Multiscale and Multispectral Characterization of Mineralogy with the ExoMars 2022 Rover Remote Sensing Payload

In 2022, the European Space Agency and Roscosmos will launch the ExoMars rover, with the scientific objective to detect evidence of life within the Martian surface via the deployment of a 2 m drill. The ExoMars Pasteur payload contains several imaging and spectroscopic instruments key to this objective: the Panoramic Camera (PanCam), Infrared Spectrometer for ExoMars (ISEM), and Close‐UP Imager (CLUPI). These instruments are able to collect data at a variety of spatial (sub‐mm to decimeter) and spectral (3.3 to 120 nm) resolutions across the 440 to 3,300 nm wavelength range and collectively will form a picture of the geological and morphological characteristics of the surface terrain surrounding the rover. We deployed emulators of this instrument suite at terrestrial analog sites that formed in a range of aqueous environments to test their ability to detect and characterize science targets. We find that the emulator suite is able to effectively detect, characterize, and refine the compositions of multiple targets at working distances spanning from 2 to 18 m. We report on (a) the detection of hydrothermal alteration minerals including Fe‐smectites and gypsum from basaltic substrates, (b) the detection of late‐stage diagenetic gypsum veins embedded in exposures of sedimentary mudstone, (c) multispectral evidence of compositional differences detected from fossiliferous mudstones, and (d) approaches to cross‐referencing multi‐scale and multi‐resolution data. These findings aid in the development of data products and analysis toolkits in advance of the ExoMars rover mission.


Introduction
In 2022, the European Space Agency (ESA) and the Russian Space Agency (Roscosmos) will deliver the ExoMars rover Rosalind Franklin and surface platform to the surface of Mars, whose communications with Earth will be supported by the already operational ExoMars Trace Gas Orbiter (TGO). The ExoMars rover Pasteur payload (Barnes et al., 2006;Vago et al., 2017) contains a suite of nine instruments plus a subsurface drill, which will be used to address ExoMars' primary objective of detecting evidence of extinct life within subsurface deposits (McMahon et al., 2018;Vago et al., 2017). Among these are a number of imaging instruments, including the Panoramic Camera (PanCam)  and the CLose UP Imager (CLUPI) , with point hyperspectral data provided by the Infrared Spectrometer for ExoMars (ISEM) . The ExoMars rover will be the first mission to combine imagers covering three different spatial scales (including stereo; Barnes et al., 2018) in combination with a point infrared spectrometer, in situ at the Martian surface. This combination presents a huge range of potential data products with which to investigate the geology along the rover traverse.
Data from PanCam, ISEM, and CLUPI will augment existing, coarser-resolution orbital information (The ExoMars 2018 Landing Site Selection Working Group (LSSWG), 2014) with multiscale observations of surface morphology, color, texture, grain size, and compositional information to provide the environmental context for tactical and strategic planning operations and support the prioritization of drilling locations where pristine subsurface material can be accessed.
Ancient terrains record a history of aqueous activity and subsequent alteration. Noachian (>3.7 Ga) terrains in particular grant access to an epoch of deposits no longer available for study on Earth (Nimmo & Tanaka, 2005). As prolonged, aqueous Martian environments may have been habitable to microorganisms (Vago et al., 2017), we deployed instrument emulators in areas hosting minerals which had formed Table 1 Filter Wheel Specifications for AUPE3 and ExoMars PanCam (Cousins et al., 2012;Coates et al., 2017) (5) 670 (5) Note.
WAC filter bandpasses are given as full width at half maximum (FWHM) values. Units are nanometers.

Infrared Spectrometer for Mars (ISEM)
Mounted below PanCam is the Infrared Spectrometer for ExoMars (ISEM) . This instrument provides point spectra for a 1 • FOV within the larger PanCam FOV and the 4.88 • FOV of HRC. ISEM uses an acousto-optic tunable filter (AOTF) to cover a spectral range from 1,150 to 3,300 nm at variable resolution. It is projected that one ISEM spectrum will be captured per HRC image. Targeting of HRC and ISEM will be driven by the discovery of mineralogical and morphological features of interest observed in PanCam imagery. Combining all of these instruments allows detailed analysis of diagnostic mineral features in the VNIR to SWIR wavelengths. For the ISEM emulator, ISEM-E, we use a handheld field spectrometer (Spectral Evolution RS-3500) with a spectral range of 350-2,500 nm and a spectral resolution of 3 nm. When in contact with the outcrop, the ISEM-E FOV is approximately 1 cm. While ISEM will not share any overlapping spectral bands with PanCam, the overlap of ISEM-E and AUPE3 allows for spectral comparison between these two instruments.

Close-Up Imager (CLUPI)
Mounted on the drill of ExoMars 2022 is the CLose-UP Imager (CLUPI) . CLUPI is a high-resolution, color camera system designed to mimic what a geologist would observe in the field through a hand lens, providing images of minerals within their original context before they are powdered for geochemical analyses. Such high-resolution data will facilitate the search for morphological features that may indicate biosignatures on outcrops and drill cores and provide grainsize constraints for sedimentary material. When at a distance of 10 cm from a target, CLUPI provides 7 micron per pixel resolution with an FOV of 1.9 × 1.3 cm. CLUPI uses a Foveon X3 detector which captures all RGB color information at the same spatial location, so, unlike detectors which use a Bayer filter, the resulting image data do not require any interpolation. CLUPI color images will closely represent RGB color from PanCam's filter set and will additionally be calibrated using the CLUPI calibration target located near the drill box . The CLUPI emulator CLUPI-E is a Sigma SD15 Digital SLR which uses the same detector as the CLUPI instrument. The lens is a Sigma 100 mm macro lens which provides a representative field of view and magnification range.

Geological Settings
Oxia Planum is the selected landing site for the ExoMars rover (European Space Agency, 2018). The landing ellipse for this site contains phyllosilicate-rich layered region which is overlain by a deltaic sedimentary fan, indicative of the ponding of water post-dating the layered formation (Quantin-Nataf et al., 2018). Later stage volcanic activity has partially covered the layered clay deposits. The entire region has been heavily eroded since these units were emplaced (The ExoMars 2018 Landing Site Selection Working Group (LSSWG), 2014). As a consequence, ExoMars will likely encounter hydrated mineral terrains along its traverse and both sedimentary and volcanic deposits. The ability of the ExoMars emulator suite to characterize geological targets is explored using four terrestrial analog sites, each described in detail below. These field sites were selected as they encompass aspects of Noachian clay-bearing sites and late Noachian to early Hesperian alluvial deposits, as well as a variety of other geological features relevant to ExoMars preparation. For the purposes of instrument testing they also had a scant vegetation coverage, (b) well-characterized geology, and (c) access to mains power. They were not specifically chosen to be analogous to locations where subsurface life has been preserved, as PanCam, ISEM, and CLUPI are not specifically designed to identify biosignatures.

Námafjall, Iceland
Námafjall is a geothermal field in the central northern region of Iceland (65 • 38 ′ 26.5 ′ ′ N, 16 • 48 ′ 43.2 ′ ′ W) ( Figure 1a). We focus on Námafjall ridge, which has exposures of subglacially erupted basalts displaying various stages of weathering and hydrothermal alteration. The first ("Pillow") is extensively weathered, with oxidized soils surrounding its base, dark glass-rich quench rinds, and small (1-5 cm) secondary zeolite deposits dispersed throughout. The second outcrop ("Pillow_Alt") is also weathered and has additionally undergone extensive secondary alteration and displays crosscutting gypsum veins. These outcrops were previously imaged with an earlier iteration of AUPE3 in Harris et al. (2015) as A07_Pillow and A08_Veins, respectively. These deposits in combination with their alteration phases provide an example of low-temperature (<150 • C) neutral-acidic alteration of basaltic material. The light-toned veins are also analogous to calcium sulfate veins detected on Mars at Endeavour and Gale craters (e.g., Arvidson

Tjörnes, Iceland
Tjörnes is located in northern Iceland at Kaldakvísl (66 Figure 1a) where the Tjörnes sedimentary beds sit on a tertiary basalt formation. This section contains the Tapes layer, the oldest and lowest layer of the Tjörnes beds. The Tjörnes sediments here are formed of fossiliferous marine sediments (Einarsson, 1994). Fluvial and lacustrine beds, as well as brown coal, are interspersed, providing an example of deposits that formed in a transitional, shallow water depositional environment. In particular, these sediments have a basaltic source, making them geochemically comparable to Martian sediments (Cannon et al., 2015), in contrast to quartzofeldspathic sediments more typically found in terrestrial systems.

Torrey, USA
The Triassic Moenkopi formation in southern Utah, USA, is characterized by mudstone interbedded with siltstone, sandstone, and gypsum beds that represent tidal marginal marine and sabkha environments at the edge of the Early Triassic epeiric sea (Young & Chan, 2017). We deploy the ExoMars emulator suite at an exposure near 38 • 18 ′ 05 ′ ′ N, 111 • 29 ′ 11.7 ′ ′ W, just outside the towns of Torrey and Teasdale, UT ( Figure 1b). This site is included as a terrestrial analog as the formation is crosscut by light-toned gypsum veins analogous to the calcium sulfate veins detected on Mars at Endeavour and Gale craters (e.g., Arvidson et al., 2014;Grotzinger et al., 2014;Nachon et al., 2014;Squyres et al., 2012).

Hanksville, USA
Alluvial deposits here are preserved as inverted channels in the Brushy Basin and Salt Wash Members of the Jurassic Morrison Formation and the Ruby Ranch member of the Cretaceous Cedar Mountain Formation (Williams et al., 2011) in the Hanksville region (38 Figure 1b). The inverted channel imaged at this site provides an example of vertical stratigraphy able to be imaged from a rover perspective. The outcrop is weathered and is composed generally of red and green mudstones with a sandstone capping unit containing layers with multiple grain sizes. Some mineral veins and large conglomerates are visibly embedded in the lower mudstones. We use this site to illustrate the multiscale capabilities of the instrument emulator suite and to capture mudstones exhibiting different oxidation states.

Field Methodology
The emulator suite was deployed at each field site at a working distance of 2-18 m from each outcrop: At this distance the spatial resolution was 1.2-10.8 mm per pixel. A Macbeth ColorChecker © was placed in front of the outcrop when capturing the scene to enable radiometric calibration post-imaging. We assume both the MacBeth ColorChecker and the outcrop surface are Lambertian scatterers, and to ensure that both were imaged under the same illumination conditions, with the same geometry and at approximately the same time, the Macbeth ColorChecker was positioned such that it had approximately the same local incidence/emission angles as the targeted features. For Námafjall and Tjörnes weather conditions were overcast, resulting in uniform lighting conditions. For the Utah sites, skies were generally clear with bright sunlight; [1-(R532/((0.53*R500)+(0.47*R568)))] Identifies ferric minerals, particularly hematite, and is related to degree of oxidation however, passing clouds meant that lighting conditions were variable while AUPE3 data were being collected, resulting in RWAC spectra having to be omitted for the Hanksville site. We replaced this wavelength range with data from ISEM-E ground truth spectra. The timing of image collection depended on the direction that outcrops were facing; for example, east-facing outcrops were typically imaged in the morning while the sun was low and directly illuminating their faces to minimize overhead shadow. After outcrops had been imaged with AUPE3 and HRC-E, ISEM-E was used in handheld mode to collect point spectra of geologic features. In handheld mode, ISEM-E is held against each feature while its internal light source illuminates the surface, collecting and averaging 30 individual spectra to provide a final result, with an FOV around 1 cm. CLUPI-E was used to collect close-range, high-resolution images of each of these point spectra locations so that their exact sampling locations could be found in AUPE3 and HRC-E images. Finally, CLUPI-E was used to collect images for high-resolution mosaics to assist in visualization of the field site, and these were taken from the same working distance as AUPE3.

Processing Methodology
For each camera position in an AUPE3 mosaic two images were captured: one LWAC and one RWAC. All raw images were processed to R* reflectance defined by Reid et al. (1999) using the ExoMars Spectral Analysis Tool (ExoSpec) developed by Allender et al. (2018). ExoSpec performs flat-fielding, radiometric correction, and correction from radiance to R* reflectance using the in-scene Macbeth ColorChecker © . As the target and surface were imaged with approximately the same geometry and at approximately the same time, we assume they are imaged under the same illumination conditions and so do not consider corrections for diffuse versus direct illumination. However, work to compensate for differences in viewing geometry between a rover mounted horizontal calibration target and a variable geometry outcrop is underway and will be incorporated into a future version of ExoSpec. Following correction to R* reflectance, ExoSpec generates spectral parameter maps for both LWAC and RWAC (Table 2). Spectral parameter maps enable band depths and shoulders to be visualized, which can indicate broad mineralogical composition, abundance, or grain size fluctuations. These parameters have proven effective for the MER Pancams (Farrand et al., 2006(Farrand et al., , 2008(Farrand et al., , 2014(Farrand et al., , 2016Rice et al., 2010 and for orbital imagers like the Compact Reconnaissance Imager for Mars (CRISM) (Pelkey et al., 2007;Viviano-Beck et al., 2014). We experimented with RGB color combinations and different spectral parameter maps to determine which combination best highlights specific geologic features depending on the challenges posed by each field site. Note that none of the spectral parameters listed in Table 2 are diagnostic of specific minerals, rather they test for features consistent with absorptions that indicate the presence of some minerals. For example, the S610_671 parameter tests for the presence of the 650 nm hematite shoulder feature, but other ferric oxides such as goethite also have similar absorptions (610 nm) that light up this parameter. Similarly, some (but not all) hydrated minerals have a negative slope from 950 to 1,000 nm, but Fe-bearing minerals with broad absorptions at around 1,000 nm may have a negative slope as well.

10.1029/2019EA000692
To confirm the presence of different compositional units at each site we also performed decorrelation stretches to highlight RGB color differences (Gillespie et al., 1986), principal components analyses to locate unique spectral classes at each outcrop, and outlier detection with the Reed-Xiaoli Detector (RXD) algorithm (Reed & Yu, 1990) to locate pixels which differ significantly from the average global spectral response of an image in order to highlight potential science target locations ( Figure 4); see supporting information for the results for all sites. We examined these images alongside the spectral parameter maps for spatial agreement. All the above visualization steps were performed on AUPE3 data in order to assess the overall spectral diversity of each scene. Next, AUPE3 spectra were extracted from regions of interest (ROI) highlighted in these images and compared to the ISEM-E spectral data. Each ROI typically contained a minimum of 50-100 spatially contiguous pixels. As an example, assuming an AUPE3 working distance of 4 m, an ROI containing 50 pixels will cover a region of approximately 1.2 by 1.2 cm, which is comparable to the ISEM-E FOV of 1 cm. The mean signature from each ROI was considered its representative, and standard deviation was also computed in order to quantify variation within each ROI. Cross-referencing AUPE3 with ISEM-E datasets allowed a greater spectral range (438-2,500 nm) to be examined for diagnostic absorption features.
Bulk mineralogy for the Torrey and Hanksville sites was characterized by X-ray diffraction (XRD). The samples were disaggregated from rock (or sediment/powder) and ground with an agate mortar and pestle. To preserve the original grain size and mineral structure, approximately 20 g of material of each sample was ground and sifted through a 500 μm sieve. The randomly oriented fine powder was front-packed onto glass slides and analyzed at the Laboratory for Stable Isotope Science (The University of Western Ontario) using a high-brilliance Rigaku Rotaflex RU-200B series X-ray diffractometer, equipped with a rotating anode (Co K-source operated at 160 mA and 45 kV) and a graphite monochromator. Scans were performed from 2 to 82 • 2 at a step size of 0.02 • 2 . Analysis was performed using the QualX phase identification program for powder diffraction data (Altomare et al., 2008). XRD pattern peaks were compared to standard XRD mineral patterns from the POW_COD Crystallography Open Database (Downs & Hall-Wallace, 2003;Grazulis et al., 2009Grazulis et al., , 2012. Quantitative analyses ("S-Quant") were additionally performed within QualX via reference intensity ratios (RIR) based on the standard chosen from the mineral database.

Crosscutting Mineral Features
Outcrops at Iceland (Pillow, Altered Pillow, and Tjörnes) and Utah (Torrey) all feature either high-albedo crosscutting mineral veins (Pillow, Altered Pillow, and Torrey) or discrete carbonate mineral features (Tjörnes), which were used to investigate the utility of the ExoMars emulator suite in the detection and characterization of high-albedo, hydrated mineral deposits. Here, targets were analyzed with radiometrically corrected AUPE3 true color imaging, false color imaging-using S532_610, R671_438, and BD610 as RGB-and ISEM-E spectra ( Figure 2). Additional AUPE3 multispectral products investigated ( Figure 4) include the outlier detection using the Reed-Xiaoli Detector (RXD) algorithm, PCA false color combinations, the ratio between red and blue wavelengths (R671_438), and the negative slope between the 950 and 1,000 nm filters (S950_1000). Products for all sites are provided in supporting information.
The Pillow outcrop was imaged with AUPE3 at a working distance of 4 m, resulting in a spatial resolution of 2.4 mm per pixel. The RGB image (Figure 2a) reveals low-albedo glass-rich quench rinds, light-toned mineral deposits varying from 1 to 5 cm in width dispersed across the pillow basalt outcrop that do not conform to any pre-existing structural features, an oxidized weathered surface that dominates the majority of the outcrop, and light-toned soils. These are represented here by four spectral units. The ISEM-E spectrum of the light-toned mineral deposit (Figure 2e) has a doublet feature at 2,300-2,350 nm, which can be the result of Mg substituting for Fe in the Mg/Fe-OH combination (Viviano-Beck et al., 2014). The ISEM-E spectrum of the mineral deposit within the basalt matrix (red spectrum in Figure 2e) displays small absorption features around 2,200 and 2,300 nm. Possible interpretations of mineralogical composition of these spectra include nontronite, zeolite (analcime/heulandite), calcite, and palagonite, which are also all products of low-temperature (<150 • C) hydrothermal alteration of volcanic substrate and possess similar absorption features in the VNIR/SWIR (Cloutis et al., 2002;Warner & Farmer, 2010). Light-toned amygdaloidal mineral deposits can be visualized at a higher-spatial resolution with CLUPI-E and are approximately 1 mm in size given the spatial resolution (<5 mm per pixel) at an imaging distance of 40 cm (Figures 3b and 3c). Bulk mineralogy results at this site from Harris et al. (2015) suggest the presence of calcite, smectite, Fe-oxide, plagioclase, and stilbite, while VNIR/SWIR results confirm the presence of zeolite and nontronite. The Altered Pillow outcrop was imaged with AUPE3 at a distance of 3 m, resulting in a spatial resolution of 1.8 mm per pixel. Light-toned mineral veins crosscut the outcrop which has eroded to form light-toned soil in the image foreground ( Figure 2b); these are consistent with the false color green spectral unit seen in the Pillow outcrop ( Figure 2a). The S950_1000 summary parameter captures the edge of the hydration feature at around 1,000 nm (Rice, Bell III et al. 2013;Rice, Cloutis et al. 2013) for these features (supporting information). Spectra extracted from targeted regions in Figure 2b are shown in Figure 2e. The mineral vein has well-defined gypsum absorption features in its ISEM-E spectrum and possesses a negative 950-1,000 nm slope in PanCam multispectral data, which is in agreement with observations of gypsum by MER Pancam (Farrand et al., 2014(Farrand et al., , 2016. The altered light material is a mix of eroded gypsum and the oxidized substrate material, as shown by the presence of a broad band centered around 900 nm. This negative slope across the 950-1,000 nm range is not apparent in the Fe-rich gypsum signature due to the presence of a broad Fe 2+ 900 nm band obscuring this feature. Fine-scale structures and grain sizes can be visualized with HRC-E. An HRC-E image is shown in Figure 3a for the region indicated within Figure 2B1. Bulk mineralogy results for light-toned soils along Námafjall ridge from Harris et al. (2015) suggest the presence of gypsum, plagioclase, smectite, zeolite, hematite, and goethite; and VNIR/SWIR results confirm the presence of these (Figures 2e  and 9).
The Torrey outcrop was measured with AUPE3 at a working distance of 18 m, resulting in a spatial resolution of 10.8 mm per pixel. In the false color AUPE3 image (Figure 2c), green pixels highlight both the crosscutting veins and the dried red unit, which is likely a result of the gypsum veins weathering out to disperse a fine amount of material across the outcrop. The S950_1000 spectral parameter separates the veins from the other units ( Figure 2C1, parameter result shown in Figure 4C1). Here, only veins are picked out using white pixels, due to their negative slope across the 950-1,000 nm range (Figure 2e). For comparison, Figure 4c shows that the spectral parameter R671_438 can highlight light-toned deposits but cannot discriminate hydrated veins from among these. CLUPI-E images depicting the Torrey geological units are shown in Figures 3d and 3e. From these it is clear that the veins are crosscutting secondary features, emplaced after the formation of the red and green mudstone units. The SWIR spectra of the high-albedo veins (Figure 2e) are consistent with that of gypsum (Young & Chan, 2017) and the 950-1,000 nm VNIR slope is truly indicative of a sharp absorption feature around 1,000 nm in the ISEM-E ground truth spectrum. The SWIR spectrum of the dry red coating material (Figure 2e) is also consistent with gypsum, showing that the mineral veins are weathering out and dispersing this material. Bulk mineralogy results for this site confirm the presence of gypsum, as well as calcite, quartz, illite, and smectite within the mudstones (see supporting information).
Finally, the cliff exposure at Tjörnes (Figure 2d) was measured with AUPE3 at a working distance of approximately 9 m, resulting in a spatial resolution of approximately 5.4 mm per pixel. Fractures are visible in places within the AUPE image, and lower strata exhibit a clear dip to one side. As with the Pillow and Altered Pillow outcrops (Figures 2a and 2b), the false color spectral parameter combination highlights the calcite layers in green and furthermore enables their fine banding patterns to be distinguished within the surrounding layers. CLUPI-E images depicting Tjörnes geological features at a higher-spatial resolution are shown in Figures 3f and 3g.

Phyllosilicates and Mudstone
Within the Pillow outcrop, ISEM-E spectra refine the composition of surface materials to indicate the presence of Fe-smectite, in this case nontronite (Harris et al., 2015). Nontronite possesses non-diagnostic absorption features in the VNIR at around 450 nm (shoulder), 640 nm (absorption), and 940 nm (absorption), which are captured by AUPE3 (Figure 2e). In the associated SWIR spectra, absorption features occur at around 1,420, 1,910, 2,208, 2,290 nm, and at 2,410 and 2,510 nm (McKeown et al., 2018). These features  are visible in ISEM-E spectra that were collected across the whole outcrop (Figure 9), suggesting pervasive alteration of this basalt. These VNIR and SWIR spectral features were also identified in AUPE3 and ISEM-E spectra for the basaltic substrate at the nearby Altered Pillow outcrop. Within the cliff exposure at Tjörnes, multispectral AUPE3 imaging delineated the darker and lighter strata (Figures 2d and 4b) beyond a simple difference in albedo: The darker sedimentary layers correspond to a deeper absorption at 610 nm, whereas the lighter-toned layers have a steeper slope between 438 and 610 nm (Figure 2e). HRC-E imaging results in images with 0.11 mm per pixel resolution, versus the 5.4 mm per pixel resolution provided by AUPE3 at this working distance. This resolution enables coarse-sand grain sizes to be resolvable (Figures 3f and 3g) and reveals fine-scale calcite deposits and sedimentary structures.
At the Torrey site ( Figure 9) the red mudstone possesses weak hydration features, while those for the green mudstone are much stronger: Absorption features at around 2,200 and 2,300 nm suggest that this unit contains Mg-smectite. The mineral vein material is identified as gypsum (using the triplet absorption from 1,400 to 1,500 nm and absorption features at 1,750, 1,900, and 2,200 nm), while the dry red material possesses very similar, yet weaker, spectral features, supporting the idea that it is the result of a gypsum weathering process.
To assess the relationship between AUPE3 spectral parameters and ISEM-E hydration bands scatter plots were created using all ISEM-E spectral ROIs available in this study (Figure 5c). ISEM-E ROIs were not plotted if they possessed no absorption features at the plotted wavelengths. It is clear that all measurements possess a strong absorption feature at 1,900 nm regardless of the S950_1000 value, confirming that the S950_1000 spectral parameter alone is not a sufficient indicator of hydration state because most hydrated minerals do not exhibit their H 2 O and/or OH absorptions in the correct location to produce this negative feature (Rice et al., 2010). There is a strong positive correlation between spectral parameters BD1900 and BD2400 (Figure 5b). The presence of absorption features at 1,900 and 2,400 nm can be indicative of zeolites, hydrated sulfates, and Al-phyllosilicates (Viviano-Beck et al., 2014). The shifting position of the 1,900 nm feature was also plotted with respect to its depth (Figure 5d). From this plot several data clusters are visible which correspond to their composition. The 1,900 nm versus 2,200 nm band depth effectively discriminates all targets (Figure 5a). The mudstones at Torrey stand apart from the general data cluster, displaying smaller BD1900 values but high BD2200 values.

Oxidation State
On top of, and surrounding, the Pillow outcrop are light-toned soils. These features are easily visible in both true and false color images of this outcrop in Figure 2a, where S532_610, S671_438, and BD610 are assigned as RGB, respectively. Using the same color mapping at Torrey, the image in Figure 2c is able to differentiate clearly between the red and green mudstone units and even shows detail in the fragmented materials in the sloping foreground.  Figure 6 illustrates that AUPE3 filter combinations can be used to refine oxidation state. Using RWAC filters at 1,000, 850, and 740 nm Fe 3+ -bearing minerals such as hematite can be identified using the presence of an absorption feature at around 900 nm. This filter combination is based on MSL Mastcam's 1,012, 867, and 751 nm false color mosaic of Vera Rubin ("Hematite") Ridge (NASA, 2017).
The Hanksville outcrop was measured with AUPE3 at a working distance of approximately 4 m, resulting in a spatial resolution of approximately 2.4 mm per pixel (Figure 7a). The spectral parameter mosaic composed of parameters BD610, BD532, and S440_610 as RGB, respectively, highlights spectral diversity in the scene with respect to Fe-bearing minerals and highlights veins exposed within the outcrop. These features can be seen in more detail in Figure 8a-d.

Multiscale Utility
At Hanksville, the panchromatic HRC-E mosaic in Figure 8c refines the resolution imaged with AUPE3 to 0.37 mm per pixel at 4 m working distance and 2.8 mm per pixel at 30 m for distant observations (Figure 7c). CLUPI-E context images of these regions were also taken with a spatial resolution of 0.23 mm per pixel at 2 m working distance ( Figure 8e) and 0.02 mm per pixel at approximately 20 cm working distance (Figure 8f). This range of resolutions allows potential strategic science targets at multiple distances to be investigated.
Combined AUPE3 RGB data, spectral parameter maps, and HRC-E textural and grain size information can define spectral and stratigraphic units (Figure 8d). An ISEM-E spectrum for each unit is displayed in Figure 8g. We also show bulk mineralogy (XRD) results for each of these units in Table 3. All units contain evidence for smectites, as indicated by their VNIR/SWIR spectral signatures in Figure 8g. The green coloring in green mudstone and smectite layers is produced by chlorite. Above the orange contact layer in Figure 8d, Table 3 shows all units contain evidence for gypsum.

Discussion
The ExoMars emulator suite was deployed at four field sites across Iceland and the United States and captured a range of geological information at multiple spectral and spatial resolutions. Data from this suite allow interpretation of surface outcrop characteristics like refined composition, morphology, texture, grain size, weathering and oxidation states, hydration state, size, and color. Furthermore, science targets, which may not be obvious in typical color images at a single scale, are able to be observed to a sub-cm level with this instrument combination.

Utility of AUPE3 Products Across Sites
The ExoMars rover PanCam, ISEM, and CLUPI emulator instruments were able to detect and confirm manually identified science targets at each site, as well as discover and refine the composition of others. Figure 9  shows AUPE3 and ISEM-E spectra of the science targets identified at each site; their spectral diversity widely varies depending on the type of aqueous environment. The spectra for the subglacially erupted site Pillow all exhibit features of the Fe-smectite nontronite, with varying albedos and oxidization states. Altered Pillow spectra exhibit features common to dark basalt and gypsum, as well as mixtures of varying oxidization state. Tjörnes spectra are flatter in the VNIR due to the low-albedo, gray material forming the Tapes beds; Note. Cell colors correspond to geological unit in Figure 8d. however, all possess hydration features at 1,400 and 1,900 nm in the SWIR. Here the dark fine-grained material spectrum possesses a deeper OH band around 1,400 nm, and a more defined 1,700 nm H 2 O band than the other targets. The Torrey site spectra are very distinctively split between the mudstones and their secondary gypsiferous material, all of which would be suitable targets for the science goals of the ExoMars rover (Vago et al., 2017). The Hanksville inverted channel spectra all exhibit hydration features consistent with Al-smectite, which is confirmed in the XRD results from Table 3. soils likely composed of nontronite. These units appear spectrally similar using this band combination and have different mineralogical compositions in the SWIR; however, the spectral parameter maps can only be used to highlight units with similar spectral characteristics at particular wavelengths in the VNIR and represent only broad compositional characteristics. A suite of parameter maps is useful to fully explore site diversity and can be used to guide ISEM-E pointing, expanding the wavelength range of AUPE3 into the SWIR and enabling diagnostic spectral features to be examined.
The Torrey gypsum veins and veins observed on Mars by Opportunity and Curiosity at Endeavour and Gale craters are late-stage diagenetic features that have experienced little alteration since emplacement (Young & Chan, 2017). The absorption feature from 0.934 to 1.009 microns observed by the Opportunity Pancam showed a marked downturn in reflectance when gypsum veins were detected due to the H 2 O overtone from hydrated calcium sulfate (Farrand et al., 2016). We also observe this downturn with the Torrey gypsum veins in ISEM-E ground truth and AUPE3 data. Late-stage mineral veins can be distinguished across multiple sites with the S950_1000 parameter ( Figure 4C1) which shows the degree of positive or negative slope across 950 to 1,000 nm. This spectral range can be indicative of hydrated minerals as it can be used to infer the presence of the H 2 O combination band and/or the 3 v OH overtone feature at around 1,000 nm (Rice, Bell III et al. 2013;Rice, Cloutis et al. 2013). However, this feature can occur at different wavelength positions for different minerals; thus, only certain hydrated minerals can be detected, and gypsum is one such mineral. We show in (Figure 4c) that this parameter is also useful to differentiate between features with a similarly high albedo.

Utility of AUPE, ISEM, and CLUPI for Strategic Planning
At each field site we have shown that the ExoMars emulator suite is capable of multiscale characterization of tactical targets of interest that may be in the vicinity of the rover. More distant, strategic targets are also able to be investigated which inform longer-term planning and operations. Finer-resolution imagery of these targets can then be requested for targeting with HRC. As an example, the vertical HRC-E mosaic inset in Figure 8c illustrates the HRC resolution available at close range: Rock textures and individual eroding clasts and grains can be seen on the outcrop face, while the HRC-E image inset in Figure 7c illustrates the detail that can be made out for distant targets. There is a tradeoff between higher-resolution imaging and multispectral imaging for long range targets-AUPE3 pixels at this distance are 20 mm per pixel yet enable spectral data to be obtained (for sufficiently large, resolvable objects), while HRC-E allows finer details to be resolved.
The ISEM targeting pattern may also be dictated by information obtained from HRC images-whether a long horizontal scan is needed to capture spectral variability across a wide feature like a ridge or a gridded scan is needed for a compact feature, for example, to distinguish a float rock from the surface it lies on. If the target is too distant and, therefore, the ISEM FOV too large, the strategic target can be flagged and prioritized for future investigation once the working distance has been reduced.
In this way, the incremental gathering of multiscale and multispectral data by this instrument suite allows a wealth of information about the terrain surrounding the rover to be collected which can in turn provide insight into the subsurface material.

Data Limitations and False Spectral Signatures
For this work, as ISEM-E is able to capture information from 300 to 2,500 nm, AUPE3 spectra were ground-truthed to ISEM-E spectra so that spectral features could be compared regardless of their perceived albedo. When comparing AUPE3 and ISEM-E spectra, a consistent offset exists between datasets due to differences in viewing geometry between the two systems, as discussed in Harris et al. (2015). In terms of spectral shape, the two datasets are consistent with each other (see supporting information). However, once bidirectional reflectance distribution functions (BRDF) are measured for a range of illumination angles this offset can be quantitatively mitigated using a lookup table applied to AUPE3 data.
Due to concentric artifacts in AUPE3s RWAC filters caused by stray light (Harris et al., 2015) a spurious band depth feature was observed at around 900 nm in AUPE3 data which was not present in ISEM-E data. Regions of interest selected off-center did not appear to be as affected and matched well to the ISEM-E ground truth. The stray light causing these concentric artifacts and spurious absorption feature is confined to the design of the AUPE3 optical system and will not be present in the PanCam system.
These concentric artifacts are also very apparent in the S950_1000 slope parameter calculation as they affect both filters at 950 and 1,000 nm to some degree. Stray light affects each RWAC filter differently, which can 10.1029/2019EA000692 introduce incorrect slope calculations as a result of differing reflectance levels. As an example, in the VNIR at Pillow, all spectra, apart from the dark quench rind, show evidence of a negative slope from 950 to 1,000 nm which can be indicative of the presence of the H 2 O combination band and/or the 3 v OH overtone feature at around 1,000 nm (Rice, Bell III et al. 2013;Rice, Cloutis et al. 2013). Free from the effects of stray light, this VNIR data may be used to infer whether an outcrop has been aqueously altered. However, comparison of AUPE3 data with ISEM-E spectra collected as ground truth over the AUPE3 wavelength range suggests that these slopes were false positives: The red spectrum (representing a mineral vein) in Figure 2e was the only ISEM-E spectrum which truly exhibited an absorption feature across this range. In agreement with Rice et al. (2010), we find that the 950-1,000 nm slope parameter alone cannot be used as an effective hydration indicator. This is because (a) the effect of stray light on AUPE3 data means that the slope parameter cannot be considered reliable and (b) most hydrated minerals do not exhibit their H 2 O and/or OH absorptions within this wavelength range. This is illustrated in Figure 5c, in which many sites that exhibit a distinct H 2 O absorption feature at 1,900 nm plot with a positive S950_1000 slope value.
It is worth noting that the outcrops examined lacked the dust cover typically present globally on Mars. Dust can lessen the spectral variability observed in an image; for example, hydrated outcrops may be covered with anhydrous dust and soil which obscures their spectral signatures. Conversely, false detections of hydrated signatures can result from dust-covered surfaces when viewed at high-emission angles (Rice et al., 2010).
Corrections for dust accumulation on the calibration target also need to be modeled and applied Kinch et al., 2007Kinch et al., , 2015 to ensure correct calibration of spectral data to R* reflectance.

Conclusions and Future Work
Individually, instruments in the emulator suite offer powerful analysis abilities, but the cross-referencing of their data has allowed for a much richer scope of information to be collected, from which tactical and strategic operational decisions can be made. Through the use of instrument emulators, we have demonstrated that the ExoMars imaging instrument suite will be able to effectively detect and characterize science targets of interest from a range of aqueous environments at a variety of spectral and spatial resolutions. This surface data can be used to augment the knowledge of geologists and instrument analysts and facilitate the prioritization of drill targets which will allow access to pristine subsurface deposits which may contain evidence of life, supporting the astrobiological goals of the ExoMars mission.
Future work will focus on improving the detection and characterization of science targets through the combination of imaging data with structural geological data provided through 3D visualization of the environment using digital elevation models (DEM) created from AUPE3 LWAC and RWAC images and visualized in Pro3D (Barnes et al., 2018). PanCam filter subsets will also be developed that can be used in specific environments to highlight specific spectral trends and assist the PanCam team in making pointing recommendations for other instruments within a tactical, sol-to-sol timeframe.