specific groof questions addressed include:

• Are the bright red, dark red, and dark gray color ununiquely indicative of distinctive groups of soils? Or does evisible-color grouping really represent a collection of spectradistinct materials that merely resemble each other at viswavelengths?• What mineralogic components are required to explain

served spectral variations? To what extent is the “classical” tcomponent bright dust/dark mafic rock and sand model cor(e.g., Christensen and Moore 1992), or are other componenquired? If so, how do additional components suggested bydata compare with those suggested by remote sensing atwavelengths and byin situanalyses?• Are any spectrally distinct soils related to underlying g

ologic units? If so, do the spectroscopic properties of the scontain evidence for the lithology or genetic mechanisms ofgeologic formations?

We analyzed the spectral properties of a representativetion of the martian surface observed by ISM (Tharsis, VaMarineris, Arabia, and Syrtis Major-Isidis). The focus of owork, bright and dark red altered materials, complementsfocus of Mustard and Sunshine (1995, 1996) on dark gray mrials and the evidence from their mafic mineral absorptionsmartian igneous evolution. Our work also complements ongoremote sensing studies of the martian surface using datadifferent speacecraft. Because ISM overlaps the wavelengtherage of IMP in the mineralogically crucial 1-µm wavelengthregion, our analysis helps constrain the relationship betwglobally distributed materials and the materials measuredin situby Mars Pathfinder (McSweenet al.1999, Bellet al.2000). Andbecause ISM covers a different wavelength range than TESsensitivity to different absorption features provides complemtary determinations of surface compositional variations.

The remainder of this paper is divided into four sections. Fprocedures for ISM data calibration, registration, and correcare summarized, and the analytical techniques used to evaspectral heterogeneity are described. Second, the data reduand analysis techniques are evaluated step-by-step to ideand to bound sources of error. We show that uncertaintiethe reduced data are small compared with the observed spheterogeneity and that results of our analysis are relatively insitive to errors in corrections for the atmosphere and aeroThird, we examine the NIR spectral properties of the surflayer. Greater complexity is found than is evident at visible walengths. Observed NIR spectral heterogeneities can be explby different relative abundances of at least four distinct min
alogic components, the canonical dust and mafic componplus a water-containing component and a low-albedo ferric co
ET AL.

trare-

thehicups

itschllyible

b-o-

rects re-SMther

e-oilsthe

por-lesr

theate-foringromcov-

een

, itsen-

st,ionluatectionntifys inctralen-ols.ce

ve-ineder-

ponent. These compare closely to four analogous componthat have been inferred from ground-based and landed stroscopy and fromin situ compositional analyses. Fourth, wconsider in greater detail three key surface materials: (a) brred and dark gray soils, which dominate most regions of theface; (b) dark red soils, which consistently exhibit a highly ferspectral signature like dust, but an enhanced 3-µm absorptiondue to H2O; and (c) the layered deposits of Valles Marinerthe only observed region with spectral heterogeneities thatrelate spatially with geologic formations. We also comparework with results fromin situ investigations, especially on MarPathfinder. The landed investigations strongly support two mjor conclusions from the ISM data, that the bright red and dgray units are physical mixtures of ferric dust and mafic roand that the dark red unit is compositionally related to brightdust.

2. DATA REDUCTION AND ANALYSIS

ISM was a scanning grating spectrometer covering the walength range 0.76–3.16µm. It measured eleven image cubes“windows” in the martian equatorial region in February–Mar1989. Two of the eleven windows were measured near periaof an elliptical transfer orbit, and they cover small regions of tTharsis plateau at∼5 km/pixel. The other nine were measurefrom the orbit of Phobos at∼22 km/pixel. The ISM instrumentand its returned data are described in detail by Bibringet al.(1990).

Six of the lower-resolution ISM windows were used in thstudy. They cover the eastern slope of the Tharsis plateau, VMarineris, Arabia, Syrtis Major, and Isidis (Fig. 1). These daare well matched with our questions about composition oftered materials and the relationship of spectral heterogeneto underlying geology. The data cover wavelengths containmineralogic absorptions due to Fe minerals and H2O- and OH-containing phases, all potentially important constituents ofsurface layer. They have an image format with spatial resoluadequate for assessing the relationships with underlying gogy. And the data cover all three visible color units as well asologic units that span nearly the whole martian stratigraphic cumn (Scott and Tanaka 1986, Greeley and Guest 1987, Taet al.1992). Bright red dust occupies nearly the entirety of tTharsis plateau and much of Arabia. Dark gray soils occur ineastern part of Valles Marineris (Eos and Capri ChasmataMargaritifer Terra, Sinus Meridiani, and Syrtis Major PlanumDark red soils occur in both Oxia and Lunae Planum (Fig. 1

The remaining windows were not used because their coveis almost entirely within the Tharsis dust region, which exhiblittle spectral contrast.

2.1. Radiometric Calibration

Most steps involved in the radiometric calibration of IS
entsm-data have been described previously (Erardet al. 1991,1994, Mustardet al. 1993): (a) subtraction of dark current;

n

a

a

e

a

a

r

ti-reeear

y)ontor

ct,uresen-

bia

arent

n-ioners,ws

therard

eedoy,

riestifi-

-th-ed

s)

ofea-toofis

nci-rdere-


(b) correction for instrumental gain settings; (c) removal of sptral order overlaps; (d) preliminary radiometric calibration usiresults from onground instrument characterization; (e) divisby the solar spectrum; and (f) refinement of the relative radmetric calibration of different channels.

The refinement of radiometric calibration is driven by the fathat ISM operated inflight at temperatures as low as−77◦C,outside the temperature range at which it was radiometriccalibrated onground (above−70◦C). Recalibration utilizes ISMobservations of several large, homogeneous regions thatbeen well characterized in telescopic data (Syrtis Major aAmazonis, dark and bright areas, respectively; McCordet al.1982). In ISM’s second spectral order (0.76–1.51µm), the plau-sible mineralogic absorptions are broad, smooth features duFe minerals which are measurable in the telescopic spectrthis spectral region both gain and offset corrections wererived from empirical line fitting of ISM measurements of thdark and bright reference areas to the telescopic reflectancthe same regions (Mustardet al.1993). These were then applieto the ISM data. The physical mechanism justifying the gcorrection is instrument performance outside of its calibratemperature range. The most probable physical explanationthe offsets is an additive signal from scattered light. Scattelight arises from up to 20 fields-of-view away or more, as edenced by detection of martian atmospheric absorptions ondisk of Phobos while it occulted Mars (Bibringet al. 1990). Asecondary source of offset could be unremoved, time-variadark current. The dark current was measured before andter each window, but it could have drifted between timesmeasurement.

A modified approach is required for the first spectral ord(1.64–3.16µm) because altered, bright soils may exhibit weanarrow absorptions in this wavelength region due to clays, cbonates, or other minerals. Such weak features may not bedent in the telescopic data, and “fitting over them” would crespectral artifacts in the ISM data after derivation and applicatof gain and offset corrections. Because the occurrence of sweak, narrow absorptions is least likely in minimally alteredgions, the dark gray region Syrtis Major was used alone to degain corrections.

Beyond 2.6µm, telescopic spectra of the reference areas wunavailable at the time the recalibration procedure was derivISM measurements of the martian moon Phobos were thereused to provide continuity in radiometric calibration from thshorter wavelengths out to 3.16µm. Phobos is well-suited forthis purpose because it exhibits no evidence of absorptionthis wavelength range (Bellet al. 1989, Rivkin et al. 2000).For calibration purposes, the spectrum of Phobos at these wlengths was assumed to be featureless with the low slope typof anhydrous chondrites (Erardet al.1994).

Since this method of recalibrating the ISM data was derivonly one significant modification has been required. Eight
the eleven ISM windows, including both windows coverinthe calibration reference areas and five of the six used in

ec-g

ionio-

ct

lly

havend

e to. In

de-es of

din

tedfor

redvi-the

bleaf-

of

erk,ar-evi-te

ionuche-rive

ereed.foree

s in

ave-ical

ed,of

study, were measured at similar detector temperatures of−76to −77◦C. Spectra of dust in these windows are nearly idencal, suggesting stable instrument performance. However, thwindows were measured at warmer detector temperatures n−71◦C, including the one covering Arabia (used in this studplus the two high-resolution windows over Tharsis. Comparisof observations of the same part of Tharsis at different detectemperatures showed that responsivity at wavelengths≤1.1µmvaried several percent with temperature. To correct this effethe measurements of Tharsis dust at two detector temperatwere used to derive the temperature-dependence of instrumtal response. This correction was then applied to the Arawindow.

2.2. Spatial Registration

The lines-of-sight of the first and second spectral ordersslightly offset due to detector configuration in the instrume(Erardet al.1991). Utilizing all of the ISM windows, we deter-mined the best-fit pointing offset to be one pixel in the dowtrack (east–west) direction and one pixel in the scan direct(northwest–southeast). We registered the two spectral ordand in further analyses used only the portions of the windowith complete spectral coverage in both orders.

Internal to the data, there are also sub-pixel differences inlines-of-sight between spectral channels in the same order (Eet al. 1991). Erardet al. partially corrected for this “misregis-tration effect” using pixel resampling. However, this techniquleaves systematic artifacts in some channels at sharp albboundaries. To minimize effects of such artifacts in this studthe data were low-pass filtered at 3× 3 pixels (∼66× 66 km)prior to analysis. In addition, areas near the sharpest boundawere neglected in subsequent analysis to prevent false idencations of “new” materials.

All six windows were registered with the Viking Digital El-evation Model (DEM) using Phobos 2 and ISM pointing information, augmented by spatially correlating maps of the depof the 2.0-µm CO2 absorption with the DEM. Subsequent comparison with Viking image mosaics of the same areas showthat ISM data are registered accurately to≤1 pixel (≤22 km) inareas with significant topographic relief (e.g., Valles Marineriand∼2 pixels (∼44 km) in flat areas such as Tharsis.

2.3. Corrections for Atmospheric and Photometric Effects

Illumination and viewing geometries and the elevationsmeasured regions have four major effects on ISM spectral msurements (Fig. 2a): variation in strengths of absorptions dueatmospheric gases, photometric variations in the brightnessthe surface, variations in the fraction of measured light thatbackscattered by atmospheric aerosols, and attenuation of ident irradiance and surface reflectance by the aerosols. In oto compare quantitatively materials observed at different geom
gthistries and through different atmospheric path lengths, correctionsfor these effects must be made.

he textpth of theof t, ov

448 MURCHIE ET AL.

FIG. 1. Spectral heterogeneities of the western and eastern equatorial regions. The base map is global topography, shown from 15◦N to 20◦S, 10◦W to 130◦W(top), and from 15◦N to 5◦S, 250◦W to 10◦W (bottom) in simple sinusoidal projections. (a) Viking visible-wavelength color. Place names mentioned in tare indicated. (b) Albedo at 1µm overlain on topography. Redder hues indicate higher reflectances and bluer hues indicate lower reflectances. (c) De2-µm pyroxene absorption, overlain on topography. Bluer hues indicate weaker absorptions and redder hues indicate stronger absorptions. (d) Depthhe 3-µmH2O absorption, overlain on topography. Redder hues indicate stronger absorptions and bluer hues indicate weaker absorptions. (e) Spectral slopeerlain ontopography. Bluer hues indicate flatter spectral continuua and redder hues more negatively sloped continuua. (f) Principal components loading map.Higher loadings
of PC1 are represented by greater brightness in the red image plane, higher loadings of PC2 by greater brightness in the blue plane, and higher loadingsof PC3 bygreater brightness in the green plane. Areas with high PC5 loadings are hachured in red, and areas with low PC5 loadings are hachured in green.

NIR SPECTRAL VARIATIONS OF THE MARTIAN SURFACE 449

FIG. 1—Continued

E

cesn

r

-de-

sum-noOmit

sorp-anyod-onsof

bynce

heric

tionsing

lunc-othrop-ov-the

n oflope

enttralcurot

rentpe isiththepro-

dhar-ssti-aze

as

rivedt-is-

.g.,

450 MURCHI

FIG. 2. Illustration of photometric correction and aerosol removal algrithms applied to the ISM data. (a) Schematic representation of the major fathat affect reflected radiance measurements of the martian surface by spac(b) Estimated NIR spectrum of light backscattered from atmospheric aeroat 0◦ emission angle and 20◦ phase angle. Data points are derived from Kirklaet al. (1997); the solid line is a second-order polynomial fit. (c) Spectra of briand dark regions observed in overlapping ISM windows at different geomet
before photometric and aerosol corrections (heavy lines) and after correc(fine lines).
ET AL.

o-torscraft.ols,d

ghties,

Atmospheric gases.The spectral contribution of atmospheric gases was partially removed using the procedurescribed by Erardet al. (1991). A spectral model of theatmosphere was constructed based on measurements of themit and base of Pavonis Mons. This 18-km-high shield volcais spectrally uniform except at wavelengths inside strong C2

absorptions, so that a ratio of spectra of the base and summeasures the increase in strength of atmospheric gas abtions over the difference in atmospheric path lengths. Forgiven spectrum, extinction due to atmospheric gases is meled by scaling strengths of absorptions in the Pavonis Mratio spectrum linearly in logarithmic space with the strengththe 2.0-µm CO2 absorption. A measured spectrum is dividedthe model atmospheric spectrum to “restore” surface reflectaproperties as they would appear in the absence of atmospgases. Because aerosol backscattering accounts for≤5% of themeasured radiance at wavelengths of atmospheric absorpexcept at 2.7µm (see below), its effects were neglected duratmosphere removal.

Photometric and aerosol effects.Photometric and aerosoeffects are intimately coupled because the martian phase ftion, as measured from space, includes contributions from bsurface materials and atmospheric aerosols. Isolating the perties of the surface materials is greatly simplified by a discery made early in analysis of ISM data. As measured byslope of the reflectance spectrum (1R/1λ), the magnitude ofaerosol backscattering is nearly independent of the elevatiothe surface. There is no significant increase in spectral sdown the flanks of the Tharsis volcanoes or over the escarpmat the edge of Valles Marineris. Such a relationship of specslope with topography would have to occur if aerosols ocuniformly throughout the atmospheric column, but this is nobserved. Instead, over deposits of similar albedo but diffeelevation, such as dust on the Tharsis plateau, spectral slocorrelated with the secant of the emission angle—that is, wpath length through a uniform layer in the atmosphere. Onbasis of this behavior, martian atmospheric aerosols wereposed to occur predominantly in a uniform haze (Drossartet al.1991, Erardet al.1994). A high altitude for the haze is favoreby the continuity of spectral slope measurements over the Tsis volcanoes (cf. Rodinet al.1997). Obviously, discrete cloudrepresent departures from this simple model and would contute concentrations of aerosols either within or outside of a hlayer.

The NIR spectrum of martian aerosols which we used wderived from ISM data by Kirklandet al. (1997, 2000a). Usingdeposits of dust in Tharsis as a reference material, they dewavelength-dependentk coefficients to the Minnaert photomeric function using observations at different incidence and emsion angles. The Minnaert function, typically used for Mars (eErard 1995, Bellet al.1997), has the form

k
tions R= R0
(cosi )

(cose)1−k,

ss

-

ind

aiasla

r

a

its

t

a

r

.

cs

oaiaval,

ataatedctra.

rtral

rials,by

ria-

in-um,

sig-therram-

eters” (Fig. 4). These include reflectance, slope of the spectral


whereR is measured reflectance,R0 is normal albedo,i is in-cidence angle, ande is emission angle. Kirklandet al. (1997,2000a) found thatk increases with wavelength from∼0.6 near0.77µm, and converges toward∼0.87 at≥2.5 µm. Kirklandet al. inferred that the wavelength dependence ofk is predomi-nantly due to wavelength-dependent backscattering by aeroand that the convergence towardk= 0.87 at longer wavelengthindicates that aerosols’ effects become nearly constant thKirkland et al.applied to the Tharsis dust spectra a wavelengindependent Minnaert correction withk= 0.87 plus an opposition correction with the form

a+ bg+ ce−g,

whereg is the phase angle. The empirical constantsa (0.371),b (−0.00185), andc (0.0725) are derived at≥2.5 µm, whereaerosol effects are relatively unimportant. They then examthe emission-angle dependence of photometrically correcteflectance of the surface dust plus overlying aerosols. At showavelengths, corrected reflectances were found to be highlyrelated with the secant of the emission angle—that is, correlwith path length through a uniform layer of aerosols. Theyferred that the emission-angle-dependent part of the reflectspectrum is due to aerosols. The spectrum in Fig. 2b reprethis aerosol reflectance spectrum, with a magnitude extrapoto a phase angle of 20◦ and an emission angle of 0◦.

We used the results of Kirklandet al.’s work to correct ISMdata for both photometric and aerosol effects, and to standareflectance to a geometry of 20◦ incidence angle, 0◦ emissionangle, and 20◦ phase angle. First, radiometrically calibrated dwere corrected using a Minnaert function withk= 0.87, inde-pendent of wavelength. Second, a correction for the opposeffect was made as described above. Third, a smoothedtrum of atmospheric aerosols (a second-order polynomial fiKirkland et al.’s results, Fig. 2b) was subtracted from the daafter the scaling of its magnitude to that expected at sec(e) and20◦ phase angle. Finally, assuming that both incident solar irrance and outgoing surface reflectance are attenuated by aebackscattering, surface reflectance was derived as

RS = Reh sec(e) sec(i ),

where R is photometrically corrected, aerosol-subtractedflectance,RS is the reflectance of the surface as it would apear if illuminated by unattenuated solar irradiance, andh is theidealized aerosol backscattering spectrum shown in Fig. 2b

The effects of photometric correction and aerosol remoon ISM data are illustrated in Fig. 3, for representative speof surface materials having a range of albedos. These reare extremely similar to aerosol-corrected spectra derivedErardet al. (1994). Figure 3a, before corrections, shows prerties typical of Mars in telescopic measurements, particula more negatively sloped continuum for lower albedo materIn Fig. 3b, after photometric correction and aerosol remo
several changes are apparent (Erardet al.1994): (a) the contrast

ols,

ere.th-

edre-

rtercor-tedn-nceentsted

dize

ta

ionpec-t toa,

di-rosol

e-p-

valtraultsbyp-rlyls.

FIG. 3. Representative spectra of normal soils of different albedo. Dat 2.6 to 2.8µm are omitted due to persistence of artifacts of the satur2.7-µm CO2 band. (a) Radiometrically calibrated, atmosphere-removed spe(b) Same spectra after photometric correction and aerosol removal.

between bright and dark regions at 1µm increases from a factoof three to a factor of four; (b) the increasingly negative specslope with decreasing albedo disappears; and (c) the 1-µm ab-sorption appears stronger, especially in lower-albedo matebecause it is no longer “drowned out” by light backscatteredaerosols.

2.4. Parameterization of Spectral Properties

Most of the variance in ISM spectra is related to albedo vations (Bibringet al.1990, Erardet al.1991). More lithologicallysignificant spectral variations, in the depths and positions of meralogic absorptions and in the slope of the spectral continuaccount for only a small part of variance in the data. Thenificant sources of variance were therefore isolated for furanalysis by representing them as five synthetic spectral “pa

continuum, and depths of three mineralogic absorptions.

E

adril

nr

t

elea

gtsmeooa

n

andrp-r-

-lcu-

ed.

n-

eg-of

ns

pla-r

-im-icalsu-oreand

ataoiseara-dis-nedtra of-ared

oper-epth, of

theldednces

ries.s.

gth-.3%

452 MURCHI

FIG. 4. Composite visible-NIR spectra of representative bright and dregions, modified from Murchieet al. (1996). ISM data are shown with solidots. Dashed lines represent the gap in wavelength coverage between the fisecond ISM spectral orders, plus wavelengths of the strongest atmosphersorptions (2.0, 2.7µm) within which surface properties were not analyzed. Solines represent spatially overlapping regions measured at visible wavele(Merenyiet al. 1996a,b). Annotations show the spectral features parametein this study.

Reflectance was represented as the average of photomcally corrected reflectance (RS) in three channels near 1µm. Ifthe surface is predominantly a mixture of bright red dust adark gray rock and sand, reflectance is expected to show srelationship to mineralogic composition.

Spectral slope was represented as the change in reflecover the wavelength interval 1.7 to 2.5µm (1RS/1λ) or

RS,1.7− RS,1.7

1λ.

Greater values of this parameter correspond to more stenegatively sloped continua. In the surface materials, the stymixture of ferric components with an underlying dark substraffects spectral slope. The NIR spectral continuum for uncopacted ferric particles much larger than a wavelength of liis nearly flat, as is the continuum of dark, coarse particulaHowever, due to wavelength-dependent transparency andtering by the ferric particles, an intimate mixture of the satwo components has a negative spectral slope (Morris and N1981). The same effect is observed where ferric material ca dark substrate, such as with an oxidized weathering rindcoating of windblown dust (Singer and Roush 1983, FischerPieters 1993).

For mineralogic absorptions, the general representatioband depth is

1− RS,λ

R∗S,λ,

whereRS,λ is the reflectance at the center of the absorption aR∗S,λ is the interpolated continuum reflectance. We represen

ET AL.

rk

st andc ab-idgthsized

etri-

ndome

ance

plyof

tem-htes.cat-eelyatsr and

of

the 2-µm pyroxene absorption using a band center of 2.15µmand a continuum that was interpolated linearly between 1.72.5µm. The latter wavelengths are on the wings of the absotion, and 2.15µm is the typical band center wavelength for matian dark regions (Mustard and Sunshine 1995, 1996). For 3µmH2O band depth, reflectance within the absorption was calated using the average of three channels near 3µm. The con-tinuum was taken as the reflectance at 2.5µm. Among the weakabsorptions purportedly existing at 2.2–2.3µm, only the∼2.2-µm feature of Bell and Crisp (1993) has been identifiin ISM data (Murchieet al. 1993, Beinroth and Arnold 1996)For this feature we assumed a band center of 2.20µm, and thecontinuum was extrapolated linearly between 2.15 and 2.30µm.Depths of the 3-µm H2O band range from 0.49 to 0.63; the miimum depths of the 2-µm pyroxene band and the 2.2-µm bandare slightly negative. For the broad pyroxene band, slightly native values in bright red dust are attributable to curvaturethe spectra approaching the 3-µm region. For the 2.2-µm band,the magnitude of slightly negative values in dark gray regiowith the weakest absorption (typically−0.002) likely representsresidual systematic calibration error. Consistent with this exnation, the effective uncertainty in 2.2-µm band depth is nea0.003 (Table II; see Section 3.3).

Characteristics of the 1-µm Fe absorption are difficult to determine accurately from ISM data in an automated fashion. Sple parameterizations like those above have limited physsignificance, because this feature actually contains multipleperimposed absorptions from ferric and ferrous minerals. Mrigorous techniques such as Gaussian modeling (MustardSunshine 1995, 1996) are difficult to implement without dhaving an even higher spectral resolution and signal-to-nratio. We therefore did not characterize this feature in a pmeterized fashion, but rather used it to help determine thetinctiveness and lithology of the spectral groupings discerusing the other parameters. To do this, representative speceach grouping were extracted, the depth and position of the 1µmabsorption were characterized, and the results were compwith the known properties of minerals.

In order to assess the independence of derived surface prties from atmospheric path length, we also calculated the dof the 2-µm CO2 band. This was done using calibrated datacourse without correction for atmospheric gas absorptions.

2.5. Intercalibration of Separate Windows

Upon calibration, parameterization, and map projection ofdata, synthetic images of reflectance and spectral slope yienearly seamless mosaics. However, small but abrupt differein strengths of the 2-µm pyroxene, 3-µm H2O, and 2.2-µmmetal–OH bands appeared across the windows’ boundaThis is in contrast to smooth continuity within the windowThese inter-window differences are equivalent to wavelendependent differences in absolute reflectance of typically 0
ndtedor less, but accounting for them is critical to meaningful analysisof the whole data set.

hdi

at

u

-r,n

eachheece,

rec-eter-t reds de-

ndninge pri-r).theow.pa-at all

thetiesma-sis,ns toin-larthe

ngare


FIG. 5. Relationship of 3-µm H2O band depth with phase angle, for brigred Tharsis dust having a 2.3-µm albedo of≥0.3. Band depths were calculateafter photometric and aerosol corrections, but without inter-window normaltion of spectral parameters. Data are fitted with a smooth curve.

In the case of 3-µm H2O band depth, the inter-window variations appear to be related to phase angle. Increasing phaseis known from previous work to increase apparent absorpband depths (Pieters 1983). To quantify the effect of phasegle on the 3-µm band, we examined spectra of Tharsis ddeposits with a 2.3-µm albedo of>0.3 observed at a range ophase angles of 1◦–18◦. Figure 5 shows that 3-µm band increaseswith phase angle, especially at low phase angles. For the 2µmpyroxene and 2.2-µm metal–OH bands the inter-window diffeences in band depth are much smaller, typically 0.5% or lessno phase angle dependence is evident. Possible explanatio

zedrst,

these inter-window differences include instrument effects suchas unremoved dark current or scattered light.

TABLE IPrincipal Components Determined After Photometric Corrections

Correlation CorrelationExplains % Correlation with Correlation with with 2-µm Correlation with with 3-µm

variance reflectance 2.2-µm band pyroxene band spectral slope H2O band Lithologic interpretation

PC1 75.4 +0.722 +0.271 −0.549 +0.013 +0.322 basic mineralogyhigh: altered, ferriclow: less altered, ferrous

PC2 12.3 −0.147 −0.016 +0.088 −0.832 +0.527 water-bearing componenthigh: enchanced contentlow: not enhanced

PC3 7.0 −0.184 −0.248 +0.104 +0.540 +0.776 ferric coatings with water-retaining componenthigh: coatings developedlow: uncoated particulates

PC4 3.8 −0.189 +0.923 +0.283 +0.125 +0.125 atmospheric effects

PC5 1.6 −0.622 +0.110 −0.774 +0.015 −0.018 low-albedo ferric componenthigh: xtalline ferric enriched

Previously, Erard (1995) performed PCA of parameteriISM data. Our approach differs from that one in two ways. Fi


t

za-

-ngle

ionan-st

f

-ands for

To intercalibrate the parameterized measurements ofwindow, inter-window corrections were applied. Although tphase angle dependence of the 3-µm band depth could havbeen corrected using a multiplicative correction to reflectanthis can be approximated with an offset provided that the cortion is small, such as the required≤2%. Offset corrections aralso appropriate for unremoved dark current effects. To demine the corrections we used the extensive deposits of brighdust as a reference material. In each window, a homogeneouposit with a 2.3-µm high albedo of 0.28–0.30 was identified aits spectral parameters were extracted. The window contaiSyrtis Major was used as the standard because it contains thmary reference region for ISM data calibration (Syrtis MajoIn the remaining windows, offsets were applied to equalizeparameters for their dust to those of dust in the Syrtis windThese intercalibrations yield a high degree of continuity inrameterized spectral properties across window boundariesalbedos (Fig. 1).

2.6. Analysis of Spectral Heterogeneity

We used principal components analysis (PCA) to isolateunderlying sources of variation in the NIR spectral properof the surface, and to define groupings of similar surfaceterials into spectral units. In this type of covariance analyspectral variations are represented as weighted contributioeach pixel (or “loadings”) of eigenvectors of the data (“prcipal components”). Pixels with similar loadings have simispectral properties. In principle the eigenvectors describesignatures of underlying lithologic variations, but interpretithem requires geological insight. The principal componentslisted in Table I, and their loadings are mapped in Fig. 1f.

low: ferric-poor pyroxenes

IE

rn

voe

r

g

agivs-

no

r

t),

e

d

inpg

ev

sisgest

heric

c-the

of thede-

ofe

tof

ed.

eresrs is

sm-ionsust

ials.h of

ichers,ats forly, a

gth.

andofrtherized

etionsoth

met-on-etricgehase

ectly

454 MURCH

Erard did not intercalibrate the different windows, so thatsults from each window are not directly comparable. Secowe normalized each spectral parameter to its total range ofation, by expressing it in units of its own standard deviatiThis can be justified because numerically larger variations (20% in 3-µm H2O band depth) are not more important indcators of soil lithology than are parameters with numericasmaller variations (e.g., 5% in pyroxene band depth). Erardstead normalized each parameter to its estimated SNR, impaa dependence on instrument sensitivity at different wavelengIn comparison, the method used here produces no weightinspectral parameters related to instrument sensitivity.

3. ERROR SOURCES AND ANALYSIS

3.1. Radiometric Calibration

ISM data compare favorably with measurements of the sregions of Mars in independent data sets. At wavelen≤1.1µm. Mustard and Bell (1994) compared ISM spectra wwell-calibrated ground-based measurements of spatially olapping regions. The ISM and ground-based measurementhighly correlated (r > 0.9) with no obvious wavelengthdependent departures. Murchieet al. (1999) compared ISMspectra of Phobos with measurements from IMP at≤1.0 µmand found similarly good agreement.

At longer NIR wavelengths, Rivkinet al. (2000) found closeagreement between>2-µm ISM spectra of Phobos and indepedent telescopic measurements. Erard and Calvin (1997) cpared ISM data at>2.2µm with spatially overlapping MarineIRS spectra and found close agreement at 2.2–2.6µm. How-ever, the IRS spectra exhibit a∼15% stronger 3-µm H2O bandthan is present in ISM spectra. This is plausibly an effect ofphase-angle dependence of the 3-µm band depth (Section 2.5though the issue remains open. IRS data were acquired at pangles of 39◦–84◦, compared to<18◦ for ISM data. The largedifferences in phase angle between the two data sets maktercomparison of the phase angle dependence of the 3-µm H2Oband in the two data sets somewhat problematic.

3.2. Atmospheric and Photometric Corrections

Atmospheric gas correction.The simple procedure we useto remove atmospheric gas absorptions assumes non-satuabsorptions, and it does not take into account temperaturepressure depedencies of band shapes or spatial variationsmixing ratios of minor species. We nevertheless used thiscedure because, unlike rigorous radiative transfer modelindoes not require highly accurate knowledge of temperature, psure, and mixing ratio profiles with altitude. Weak atmospheabsorptions near the 1-µm Fe band are consistently removwithout artifacts evident above the noise in the data. Howethere are artifacts around 2.7µm where the very strong CO2
absorption is saturated at lower surface elevations, and nearwings of the strong 2.0-µm CO2 absorption.
ET AL.

e-d,

ari-n..g.,i-llyin-tingths.

of

methsther-are

-m-

he

hase

s in-

ratedand

thero-, it

res-ricder,

In order to minimize effects of these artifacts, our analywas restricted to channels away from the centers of the stronatmospheric absorptions at 1.4, 2.0, and 2.7µm. Several of thewavelengths we used are entirely outside significant atmospabsorptions, including the three channels near 1µm used to pa-rameterize albedo, the 1.7-µm channel used to calculate spetral slope and to define the short-wavelength shoulder ofpyroxene absorption, and the three channels at the center3-µm H2O absorption. However, all three channels used tofine the 2.2-µm band, the 2.15-µm channel near the centerthe pyroxene absorption, and the 2.5-µm channel used to definspectral slope and the shoulders of the pyroxene and H2O ab-sorptions are all within the wings of CO and CO2 absorptions a2.0, 2.35, and 2.7µm. We carefully analyzed the relationshipeach spectral parameter to the depth of the 2.0-µm CO2 absorp-tion, to determine if significant atmospheric artifacts remainTwo methods were used.

First, as shown in Fig. 6, the values of all five parameters wcompared directly with the 2.0-µm CO2 band depth to assesthe occurrence of strong artifacts. None of the five parametemeaningfully correlated with the 2.0-µm CO2 band depth. Un-usually high 1-µm albedos and low 2-µm pyroxene band depthdo occur at low CO2 band depths, but this is due to the predoinance of high-albedo, bright red dust at the highest elevatstudied (in the Tharsis plateau). Any atmospheric artifacts mbe small compared with instrinsic variations in surface mater

Second, as described in Section 4.2, the values of eacthe five principal components were compared with the 2.0-µmCO2 band depth. The fourth principal component (PC4), whexplains<4% of the variance in normalized spectral parametis correlated with the 2-µm CO2 band depth even though thparameter was not an input to the analysis. PC4 accountsmall atmospheric artifacts that were unrecognized previousslight strengthening of both the 2.2-µm band depth and the 2-µmpyroxene band depth with increasing atmospheric path lenThese artifacts arise from over-removal of the 2- and 2.7-µmCO2 absorptions. The short-wavelength shoulder of the 2.2-µmband and the long-wavelength shoulder of the pyroxene bare both artificially elevated, yielding an artificial inflationband depths. To remove these atmospheric artifacts from fuanalysis of surface spectral properties (Section 4.3), we utilthe other principal components and neglected PC4.

Photometric and aerosol effects.There are five possiblsources of uncertainty in the photometric and aerosol correcwe used. First, all reflectance data (including contributions bfrom the surface and from aerosols) were corrected photorically using the phase function derived for Tharsis dust. Ndust surfaces and aerosols may exhibit different photombehaviors. Unfortunately, ISM did not return a sufficient ranof measurements of dark regions to formulate separate pfunctions for different materials.

Second, aerosol removal was performed assuming a perf
theuniform aerosol layer. In fact, there are spatial variations inaerosol opacity of±20–30% as shown by both Erardet al.

as


FIG. 6. Scatterplots of spectral parameters calculated from fully reduced data, against depth of the 2-µm CO2 band from data in which the atmosphere w
not removed: (a) 1-µm albedo; (b) 2.2-µm band depth; (c) 2-µm pyroxene band depth; (d) spectral slope; (e) 3-µm H2O band depth. Slightly negative values forweaker bands result from superimposed spectral curvature, especially in bright red dust.

E

tretimtdgfhrle

r

n

rcel

it

mshe

e

nero

Tbe

at

rra

s

atedones

ageity).xelsphic

edtarypec-hatns

ctthanths,

inma-

456 MURCHI

(1994) and Kirklandet al.(2000b). These should corresponddifferences of 0.01–0.02 in 1-µm reflectance. The combined erors from the photometric correction and from this simplifiaerosol model can be bounded by comparing fully correcspectra of the same region that were measured at different tand geometries. Figure 2c shows two examples, a brighgion (Tharsis) observed at divergent emission angles and aregion (Melas Chasma) observed through an unusually lonmospheric column. The bright region spectra, especially, disignificantly in their uncorrected reflectances because of tdifferent photometric geometries. After correction each paispectra agrees to within 1% reflectance, consistent with theof artifacts expected from differences in aerosol opacity. Tresidual is small compared to the 25% albedo variation obsebetween the brightest and darkest surface regions.

Third, there may be discrete water-ice clouds. If preseclouds might be recognized as discrete patches with an enhaspectral slope and possibly a stronger 3-µm H2O band. If ob-served more than once, the spatial patterns in these two paeters should be time-varying. Several regions with enhanspectral slope correspond with well-known albedo or color ftures and can be attributed to surface materials. These inceastern Syrtis Major, Sinus Meridiani, and the small deposdark material northwest of Sinus Meridiani. Several spotsTharsis with enhanced spectral slopes do not correlate withjor albedo features, and these were identified as candidateoptically thin clouds. Their locations were compared with tViking red-albedo image mosaic acquired over 10 yearslier and were found to correspond with surface deposits havlong-lasting elevated visible-wavelength albedos (Fig. 7). Thpatches, too, are attributable to surface materials.

Only two regions are strongly suspected of having opticathin clouds during their measurement by ISM. The first regioPavonis Mons. In the window we used which covers it, therno evidence for cloud-like properties. However, one of the thwindows we did not use overlaps Pavonis, and in that windPavonis is overlain by a circular patch with enhanced specslope and 3-µm band depth. This cloud does not pertain to oanalysis because it does not occur in the data we utilized.second region is the eastern half of Candor Chasma. There,spectral slope and 3-µm band depth show mismatches betweadjacent windows. The data covering western Candor shosmooth continuity between windows, such that western Cdor appears cloud-free. We therefore exclude eastern (buwestern) Candor Chasma from further analysis.

Fourth, as described earlier, the depth of the 3-µm H2O bandis weakly related to phase angle. We did not attempt to corfor this effect within single windows because it is so poounderstood and likely material-dependent. Uncorrected phangle dependence can lead to band-depth variations of±1% outof the 20% range observed.

Fifth, our aerosol spectrum lacks well-defined absorption
any wavelength due to ferric minerals, silicate minerals, or waice (Fig. 2b). The aerosols should exhibit ferric and H2O features
ET AL.

o-ded

esre-arkat-

fereirofvelheved

nt,ced

am-eda-udeofina-for

ear-ingse

llyisis

eew

tralurheothn

w an-not

ectlyse-

at

FIG. 7. Correspondence of spots on the Tharsis plateau having elevspectral slopes with visible-albedo variations. (top) Spectral slope. Brighter tcorrespond to more negative spectral slopes. (bottom) Viking red-albedo immosaic of the same region (shown only in areas covered by ISM, for clarArrows denote corresponding features. Spatial offsets of up to 2 ISM pioccur due to poorer control of ISM data registration away from large topografeatures.

if they are equivalent to surface dust. The lack of well-definaerosol absorptions is strongly corroborated by complemenmeasurements made from Mars Pathfinder and by the IRS strometer on Mariner 6. IMP spectra of sky radiance show taerosols lack the visible and NIR ferric mineral absorptiopresent in dust on the surface (Thomaset al. 1999; Bellet al.2000). Bellet al.in fact concluded that aerosols must be distinfrom dust on the surface. IRS observed several regions moreonce at different emission angles and atmospheric path lengand found no variations in H2O band depths at either 3 or 6µm.This corroborates the lack of an aerosol 3-µm band in ISM data.Aside from effects of scattered water-ice clouds, variations3-µm band depth must therefore be attributable to surface
terterials (Kirklandet al.2000b). Mars Pathfinder and IRS resultscorroborate the aerosol spectrum we have used in data reduction,

ascs

ro-b

pic

od

,erfnowctaee

ml

iredin-

rin-

sim-eters

riben-, andctralrmalffec-iallyntly

et-wastionscom-thema-

t on

in-allm

yzedset-t theions ofny

end-The


and results from all three data sets show that spectral modemartian aerosols are due for reconsideration.

In their earlier analysis of aerosols from ISM spectra, Eret al. (1994) found possible water-ice bands near the limitdetectability. The data they used cover the suspected cloud orence in Pavonis Mons, possibly yielding a spectrum of aeroplus a discrete cloud instead of a cloud-free spectrum.

Errors in the atmospheric opacity that we assumed for aeremoval would be manifested as errors in 1-µm albedo or spectral slope. Few to no artifacts are expected in parameterizeddepths, because of the featureless shape of the aerosol speand because of the low atmospheric opacity at the>2-µm wave-lengths of the bands. The only possible artifact in band dewould be in the 3-µm band in areas of unrecognized water-clouds.

3.3. Effects on Data Analysis

The uncertainties discussed above can all propagate tsteps of spectral parameterization and PCA. Their summefects were estimated in three ways. First, worst-case “effecuncertainties” in spectral parametersbetweendifferent windowswere determined by measuring the RMS differences betwspatially overlapping observations acquired at different timOverlaps occur in Tharsis, Melas Chasma, Candor Chasmathe intervening plateau. These regions exhibit the full rangspectral heterogeneity in the data, they were observed atcally different emission angles, and they cover nearly therange of elevation differences in the data set. The uncertaiare worst-case because (a) they include errors in registratiwell as data reduction, (b) one of the overlapping windowsacquired with the lowest instrumental gain setting (1 on a sof 1 to 3), and (c) the overlap south of Candor Chasma conone of only two suspected water-ice clouds in the area covby ISM. Worst-case effective uncertainties are listed Tabland shown graphically in Fig. 8. They are 9 to 27 times smathan the range of the spectral parameters’ values, so thatsured spectral heterogeneity is highly significant. In absoterms, the worst-case uncertainties in 1-µm albedo and spectraslope are relatively large because they are dominated by ina
racies in photometric and aerosol corrections. Band depths are
nge

radiometric calibration of ISM data has been shown to yieldts of
largely unaffected by these corrections, so their uncertainties are
TABLE IIEffective Uncertainties and Dynamic Ranges of Spectral Parameters

Effective EffectiveParameter Minimum, maximum Range uncertainty dynamic ra

1-µm albedo, 20◦ phase angle 0.055 to 0.316 0.261 0.0098 272.2-µm M–OH band depth −0.0047 to 0.0071 0.0118 0.0010 122-µm pyroxene band depth −0.0157 to 0.0371 0.0528 0.00288 18Spectral slope −0.0158 to 0.0170 0.0328 0.0037 9

spectra that compare well with independent measuremen

3-µm H2O band depth 0.487 to 0.625


ls of

rdofcur-ols

sol

andctrum

thse

theef-

tive

eenes.andof

adi-ulltiesn asasaleinsredII

llerea-

utelccu-

smaller and more affected by SNR of the data. For data acquat the higher instrumental gains, which cover five of the six wdows, uncertainties in band depths could be up to∼1.7 timessmaller.

In the results of PCA discussed in Section 4.2, the first pcipal component describes “normal” soils that occupy>80%of the study areas. The normal soils are characterized byple relationships between albedo and other spectral param(Fig. 8). Three of remaining four principal components desc“anomalous” data that form outliers from the normal relatioships among albedo, pyroxene band depth, spectral slopeH2O band depth. In the four-dimensional space of these speparameters, the outlying data occur at distances from the norelationships at least 2–6 times larger than the worst-case etive uncertainties. The outlying data clusters are also spatcoherent, and nearly all of them correspond to independeidentifiable geologic or visible-color features.

Second, we bounded the effects of small errors in photomric correction and atmospheric and aerosol removal. PCArepeated using data with all photometric and aerosol correcmade and using data with none made, and the results werepared. Table III shows that the principal components fromunprocessed and fully processed data are quite similar. Thejor results of our work are therefore not strongly dependeneven large errors in data reduction.

Third, the effects of data artifacts peculiar to any single wdow were investigated by running PCA independently onpossible sub-groups of five of the six windows. Artifacts frosingle windows are conceivable because four of the six analwindows were measured at the medium instrumental gainting, one window was measured at the low setting, and one ahigh setting. The results are shown in Table IV. The exclusof any single window has no large effect, showing that resultPCA are not driven by errors in calibration or reduction of asingle window.

3.4. Summary of Error Analysis

We have performed an exhaustive error analysis on theto-end procedures used to reduce ISM data for this study.

0.138 0.0127 11

tion

s of

458 MURCHIE ET AL.

TABLE IIIPrincipal Components Determined Before Photometric Corrections

Correlation Correlation Comparison with principalExplains % Correlation with Correlation with with 2-µm Correlation with with 3-µm component from fully

variance reflectance 2.2-µm band pyroxene band spectral slope H2O band corrected data

1 76.4 +0.612 +0.230 −0.673 −0.278 +0.206 identical to PC1

2 14.2 +0.268 −0.039 −0.218 +0.908 −0.236 spectral slope component ofPC3, plus aerosol effects

3 5.0 −0.263 −0.221 +0.151 +0.263 +0.889 enhanced 3-µm band depthcomponent of both PC2, PC3

4 2.8 −0.180 +0.807 +0.457 +0.165 +0.283 closely similar to PC4

5 1.6 −0.673 +0.495 −0.517 +0.049 −0.178 closely similar to PC5

FIG. 8. Scatterplots of parameterized spectral properties against albedo at 1µm, derived after photometric and aerosol corrections and inter-window correcof spectral parameters. Slight negative values for weaker bands result from superimposed spectral curvature, especially in bright red dust. Error bars show 1-σworst-case effective uncertainties in inter-window comparison of parameters. (a) 2-µm pyroxene band depth. Areas with highly positive and negative loading
PC5 are noted. (b) 3-µm H2O band depth. Areas with high loadings of PC2 and PC3 are noted. (c) 2.2-µm band depth. (d) Spectral slope. Areas with high loadingsof PC2 and PC3 are noted.


TABLE IVPrincipal Components from Subsets of Five Windowsa

Correlation CorrelationExplains % Correlation with Correlation with with 2-µm Correlation with with 3-µm

variance reflectance 2.2-µm band pyroxene band spectral slope H2O band

PC1 75.4 +0.722 +0.271 −0.549 +0.013 +0.322Mean, st. dev. 75.2± 3.6 +0.718± 0.023 +0.274± 0.039 −0.545± 0.022 +0.002± 0.076 +0.322± 0.050Min., max. 68.9, 79.2 +0.694,+0.760 +0.222,+0.344 −0.564,−0.503 −0.144,+0.071 +0.224,+0.356

PC2 12.3 −0.147 −0.016 +0.088 −0.832 +0.527Mean, st. dev. 12.5± 2.0 −0.150± 0.091 −0.028± 0.122 +0.089± 0.063 −0.815± 0.089 +0.514± 0.110Min., max. 10.610, 16.170 −0.257,−0.006 −0.241, 0.098 −0.014,+0.161 −0.931,−0.668 +0.357,+0.684

PC3 7.0 −0.184 −0.248 +0.104 +0.540 +0.776Mean, st. dev. 7.1± 1.4 −0.170± 0.055 −0.252± 0.154 +0.096± 0.038 +0.529± 0.111 +0.761± 0.076Min., max. 5.6, 9.2 −0.265,−0.117 −0.415,−0.042 +0.053,+0.136 +0.352,+0.667 +0.622,+0.829

PC4 3.8 −0.189 +0.923 +0.283 +0.125 +0.125Mean, st. dev. 3.6± 0.4 −0.184± 0.158 0.887± 0.039 0.281± 0.137 0.124± 0.134 0.125± 0.142Min., max. 3.1, 4.2 −0.462,−0.032 +0.827,+0.931 +0.035,+0.430 −0.073,+0.323 −0.085,+0.356

PC5 1.6 −0.622 +0.110 −0.774 +0.015 −0.018Mean, st. dev. 1.6± 0.1 −0.604± 0.068 +0.113± 0.176 −0.765± 0.051 +0.028± 0.049 −0.009± 0.061Min., max. 1.4, 1.7 −0.642,−0.467 −0.197, 0.299 −0.853,−0.703 −0.029, 0.106 −0.073, 0.093

a Numbers shown are for whole data set (first line), mean and standard deviation of six sub-sets (second line), minimum and maximum among six sub-sets(third line).

h

.h

don

s

cntn

.eifties

of

nsionarkut

tentreass,nly

its

p-r,

tion

c-herayfer-

slyedns,

so dark-region spectra are simply most affected by the nega-

the same parts of Mars, with the exception of some publisIRS spectra at 2.6–3.1µm. In the latter case, differences in thphase angles of the observations may be partly responsiblecorrections we used for atmospheric gases have been sto yield only minor artifacts at wavelengths which were anlyzed further, and these artifacts can be isolated using PCA.aerosol spectrum we used to remove the spectral effects omospheric particulates is strongly corroborated by indepenMariner 6/7 IRS measurements. We have also shown that prgated uncertainties in our procedures are much smaller thaspectral variations in surface materials, and that results ofdata analysis would be only weakly affected by large errortreatment of the atmosphere, had they occurred.

4. RESULTS AND DISCUSSION

4.1. Overview of Spectral Heterogeneities

In the “classical,” two-component bright dust/dark mafic roand sand model of the martian surface layer (e.g., Christeand Moore 1992), all surface lithologies and spectral propershould be explainable by physical mixing of dust with rock asand. The measured relationships among albedo, band deand spectral slope are displayed graphically in Fig. 8, andspatial variations in spectral parameters are mapped in Figs1e. The 3-µm absorption, the 2-µm pyroxene absorption, and th2.2-µm metal–OH absorption do exhibit some correlation walbedo. However, there are discrete regions with departuresthe normal relationships much larger than effective uncertain
in the spectral parameters. Regions with such anomalous p
edeTheowna-Thef at-entpa-the

ourin

kseniesd

pths,the1b–

throm

erties are candidates for different lithologies or the presenceadditional mineralogical components.

Figure 8a shows that depth of the 2-µm pyroxene absorption(Fig. 1c) is inversely correlated with albedo in most regio(Fig. 1b). However, several regions exhibit a weaker absorptthan expected purely from albedo. The clearest example is dred soil in Lunae Planum, which is intermediate in albedo bnearly lacks a pyroxene absorption, like dust. To a lesser exeastern Syrtis Major has a weakened pyroxene band, whedark gray soils in western Syrtis Major and on the floors of MelaEos, and Capri chasmata in Valles Marineris have uncommostrong absorptions. The 2.2-µm absorption (Fig. 8c) is morehighly correlated with albedo, so that the small variations indepth (≤1%) can mostly be explained by mixing.

The depth of the 3-µm band (Fig. 1d) exhibits a complicatedrelationship with albedo. The strength of this absorption is tyically greater in bright than in dark regions (Fig. 8b). Howevethere are discrete regions of all albedos in which the absorpis stronger than in normal bright regions. Overall, 3-µm banddepth is therefore only loosely related to reflectance.

The most complicated relationship with albedo is that of spetral slope (Fig. 1e and Fig. 8d). Spectral slope is low in tbrightest and darkest materials. Previous claims that dark gsoils have a consistently negative slope due to a pervasiveric contaminant (e.g., Singeret al. 1979) result from viewingthe soils through atmospheric aerosols. As shown previouby Erardet al. (1994), the additive component of backscatterlight is largest relative to surface reflectance in dark regio

rop-tively sloped aerosol spectrum. However, some intermediate- to

t-

rke

stern

460 MURCHIE ET AL.

TABLE VProperties and Geologic Correlations of ISM Spectral Units

False color, Princ. compon. 3-µm “water” 2-µm pyrox. SpectralDesignation Visible color PCA map loadings band deptha band deptha slopea 1-µm banda Geologic correlations

Normal soilsNormal bright bright red deep pink 1 high intermed/strong absent flat 0.85–0.88µm, bright red, low thermal

2 low shallow inertia “dust”3 low

Normal bright red dull orange 1 intermed weak/intermed weak/intermed negative broad, Libya Montes, brightransitional 2 low intermed dark borders

3 intermed wvl.

Normal dark I dark gray dark blue 1 low weak v. strong flat ∼1µm, W Syrtis Major,2 low deep Margaritifer Terra3 low

Normal dark II dark gray dark green 1 low weak intermed./ negative∼1µm, E Syrtis Major, Ophir2 low strong moderate Planum3 intermed depth5 high

Anomalous soilsAnomalous dark gray light blue 1 low strong v. strong flat ∼1µm, mottled unit in Melas,

dark 2 high very deep Eos chasmata3 low5 low

Anomalous dark gray to bright green 1 intermed strong weak to negative broad, soils at borders of datransitional dark red to yellow 2 low intermed. intermed wvl. red regions: SE Luna

3 high to 1µm Planum, Sinus Meridiani

Anomalous dark red magenta 1 high v. strong weak to absent flat 0.85–0.90µm, “dark red” plains in Oxia,bright I 2 high moderate Lunae Planum

3 low depth5 high

Anomalous bright red bright bluish 1 intermed strong weak highly broad, layered material in webright II green to high negative intermed Candor Chasma

2 low wvl.3 high

n

es

r

ndw

or

lsasie

dakarknly

oilsate-doestig-g

d,

a Qualitive descriptors pertain to the range of values shown in Figs. 6b a

high-albedo soils do exhibit the negative continum expectedferric–ferrous mineral mixtures, for example, in Libya Montand Noctis Labyrinthus. There are also restricted locationall albedos with locally very strong negative spectral slopespecially eastern Syrtis Major, Sinus Meridiani, and layematerial in western Candor Chasma.

Overall, the principal source of spectral heterogeneities ulated to albedo is the 3-µm H2O band. In 81% of the analyzesurface, band depth is closely related to albedo (Fig. 8b),higher albedos corresponding to stronger absorptions. As sthese areas are given the short-hand designation “normal sThe remaining 19% of the study area consists of those awhich have a stronger 3-µm H2O absorption than in normal soiof comparable reflectance, hence our short-hand design“anomalous soils.” They include dark gray layered depositMelas and Eos chasmata, intermediate-albedo dark red soOxia and Lunae Planum, bright red layered deposits in w
ern Candor Chasma, and intermediate-albedo materials in SMeridiani.
d 6c and 8a and 8c.

forsof

es,ed

re-

ithuch,ils.”eas

tionin

ls inst-

In their attributes other than 3-µm band depth, anomaloussoils are highly varied (Table V). For example, the dark resoils in Oxia and Lunae Planum have an uncommonly wepyroxene band for materials of their albedo, whereas the dgray materials in Melas and Eos chasmata have an uncommostrong pyroxene band (Fig. 1c). Both of these anomalous stypes have the lowest observed spectral slopes. Intermedialbedo anomalous soils in Sinus Meridiani and high-albeanomalous soils in western Candor Chasma have the highobserved spectral slopes (Fig. 1e). The possible lithologic snificances of these variations are explored in the followinsection.

4.2. Principal Components and their PossiblePhysical Significances

The results of PCA of photometrically and aerosol-correcte
inusnormalized, parameterized data are summarized in Table I. Es-timated uncertainties in each of the principal components are

p

ly

re

ea

oa

ni

no

h

c

andntatheondatab-

gh-umir.m-

opeght inandibitavehedoeenst-in

bedy tover,

tionsor-

that

uldl-rred

edctedns

ne oricharects

insndose

nt


given in Table IV. In summary, PC1, PC2, and PC3 account∼94% of the variance in the spectral parameters. PC1 resents the first-order spectral variations correlated with albePC2 and PC3 represent enhancements in 3-µm H2O band depth,each accompanied by different variations in the other spectrarameters. All three principal component loadings are displain Fig. 1f in the red, blue, and green image planes, respectivLarge areas of the surface have distinctive principal componloadings, forming units with spatially coherent, similar spectproperties. These spectral units correspond to albedo featurgeologic formations. PC4 and PC5 each accounts for<4% ofthe variance in the spectral parameters. PC4 is closely relateatmospheric path length, and it encapsulates small artifacts fremoval of atmospheric gas absorptions. PC5 loading is sigicant in regions with an unusually strong or unusually we2-µm pyroxene band. High-PC5 regions, with weak pyroxebands, also have 1-µm absorptions which are unusually shoin wavelength for materials of their albedo. That is, they appunusally ferric in character. Areas with a high PC5 loadinghachured in red in Fig. 1f, and areas with a low PC5 loadinghachured in green.

PC1. The first principal component, which accounts f75.4% of the variance in the normalized parameterized dis dominated by moderate positive correlations of albedo w2.2-µm metal–OH and 3-µm H2O band depths, and a stronanticorrelation of albedo with 2-µm pyroxene band depth. Ahigh loading of PC1 occurs in all bright soils, and a low loadioccurs in all dark soils. High-PC1 soils have a ferric absorptnear 0.86µm, whereas low-PC1 soils have 1- and 2-µm bandsconsistent with pyroxene.

Interpretation of PC1. Laboratory experiments show thaspectral variations analogous to those represented by PC1 rfrom physical mixing of pyroxene-rich material with brighteferric- or clay-containing material (Singer 1981, 1982, Singand Roush 1983, Fischer and Pieters 1993). The mineralsgested by high loadings of PC1 include ferric minerals (edenced by the short wavelength position of the 1-µm band), hy-droxylated minerals (suggested by the 2.2-µm absorption), andwater-bearing salts or silicates (indicated by the greater streof the 3-µm band). These are predicted by thermodynamic mels to result from oxidation and hydration of a pyroxene-ricbasaltic parent material (Gooding 1978, Goodinget al. 1992).PC1 is therefore most simply interpreted as representing thedient in overall spectral properties between relatively unalterbasaltic rock fragments and bright red dust (an altered weating product).

PC2 and PC3. One or the other of these two principal components has a high loading in all of the regions with enhan3-µm H2O band depths. However, PC2 and PC3 have differrelationships to variations in spectral slope. PC2 accounts12.3% of the variance in the normalized parameterized data
is dominated by a strong positive correlation with 3-µm banddepth and a strong anticorrelation with spectral slope. High P

forre-

do.

pa-edely.entals or

d toromnif-aknertarre

are

rta,

ithg

gon

tesultrersug-vi-

gthd-h,

gra-ed,er-

-ed

entforand

loadings correspond to enhanced 3-µm absorptions (Figs. 1dand 8b) and the lowest observed spectral slopes (Figs. 1e8d). Areas with high PC2 loadings have a light bluish to magehue in Fig. 1f, depending on the loading of PC1 (shown inred image plane). Low-albedo, high-PC2 materials correspwith restricted dark gray soils in Melas, Eos, and Capri chasm(light blue in Fig. 1f). These dark soils exhibit the strongest oserved 2-µm pyroxene band depths. Intermediate-albedo hiPC2 materials correspond with dark red soils in Lunae Planand Oxia (light pink in Fig. 1f). The short wavelength of the1-µm band also indicates a ferric-dominated Fe mineralogy

PC3 accounts for 7% of the variance in the normalized paraeterized data and is strongly correlated both with spectral sland with 3-µm band depth. In Figs. 8b and 8d, areas with hiloadings of PC3, like areas with high PC2 loadings, stand outhe plot of 3-µm band depth vs albedo as having enhanced bstrengths. In the plot of spectral slope vs albedo they exhthe most negatively sloped continua. High-PC3 regions ha light greenish to yellowish hue in Fig. 1f, depending on tloading of PC1. Spatially, they correspond with the high-albelayered deposits in western Candor Chasma (light yellow grin Fig. 1f) and low- to intermediate-albedo soils in southeaern Lunae Planum and Sinus Meridiani (yellow to light greenFig. 1f).

Interpretation of PC2 and PC3.The two major possibilitiesfor the physical significance of PC2 are differences in absorwater content and compositional variations. We have no waascertain the possible importance of adsorbed water. Howeif adsorbed water is responsible for enhanced 3-µm band depth,the abruptness of the boundaries in some band-depth variaand the variations’ correspondence with albedo, color, and mphologic features are all suggestive of material propertieslead to differing adsorption of H2O.

Among compositional variations, at least four phases coplausibly account for enhanced 3-µm band depths. One is sufates, whose presence in the surface layer has been infefrom Viking Lander and Mars Pathfinder data (Toulminet al.1977, McSween and Ghosh 1999, Bellet al.2000). Bishop andPieters’s (1995) and Bishop and Murad’s (1995) works showthat ferric sulfates such as schwertmannite, which are predithermodynamically to form under martian surface conditioby acid aqueous weathering (Burns 1994), retain a strong 3-µmband against desiccation. The second possible phase is omore hydrated ferric oxyhydroxides such as ferrihydrite, whexhibit similar water-retention under desiccation and whichalso geochemically reasonable martian alteration produ(Burns 1994). The third is hydrovolcanic glass, which retawater due to its dissolution in the mineraloid matrix (Farraand Singer 1991). A fourth is hydrous carbonates, whpresence in the surface layer was proposed by Calvinet al.(1994). Hydrous carbonates have only weak CO3 overtones at2.35 and 2.55µm, so their presence might not be inconsiste

C2with the weakness or lack of these features in ISM data (cf. Erardet al.1991).

IE

-

tMea

io

ix

n

tih

-nsriicif

-e

oc

d

dnSn

i

uax

-n

i

nger) didgthheogy,ar-t-Ca

a py-le.uld., orPC5areda).

alonly

gths.tions.

462 MURCH

The strong correlation of PC3 with 3-µm band depth suggests that high PC3 surface materials also retain H2O againstdesiccation, and it may thus include one or more of the waretaining phases suggested above. Laboratory studies ofanalog materials (Singer and Roush 1983, Fischer and Pi1993) have shown that either a bright ferric coating on a dsubstrate or mixing of bright ferric material with dark particulamaterial will introduce a negative spectral slope consistent wPC3. A simple physical explanation for PC3 and its distinctfrom PC2 is that in PC2 the water-retaining phase is mixed wbright red dust whereas in PC3 the mixture coats or is admwith dark substrate.

PC4. The fourth principal component accounts for on3.8% of variance in the normalized parameterized data, ais dominated by a strong positive correlation with strengththe 2.2-µm absorption. There is also a weak positive correlawith the 2-µm pyroxene band depth. PC4 accounts for one-tof the variance in 2.2-µm band depth and<1% of the variancein 2-µm pyroxene band depth.

Interpretation of PC4. Alone among the principal components, PC4 loading is correlated with atmospheric path lein each window (meanr , 0.46). High loadings occur consitently in the topographically low regions (e.g., Valles MarineIsidis) regardless of their other spectral attributes, and loaddecrease with increasing elevation. The lowest loadings oover Ascraeus Mons. PC4 appears to represent small artfrom over-removal of the 2- and 2.7-µm CO2 absorptions. Theshort-wavelength shoulder of the 2.2-µm band and the longwavelength shoulder of the pyroxene band are artificiallyvated, yielding an artifical inflation of band depth.

PC5. The fifth principal component accounts for 1.6%the variance in the normalized parameterized data and isacterized by anticorrelations with both albedo and 2-µm pyrox-ene band depth. High loadings imply darkening accompanieweakening of pyroxene absorptions, the reverse of the strenening of pyroxene absorptions at lower albedos representePC1. The highest PC5 loadings occur in dark red soil in LuPlanum and in restricted dark gray soils, principally easterntis Major (hachured red in Fig. 1f). High-PC5 regions are distion the plot of 2-µm pyroxene band depth against reflectan(Fig. 8b), where they form a “low pyroxene band depth” outlto the quasi-linear, inverse relationship between pyroxene bdepth and reflectance. Conversely, low PC5 loadings (hachin green in Fig. 1f) imply enhanced pyroxene absorptions,correspond with the darkest soils having the strongest pyrobands.

Interpretation of PC5. Variations in PC5 loading could result from particle size differences that cause absorption strevariations, or from mineralogic differences. One possible meralogic difference is the composition of pyroxene itself, whwould be evidenced by variations in the relative depths and
wavelength positions of the 1- and 2-µm pyroxene absorptions.Low-Ca pyroxenes have a stronger 2-µm band than high-Ca py-
ET AL.

er-ars

tersrk

teithn

ithed

lyd itofonird

gth-s,ngscuracts

le-

fhar-

bygth-

byaeyr-ctceerandredndene

gthin-chthe

roxenes and might account for areas of low PC5 (i.e., stropyroxene absorption). Mustard and Sunshine (1995, 1996identify variations within martian dark regions in the wavelenposition of the 2-µm band and in its depth relative to that of t1-µm band. Using Gaussian modeling to estimate mineralthey inferred differences in Ca content. However, low-PC5eas include examples both of Mustard and Sunshine’s lowespyroxene (Eos and Melas Chasmata) and of the highest-Croxene (Syrtis Major), so this explanation appears untenab

Alternatively, in high-PC5 areas, a low-albedo mineral codarken soils without contributing a 2-µm band like pyroxenePlausible, low-albedo minerals include magnetite, ilmeniteferric minerals. To test these hypotheses, spectra from high-dark red soils in Lunae Planum were extracted and compagainst spectra of normal soils with similar albedos (Fig. 9

FIG. 9. Comparison of dark red “anomalous bright I” soils with normbright and transitional soils having comparable albedos. Data are shownout to 2.6µm to accentuate the spectral differences at shorter wavelen(a) Calibrated spectra after atmospheric, photometric, and aerosol correc
(b) Same spectra, scaled to unity at 2.3µm. Lower-albedo dark red soils exhibita characteristic ferric absorption near 0.9µm.

In

e

xnu

enfren

iaeigw

e

ard

dro

fi

ea

heag

i

edarkectraosols

nsi-ght

reer,ned

heveeirend-

tion

htgs,p-11.ts inoothr-ark

sng

ent aarklow


The dark red soils exhibit a consistently lesser spectral slopea shorter-wavelength 0.9-µm band than do the normal soils.Fig. 9b the dark red soil spectra are scaled to unity at 2.3µm.With decreasing albedo, the ferric absorption increases in dand shifts slightly in wavelength position, from∼0.86 µm tonear 0.90µm. This wavelength shift is the opposite of that epected from the change in spectral slope, implying a differein the 0.9-µm absorption. Nevertheless, the overall spectrremains dust-like in appearance, and there is no accompan2-µm absorption to suggest the presence of pyroxene. The dest, longest-wavelength ferric absorption occurs in the regioLunae Planum having the highest loading of PC5. We thereinterpret high loadings of PC5 to represent a crystalline femineral component. The deepened pyroxene bands represby low loadings of PC5 might indicate “clean,” relatively uweathered pyroxenes or, alternatively, particle size differen(cf. Crown and Pieters 1987).

4.3. Principal Components, Spectral Units

Normal soils. In the principal components loading mapsFig. 1f, normal soils form four groupings whose propertiessummarized in Table V. “Normal bright soil,” which has a depink hue in Fig. 1f, corresponds to most soils that are brred at visible wavelengths. Representative spectra are shoFig. 3b. Normal bright soil exhibits a relatively strong 3-µmabsorption, a very weak but still evident 2.2-µm absorption, andno 2-µm absorption indicating the absence or near absencpyroxene. A shallow, ferric absorption band with a minimunear 0.86µm and a flat-to-positive spectral slope also typifyof these soils. This grouping corresponds spatially to the labland areas of low thermal inertia interpreted as deposits of(Christensen 1986, Christensen and Moore 1992), such thaterms “normal bright soil” and “dust” are synonymous.

“Normal dark I soil” is dark blue in Fig. 1f and corresponto a sub-group of dark gray soils occurring in western SyMajor and Margaritifer Terra. Representative spectra are shin Fig. 3b. These soils exhibit a relatively weak 3-µm band,no 2.2-µm feature, and moderately strong, broad 1- and 2-µmabsorption bands consistent with pyroxene. Spectral slopflat. All of these properties are consistent with criteria deing weakly altered mafic soils (cf. Mustard and Sunshine 191996), such that normal dark I soil apparently represents rtively clean exposures of mafic crustal material with the leferric contamination or alteration.

“Normal dark II” soils are dark green in Fig. 1f. Althougtheir albedo is similar to that of normal dark I soils, they havmore negatively sloped spectral continuum (Fig. 10) and wepyroxene absorptions. They also exhibit an elevated loadinPC5, which as explained above is interpreted as an enrichmecrystalline ferric minerals. Geographically, normal dark II sooccur predominantly in eastern Syrtis Major.

Principal components analysis also distinguishes “transitio
soils” along the boundaries of normal bright and normal dasoils. Transitional soils have a dull to medium orange hue

and

pth

-cem

yingep-of

orericnted-ces

nrephtn in

ofmllge,ust

t the

stiswn

e isn-95,la-st

akerof

nt inls

nal

FIG. 10. Spectrum of normal dark II soil in eastern Syrtis Major, comparwith other normal soils. Eastern Syrtis Major is distinguished from other dsoils by a more negative spectral slope and weakened absorptions. All sphave been corrected photometrically and had light backscattered by aerremoved.

Fig. 1f. Representative spectra are shown in Fig. 3b. Trational soils have loadings of PC1 intermediate to normal briand dark soils, as well as intermediate 1-µm albedo, absorptiondepths, and 1-µm band positions. All of these properties aconsistent with a mixed ferric-ferrous composition. HowevPC3 loading is slightly elevated, consistent with the steepespectral slope (e.g., Fig. 8d).

Anomalous soils. Anomalous soils are also separable on tbasis of principal component loadings into four groups. All hain common a high loading of either PC2 or PC3, due to thenhanced 3-µm bands. Which loading is high, as well as thloadings of PC1 and PC5, differs between the groups depeing on their albedo, spectral slope, and Fe-mineral absorpcharacteristics.

The first anomalous group is “anomalous dark” soils (ligblue in Fig. 1f), which are characterized by low PC1 loadinhighly positive loadings of PC2, and low loadings of PC5. Reresentative spectra are compared with normal soils in Fig.These soils occur on low-albedo layered and related deposiEos and Melas Chasmata, and in Capri Chasma where smplains material is ponded within an outflow channel. All occurences are dark gray at visible wavelengths. Anomalous dsoils exhibit very strong absorptions at 1 and 2µm indicativeof pyroxene, a flat spectral slope, and a strong H2O absorption.The high strength of both the H2O and the pyroxene absorptionis in contrast with the situation in normal soils, where a stropyroxene band is accompanied by a relatively weak H2O band.As such, the anomalous dark soils apparently do not represdistillation of the pure dark component present in normal dI soils. Their low albedo, strong pyroxene absorptions, and
rkinspectral slope are consistent with a pyroxene-rich lithology hav-ing minimal or no ferric contamination.

E

os

ovrtt

I

)dLl

t

h

ioee

teyh

al

Datarrec-s

t ass arema-andred,

nal

. Thein

lly

edicalely

ex-tralbers.po-

form-

464 MURCHI

By far the largest geographic occurrence of anomalous sare “anomalous bright I,” corresponding to the dark red vible color unit in Oxia and Lunae Planum. These are magein Fig. 1f. Representative spectra are shown in Fig. 9. Sintermediate between this unit and normal bright soils haveible color intermediate to bright and dark red, and are colomedium pink in Fig. 1f. They are widespread and occupyTharsis plateau adjacent to Noctis Labyrinthus, the floor ofIsidis basin, and parts of Libya Montes. Anomalous bright I sohave a medium reflectance, a shorter-wavelength 0.9-µm bandthan in normal soils of comparable reflectance, a flat specslope, and a strong 3-µm H2O absorption. Anomalous brightsoils have a medium-to-high PC1 loading, plus a very high Ploading reflecting their unusually flat spectral slope and stro3-µm band. The 3-µm band in Lunae Planum is the strongeobserved in any region.

“Anomalous bright II” soils (bright yellow green in Fig. 1fare unique to the bright red interior deposits of western CanChasma. Representative spectra are shown in Fig. 11.the lower-albedo anomalous dark soils elsewhere in VaMarineris, anomalous bright II soils exhibit an enhanced 3-µmH2O band. However, the soils of western Candor are disguished by the most negatively sloped spectral continuumany observed soils. PC1 loadings are high, consistent with tintermediate to high albedo and the ferric 1-µm bands in Fig. 11.PC3 loadings are also very high, consistent with their enhan3-µm band strength and spectral slope.

“Anomalous transitional” soils (bright green to yellow huin Fig. 1f) cover a range of low to intermediate albedos. Th

FIG. 11. Spectra of anomalous soils covering Valles Marineris interdeposits, compared with normal soils. Anomalous dark soils, in Eos and MChasmata, are characterized by strong pyroxene absorptions and flat spslopes. Anomalous bright II soils, which occur on layered material in wesCandor Chasma, are characterized by a ferric 1-µm band and the most negativelsloped spectral continuum. Data shown are calibrated spectra after atmosp
photometric, and aerosol corrections. They are plotted only out to 2.6µm toaccentuate the spectral differences at shorter wavelengths.
ET AL.

ilsi-ntailsis-edheheils

tral

C2ngst

orikeles

in-ofeir

ced

eey

rlasctralrn

eric,

FIG. 12. Spectra of anomalous transitional soils compared with normsoils. Anomalous transitional soils are characterized by enhanced 3-µm H2Oabsorptions, but in other respects they resemble normal transitional soils.shown are calibrated spectra after atmospheric, photometric, and aerosol cotions. They are plotted only out to 2.6µm to accentuate the spectral differenceat shorter wavelengths.

are characterized by an elevated PC3 loading, though nohigh as in western Candor Chasma, and their spectral slopenot as negative as in western Candor. Geographically, anolous transitional soils occupy southeastern Lunae PlanumSinus Meridiani. Both occurrences are at the margins of darkanomalous bright I soils. Aside from their strong 3-µm band,anomalous transitional soils mostly resemble normal transitiosoils—with intermediate reflectances, intermediate 2-µm py-roxene absorptions, and a moderately steep spectral sloperepresentative spectra, shown compared to normal soilsFig. 12, would be undistinguished except for their unusuastrong 3-µm band.

5. IMPLICATIONS FOR SURFACE COMPOSITION

5.1. Components of the Surface Layer: Comparisonwith Other Data

Most of the spectral variance in ISM data can be explainby differences in the relative abundances of the two canoncomponents of the surface layer, altered ferric dust and relativunaltered, pyroxene-containing mafic material. PC1, whichplains three-fourths of the normalized parameterized specvariance, represents a gradient between these two end-memThe NIR spectral attributes of these two major surface comnents are summarized in Table VI.

We interpret the results of PCA also to reveal evidenceat least two additional components (Table VI). The third co
ponent is the crystalline ferric phase(s) interpreted to occur inhigh-PC5 regions, especially Lunae Planum and eastern Syrtis

nophase,


TABLE VISummary of Spectral Evidence for Components of Surface Layer

Possible lithologicComponent Spectral characteristics PCA characteristics Type locations interpretation(s)

Bright ferric component high albedo, ferric absorption at high PC1, low loadings Tharsis plateau, eastern dust: assemblage of na0.85–0.88µm, no pyroxene, of other principal Arabia and crystalline ferric phasemoderate 3-µm band, 2.2-µm components poorly crystalline clays,band present ±other minerals

Dark ferrous component low albedo, 1- and 2-µm pyroxene low PC1, low loadings Western Syrtis Major, mafic rocks and sand: richabsorptions, weak 3-µm from other principal Margaritifer Terra in pyroxeneband, 2.2-µm band absent components

Water-retaining component enhanced 3-µm band high PC2 or PC3 loading Oxia, Lunae Planum water-containing phase(s)

Dark ferric component lowered 1-µm albedo without high PC5 loading Lunae Planum, eastern crystalline ferric mineral(s)increased pyroxene band, Syrtis Major0.66-µm and/or 0.9-µm band

n

ed

mpnu

oe

e

ein

ds

r-

ates.ow

er-ting

segy

beheyred

cor-fateonallainer

soills atnse tos.f

ix-thatby ance,s for

Major. The fourth component is the phase(s) responsibleenhancements in the depth of the 3-µm H2O band.

A variety of spectral data from both ground- and space-basensors corroborates the occurrence of regional concentraof low-albedo, crystalline ferric minerals (Bell 1992, Bellet al.1997, Merenyiet al. 1996a, 1996b). Eastern Syrtis Major, oof two large low-PC5 regions, has among the strongest obseferric absorptions near 0.66 and 0.86µm (Bell 1992, Bellet al.1997). These bands are present in several of the ferric minproposed to occur on Mars based on thermodynamics, incluhematite, goethite, and schwertmannite (Shermanet al. 1982,Morris et al.1985, Bishop and Murad 1997). Although the seond large low-PC5 area (Lunae Planum) is not well-coveredtelescopic measurements, analogous dark red soils weresured at the Mars Pathfinder landing site. IMP revealed stral properties similar to those in ISM measurements of LuPlanum. That is, the dark red soils have a spectral contincomparable to that in the dust, but a deeper ferric absorpcentered near 0.90µm. The IMP spectra of dark red soils do nreveal a 0.66-µm absorption like that in eastern Syrtis, howevand more resemble maghemite or ferrihydrite (McSweenet al.1999).

Ground- and space-based observations also corroboratexistence of regional enhancements in 3-µm H2O band depth.Telescopic spectra (e.g., Bell and Crisp 1993) reveal broscale regional differences in band depth. Higher spatial restion measurements from IRS, covering eastern Valles Marinshow regional variations in 3-µm band depth comparablemagnitude to those seen in the same region in ISM data (Ca1997).

Phases possibly responsible for enhancements in 3-µm banddepth could include hydrated carbonates, ferric oxyhydroxiand hydrated sulfates. Remote sensing at other wavelength
in situanalyses of elemental compositions are useful to evaluwhich phase may be abundant enough to account for the s
for

sedtions

erved

ralsing

c-byea-

ec-aeumtiontr,

the

ad-olu-ris,

lvin

es,and

percentage of water in surface materials (Houcket al.1973, Yenet al. 1998) yet spatially variable enough to account for diffeences in 3-µm band depth. Neitherin situcompositional analysisnor orbital remote sensing (Bell 1996, Christensenet al.2000a,2000b) has revealed evidence for significant exposed carbonIn situ compositional measurements by Mars Pathfinder shthat there is little variation in the total abundance of iron minals in altered soils, yet the spectral properties vary, suggesdifferences in the mineralogy present (McSweenet al. 1999,Bell et al.2000). If some of the iron occurs in a hydrated phasuch as ferrihydrite, then differences in ferric iron mineralocould be partly responsible for differences in 3-µm band depth.

Another possible water-retaining phase thought both toabundant and to vary spatially in abundance is sulfates, if toccur in hydrous form. Sulfates’ presence on Mars is inferfrom in situ analysis (Toulminet al. 1977, Riederet al. 1997),and at the Viking landing sites the cohesiveness of the soil isrelated with sulfur content, suggesting the occurrence of sulsalts as a cementing phase (Clark and Van Hart 1981). Regivariations in abundance of sulfates may be required to expthe compositions of bright red soils at the Viking and Pathfindlanding sites (McSween and Ghosh 1999). The bright redat the Pathfinder site has a much lower S content than soithe Viking sites. If surface materials are simply combinatioof rock and dust, then dust at the Pathfinder site would havbe significantly poorer in S than dust at the two Viking siteThis is difficult to reconcile with the extensive eolian mixing odust thought to occur on Mars (e.g., Greeleyet al. 1992). Thisproblem is resolved if elemental compositions result from ming of three components, rock, dust, and sulfate cements. Incase, differences between the sites can be explained simplygreater sulfate abundance at the Viking sites. From this evideferric minerals and sulfate salts both appear to be candidate
ate
malla water-retaining phase evidenced by regional enhancements inthe 3-µm H2O band.

E

no

iin

n

9

er

een

howlorea-ent.hysi-fur-

blyockyaedo

othofLeeaxrmsrvedight

rtisrate

allyer-y-rys-

466 MURCHI

5.2. Normal Soils: Evidence for Mixing of Maficand Dust Components

Normal bright, normal transitional, and normal dark soils ehibit a continuum of albedos, band depths, and PC1 loadiThis is strongly suggestive of physical mixing of two compnents, dark mafic rock and sand and the altered, ferric dust cposing bright red dust/normal bright soil. In fact, all regions wdark gray visible color, whether normal or anomalous, exh1- and 2-µm absorptions consistent with pyroxene. The ospectral attribute of normal soils which does not vary lineawith albedo is spectral slope, which is low in dark and brignormal soils but higher in the intermediate-albedo transitiosoils. These properties provide evidence for physical mixingmafic and dust end-members. The spectral slopes of both loporous ferric particles and dark, coarse particulates are neflat. However, an intimate mixture of the same two componeor a ferric coating on a dark, mafic substrate has a negaspectral slope (Morris and Neely 1981, Singer and Roush 1Fischer and Pieters 1993).

Based on their strongly negative spectral slope but intermate values of other spectral attributes, we interpret normal tsitional soils to be simple physical mixtures of normal brigsoil and normal dark I soil. For illustration, in Fig. 13, the thrnormal soil types are represented on a continuum between fdust and mafic rock end-members. This interpretation is strosupported by results from Mars Pathfinder.In situ elementalcomposition measurements and spectral measurements by
covered the gray rock, red rock, and drift units of McSweenet al. talline ferric mineral in addition to pyroxene. Mustardet al.also
over
(1999). These correspond with normal dark I, transitional, andnoted that albedo characteristics of eastern Syrtis are stable
FIG. 13. Unfolded tetrahedron illustrating possible lithologic interpretatboldface type.

ET AL.

x-gs.-

om-thbitlyrlyhtalof

ose,arlyntstive83,

di-an-hterricgly

IMP

normal bright soils. The Mars Pathfinder measurements sthat albedo is strongly correlated with visible color, that cois highly correlated with sulfur content, and that nearly all msured elements vary in abundance linearly with sulfur contThese relationships indicate that the surface materials are pcal mixtures of bright, red, sulfur-rich dust and dark, gray, sulpoor rock.

In many normal transitional soils, bright red dust probaforms a loose, physically decoupled coating on normal dark ror sand. The largest region of normal transitional soil, in LibMontes (Fig. 1a), has undergone historic changes in its albbetween relatively bright and relatively dark (Martinet al.1992).ISM observed the region during a bright phase. Similarly, blarge regions of soils that were “normal dark I” at the timeISM observations, western Syrtis Major (Christensen 1986,1987) and Margaritifer Terra (McEwen 1992), are known to wand wane in albedo, presumably as thin fallout from dust stois deposited and then removed by eolian stripping. If obseat another time, either or both of these normal dark I soils mhave been considered “transitional.”

In contrast, in normal dark II soils such as in eastern SyMajor, the dust may be firmly cemented to the mafic substpossibly by a crystalline ferric phase. Bell (1992) and Bellet al.(1997) showed that eastern Syrtis also exhibits an unusustrong 0.66-µm absorption consistent with any of several fric minerals. Mustardet al. (1993) corroborated this work bshowing that eastern Syrtis’s 1-µm absorption is elongated toward shorter wavelengths, indicating the presence of a c

ions of the results of this study. Inferred compositional end-members are indicated in

ecsh

reno

u

t

soaob

mi

ze

nn

trma

c

sm

a

w

ofnal-

otofas-t aplyficsntedbeby

ousht Ier-o-wning

ostgicnt.

ped-ialstion,

ndand

tleduniteri-rialt II

ris-

ther-

a—igna-rvedech-rialsred


time and suggested that the dust is cemented to rock or alttively that rock surfaces could be oxidatively weathered. Reanalysis of IMP imagery of the Mars Pathfinder landingsupports Mustardet al.’s cemented dust model. Barnouin-Jet al.(2000) showed that dust-coated rocks exhibit evidencea 0.66-µm band absent from both loose dust and uncoatedand that this band becomes stronger as the rocks becomder and progressively more dust-coated. The band is strowhere all mafic spectral characteristics of the underlying rhave been hidden by the coatings. A physical mechanismgrowth of new ferric minerals in dust coatings was proposedBishop (1999). In her model, sulfate-containing dust adhererock and, in the presence of thin films or monolayers of liqwater, the sulfates react with dust diagenetically to form nferric minerals.

5.3. Dark Red and Associated Soils: Evidence for Duricrus

One of the most important results of this study is the content, distinctive spectral signature of the dark red visible-cunit, which corresponds to anomalous bright I soil (Figs. 1f9). Presley and Arvidson (1988) and Christensen and M(1992) synthesized evidence from radar, thermal, and alproperties and concluded that the dark red color unit is mlikely an exposure of a duricrust substrate to the overlying,bile sediment, like the indurated hardpan found at the VikLander sites (Mutchet al. 1977). Results from ISM are consistent with this interpretation. Samples of duricrust analyat the Viking Lander sites are enriched in sulfur most likoccurring as sulfates (Clark and VanHart 1981). Based ongeochemical arguments summarized above, sulfates are othe plausible water-retaining phases. The high-PC2 compowhich is more concentrated in dark red soils, is most siminterpreted as a water-bearing mineral phase.

The spectral attributes of the dark red soils suggest thatare predominantly composed of dust rather than a dust–mixture, at least at the optical surface. Compared with norsoils of similar reflectance, both major dark red areas (OxiaLunae Planum) have a relatively short-wavelength 0.9-µm ab-sorption, like that in dust except deepened. Neither occurrendark red soil exhibits significant evidence for a 2-µm pyroxeneband. The only compositional differences from dust suggeby spectral evidence are an enrichment in water-retainingerals and, in some cases such as in Lunae Planum, one orcrystalline ferric minerals.

Our interpretation of dark red soils as dust with minorditions of water-bearing and crystalline ferric phases is wsupported byin situ compositional measurements. At the MaPathfinder landing site, IMP measured dark red soils and fothe same spectral contrast with dust as in ISM data: a loalbedo and deeper 0.9-µm absorption, but a comparable spetral continuum. The elemental analysis of one of these darksoils proved indistinguishable from dust. The deeper 0.9-µm fer-
ric absorption must be explained by either a difference in oveferric mineral crystallinity or the presence of perhaps∼2% of

rna-entiteaforockred-

gestckforby

s toidew

is-lorndoreedoosto-

ng-edlythee ofent,ply

heyockalnd

e of

tedin-

more

d-ellrsund

erc-red

the Fe in a more crystalline ferric mineral (cf. McSweenet al.1999, Bellet al.1990, 2000).

Anomalous transitional soils, which occur at the marginsdark red soils, have most of the attributes of normal transitiosoils plus an enhanced 3-µm band. Their intermediate PC1 loading and 1-µm absorption indicate that their mineralogy is ndominated strictly by dust, but is consistent with a mixturedust with mafic rock or sand. However, their close spatialsociation with dark red (anomalous bright I) soils hints arelationship. Some anomalous transitional soils could simbe a variant of the dark red soils, a mixture of dust and macemented by a water-retaining phase rather than just cemedust. Alternatively, some anomalous transitional soils couldvariants of the normal dark II soils, dust cemented onto rocka hydrated mineral.

Possible interpretations of anomalous bright I and anomaltransitional soils are represented in Fig. 13. Anomalous brigsoil is shown on a mixing line between ferric dust and a watretaining phase, with admixture of a crystalline ferric compnent in Lunae Planum. Anomalous transitional soil is shoas a mixture of the ferric dust, mafic rock, and water-retaincomponents.

5.4. Interior Deposits of Valles Marineris: Evidencefor Volcanism

The interior deposits of Valles Marineris are some of the mintriguing geologic formations on Mars, because of morpholoevidence that they were formed in a liquid water environmeThe most famous are the layered deposits (Nedellet al.1987).Transitional to the “layered” deposits is the massive unit mapby Witbecket al.(1991) as unit Hml or “mottled material.” Several hypotheses for the origin of layered and mottled materhave been proposed. These include subaerial fluvial deposivolcanic flows (Lucchitta 1981, Lucchittaet al. 1992), depo-sition of carbonates in a lacustrine environment (McKay aNedell 1988), subaerial deposition of dust (Peterson 1981),hydrovolcanism in a lacustrine environment (Nedellet al.1987).

ISM observations cover both major occurrences of the motunit plus layered material in Candor Chasma (Fig. 11). Eachis covered by a distinctive, anomalous soil. The mottled matals coincide with anomalous dark soils, and the layered matein western Candor Chasma coincides with anomalous brighsoil. Small deposits with anomalous visible color charactetics, identified by Geissleret al. (1994), lie within the west-ern Candor layered material but are below the resolution of3× 3 pixel filtered data. This correlation of spectral propeties with geology—not observed elsewhere in our study aresuggests that these two anomalous units are the spectral stures of either fresh or weathered local bedrock, so that obsespectral properties may provide tests for proposed genetic manisms. The strong pyroxene absorptions in mottled mateand the mixed ferric–ferrous spectral character of the laye

ralldeposits rule out an origin by sedimentation of dust. Mustardand Sunshine (1995, 1996) found that the pyroxene absorptions

E

d

i

a

k

r

avs9

wa

et

andn-llelslousothsave

s,inde-

anic

IRsult-esrric

crys-alresareo-hedous

eithox-ctralnd

n ofark

al”e isb-

--ical

ilynor-ortedils

nd

468 MURCHI

in mottled materials are best fit by a composition significanricher in low-Ca pyroxene than in the surrounding highlanThis is inconsistent with either subaerial or fluvial depositionmaterials that originate from the surrounding plateau. Also,dedicated search for carbonate absorptions at 2.35 and 2.55µm,Erardet al. (1991) found no evidence for carbonates in anythe Valles Marineris interior deposits. Thus, neither the layenor the mottled materials can be limestones, although accescarbonates cannot be ruled out.

The only previously proposed genetic mechanisms content with spectral constraints from ISM involve volcanism rstricted to the interiors of the chasmata, of either a subaeor subaqueous nature. The enhanced 3-µm band in the mot-tled material might be attributable to the presence of the swater-retaining phase as proposed in other anomalous soilsstronger pyroxene absorptions than in normal dark soils coresult from particle size differences or the absence of masferric phases, as suggested by the low loadings of PC5. Inlayered material of western Candor Chasma (anomalous bII soil), the negative spectral slope and strong 3-µm absorptioncould result from a coating or rind containing a hydrated fermineral, or a mixture of dust, mafics, and the water-retainphase as in anomalous transitional soils.

Alternatively, the morphologic evidence for past liquid wter in the chasmata inevitably draws attention to the hydrocanic hypothesis of Nedellet al. (1987). The spectral effectof hydrovolcanism, described by Farrand and Singer (19are summarized in Fig. 14. In hydrovolcanism, lava can incporate and react with water to widely varying degrees. Lahaving undergone the least incorporation of water form dpsideromelanes with moderate mafic glass absorptions andor absent H2O bands (Fig. 14, bottom). Slight alteration by wter increases the depth of the mafic absorptions while inducing significant H2O absorptions (Fig. 14, middle). Greatalteration leads to formation of oxidized palagonite charac

FIG. 14. Spectra of hydrovolcanic deposits having undergone differidegrees of palagonitization, from Farrand and Singer (1991).

ET AL.

tlys.ofn a

ofredsory

sis-e-rial

me; theuldingtheight

ricing

-ol-

1),or-vasarkeak-

tro-rer-

ized by higher reflectance, the presence of ferric minerals,further increases in H2O absorptions due to greater water cotent (Fig. 14, top). This sequence of spectral changes parathat observed among layered and mottled materials. Anomadark soils (mottled material) have strong absorptions due to bmafic minerals and H2O, as in slightly altered lavas. Anomaloubright II soils (western Candor Chasma layered material) hstrongly elevated albedos, a mixed ferrous–ferric 1-µm band,and a strong 3-µm absorption as in oxidized palagonites. Thualthough ISM spectral data do not uniquely identify the origof the layered deposits, they do establish that the layeredposits have a unique spectral signature consistent with a volcorigin.

5.5. Synthesis of Lithology of the Surface Layer

In Fig. 13 we summarize possible interpretations of the Nspectral heterogeneities of the martian surface layer, as reing from mixtures of four mineralogic components. The verticof the unfolded tetrahedron represent four end-members, fedust, mafic rock and sand, the water-retaining phase, and atalline ferric component. In this diagram, normal bright, normtransitional, and normal dark I soils are simple physical mixtuof ferric dust and mafic rocks and sand. Normal dark II soilsmafic-dominated, with an additional crystalline ferric compnent. Most of the anomalous soils represent ferric dust enricin the water-retaining phase, sometimes intermixed with ferrmafics or the crystalline ferric component.

6. SUMMARY

(1) At NIR wavelengths, the three visible color units of thmartian surface layer break down into smaller groupings wdifferences in strengths of mineralogic absorptions due to pyrene, ferric minerals, and water, as well as differences in speslope. Regional differences in composition of the dark gray abright red units are hidden at visible wavelengths by reasoinsufficient wavelength coverage and resolution. Only the dred soils remain reasonably distinctive in the near-infrared.

(2) Eighty-one percent of the study area consists of “normbright red and dark gray soils in which increasing reflectancaccompanied by a shift from pyroxene to ferric mineralogic asorptions, deepening of the 3-µm H2O absorption, and development of a weak 2.2-µm band possibly attributable to a hydroxylated mineral. These trends are interpreted to result from physmixing of weakly altered basaltic crustal material with heavaltered dust. A steep spectral slope in intermediate-albedomal soils is explainable as a textural effect of dust mixing withcoating rocks or sand. This interpretation is strongly supporby in situ measurements of elemental compositions of soand rocks at the Mars Pathfinder landing site (McSweenet al.1999).

(3) The remaining 19% of the study area is “anomalous” a

ngconsistently exhibits a stronger 3-µm band than in normal soilsof comparable albedo. This enhancement is plausibly explained

dnsg

bt

a

l-o

t

r

a

a

y

t

0.

the

line

n-

ur-95:

is,A.sen,icars

az,n,b-

ion

naly-ible

),

sure

ro-

-ter.

ncectra.

9.

n,,

itefor

nti-ion

nce


by increased abundance of a minor water-bearing mineral phpossibly hydrated ferric minerals or sulfates. In other waanomalous soils are compositionally varied.

(4) Dark red soils consistently exhibit a low spectral sloand a 0.9-µm ferric mineral absorption similar to that in dusHowever, they also consistently exhibit a stronger 3-µm bandthan in dust. These soils are interpreted to consist largely ofenriched in the water-retaining phase, and possibly cemeinto a duricrust.In situmeasurement of the elemental compotion of dark red soil at the Mars Pathfinder landing site stronsupports a composition nearly identical to dust.

(5) Significant parts of the martian surface are darkeneda mineral lacking the 2-µm pyroxene absorption. ISM spectrof one such region, Lunae Planum, show an enhanced 0.9µmabsorption suggesting that the responsible phase is a crystaferric mineral. IMP visible-wavelength spectra of comparadark red soil at the Mars Pathfinder landing site also show0.9-µm feature. The IMP spectra have been interpreted todicate either ferrihydrite or maghemite (McSweenet al. 1999,Bell et al. 2000). However, the second major darkened regieastern Syrtis Major, has been shown in telescopic data to exa 0.66-µm absorption indicative of a different ferric mineralogperhaps hematite, schwertmannite, or goethite.

(6) These variations in NIR spectral properties imply at lefour components of the surface layer, the canonical dustweakly altered mafic particulates, plus a low-albedo crystalferric component and a component that retains a strong 3µmH2O absorption under the desiccating martian climatic cditions.

(7) In contrast with other observed geologic formations,layered and mottled materials in Valles Marineris exhibit distintive spectral signatures. These include an anomalously st3-µm absorption and pyroxene absorptions which imply a meralogy distinct from normal dark soils on the surrounding higlands. No evidence of carbonates is observed in these mateThese results are consistent with proposed genetic mechanwhich involve volcanism restricted to the interiors of the chmata, of either a subaerial or subaqueous nature.

ACKNOWLEDGMENTS

This manuscript benefitted greatly from discussions with A. TreimanD. Domingue, and thoughtful reviews by J. Bell, D. Blaney, and W. Calvin.

REFERENCES

Adams, J., and T. McCord 1969. Mars: Interpretation of spectral reflectivitbright and dark regions.J. Geophys. Res.74, 4851–4856.

Beinroth, A., and G. Arnold 1996. Analysis of weak surface absorption bandthe near-infrared spectra of Mars obtained by Phobos 2.Vibrational Spectrosc.11, 115–121.

Barnouin-Jha, O., S. Murchie, J. Johnson, and J. F. Bell III 2000. Rock coaat the Mars Pathfinder landing site.Proc. Lunar Planet. Sci. XXXI, 1262.

Bell, J. 1992. Charge-coupled device imaging spectroscopy of Mars.Icarus100,575–597.


ase,ys,

pet.

usttedi-ly

bya-llinelehisin-

on,hibity,

standine

n-

hec-ongin-h-rials.ismss-

nd

of

s in

ings

Bell, J. 1996. Iron, sulfate, carbonate, and hydrated minerals on Mars. InMineralSpectroscopy(H. Dyar, C. McCammon, and M. Schaefer, Eds.), pp. 359–38Geochemical Society, Special Publication 5.

Bell, J., and D. Crisp 1993. Ground-based imaging spectroscopy of Mars innear-infrared: Preliminary results.Icarus104, 2–19.

Bell, J., T. McCord, and P. Owensby 1990. Observational evidence of crystaliron oxides on Mars.J. Geophys. Res.95, 14,447–14,461.

Bell, J., J. Piscitelli, and L. Lebofsky 1989. Deimos: Hydration state from ifrared spectroscopy.Proc. Lunar Planet. Sci. XX, 58–59.

Bell, J., M. Wolff, P. James, T. Clancy, S. Lee, and L. Martin 1997. Mars sface mineralogy from Hubble Space Telescope imaging during 1994–19Observations, calibration, and initial results.J. Geophys. Res.102, 9109–9123.

Bell, J. F., III, H. McSween, S. Murchie, J. Johnson, R. Reid, R. MorrR. Anderson, J. Bishop, N. Bridges, D. Britt, J. Crisp, T. Economou,Gosh, J. Greenwood, H. Gunnlaugsson, R. Hargraves, S. Hviid, J. KnudM. Madson, H. Moore, R. Rieder, and L. Soderblom 2000. Mineralogand compositional properties of martian soil and dust: Results from MPathfinder.J. Geophys. Res.105, 1721–1756.

Bibring, J.-P., M. Combes, Y. Langevin, C. Cara, P. Drossart, T. EncrenS. Erard, O. Forni, B. Gondet, L. Ksanfomality, E. Lellouch, P. MassoV. Moroz, F. Rocard, J. Rosenqvist, C. Sotin, and A. Soufflot 1990. ISM oservations of Mars and Phobos: First results.Proc. Lunar Planet. Sci. Conf.XX, 461–471.

Bishop, J. 1999. Hydrothermal alteration products as a key to formatof duricrust and rock coatings on Mars.Proc. Lunar Planet. Sci. XXX,1887.

Bishop, J., and E. Murad 1996. Schewertmannite on Mars? Spectroscopic ases of schwertmannite, its relationship to other ferric minerals, and its posspresence in the surface material of Mars. InMineral Spectroscopy: A Trib-ute to Roger G. Burns(M. Dyar, C. McCammon, and M. Scheaer, Eds.pp. 337–358. Geochemical Society, Special Publication 5, Houston.

Bishop, J., and C. Pieters 1995. Low-temperature and low atmospheric presinfrared reflectance spectroscopy of Mars soil analog materials.J. Geophys.Res.100, 5369–5379.

Burns, R. 1994. Schwertmannite on Mars: Deposition of this ferric oxyhydxysulfate mineral in acidic saline meltwaters.Proc. Lunar Planet. Sci. XXV,203–204.

Calvin, W. 1997. Variation of the 3-µm absorption feature on Mars: Observations over eastern Valles Marineris by the Mariner 6 infrared spectromeJ. Geophys. Res.102, 9085–9096.

Calvin, W., T. King, and R. Clark 1994. Hydrous carbonates on Mars? Evidefrom Mariner 6/7 infrared spectrometer and groundbased telescopic speJ. Geophys. Res.99, 14,659–14,676.

Christensen, P. 1986. The spatial distribution of rocks on Mars.Icarus 68,217–238.

Christensen, P., and H. Moore 1992. The martian surface layer. InMars(H. Kieffer, B. Jakosky, C. Snyder, and M. Matthews, Eds.), pp. 686–72Univ. of Arizona Press, Tucson.

Christensen, P., J. Bandfield, R. Clark, K. Edgett, V. Hamilton, T. HoefeH. Kieffer, R. Kuzmin, M. Lane, M. Malin, R. Morris, J. Pearl, R. PeasonT. Roush, S. Ruff, and M. Smith 2000b. Detection of crystalline hematmineralization on Mars by the thermal emission spectrometer: Evidencenear-surface water.J. Geophys. Res.105, 9623–9642.

Christensen, P., J. Bandfield, M. Smith, V. Hamilton, and R. Clark 2000a. Idefication of a basaltic component on the martian surface from thermal emissspectrometer data.J. Geophys. Res.105, 9609–9622.

Clark, B., and D. Van Hart 1981. The salts of Mars.Icarus45, 370–378.

Clark, R., G. Swayze, R. Singer, and J. Pollack 1990. High-resolution reflecta
spectra of Mars in the 2.3-µm region: Evidence for the mineral scapolite.J. Geophys. Res.95, 14,463–14,480.

r

x

viM

d

o

t

sk

.)

o8

s.

a

t

ni

n

ta

88-

sin

S.

ss,

ser-

ss,

pec-

nceetry

rs:

ris,

ils,

III,k,, R.intsite.

nce

on

edctral

ix-

9.and

ike

85.of

dis-

470 MURCHIE

Cloutis, E., and M. Gaffey 1991. Pyroxene spectroscopy revisited: Spectcompositional correlations and relationship to geothermometry.J. Geophys.Res.96, 22,809–22,826.

Cooper, C., and J. Mustard 1997. Loss of spectral contrast in hyperfine paonite: Implications for smectite on Mars.Proc. Lunar Planet. Sci. XXVIII,257–258.

Crown, D., and C. Pieters 1987. Spectral properties of plagioclase and pyromixtures and the interpretation of lunar soil samples.Icarus72, 492–506.

Drossart, P., J. Rosenqvist, S. Erard, Y. Langevin, J.-P. Bibring, and M. Com1991. Martian aerosols properties from the Phobos/ISM experiment.Ann.Geophys.9, 754–760.

Erard, S. 1995. Variability of surface materials in the equatorial region of MaProc. Lunar Planet. Sci. XXVI, 377–378.

Erard, S., and W. Calvin 1997. Composite spectra of Mars, 0.4–3.7µm. Icarus130, 449–460.

Erard, S., J.-P. Bibring, J. Mustard, O. Forni, J. Head, S. Hurtrez, Y. LangeC. Pieters, J. Rosenqvist, and C. Sotin 1991. Spatial variations in composof the Valles Marineris and Isidis Planitia regions of Mars derived for ISspectra.Proc. Lunar Planet. Sci. Conf. XXI, 437–455.

Erards, S., J. Mustard, S. Murchie, J.-P. Bibring, P. Cerroni, and A. Cora1994. Martian aerosols: Near-infrared spectral properties and effects onobservation of the surface.Icarus111, 317–337.

Farrand, W., and R. Singer 1991. Spectral analysis and mapping of palagtuffs of Pavant Butte, Millard County, Utah.Geophys. Res. Lett.18, 2237–2240.

Fischer, E., and C. Pieters 1993. The continuum slope of Mars: Bi-directioreflectance investigations and applications to Olympus Mons.Icarus 102,185–202.

Geissler, P., R. Singer, G. Komatsu, S. Murchie, and J. Mustard 1994. An unuspectral unit in western Candor Chasma: Evidence for aqueous or hydromal alteration in the martian canyons.Icarus106, 380–391.

Gooding, J. 1978. Chemical weathering on Mars: Thermodynamic stabilitieprimary minerals (and their alteration products) from mafic igneous rocIcarus33, 483–513.

Gooding, J., R. Arvidson, and M. Zolotov 1992. Physical and chemical weaering. InMars (H. Kieffer, B. Jakosky, C. Snyder, and M. Matthews, Edspp. 626–651. Univ. of Arizona Press, Tucson.

Greeley, R., and J. E. Guest 1987. Geologic Map of the Eastern EquatRegion of Mars. U.S.G.S. Miscellaneous Investigation Series, Map I-103

Greeley, R., N. Lancaster, S. Lee, and P. Thomas 1992. Martian eolian procesediments, and features. InMars (H. Kieffer, B. Jakosky, C. Snyder, and MMatthews, Eds.), pp. 730–766. Univ. of Arizona Press, Tucson.

Hanel, R., B. Conrath, W. Hovis, V. Kunde, P. Lowman, W. Maguire, J. PeJ. Pirraglia, C. Prabhakara, and B. Schlachman 1972. Investigation ofmartian environment by infrared spectroscopy on Mariner 9.Icarus 17,423–442.

Houck, J., J. Pollack, C. Sagan, D. Schaak, and J. Decker 1973. High-altiinfrared spectroscopic evidence for bound water on Mars.Icarus18, 470–480.

Kieffer, H., T. Martin, A. Peterfreund, B. Jakosky, E., Miner, and D. Palluco1977. Thermal and albedo mapping of Mars during the Viking primary msion.J. Geophys. Res.28, 4249–4921.

Kirkland, L., A. Treiman, and S. Murchie 1997. Photometry of bright regioon Mars: ISM results.Proc. Lunar Planet. Sci. XXVIII, 729–730.

Kirkland, L., A. Treiman, and S. Murchie 2000a. Photometric correctionsnear-infrared spectra of Mars using Phobos 2 ISM. Submitted toJ. Geophys.Res.

Kirkland, L., K. Herr, J. Salisbury, P. Forney, and P. Adams 2000b. Water inmartian aerosol dust and surface material: Results from Mariner 7 IRSPhobos 2 ISM. Submitted toJ. Geophys. Res.

Lee, S. 1987. Regional sources and sinks of dust on Mars: Viking obsertions of Cerberus, Solis Planum, and Syrtis Major. InMECA Symposium on

ET AL.

al–

lag-

ene

bes

rs.

in,tion

inithe

nite

nal

sualher-

ofs.

th-,

rial.

ses,

rl,the

ude

is-

s

to

hend

Mars: Evolution of its Climate and Atmosphere(V. Baker, M. Carr, F. Fanale,R. Greeley, R. Haberle, C. Leovy, and T. Maxwell, Eds.), LPI Tech. Rep.01, pp. 57–58.

Lucchitta, B. 1981. Valles Marineris—Faults, volcanic rocks, channels, babeds. InReports of the Planetary Geology Program, NASA TM 84211,pp. 419–421.

Lucchitta, B., A. McEwen, G. Clow, P. Geissler, R. Singer, R. Schultz, andSquyres 1992. The canyon system on Mars. InMars (H. Kieffer, B. Jakosky,C. Snyder, and M. Matthews, Eds.), pp. 453–492. Univ. of Arizona PreTucson.

Martin, L., P. James, A. Dollfus, K. Iwasaki, and J. Beish 1992. Telescopic obvations: Visual, photographic, polarimetric. InMars (H. Kieffer, B. Jakosky,C. Snyder, and M. Matthews, Eds.), pp. 34–70. Univ. of Arizona PreTucson.

McCord, T., R. Clark, and R. Singer 1982. Mars: Near-infrared reflectance stra of surface regions and compositional implications.J. Geophys. Res.87,3021–3032.

McCord, T., R. Huguenin, D. Mink, and C. Pieters 1977. Spectral reflectaof martian areas during the 1973 opposition: Photoelectric filter photom0.33–1.10µm. Icarus31, 25–39.

McEwen, A. 1992. Temporal variability of the surface and atmosphere of MaViking Orbiter color observations.Proc. Lunar Planet. Sci. XXIII, 877–878.

McKay, C., and S. Nedell 1988. Are there carbonates in the Valles MarineMars?Icarus73, 142–148.

McSween, H., and A. Ghosh 1999. Sulfur mixing relationships in martian soand possible implications for a globally homogeneous dust layer.Proc. LunarPlanet. Sci. XXX, 1083.

McSween, H., S. Murchie, J. Crisp, N. Bridges, R. Anderson, J. F. BellD. Britt, J. Bruckner, G. Dreibus, T. Economou, A. Ghosh, M. GolombeJ. Greenwood, J. Johnson, H. Moore, R. Morris, T. Parker, R. RiederSinger, and H. Wanke 1999. Chemical, multispectral, and textural constraon the composition and origin of rocks at the Mars Pathfinder landing sJ. Geophys. Res.104, 8679–8715.

Merenyi, E., K. Edgett, and R. Singer 1996a. Deucalionis Regio, Mars: Evidefor a new type of immobile weathered soil unit.Icarus124, 296–307.

Merenyi, E., R. Singer, and J. Miller 1996b. Mapping of spectral variationsthe surface of Mars from high spectral resolution telescopic images.Icarus124, 280–295.

Moroz, V. 1964. The infrared spectrum of Mars (1.1–4.1µm). Soviet Astron. 8,273–281.

Morris, R., and H. Lauer 1990. Matrix effects for reflectivity spectra of dispersnanophase (superparamagnetic) hematite with application to martian spedata.J. Geophys. Res.95, 5101–5109.

Morris, R., and S. Neely 1981. Optical properties of hematite–magnetite mtures: Implications for Mars.Proc. Lunar Planet. Sci. XIII, 548–549.

Morris, R., D. Agresti, H. Lauer, J. Newcomb, T. Shelfer, and A. Murali 198Evidence for pigmentary hematite on Mars based on optical, magnetic,Mossbauer studies of superparamagnetic (nanocrystalline) hematite.J. Geo-phys. Res.94, 2760–2778.

Morris, R., J. Gooding, H. Lauer, and R. Singer 1990. Origins of the Mars-lspectral and magnetic properties of a Hawaiian palagonite.J. Geophys. Res.95, 14,427–14,435.

Morris, R., H. Lauer, C. Lawson, E. Gibson, G. Nace, and C. Stewart 19Spectral and other physiochemical properties of submicron powdershematite (α-Fe2O3), maghemite (γ -Fe2O3), magnetite (Fe3O3), goethite(α-FeOOH), and lepidocrocite (γ -FeOOH). J. Geophys. Res.90, 3126–3144.

Murchie, S., and J. Mustard 1994. Spectral units of martian soil: Possible
va-crimination of mobile sediment and substrate.Proc. Lunar Planet. Sci. XXV,
955–956.

0

a

c

s

.

n

s as:,

ion

iron

ys-

and

laters.

ace

lay

ace

na

2.

rtic

ris

oil:


Murchie, S., O., Barnouin-Jha, J. Johnson, J. F. Bell III, and H. McSween 2Diverse rock types at the Mars Pathfinder landing site.Proc. Lunar Planet.Sci. XXXI, 1267.

Murchie, S., E. Merenyi, R. Singer, and L. Kirkland 1996. Visible-NIR spetroscopic evidence for the composition of low-albedo altered soils on MProc. Lunar Planet. Sci. XXVII, 919–920.

Murchie, S., J. Mustard, J. Bishop, J. Head, C. Pieters, and S. Erard 1Spatial variations in the spectral properties of bright regions on Mars.Icarus105, 454–468.

Murchie, S., N. Thomas, D. Britt, K. Herkenhoff, and J. F. Bell III 1999. MaPathfinder spectral measurements of Phobos and Deimos: Comparisonprevious data.J. Geophys. Res.104, 9069–9079.

Mustard, J. 1995. Are dark red soils and bright red dust on Mars related thromineralogic phase changes?Proc. Lunar Planet. Sci. XXVI, 1021–1022.

Mustard, J., and J. Bell 1994. Bew composite reflectance spectra of Mars0.4–3.14µm. Geophys. Res. Lett.21, 353–356.

Mustard, J., and J. Sunshine 1995. Seeing through the dust: Martian crheterogeneity and links to the SNC meteorites.Science267, 1623–1626.

Mustard, J., and J. Sunshine 1996. Mafic mineralogy of martian volcanics: Vations in space and time.Proc. Lunar Planet. Sci. XXVII, 925–926.

Mustard, J., S. Erard, J.-P. Bibring, J. Head, S. Hurtrez, Y. Langevin, C. Pieand C. Sotin 1993. The surface of Syrtis Major: Composition of the volcasubstrate and mixing with altered dust and soil.J. Geophys. Res.98, 3387–3400.

Mutch, T., R. Arvidson, A. Binder, E. Guiness, and E. Morris 1977. The geoloof the Viking 2 lander site.J. Geophys. Res.82, 4452–4467.

Nedell, S., S. Squyres, and D. Andersen 1987. Origin and evolution of the laydeposits in the Valles Marineris, Mars.Icarus70, 409–441.

Peterson, C. 1981. A secondary origin for the central plateau of Hebes ChaProc. Lunar Planet. Sci. Conf. XII, 1459–1471.

Pieters, C. 1983. Strength of mineral absorption features in the transmittedponent of near-infrared reflected light: First results from RELAB.J. Geophys.Res.88, 9534–9544.

Pimental, G., P. Forney, and K. Herr 1974. Evidence about hydrate andwater in the martian surface from the 1969 Mariner infrared spectromeJ. Geophys. Res.79, 1623–1634.

Presley, M., and R. Arvidson 1988. Nature and origin of materials exposethe Oxia Palus–Western Arabia–Sinus Meridiani region, Mars.Icarus 75,499–517.

Rieder, R., T. Economou, H. Wanke, A. Turkevich, J. Crisp, J. Bruckner,Dreibus, and H. McSween, Jr. 1997. The chemical composition of mar
soil and rocks returned by the mobile alpha proton X-ray spectrometer: Pliminary results from the X-ray mode.Science278, 1771–1774.

00.

c-rs.

993.

rswith

ugh

from

ustal

ari-

ters,nic

gy

ered

sma.

om-

olidter.

d in

G.tian

Rivkin, A., R. Brown, D. Trilling, J. F. Bell III, and J. Plassman 2000Infrared spectrophotometry of Phobos.Proc. Lunar Planet. Sci. XXXI,1481.

Rodin, A., O. Korablev, and V. Moroz 1997. Vertical distribution of water ithe near-equatorial troposphere of Mars: Water vapor and clouds.Icarus125,212–229.

Roush, T., D. Blaney, and R. Singer 1993. The surface composition of Marinferred from spectroscopic observations. InRemote Geochemical AnalysisElemental and Mineralogical Composition(C. Pieters and P. Englert, Eds.)pp. 367–393. Cambridge Univ. Press, New York.

Scott, D., and K. Tanaka 1986. Geologic Map of the Western Equatorial Regof Mars. U.S.G.S. Miscellaneous Investigation Series, Map I-1802-A.

Sherman, D., R. Burns, and V. Burns 1982. Spectral characteristics of theoxides with application to the martian bright region mineralogy.J. Geophys.Res.87, 10,169–10,180.

Singer, R. 1981. Near-infrared spectra: Reflectance of mineral mixtures: Stematic combinations of pyroxenes, olivine, and iron oxides.J. Geophys. Res.86, 7967–7982.

Singer, R. 1982. Spectral evidence for the mineralogy of high albedo soilsdust on Mars.J. Geophys. Res.87, 10,159–10,168.

Singer, R., and T. Roush 1983. Spectral reflectance properties of particuweathered coatings on rocks: Laboratory modeling and applicability to MaProc. Lunar Planet. Sci. XIV, 708–709.

Singer, R., T. McCord, R. Clark, J. Adams, and R. Huguenin 1979. Mars surfcomposition from reflectance spectroscopy: A summary.J. Geophys. Res.84,8415–8426.

Singer, R., P. Owensby, and R. Clark 1985. Observed upper limits for cminerals on Mars.Proc. Lunar Planet. Sci. Conf. XVI, 787–788.

Soderblom, L. 1992. The composition and mineralogy of the martian surffrom spectroscopic observations: 0.3µm to 50µm. In Mars (H. Kieffer, B.Jakosky, C. Snyder, and M. Matthews, Eds.), pp. 557–593. Univ. of ArizoPress, Tucson.

Tanaka, K., D. Scott, and R. Greeley 1992. Global stratigraphy. InMars(H. Kieffer, B. Jakosky, C. Snyder, and M. Matthews, Eds.), pp. 345–38Univ. of Arizona Press, Tucson.

Toulmin, P., A. Baird, B. Clark, K. Keil, H. Rose, R. Christian, and K. Stewa1977. Geochemical and minealogic interpretation of the Viking inorganchemical results.J. Geophys. Res.82, 4625–4634.

Witbeck, N., K. Tanaka, and D. Scott 1991. Geologic Map of the Valles MarineRegion, Mars. U.S.G.S. Misc. Inv. Series, Map I-2010.

Yen, A., B. Murray, and G. Rossman 1998. Water content of the martian s
re-Laboratory simulations of reflectance spectra.J. Geophys. Res.103, 11,125–
11,133.

near-infrared spectral variations of martian surface materials from ism imaging spectrometer data

Documents