CRISM

Icarus 419 (2024) 115612
Contents lists available at ScienceDirect
Icarus
journal homepage: www.elsevier.com/locate/icarus
The CRISM investigation in Mars orbit: Overview, history, and delivered

data products
Frank P. Seelos a, Kimberly D. Seelos a, Scott L. Murchie a, *, M. Alexandra Matiella Novak a,
Christopher D. Hash b, M. Frank Morgan a, Raymond E. Arvidson c, John Aiello a,
Jean-Pierre Bibring d, Janice L. Bishop e, John D. Boldt a, Ariana R. Boyd a,
Debra L. Buczkowski a, Patrick Y. Chen a, R. Todd Clancy f, Bethany L. Ehlmann g, p,
Katelyn Frizzell h, Katie M. Hancock a, John R. Hayes a, Kevin J. Heffernan a, David C. Humm a, i,
Yuki Itoh a, Maggie Ju a, Mark C. Kochte a, Erick Malaret b, J. Andrew McGovern j,
Patrick McGuire k, Nishant L. Mehta a, Eleanor L. Moreland l, John F. Mustard m, A. Hari Nair a,
Jorge I. Núñez a, Joseph A. O’Sullivan n, Liam L. Packer a, Ryan T. Poffenbarger a,
Francois Poulet d, Giuseppe Romeo a, Andrew G. Santo a, Michael D. Smith o, David C. Stephens a,
Anthony D. Toigo a, Christina E. Viviano a, Michael J. Wolff f
a
Johns Hopkins University Applied Physics Laboratory, Laurel, MD, USA
b
Applied Coherent Technology, Herndon, VA, USA
c
Department of Earth and Planetary Sciences, Washington University in Saint Louis, Saint Louis, MO, USA
d
Institut d’Astrophysique Spatiale, Université Paris Sud, Orsay, France
e
SETI Institute/NASA-ARC, Mountain View, CA, USA
f
Space Science Institute, Boulder, CO, USA.
g
Division of Geological and Planetary Sciences, California Institute of Technology, Pasadena, CA, USA
h
Department of Earth and Planetary Sciences, Rutgers University, Piscataway, NJ, USA
i
Space Instrument Calibration Consulting, Annapolis, MD, USA
j
Aerospace Corporation, 14745 Lee Rd., Chantilly, VA, USA
k
Department of Meteorology and National Centre for Atmospheric Science (NCAS), University of Reading, Reading, England, United Kingdom
l
Department of Earth, Environmental, and Planetary Sciences, Rice University, Houston, TX, USA
m
Department of Earth, Environmental and Planetary Sciences, Brown University, Providence, RI, USA
n
Department of Electrical Engineering, Washington University, Saint Louis, MO, USA
o
NASA Goddard Space Flight Center, Greenbelt, MD, USA
p
Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA
A R T I C L E I N F O A B S T R A C T
Keywords: The Compact Reconnaissance Imaging Spectrometer for Mars (CRISM) on the Mars Reconnaissance Orbiter
Mars, atmosphere (MRO) collected hyperspectral images of the Martian surface and atmosphere from September 27, 2006, through
Mars, surface May 7, 2022. Over that time, nearly twenty scientific investigations were completed, most of which arose as a
Mineralogy
result of the findings from previous investigations. Two review papers published in 2009 (Murchie et al., 2009a,
Instrumentation
b) described the initial two-year investigation during MRO’s Primary Science Phase, its key findings, and the
CRISM data products that were developed and released to the community through that time. Here we describe
the conduct and evolution of the CRISM investigation since then, which includes MRO’s Extended Science Phase
and first five Extended Missions. We document the physical changes in the instrument as it aged, including
capabilities that were lost as well as new modes of operation not initially envisioned; the new science questions
that were investigated and their key findings; anatomy of the extensive collection of data products that have been
released to the Planetary Data System; the “final” radiometric calibration; high-order derived products produced
* Correspondence author at: MS 200-W230, Johns Hopkins Applied Physics Laboratory, 11100 Johns Hopkins Rd., Laurel, MD 20723, USA.
E-mail address: [email protected] (S.L. Murchie).
https://doi.org/10.1016/j.icarus.2023.115612
Received 23 February 2023; Received in revised form 13 April 2023; Accepted 2 May 2023
Available online 3 May 2023
0019-1035/© 2023 The Authors. Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
F.P. Seelos et al. Icarus 419 (2024) 115612
from high-resolution targeted observations and global mapping campaigns; and data processing and analysis
tools which have been developed and released by the CRISM team.
1. Overview of CRISM design and operation concepts ice aerosols modulate the low spectral frequency signatures of primary
igneous minerals (pyroxene, olivine, and Fe-bearing plagioclase), and
In the early 2000’s, a series of missions had been sent to Mars to aerosol opacity must be accounted for in modeling the spectra of pri
recover science lost by the failure of Mars Observer. In terms of spec mary minerals.
troscopy and mapping of surface composition, the Thermal Emission CRISM’s design and operation emphasized spatial resolution first
Spectrometer (TES) on Mars Global Surveyor was beginning its several among requirements, to maximize ability to resolve stratigraphic re
kilometers-per-pixel, 6–50 μm thermal infrared hyperspectral global lations and small outcrops. The native resolution of high-resolution
mapping (Christensen et al., 2001). The Thermal Imaging System (“targeted”) observations was designed to be 18 m/pixel, ≥100%
(THEMIS) on Mars Odyssey (Christensen et al., 2004) would comple margin on MRO requirements. Targeted observation spatial resolution
ment TES results by imaging in selected 7–14 μm bands at pixel scales as was further enhanced during operation through introduction of a
small as 100 m. The Observatoire pour la Minéralogie, l’Eau, les Glaces “super-resolution” observing mode (see Section 2.4.1). CRISM’s
et l’Activité (OMEGA, on the European Space Agency’s Mars Express contextual global mapping, at 90–180 m/pixel, provides several times
mission) would acquire images of Mars at hundreds of meters to several higher spatial resolution than previous spectral data sets.
kilometers per pixel at 0.4–5.0 μm (Bibring et al., 2004). OMEGA’s Three design choices enabled CRISM simultaneously to meet the
sensitivity to fine-grained, hydrated and hydroxylated secondary min requirements of high spatial and spectral resolution and high signal-to-
erals formed by past aqueous activity was in the line with NASA’s theme noise ratio (SNR), while providing an ability to correct for atmospheric
for Mars exploration to “follow the water”, and complemented TES’s and gases and aerosols (Silverglate and Fort, 2004), the benefits of which
THEMIS’s sensitivity to the primary silicate minerals that formed unfortunately necessitated using lifetime-limiting components.
igneous rocks and more crystalline secondary minerals. (1) Two-dimensional spatial images were constructed by collecting
With that context, a small workshop was hosted at the Lunar and one line at a time of a two-dimensional spatial image, using separate
Planetary Institute (LPI) entitled “Mars Spectroscopy: What Next?” visible/near-infrared (VNIR, 0.36–1.05 μm) and infrared (IR, 1.01–3.92
(Kirkland et al., 1999). The consensus recommendation was that the μm) detectors in either a gimbaled or nadir-pointed mode, and each
next Mars orbital spectrometer should be visible-infrared (Vis-IR) and pixel in a frame contained the full wavelength sampling. This resulted in
targeted at specific regions of interest, with tens-of-meters per pixel a spectral image cube, with dimensions comprising the x, y, and wave
scale, and hyperspectral wavelength coverage from visible wavelengths length domains. For the “targeted” observations, image motion
(beginning at 0.4 μm) to as far into the infrared as possible to marry compensation was used to attain longer integrations and higher SNR.
complementary capabilities of different wavelength ranges. Following The entire Optical Sensor Unit (OSU) was gimballed over up to a 120◦
the loss of both Mars Climate Orbiter and Mars Polar Lander, one of the range: a targeted observation was acquired by tracking the center-point
new missions defined to recover scientific leadership in Mars explora of the target, and superimposing a slow scan relative to the center point
tion was the Mars Reconnaissance Orbiter (MRO) (Zurek and Smrekar, to sweep out the spatial extent. The gimbal’s range of motion and on
2007), which was tasked with finding and characterizing sites where board data editing allowed acquisition of reduced spatial resolution
ancient surface water had persisted. MRO’s top level science re image segments at higher, bracketing emergence angles, forming an
quirements focused on complementary, synergistic imaging between “emission phase function”, or EPF), and enabling retrieval and separa
payload elements. Requirements for one of the imaging investigations, tion of atmospheric aerosol from surface reflectance (Murchie et al.,
Vis-IR hyperspectral imaging, closely followed the LPI workshop’s rec 2007). Note that the term “EPF” may refer to either an observing mode
ommendations: image dozens of sites at ≤40 m/pixel, hyperspectrally at (i.e., class of data delivered to the PDS) or the set of incoming and
0.4–3.6 μm, at ≤10 nm/channel). The Compact Reconnaissance Imaging outgoing scans acquired as part of other, gimbaled observing modes.
Spectrometer for Mars (CRISM) was competitively selected to meet these Mapping images in contrast were acquired with the gimbal pointing
requirements (Murchie et al., 2007). CRISM high-resolution observa fixed at nadir, resulting in greater ground coverage but reduced spatial
tions typically were accompanied by tens-of-centimeters per pixel im and spectral resolution.
ages from the High-Resolution Imaging Science Experiment (McEwen 2) The IR detector had to be cooler than ambient temperatures to
et al., 2007) and a 6 m/pixel Context Imager (Malin et al., 2007). These enable sufficiently low dark current noise. MRO’s nominal ~3 PM local
coordinated observations would provide an unprecedented basis for solar time (LST) orbit, while nearly circular and relatively stable ther
studying Mars geology at the outcrop scale. mally, made a passive cooling approach problematic to implement.
The CRISM team, led by the Johns Hopkins Applied Physics Labo Instead, CRISM used three redundant, mechanical Ricor cryocoolers
ratory (APL), based its investigation on both the MRO requirements for a connected in parallel to the IR detector via diode heat pipes. Because of
Vis-IR hyperspectral imager and orbital remote sensing objectives their mechanical and thermal separation, however, when all three cry
defined by the Mars Exploration Program Analysis Group (MEPAG). ocooler lifetimes were expired, VNIR observing would continue
These also included Vis-IR global mapping for placing the high- unaffected.
resolution observations in context, monitoring of polar volatile (3) A key question about Martian alteration mineralogy was whether
cycling, and characterization of atmospheric trace gases and aerosols carbonate minerals were present and if so where and in what abun
(Table 1). Murchie et al. (2007) provided a detailed description of in dances. The strongest fundamental doublet absorptions are at 3.3–3.5
strument design and operation and how CRISM was responsive to these and 3.7–3.9 μm, the latter beyond the minimum wavelength range
objectives. required by the MRO mission. CRISM’s design adopted a long-
The requirements for observing Mars’ surface overlapped with ob wavelength cutoff of ~4 μm to capture these features. The extended
jectives for observing Mars’ atmosphere. Secondary minerals on Mars’ wavelength coverage enabled detection and mapping of carbonates
surface typically have narrow, diagnostic mineral absorptions at 1.4-, (Ehlmann et al., 2008), but also required a lower detector temperature
1.9-, 2.1-, and 2.2–2.5 μm that overlap with atmospheric gas absorp to control dark current, with the effect of causing more wear on the
tions. Thus, an observing strategy to accurately measure secondary coolers.
mineral signatures must also measure signatures of atmospheric gases, Over the course of the investigation CRISM met all of its objectives,
particularly CO2, CO, and H2O. Similarly, atmospheric dust and water- discovering and mapping the occurrences of dozens of minerals.
2
Table 1 observing modes were modified or retired, and new ones were intro
CRISM investigation objectives and implementation. duced in response to unexpected discoveries and new objectives. Cali
Objective Implementation / Observation / Planning bration algorithms were improved, and new data products were
Measurement Strategy Design introduced that minimized users’ needs for specialized knowledge about
Target observations using the instrument.
previous geologic studies Murchie et al. (2009a) described CRISM’s science investigations
and results from the
Target list assembled
during MRO’s Primary Science Phase (PSP) and their key results. Here,
Thermal Emission we summarize evolution of the CRISM investigation from that time
from previous studies
Spectrometer (TES),
Thermal Emission Imaging
Screened / added to through the conclusion of its orbital measurements in May 2022. We
using map products provide detailed descriptions of types of observations that were
System (THEMIS), Mars
Orbiter Camera (MOC), collected, the observing campaigns that used them, generation of
etc. higher-order derived data products, and the CRISM team-provided tools
Use Mars Express OMEGA
Find new targets of Derived products from that are available to the community for analysis of the data products.
data to find new targets
interest: (aqueous
lacking morphologic or
OMEGA included in Section 2 summarizes instrument design and how its use evolved over
deposits, crustal targeting time in response to new discoveries, and as the instrument aged and the
thermal IR signatures
composition) and
provide geologic
72-channel, 180-m/pixel gimbal and cryocooling systems degraded. Section 3 describes the new
VNIR+IR multispectral science investigations introduced over the Extended Science Phase and
context
mapping
Find targets below
262-channel, 180 m/
multiple Extended Missions, and provides an overview of the data
OMEGA’s resolution using products generated by the CRISM Science Operations Center (SOC).
pixel VNIR+IR
near-global mapping at
hyperspectral mapping Section 4 describes the “final” radiometric calibrations and noise
key wavelengths
(including IR through
107-channel, ~180 m/ remediation applied to the data. Section 5 describes advanced products
pixel VNIR hyperspectral
2017); Provide context for that have been derived from targeted observations, both in sensor space
mapping
targeted observations
92-channel, ~90 m/pixel and with map projection, to facilitate analysis of CRISM data by the
VNIR hyperspectral planetary science community. Section 6 describes assembly, correction
mapping for time-variable atmospheric attenuation of surface reflectance, and
Observe emission phase ±60◦ or − 60◦ to +30◦ normalization of data acquired as part of the original global, 72-channel
function (EPF) at each EPF inherent to full and
multispectral mapping campaign. Finally, Section 7 briefly describes
targeted observation to half resolution targeted
quantify atmospheric observations through analytical tools developed since the PSP and additional data products
effects (through 2012) 2012 that are in development as of early 2023.
High time-resolution
Regularly acquire global
(atmospheric monitoring
grids of EPFs to monitor 2. History of CRISM in Mars Orbit
Separate the surface and campaign)
seasonal variations in
atmosphere High spatial-resolution
surface and atmospheric This section briefly summarizes the basic design of the CRISM in
Provide information (seasonal change
properties (through 2012)
on spatial/ seasonal campaign) strument and the technical basis of different observing modes, including
variations in dust, ice Acquire limb scans to the history of using different observing modes as the instrument aged
2 orbits of limb scans at
aerosols, H2O, CO2, CO measure vertical
representative longitudes and crycooling and gimbaling capabilities gradually declined.
variations in gases,
every 2 months
aerosols (2009–2017)
Key areas targeted with 2.1. Instrument hardware overview
Sample compositional full and half resolution
layering and seasonal Monitor seasonal cap
The CRISM system, design, and function are illustrated in Fig. 1 (a-d)
change of polar ices with multispectral
observations and detailed at length in Murchie et al. (2007). Briefly, CRISM consisted
Full and half resolution of three boxes: the OSU which included the optics, gimbal, focal planes,
targeted observations cryocoolers and radiators, and focal plane electronics; the Gimbal Motor
Measure thousands of provide coverage at 6.55 Electronics (GME), which commanded and powered the gimbal motor
targets at full spectral nm/channel, 18–36 m/
resolution and high spatial pixel; spatially
and encoder and analyzed data from the encoder in a feedback loop; and
resolution oversampled the Data Processing Unit (DPU), which accepted and processed com
observations rendered at mands from the spacecraft and accepted and processed data from the
12 m/pixel OSU and communicated it back to the spacecraft. The CRISM OSU
Measure surface targets
Measure key regions of the Hyperspectral mapping
with high spatial and contained a telescope which imaged a scene onto a slit which defined a
surface at key wavelengths provides coverage of
spectral resolutions
at higher spectral large regions of the single detector row, after which a beamsplitter directed VNIR and IR
and high SNR
resolutions than planet at 90 or 180 m/ light each into a separate spectrometer, within which a grating and
multispectral mapping pixel optics reimaged light onto a detector. Different wavelengths were
Radiometric calibration dispersed in the detector row direction with multiple grating orders (λ,
Conduct inflight
using integrating sphere
calibration of background
Background calibrations
λ/2, λ/3, etc.) superimposed. The higher-order light was blocked by an
and responsivity to order sorting filter mounted to the detector, two on the VNIR focal plane
using shutter are
provide radiometric
integrated with each covering 362–1053 nm at 6.55 nm/channel, and three on an IR focal
accuracy
observation plane covering 1002–3920 nm at 6.55 nm/channel. The IR focal plane
was cooled during operation to ≤125 K by one of three cryocoolers
(selectable by the DPU). Each of the detectors’ field-of-view (FOV) was
Beginning on September 27, 2006, and ending May 7, 2022, over 33,000
approximately 605 61.5-μrad pixels wide. On each side of the VNIR FOV
targeted observations were acquired, 86% of the surface of Mars was
and one side of the IR FOV, there are detector elements not illuminated
covered in full-wavelength VNIR+IR mapping modes, and > 99% of the
through the spectrometer slit, that measured the magnitude of instru
surface of the planet was mapped hyperspectrally at VNIR wavelengths.
ment scattered light in the spectrometer cavity as a function of wave
In addition, many thousands of observations characterized temporal,
length simultaneously with each scene measurement. The cross-track
spatial, and vertical variations in atmospheric trace gas abundances and
(spatial) shape of scattered light was estimated from the short-
dust and ice opacities. As the instrument aged, many of the original
wavelength end of each detector where there was effectively no in-
3
Fig. 1. a. Photographs of the CRISM OSU, DPU, GME.

b. CRISM block diagram.
c. CRISM Optical Sensor Unit (OSU) configuration.
d. CRISM optical design.
focus dispersed scene light, only scatter from the grating. A shutter focal plane had a dedicated electronics board that digitized the video
periodically blocked the FOV to interleave full-field background mea data from the focal plane. The digitized data were transmitted through a
surements. The shutter could also be positioned to view a closed-loop- twist capsule to the DPU. For gimballed targeted observations the
controlled integrating sphere simultaneously by each focal plane, scanning subsystem, controlled by software in the DPU, followed a
providing radiometric response and flat-field calibration (Fig. 2). Each programmed scan pattern carefully designed to compensate for orbital
4
Fig. 2. Sample VNIR and IR image frames (viewing internal integrating sphere).
motion and to accomplish the desired scan pattern across the Martian as the calibration reference due to its more uniform illumination of the
surface; for mapping it pointed at nadir relative to the instrument deck. sphere interior. A background measurement for a sphere observation
The DPU received and processed commands from the MRO, controlled was acquired looking into the sphere before lamps were powered (to
the CRISM subsystems, and acquired, edited, and formatted CRISM include the sphere’s thermal background).
science and housekeeping data that were then sent to the spacecraft
solid state recorder for downlink to Earth.
2.3. Overview of CRISM operations
2.2. Key variables in observing modes CRISM’s combinations of instrument settings to acquire various
types of observations is described in Table 2. “Targeted mode” provided
The DPU-controlled basic detector and OSU observing parameters high spatial resolution hyperspectral measurements of the surface, that
“switched” CRISM between gimballed (targeted) or push-broom (map included (through 2012) accompanying measurements of atmospheric
ping) observing modes or placed it into an internal calibration mode that opacity and trace gases. The original configuration is shown in Fig. 3;
was central to the processing of raw into calibrated data. Fig. 4 shows temporal evolution of targeted mode design as gimbal
range degraded. As a target was over flown it was covered by a slow,
2.2.1. Detector row selection continuous scan of the FOV, taking out most ground track motion.
On each detector, rows represent different wavelengths of a push- During this operation, the instrument gimbal covered angles ±35◦ . This
broom image, and columns represent spatial pixels (see Fig. 2). At any central scan was bracketed by five incoming and five outgoing ±0.3◦
given time, instrument software allowed four options for wavelength scans centered on the center point of the target, at 5◦ increments in
selection for each detector; at any given time, the observing modes gimbal position over the range of 40◦ -60◦ in gimbal angle. The total of
described in Section 2.3 used three VNIR and three IR configurations: all eleven scans provided an 11-angle emission phase function that contains
rows with measurable signal (targeted hyperspectral observations), information needed for photometric and atmospheric correction of the
selected rows that captured shapes of mineral absorptions (multispectral central targeted scan. In atmospheric EPF mode, during early operations
mapping,) or most rows with useful signal (hyperspectral mapping). the central scan was replaced by a single ±0.3◦ scan; then in 2007 two
additional ±0.3◦ scans were added to the central region of atmospheric
2.2.2. Pixel binning EPF, for a total of 13 images. Fig. 7 shows the full set of EPFs that
Pixels were unbinned or binned 2× in the spatial dimension (for accompany one of the original designs of targeted observations (FRT).
targeted observations), 5× in the spatial direction for higher-resolution Compared to the central scan, each EPF image covers a larger cross-track
mapping, or 10× in the spatial direction for EPF measurements or lower- dimension due to the greater line-of-sight range. Beginning in August
resolution mapping. The combination of row selection and pixel binning 2010, aging of the gimbal led to restriction of the range of gimbal mo
allowed for CRISM’s different mapping modes. tion, eliminating the 5 incoming ±0.3◦ scans. Beginning in October
2012 further restriction of gimbal motion to ~0◦ to +30◦ led to cessation
2.2.3. Frame rate of EPF measurements altogether, plus reduction in the along-track
Frame rates were matched to the observing mode to yield approxi dimension of the central scan. Three-letter acronyms for the different
mately square pixel footprints, targeted observations at a frame rate of targeted observing modes include FRT, HRL, HRS, and EPF through
3.75 Hz, or lower and higher spatial resolution mapping at 15 and 30 Hz 2012, and FRS, ATO, and ATU thereafter. More details on the configu
respectively. rations of each are in Table 2.
Atmospheric EPF mode was used every ~9◦ of solar longitude (Ls) to
2.2.4. Calibration lamps and shutter position acquire a low spatial density global grid of EPFs over 1 solar day to track
Three configurations of the shutter and integrating sphere were used seasonal variations in surface and atmospheric properties. This provided
to provide calibration observations to process every observation of a regular spatial and temporal atmospheric sampling, filling in between
“scene” on Mars. (1) For scene observations the shutter was held open and supplementing EPFs acquired as part of the targeted hyperspectral
and the integrating sphere lamps off. (2) Dark backgrounds acquired observations selected based on surface characteristics. Every ~36◦ of Ls,
with the shutter closed bracketed each scene image to measure the sum a cluster of grids was taken on non-contiguous days to provide a higher
of dark current and thermal background on the IR detector. (3) For spatial density grid to monitor seasonal change in surface material
observations of the internal, closed-loop-controlled integrating sphere spectral properties. The EPF grids were interspersed with other CRISM
that served as a calibration reference, the shutter was held in an inter and MRO observations on a best-effort basis, with repeat coverage to
mediate position. The sphere contained redundant lamps controlled by ±25 km. The restrictions on gimbal motion that began in August 2010
VNIR and IR focal plane electronics, but the IR-controlled lamp was used left an 8-position EPF measurement in this type of observation.
5
Table 2 Table 2 (continued )

CRISM Observing Modes. Mode Pointing Description and Coverage,
Mode Pointing Description and Coverage, Observation Type YEAR_DOY
Observation Type YEAR_DOY Acronym Used
Acronym Used
VNIR+IR
Full resolution (FRT) Hyperspectral survey ~39% of Mars,
Spatial pixels unbinned (HSP) focused on
for target – 18 m/pixel 262 channels, spatial selected areas
@300 km, 10× binned pixels 10× binned 2010_001 thru
for EPF (~180 m/pixel @300 2018_046
Half resolution short km)
(HRS) VNIR-only
Spatial pixels 2× Hyperspectral map
binned for target – 36 (HSV) ~87% of Mars
Tracking as 2006_270 thru
First-generation m/pixel @300 km, 107 channels, spatial 2009_174 thru
shown in 2012_146
targeted 10× binned for EPF; pixels 10× binned 2022_127
Fig. 4, once per (1-sided EPF
(hyperspectral) similar swath length as (~180 m/pixel @300
target after 2010_080)
above km)
Half resolution long VNIR-only
(HRL) Hyperspectral map
~96% of Mars
Spatial pixels 2× (MSV)
2012_013 thru
binned for target – 36 92 channels, spatial
2022_127
m/pixel @300 km, pixels 5× binned (~90
10× binned for EPF; m/pixel @300 km)
twice swath length as Observations of
Radiometric
above – onboard integrating Daily
Calibration
Along-track sphere
oversampled (ATO) Dark observations at
With 1-sided EPF thru Bias Calibration – multiple exposure Daily
2012_142; 2011_079 thru times
Spatially oversampled for on-ground subsequently without, 2012_142 Observations of bland
rendering at ≤12 m/pixel and with shorter 2012_293 thru regions of Mars (FFC)
Flat-Field 2006_270 thru
central scan with 2022_127 Nadir-pointed Spatial pixels
Calibration 2022_105
compressed spatial unbinned, 18 × 800 m/
pixels - ~12 × 18 m/ pixel
pixel
Full resolution short
(FRS)
Spatial pixels unbinned
Second- Central swath
for target – 18 m/pixel
Beginning in late 2008 through 2017, two new types of observations
generation only, scanned 2012_275 thru supplemented EPF grids with additional atmospheric measurements.
Along-track
targeted at 0◦ -30◦ 2022_127
undersampled (ATU) Approximately every 2 months a cluster of two or more orbits of limb-
(hyperspectral) gimbal angle
Spatial pixels scan measurements (abbreviated LMB), located at longitudes nearest
unbinned, smeared –
to Nili Fossae and Ascraeus Mons, were acquired to vertically resolve
18 × 36 m/pixel
Emission phase Lat./lon. Grids aerosols and trace gases. These longitudes were chosen to sample at
function (EPF) every ~9◦ of Ls mospheric properties of typical and high-elevation surfaces, respec
Tracking as Spatial pixels 10× 2006_270 thru tively; the Nili Fossae location additionally was a reported location of
shown In Fig. 4 binned 2012_146 transient increases in CH4 abundance [Holmes et al., 2015]. The MRO
(~180 m/pixel @300 (1-sided after
spacecraft was pitched to allow the instrument FOV access to the limb,
km) 2010_280)
Tracking Optical Depth and gimbal motion was used to scan the FOV from below the horizon to
Between other
(TOD) at least 120 km above it. Over each orbit ~15 scans were collected over
Atmospheric scheduled
(hyperspectral) Nadir-pointed
Spatial pixels 10×
observations the day side of Mars, and several additional scans past the ascending and
binned, 3.75 Hz (~180
2007_217 thru descending terminators looked for airglow over the polar regions. The
× 800 m/pixel @300
km)
2017_112 spacecraft pitch direction varied seasonally, biased to observe the illu
Limb scan minated pole. An optical depth tracking mode (abbreviated TOD) also
Limb scans w/ 2 orbits every 2
special
measurements (LMB)
months
supplemented EPF grids with high spatial density measurements with
Spatial pixels 10× full hyperspectral capability, to provide additional monitoring of trace
spacecraft 2009_192 thru
binned; ~180 m
pointing 2017_119 gases. The instrument was fixed pointing at nadir, and a short burst of
vertical sampling)
VNIR +IR data was taken periodically with CRISM’s full wavelength selection, but
Multispectral survey at reduced spatial resolution to manage data volume.
(MSP)
~86% of Mars In each of several “mapping modes”, the instrument was pointed at
73/74 or 94/95
2006_270 thru nadir, and selected wavelengths were measured at a reduced spatial
selected channels,
spatial pixels 10×
2018_046 resolution accomplished by binning pixels in the x and y spatial di
Mapping binned (~180 m/pixel rections, to manage data volume and maintain square pixels. Two modes
(selected Nadir-pointed @300 km) of multispectral operation were used initially: ~180 m/pixel “multi
wavelengths) VNIR+IR Multispectral spectral mapping” (MSP) mode which was designed to accomplish
windows (MSW)
coverage rapidly, and ~ 90 m/pixel “multispectral window” (MSW)
73/74 or 94/95 Selected targets
selected channels, 2006_270 thru mode which was intended for higher spatial resolution in key areas.
spatial pixels 5× 2008_072 Multispectral window observations were discontinued in March 2008
binned (~90 m/pixel due to electronic interference with control of the cryocoolers. Beginning
@300 km)
in January 2010, a ~ 180 m/pixel “hyperspectral mapping” (HSP) mode
using all VNIR wavelengths and an expanded set of IR detector rows was
6
Fig. 3. Elements to a CRISM targeted observation, shown in a Mars-centered reference frame.
Fig. 4. Temporal evolution of the elements of a CRISM targeted observation, shown in an MRO-centered reference frame.
initiated, to provide a contextual hyperspectral (at 0.4–1.0, 1.3–1.5, and was found that thermal expansion and contraction of the instrument at
1.9–2.5 μm) mapping mode for target-rich regions of the planet. When the beginning and end of decontamination periods caused the slit
the cryocoolers began to age and portions of CRISM’s observing time mechanism to settle into a position with a several hundredths of a
were spent in VNIR-only mode, two additional mapping modes were nanometer different position in the instrument wavelength direction. To
added: the hyperspectral VNIR (HSV) and the multispectral VNIR (MSV) quantify these variations, quasi-periodically, a flat-field calibration was
modes. Despite their names, both HSVs and MSVs comprised all or a acquired that crossed Olympus Mons, such that the ratio of atmospheric
majority of VNIR wavelengths with the main difference being spatial gas absorptions at the base and summit of the volcano (where there was
resolution at ~180 m/pixel and ~ 90 m/pixel, respectively. (Note that it nearly no atmosphere) could be used to accurately normalize gas ab
was also possible to obtain VNIR-only targeted observations but the sorptions during the periods between thermal excursions. Algorithms to
abbreviation for these modes remained unchanged.) correct for absorptions due to atmospheric gases have been coded to
Radiometric calibrations using the onboard integrating sphere and recognize these temporal discontinuities in wavelength calibration, as
measurements of detector bias were taken at least daily, bracketing the part of the “volcano scan” atmospheric correction (Murchie et al.,
thermal environment experienced by the instrument in the MRO orbit. 2009a).
In parallel, early results from Mars orbit showed that ground-derived As the three cryocoolers aged, two effects modulated CRISM’s
flat-field calibrations had excessive residuals from systematic instru overall investigation strategy. First, cooling capacity diminished, and
ment noise. To reduce noise, on at least a monthly basis, the flat-field the coolers were run at progressively higher temperatures, eventually
response was refined by imaging bland regions of Mars, and blurring exceeding the desired detector temperature of ≤120 K (Fig. 5). In
images in the along-track direction, a so-called “flat field calibration”. addition, an individual cooler would lose ability to hold temperature
Typically, this “bland region” was Arabia Terra, a relatively high-albedo until its power was cycled, introducing a gap in useful IR data while the
terrain. As data volume allowed, flat field calibrations were also ac same or another cooler was brought back to cryogenic temperature.
quired over Syrtis Major, a relatively low-albedo terrain, and the Whereas this approach prolonged the overall investigation, it resulted in
southern summer residual polar cap. increased stochastic noise in IR scene data, and stochastic noise in IR
Early in the Primary Science Phase, it was understood that water dark and sphere data that in the calibration pipeline became manifested
adsorbed into the CRISM instrument would migrate to the cooled IR as systematic noise in the calibrated data. Second, scheduling priority
detector and form ice, and so water needed to be baked out in order to was given to CRISM targeted observations rather than mapping
attain a stable radiometric calibration and minimize radiative transfer of observing modes. For use when the cryocoolers were off, a new mode of
heat to the cryogenic detector. Therefore, from time to time during the hyperspectral mapping at ~180 m/pixel using all 107 VNIR wave
cruise to Mars and during early Mars orbit, observing was suspended and lengths was initiated (abbreviated HSV), with an objective of globally
spacecraft-controlled survival heaters were used for “decontamination” mapping ferric iron minerals and low-Ca pyroxene. A related mode with
to raise internal instrument temperature to 30C◦ to drive out water. It 90 m/pixel spatial sampling using all 92 channels with useful signal was
7
Fig. 5. Evolution of the temperature to which the cryocoolers were able to cool the IR detector, over the full life of functional cryogenic cooling of the IR detector, as
produced by CRISM SOC automated telemetry processing software. Excursions to temperature temperatures above -147C occurred when a cooler was power-cycled
or the MRO spacecraft entered safe mode.
Fig. 6. Timeline of the configuration of CRISM targeted observations. In this and following plots, PSP = Primary Science Phase, ESP = Extended Science Phase, and
EM* = Extended Mission phase.
inaugurated in January 2012. The three-letter abbreviation for the new 2.4.1. Gimbaled observations: targeted, EPF, and limb scan measurements
mode is MSV. Acquisition of the last IR data with the detector at cryo The design of targeted observations of Mars’ surface progressed
genic temperature occurred on December 17, 2017. through two generations, the first tuned to as-built instrument capabil
ities, and the second adjusted to accommodate loss of gimbal angle
2.4. CRISM observation modes through time range in 2012. Fig. 6 shows the timeline of different design variants.
Fig. 7 shows map-projected versions of first-generation targeted obser
Below is a summary of the structure of CRISM data acquired in vations, and Fig. 8 shows map-projected versions of second-generation
different observing modes as usage of the instrument evolved over 16 targeted observations. Several gimbaled observation types were inten
years in Mars orbit. Additional information and the start and stop dates ded primarily to characterize the atmosphere. The timeline of these
of each observation mode are given in Table 2. observations is shown in Fig. 9, and examples are shown in Fig. 10.
The first generation of targeted observation modes was collected
from September 2006, through May 2012. The gimbal would operate by
8
Fig. 7. Map projected views of first-generation targeted observations. RGB image planes are 2.53, 1.51, and 1.08 μm. Width of the central scan at its narrowest point
is 10 km.
sequence was repeated outgoing, except in reverse order (part 3 in

Fig. 4). Four “dark” measurements of instrument background were also
taken, at the start and end of the EPF scans and bracketing the central
scan.
In a full resolution targeted observation (FRT), the gimbal was
scanned at a rate of 1 pixel (approximated as 60 μrad) per integration
time, and crossed the target at mid-scan yielding an effective pixel size of
~18 m from MRO’s orbit. The central scan and bracketing dark data
were taken without pixel binning. To conserve data volume, the EPF
Fig. 8. Map projected views of second-generation targeted observations. RGB

image planes are 2.53, 1.51, and 1.08 μm. Width of the central scan at its
narrow point is 10 km.
first pivoting to +60◦ , looking ahead of spacecraft nadir (reduced to

+30◦ beginning in August 2010). During approach to the target, the scan
profile slowly swept the FOV back and forth across the target, and five
±0.3◦ scans were superimposed (part 1 in Fig. 4, only prior to August
2010). These are EPF scans. During target over-flight (+35◦ to − 35◦
gimbal angle through August 2010; +30◦ through − 35◦ through May
2012; part 2 in Fig. 4), the gimbal made a much longer continuous sweep Fig. 10. Views of example observations intended primarily for atmospheric
across the target; configuration of this long central scan differentiates characterization (an example EPF is shown in Fig. 7). RGB image planes are
the classes of first-generation targeted observations. The incoming EPF 2.53, 1.51, and 1.08 μm.
Fig. 9. Timeline of observations intended primarily for atmospheric characterization.
9
Fig. 11. Timeline of mapping observations.
scans where present were taken with 10× pixel binning and the gimbal
scan rate for the EPFs yielded approximately square pixels projected
onto the surface. In half resolution long mode (HRL), the central scan
covered roughly twice the area of an FRT, at half the spatial resolution.
The FOV was crossed at a rate of 2 pixels per integration time, and the
data were taken with 2× spatial pixel binning; the higher scan rate
yielded approximately square pixel footprints. The dark data were bin
ned similarly. The EPF scans (one-sided from August 2010 through May
2012) were as with an FRT. In half resolution short mode (HRS) the area
covered by the central scan was similar to that in an FRT, but scan rates
and pixel binning were as with an HRL. These three observation designs
offered a menu of spatial resolution and areal coverage for different
surface priority targets. In an EPF measurement for atmospheric science,
the high spatial resolution central scan was replaced with 1 or 3 addi
tional 10× binned scans. The gimbal angle range used was +60◦ through Fig. 12. Views of mapping observations. RGB image planes are 2.53, 1.51, and
1.08 μm. Images are 10 km wide.
− 60◦ through August 2010, and + 30◦ through − 60◦ from August 2010,
through May 2012.
In 2011 as gimbal functionality began to decline, a variant of the FRT ATU central scan was bracketed by dark measurements, but the data
observing mode was introduced. By adjusting the gimbal scan rate over were taken without pixel binning. Thus, ATU pixels have a rectangular
Mars’ surface to be approximately 3 times slower, a spatially over footprint twice as long in the along-track direction, restoring the length
sampled observation akin to “super-resolution” could be achieved, with of the central scan almost to that of an FRT, or about 10 km.
the oversampling occurring in the along-track direction but only in the During these years of reinvention to compensate for loss of gimbal
center-most frames. In 2012, this “along track oversampled” (ATO) range, CRISM operations also made modifications in response to aging
mode was formalized and improved to have consistent oversampling and degraded performance of the cryosystem. To try to stretch out
across the entire length of the scene, although at the expense of overall CRISM IR acquisition lifetime, cryocooler operation was concentrated
ground coverage. The difference between the two types of ATOs is into 1 of every 4 planning periods; VNIR+IR data collection was only
apparent when map projected as shown in Fig. 8. attempted in these 2-week “cold cycles”, and VNIR-only data were ac
Procedures performed to process and render the spatially over quired in the intervening 6 weeks when a cooler was not operating.
sampled measurement at a higher spatial resolution of ≤12 m/pixel are Surface targets were sorted by the type of expected minerals, and then
described by Kreisch et al. (2017) and He et al. (2019). scheduled accordingly. For example, targets geared toward minerals
In a limb scan measurement (LMB), a spacecraft pitch maneuver put with absorptions at 0.9–1 μm (e.g., low-Ca pyroxene, ferric minerals)
Mars’ limb just within CRISM’s gimbal range. Images were binned 10× were relegated to VNIR-only cycles. Mineral targets expected to have
spatially, but the FOV was scanned across the limb at a rate of 1 pixel per hydroxysilicates, carbonates, and non-ferric sulfates were scheduled in
integration time to maximize vertical sampling. Multiple sets of limb “cold” cycles for VNIR+IR measurements. Examples of advanced data
scans were acquired during an entire dayside orbit while the orbiter products in Sections 5 and 6 show that this approach succeeded, with
remained pitched, with dark measurements located between each VNIR wavelengths successfully distinguishing ferric phases and differ
measurement of the limb. ences in pyroxene composition, as expected from Mars landed missions
After further restriction of gimbal range of motion to 0◦ to +30◦ in (Bell III et al., 2019).
2012, the second generation of targeted observation types was defined.
EPF measurements were no longer collected as part of any targeted 2.4.2. Nadir observations: mapping observations, TODs, and FFCs
observing mode, and EPF mode was retired. Instead of FRT mode, the Nadir observations also went through design evolution in response to
full resolution short (FRS) became the default high resolution targeted the discovery of unexpected mineralogical diversity on Mars and cry
observing mode for surface science. This abbreviated central scan at 1 osystem performance. Fig. 11 shows the timeline of the types of mapping
pixel per integration time was bracketed by dark measurements, both observation modes, and Fig. 12 shows map-projected examples.
acquired without pixel binning. In addition to the ATO described pre VNIR+IR multispectral mapping (MSP) was CRISM’s original map
viously, an along-track undersampled targeted observation (ATU) was ping mode, intended to cover ≥80% of Mars at 180 m/pixel to find sites
created that provided extended spatial coverage albeit at reduced along- for targeted measurements and to characterize composition over large,
track resolution scanned at a rate of 2 pixels per integration time. The contiguous areas. The basic configuration was a repeating sequence of
alternating Mars-viewing and background measurements. The Mars-
10
viewing periods were constrained to be in blocks of 3 min so that Table 3

adequate interpolation of background was possible. CRISM spent most CRISM Extended Mission Science Objectives and Selected Results.
of its observing time in this mode or its derivatives while cryocooling of Science investigation Selected results References
the IR detector remained an option. The 72 channels of MSP mode were
(1-ESP) Sample new targets
designed to characterize depths of mineralogic and atmospheric gas identified in the
Layered clay deposits
south of Coprates Chasma Buczkowski et al.,
absorptions known to be present from OMEGA data. Spatial sampling at multispectral survey that
required multiple 2020
10× spatial binning yielded a pixel scale of 180 m, and covering 86% of probe the structure of the
episodes of pedogenesis
Mars in the end. Hyperspectral Survey observations (HSP) were an Noachian crust
Vertical structure of
augmented form of MSP, with the same spatial sampling but increased (2-ESP) Map lateral primary mineralogy
spectral sampling within selected wavelength ranges in the IR and at all variations in crustal varies E-W. Alteration is
Viviano-Beck et al.,
VNIR wavelengths. The higher spectral sampling provided the capability layering and alteration discontinuous with
2017
to separate subtly different spectral signatures for minerals that were not exposed in the walls of evidence for elevated
Valles Marineris temperature along
known to exist prior to their discovery in CRISM targeted observations
regional fracture zone.
(e.g., Viviano-Beck et al., 2014; Ehlmann and Edwards, 2014), and to Regional geologic history,
detect and map absorptions from carbonates with different cation (3-ESP) Densely sample mineralogy and plausible
compositions. Global coverage in HSP mode reached ~39%. geologically complicated origins of Noachian Scheller and Ehlmann,
Multispectral Window observations (MSW) differ from MSPs in that aqueous deposits with materials on the west rim 2020
targeted observations of Isidis basin that form
they were acquired at 30 Hz with 5× pixel binning and only 15-s drainage basin of Jezero
duration, yielding 90-m effective pixels. MSWs were used mainly as Chlorides post-date clay-
(4-ESP) Characterize the
“ride-along” observations to a HiRISE or CTX image that was not rich units; deposition Leask and Ehlmann,
stratigraphy of chlorides
otherwise coordinated with a CRISM targeted observation. After MSW and associated
extended to Amazonian; 2022;
no ancient buried Beck et al., 2020
observations were discontinued in 2008, MSP or HSP data served the phyllosilicates
chlorides
same purpose. Locations and relations of
VNIR-only hyperspectral mapping was performed while cryocoolers widespread occurrences
were off, from 2009 to 2022. There are two variants: hyperspectral VNIR of Fe-Mg clays and Fraeman et al., 2013;
(HSV) observations are analogous to the HSP except with all 107 VNIR (5-ESP thru EM5) Sample carbonates, locations of Fraeman et al., 2016;
possible landing sites, restricted Al-clays, Ehlmann and Buz,
wavelengths and no IR wavelengths. HSV data and accompanying
densely sample and map hydrated silica in Jezero 2015; Goudge et al.,
background calibrations were taken in 10× pixel binning mode at 15 Hz, selected sites to crater. Enhanced spatial 2017; Tarnas et al.,
yielding 180-m effective pixels. Multispectral VNIR (MSV) observations characterize surface resolution from ATO 2019; Horgan et al.,
are analogous to MSW except for being limited to the 92 VNIR channels compositional and observations. Retrieval of 2020; Kreisch et al.,
physical properties and apparent thermal inertia 2017; He et al., 2019,
having useful signals. MSV data and accompanying background cali
support rover traverse from 3 to 4 μm 2022; Christian et al.,
brations were taken in 5× pixel binning mode at 30 Hz, yielding 90-m planning wavelength thermal 2022; Moreland et al.,
effective pixels. emissions. Variations in 2022
Tracking Optical Depth observations (TOD) were designed to fill mafic mineral
gaps between other hyperspectral observations to maintain a high compositions of sands
Originate from NPLD
spatial density of sampling and detection of atmospheric trace gases.
Basal Unit (mixture of
They were taken with all wavelengths at 3.75 Hz and 10× pixel binning (6-ESP) Search for sources of
water-ice, sand/dust,
hydrated sulfates exposed
in sets of 4 bursts every approximately 48 s, yielding 10 × 10 km foot gypsum, polyhydrated Mg Das et al., 2022
in Olympia Undae in north
prints every 2◦ of latitude. polar region
sulfate); implies basal
melting during
Flat-Field Calibration observations (FFC) were acquired over bland,
Amazonian
dusty terrain in Arabia Terra and across Tharsis over Olympus Mons to, Interannually repeatable
respectively, (a) provide empirical correction to the ground-derived flat- (7-ESP) Interannual cycles of H2O and CO2
field calibration, and (b) provide a time-segmented empirical measure variations in condensation deposition / ablation at Brown et al., 2012,
ment of attenuation of solar radiance by Martian atmospheric gases. / sublimation of seasonal each pole; cycles 2014
polar caps distinctly different
FFCs were taken with all wavelengths at 3.75 Hz without pixel binning
between poles
throughout the duration of the CRISM investigation; no changes in Resolution of small
designation or commanding were made in response to cryosystem (8-EM1) Use the ATO spatial variations in
functionality. observing mode to hematite crystallinity Fraeman et al., 2020
improve spatial resolution investigated in situ by
MSL
2.4.3. Internal calibrations (CAL and ICL observations) Hellas basin-ring massifs
Instrument radiometric and bias calibration was performed at least (9-EM1) HSP mapping of
include feldspathic
mineralogically complex
daily. An Internal Calibration observation (ICL) consisted of a set of regions to extend
compositions; original
integrating sphere measurements (with the sphere operated under location in lowermost Phillips et al., 2022
hyperspectral coverage
crust consistent with
closed-loop control) and bracketing measurements of the ambient and mineral
preservation of primary
background with the shutter viewing the darkened sphere. These data discrimination
crust
were used to recover radiometric responsivity. Although data were ac Subsurface aqueous
quired using the lamps controlled by each of the VNIR and IR focal plane (10-EM1) Map 3D structure alteration widespread but
Carter et al., 2013,
electronics, the flight calibration and ground calibration pipeline only of Noachian crust by most prevalent in certain
2015
systematically imaging regions; widespread
used sphere images illuminated by the IR-controlled lamp because of its impact craters occurrences of pedogenic
Pan et al., 2017
better along-slit uniformity. The integrating sphere provided sufficient deposits
signal for measurements in the IR, but at VNIR wavelengths <600 nm (11-EM1) Improve vertical Distinct seasonality,
there was insufficient signal at a single detector element to determine resolution of atmospheric particle size / latitude /
Clancy et al., 2019
structure and aerosols longitude / altitude
non-uniformity at the desired accuracy of 0.001. For those wavelengths,
using limb scans distributions of H2O and
pre-launch radiometric calibrations have been used.
(continued on next page)
Bias Calibration observations (CAL) were performed at least daily. A
bias calibration consisted of a set of shutter-closed measurements at
11
Table 3 (continued ) and highlights selected results. A more detailed synthesis of results of the
Science investigation Selected results References CRISM investigation over all mission phases – and those of OMEGA – is
provided by Murchie et al. (2019).
CO2 mesospheric ice
aerosols
During ESP and the EMs, radiometric calibration of the data reached
Clays associated with fans its final version (version 3) and – for targeted observations – a parallel
(12-EM2) Characterize clay-
and deltas are detrital version of the data was developed and produced in which noise due to
rich deposits in fluvial and
deltaic fans, central peaks
from watershed. Both Goudge et al., 2017; elevated IR detector temperature was remediated by filtering (Section
landform types overlain Pan et al., 2021
(15-EM3) Complete 4). The availability of filtered version 3 calibrated data enabled the
by hydrated silica with
coverage of deltas
distinct spatial patterns discovery and analysis of smaller outcrops of materials having weaker
Amazonian volcanics spectral signatures.
(13-EM2) Use fresh craters
to penetrate dust cover
mineralogically similar to In addition, two major classes of new, highly processed “special
Hesperian, distinct from products” were developed and released by the CRISM Science Opera
and map the compositions Viviano et al., 2019
Noachian; only major
of volcanic rocks in
change at Noachian/
tions Center (SOC). The first new class of products was “Map-projected
Tharsis and Elysium Targeted Reduced Data Records” using the version 3 calibration
Hesperian boundary
Range of Fe,Ca rich (MTRDRs; Section 5). In these products, corrections to I/F in targeted
(14-EM3) Characterize the
carbonate compositions observations for atmospheric gas absorptions and solar incidence angle
in pre-Noachian rock
diversity of carbonate Wray et al., 2016; were performed using standard procedures from the CRISM Analysis
exposed from depth;
minerals exposed from
widespread occurrence at
Amador et al., 2018 Tool (CAT; Murchie et al., 2009a). In addition, cross-track instrument
depth artifacts were remediated using an empirical correction, and along-track
low concentration may be
CO2 sink radiometric variations induced by the continuously variable viewing
(16-EM3) Global VNIR Work in progress; focus geometry were normalized using a fit of I/F through the geometries of
mapping to inventory and has been maturation of Frizzell et al., 2020
map ferric minerals data processing
the central scan and accompanying EPF images. The latter correction
Opal- and acid sulfate- had the effect of normalizing atmospheric aerosol opacity to that at the
(17-EM4) Nature and extent
rich materials detected minimum emission angle in the central scan – an observation-dependent
of Hesperian- to
Amazonian-aged aqueous
and mapped at depth,
Smith et al., 2019 value. Data from the VNIR and IR detectors were co-registered, map-
covering wide band
deposits on the rim of projected, and bad bands (i.e., those having degraded calibration ac
adjacent to Valles
Valles Marineris
Marineris curacy) were removed. By definition, MTRDRs were generated only for
Identification and targeted observations acquired during or prior to May 2012, when the
mapping of amorphous
Sun and Milliken,
gimbal range of motion permitted acquisition of EPF images. The data
silica; additional evidence were formatted as multiband images of corrected I/F with companion
(18-EM5) Distribution and 2018;
that crystallized hydrated
age range of impact glass Sinha and Horgan, images showing spectral indices or “summary parameters” and com
silica is important
2022 posite, quick-look thematic browse images (Viviano-Beck et al., 2014) to
component of northern
plains sand facilitate an overview of minerals detected. An analogous product set
Short-term warm and wet with data still in sensor space and all channels present (“Targeted
(19-EM5) Transitional environments during a
Empirical Records”, TERs, Section 5) was also generated.
environments recorded in generally cold early Mars Bishop and Rampe,
layered clay deposits and transitioned to an acidic 2016; The second new class of products was the final generation (version 4)
stratigraphically adjacent environment with Bishop et al., 2018 of Multispectral Reduced Data Records (MRDRs; Section 6) – a series of
acid sulfates sporadic water 1964 pieces or “tiles” of the global, 72-channel global multispectral map
availability in which data were map-projected, corrected for photometric effects and
cross-track instrument artifacts, and formatted similarly to MTRDRs. A
fundamental difference from MTRDRs was how the atmospheric
correction was performed and its effects on the derived multiband im
each frame rate, at 4–5 integration times per frame rate. These data were ages. The underlying strips of mapping data fell into two groups: those
used to recover IR detector bias, i.e., the offset image with zero scene acquired during the time range of globally distributed EPF measure
radiance or thermal background. This is a necessary input to lineariza ments, and those acquired after the gimbal lost that capability. For the
tion of the data, in the calibration procedure described by Murchie et al. first group, atmospheric corrections were performed using DISORT
(2009a). For the VNIR detector, bias dominates background measure radiative code (Stamnes et al., 1988), adapted for Mars conditions
ments so a separate measurement was not needed. (Wolff et al., 2009), with inputs including observation geometry, ele
vations and solar longitudes of pixel footprints, estimated dust opacity
3. Science investigations and data products over the extended τdust from CRISM EPFs, and estimated water-ice opacity τice from MARCI
science phase and extended missions images (McGuire et al., 2008, 2013). In addition to modeled gas ab
sorptions being divided out, the aerosol contribution to measured
Murchie et al. (2009a) described key science objectives and corre radiance was scaled to τdust = 0.2 and τice = 0.0. For the second group of
sponding observing campaigns for CRISM during the PSP (2006–2008), mapping strips, corrections for atmospheric gas absorptions and solar
and Murchie et al. (2009b) summarized the major results regarding incidence angle were performed using standard CAT procedures. Then,
surface composition. Since then, an Extended Science Phase (ESP) and using scaling factors weighted by adjacency and overlap of strips, all of
five Extended Missions (EM1-EM5) have occurred, enabling new con the strips from both groups were normalized to strips in the first group
tributions to Mars science and planetary exploration. Table 1 gives a collected under the clearest atmospheric conditions of the mission (late
summary of aggregated CRISM objectives over all mission phases, how 2007; τdust = 0.2). The result was corrected I/F as it would appear in the
they were addressed by different types of observations, and an overview absence of atmospheric gases and with standard aerosol opacity.
of the resulting data products. CRISM’s investigations during the
Extended Science Phase (ESP) and five subsequent Extended Missions 4. Final radiometric calibration and standard data products in
(EM1 through EM5) built upon new questions raised in the aggregated sensor space
results from earlier phases, by introducing new, focused science ques
tions and corresponding observing campaigns that would each extend By the close of PSP, both VNIR and IR radiometric calibration had
across multiple mission phases. Table 3 summarizes these campaigns proceeded to version 2. Murchie et al. (2009a) described evolution of the
12
algorithms over versions 0, 1, and 2, and remaining issues. During ESP by a ground-derived radiance model of the sphere to which a non-
and EM1 radiometric calibration was further upgraded to version 3. In uniformity correction derived from FFC measurements was applied.
addition, a filtering procedure was applied to IR targeted observation • To convert radiance to I/F, radiance values were multiplied by solar
central scan images (all hyperspectral, with spatial pixels unbinned or distance in AU squared, and divided by pi times the solar spectral
2× binned) to remediate noise from elevated IR detector temperature. irradiance at 1 AU scaled to CRISM bandpasses Solar spectra
The following discussion briefly summarizes issues with radiometric resampled to CRISM observation types are available from the PDS at
calibration that were corrected in version 3, and then describes appli https://pds-geosciences.wustl.edu/mro/mro-m-crism-2-edr-v1/
cation of filtering in selected version 3 calibrated data to remediate mrocr_0001/cdr/sf/
systematic and stochastic instrument noise.
For context of what parts of the CRISM data set are affected by such a
calibration update, CRISM “standard” data products (raw and calibrated 4.1. VNIR and IR current calibration versions
data, calibration files, and geometric information) have been delivered
to the Planetary Data System (PDS) as five product types that support Through 2010, version 2 calibration was applied to Mars scenes
derivation of the higher order “special products” described above processed to radiance or I/F, where Mars overfilled the scene and scat
(MTRDR, MRDR, TER), with a separate product for data from each of the tered light corrections were applied using the unmasked pixels outside
VNIR and IR focal planes. The structure and contents of all standard the illuminated FOV. There were two alternate versions of version 2
products are described in the Data Product Software Interface Specifi specifically for special types of observations. (a) The bland Mars scenes
cation (DPSIS) available online from the PDS (Murchie et al., 2022). An measured with FFCs to derive spatial non-uniformity calibrations were
“Experiment Data Record” (EDR) is a collection of raw VNIR or IR processed to version 9 TRDRs. Version 9 processing is the same as
frames in units of DN collected sequentially using a consistent instru version 2 except that a non-uniformity correction is not applied to the
ment configuration., e.g., scene measurements, background measure integrating sphere model (rather, it is the basis for that correction). (b)
ments, or sphere measurements, in the format of a multiband spatial Images of Deimos, Phobos, and any other compact target were processed
image. A single observation yielded multiple EDRs on account of to version 8 TRDRs. The major difference in the processing for such
interleaved scene and background measurements. Geometric and scenes from version 2 is that within-scene scattered light from the
ancillary information needed to derive map-projected special products grating is much reduced, so the correction for intra-scene scattered light
or to support data analysis were packaged into two other types of files is skipped.
whose spatial dimensions match those of an EDR containing scene data, Version 3 was the final version of radiometric calibration, released in
a “Derived Data Record” (DDR) for an observation of the surface or a late 2010. For the special scene types described above, version 7
“Limb Data Record” (LDR) for an observation of the limb. EDRs, DDRs, replaced version 9 for processing FFCs, and version 6 replaced version 8
and LDRs are independent of radiometric calibration and, once created, for processing images of Phobos and Deimos. There were several im
are not replaced by subsequent versions except in case of errors in file provements over version 2 and its derivatives. First, in scene measure
creation. ments, the shutter slightly vignetted the shorter-wavelength zone of the
Two product types incremented their versions upon introduction of a 2-zone IR diffraction grating. In version 2, this was treated as a fixed
new calibration version. A “Calibration Data Record” (CDR) was created effect. However, slight irreproducibility in shutter mirror position was
from ground processing of an inflight calibration or a ground calibra inferred to occur between scene measurements, resulting in calibrated
tion, and one or more types of CDRs may have been updated in a new data values being systematically too high or low at 1000–1700 nm.
calibration version. The main product type that was updated by a new Instead, the magnitude in the shutter position correction was scaled
calibration version is a “Targeted Reduced Data Record” (TRDR), a pair based on the shape of the attenuated response along the boundary of the
of multiband images created from a scene EDR, a radiance image and an two VNIR order-sorting filters. (Shutter position affected the shape of
I/F (reflectance) image. The file name itself embeds the calibration the shadow between the two filters.) This approach was found to yield
version, and an associated text label lists all of the CDRs used. greatly improved reproducibility of different TRDRs covering the same
The related files for a given observation and image segment within scene. Second, the correction of leaked 2nd-order light at IR wave
the observation are linked by a unique “observation ID” and by an image lengths near 3100–3300 nm was improved. The algorithm used in
counter within the observation, together providing unique identifier for version 2 had propagated an artifact at an order-sorting filter boundary
any given raw, calibrated, or derived CRISM image. near 1630 nm (in the wavelength range from which the leakage was
The DPSIS describes the radiometric calibration algorithm in detail being estimated) to 3180 nm (in the wavelength range within which the
(Murchie et al., 2022). Key steps include the following, and uses the leakage was being removed). Interpolation of the leakage correction
matrices of coefficients recorded in a library of derived CDRs: over a wider band at the 1630-nm boundary was found to greatly reduce
the artifact near 3180 nm. Third, a correction for atmospheric water
• Subtraction of a detector bias measured by shutter-closed measure vapor in ground calibration measurements was modified to reduce
ments in the VNIR and estimated from inflight CAL observations in spectral artifacts near 1850 and 2550 nm. Fourth, the width of the band
the IR; near 2000 nm within which flat-fielding is not applied was increased, to
• Linearization of raw signal using a linearity relationship derived remove distortions in Mars’ atmospheric CO2 absorption near 1830 nm.
onground; Absolute and relative accuracy over hundreds of nanometers wave
• Subtraction of instrument electronics artifacts using relationships length of version 3 calibration was tested at ≤2600 nm by the mineral
derived onground; mixture modeling of Lapôtre et al. (2017). They modeled
• Subtraction of thermal background in the IR spectrometer using atmospherically-corrected CRISM reflectance of the Bagnold Dunes at
shutter-closed measurements; Gale crater as mixtures of primary minerals of different grain sizes,
• Subtraction of scattered light from the grating, estimated in shape whose abundances were measured in situ by the CheMin investigation
from the shortest-wavelength data and scaled in magnitude using on MSL (Blake et al., 2012). CheMin in situ measurements were within
unmasked pixels outside the FOV; error bars of the best-fit model mineral mixtures. Residuals exhibited
• Subtraction of higher-order light from the IR grating using scaled little to no organized wavelength structure, which indicates a lack of
values from shorter wavelengths representing λ/2; meaningful structured error in relative calibration over wavelength
• Division of data in units of DN by exposure time; and ranges comparable to mineral absorptions due to feldspar, pyroxene,
• Correction to radiance by dividing by a sphere observation close in and olivine at 1000–2500 nm. The ratio of bright and dark minerals was
time and processed through all of the above steps, then multiplying correctly modeled to within uncertainties, which is predicated on
13
Fig. 13. CRISM hyperspectral data filter processing flow chart. Boxes represent data processing steps described in the text; stacks represent CRISM multiband images.
All procedures and intermediate products are employed in filtering of IR data. Elements shown in green are bypassed by the VNIR filtering process. (For inter
pretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
accurate absolute radiometric calibration. This result provides addi 4.2. Noise-mitigation filtering in version 3 I/F images
tional evidence for absolution calibration accuracy to the evidence
summarized by Murchie et al. (2009a). Full and half resolution hyperspectral data products (FRT, HRS, HRL,
Several data quality issues remain in the version 3 radiometric FRS, ATO, ATU central scans only) processed to I/F using version 3
calibration. (a) Radiance at <410 nm is low, and is susceptible to arti radiometric calibration have been processed with a custom filtering
facts from scattered light subtraction. Such artifacts are largest where procedure prior to PDS delivery to remediate instrument optical artifacts
the scene contains material that is brighter at ultraviolet wavelengths and systematic and stochastic noise (Fig. 13; Seelos et al., 2009, 2011).
than typical Martian regolith, e.g., ice or frost. As a general practice, The companion radiance spectral data products have not been filtered,
wavelengths <410 nm should be ignored, and they are included among allowing for the recovery of unmodified reflectance spectra (Fig. 14a).
the “bad bands” flagged in TERs and excluded from MTRDRs. (b) An IR observations acquired at higher detector temperatures exhibited
artifact remains at the wavelengths of the VNIR filter zone boundary. In increased systematic and stochastic noise. The primary resulting sys
some parts of the field of view, at sharp brightness contrasts, artifacts tematic noise component in TRDRs was along-track, column-oriented
extend over a wavelength range as broad as 644–684 nm. This artifact is striping due artifacts in ground calibration files applied to inflight
scene-dependent, but no forward model of its magnitude has been integrating sphere image observations. This type of systematic noise was
developed. (c) Scenes with large spatial coverage by ice, especially near addressed by the Ratio Shift Correction (RSC), a scene-dependent mul
either edge of the field of view, have more significant artifacts from the tiplicative correction frame developed through the serial evaluation of
scattered light correction because, unlike typical Martian soils, ice has channel-specific inter-column ratio statistics. The dominant stochastic
significant UV reflectance and that impacts accuracy of the correction. noise component was in IR data, and appeared as isolated positive or
Typical effects may include degradation at wavelengths below 480 nm negative data spikes or as column-oriented groups of pixels with erro
and above 1010 nm. (d) Due to spectral smile, mineralogic absorptions neously high values. The spikes are thought to have originated from the
located where instrument response varies steeply with wavelength delayed response of specific detector elements to high time frequency
exhibit variation with field angle. This effect is most prominent within changes in brightness as the slit moved across a scene on Mars having
Fe mineral absorptions near 900 nm and at the edge of atmospheric gas large brightness contrasts at small spatial scales; column-oriented fea
absorptions near 1400 and 2000 nm. These effects are largely corrected tures came from transient elevated values of detector dark current or
within the TER and MTRDR data sets described in Section 5. (e) Many IR from noise that propagated from ground calibrations. Noise in IR data
images exhibit an artifact at 1630 nm that originated from a boundary was identified and corrected through the application of an Iterative
between order-sorting filters. The affected wavelengths are included Kernel Filter (IKF), which employed a formal statistical outlier test as the
among the “bad bands” flagged in TERs and excluded from MTRDRs. (f) iteration control and recursion termination criterion. This allowed the
There is commonly a mismatch between calibrated radiances from the filtering procedure to make a statistically supported distinction between
VNIR and IR detectors at the wavelength boundary. It is likely that this high frequency (spatial/spectral) signal and high frequency noise based
results from differences between the detectors in residual, uncorrected on the information content of a given three dimensional (x,y,λ) data
scattered light. This effect is recorded within ancillary products that are kernel.
part of the TER/MTRDR data deliveries as described in Section 5. (g) In
version 3 IR I/F images that have been filtered to remediate noise (see 4.2.1. Data processing overview
below), coincidence of pixels having elevated noise (“spikes” as The CRISM hyperspectral data filtering process took as input a three
described below) within shoulders of the 2-μm CO2 absorption led to dimensional (x,y,λ) VNIR or IR I/F full or half resolution hyperspectral
creation of narrow absorption-like features. The shape and spacing of multiband image (Fig. 13; TRR I/F) and generated a filtered version of
the false absorptions closely resemble those of real features in perchlo the input data (Fig. 13; Restore I/F) with a 1:1 correspondence between
rate minerals. Leask et al. (2018) developed detailed documentation of a given input and output pixel. The two main data filtering procedures –
this artifact. RSC and IKF – were performed on continuum-normalized spectral data
To control data quality, only IR data acquired with a detector tem to avoid filtering out real high spatial-frequency information. Thus, the
perature ≤ 127.5 K were routinely calibrated and delivered to the PDS. initial data processing steps were the calculation of a reference image
cube (Calculate Reference in Fig. 13) and the calculation of a ratio image
cube using the reference image cube as the denominator (TRR I/F /
14
Fig. 14. Succession of 3-panel composite views of of FRT0000C202 (central scan), showing the processing of a single scene through the filtering procedure shown in
Fig. 13. Each composite plot consists of (top right) a 2300-nm (x,y) ground plane image with a 1% linear stretch; (top left) a boxplot showing the data distribution
and stretch limits for the displayed band; and (bottom) a median spectrum and [1,99]% envelope for the input image. Observation FRT0000C202 was acquired with
an IR detector temperature of ~120.8 K. The systematic and stochastic noise components to be addressed by the filtering procedure are evident as bright pixels and
vertical lines. (a) Input I/F hyperspectral multiband image cube. (b) Reference cube. High frequency spatial information was retained while high frequency spectral
information was discarded. (c) Ratio cube. Pixel scale variability due to spatial variations in reflectance were normalized out of the ratio cube; both high frequency
noise and high frequency signal components of spectral variations were retained. The spectral plot corresponds to the center pixel in the scene. (d) Ratio RSC cube.
Systematic column-to-column (along track) artifacts have been mitigated using the RSC. (e) Resolve cube. Stochastic noise has been highly reduced using the IKF
while real high frequency spatial/spectral variability has been retained. (f) Resolve RSC cube. Systematic along track banding has been mitigated. (g) Restore cube,
output I/F. Compare the ground plane image, median spectrum and spectral percentile envelope, and data distribution boxplot to (a). (h) Ratio of input / output
showing the effect of the data filtering procedure in the ground plane. Note that the single pixel input/output ratio spectrum and 2300-nm ratio image are centered
on unity.
Reference in Fig. 13). In this normalized space, the filtering procedures to the application of the IKF procedure. The image cube resulting from
applied depended on the wavelength range of the input image cube. For the ratio space data processing (VNIR: Ratio RSC; IR: Resolve RSC in
VNIR data, only a single application of the RSC procedure was required. Fig. 13) was then recombined with the previously calculated reference
The noise structure in CRISM IR data is more complex, and required the cube (Resolve RSC * Reference in Fig. 13) to transform the data back to
use of the RSC procedure as both a pre- and post- processing step relative I/F (Restore I/F in Fig. 13). The image cubes used and/or produced and
15
Fig. 14. (continued).
corrections applied to them are described further below. (Fig. 14c). High-frequency real spatial variability in I/F was normalized
out of the ratio to preserve real small scale reflectance variations,
4.2.2. TRR I/F whereas high frequency spectral variability (either signal or noise) was
This was the input multiband hyperspectral image “cube” (e.g. retained. Subsequent data processing occurred either in ratio space or in
FRT0000C202 in Fig. 14) calibrated to I/F (Fig. 14a). a space resulting from further transformations of the ratio cube.
4.2.3. Reference cube 4.2.5. Ratio Shift Correction (RSC)

The reference cube was a low spectral frequency / high spatial fre The RSC procedure was applied to the IR ratio cube as both a pre-
quency representation of the input image cube (Fig. 14b). The spectrum processing (Fig. 14.d) and post-processing (Fig. 14f) step, before and
for each spatial pixel was processed with a series of boxcar low-pass after application of the IKF procedure, but as only a single filtering step
filters tuned implementations of the RSC procedure to produce a for VNIR data. Within a given spectral band, each spatial column cor
“noise-free” reference spectrum for each pixel. Any residual noise in the responds to a single detector element. The Ratio Shift Correction char
reference cube would be propagated into the final result. acterizes the calibration residual at each detector element through the
evaluation of inter-column (i.e., cross-track) ratio statistics relative to a
4.2.4. Ratio cube cross-track model that resembles shape of the instrument’s spectral
This was the input image cube divided by the reference cube smile (Murchie et al., 2007). Modifying the complexity of the underlying
16
Fig. 15. Photometric geometry and 760-nm I/F for FRT00003E12, which is located at low latitude and was observed with 5.7◦ of spacecraft roll off-nadir. Blue,
green, and red colors indicate the incoming EPF images, central scan image, and outgoing EPF images as shown in Fig. 3. (a) Phase angle vs. absolute value of the
emission angle. (b) 760-nm I/F vs. directional emission angle, showing enhancement of radiance from aerosols at higher emission angles. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)
cross track model allowed the RSC procedure to remediate both high was iterated. Model iteration was terminated when no further outliers
spatial frequency column striping (Fig. 14c to 14d) and low spatial were detected. The filtered value for the target pixel at the center of the
frequency banding (Fig. 14e to Fig. 14f), while retaining real scene input kernel was then given by a proximity-weighted model of the kernel
cross-track spatial variability. elements that were not marked as outliers. The confidence level
threshold for the Grubbs test was conservative, so the filter retains some
4.2.6. Iterative Kernel Filter (IKF) marginal noise (Fig. 14e) rather than erroneously removing real spectral
The IKF modeled as a multidimensional polynomial the information structure.
content of a given three dimensional (x,y,λ) normalized data kernel
several detector elements in each dimension that moves through the 4.2.7. Restore I/F
cube. Model residuals were treated as a sample set and examined for The result of the ratio space data processing (Fig. 14f) was multiplied
outliers using the Grubbs test. If an outlier was detected, the corre by the reference cube (Fig. 14b) to transform the filtered data back to I/F
sponding pixel was removed from consideration and the kernel model (Fig. 14g). This represents the noise-remediated I/F multiband image.
Fig. 16. Summary of TER and MTRDR processing flow. Colors show which of VNIR and IR data to which processing steps are applied, and whether the outcome of
the processing step is deliver to the PDS.
17
Fig. 14h shows the systematic and stochastics artifacts that were information was derived from the ‘INA at areoid’ band (incidence angle
removed from the example input image cube (Fig. 14a). with respect to the Mars areoid) in the corresponding DDR associated
with the targeted observation central scan. The DDR INA information
5. TER and MTRDR data processing and product description was calculated with respect to the Mars Orbiter Laser Altimeter (MOLA)
areoid, which is stored in the PDS as a 16 pixel/degree gridded data
5.1. Overview of input data and processing product. CRISM targeted observations spatially oversample the gridded
areoid data; to prevent the propagation of tessellation into the TER/
As described in Section 2.3 and Fig. 4, for the first ~40% of CRISM’s MTRDR data products, the INA data were fit by a second-order two-
orbital investigation, each high-resolution targeted observation “central dimensional polynomial in CRISM sensor space, and the polynomial
scan” was accompanied by 5 companion, reduced spatial resolution EPF model was used as the incidence angle information source in the spatial
images either inbound or both inbound and outbound along the MRO pixel specific photometric correction. The variation in incidence angle
spacecraft ground track. The EPF images of each target covered a much (at the areoid) across a CRISM targeted observation central scan is very
larger range of photometric angles than did the central scan by itself, as small, so the photometric correction does not differ greatly from a scalar
well as – at higher emission angles – a greater enhancement of the multiplicative change in scale. The correction was applied to all images
fraction of measured radiance that came from atmospheric aerosols within a targeted observation to maintain consistency across the central
instead of the surface, due to greater atmospheric path length (Fig. 15). scan and EPF images for downstream data processing.
Fitting spectral radiance through the aggregated geometries of the There was no attempt made to correct for photometric effects at
central scan and EPF images provided a basis for normalizing the aerosol scales smaller than the aeroid (i.e., small-scale topography). Variations
contribution to that at the lowest emission angle in the central scan, thus in incidence and emission angles are correctable in principle, but the
minimizing the radiometric effects of the continuously variable gimbal effects of associated differences in direct and sky illumination would be
geometry. far more complicated, and any corrections more prone to error. Thus it is
The MTRDR and TER data product sets used the EPF suite acquired left to the user of the data to ascertain such effects. At a very high level,
with each targeted observation, along with photometric and atmo the effect of shading is to impart differences in the spectral continuum
spheric gas correction algorithms standardly applied to CRISM data (measurable in indicators of spectral slope) but not in absorption depths.
using CAT, plus corrections for known instrument optical artifacts, to
render the observation as it would be seen by an idealized version of 5.4. Modified ‘volcano scan’ atmospheric correction
CRISM over the instrument’s full wavelength range, without absorption
by atmospheric gases, with uniform near-nadir pointing. The corrected The volcano scan atmospheric correction used empirically derived
values of I/F were accompanied by a suite of summary parameter images Mars atmospheric transmission spectra to correct CRISM IR I/F for at
and thematic browse product images to provide an overview of the mospheric gas absorptions due to CO2, H2O, and CO (Langevin et al.,
spectral content of the central swath. The process flow to produce these 2005; McGuire et al., 2008; Morgan et al., 2011). There are no signifi
product sets is shown in Fig. 16 and explained below. At several points in cant atmospheric gas absorptions over the CRISM VNIR wavelength
the discussion, the geometry, processing effects, and outputs of the range, so this correction was not applied to VNIR data. Detector column-
targeted observation FRT00003E12 (located in Nili Fossae and the type specific reference atmospheric transmission spectra (to accommodate
location for Mg carbonate; Viviano-Beck et al., 2014) are shown for spectral smile) were derived from a suite of FFC observations over
illustrative purposes. Olympus Mons, and were matched to a given targeted observation based
on fitted position of the 2000-nm CO2 triplet to minimize high spectral
5.2. Down-selection of targeted observations frequency residual structure due to sub-nm temporal variability in
wavelength calibration of the IR detector. The depth of the 2000-nm CO2
Pre-processing of input targeted observations calibrated to I/F con absorption was determined for the spectrum of each spatial pixel [xi,yj]
sisted of two further down-selections of the input data. First, only data in the multiband images under consideration, the selected reference
acquired with an IR detector temperature of ≤127.5 K are included. This transmission spectrum was scaled to match according to the Beer-
was determined using the calibrated value of IR_DETECTOR_TEMP1 Lambert Law, and the resulting model transmission spectrum was rati
from the PDS label on the multiband image. Second, only data from time oed out of the I/F data. The TER/MTRDR pipeline implementation of the
periods of relatively low atmospheric dust and ice dust opacity as volcano correction also included a CRISM-specific patch that was
determined by analysis of the accompanying EPFs and smultaneous applied after the initial correction to reduce the influence of varying
color imaging from the Mars Color Imager (Bell III et al., 2009; Wolff path length and pressure broadening of the 2-μm CO2 absorption in the
et al., 2009) were included. Cutoff values used based on inspection of the reference transmission spectra. The atmospheric correction was applied
results were τdust < 1.39 (at 900 nm from CRISM) and τice < 0.29 (at 320 to all images within a targeted observation to maintain radiometric
nm from MARCI). These approximate the 90th and 95th percentiles of consistency between the central scan and EPF images for downstream
the distributions of the tau data, and so only the haziest scenes were data processing.
eliminated. Other than the selection cutoff, the retrievals were not used
to inform further processing. 5.5. Ratio shift correction (empirical flat field / along-track ‘destriping’)
The Ratio Shift Correction (RSC) was a spatial column-oriented

5.3. Lambertian photometric correction “destriping” procedure applied after the correction for atmospheric
gas absorptions in MTRDRs and TERs to mitigate the reintroduction of
The TER/MTRDR data processing pipeline photometric correction column stripe residuals [xi,λk] that arise due to inconsistencies at the
used a Lambert photometric correction, detector-element level in processing accuracy of the selected file for the
IOF 1 = IOF 0 /cos(θ) volcano scan atmospheric correction. The RSC procedure was initially
developed as a component of the TRR3 hyperspectral data filtering
where IOF0 was the observed spectral reflectance (I/F), IOF1 was procedure (Section 4.2.5; Seelos et al., 2009) and sought to isolate and
photometrically corrected spectral reflectance, and θ is solar incidence mitigate systematic calibration residuals traceable to individual detector
angle. The photometric correction was calculated for each spatial pixel elements. Residuals in the “AT” CDR used for atmospheric correction
[xi,yj] in a multiband image, and was applied uniformly to all spectral reintroduced similarly structured systematic noise. The RSC procedure
channels [λ] at that location. The spatial pixel-specific incidence angle was a band-independent operation that gathered adjacent inter-column
18
Fig. 17. Magnitude of the EGN correction for targeted observation central scans acquired at high dust opacity (FRT000064D9; Fig. 17a,b) and typical dust opacity
(FRT00003E12, Fig. 17c,d, also shown in Figs. 18, 19, and 20). Figs. 17a,c show corrections at VNIR wavelengths and Fig. 17b,d shows corrections at IR wavelengths.
Red, green and blue curves show the magnitude of EGN corrections at index VNIR and IR wavelengths in Fig. 18. Plots like these accompany every observation
delivered to the PDS as part of the TER/MTRDR data sets, in order to document the effects of processing for atmospheric aerosol correction. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)
ratio statistics from continuum-normalized spectral data and con variability attributable to the associated gimbal motion, and trans
structed a cross-track multiplicative aggregate profile from robust esti formed I/F within central scan to that at a synthetic idealized fixed data
mates (median values) of the inter-column ratios. The ratio of the cross- acquisition geometry. This procedure effectively normalized the aerosol
track aggregate profile to a low-order Legendre polynomial (typically l contribution to spectral reflectance to that at the minimum emission
= 5, m = 0) model profile fit results in a correction profile [x] for the angle (more-or-less equivalent to the MRO spacecraft roll angle off-
subject band [λk], and the accumulation of correction profiles with nadir) within the central scan.
wavelength results in a correction frame [x,λ]. This correction frame was For this correction, the CRISM spectral data were binned as a func
applied to all frames [y] in the continuum normalized multiband image, tion of cosine of emission angle [cos(e)], which is proportional to the
and the data were then transformed back to I/F. The accumulation of the atmospheric path length under a plane-parallel simplifying assumption,
cross-track aggregate profile was subject to memory (history) effects, so and the phase angle [g], which is a strong discriminator for atmospheric
the inter-column ratio statistics were collected with both a + 1 and − 1 aerosol scattering (see Fig. 14a). The [cos (e), g] sampling also captured
column shift (+y, − y), and the profile accumulation was conducted in variability due to surface photometry. The data were binned in one-
both cross-track directions (+x, − x). The average of the four resulting degree increments in phase angle [g] and an equivalent sampling fre
correction profiles was forwarded into the correction frame. The refer quency in cos (e) (90 bins over the interval [0,1]).
ence Legendre polynomial profile was determined using a least absolute Spatial pixel-specific emission angles were taken from the ‘EMA at
deviation figure of merit that reduced the influence of spurious inter- areoid’ (emission angle with respect to the Mars areoid) and ‘Phase
column ratio median samples on the model profile fit. angle’ bands in the DDRs associated with the targeted observation under
consideration, smoothed using a polynomial function as described in
Section 5.3. An abstract polynomial model of the form:
5.6. Empirical geometric normalization
IOF model = c0 + c1 cos(e) + c2 cos(e)2 + c3 g + c4 g2
The Empirical Geometric Normalization (EGN) procedure charac
terized the dependence of spectral radiance across all segments of a was fit to spectral reflectance so that the coefficients c0, c1, …, c4
∑
CRISM targeted observation (the central scan and associated EPF im minimized n(IOFdatan − IOFmodeln)2 where 1 ≤ n ≤ N for the N binned
ages) on phase and emission angle, isolated the spectral continuum data points. The sampled spectral data for each wavelength were
19
Fig. 18. Before (top) and after (bottom) images showing TER processing for VNIR (left) and IR (right) data for targeted observation FRT00003E12. Construction of
the plots is as in Fig. 14. The vertical blue, green, and red dashed lines in the spectral plots indicate bands shown in the RGB image composites. (a) VNIR data, before
correction. (b) IR data, before correction. (c) VNIR data, after correction. (d) IR data, after correction. The rise in I/F at the longest IR wavelengths is due in large part
to surface thermal emission. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
modeled independently, but high channel-to-channel correlation in forward model spectral cube was scaled by the forward model spectrum
hyperspectral data allowed model parameters for band λk to seed the at the center of the cross-track field of view [x] that included the min
optimization for band (λk + 1). imum sampled emission angle, to product a “correction cube”. The
The model parameters for each spectral channel were then used to correction cube had a value of unity at all wavelengths at the reference
calculate a central-scan image spatial pixel-specific I/F forward model minimum emission angle geometry, and a relative multiplicative
using the EMA and phase angle bands from the central scan DDR as correction at all other geometries. Data from the previous RSC pro
independent variables. The aggregated forward model bands produced a cessing step were ratioed to the scaled correction cube to complete the
forward model spectral cube that encompassed the geometrically Empirical Geometric Normalization.
dependent variation in spectral reflectance across the central scan. The In high spectral contrast scenes (e.g., observations of polar layered
20
deposit scarps), particularly those with spatially organized disparate spectral range image cube in IR sensor space.
spectral classes (e.g., having both regolith and ice within the central
scan), the empirical geometric modeling could have been unduly 5.9. TER/MTRDR data processing pipeline information maps
influenced by surface spectral variability leading to the incorporation of
erroneous spectral structure in the correction cube. To ensure that the Spatial pixel-specific performance and traceability information for
resulting correction cube tracked the spectral continuum and main processing steps was stored in a suite of dedicated processing informa
tained channel-to-channel continuity, the correction cube was spectrally tion data products that were delivered to the PDS and were described in
stabilized. The stabilization took the form of a low-order Legendre detail in the DPSIS (Murchie et al., 2022). Among the information
polynomial least absolute deviation figure of merit model that was fit as shown, spectral reflectance discontinuity at the ~1000 nm VNIR/IR
a function of wavelength, weighted by the reference forward model interface could have been inherited from the source I/F data, created as
spectrum. The resulting correction cube was also approximately unity at a byproduct of the TER/MTRDR data processing pipeline, and/or
all wavelengths at the reference minimum emission angle geometry, and resulted from the nearest-neighbor mapping of VNIR data into IR sensor
a multiplicative correction at all other geometric coordinates. space. The most common atmospheric correction residual was a high
Fig. 17 shows examples of magnitudes of fitted EGN corrections as a spectral frequency oscillation across the wavelengths affected by the
function of wavelength, for scenes with typical and high dust opacities. ~2000 nm CO2 absorption, which ultimately was traceable to small
The correction is greatest near 1.1 μm, and decreases toward both longer variations in the wavelength calibration between the observation being
and shorter wavelengths. For a dust opacity near the upper allowable corrected and the best matching reference atmospheric transmission
value for TER/MTRDR data set, the correction approaches 40% at the frame.
highest emission angles within the central scan. For more typical
opacities, it approaches 15–20%. 5.10. Spectral summary parameters
5.7. Empirical smile correction The CRISM spectral summary parameter library, as revised in 2014,
takes full advantage of hyperspectral sampling to characterize the
Spectral smile effects were addressed by the Empirical Smile diverse mineralogy identified in CRISM data (Pelkey et al., 2007;
Correction (ESC). This correction used the “WA” CDR (detector center- updated by Viviano-Beck et al., 2014). The revised library includes a
wavelength sampling map) and “SW” CDR (detector “sweet-spot” suite of 56 parameterizations of the spectral structure in TER and
wavelength vector, at the cross-track x coordinate at which the first MTRDR I/F data products. The hyperspectral implementation of each
derivative of spectral smile is zero) to transform data acquired across the summary parameter samples multiple adjacent spectral channels for a
detector to the reference (SW CDR) center wavelength sampling vector. given reference wavelength used in the band math calculation for a
The ESC procedure used the detector wavelength map to construct an given parameter, and fits the data in the sampling kernel to each target
intra-channel, cross-track [x] wavelength sampling histogram. The bin wavelength to reduce noise in the result. The dimensionality and
size was 0.125 nm for full resolution data (FRT) and 0.250 nm for half handling of the spectral sampling boxcar filter depends on the nature of
resolution data (HRL, HRS). Continuum-normalized spectral data were the spectral feature being parameterized, and is described in detail in the
binned according to the histogram map, and resulting data points were DPSIS (Murchie et al., 2022). Summary parameter calculation results
modeled as a linear function of wavelength. A forward model was were concatenated in a “SU” multiband image that was delivered to the
calculated for the detector wavelength map cross-track profile for each PDS.
spectral channel, and the reference wavelength for each channel was
used to calculate a reference model value. Each channel-specific forward 5.11. Refined spectral summary parameters
model was then normalized to the “sweet spot” reference, the resulting
correction frame [x,λ] was applied to all frames [y] in the continuum The procedures described above mitigated noise in IR data, yet noise
normalized spectral cube, and the result was transformed back to I/F. that was passed though conservative processing thresholds degraded the
The cumulative effects of the corrections described in Section 5.3 utility of those summary parameters that encode small spectral varia
through 5.7 are illustrated in Fig. 18, which compares input data (top tions. To maximize the scientific utility of summary parameters, a
row) and output data (bottom row) for the VNIR (left column) and IR custom noise remediation procedure was applied to the majority of
(right column) detectors. The most evident changes from top to bottom them. This procedure had heritage from the CRISM hyperspectral noise
are increased I/F values due to photometric correction, ratioing out of filtering procedure implemented in version 3 radiometric calibration
atmospheric gas absorptions, and disappearance of the “brightening” at (Section 4.2), but was modified to operate on individual parameter
the tops and bottoms of the spatial images due to greater atmospheric bands rather than hyperspectral I/F data. The filtering procedure iden
path lengths that enhanced the contributions of aerosol radiance. tified spurious pixels using a statistical outlier test applied to an iterative
sampling kernel. Pixels that were flagged as outliers were subsequently
5.8. VNIR to IR sensor spatial transform interpolated through using neighboring non-spurious data elements.
The parameter filtering procedure resulted in the “SR” multiband image
The CRISM VNIR and IR optical designs yielded slightly different that was delivered to the PDS.
spatial samplings. To merge data from the two detectors, a VNIR/IR
sensor space transform used the known ground location of every VNIR 5.12. Browse products
and IR spatial pixel to construct a spatial transformation that mapped
VNIR data into the IR sensor space. This transformation allowed for the Browse products are a series of 8-bit scaled RGB color composites
concatenation of VNIR and IR spectral information and the generation of that display 1- or 3-band combinations of thematically related summary
full spectral range IR sensor space (TER) and map projected (MTRDR) parameters (Viviano-Beck et al., 2014) to provide a high-level visuali
data products. The spatial transformation of the VNIR data was con zation of the information content in the source spectral image cube.
ducted as a nearest neighbor resampling to avoid spectral averaging. Example themes (e.g., Figs. 20, 25, 26) include VNIR enhanced visible
After the corrected and transformed VNIR data and corrected IR data color, IR false color, ferric mineralogy, mafic mineralogy, hydroxylated
were concatenated in IR sensor space, an image mask was generated that silicate mineralogy, and hydrated mineralogy. Summary parameters
set all the spectral channels in any spatial pixels that did not have both that are measures of the spectral continuum at a given wavelength (e.g.
VNIR and IR data to the CRISM not-a-number (NaN) value (65,535.0). R770, R1330) and those that encode spectral ratio or slope information
The result was the TER I/F data product – a fully corrected, full CRISM (e.g. ISLOPE1, IRR2) were stretched according to the scene statistics.
21
5.14. Map-projection
Final TER data products are in IR sensor space, allowing for the
calculation of a single map-projection spatial transformation that map
ped the data from both the VNIR and IR detectors into a common map-
projected space. The MTRDR data processing pipeline used the ESRI
Projection Engine available through the ENVI APIs (ENVI version 4.7+)
in the generation of a Geographic Lookup Table (GLT) that encoded this
transformation. The map projection transformation was generated in
accordance with MRO project standards, and is described in a file
delivered to the PDS as part of the MTRDR data collection. Briefly, a
rolling equirectangular projection with the reference latitude at the
equatorward edge of a 5◦ latitude increment containing the observation
was used for observations centered between − 65◦ and 65◦ N. A north or
south polar stereographic projection was used for observations poleward
of that latitude range. The pixel scale was optimized to each observation.
The GLT was applied to each TER data product (minus bad bands) to
produce a joined, map-projected VNIR+IR MTRDR data product.
Fig. 19 (compare to Figs. 18c,d) shows the map-projected “joined”
VNIR+IR MTRDR I/F data product. Fig. 20 shows several browse
products that provide an overview of the spectral content of the
FRT00003E12 central scan. These products are typical samples of the
information included with each targeted observation delivered to the
PDS as part of the MTRDR data set.
5.15. Human-in-the-loop product validation
The complexity of the TER/MTRDR processing pipeline created the

opportunity for non-intuitive failure modes that introduced either
Fig. 19. Overview of the MTRDR resulting from targeted observation spurious spatial/spectral features that make an output image unsuitable
FRT00003E12. Construction of the plot is as in Fig. 14. The vertical blue, green, for scientific interpretation, or minor artifacts that a data user should be
and red dashed lines in the spectral plots indicate bands – drawn from both the aware of. To deal with either possibility, human-in-the-loop data vali
VNIR and IR detectors – shown in the RGB image composite. (For interpretation dation was introduced. For each TER/MTRDR product set, one or two
of the references to color in this figure legend, the reader is referred to the web APL staff or interns with experience in both Mars and spectroscopy
version of this article.)
reviewed the files and approved or rejected release; only a few percent
were rejected. In many cases minor artifacts were found and are noted in
Summary parameters that are intended to quantify the presence and an Excel spreadsheet that accompanies the data collections in the PDS.
magnitude of a particular spectral feature (i.e., those with the embedded
character strings BD, INDEX, and MIN) were scaled using a stretch that is 6. Global VNIRþIR multispectral map
constrained by both the range of physically meaningful parameter cal
culations and cumulative statistics representative of the parameter The portion of the global multispectral mapping data set that was
variability for the entire data set. The stretch floor was set by the min included in the MRDR "tiles" consisted of over 51,000 individual strips of
imum meaningful value for a featureless spectrum (e.g., band depth data each typically ~550 km in length, acquired over the time from
(BD) parameters are designed to return zero for a featureless spectrum), September 2006 through late 2017. Data were collected over a wide
and the stretch ceiling was set by the scene statistics, subject to a min range of illumination conditions, atmospheric opacities, and IR detector
imum reasonable ceiling value). The ceiling value was based on cumu temperatures. This diversity presented a significant challenge to
lative statistics for each parameter band across a TER sample set mosaicking the component strips into composite image products, within
encompassing ~5% of all targeted observations, containing scenes with which within-strip noise and strip-to-strip residuals from differences in
a high degree of spectral contrast including type locations of minerals instrument state and atmospheric opacity were both much smaller than
identified in CRISM data (Viviano-Beck et al., 2014). This protocol real variations in surface spectral properties. The processing chain that
prevented saturation in scenes where parameter values are in the high was developed to accomplish all of these data processing objectives,
tail of the global distribution, yet also prevented spectral noise from resulting in version 4 MRDRs, is shown in Fig. 21. Mitigation of noise
masquerading as a meaningful detection. Generation of the TER browse and inter-strip residuals required development of several processing
product suite marked the end of the TER/MTRDR sensor space data steps that were not included in earlier deliveries of map tiles in 2009
processing pipeline. (version 1) and 2012 (version 3).
5.13. Spectral bad bands 6.1. CRISM VNIR+IR map inputs
A canonical set of spectral “bad bands” – whose I/F values are sus Inputs to the CRISM global multispectral map MRDR tiles included
ceptible to errors that are not well constrained by instrument telemetry both the 72-channel MSP mapping data, and the 262-channel HSP data
or the calibration pipeline – was identified through the evaluation subsetted to the embedded 72-channel I/F by deleting the detector rows
numerous central scan images. The TER/MTRDR “bad bands” list was in dimension λ that were outside of the core VNIR+IR 72-channel set in
stored in the TER text file of “sweet spot” wavelengths, and was applied MSP data. Each single input strip was a three dimensional multiband
to the TER I/F spectral data prior to map projection so that “bad bands” image with dimensions [x,y,λ] where individual [x,λ] frames are accu
were excluded from MTRDR I/F data products. mulated over different along-track positions [y].
22
Fig. 20. Browse versions of targeted observation FRT00003E12. See Viviano-Beck et al. (2014) for descriptions of composite bands and their interpretation. (a)
“TRU” consisting of autostretched VNIR “true color”. (b) “FAL” consisting of autostretched IR color. (c) “FM2” in which red = BD530_2 showing nanophase ferric
oxide, green = BD920_2 showing crystalline ferric oxide in lime green and low-Ca pyroxene in cyan, and blue = BDI1000VIS showing mafic minerals. (d) “PHY” in
which red = D2300 showing Fe/Mg hydroxylated silicate or Mg carbonate, green = D2210 showing Al-OH and Si-OH phases, and blue = BD1900R2 showing bound
H2O including in this case hydrated Fe-hydroxysilicates in magenta and hydrated carbonates in light blue. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)
6.2. Down-selection of mapping strips stochastic and systematic noise in two steps adapted from version 3
TRDR processing. The first step to remediate stochastic noise was a
Pre-processing of input strips calibrated to I/F consisted of two simplified version of the IKF filter used for noise remediation in I/F
further down-selections of the input data. First, only data acquired with version 3 TRDRs. The second step to remediate systematic noise was the
an IR detector temperature of ≤127.5 K are included. This was deter RSC correction also applied during TRDR filtering.
mined using the calibrated value of IR_DETECTOR_TEMP1 from the PDS The Iterative Kernel Filter (IKF) procedure as applied to a given
label on the multiband image. Second, only data from time periods wavelength [λk] of IR multispectral data was a kernel-based filtering
when τdust < 1.39 (at 900 nm from CRISM) and τice < 0.29 were algorithm that modeled the information content of a given two dimen
included. The opacities used were taken from multiannual climatology sional [x,y] normalized data kernel as a multidimensional polynomial.
of airborne dust from Martian Years 24 to 35 derived from observations The model residuals were treated as a sample set and examined for
of the Martian atmosphere from April 1999 to February 2021 by the outliers using the Grubbs test. If an outlier was detected, the corre
Thermal Emission Spectrometer (TES) aboard Mars Global Surveyor, the sponding pixel was removed from consideration and the kernel model
Thermal Emission Imaging System (THEMIS) aboard Mars Odyssey, and was iterated. Model iteration was terminated when no further outliers
the Mars Climate Sounder (MCS) aboard MRO (Montabone et al., 2015, were detected. The filtered value for the target pixel at the center of the
2020). The dust climatology is accessible at http://www-mars.lmd.jussi input kernel was then given by a proximity weighted model of the kernel
eu.fr/mars/dust_climatology/. The scaling factor in dust opacity from elements that were not marked as outliers. The confidence level
thermal IR to CRISM wavelengths is 2.0. threshold for the Grubbs test was conservative so the filter retained some
marginal noise.
The Ratio Shift Correction (RSC) procedure as applied to multi
6.3. Filtering
spectral data was the only filtering process for VNIR data and a second
filtering step for IR data. Within a given spectral band, a spatial column
The subsetted 72-channel I/F mapping strips were then filtered for
23
Fig. 21. Processing pipeline used to create version 4 MRDR multispectral map tiles.
corresponds to a single detector element. The RSC characterized sys correction from I/F to Lambert albedo with τdust = 0.2, τice = 0.0 with no
tematic calibration error of each detector element through the evalua absorptions due to atmospheric gases. The Lambert albedo also became
tion of inter-column (or cross-track) ratio statistics relative to a cross- the basis to which mapping strips acquired after May 2012 (when EPFs
track model of spectral smile. Modifying the complexity of the under were no longer acquired) were corrected.
lying cross-track model allowed the RSC procedure to address both high- For mapping data acquired after May 2012, beginning with version 4
frequency column striping and low-frequency banding while retaining of the MRDRs, a two-step correction was used that normalized out at
real scene cross-track variability. mospheric gas absorptions and corrected for photometric effects, but did
not explicitly correct for aerosol effects. This processing closely followed
that applied to TERs and MTRDRs (Section 5). The first step was the
6.4. Two alternative paths for correction of atmospheric and photometric volcano scan atmospheric correction that used an empirically derived
effects Mars atmospheric transmission spectra to correct CRISM IR spectral
reflectance data for atmospheric gas absorptions due to CO2, H2O, and
Mapping data acquired prior to May 2012 fell within the time range CO. The second step for data corrected using the volcano scan was
of acquisition of CRISM EPF data, from which retrievals provided a re application of a Lambert photometric correction for each spatial pixel
cord of τdust sampled at CRISM’s wavelength sampling and resolution as [xi,yj], applied uniformly to all spectral channels [λ]. As with TERs and
a function of latitude, longitude, Ls, and Mars year. MRO MARCI data MTRDRs, spatial pixel specific incidence angles were derived from the
acquired in the same time range provided retrievals of atmospheric ‘INA at areoid’ band (incidence angle with respect to the Mars areoid) in
water-ice opacity (τice) also as a function of latitude, longitude, Ls, and the CRISM DDR (Derived Data Record) associated with the mapping
Mars year. Where these opacities were present, using the procedures strip under consideration.
described by McGuire et al. (2008, 2013), a full radiative transfer model
was used to correct I/F to a standard reference photometric geometry (0◦
6.5. Empirical smile correction
emission angle, 0◦ incidence angle) with atmospheric gas absorptions
removed and aerosol contributions to I/F normalized to τdust = 0.2, τice
Application of the ESC correction began with version 4 MRDRs. The
= 0.0. Rather than calculations being repeated for each pixel with minor
procedures used followed exactly those used in the TER/MTRDR pro
inter-pixel differences, calculations were performed using DISORT for
cessing pipeline, with the only differences being the smaller spatial
the range of expected variables and stored as look-up tables for discrete
dimension [x] of the data that were 10×-binned in the cross-track di
values. The corrections to be applied were interpolated to the specific
rection, and the non-adjacency of the spectral bands to which the
conditions of each element in the I/F multiband image [xi,yj,λk].
correction was applied.
Information on Ls was extracted from the PDS label of the TRDR
containing the calibrated I/F data, and information on latitude, longi
tude, and altitude was extracted from the companion DDR. These data 6.6. Reconciliation of inter-strip residuals in corrected I/F and Lambert
were then used to extrapolate to the best estimate of the atmospheric albedo
opacites from lookup tables. Then the photometric angles, altitude, Ls,
and wavelength plus the dust and ice opacities extracted from the Application of this processing step began with version 4 MRDRs. At
lookup table were used to interpolate wavelength-by-wavelength to a this stage of processing there remained significant residual differences in
24
Fig. 22. Visual representation of the approach to reconciling inter-strip residuals in mapping data contained within the MRDRs.
I/F between overlapping strips of mapping data due to several effects: identified. The differences between each such pair of strips were
uncorrected errors in radiometric calibration between strips; inaccuracy analyzed using graph theory, and the best-fit gain and offset describing
in the assumption of a wavelength-independent Lambert photometric the differences were recorded. Of course, each such solution retained
function; for strips processed to Lambert albedo, differences between the some residual systematic error.
modeled and actual atmospheric dust and ice opacities; for strips pro In the optimization procedure, overall error was minimized (and the
cessed using the volcano scan correction, the presence of atmospheric data set was optimized) by applying the gains and offsets in a weighted
dust and ice opacities different than the target values of τdust = 0.2, and fashion, anchoring the output values to that part of the data set which
τice = 0.0; and differences between the actual and modeled values of was closest to ideal. Those “anchor” strips have the following attributes:
H2O vapor and CO due to seasonal or meteorological variations. The low IR detector temperature, as close as possible to the minimum value
calibration residuals were only somewhat notable in I/F, but the mag used in flight (106-110 K); low solar incidence angles; and low τdust in
nitudes were in family with real spectral variations for many of the the data as expected from the Montabone et al. (2015, 2020) dust
weaker mineralogical spectral indices represented as summary climatology. The mapping strips with such attributes have large weights
parameters. as represented by the dumbbells in Fig. 22. The relative weightings
To remediate these interstrip differences, an optimization procedure among competing overlaps in correcting an individual strip are repre
was performed in which derived values of surface reflectance were sented by the springs, with higher “spring constants” representing larger
corrected to those values in the closest to ideal data among the mapping overlaps, closer adjacencies, and smaller radiometric differences.
strips, using the millions of overlap and proximity relations among the Only mapping strips corrected to τdust = 0.2 and τice = 0.0 using
approximately >51,000 strips of VNIR+IR data in the map. The opti DISORT-generated lookup tables were used to populate the anchor
mization procedure is shown schematically in Fig. 22. Prior to optimi strips. Provided that there were overlaps and/or adjacencies with such
zation, the millions of areas of intersection and close proximity were strips, other strips corrected with the volcano scan approach effectively
Fig. 23. Mosaic of VNIR wavelengths R = 716 nm, G = 599 nm, B = 534 nm in mapping strips covering the region 17.0◦ -19.5◦ N, 76◦ -79◦ E centered on Jezero crater,
both autostretched. (a) Processed through filtering, photometric and atmospheric correction, and empirical smile correction. Residuals between strips result from the
photometric correction used and uncorrected differences in aerosol opacity. (b) Processed through inter-strip reconciliation.
25
Fig. 24. Tiling scheme for MRDRs, shown in global, equirectangular map form.
had their dust and ice opacities normalized to the values in the over space was performed using the same procedures that were employed in
lapping DISORT-traceable strips by the reconciliation procedure. Thus, the generation of TERs (Section 5.7).
correction to Lambert albedo was propagated among all strips. Resulting
data were populated in a Lambert albedo multiband image, one per tile. 6.10. Map-projection
Fig. 23 shows the effects of reconciliation in eliminating strip-to-strip
residuals at shorter VNIR wavelengths, which are most susceptible to The transformed IR mapping strips and coregistered VNIR data were
scattering by aerosols. The reconciliation step removed the vast majority propagated into the map projected tile space using IR DDR per-pixel
of the residuals; remaining inter-strip differences likely include latitude and longitude information. The map projection was generated
unmodeled photometric effects and discrete dust or ice hazes; in fact, ice in accordance with MRO project standards described in Section 5.13 for
hazes are detected directly by BD1500 in IR data. MTRDRs, using a pixel scale of 327 pixels per degree (ppd; 181 m/pixel),
(a) (b) which is a close match to native resolution with 10× spatial pixel
binning. This pixel scale avoided the aliasing of the data that occurred in
6.7. Spectral summary products v1 and v3 MRDRs that were projected at 256 ppd.
The CRISM spectral summary parameter library was revised between 6.11. Stacking, tiling, and traceability information
MRDR versions 3 and 4, and the formulae used in version 4 have been
augmented to span the diverse mineralogy identified and/or reported in The tiling scheme for MRDRs is shown in Fig. 24. Each tile has a
analyses of CRISM data (Pelkey et al., 2007; updated by Viviano-Beck dimension of 5 degrees of latitude and a longitude dimension that yields
et al., 2014). The library was used to generate a suite of parameteriza an approximately square tile based on scale at the reference latitude.
tions of the spectral structure in the MRDR Lambert albedo data prod Reference latitude for map projection is the latitude within the tile that
ucts, which were recorded in a summary product multiband image, one is nearest the equator. For version 4 MRDRs, constituent mapping strips
per tile. Unlike summary products for TERs and MTRDRs, which were were ordered so that the strips acquired under the highest atmospheric
evaluated hyperspectrally at wavelengths specified in the summary optical depth (from Montabone et al., 2015, 2020) are placed first, and
product formulas, MRDR summary products were evaluated at the de strips are then placed in order of decreasing optical depth with the strips
tector row closest to the specified wavelengths. Thus, the values acquired under the clearest atmospheric conditions “on top”. In contrast
resulting from the calculation are different for MRDRs. in the v1 and v3 tile production process, the constituent mapping strips
were ordered to place minimum solar incidence angle rather than lowest
6.8. Reconciliation of summary products inter-strip residuals atmospheric opacity “on top”.
In addition to spectral reflectance and summary product multiband
Implementation of this processing step also began with version 4 images, for each tile there is also a multiband image that provides
MRDRs. Even after reconciliation of inter-strip residuals in Lambert al traceability back to the source observation, [xi,yj] pixel location within
bedo data, very small wavelength-dependent strip-to-strip residuals it, and observing conditions. The map-projected information includes
remained in summary parameters. To remediate the residuals, the layers of the DDRs corresponding to the mapping strips, observation ID,
optimization procedure was applied a second time to summary param image segment, and [xi,yj] pixel locations.
eter strips, using the same weights and relative weightings as applied In the three MRDR versions, up to five map projected multiband
earlier to correction of strips to Lambert albedo. images were included: (a) uncorrected I/F, (b) Lambert albedo, (c)
summary products, (d) data from the VNIR and IR DDRs corresponding
6.9. VNIR to IR sensor space transform to the data strips that are part of the I/F file, and (e) data from the VNIR
and IR DDRs corresponding to the data strips that are part of the Lambert
Transformation of VNIR mapping data in sensor space to IR sensor albedo file. Version 4 MRDRs include items (b), (c), and (e).
26
Fig. 25. Browse versions of tile 1249, located in Nili Fossae near FRT00003E12. This tile covers the latitude/longitude range 17.5◦ -22.5◦ N, 70◦ -75◦ E. See Viviano-
Beck et al. (2017) for descriptions of summary and browse products and their interpretation. (a) “TRU” consisting of autostretched VNIR “true color”. (b) “FAL”
consisting of autostretched IR color. (c) “MAF” in which red = OLINDEX3 showing olivine and Fe-hydroxysilicates, green = LCPINDEX2 showing low-Ca pyroxene,
and blue = HCPINDEX2 showing high-Ca pyroxene. (d) “PHY” in which red = D2300 showing Fe/Mg hydroxylated silicate, green = D2210 showing Al-OH and Si-
OH phases, and blue = BD1900r2 showing bound H2O in this case in hydrated silicates and carbonates. (For interpretation of the references to color in this figure
legend, the reader is referred to the web version of this article.)
6.12. Browse products parameter variability for the entire data set. The stretch floor was set by
the expected minimum meaningful value for a featureless spectrum
Version 4 MRDRs include revised browse products based on updated according to the underlying band math calculation (e.g. band depth (BD)
summary parameter formulations (Viviano-Beck et al., 2014). Browse parameters are designed to return zero for a featureless spectrum), and
products are a series of up to 18 scaled RGB color composites that the stretch ceiling was set by the tile statistics, subject to a minimum
display 3-band combinations of thematically related summary products. reasonable ceiling value. That ceiling value value was based on cumu
They provide a high-level visualization of the information content in the lative statistics for each parameter band across a sample of all MRDRs.
source spectral image cube. The utility of a browse product visualization This parameter stretching protocol prevents saturation in tiles where the
is dependent on the quality of the constituent summary parameter bands parameter values are in the high tail of the global distribution (when the
and the stretch limits applied in scaling the parameter bands from 32-bit stretch ceiling is set by tile statistics), and prevents tiles that lack the
(floating point) to 8-bit (byte). Parameters that are measures of the spectral structure a given parameter is designed to isolate from
spectral continuum at a given wavelength (e.g. R770, R1330) and those masquerading spectral noise as a meaningful detection (when the stretch
that encode spectral ratio or slope information (e.g. ISLOPE1, IRR2) ceiling is set by cumulative statistics). Scaling information for each band
were stretched according to the tile statistics. Parameters that are was stored in the associated PDS label. Each resulting browse product
seeking to quantify the presence and magnitude of a particular spectral was stored as both a 3-band raw binary file (.IMG) and a Portable
feature (i.e., those with the embedded character strings BD, INDEX, and Network Graphics file (.PNG) with an alpha transparency channel cor
MIN) were scaled using a stretch that was calculated from the tile sta responding to the non-scene spatial pixels. The generation of the browse
tistics, but constrained by both the range of physically meaningful product suite marked the end of the MRDR data processing pipeline.
parameter calculations and cumulative statistics representative of the Figs. 25 and 26 show browse products that provide an overview of
27
Fig. 26. Browse versions of tile 793, located in Eos Chasma. This tile covers the latitude/longitude range 17.5◦ -12.5◦ S, 310◦ -315◦ E. See Viviano-Beck et al. (2017) for
descriptions of summary and browse products and their interpretation. (a) “TRU” consisting of autostretched VNIR “true color”. (b) “FAL” consisting of autostretched
IR color. (c) “FM2” in which red = BD530_2 showing nanophase ferric oxide, green = BD920_2 showing crystalline ferric oxide in yellow to green hues and blue =
BDI1000VIS showing mafic minerals. (d) “HYD” in which red = SINDEX2 showing hydrated sulfates, green = BD2100_2 showing monohydrated sulfates in yellow to
green hues, and blue = BD1900r2 showing bound H2O in this case in polyhydrated sulfates in magenta to blue hues (the color variation resulting for elevation of
SINDEX values by by classes of hydrated sulfate). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of
this article.)
the processing outcome for two tiles in very different geologic envi monohydrated and polyhydrated sulfates dominate different exposures.
ronments which exhibit strong spectral variations due to both primary
and secondary minerals. Tile 1249 (Fig. 24), located in Nili Fossae, ex
hibits a range of primary mafic mineral compositions and extensive 6.13. Validation and follow-up observing
alteration to phyllosilicate. Tile 793, in Eos Chasma at the eastern end of
Valles Marineris, is a region where sulfate- and ferric mineral-bearing, One concern with VNIR+IR multispectral data had been composi
mafic sand-rich layered deposits overlie mafic sediments and bedrock tional interpretability of discontinuous wavelength coverage, including
on the chasma floor. IR false color (Figs. 25b, 26b), which is relatively possible ambiguity between phases, or non-detection of phases that
insensitive to uncorrected aerosols, shows a high degree of correlation were present whose characteristic absorptions might have fallen “be
across mosaic seams whereas visible color (where ice clouds scatter tween” multispectral wavelengths. A converse concern, regarding tar
strongly), show some slight mismatches due to thin, bluish water ice geted observations, had been that their sparse spatial coverage (≤3%
clouds (Figs. 25a, 26a). (The same regions have an elevated signature in globally) could have led to misinterpretation of regional compositional
the BD1500 summary product that is diagnostic of water ice.) In trends, for example, which parts of the Noachian crust had been more
Fig. 25c, and d, the signatures of low-Ca pyroxene, high-Ca pyroxene, intensely altered to secondary minerals. Carter et al. (2013) used tar
and olivine-rich mafic materials demonstrate continuity across seams, as geted observations to infer which regions of the Noachian crust were
do signatures of phyllosilicates that exhibit differences in hydration more highly altered, and to higher metamorphic grades, as indicated by
between exposures as indicated by the range of red to magenta colors. In a variety of hydroxysilicate, carbonate, and sulfate phases. Viviano et al.
Fig. 26c and d, crystalline ferric minerals (here, hematite) indicated in (2023) used multispectral mapping data covering regions of relatively
VNIR data co-occur with hydrated sulfates indicated by IR data, and low, medium, and high degrees of alteration as inferred by Carter et al.,
to compare results regarding coverage and identification of minerals
28
7.1. VNIR hyperspectral map
The VNIR hyperspectral map is being delivered to the PDS as a series

of VNIR Hyperspectral Data Records (VRDRs), whose processing and
formatting parallels that of the MRDR VNIR+IR hyperspectral map tiles
as described in Section 6, with a number of differences described below.
The overall processing flow still remains the same:
• Input data consisted of two types of mapping strip multiband images

whose formats were comparable to MSW and HSP data respectively:
VNIR hyperspectral data in 92-channel MSV strips where all 92
channels with useful signal were included, at 90 m/pixel; and VNIR
107-channel hyperspectral data in HSV strips at 180 m/pixel.
• Pre-processing consisted of subsetting HSV strips to the 92 channels
present in MSV data.
• No atmospheric correction was applied to the data, only a correction
to Lambert albedo.
• The Ratio Shift Correction (RSC) and Empirical Smile Correction
(ESC) were applied exactly as in MRDRs.
• The Iterative Kernel Filter (IKF) correction was not applied.
• However an additional filtering step was required, the Frame Set
Correction described by Frizzell et al., 2020; Fig. 27). In 30 Hz MSV
mode, VNIR bias oscillated chaotically at sub-second to seconds time
scales. Differences in the oscillation pattern between channels of the
input multiband images were of a magnitude comparable to weaker
summary parameters, e.g. BD920, strength of the ~0.9-μm band due
Fig. 27. Frame set correction, illustrated using MSV mapping strip to low-Ca pyroxene and ferric minerals. Bracketing background/bias
MSV0003B303_01 covering the Jezero Crater delta. (a) 416-nm band displayed measurements have insufficient temporal sampling to remove this
with no corrections. The low reflectance of Mars’ surface at this wavelength source of systematic noise. The Frame Set Correction removed the
accentuates the artifact. (b) Same scene after application of the frame-set
resulting spatial row- oriented banding (in sensor space, parallel to
correction. (c) Same scene after application of all processing pipeline correc
single frames within the strip) by gathering adjacent [y] statistics
tions. (d) Ratio of corrected to uncorrected data. (e) RGB image of the fully
corrected MSV strip with R: 696 nm, G: 592 nm, and B: 520 nm.
and constructing an along-track additive aggregate profile from the
median values of the discrepancies. The profiles were accumulated
from both slightly offset along-track directions (+y and –y) and the
present. They found a high degree of correlation between inferred de
difference statistics were also collected with a + 1 and − 1 shift. The
gree of regional alteration from targeted observations and more spatially
aggregate profile was smoothed using a sliding boxcar median filter
continuous measurements from mapping data. In addition, analysis of
with a length scale of 50 frames (approximately the size of the largest
the mapping data was able to correctly identify most minerals present,
residual structures). The boxcar result was subtracted from the
within limitations of reduced ability to discriminate spectrally similar
aggregate profile to produce a correction profile for each spectral
phases: Fe vs. Mg hydroxysilicates and hydrated salts, some Fe/Mg-
band [λk]. The along-track correction was then applied to all samples
phyllosilicates from hydrated Mg‑carbonates, and silica with differing
[yj] in the input multiband image.
degrees of hydration.
• Anchor strips for the map were those acquired under conditions of
Many targeted observations were coordinated with contempora
lowest atmospheric opacity as recorded in the data set of Montabone
neous HiRISE or CTX observations to provide a full complement of im
et al. (2015, 2020). The strips included some that anchor the MRDRs,
aging and spectroscopic observations of key locations of Martian
but also incorporated strips acquired after cessation of IR data
compositional heterogeneity. Completion of version 4 MRDR map tiles,
collection. Inter-strip reconciliation proceeded as with MRDRs
which reveal mineralogically diverse outcrops that were not noticed in
except only for VNIR wavelengths.
OMEGA data or targeted observations, came after >15 years of MRO
• The only summary parameters and browse products generated were
orbital operations. To provide HiRISE observations of hundreds to
those that require only VNIR wavelengths.
thousands of secondary mineral exposures newly identified in MRDRs,
• VRDRs were constructed at a pixel scale of 654 ppd. (twice the value
hundreds of new targets have been submitted to the HiRISE observa
as for v4 MRDRs, 90 m/pixel), a close match to the native resolution
tional database and are being imaged by HiRISE as of this writing.
of MSV data. Thus resulting tiles have 2× the linear dimension of v4
MRDRs.
7. Products in development and data analysis tools
• The maps were constructed in two steps, each with higher values of
τdust laid in first, and lower values of τdust laid in on top. In the first
At the time of this writing (early 2023) two product sets are under
step, 90 m/pixel MSV strips were laid in. Then, 180 m/pixel HSV
development from MRO project funding and one from a Planetary Data
strips are laid in the in same order, but placed “in back of” the MSV
Archiving, Restoration, and Tools (PDART) grant. Four analysis tools
strips. This procedure resulted in lower spatial resolution data filling
funded by the MRO project have become available since Murchie et al.
coverage gaps in higher spatial resolution data.
(2007, 2009a) described CAT and the ground-based spectral library of
Mars analog minerals measured under Mars-like desiccating conditions.
7.2. ATOs rendered at 12 m/pixel
The tools are available online from the PDS at https://pds-geosciences.
wustl.edu/missions/mro/crism.htm. In addition, targeted observations
Spectral radiance sensed by CRISM was influenced not only by
that lack accompanying EPF observations have been corrected for at
scattering, absorption, and reflection of solar radiation by surface and
mospheric effects and delivered to the PDS, funded by PDART outside of
atmospheric components. For wavelengths longer than ~2.7 μm the
CRISM project funding.
surface also emitted thermal radiation, which was subjected to
29
Fig. 28. Sample QuickMap views of CRISM data, showing capabilities to access different data types and to query the data. (a) Targeted observations as represented
by the TAN browse products, overlain on a reduced-opacity background of TAN browse products from MRDRs. (b) One of the browse products for hydroxylated
silicates, constructed from multiple targeted observations and MRDRs, with an inset box showing a retrivied reflectance spectrum from the MRDRs. This interface has
been enabling in identifying new targets for follow-up HiRISE observations, and promises to be an enduring tool for the planetary science community.
scattering and absorption during its outgoing transit through the at DISORT was employed to model atmospheric gases and aerosols, with a
mosphere (He et al., 2022). Both thermal and solar regimes were present surface boundary condition based on the Hapke (2012) photometric
for the longer wavelengths, with an increasing contribution of thermal function. Output from the DISORT computations included surface single
radiation as the wavelength increased (e.g., Fig. 19). This mixture was a scattering albedo (SSA, the independent variable in the Hapke photo
consequence of the midafternoon observing time and relatively warm metric function) estimates for each CRISM wavelength as a function of
temperatures for CRISM observations, typically 15:00 to 15:30 LTST pixel-dependent atmospheric surface pressure, scene-dependent τdust
(local true solar time), with retrieved surface temperatures ranging up to and τice, pixel-dependent incidence, emission, and phase angles, and
290 K (He et al., 2022). pixel-dependent surface kinetic temperatures. This approach followed
Construction of rendered ATOs focused on regularization of the data work reported by Arvidson et al. (2006, 2014), Fox et al. (2016), Powell
to remediate non-uniform spatial sampling at the sub-pixel scale, et al. (2016), Kreisch et al. (2017), and He et al. (2019, 2022). For the
retrieval of the best estimate of surface spectral reflectance over the full shorter wavelengths, which lack measurable thermal emission, the
CRISM wavelength range (i.e., both VNIR and IR data) free from at retrieval of SSA for each band and pixel was a one-to-one mapping,
mospheric effects, and explicit consideration of both solar and thermal given the pixel-dependent lighting and viewing conditions, and pixel-
terms that contributed to spectral radiance on the sensor. To meet the dependent surface pressure values. For the mixed solar and thermal
objective of isolating the surface reflected component of radiance, regimes, a neural network approach was used to retrieve the surface SSA
30
values with both solar and thermal effects simultaneously modeled and published targeted MTRDR RGB browse product views, all targeted
removed (He et al., 2022). observations can be searched as color-coded footprint area layers with
Retrieval of SSA spectra for each pixel was performed in sensor an observation ID overlay. This CRISM-focused Mars QuickMap site was
space. Statistical testing of multiple CRISM scenes demonstrated that developed by Applied Coherent Technology Corporation (ACT), which is
once noise spikes were removed, what was left was Poisson-distributed a part of the CRISM Co-Investigator team.
instrument-generated stochastic noise superimposed on the true surface
SSA spectra (Kreisch et al., 2017; He et al., 2019). To remove the noise, 7.6. MICA files
an iterative log maximum likelihood approach was used to retrieve the
best estimate of the surface SSA spectra in the presence of Poisson- Minerals Identified through CRISM Analysis (MICA Files) is a spec
distributed noise for each pixel. The processing also generated map- tral library of type examples of minerals identified in CRISM data, as
projected-SSA multiband images at 12 m/pixel, taking into account measured at Mars, accompanied by terrestrial analog mineral spectra
the overlap of the data associated with the along-track oversampling, measured in laboratories (Viviano-Beck et al., 2014). An accompanying
that enables a form of “super resolution” processing. Regularization of document with descriptions of the type locations and CRISM and HiRISE
the data from sensor to projected space included inversion of both the images of the locations is available from the CRISM team web site at
spatial and spectral transfer functions to sharpen the data, while http://crism.jhuapl.edu/data/mica/
removing Poisson-distributed noise using penalties for anomalous
excursions. 7.7. TERs and MTRDRs for Post-2012 observations without EPFs
CRISM team members and collaborators from Washington University
in Saint Louis processed and delivered to the PDS 303 VNIR ATO scenes The aforementioned work was completed under the auspices of
as I/F values that have been processed through the log maximum like funding by the MRO project. Extension of TER/MTRDR processing to
lihood procedures. The scenes were selected based on scientific content post-2012, EPF-less targeted observations was sponsored by a NASA
and a high degree of spatial overlap that allows processing to 12 m/pixel PDART grant.
map projected data sets with sharpened spatial features relative to other The EGN correction is a key component of the TER/MTRDR data
processing methods. In addition, ten ATO scenes were processed to processing pipeline which characterizes and corrects wavelength-
single scattering albedo values for the full CRISM wavelength region. dependent, along-track radiance gradients present in central scan
The S and L data were also co-registered to produce integrated S and L multiband images due to gimbal motion and the resulting continuously
spectral cubes. The scenes were selected after careful scrutiny of the full variable observation geometry. The EGN correction applied to pre-2012
S and L ATO scenes. The selected scenes are those in which spectral data was fitted to the accompanying EPF images. In mid-2012, acqui
evidence is present in the data for minerals produced and/or altered in sition of accompanying EPF images was suspended due to aging of the
the presence of water. gimbal and the onset of angular range restrictions. To overcome this
limitation, a modified “proxy” EGN correction was developed that does
7.3. JCAT not rely on accompanying EPF images, and allows for the production of
TER/MTRDR data product suites for the restricted gimbal range later-
The CAT tool requires an ENVI license to operate. To provide a free mission targeted observations. This was applied to the highest-value
tool that shared CAT capability to access, display, process, and plot re data containing both VNIR and IR wavelengths.
sults from CRISM data products, CRISM SOC team members developed The pre-2012 EGN procedure (Section 5.6) approximated the
and released the Java-based tool JCAT. JCAT allows users to perform gradient of atmospheric radiance with a polynomial function of two
basic processing and analysis tasks including view and stretch multiband observation angles, the cosine of emission angle (proportional to at
images, perform I/F comparisons, alter wavelength sets for RGB images, mospheric path length) and phase angle (dependent on aerosol scat
plot spectra, save data and images, and perform atmospheric and tering). Due to the lack of EPF images after mid-2012, it is difficult to
photometric corrections. It works either on TRDRs or advanced products directly estimate the underlying intensity gradient caused by the varying
described above. Along with the source code and required CDRs, a user observation angles. Instead, a proxy EGN correction transferred gradient
tutorial is downloadable from the PDS and at http://crism.jhuapl. model obtained from early-mission observations to the gimbal-restricted
edu/JCAT/. later-mission observations was used. The successful transfer of the EGN
gradient model required identification of an earlier observation whose
7.4. MICA files gradient model approximated that of a given later-mission EPF-free
observation. Because the gradient was caused mainly by atmospheric
Minerals Identified through CRISM Analysis (MICA Files) is a spec interactions, observations with a transferrable model should have been
tral library of type examples of minerals identified in CRISM data, as collected under atmospheric conditions close to those of the scene of
measured at Mars, accompanied by terrestrial analog mineral spectra interest. In particular, the scattering characteristics of the atmosphere
measured in laboratories (Viviano-Beck et al., 2014). An accompanying significantly affected the shape of the gradient. Radiance typically
document with descriptions of the type locations and CRISM and HiRISE increased with emission angle, implying that the gradient was strongly
images of the locations remains available from the CRISM team web site correlated with atmospheric path length and with the amount of at
at http://crism.jhuapl.edu/data/mica/ mospheric scattering. In addition, the gradient models could be trans
ferred more successfully to the scene of interest if they were obtained at
7.5. QuickMap web Interface to the CRISM data a similar observation geometry (incidence, emission, and phase angles).
Additional observation attributes such as geographic proximity and
Mars QuickMap (http://mars.quickmap.io/; Fig. 28) provides an seasonal similarity also informed the proxy identification.
easy to use yet powerful web GIS interface to view CRISM map products. The set of early-mission observations was down-selected based on
At the time of this writing, it is being upgraded to include CRISM End-of- the complexity of the scenes, to retain only images that are sufficiently
Mission data products. Thematic RGB browse products for the global uniform to allow accurate modeling of the EGN gradient. 2589 early-
multispectral survey (MRDR) and over 9000 higher-resolution targeted mission FRT and HRL reference observations were sufficiently
products (MTRDRs) can be layered together with many other published spatially uniform to support the calculation of an accurate reference
Mars datasets such as HiRISE and CTX images. Also, it can provide gradient model for all wavelength bands of both VNIR and IR multiband
spectral plots and downloadable data values for user-entered points images. To ensure that there were similar scattering properties between
based on the MRDR multispectral global map. In addition to the the target scene and a candidate scene, the shortest wavelength bands
31
Fig. 29. Comparison of the RGB composite images of FRS00031524 VNIR (left; R: band 37 (599 nm), G: band 26 (527 nm), B: band 13 (443 nm) and IR (right; R:
band 206 (2530 nm), G: band 351 (1572 nm), B: band 426 (1080 nm)) data before (bottom) and after (top) the proxy EGN correction. The columns on the left of the
VNIR and IR RGB images are the boxplot of the intensity statistics of each RGB band image before (left) and after (right) the correction.
(410–440 nm) of the VNIR image cube were inspected. These bands Declaration of Competing Interest
contain the most distinct gradients across the CRISM wavelength range
even in topographically and mineralogically complex scenes due to The authors declare that they have no known competing financial
strong atmospheric scattering and minimal surface spectral contrast. interests or personal relationships that could have appeared to influence
Thanks to these properties, the shape of these short-wavelength gradi the work reported in this paper.
ents could be used to indirectly assess the scattering characteristics of
the atmosphere. Data availability
Model gradients were obtained by a tailored convex optimization
that operated on the short- wavelength bands of EPF-free later-mission Data shared in thee PDS
observations. The optimization problem learned underlying vertical
(along-track) variation/curvature commonly present along the image Acknowledgements
columns that largely corresponded to the gradient. Polynomial fitting to
the uniform observations was performed on the gradient model to This work was funded by Jet Propulsion Laboratory subcontract
minimize the effect of surface variability and to accurately fit the geo 1277793 to the Johns Hopkins University Applied Physics Laboratory
metric dependence. for conduct of the CRISM Investigation, and NASA Planetary Data
The similarity in observation geometry was computed using pixel- Archiving, Restoration, and Tools (PDART) grant 80NSSC19K0418. The
dependent observation angles stored in DDRs. Similarity in incidence CRISM team gratefully acknowledges support from the MRO Project and
angle was measured by the scene-average as it varies only slowly over a the PDS Geosciences Node.
scene (typically by less than one degree). The configuration of emission
and phase angles was represented by a vectorized profile of the pair of References
angles along the boresight pixel in each component frame of a multiband
image, as these angles changed primarily with gimbal motion. In order Amador, E.S., Bandfield, J.L., Thomas, N.H., 2018. A search for minerals associated with
serpentinization across Mars using CRISM spectral data. Icarus 311, 113–134.
to compare geometric configurations between scenes, a candidate pro https://doi.org/10.1016/j.icarus.2018.03.021.
file was compared with a reference scene profile based on the number of Arvidson, R.E., Poulet, F., Morris, R.V., Bibring, J.P., Bell, J.F., Squyres, S.W., et al.,
shared emission angle samples, and on the consistency of the phase 2006. Nature and origin of the hematite-bearing plains of Terra Meridiani based on
analyses of orbital and Mars Exploration rover data sets. J. Geophys. Res. 111,
angles associated with the shared emission angles. E12S08. https://doi.org/10.1029/2006JE002728.
Once a closely matching reference observation was identified, its Arvidson, R.E., et al., 2014. Ancient aqueous environments at Endeavour Crater, Mars.
gradient model for band was applied to the scene of interest The refer Science 343, 1248097. https://doi.org/10.1126/science.1248097.
Beck, A.W., Murchie, S.L., Viviano, C.E., 2020. A search for early- to mid-Noachian
ence model was scaled with a single wavelength-independent factor chloride-rich deposits on Mars. Icarus 338, 113552. https://doi.org/10.1016/j.
obtained from the fit of the gradient model to the reference short icarus.2019.113552.
wavelength band images. The gradient was is then shifted to have zero Bell III, J.F., et al., 2009. Mars reconnaissance orbiter Mars color imager (MARCI):
instrument description, calibration, and performance. J. Geophys. Res. 114, E08S92
value at the smallest emission angle in the scene, and the correction was
https://doi.org/10.1029/2008JE003315.
subtracted from the image of interest. Bell III, J.F., et al., 2019. In: Bishop, J., et al. (Eds.), Compositional and Mineralogic
The new proxy EGN algorithm was applied to post-May 2012 Analyses of Mars Using Multispectral Imaging on the Mars Exploration Rover,
restricted gimbal range targeted observations including FRS, ATO, and Phoenix, and Mars Science Laboratory Missions. Cambridge Univ., New York,
pp. 513–537.
ATU class types that do not have accompanying EPF measurements. Bibring, J.-P., et al., 2004. OMEGA: Observatoire pour la Minéralogie, l’Eau, les Glaces et
Fig. 29 shows a comparison of RGB composite images for VNIR and IR l’Activité. Mars Express: the scientific payload. In: Wilson, Andrew (Ed.), ESA SP-
image cubes for the scene FRS00031524 before and after the correction 1240. ESA Publications Division, Noordwijk, Netherlands, pp. 37–49.
Bishop, J.L., Rampe, E.B., 2016. Evidence for a changing Martian climate from the
was applied. The proxy EGN method successfully removed and flattened mineralogy at Mawrth Vallis. Earth Planet. Sci. Lett. 448, 42–48. https://doi.org/
the vertical gradient (brighter toward the top of the image) in both data 10.1016/j.epsl.2016.04.031.
sets. The distribution of the image intensity shown in the RGB boxplots Bishop, J.L., Fairén, A.G., Michalski, J.R., Gago-Duport, L., Baker, L.L., Velbel, M.A.,
Gross, C., Rampe, E.B., 2018. Surface clay formation during short-term warmer and
confirms the mitigation of geometric dependence. wetter conditions on a largely cold ancient Mars. Nature Astron. 2, 206–213. https://
The proxy EGN correction was integrated into the larger TER/ doi.org/10.1038/s41550-017-0377-9.
MTRDR data processing workflow to support the production, review, Blake, D., et al., 2012. Characterization and calibration of the CheMin mineralogical
instrument on Mars science laboratory. Spaced. Scci. Rev. 170, 341–0399. https://
and Planetary Data System (PDS) delivery of the TER/MTRDR data doi.org/10.1007/s11214-012-9905-1.
product suites for the post-May 2012 EPF-free CRISM hyperspectral Brown, A.J., Calvin, W.M., Murchie, S.L., 2012. Compact Reconnaissance Imaging
targeted observations. Spectrometer for Mars (CRISM) north polar springtime recession mapping: First 3
32
Mars years of observations. J. Geophys. Res. 117, E00J20 https://doi.org/10.1029/ Kirkland, L, Salisbury, J., Mustard, J., Clark, R., P. Lucey, P., Murchie, S., 1999.
2012JE004113. Spectroscopy of the Martian surface: What next?, LPI Contrib. 1149, Lunar and
Brown, A.J., Piqueux, S., Titus, T.N., 2014. Interannual observations and quantification Planet. Inst., Houston, Tex.
of summertime H2O ice deposition on the Martian CO2 ice south polar cap. Earth Kreisch, C.D., O’Sullivan, J.A., Arvidson, R.E., Politte, D.V., He, L., Stein, N.T., et al.,
Planet. Sci. Lett. 406, 102–109. https://doi.org/10.1016/j.epsl.2014.08.039. 2017. Regularization of Mars reconnaissance orbiter CRISM along-track
Buczkowski, D.L., Seelos, K.D., Viviano, C.E., Murchie, S.L., Seelos, F.P., Malaret, E., oversampled hyperspectral imaging observations of Mars. Icarus 282, 136–151.
Hash, C., 2020. Anomalous phyllosilicate-bearing outcrops south of Coprates https://doi.org/10.1016/j.icarus.2016.09.033.
Chasma: a study of possible emplacement mechanisms. J. Geophys. Res. Plan. 125, Langevin, Y., Poulet, F., Bibring, J.-P., Gondet, B., 2005. Sulfates in the North Polar
e2019JE006043 https://doi.org/10.1029/2019JE006043. Region of Mars Detected by OMEGA/Mars Express. Science 307, 1584–1586.
Carter, J., Poulet, F., Bibring, J.-P., Mangold, N., Murchie, S., 2013. Hydrous minerals on https://doi.org/10.1126/science.1109091.
Mars as seen by the CRISM and OMEGA imaging spectrometers: updated global Lapôtre, M.G.A., Ehlmann, B.L., Minson, S.E., 2017. A probabilistic approach to remote
view. J. Geophys. Res. Plan. 118, 831–858. https://doi.org/10.1029/ compositional analysis of planetary surfaces. J. Geophys. Res. Plan. 122, 983–1009.
2012JE004145. https://doi.org/10.1002/2016JE005248.
Carter, J., Loizeau, D., Mangold, N., Poulet, F., nBibring, J.-P., 2015. Widespread surface Leask, E.K., Ehlmann, B.L., 2022. Evidence for deposition of chloride on Mars from small-
weathering on early Mars: a case for a warmer and wetter climate. Icarus 248, volume surface water events into the late Hesperian-early Amazonian. AGU Adv. 3,
373–382. https://doi.org/10.1016/j.icarus.2014.11.011. e2021AV000534 https://doi.org/10.1029/2021AV000534.
Christensen, P.R., et al., 2001. Mars global surveyor thermal emission spectrometer Leask, E.K., Ehlmann, B.L., Dundar, M.M., Murchie, S.L., Seelos, F.P., 2018. Challenges in
experiment: investigation description and surface science results. J. Geophys. Res. the search for perchlorate and other hydrated minerals with 2.1-μm absorptions on
106, 23823–23871. https://doi.org/10.1029/2000JE001370. Mars. Geophys. Res. Lett. 45, 12180–12189. https://doi.org/10.1029/
Christensen, P.R., et al., 2004. The Thermal Emission Imaging System (THEMIS) for the 2018GL080077.
Mars 2001 Odyssey mission. Space Sci. Rev. 110, 85–130. https://doi.org/10.1023/ Malin, M.C., et al., 2007. Context camera investigation on board the Mars reconnaissance
B:SPAC.0000021008.16305.94. orbiter. J. Geophys. Res. 112, E05S04. https://doi.org/10.1029/2006JE002808.
Christian, J.R., Arvidson, R.E., O’Sullivan, J.A., Vasavada, A.R., Weitz, C.M., 2022. McEwen, A.S., et al., 2007. Mars Reconnaissance Orbiter’s High Resolution Imaging
CRISM-based high spatial resolution thermal inertia mapping along Curiosity’s Science Experiment (HiRISE). J. Geophys. Res. 112, E05S02. https://doi.org/
traverses in Gale crater. J. Geophys. Res. Plan. 127, e2021JE007076 https://doi.org/ 10.1029/2005JE002605.
10.1029/2021JE007076. McGuire, P.C., et al., 2008. MRO/CRISM retrieval of surface Lambert albedos for
Clancy, R.T., Wolff, M.J., Smith, M.D., Kleinböhl, A., Cantor, B.A., Murchie, S.L., multispectral mapping of mars with DISORT-based radiative transfer modeling:
Toigo, A.D., Seelos, K., Lefèvre, F., Montmessin, F., Daerden, F., Sandor, B.J., 2019. phase 1 – using historical climatology for temperatures, aerosol optical depths, and
The distribution, composition, and particle properties of Mars mesospheric aerosols: atmospheric pressures. Trans. Geosci. Rem. Sens. 46, 4020–4040.
an analysis of CRISM visible/near-IR limb spectra with context from near-coincident McGuire, P.C., et al., 2013. Mapping Minerals on Mars with CRISM: Atmospheric and
MCS and MARCI observations. Icarus 328, 246–273. https://doi.org/10.1016/j. photometric correction for MRDR map tiles, version 2, and comparison to OMEGA.
icarus.2019.03.025. In: 44th Lunar Planet. Sci. 44, abstract #1581.
Das, E., Mustard, J.F., Tarnas, J.D., Pascuzzo, A.C., Kremer, C.H., 2022. Investigating the Montabone, L., et al., 2015. Eight-year climatology of dust optical depth on Mars. Icarus
origin of gypsum in Olympia Undae: characterizing the mineralogy of the basal unit. 251, 65–95. https://doi.org/10.1016/j.icarus.2014.12.034.
Icarus 372, 114720. https://doi.org/10.1016/j.icarus.2021.114720. Montabone, L., et al., 2020. Martian year 34 column dust climatology from Mars climate
Ehlmann, B.L., Buz, J., 2015. Mineralogy and fluvial history of the watersheds of Gale, sounder observations: reconstructed maps and model simulations. J. Geophys. Res.
Knobel, and Sharp craters: A regional context for the Mars Science Laboratory Planets 125, e2019JE006111. https://doi.org/10.1029/2019JE006111.
Curiosity’s exploration. Geophys. Res. Lett. 42, 264–273. https://doi.org/10.1002/ Moreland, E.L., Arvidson, R.E., Morris, R.V., Condus, T., Hughes, M.N., Weitz, C.M.,
2014GL062553. VanBommel, S.J., 2022. Orbital and in-situ investigation of the Bagnold dunes and
Ehlmann, B.L., Edwards, C.S., 2014. Mineralogy of the Martian surface. Annu. Rev. Earth Sands of Forvie, Gale crater, Mars. J. Geophys. Res. Planets. https://doi.org/
Planet. Sci. 42, 291–315. https://doi.org/10.1146/annurev-earth-060313-055024. 10.1029/2022JE007436.
Ehlmann, B.L., Mustard, J., Murchie, S., Poulet, F., Bishop, J., Brown, A., Calvin, W., Morgan, F., Mustard, J.F., Wiseman, S.M., Seelos, F.P., Murchie, S.L., McGuire, P.C.,
Clark, R., Des Marais, D., Milliken, R., Roach, L., Roush, T., Swayze, G., Wray, J., CRISM Team, 2011. Improved algorithm for CRISM volcano scan atmospheric
2008. Orbital identification of carbonate-bearing rocks on Mars. Science 322, correction. Lunar Planet. Sci. 42 abstract #2453.
1828–1832. https://doi.org/10.1126/science.1164759. Murchie, S., et al., 2007. Compact Reconnaissance Imaging Spectrometer for Mars
Fox, V.K., Arvidson, R.E., Guinness, E.A., McLennan, S.M., Catalano, J.G., Murchie, S.L., (CRISM) on Mars Reconnaissance Orbiter (MRO). J. Geophys. Res. 112, E05S03.
Powell, K.E., 2016. Smectite deposits in Marathon Valley, Endeavour crater, Mars, https://doi.org/10.1029/2006JE002682.
identified using CRISM hyperspectral reflectance data. Geophys. Res. Lett. 43, Murchie, S.L., et al., 2009a. The CRISM investigation and data set from the Mars
4885–4892. https://doi.org/10.1002/2016GL069108. Reconnaissance Orbiter’s Primary Science Phase. J. Geophys. Res. 114, E00D07.
Fraeman, A.A., et al., 2013. A hematite-bearing layer in Gale crater, Mars: mapping and https://doi.org/10.1029/2009JE003344.
implications for past aqueous conditions. Geology 41, 1103–1106. https://doi.org/ Murchie, S., et al., 2009b. A synthesis of Martian aqueous mineralogy after 1 Mars year
10.1130/G34613.1. of observations from the Mars reconnaissance orbiter. J. Geophys. Res. 114, E00D06.
Fraeman, A.A., Ehlmann, B.L., Arvidson, R.E., Edwards, C.S., Grotzinger, J.P., https://doi.org/10.1029/2009JE003342.
Milliken, R.E., Quinn, D.P., Rice, M.S., 2016. The stratigraphy and evolution of lower Murchie, S.L., et al., 2019. VSWIR spectral analyses of Mars using CRISM & OMEGA. In:
Mount Sharp from spectral, morphological, and thermophysical orbital data sets. Bishop, J., et al. (Eds.), Remote Compositional Analysis. Cambridge Univ, New York,
J. Geophys. Res. Planets 121, 1713–1736. https://doi.org/10.1002/2016JE005095. pp. 453–483.
Fraeman, A.A., Johnson, J.R., Arvidson, R.E., Rice, M.S., Wellington, D.F., Morris, R.V., Murchie, S.L., et al., 2022. Mars Reconnaissance Orbiter CRISM Data Product Software
et al., 2020. Synergistic ground and orbital observations of iron oxides on Mt. Sharp Interface Specification. Available online at. https://pds-geosciences.wustl.edu/mro
and Vera Rubin ridge. J. Geophys. Res. Plan. 125, e2019JE006294 https://doi.org/ /mro-m-crism-2-edr-v1/mrocr_0001/document/crism_dpsis.pdf.
10.1029/2019JE006294. Pan, L., Ehlmann, B.L., Carter, J., Ernst, C.M., 2017. The stratigraphy and history of
Frizzell, K.R., Seelos, F.P., Humm, D.C., Murchie, S.L., Hash, C.D., 2020. Implementation Mars’ northern lowlands through mineralogy of impact craters: a comprehensive
of a residual correction in the MRO CRISM VNIR mapping strip data processing survey. J. Geophys. Res. Planets 122. https://doi.org/10.1002/2017JE005276.
pipeline. 51st. Lunar Planet. Sci. 51 abstract #2377. Pan, L., Carter, J., Quantin-Nataf, C., Pineau, M., Chauviré, B., Mangold, N., Le Deit, L.,
Goudge, T.A., Milliken, R.E., Head, J.W., Mustard, J.F., Fassett, C.I., 2017. Rondeau, B., Chevrier, V., 2021. Voluminous silica precipitated from Martian waters
Sedimentological evidence for a deltaic origin of the western fan deposit in Jezero during late-stage aqueous alteration. Plan. Sci. J. 2, 65. https://doi.org/10.3847/
crater, Mars and implications for future exploration. Earth Planet. Sci. Lett. 458, PSJ/abe541.
357–365. Pelkey, S.M., Mustard, J.F., Murchie, S., Clancy, R.T., Wolff, M., Smith, M., Milliken, R.,
Hapke, B., 2012. Bidirectional reflectance spectroscopy 7. The single particle phase Bibring, J.-P., Gendrin, A., Poulet, F., Langevin, Y., Gondet, B., 2007. CRISM
function hockey stick relation. Icarus 221, 1079–1083. https://doi.org/10.1016/j. multispectral summary products: parameterizing mineral diversity on Mars from
icarus.2012.10.022. reflectance. J. Geophys. Res. 112, E08S14. https://doi.org/10.1029/2006JE002831.
He, L., O’Sullivan, J.A., Politte, D.V., Powell, K.E., Arvidson, R.E., 2019. Quantitative Phillips, M.S., Viviano, C.E., Moersch, J.E., Rogers, A.D., McSween, H.Y., Seelos, F.P.,
reconstruction and denoising method HyBER for hyperspectral image data and its 2022. Extensive and ancient feldspathic crust detected across north Hellas rim, Mars:
application to CRISM. IEEE J. Select. Top. Appl. Earth Observ. Rem. Sens. 12, Possible implications for primary crust formation. Geology. https://doi.org/
1219–1230. https://doi.org/10.1109/JSTARS.2019.2900644. 10.1130/G50341.1 published online.
He, L., Arvidson, R.E., O’Sullivan, J.A., Morris, R.V., Condus, T., Hughes, M.N., Powell, K.E., Arvidson, R.E., Zanetti, M., Guinness, E.A., Murchie, S.L., 2016. The
Powell, K.E., 2022. Surface kinetic temperatures and nontronite single scattering structural, stratigraphic, and paleoenvironmental record exposed on the rim and
albedo spectra from Mars Reconnaissance Orbiter CRISM hyperspectral imaging data walls of Iazu crater, Mars. J. Geophys. Res. Planets 122, 1138–1156. https://doi.org/
over Glen Torridon, Gale Crater, Mars. J. Geophys. Res. Plan. 127, 1–26. https://doi. 10.1002/2016JE005196.
org/10.1029/2021JE007092. Scheller, E.L., Ehlmann, B.L., 2020. Composition, stratigraphy, and geological history of
Holmes, J.A., Lewis, S.R., Patel, Manish R., 2015. Analysing the consistency of martian the Noachian basement surrounding the Isidis impact basin. J. Geophys. Res. Plan.
methane observations by investigation of global methane transport. Icarus 257, 125, e2019JE006190 https://doi.org/10.1029/2019JE006190.
23–32. https://doi.org/10.1016/j.icarus.2015.04.027. Seelos, F.P., Parente, M., Clark, T., Morgan, F., Barnouin-Jha, O.S., McGovern, A.,
Horgan, B.H.N., Anderson, R.B., Dromart, G., Amador, E.S., Rice, M.S., 2020. The Murchie, S.L., Taylor, H., 2009. CRISM hyperspectral data filtering with application
mineral diversity of Jezero crater: evidence for possible lacustrine carbonates on to MSL landing site selection. AGU Fall Meeting Abstr. 23, 1234.
Mars. Icarus 339, 113526. https://doi.org/10.1016/j.icarus.2019.113526.
33
Seelos, F.P., Murchie, S.L., Humm, D.C., Barnouin, O.S., Morgan, F., Taylor, H.W., MRO/CRISM reflectance spectra. Icarus 328, 274–286. https://doi.org/10.1016/j.
Hash, C.D., 2011. CRISM data processing and analysis products update — icarus.2019.03.001.
Calibration, correction, and visualization. Lunar Planet. Sci. 42 abstract #1438. Viviano, C.E., Beck, A.W., Murchie, S.L., Dapremont, A.M., Seelos, F.P., 2023.
Silverglate, P.R., Fort, D.E., 2004. System design of the CRISM (Compact Reconnaissance Heterogeneity of the Noachian crust of Mars using CRISM multispectral mapping
Imaging Spectrometer for Mars) hyperspectral imager. In: Imaging Spectrometry IX, data. Geophys. Res. Lett. 50, e2022GL102711 https://doi.org/10.1029/
edited by S. S. Shen and P. E. Lewis, Proc. SPIE Int. Soc. Opt. Eng., 5159, 2022GL102711.
pp. 283–290. Viviano-Beck, C.E., Murchie, S.L., Beck, A.W., Dohm, J.M., 2017. Compositional and
Sinha, P., Horgan, B., 2022. Sediments within the icy north polar deposits of Mars record structural constraints on the geologic history of eastern Tharsis rise, Mars. Icarus
recent impacts and volcanism. Geophys. Res. Lett. 49 https://doi.org/10.1029/ 284, 43–58. https://doi.org/10.1016/j.icarus.2016.09.005.
2022GL097758 e2022GL097758. Viviano-Beck, C.E., Seelos, F.P., Murchie, S.L., Kahn, E.G., Seelos, K.D., Taylor, H.W.,
Smith, I.B., Viviano, C., Chojnacki, M., Putzig, N.E., Quantin, C., Rodriguez, J.A.P., 2019. Taylor, K., Ehlmann, B.L., Wisemann, S.M., Mustard, J.F., Morgan, M.F., 2014.
Characterization of hydrated, layered deposits at Valles Marineris plateau, a Revised CRISM spectral parameters and summary products based on the currently
multidisciplinary approach. Lunar Planet. Sci. 50 abstract #2713. detected mineral diversity on Mars. J. Geophys. Res. Planets 119, 1403–1431.
Stamnes, K., Tsay, S., Wiscombe, W., Jayaweera, K., 1988. Numerically stable algorithm https://doi.org/10.1002/2014JE004627.
for discrete-ordinate-method radiative transfer in multiple scattering and emitting Wolff, M.J., Smith, M.D., Clancy, R.T., Arvidson, R.E., Kahre, M., Seelos, F.P.,
layered media. Appl. Opt. 27, 2502. Murchie, S., Savijärvi, H., 2009. Wavelength dependence of dust aerosol single
Sun, V.Z., Milliken, R.E., 2018. Distinct geologic settings of opal-a and more crystalline scattering albedo as observed by the compact reconnaissance imaging spectrometer.
hydrated silica on Mars. Geophys. Res. Lett. 45, 10221–10228. https://doi.org/ J. Geophys. Res. Planets 114, 1–17. https://doi.org/10.1029/2009JE003350.
10.1029/2018GL078494. Wray, J.J., Murchie, S.L., Bishop, J.L., Ehlmann, B.L., Milliken, R.E., Wilhelm, M.B.,
Tarnas, J.D., Mustard, J.F., Lin, H., Goudge, T.A., Amador, E.S., Bramble, M.S., et al., Seelos, K.D., Chojnacki, M., 2016. Orbital evidence for more widespread carbonate-
2019. Orbital identification of hydrated silica in Jezero crater, Mars. Geophys. Res. bearing rocks on Mars. J. Geophys. Res. Planets 121, 652–677. https://doi.org/
Lett. 46, 12771–12782. https://doi.org/10.1029/2019GL085584. 10.1002/2015JE004972.
Viviano, C.E., Murchie, S.L., Daubar, I.J., Morgan, M.F., Seelos, F.P., Plescia, J.B., 2019. Zurek, R.W., Smrekar, S.E., 2007. An overview of the Mars reconnaissance orbiter (MRO)
Composition of Amazonian volcanic materials in Tharsis and Elysium, Mars, from science mission. J. Geophys. Res. 112, E05S01 https://doi.org/10.1029/
2006JE002701.
34

CRISM

Uploaded by

Copyright:

Available Formats

CRISM

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CRISM

Uploaded by

Copyright:

Available Formats

Icarus 419 (2024) 115612

Contents lists available at ScienceDirect

The CRISM investigation in Mars orbit: Overview, history, and delivered

Fig. 1. a. Photographs of the CRISM OSU, DPU, GME.

Table 2 Table 2 (continued )

Fig. 3. Elements to a CRISM targeted observation, shown in a Mars-centered reference frame.

sequence was repeated outgoing, except in reverse order (part 3 in

Fig. 8. Map projected views of second-generation targeted observations. RGB

first pivoting to +60◦ , looking ahead of spacecraft nadir (reduced to

Fig. 9. Timeline of observations intended primarily for atmospheric characterization.

Fig. 11. Timeline of mapping observations.

viewing periods were constrained to be in blocks of 3 min so that Table 3

Fig. 14. (continued).

4.2.3. Reference cube 4.2.5. Ratio Shift Correction (RSC)

The Ratio Shift Correction (RSC) was a spatial column-oriented

5.15. Human-in-the-loop product validation

The complexity of the TER/MTRDR processing pipeline created the

5.13. Spectral bad bands 6.1. CRISM VNIR+IR map inputs

7.1. VNIR hyperspectral map

The VNIR hyperspectral map is being delivered to the PDS as a series

• Input data consisted of two types of mapping strip multiband images

You might also like