low abundance materials at the mars pathfinder landing site: an

Icarus 158, 56–71 (2002)doi:10.1006/icar.2002.6865

Low Abundance Materials at the Mars Pathfinder Landing Site: AnInvestigation Using Spectral Mixture Analysis and Related Techniques

J. F. Bell III

Cornell University, Department of Astronomy, Ithaca, New YorkE-mail: [email protected]

W. H. Farrand

Space Science Institute, Boulder, Colorado

J. R. Johnson

U.S. Geological Survey, Astrogeology Team, Flagstaff, Arizona

and

R. V. Morris

NASA Johnson Space Center, Code SR, Houston, Texas

Received October 29, 2001; revised February 25, 2002

Recalibrated and geometrically registered multispectral imagesfrom the Imager for Mars Pathfinder (IMP) were analyzed usingSpectral Mixture Analysis (SMA) and related techniques. SMAmodels a multispectral image scene as a linear combination of end-member spectra, and anomalous materials which do not fit themodel are detected as model residuals. While most of the IMPdata studied here are modeled generally well using “Bright Dust,”“Gray Rock,” and “Shade” image endmembers, additional anoma-lous materials were detected through careful analysis of root meansquare (RMS) error images resulting from SMA. For example, ana-lysis of SMA fraction and RMS images indicates spectral differenceswithin a previously monolithologic Dark Soil class. A type of DarkSoil that has high fractional abundances in rock fraction images(Gray Rock Soil) was identified. Other anomalous materials identi-fied included a previously noted “Black Rock” lithology, a class ofpossibly indurated, compacted, or partially cemented soils (“Inter-mediate Soil”), and a unit referred to as “Anomalous Patches” on atleast one rock. The Black Rock lithology has a strong 900–1000-nmabsorption, and modeling of the derived image endmembers usinga laboratory reference endmember modeling (REM) approach pro-duced best-fit model spectra that are most consistent with the pre-sence of high-Ca pyroxenes and/or olivine, crystalline ferric oxideminerals, or mixtures of these materials as important componentsof the Black Rock endmember. More unique mineralogic identifi-cations could not be obtained using our initial REM analyses. BothIntermediate Soil and Anomalous Patches units exhibit a relativelynarrow 860–950-nm absorption that is consistent with the presenceof either low-Ca pyroxenes or a cementing crystalline ferric oxidemineral. c© 2002 Elsevier Science (USA)

Key Words: Mars, surface; surfaces, planets; mineralogy; imageprocessing.

INTRODUCTION

56

0019-1035/02 $35.00c© 2002 Elsevier Science (USA)

All rights reserved.

On July 4, 1997 the Mars Pathfinder (MPF) spacecraft landedon the Ares Vallis floodplain, where it returned data for 83 solsfrom a landscape strewn with rocks, boulders, and windblownfines. The landing site was selected as a “grab bag” site, so namedbecause it was hoped to contain rocks transported from severalgeologic provinces by the Ares and Tiu Vallis outflow chan-nel flood events (Golombek et al. 1997). The expectation wasthat rocks from the ancient Noachian plateau and the Hesperianridged plains would be present at the landing site. However,initial published results suggested only a single lithology. Thisinference followed from the very similar major element com-positions for a small number of rocks as measured by the alphaproton X-ray spectrometer (APXS) mounted on the Sojournerrover (Rieder et al. 1997, Bruckner et al. 2001). The extra-polated dust-free composition of the putative single lithology,as measured by the APXS, corresponds to a basaltic andesite(McSween et al. 1999). These authors also report that the lithol-ogy corresponds well with the dominant “Gray Rock” spectralclass observed in visible and near-IR multispectral data obtainedby the Imager for Mars Pathfinder (IMP) instrument (Smith et al.1997a,b). Most differences in the spectral reflectance of rockswere attributed to variable coatings of dust. Soil at the MPF

H
ANOMALOUS MATERIALS AT T
landing site consisted of bright drift material, darker fines, and avariety of related, but spectrally distinct soils (Bell et al. 2000).The chemical composition of the six soil measurements madeby the APXS indicates a soil very similar in composition to thatmeasured at the two Viking landing sites, supporting the infe-rence of a component of the soils having been globally homoge-nized by aeolian redistribution. Morris et al. (2000) calculated achemical composition for this global-average soil that was con-sistent with a mixture of basaltic andesites and high-Al, Mg SNCmeteorites. Chemical modeling by Bell et al. (2000) indicatesthat the soils measured by the APXS were not derived from thedominant andesitic Gray Rock lithology at the landing site, butwere more consistent with palagonitic alteration of a rock of amore basaltic composition.

The original interpretation of the monolithologic makeup ofthe landing site has been challenged by Murchie et al. (2000,2001). In a reexamination of a more accurately calibrated andgeometrically registered IMP dataset (e.g., Johnson et al. 2001),these authors observed the presence of at least two other sets ofrocks whose spectra were distinct from that of Gray Rock andcoated variants thereof. Assuming that these spectral classesequate with different lithologies, Murchie et al. (2000, in prepa-ration). Noted several examples of a Black Rock lithology witha deep 1000-nm absorption band. They also noted a single ex-ample of an “Orange Rock” lithology with a distinct 750-nmpeak and a strong absorption band centered near 900 nm. Thesenewly discovered minor lithologies occur at the landing site assmall rocks, cobbles, or pebbles. The presence of these uniquerocks indicates more lithologic diversity at the MPF landing sitethan was previously suspected; however, their small size indi-cates that the diversity is found in relatively low abundance andapparently only among small rocks.

Searching for low-abundance materials in spectral image datarequires specialized data processing tools. Such tools have beendeveloped and successfully applied to terrestrial multi- and hy-perspectral datasets by a number of researchers. Particularlyimportant are tools such as image endmember-based and refer-ence endmember-based spectral mixture analysis (Adams et al.1986, 1993) and foreground–background analysis (Smith et al.1994), which account for the reality of multiple materials withina sensor’s field of view by modeling the recorded pixel radi-ance as a linear combination of endmember spectra. Materialspresent within the scene that have spectral signatures differentfrom those of the endmembers, e.g., the low abundance mate-rials noted above, can be detected as anomalies from modeledimages.

In this paper, we describe a detailed search for spatially re-stricted, low-abundance materials within the most spatially andspectrally extensive, highest fidelity multispectral data set ob-tained during the Mars Pathfinder mission. First we describethe data set and the specific calibration and image processingsteps used in its creation. Then we describe the spectroscopic
analysis procedures that we used, focusing on techniques foridentification of the most extreme spectral units in the scene and
E PATHFINDER LANDING SITE 57

subsequent spectral mixture analyses using those units as end-members. We then present our results from mapping the spatialdistribution of these endmembers and describe the properties ofthree different classes of spectrally anomalous materials that weidentified using our analysis tools. We also describe some pre-liminary results using reference endmember modeling methods,attempting to tie our derived image endmembers to laboratoryspectra of specific rocks, minerals, and meteorites. Finally, weconclude with a discussion of the implications of our results forexpanding our understanding of the composition and mineralogyof the Mars Pathfinder landing site.

DATA SET

We analyzed a recalibrated and geometrically registered ver-sion of the “Super Pan” data set, acquired in all 12 IMP bands(440–1000 nm) and covering most of the MPF landing site(Smith et al. 1997b, Golombek et al. 1999a). The Super Panwas acquired in eight separate image mosaics (“octants” labeledS0181 through S0188), with each mosaic consisting of manyseparate but contiguous camera azimuth and elevation pointingpositions (each referred to as “segments”). The results describedhere focus on a subset of 10 segments shown in Table I. Version 3of the IMP calibration algorithm (e.g., Johnson et al. 2001) wasused to calibrate the images to relative reflectance using imagesof the radiometric calibration targets (RCTs) acquired close intime to a given Super Pan image sequence (Reid et al. 1999).There are a number of sources of potential systematic errors inthe IMP calibration, including temporal and viewing geometry(phase function) differences between images of the scene and ofthe RCTs, and the changing balance of direct vs diffuse illumi-nation of the surface depending on topographic slope, viewingazimuth, and time of day, and changes in the scattering propertiesof atmospheric aerosols (e.g., Tomasko et al. 1999, Thomas et al.1999). Version 3 of the IMP calibration algorithm described byReid et al. (1999) and Johnson et al. (2001) attempts to compen-sate for some of these potential sources of systematic errors byscaling the radiance values of a scene image by the ratio of thetotal downward flux at the time of RCT acquisition to the totaldownward flux at the time of scene acquisition. The algorithmuses an approximation to the sky model of Tomasko et al. (1999)to essentially convert the scene brightnesses to values that theywould be predicted to have had if observed simultaneously withthe RCT images. The times for RCT image acquisitions associ-ated with each of the Super Pan segments used here are shownin Table I. For 6 of the 10 segments used, RCT images wereacquired within about 30 min of the scene images. For the other4 segments, RCT images were acquired within about 30 min ofthe same local solar time as the scene images, but up to 4 solsapart.

ISIS was used to compute the subpixel registration offsetsamong different bands for a given scene and camera eye. The
USGS geometric control network of the landing site was alsoused with ISIS algorithms to compute the appropriate map

58 BELL ET AL.

TABLE ILink between Segment Definitions and IMP Image Information Retrievable from the NASA PDS

Octant Seg. Sol Az. EL. SCLK range LST RCT SCLK range LST of RCT Note

S0182 4 18 291.13 −11.20 1248280287–1248280594 1500–1505 1248278587–1248278877 1432–1437S0183 5 54 262.24 −21.43 1251454484–1251454639 0910–0913 1251454137–1251454343 0904–0908S0184 8 32 217.37 −9.42 1249503045–1249503293 0934–0938 1249859509–1249859735 1000–1004 (Sol 36)S0184 14 32 237.49 −20.91 1249504772–1249505061 1002–1007 1249859509–1249859735 1000–1004 (Sol 36)S0185 6 13 168.14 −11.59 1247813308–1247813548 0845–0849 1247549055–1247549346 0919–0923 (Sol 10)S0185 12 66 187.42 −22.12 1252519540–1252519693 0905–0907 1252520553–1252520792 0921–0925S0187 8 42 78.33 −5.28 1250388941–1250389091 0908–0910 1250566519–1250566740 0908–0912 (Sol 44)S0188 8 20 30.2 2.94 1248458172–1248458307 1505–1508 1248457313–1248457586 1452–1456S0188 13 20 42.2 −5.6 1248458931–1248459066 1518–1520 1248457313–1248457586 1452–1456S0188 15 20 43.17 −17.24 1248460062–1248460289 1536–1540 1248457313–1248457586 1452–1456S0188 20 20 42.31 2.7 1248459742–1248460030 1531–1535 1248457313–1248457586 1452–1456

Note. Seg. = image segment number within each octant mosaic. Az. = IMP camera azimuth pointing. El. = IMP camera elevation pointing. SCLK = spacecraft
clock time (number of seconds since launch). LST = local solar time of the observations, assuming 1200 is the time of solar zenith crossing at the landing site. RCT = radiometric calibration target.
projection parameters (cf. Kirk et al. 1999). Using a single pixelresampling, the registration offsets and projection parameterswere simultaneously used to produce left- and right-eye multi-spectral image mosaics for each of the eight separate image se-quences (octants S0181 through S0188) that comprise the SuperPan. For reference, left-eye segments consisted of images ac-quired at 443, 671, 802, 858, 898, 931, 968, and 1003 nm andright-eye segments consisted of images acquired at 443, 480,531, 600, 671, 752, and 967 nm (Smith et al. 1997a).

PROCESSING APPROACH

While the recalibration of the Super Pan data has mitigatedphotometric differences between the image segments that makeup each octant, these differences have not been entirely elim-inated. Within a given octant, residual artifacts such as subtlemosaic seams and photometric differences resulting from vari-ability in image acquisition times may be enhanced by mathe-matical processing. In order to mitigate these effects, individualsegments of a mosaic (see Table I) were examined here one at atime, and spectra were extracted only from segments that weregeometrically coregistered but not map-projected. Data fromeach eye of IMP were examined separately, though spectra ex-tracted from representative units were later merged into a singlespectrum spanning all IMP wavelengths (Fig. 1) in order to betterdetermine, through examination of the spectral characteristics,the identity of the endmember and anomalous materials. Imageregions that contained portions of the MPF spacecraft or skyabove the horizon were masked and eliminated from the analysis.

To assess the degree of subpixel spectral heterogeneity, weapplied linear spectral mixture analysis (SMA) to the imagedata. The basic equation for SMA, in vector form, is

r (x, y) = αM + n, (1)

where r (x, y) = the relative reflectance spectrum for the pixel at

position (x , y), α = the vector of endmember abundances, M =the matrix of endmember spectra, and n = the vector of residualsbetween the modeled and the measured relative reflectances.

The endmember spectra are those spectra that can be usedto model, by linear combination, the measured reflectance ofeach pixel in the scene. This linear mixing approach assumesthat different materials each occupy a distinct fraction of eachpixel (checkerboard style) and that the resulting spectrum ofthat pixel can be represented by adding the component spectrain proportion to their fractional abundances (e.g., Adams et al.1986, 1993). Nonlinear mixing effects, such as extremely thindust coatings or weathering rinds which allow photons to pene-trate to different substrate materials, are often not well modeledusing such a simple linear approach.

The parameter n in Eq. (1) can be represented by a root meansquare (RMS) error image in which the RMS error is calculatedover all bands following Eq. (2),

ε =[

M−1M∑

c=1

n2c

]1/2

, (2)

where ε = the RMS error, M = the number of channels, c =channel number, and n = the vector of residuals between themodeled and the measured relative reflectances. The RMS errorimage provides an indication of how well the chosen endmem-bers model the spectral variability of the scene and whether thereare additional materials present which are not well modeled bythe endmembers. Examples of such additional materials includethe low-abundance, spectrally unique materials sought in thisinvestigation. Also, effects of nonlinear mixing could contributeto errors between the modeled and measured spectra; however,as discussed below, we identified only a few possible examplesof such nonlinear effects.

In most image segments, we found that using only three end-members (Bright Dust, Gray Rock, and Shade) could produce

ANOMALOUS MATERIALS AT THE PATHFINDER LANDING SITE 59

FIG. 1. Combined left- and right-eye relative reflectance (Reid et al. 1999, Bell et al. 2000) spectra of multipixel averages of typical (a) Bright Dust, (b) GrayRock, and (c) Shade endmembers from a variety of different SuperPan octants. For this and all other spectra presented in this paper, left- and right-eye IMP spectrawere extracted from separate image cubes and combined by comparing the relative reflectance values of each spot in the common 671-nm filter. The spectra werecombined by: (1) determining which eye had the higher 671-nm reflectance and doing nothing to the relative reflectance values for that eye, and (2) adding the
difference between the 671-nm relative reflectances to the relative reflectance values of all filters in the other eye. The error bars represent the standard deviation of the individual pixel spectra used to generate each average.
low RMS model simulations of the observed spectra (Fig. 1).The Shade endmember is not a physical material but rather re-presents the “dark point” in the image scene; i.e., an ideal shadeendmember would be a shadowed surface with zero reflectanceat all wavelengths. In the SMA implementation used in thisstudy, the shade image endmember was first subtracted from thematerial endmember (Gray Rock or Bright Dust) spectra (Adamset al. 1986). SMA was then run using only the nonShade mate-rial endmembers. The Shade fraction image was then calculatedby subtracting the sum of the material endmember fraction im-ages from 1.0. We note that the Shade endmember certainly in-cludes a diffuse red (and nonlinear) “skylight” component (e.g.,Thomas et al. 1999). While subtracting Shade from the materialendmember spectra does not rigorously remove the diffuse illu-mination component from the scene, it does allow us to makeestimates of material endmember fractions that are much less(if at all) influenced by sky illumination. Analysis by Bell et al.(2000) has also shown that the sky illumination component isquite spectrally bland, with a smooth spectrum consisting of a
reddish slope in the visible and a flat slope in the near-IR. Thelack of distinct crystal field absorption bands in the spectrum
of martian sky light (consistent with its origin from scatteringby poorly crystalline, micron-sized ferric-bearing dust grains)further argues against diffuse illumination being a significantsource of additional near-IR spectral variability in the IMP dataanalyzed here.

The endmembers used in this part of the analysis were re-ferred to as image endmembers; i.e., averages of a small num-ber of pixels taken to be the best, “purest” pixels of the materialpresent in the scene (Adams et al. 1986, 1993). Image end-members are not necessarily the same as reference endmembers,which are laboratory or field measured spectra of pure minerals,mineral mixtures, or rocks of known compositions. Findingthe best image endmember pixels is often difficult. However,a processing approach implemented with the commercial ENVIsoftware package (Research Systems, Inc.) and developed ini-tially for use with terrestrial airborne hyperspectral datasets(Boardman et al. 1995) has proven to be quite effective in findingappropriate image endmember pixels. This approach consistsof first reducing the number of required channels by subject-
ing the multispectral data to a variant of principal componentsanalysis known as the minimum noise fraction (MNF) transform

E
60 BELL
FIG. 2. Two-dimensional scatter plot representation of first three MNFbands of a segment from octant S0184. The data cloud represented by the firstthree MNF bands (axes shown) was rotated to an orientation where it showed thelargest amount of variation, and then it was collapsed onto this two-dimensionalplot. The ellipse at the upper vertex of the triangular shaped data cloud definesshade image endmember pixels; the ellipse on the lower vertex defines the BrightDust image endmember and the ellipse on the right defines the Gray Rock imageendmember.

(Green et al. 1988). The MNF segregates signal into the first sev-eral components and noise into the later components. For largerscenes there is normally an intermediate step to reduce the num-ber of spatial elements that are examined; however, in this casethe image segments making up the octants were relatively smallin size and could be considered as a whole (except for thosecases of masked pixels noted above). A small number of end-members accounted for most of the spectral variability in eachsegment. Hence, for each segment all pixels from the first threeto four MNF bands were examined in three- to four-dimensionalspace through the use of an n-dimensional visualization tool inthe commercial ENVI software package. Figure 2 shows onerepresentation of data from the first three MNF bands of a givensegment. The triangular shape of the data cloud reveals the lo-cation of at least three potential image endmembers (Gillespieet al. 1990, Adams et al. 1993). Endmembers determined here bythe n-dimensional visualization approach generally consisted ofShade, Bright Dust, and Gray Rock. Having determined a set ofpixels corresponding to the extreme borders of the MNF trans-formed data cloud, averages of these pixels (e.g., averages of thepixels occurring within each of the ellipses in Fig. 2 for octantS0184) were used as the image endmembers in SMA (cf. Smithet al. 1990).

Alternatively, some endmembers or anomalous materialswere mapped using the foreground/background analysis (FBA)

T AL.

technique (Smith et al. 1994). FBA is an approach similar toSMA but computationally different in that it produces abun-dances not necessarily for single endmembers but rather forendmember groups, namely the desired “foreground” spectrawith undesired “background” spectra nulled out. Like SMA,FBA produces fraction images in which the value of each pixelis related to the fractional abundance of the material (or group offoreground materials) in question. Fraction images produced byFBA can have superior contrast between foreground and back-ground materials because the approach is not restricted to using asingle image endmember spectrum to represent each image end-member. For example, multiple spectra of a single image end-member, illuminated under different geometries, can be usedto achieve superior suppression of the undesired backgroundspectra. In this study, FBA was used with select spectra of low-abundance materials for the purpose of producing high-contrastfraction images of those materials.

RESULTS

Endmember Fraction Images

As noted above, the MNF and n-D visualization approach toendmember selection followed by examination of the RMS errorimages generally shows that most segments are well modeledby just three endmembers (Figs. 1 and 2). For example, Fig. 3shows representative fraction images of Shade, Bright Dust, andGray Rock from segment 20 of octant S0188.

The fraction images themselves generally indicate the ex-pected patterns, e.g., relatively bare rock surfaces have fractionsnear 1.0 in the rock fraction images. Closer inspection, how-ever, reveals several interesting patterns. The endmember usedto model soils in Fig. 3 was a “Bright Dust” endmember. Ingeneral, each segment was modeled using a set of image end-members drawn from within that segment itself. Consequently,there could be minor, but noticeable, spectral differences be-tween the Bright Dust endmember used in one segment and thatused in another. Comparison with the spectra published by Bellet al. (2000) indicates that the Bright Dust endmembers selectedin the present work correspond to their “Bright I and II” classes.The “Dark” class of Bell et al. (2000) did not constitute a spec-trally distinct endmember per se, but as part of the n-dimensionalvisualization stage of this analysis, the Dark Soils did constitutea discrete cluster within the confines of the larger data envelope(if Fig. 2 were three-dimensional, this cluster would be just be-low the Shade cluster and out of the plane of the page towardsthe reader). When viewed in the fraction images, some DarkSoils have detectable fractions even in the Bright Dust fractionimages. This suggests that there may be more variability amongthe low albedo soils than was detected or inferred in the ear-lier study by Bell et al. (2000). Those authors defined a singleDark Soil class; however, the present work indicates at least twospectrally distinct types of Dark Soil. These include (1) the afore-
mentioned Dark Soils that have intermediate to low fractionalvalues in Bright Dust fraction images and low fractional values

ANOMALOUS MATERIALS AT THE PATHFINDER LANDING SITE 61

FIG. 3. Fraction images and RMS error image for octant S0188 left-eye data segment 20. (A) Left-eye 443-nm band. (B) Shade fraction image. (C) Bright Dustfraction image. (D) Rock fraction image (the arrow points to a crescent of Rock Soil in the foreground, to the left of and under the 28-cm-wide rock named Calvin).
(E) RMS error image; arrow indicates Black Rock occurrence. For the fraction images, white = 100% abundance and black = 0% abundance. (F) Endmemberspectra used to generate the model images in (A) through (E).

62 BELL ET AL.

FIG. 4. Example of an occurrence of Gray Rock Soil. (A) Composite of left-eye bands 3, 2, and 1 (802, 671, and 443 nm) from segment 6 of octant S0185.Bare rock surfaces have a blue hue and so does the patch of smooth, unperturbed soil that runs diagonally across the ellipse. (B) Gray Rock fraction image, fromSMA analysis of this segment. The patch of light-toned material that runs diagonally across the ellipse in this image exhibits relatively high Gray Rock fraction
values in the mixing model of this segment, but it is clearly a soil-like or duneform deposit. As discussed in the text, this may indicate that this Gray Rock Soil unit
ie
could include or consist of comminuted fragments of rocks with spectral propert
in the Rock fraction images and (2) a soil unit that is bright inthe Rock fraction images. The variability of low albedo soils isdiscussed further below.

In certain scenes, patches of soil, generally surrounding largerocks but sometimes isolated, have relatively high Gray Rockfraction values. Figure 3d shows an example of this “Gray RockSoil” unit near the rock named Calvin. Figure 4 shows an occur-rence of this unit associated with a low albedo duneform deposit.The fact that there are occurrences of this unit away from largerocks (as in Fig. 4) strongly indicates that it is not just the resultof scattered light from adjacent rock surfaces. However, the unitgenerally is seen surrounding larger rocks (Fig. 3c) and the re-flectance spectrum of the unit shows similarities to a Gray Rockspectrum in the mineralogically diagnostic 800–1000-nm region(Fig. 5). Thus, there may be a genetic relationship between therocks and the Gray Rock Soil unit. This possibility is exploredin further detail below.

There were some data segments that could be better modeledusing four endmembers instead of three. Because the processingwas done on left- and right-eye segments separately, data fromthe left and right eyes did not necessarily require the same num-ber of endmembers, mainly because of the different wavelengthbands available in each eye (Smith et al. 1997a). For example,in left-eye segments where disturbed soil was relatively abun-dant, that material was spectrally distinct enough in the 800–1000-nm region to be used as a fourth endmember. Alterna-tively, in right-eye segments where dust-covered rocks wereprominent, including notably some rocks that were caked withdust on their lower portions but were relatively clean near theirtops, the lower rock portions were bright in the RMS error im-ages, indicating that these portions were not well modeled byeither rock or dust. Such instances are likely examples of non-
linear mixing between optically thin dust coverings or rinds thatdo not completely spectrally mask the underlying rock substrate
s similar to those at the landing site.

FIG. 8. (a) RGB composite of right-eye bands at 671, 531, and 443 nm forsegment 8 of octant S0187. Patches of Intermediate Soil discussed in the textoccur in the lower left (arrows). (b) Subsection of octant S0188 mosaic centeredon the rock called Desert Princess. Left-eye bands 3, 2, and 1 (802, 671, and443 nm) are displayed in R, G, and B for the color composite. (c) Subsection ofRMS error image for segment 13 of octant S0188 with Desert Princess at lower
center (aligned the same relative to panel (b)). “Anomalous Patches” materialsseen on this rock are noted with an arrow.

H
FIG. 5. Combined left- and right-eye relative reflectance spectra of BrightDust, Dark Soil, Gray Rock, and Gray Rock Soil materials from SuperPan octantS0185. Each spectrum is the average of many tens of spectra of each material,identified through SMA and RMS images. The error bars represent the standarddeviation of the spectra used to generate each average.

e.g., Fischer and Pieters 1993, McSween et al. 1999, Johnsonand Grundy 2001).

Anomalous Materials Identified in RMS Error Images

Black Rock. The best examples of anomalous, low-abundance materials identified in the RMS error images thatresulted from SMA of the left-eye segments were Black Rockcobbles and pebbles. Figure 3e shows the RMS error imagefrom a portion of octant S0188. The bright region in the upperleft (noted with an arrow in the RMS error image in Fig. 3e)was not well modeled by the three image endmembers used andis an occurrence of the Black Rock lithology. Another exampleof an identification of the Black Rock lithology is provided inFig. 6. Figure 6a shows a 443-nm image of a portion of octantS0184. Figure 6b shows the RMS error image for that segmentin which the Black Rock cobble has an anomalously high RMSvalue. Figure 6c shows a fraction image for that segment de-
rived from foreground/background analysis, providing superiorcontrast between the Black Rock cobble and the background.

Black Rock is one of the anomalous rock types also discussedby Murchie et al. (2000). While the pebbles and cobbles that wegroup together under the category of Black Rock share manyimportant spectral characteristics, they differ in others. Figure 7compares relative reflectance spectra of four types of BlackRock occurrences to a typical Gray Rock spectrum. None ofthe Black Rock spectra display a complete “1-µm” band. Theright half of the absorption is truncated because IMP spectralcoverage ended at 1 µm. However, the S0183 segment 5 ex-ample and, to a lesser extent, the S0185 segment 12 exampleshow increases in reflectance in the longest wavelength band,whereas the Black Rock spectra from segments 8 and 14 appar-ently have minima at wavelengths longer than 1000 nm. Bandminima near 950 nm and lower for ferrous-iron-bearing mate-rials are consistent with low-calcium pyroxene (orthopyroxeneand pigeonite), and band minima longer than 1000 nm are consis-tent with high-calcium pyroxene (clinopyroxene) and/or olivine(e.g., Adams 1974, Cloutis et al. 1986, Sunshine and Pieters1993). These band positions are also observed for the SNC me-teorites, with the orthopyroxenite and basaltic SNCs having bandminima between 900 and 1000 nm and the clinopyroxenite anddunite SNCs having band minima at wavelengths longer than1000 nm (e.g., Gaffey 1976, Bishop et al. 1998, Morris et al.2000).

Intermediate soil. Our analysis identified a correspon-dence between small clusters of high RMS error pixels andlarger patches of soil with a morphology suggestive of an in-creased level of cohesion that may indicate induration or com-paction (e.g., Arvidson et al. 1989, Moore et al. 1999), thoughnot to the degree attained by “Scooby Doo” and other moreclearly indurated soil deposits observed elsewhere at thelanding site (e.g., Moore et al. 1999, McSween et al. 1999,Bell et al. 2000). A good example of this material occursin octant S0187. A color composite of a segment from oc-tant S0187 is shown in Fig. 8a. Composite left- and right-eyespectra of this unit, which we are calling “Intermediate Soil,”are shown in Fig. 9a. The S0187 segment 8 #1 and S0187segment 8 #2 spectra shown in Fig. 9a are 8- and 12-pixelaverages, respectively, from the arrowed regions in Fig. 8a.The S0187 segment 8 #3 spectrum represents the average ofsmall, 1-to-3-pixel occurrences of a material with an even deeper900-nm absorption, scattered throughout this unit. The smalloccurrences of this subset of Intermediate Soil material havedeeper 900-nm band depths (∼10% versus ∼5%) than in thebroader exposures of Intermediate Soil indicated by the seg-ment 8 #1 and segment 8 #2 spectra in Fig. 9a. Associationof these small clusters with the morphologically distinct In-termediate Soil unit in octant S0187 indicates that the small1-to-3-pixel clusters might be a coarser-grained or fractionallymore abundant component of the surrounding Intermediate Soildeposits.

Freckles on the “Desert Princess” in S0188 segments 13,
15. Another unusual material was found in the S0188 octant(Fig. 8b). The RMS error image (Fig. 8c) showed a number of

64 BELL ET AL.

FIG. 6. (A) 443-nm image from a portion of S0184. Black Rock cobble is circled. (B) RMS error image resulting from SMA analysis of segment shown ina
(A). (C) Foreground/background analysis fraction image of Black Rock for the s
groupings of high RMS error pixels scattered across the face ofthe ∼27-cm-wide rock called Desert Princes. These high RMSerror pixels corresponded to pixels that displayed a lighter tonewhen viewed in a composite of the 802, 671, and 443-nm left-eye channels, shown in Fig. 8b. Examination of the spectra ofthese pixels showed a discrete ∼900-nm absorption with, un-like the Intermediate Soil material, a high reflectance at the671 through 802-nm wavelengths. The 900-nm absorption hasa ∼7% band depth. A combined right- and left-eye spectrumof this unit, which we are calling “Anomalous Patches,” is pre-sented in Fig. 9b.

Reference Endmember Modeling

In an attempt to constrain further the mineralogic interpre-tation of image endmember spectra, we also performed a pre-

me segment.

liminary study using reference endmember modeling (REM)techniques. REM was one of the methods used by Adams et al.(1986) in their spectral mixture analysis of Viking Lander 1multispectral image data. As noted above, image endmembersare pixel spectra taken to represent the purest materials in thescene; reference endmembers are laboratory or field measuredreflectance spectra of known materials. While later iterationsof the Viking lander multispectral data were more fully cali-brated, the data set used in the Adams et al. (1986) study wasincompletely calibrated, so that the equations used to solve forimage and reference endmembers included terms for gain andoffset. The well-calibrated data used in this study enabled com-parison of image endmember spectra with potential referenceendmembers.

Our REM analysis used a sample set of combined left- andright-eye 12-channel image endmember spectra of Shade, Bright

H
FIG. 7. Four examples of Black Rock occurrences compared with a typicalGray Rock spectrum.

Dust, Gray Rock, and Black Rock from octant S0185, segment12 and octant S0183, segment 5. We created a reference end-member library (Table II) consisting of 43 spectra, includingpalagonitic and sulfatetic tephras from Mauna Kea (Morris et al.2000), ferric oxyhydroxide precipitates (e.g., Farrand 1997),palagonite tuffs from tuff cones (Farrand and Singer 1993), SNCmeteorites (Morris et al. 2000), komatiites and columnar basalts(Morris et al. 2000), select pure mineral spectra from the USGSspectral library (Clark et al. 1993), and reference shade (con-stant 0.01 reflectance), used essentially as a proxy for uniformlyincreasing or decreasing the reflectivity of model mixtures. Thereference endmember spectra were convolved to the bandpassesof the IMP filters.

We modeled the image endmembers as successive combina-tions of the candidate reference endmembers. Specifically, eachimage endmember was modeled using a linear combination ofthree reference endmembers from the library: shade, a bright,altered “dust” endmember, and a darker, less altered “rock” end-member. In the model, the sum of the fractions was constrained
to equal 1.0, and to allow for noise and calibration uncertaintiesin the data, individual reference endmember fractions were al-

lowed to range from −0.1 to 1.1. The RMS error (as defined byEq. (2)) was also calculated for each combination, and the modelresults were ranked by this goodness-of-fit metric. Because themodels produced typical best-fit RMS residuals comparable inmagnitude to the uncertainties in the data, additional model-ing with greater numbers of reference endmembers per imageendmember was not warranted.

Unfortunately, our REM results using the library in Table IIand the approach outlined above did not yield solutions thatcould uniquely constrain the mineralogy of the Bright Dust,Gray Rock, and Black Rock image endmembers that we exam-ined. For example, for each image endmember considered, eachmodeling run through the many thousands of possible REM li-brary combinations returned from 20 to 30 cases with equallygood RMS values and having reference endmember fractionsthat exceeded 0.8 for their associated image endmember (an ar-bitrary metric for a “close fit”). Examples are shown in Fig. 10and Table III. While many model spectra were generated whichfit the image endmembers quite well, in general the best-fit

FIG. 9. (a) Spectra of Intermediate Soil material from octant S0187(Fig. 8a) and of the indurated soil Scooby Doo (spectrum 19 1 of Bell et al.2000). The “S0187 segment 8 #1” and “S0187 segment 8 #2” spectra are com-bined left- and right-eye spectra over 8- and 12-pixel averages, respectively, ofthe Intermediate Soil material. The “S0187 segment 8 #3” spectrum is a left-eye3-pixel average representative of the 1-to-3-pixel clusters observed in the left-eyedata that have a deeper 900-nm band and that can occur in isolation from larger,morphologically distinct deposits of Intermediate Soil. (b) Combined left- andright-eye spectrum of an average of 9 pixels from several clusters of Anomalous
Patches material on the rock called Desert Princess in segments 13 and 15 ofoctant S0188 (Fig. 8b).

66 BELL ET AL.

FIG. 10. Examples of reference endmember modeling (REM) analysis of image endmember spectra identified from our SuperPan multispectral imageanalyses. Gray Rock, Black Rock, and Bright Dust image endmembers from octant S0183 segment 5 (left column of plots) and S0185 segment 12 (rightcolumn of plots) were modeled using the REM library in Table II. In each plot, open symbols and dotted lines are the library spectra used to model each image

e
endmember (open square symbols), and the lowest RMS model spectrum fromdetails.
models did not provide consistent mineralogic inferences foreach image endmember. For example, models of the Black Rockimage endmember having equally low RMS values were
generated (a) using an altered, bright dust endmember with astrong crystalline ferric absorption in the 900–1000-nm region
ach entry in Table III is plotted as a solid line and filled circles. See text for

mixed with a dark, relatively spectrally neutral rock endmember,and (b) using an altered, nanophase iron-oxide-bearing brightdust endmember mixed with a dark rock endmember that ex-
hibits strong ferrous absorption in the 900–1000-nm region.By the metric of RMS error, both of these models are equally

H
TABLE IIReference Endmember Library Used in Our Analysis

No. Material Ref.

Shade1 Constant 0.01 reference shade —

“Dust” REM suite2 HWMK-602 < 1 mm 13 HWMK-600 2–5 mm 14 HWMK-600 < 1 mm 15 VOL02A < 2 mm 16 PN-9 < 1 mm 17 VOL02A < 1 mm 18 HWMK-515 < 5 mm 19 HWMK-612 < 1 mm 1

10 HWMK-511 < 1 mm 111 HWMK-606 < 1 mm 112 Clear Creek Avg. (acid mine drainage precipitate) 213 CCnf-3 (acid mine drainage precipitate) 214 CC-2c (palagonite tuff from Cerro Colorado tuff cone) 315 PB-2b (palagonite tuff from Pavant Butte tuff cone) 316 Sv98-17 (jarosite) 217 Sv98-14 (acid mine drainage precipitate) 218 Sv98-5 (acid mine drainage precipitate) 2

“Rock” REM suite19 Komatiite basalt 1 120 Komatiite basalt 3 121 Komatiite basalt 4 122 Komatiite basalt 5 123 WD232 columnar basalt 2 124 WD232 columnar basalt 3 125 Utah basalt 1b (Markagunt Plateau) 426 Utah basalt 1a (Markagunt Plateau) 427 Lunar Crater 96-5 (Nevada) 428 Lunar Crater 96-9 (Nevada) 429 Cuprite basalt (Nevada) 430 EETA79001 lithology A 131 EETA79001 sample 65 132 ALH77005 133 Shergotty 134 Zagami 135 Lafayette 136 Nakhla 137 Governador Valadares 138 ALH84001 139 Augite NMNH120049 540 Bronzite HS9.3B 541 Hypersthene PYX02.h 542 Hypersthene NMNHC2368 543 Pigeonite HS199.3B 5

Note. References: [1] Morris et al. 2000; [2] Farrand 1997; [3] Farrand andSinger 1993; [4] W. H. Farrand, unpublished data; [5] Clark et al. 1993.

good however, each has substantially different compositionalimplications for the interpretation of the Black Rock imageendmember (i.e., one used the broad 900-nm ferric band tomimic the strong Black Rock 900–1000-nm reflectance fea-
ture, and the other used the broad 950-nm ferrous band). Sim-ilar interpretive nonuniqueness was encountered in our REM

analysis of Bright Dust and Gray Rock image endmemberspectra.

We believe that it was important to attempt this simple ini-tial REM modeling approach, because several previous multi-spectral imaging and imaging spectroscopic studies have shownthat REM analysis can reveal unique mineralogic insights (e.g.,Farrand 1997, Farrand and Singer 1993, Farrand et al., in prepa-ration). However, it is clear from our study that either an ex-panded reference library (including additional mineral endmem-bers as well as consideration of physical effects like grain sizeand sample texture), higher spectral resolution imaging, and/ora more refined similarity metric than RMS is required in orderto derive more compositionally unique and diagnostic solutionsfrom quantitative mixing methods like REM. Indeed, we areexploring additional quantitative metrics such as spectral anglemeasures, spectral feature (band depth) fitting, and spectral cor-relation measures to compare “unknown” spectra to a spectrallibrary as part of an expanded companion study to this paper(Farrand et al., in preparation).

DISCUSSION

Observations of previously unreported materials made in thisstudy fall into two broad classes: widespread, but spectrallysubtle materials, and isolated, small occurrences of spectrallyvery distinct materials. The Gray Rock Soil unit falls into thefirst class. There are numerous occurrences of this unit, mainlyconcentrated around rocks and boulders, but also in isolateddrifts (e.g., Figs. 3d and 4). It is widespread, and as the spectrain Fig. 5 show, it is superficially spectrally similar to “normal”Dark Soil. However, Gray Rock Soil is spectrally distinguishablefrom Dark Soil using SMA because it exhibits a near-IR bandwith a center near 950 nm (more like Gray Rock) rather than near860–900 nm (more like Bright Dust). In fact, Gray Rock Soilmight not have been recognized as a separate unit were it not forthe use of spectral mixture analysis of the scene using just GrayRock, Bright Dust, and Shade endmembers (e.g., Figs. 3 and4). In those models, some regions exhibiting the morphologyand texture of soils showed up prominently in the Gray Rockfraction images. These regions were subsequently identified asGray Rock Soil patches.

Because of its general physical association with nearby rocks,the Gray Rock Soil unit could be the result of mechanical weath-ering of the rocks. That is, the Gray Rock Soil unit could be thecomminuted remains of spalls and flakes from the rocks at thelanding site. However, the existence of some isolated Gray RockSoil patches (not directly or obviously associated with an adja-cent rock) indicates that this unit may be locally mobile—that is,transported by occasional strong winds and collected into smalldrift or duneform deposits. Spectrally, Gray Rock Soil appearsto be similar to Gray Rock in the 800–1000-nm region, but toDark Soil in the 440–800-nm region. This implies that Gray
Rock Soil may be a physical mixture of coarser-grained aeolianfines (Bell et al. 2000) and small Gray Rock spalls and flakes.

68

are the Black Rofirst described by

BELL ET AL.

TABLE IIIRepresentative Reference Endmember Modeling (REM) Results for Each Image Endmember

from IMP Super Pan Octant S0183 Segment 5 and Octant S0185 Segment 12

Gray Rock image endmember

Octant S0183, Segment 5

“Dust” REM “Dust” fraction “Rock” REM “Rock” fraction Shade fraction RMS

HWMK606 < 1 mm 0.2081 ALH77005 0.7583 0.0336 0.0049HWMK606 2–5 µm 0.1660 ALH77005 0.7676 0.0664 0.0050PN9 < 1 mm 0.3422 Kom. Basalt 1 0.7173 −0.0595 0.0086



PN9 < 1 mm 0.3065 Kom. Basalt 1 0.7296 −0.0361 0.0044HWMK606 2–5 µm 0.1457 ALH77005 0.7812 0.0731 0.0049HWMK606 < 1 mm 0.1824 ALH77005 0.7735 0.0440 0.0051

Black Rock image endmember



VOL02A 0.2184 Lafayette 0.8491 −0.0675 0.0112VOL02A 0.2105 ALH77005 0.8164 −0.0269 0.0123HWMK606 < 1 mm 0.1240 ALH77005 0.9515 −0.0755 0.0125



HWMK515 < 5 µm 0.1165 ALH77005 0.9152 −0.0317 0.0088HWMK515 < 5 µm 0.1500 Lafayette 0.8042 0.0458 0.0116PN-9 < 1 mm 0.1373 Nakhla 0.7503 0.1124 0.0157

Bright Dust image endmember



HWMK606 < 1 mm 0.9940 Hypersthene NMNH 0.1025 −0.0965 0.0056HWMK606 < 1 mm 1.0244 EETA79001 lith. A 0.0455 −0.0699 0.0060HWMK600 2–5 µm 0.8195 EETA79001 lith. A 0.0591 0.1214 0.0061



HWMK606 < 1 mm 0.8505 EETA79001 1–10 mm 0.0324 0.1171 0.0064HWMK606 < 1 mm 0.8505 Bronzite HS9.3B 0.0190 0.1305 0.0064
HWMK606 < 1 mm 0.8555 Hypersthene PYX 0.0320 0.1125 0.0066
Alternately, individual Gray Rock Soil particles may be smallflakes or spalls of dust-coated Gray Rock surfaces. For example,Adams et al. (1986) performed an SMA study of Viking Lander1 multispectral data and also observed the presence of a darksoil, putatively derived by mechanical weathering of rocks.

Prominent among the second category of spectrally very dis-tinct, small in size, and relatively sparse in occurrence materials

ck pebbles and cobbles. This lithology wasMurchie et al. (2000). These authors noted

that while the elemental abundances of the Gray Rock litho-logy (the “soil free rock” of Rieder et al. 1997 and McSweenet al. 1999) are consistent with a normative high-Ca pyroxene,the shorter wavelength absorption evident in the Black Rocklithology could be indicative of a greater abundance of low-Ca pyroxene. While our reference endmember modeling studydid not generate unique compositional solutions for the Black
Rock image endmembers, the best-fit models consistently in-cluded rock reference endmembers containing or consisting of

H
relatively high-Ca pyroxene and/or olivine (e.g., ALH77005,Lafayette, Governador Valaderes; Clark et al. 1993, McSween1994) as well as dust reference endmembers containing well-crystalline ferric oxide minerals (e.g., altered palagonitic tephrasand ferric precipitates; Farrand 1997, Morris et al. 2000). Thepresence of these kinds of materials in models of the Black Rockspectra is consistent with the need for a spectral component thatcan match the increased strength and perhaps shorter wavelengthposition of the 900–1000-nm absorption seen in Black Rockimage endmember spectra compared to Gray Rock image end-member spectra (Fig. 7). In fact, it is possible that both ferric andferrous components could be responsible for the nature of the900–1000-nm absorption seen in Black Rock image endmem-ber spectra. For example, Morris et al. (1995) analyzed spectraof mixed ferric/ferrous impact melt rocks from Manicouaganimpact crater and found that such mixtures could provide goodmatches to some Mars telescopic spectra in the 900–1000-nmregion.

The best examples of the Black Rock lithology (i.e., thoseoccurrences that are most prominent in the RMS error imagesresulting from SMA) all occur among the smallest size frac-tion of rocks (cm-sized cobbles and pebbles; Golombek et al.1999b). While other explanations are possible, this observationis consistent with these materials having been transported thegreatest distances from the landing site by the Ares and TiuVallis outflow channel flood events. Another possibility is thatthese small anomalous rocks are allochthonous fragments ofmaterial ejected laterally by impact events.

Murchie et al. (2000, 2001) also noted what they termed anOrange Rock lithology with a reflectance maximum at 750 nmand a relatively short band minimum in the 900-nm channel.While the particular example of Orange Rock cited in Murchieet al. (2000) did not show up as anomalous in our SMA analysisof the segment containing it (S0188), their example is compara-ble to some Black Rock occurrences that we found, such as theone from octant S0183, segment 5 shown in Fig. 7.

Also in the category of spectrally distinct but rare in occur-rence are the small, spectrally Anomalous Patches material onthe rock called Desert Princess from the S0188 octant. Thesepatches could be vesicles filled with altered soil or dust, cobblesin a conglomerate rock, or perhaps even phenocrysts of a compo-sition distinct from the rock matrix. Unfortunately, the patchesare near the limit of resolution of the IMP camera, and a searchof the Sojourner rover camera image archive revealed no close-up rover images of this particular rock. The material’s limitedspatial extent, coupled with the wavelength-dependent decreasein IMP spatial resolution and the generally lower calibrationfidelity of the IMP 965-nm filter data (e.g., Reid et al. 1999), re-sults in relatively large uncertainties in the spectrum of this unit,especially at the longest wavelengths. The Anomalous Patchesmaterial spectrum in Fig. 9b shows evidence for a narrow 900–950-nm absorption band that could be consistent with the pres-ence of low-Ca pyroxenes such as those studied in lab samples by
Sunshine and Pieters (1993) or in SNC meteorites (Bishop et al.1998, Morris et al. 2000). However, signal-to-noise ratio limi-

tations prevent a more detailed compositional analysis of thismaterials’s spectral properties using the methods described here.

Small occurrences of Intermediate Soil material with a dis-tinct 900-nm band minimum share some spectrally distinct fac-tors with both Black Rock and Anomalous Patches materials.Intermediate Soil is widely distributed in small, discrete spotsor clusters of pixels seen to occur in the RMS error images froma large number of image segments. It is associated in severalinstances with larger occurrences of morphologically distinct“outcrops” of what appears to be indurated, compacted, or par-tially cemented soil. These soil deposits generally exhibit a ∼5%deep 900-nm absorption, with an even deeper band occurring inthe smaller occurrences noted in the RMS error images. Thesmaller clusters with the deeper 900-nm band might be smalleroccurrences of a more fractionally abundant or coarser-grainedcementing agent like a ferric oxide or oxyhydroxide (e.g., Clarkand Van Hart 1981, Bishop et al. 1995, Morris et al. 2000),which in the larger occurrences of Intermediate Soil is spectrallymasked by other lower albedo sedimentary grains. Alternately,this narrow short wavelength feature may indicate the presenceof low-Ca pyroxene (e.g., Sunshine and Pieters 1993, Bishopet al. 1998, Morris et al. 2000), presumably from intermixedsubresolution rock or soil fragments.

CONCLUSIONS

We have examined recalibrated and spatially registered Im-ager for Mars Pathfinder multispectral panoramas and havefound evidence for small, spatially coherent and spectrally dis-tinct deposits of previously unrecognized materials. These mate-rials fall into four main categories: (1) “Gray Rock Soil,” whichconsists of low albedo soil or drift deposits that are spectrallysimilar to the Gray Rock spectral unit of McSween et al. (1999)in the 800–1000-nm region and to the Dark Soil spectral unit ofBell et al. (2000) in the 440–800-nm region and that are proba-bly physically weathered spalls or fragments of locally occurringGray Rocks; (2) “Black Rock,” which consists of small possiblypyroxene-bearing pebbles and cobbles with a distinctly strongerand slightly shorter-wavelength “1-µm” absorption feature thanthe ubiquitous Gray Rock unit of McSween et al. (1999) andwhich is similar to the anomalous materials initially identifiedby Murchie et al. (2000); (3) “Intermediate Soil,” which consistsof widespread but small deposits of bright, possibly indurateddrift material with a deeper than average 900-nm band that mightbe caused by a ferric-rich cementing agent; and (4) “AnomalousPatches,” observed so far only on the rock known as DesertPrincess, and consisting of small occurrences of material withan even deeper 900-nm band than Intermediate Soil with a highreflectance at the 670 through 800 nm wavelengths.

These results reinforce the common assumption that higherspatial resolution, well-calibrated multispectral imaging, andspectroscopy investigations on future Mars lander and rover mis-sions are likely to enable the identification of even more spec-
trally diverse but perhaps spatially rare rock and soil lithologieson the surface of Mars.

E
70 BELL
ACKNOWLEDGMENTS

We thank T. Sucharski, J. Anderson, K. Becker, A. Howington-Kraus, R. Kirk,and K. Herkenhoff from USGS/Flagstaff for their assistance with the softwaredevelopment used for the calibration, registration, and mosaicking of the IMPdata described here. We also thank S. Murchie and O. Barnouin-Jha for helpfulinitial discussions on the accuracy of the IMP SuperPan calibration and registra-tion. We are grateful to Lisa Gaddis, Janice Bishop, and Peter Smith for insightfulcomments and reviews of an earlier draft of this paper. This research was sup-ported by Grant NAG5-9774 (J.F.B) and by RTOP 334-34-72-01 (R.V.M.) fromthe NASA Mars Data Analysis Program.

REFERENCES

Adams, J. B. 1974. Visible and near-infrared diffuse reflectance spectra of py-roxenes as applied to remote sensing of solid objects in the Solar System.J. Geophys. Res. 79, 4829–4836.

Adams, J. B., M. O. Smith, and P. E. Johnson 1986. Spectral mixture modeling:A new analysis of rock and soil types at the Viking Lander 1 site. J. Geophys.Res. 91(B8), 8098–8112.

Adams, J. B., M. O. Smith, and A. R. Gillespie 1993. Imaging spectroscopy:Interpretation based on spectral mixture analysis. In Remote GeochemicalAnalysis: Elemental and Mineralogical Composition (C. M. Pieters andP. A. J. Englert, Eds.), pp. 145–146. New York: Cambridge Univ. Press,Cambridge, UK.

Arvidson, R. E., J. L. Gooding, and H. J. Moore 1989. The martin surface asimaged, sampled, and analyzed by the Viking landers. Rev. Geophys. 27,39–60.

Bell, J. F., III, and 23 colleagues 2000. Mineralogic and compositional propertiesof martin soil and dust: Results from Mars Pathfinder. J. Geophys. Res. 105,1721–1756.

Bishop, J. L., C. M. Pieters, R. G. Burns, J. O. Edwards, R. L. Mancinelli, andH. Froschl 1995. Reflectance spectroscopy of ferric sulfate-bearing montmo-rillonites as Mars soil analog materials. Icarus 117, 101–119.

Bishop, J. L., C. M. Pieters, T. Hiroi, and J. F. Mustard 1998. Spectroscopicanalysis of martian meteorite Allan Hills 84001 powder and applicationsfor spectral identification of minerals and other soil components on Mars.Meteorit. Planet. Sci. 33, 699–707.

Boardman, J. W., F. A. Kruse, and R. O. Green 1995. Mapping target signaturesvia partial unmixing of AVIRIS data. Summaries of the Fifth Annual JPLAirborne Earth Science Workshop, Washington, D. C., Vol. 1. JPL Publ. 95-1,pp. 23–26. Jet Propulsion Laboratory, Pasadena.

Bruckner, J., G. Dreibus, R. Rieder, and H. Wanke 2001. Revised data of theMars Pathfinder Alpha Proton X-Ray Spectrometer: Geochemical behaviorof major and minor elements. Lunar Planet. Sci. 32, 1293 (abstract).

Clark, B. C., and D. C. Van Hart 1981. The salts of Mars. Icarus 45, 370–378.

Clark, R. N., G. A. Swayze, A. J. Gallagher, T. V. V. King, and W. M. Calvin1993. The U.S.G.S. Digital Spectral Library: Version 1: 0.2 to 3.0 µm. U.S.Geol. Surv. Open File Rep. 93-592.

Cloutis, E. A., M. J. Gaffey, T. L. Jackowski, and K. L. Reed 1986. Cali-bration of phase abundance, composition, and particle size distribution forolivine–orthopyroxene mixtures from reflectance spectra. J. Geophys. Res.91, 11,641–11,653.

Farrand, W. H. 1997. Identification and mapping of ferric oxide and oxyhydrox-ide minerals in imaging spectrometer data of Summitville, Colorado and thesurrounding San Juan Mountains. Int. J. Rem. Sens. 18, 1543–1552.

Farrand, W. H., and R. B. Singer 1993. A comparison of the visible and nearinfrared reflectance of hydrovolcanic palagonite tuffs and martian weatheredsoils. Lunar Planet. Sci. 24, 465–466.

Fischer, E. M., and C. M. Pieters 1993. The continuum slope of Mars: Bidirec-
tional reflectance investigations and applications to Olympus Mons. Icarus102, 185–202.
T AL.

Gaffey, M. J. 1976. Spectral reflectance characteristics of the meteorite classes.J. Geophys. Res. 81, 905–920.

Gillespie, A. R., M. O. Smith, J. B. Adams, S. C. Willis, A. F. Fischer III,and D. E. Sabol 1990. Interpretation of residual images: Spectral mixtureanalysis of AVIRIS images, Owens Valley, California. In Proceedings of the2nd AVIRIS Workshop, Pasadena, pp. 243–270.

Golombek, M. P., R. A. Cook, H. J. Moore, and T. J. Parker 1997. Selec-tion of the Mars Pathfinder landing site. J. Geophys. Res. 102, 3967–3988.

Golombek, M. P., and 53 colleagues 1999a. Overview of the Mars Pathfindermission: Launch through landing, surface operations, data sets, and scienceresults. J. Geophys. Res. 104, 8523–8554.

Golombek, M. P., H. J. Moore, A. F. C. Haldemann, T. J. Parker, and J. T.Schofield 1999b. Assessment of Mars Pathfinder landing site predictions.J. Geophys. Res. 104, 8585–8594.

Green, A. A., M. Berman, P. Switzer, and M. D. Craig 1988. A transformationfor ordering multispectral data in terms of image quality with implicationsfor noise removal. IEEE Trans. Geosci. Rem. Sens. 26, 65–74.

Johnson, J. R., and W. M. Grundy 2001. Visible/near-infrared spectral andtwo-layer modeling of palagonite-coated basalts. Geophys. Res. Lett. 28,2101–2104.

Johnson, J. R., R. Kirk, L. A. Soderblom, L. Gaddis, R. J. Reid, D. T. Britt,P. Smith, M. Lemmon, N. Thomas, J. F. Bell III, N. T. Bridges, R. Anderson,K. E. Herkenhoff, J. Maki, S. Murchie, A. Dummel, R. Jaumann, E. Hauber,and G. Arnold 1999. Preliminary results on photometric properties ofmaterials at the Sagan Memorial Station, Mars. J. Geophys. Res. 104, 8809–8830.

Johnson, J. R., T. Sucharski, and R. J. Reid 2001. Implementation of the Imagefor Mars Pathfinder Calibration Algorithms in ISIS. Lunar Planet. Sci. 32,2062 (abstract).

Kirk, R. L., E. Howington-Kraus, T. Hare, E. Dorrer, D. Cook, K. Becker,K. Thompson, B. Redding, J. Blue, D. Galuszka, E. M. Lee, L. R. Gaddis,J. R. Johnson, L. A. Soderblom, A. W. Ward, P. H. Smith, and D. T. Britt 1999.Digital photogrammetric analysis of the IMP camera images: Mapping theMars Pathfinder landing site in three dimensions. J. Geophys. Res. 104, 8869–8887.

McSween, H. Y., Jr. 1994. What we have learned about Mars from SNCmeteorites. Meteoritics 29, 757–779.

McSween, H. Y., Jr., and 18 colleagues 1999. Chemical, multispectral andtextural constraints on the composition and origin of rocks at the MarsPathfinder landing site. J. Geophys. Res. 104, 8679–8716.

Moore, H. J., D. B. Bickler, J. A. Crisp, H. J. Eisen, J. A. Gensler, A. F. C.Haldemann, J. R. Matijevic, L. K. Reid, and F. Pavlics 1999. Soil-like depositsobserved by Sojourner, the Pathfinder rover. J. Geophys. Res. 104, 8729–8746.

Morris, R. V., D. C. Golden, J. F. Bell III, and H. V. Lauer, Jr. 1995. Hematite,pyroxene, and phyllosilicates on Mars: Implications from oxidized impactmelt rocks from Manicougan Crater, Quebec, Canada. J. Geophys. Res. 100,5319–5329.

Morris, R. V., and 11 colleagues 2000. Mineralogy, composition and alterationof Mars Pathfinder rocks and soils: Evidence from multispectral, elemental,and magnetic data on terrestrial analogue, SNC meteorite, and Pathfindersamples. J. Geophys. Res. 105, 1757–1818.

Murchie, S. L., O. Barnouin-Jha, J. Johnson, J. F. Bell III, H. McSween, andR. V. Morris 2000. Diverse rock types at the Mars Pathfinder landing site.Lunar Planet. Sci. 31, 126 (abstract).

Murchie, S. L., O. Barnouin-Jha, J. R. Johnson, J. F. Bell III, and R. V. Morris2001. Spectral differences between shape classes of rocks at the MarsPathfinder Landing Site. Lunar Planet. Sci. 32, #1825 (abstract).

Reid, R. J., P. H. Smith, M. Lemmon, R. Tanner, M. Burkland, E. Wegryn,
J. Weinberg, R. Marcialis, D. T. Britt, N. Thomas, R. Kramm, A. Dummel,D. Crowe, B. J. Bos, J. F. Bell III, P. Rueffer, F. Gliem, J. R. Johnson,

H
J. N. Maki, K. E. Herkenhoff, and R. B. Singer 1999. Imager forMars Pathfinder (IMP) image calibration. J. Geophys. Res. 104, 8907–8926.

Rieder, R., H. Wanke, T. Economou, and A. Turkevich 1997. Determination ofthe chemical composition of Martian soil and rocks: The alpha proton X-rayspectrometer. J. Geophys. Res. 102, 4027–4044.

Smith, M. O., S. L. Ustin, J. B. Adams, and A. R. Gillespie 1990. Vegetationin deserts: 1. A regional measure of abundance from multispectral images.Remote Sens. Env. 31, 1–26.

Smith, M. O., D. A. Roberts, J. Hill, W. Mehl, B. Hosgood, J. Verdebout,G. Schmuck, C. Koechler, and J. B. Adams 1994. A new approach to determin-ing spectral abundances of mixtures in multispectral images. In Proceedings
of the IEEE Geoscience Remote Sensing Symposium IGARSS ’94, 2372–2374.

Smith, P. H., and colleagues 1997a. The imager for Mars Pathfinder experiment.J. Geophys. Res. 102, 4003–4025.

Smith, P. H., and colleagues 1997b. First results from the Pathfinder camera.Science 278, 1758–1765.

Sunshine, J. M., and C. M. Pieters 1993. Estimating modal abundances fromthe spectra of natural and laboratory pyroxene mixtures using the modifiedGaussian model. J. Geophys. Res. 98, 9075–9087.

Thomas, N., W. J. Markiewicz, R. M. Sablotny, M. W. Wuttke, H. U. Keller,J. R. Johnson, R. J. Reid, and P. H. Smith 1999. The color of the martian skyand its influence on the illumination of the Martian surface. J. Geophys. Res.104, 8795–8808.

Tomasko, M. G., L. R. Doose, M. Lemmon, P. H. Smith, and E. Wegryn 1999.
Properties of dust in the martian atmosphere from the Imager on MarsPathfinder. J. Geophys. Res. 104, 8987–9008.

low abundance materials at the mars pathfinder landing site: an

Documents