Data Set Overview
    =================

      This volume contains portions of the CRISM Targeted Reduced Data
      Record (TRDR) Archive, a collection of multiband images from the
      Compact Reconnaissance Imaging Spectrometer for Mars on the Mars
      Reconnaissance Orbiter spacecraft.  Images consist of
      data calibrated to units of radiance or I/F plus a text file with
      housekeeping information, and optionally image data to which
      further corrections have been applied. Image data are in sensor
      space and non-map-projected. The data are stored with PDS labels.

      For more information on the contents and organization of the
      volume set refer to the aareadme.txt file located in the root
      directory of the data volumes.


    Parameters
    ==========

      CRISM observing scenarios are constructed using a set of key
      variables ('configurations') which include the following. (All
      are selectable separately for the VNIR and IR detectors. Only a
      subset of the configurations represent 'scene' data, as indicated
      by the keyword MRO:ACTIVITY_ID. Only scene data that are aimed
      at planetary objects are processed to TRDRs. Only those
      configurations that affect the contents or dimensionality of a
      TRDR are discussed below):

      Image source: Image data may be generated using digitized output
      from the detector, or using one of up to seven test patterns. Only
      data from the detector are processed to a TRDR.

      Pixel binning: Pixels can be saved unbinned or binned 2x, 5x, or
      10x in the spatial direction. No pixel binning in the spectral
      direction is supported. Data with any pixel binning
      configuration may have a corresponding TRDR, but the pixel
      binning configuration will affect the dimensionality of the
      TRDR.

      Row selection: Detector row selection. All detector rows,
      sampling different wavelengths, having useful signal can be
      saved. Alternatively an arbitrary, commandable subset of rows can
      be saved. The number of rows with useful signal is 545, 107 in
      the VNIR and 438 in the IR, and these subsets of VNIR and IR
      detector rows are used for hyperspectral observing. Prior to
      10 Dec 2006 the nominal number of rows for multispectral mode
      was 73, 18 in the VNIR and 55 in the IR. On 10 Dec 2006 an
      extra channel was added to the VNIR for calibration purposes,
      for a total of 19. For each detector, there are four options
      of channel selection to choose from rapidly by command:
      hyperspectral (545 total channels), multispectral (73 total
      channels prior to 10 Dec 2006, 74 total channels on and after
      10 Dec 2006), and two sets of expanded multispectral (84 and 92
      channels prior to 10 Dec 2006, 85 and 93 channels on and after
      10 Dec 2006). New options are set by uploading a data structure
      to the DPU. On 12 Jan 2010, the smaller of the two expanded
      multispectral channel selections was replaced with the largest
      value supportable at a 15 Hz frame rate, all 107 VNIR channels
      and 155 IR channels; most of the IR channels are contiguous from
      1.8-2.55 ?m to cover key mineral absorptions.

      Calibration lamps: 4095 levels are commandable in each of two
      lamps at each focal plane, and in two lamps in the integrating
      sphere. All lamps can be commanded open-loop, meaning that
      current is commanded directly. For the integrating sphere only,
      closed loop control is available at 4095 settings. For closed
      loop control, the setting refers to output from a photodiode
      viewing the interior of the integrating sphere; current is
      adjusted dynamically to attain the commanded photodiode output.
      Note: lamps reach maximum current at open- or closed-loop
      settings <4095. Only data for which the calibration lamps are
      off may be processed to a TRDR.

      Shutter position:  Open, closed, or viewing the integrating
      sphere. The shutter is actually commandable directly to position 0
      through 32.  In software, open=3, sphere=17, closed=32. NOTE:
      during integration and testing, it was discovered that at
      positions <3 the hinge end of the shutter is directly illuminated
      and creates scattered light. Position 3 does not cause this
      effect, but the other end of the shutter slightly vignettes
      incoming light. Only data in which the shutter is open, and at
      position 3, may be processed to a TRDR.

      Pointing:  CRISM has two basic gimbal pointing configurations and
      two basic superimposed scan patterns.  Pointing can be (1) fixed
      (nadir-pointed in the primary science orbit) or (2) dynamic,
      tracking a target point on the surface of Mars and taking out
      ground track motion.  Two types of superimposed scans are
      supported: (1) a short, 4-second duration fixed-rate ('EPF-type')
      scan which superimposes a constant angular velocity scan on either
      of the basic pointing profiles, or (2) a long, minutes-duration
      fixed-rate ('target swath-type') scan. Pointing configuration
      affects the contents but not the dimensionality of a TRDR.

    Processing
    ==========

      The CRISM data stream downlinked by the spacecraft unpacks into a
      succession of compressed image frames with binary headers
      containing housekeeping. In each image, one direction is spatial
      and one is spectral.  There is one image for the VNIR focal plane
      and one image for the IR focal plane. The image from each focal
      plane has a header with 220 housekeeping items that contain full
      status of the instrument hardware, including data configuration,
      lamp and shutter status, gimbal position, a time stamp, and the
      target ID and macro within which the frame of data was taken.
      These parameters are stored as part of an Experiment Data Record
      (EDR), which consists of raw data, or a Targeted Reduced Data
      Record, or TRDR, the 'calibrated' equivalent of an EDR.

      The data in one EDR or TRDR represent a series of image frames
      acquired with a consistent instrument configuration (shutter
      position, frame rate, pixel binning, compression, exposure time,
      on/off status and setting of different lamps). Each frame has
      dimensions of detector columns (spatial samples) and detector rows
      (wavelengths, or bands). The multiple image frames are
      concatenated, and are formatted into a single multiple-band
      image (suffix *.IMG) in one file, plus a detached list file in
      which each record has housekeeping information specific to one
      frame of the multiple-band image (suffix *.TAB).

      The text file is based on the 220 housekeeping items. Five
      of the items are composite in that each byte encodes
      particular information on gimbal status or control. These separate
      items are not broken out, except for the gimbal status at the
      beginning, middle, and end of each exposure, from which gimbal
      position is broken out (3 additional items).  The housekeeping is
      pre-pended with 10 additional frame-specific items useful in data
      validation, processing, and sorting, for a total of 233 items per
      frame.  Further information can be found in the data product SIS
      in the DOCUMENT directory.

      The multiple-band image has dimensions of sample, line, and
      wavelength. The size of the multiple-band image varies according
      to the observation mode but is deterministic given the macro ID.
      A typical multiple-band image might have XX pixels in the sample
      (cross-track) dimension, YY pixels in the line (along-track)
      dimension, and ZZ pixels in the wavelength dimension, where:

      XX (samples) = 640/binning, where 640 is the number of columns
      read off the detector, and binning is 1, 2, 5, or 10;

      YY (lines) = the number of image frames with a consistent
      instrument configuration; and

      ZZ (bands) = the number of detector rows (wavelengths) whose
      read-out values are retained by the instrument.

      Once data are assembled into EDRs, they are calibrated into TRDRs.
      Image data are converted to units of radiance using level-4 and
      level-6 CDRs, and analog housekeeping items in the text file
      (voltages, currents, and temperatures) have been converted into
      physical units using a level-6 CDR. Both files share a common
      label. The calibration algorithms are discussed at length in an
      Appendix in the CRISM Data Product SIS.

      A TRDR may also contain separately labeled multiband images in
      which radiance has been processed to one of the following:

        I/F (radiance divided by (pi * solar flux at 1 AU * heliocentric
        distance^2)),

        Lambert albedo, or

        a set of derived spectral parameters (summary products) that
        provide an overview of the data set. The summary products
        include Lambert albedo at key wavelengths, or key band depths
        or spectral reflectance ratios.

      To create Lambert albedo or most summary products, estimated
      corrections for atmospheric, photometric, and thermal effects are
      applied to the I/F data using corrections given in ADRs. The
      formulations for all of the summary products have been validated
      using data from Mars Express/OMEGA.

      The sequence of processing that creates a TRDR is as follows:

      (a) EDRs are assembled from raw data.

      (b) The radiance multiband images in TRDRs are created from the
      EDRs and Calibration Data Records, or CDRs, using a calibration
      algorithm discussed at length in an Appendix in the CRISM Data
      Products SIS.

      (c) Gimbal positions are extracted from the EDR housekeeping and
      formatted as a gimbal C kernel.

      (d) Using the gimbal C kernel and other SPICE kernels, DDRs are
      created. The surface intercept on the MOLA shape model is
      calculated for each spatial pixel (sample at the reference
      detector row). The angles of this pixel relative to the equatorial
      plane and reference longitude constitute the latitude and
      longitude of the pixel. For that latitude and longitude, solar
      incidence, emission, and phase angles are determined at a
      surface parallel to the areoid but having a radius from
      planetary center equivalent to that of the surface intercept of
      the shape model. Solar incidence and emission are also
      determined relative to the shape model itself. Using the
      latitude and longitude of the surface intercept of each spatial
      pixel, TES bolometric albedo and thermal inertia are retrieved
      from global map products, and resampled into CRISM sensor space
      using nearest neighbor resampling. The same procedure is used to
      retrieve MOLA elevation, and the local slope magnitude and slope
      azimuth of the MOLA elevation model.

      (e) Optionally, radiance is converted to I/F by dividing by (pi *
      solar flux at 1 AU * heliocentricdistance^2)). Solar flux is
      maintained in a level 4 CDR, and solar distance is written in the
      label to the radiance image.

      (f) Beginning with version 3 of TRDRs, additional processing is
      applied to I/F multiband images to remediate the effect of noise
      in the flight data and propagated noise from ground and flight
      calibrations. The IR multiband image only is processed using an
      iterative kernel filter. A kernel with settable, typical dimensions
      of 3x3x5 spatial x spatial x spectral pixels is used to identify
      afflicted pixels using a Grubbs test for outliers; an outlying
      value is replaced by interpolation linearly in the spatial
      direction and using a 2nd degree polynomial in the spectral
      direction. Both the VNIR and IR multiband images are then
      processed using the ratio shift correction, which searches for
      values from detector elements that are systematically high or low
      and using a ratio to nearby pixels attempts to remove systematic
      error. Testing on a variety of multiband images shows that most
      systematic errors propagated from calibration files are removed,
      and the spurious data values due to elevated IR detector operating
      temperatures are greatly reduced in number and magnitude.

      (f) Optionally, I/F is converted to Lambert albedo. Some or all
      of the following corrections may be made:

        I/F is divided by cosine of the solar incidence angle, the
        estimated contribution to and attenuation of the signal by
        atmospheric aerosols is normalized or removed, and the
        estimated attenuation of the signal by atmospheric gases
        is removed.

        The thermal emission from longer IR wavelengths is removed.

      MRO:ATMO_CORRECTION_FLAG and MRO:THERMAL_CORRECTION_MODE
      indicate whether such corrections have been performed.

      CRISM data products are described in greater detail in the Data
      Product Software Interface Specification and the Data Archive
      Software Interface Specification in the DOCUMENT directory.

      Details of DN to radiance conversion
      ====================================

      FLIGHT AND GROUND CALIBRATION DATA:
      ----------------------------------

      All calibration matrices are stored in 'calibration data
      records' or CDRs, separate from the main algorithm coded in
      software. There are two general classes of calibration matrices,
      those derived from ground data and updated infrequently,
      and those that represent highly time-variable properties of
      the instrument. Examples of the former include the constants
      needed to uncompress data, correct non-linearity, or correct
      bias for effects of detector or focal plane electronics
      temperature. Examples of the latter include bias and IR thermal
      background, which depend on detector and spectrometer housing
      temperature respectively.

      There are two formats for storing the values in the matrices,
      distinguished by the levels of processing. Level 6 CDRs, or
      CDR6s, are tabulated numbers in ASCII format, and level 4 CDRs,
      or CDR4s, are images each derived from a collection of flight or
      ground calibration measurements.

      Calibration matrices that are highly time variable are
      measured inflight, and include the following.

        For the VNIR detector bias is measured directly, with the
        shutter closed and at the same integration time as
        accompanying measurements of Mars. For the IR, shutter-closed
        measurements also include thermal background. The bias is
        therefore measured for the IR detector by taking data at
        several integration times and extrapolating to zero exposure.
        The step function, a discontinuity in measured bias at some
        detector row, is modeled deterministically as a function of
        integration time. Currently, the default is to take a VNIR
        bias measurement with every observation and IR bias
        observations several times daily.

        IR thermal background is the response of the IR detector to
        'glow' of the inside of the instrument predominantly at >2300
        nm. The change in spectrometer housing temperature that perturbs
        thermal background by the equivalent to read noise is about
        0.02K, whereas the spectrometer housing is predicted to change
        by several degrees over the course of an orbit. Therefore
        shutter-closed IR measurements are taken interspersed
        within all observations, at an interval of approximately once
        per 3 minutes. To correct any given frame of scene data,
        temporally adjacent shutter-closed measurements are used.

        The onboard integrating sphere serves as the radiance reference
        against which CRISM's radiometric responsivity as a system is
        pegged. Multiple measurements are conducted daily. The signal
        from the sphere is adequate at wavelengths greater than 560 nm;
        at shorter wavelengths a fixed responsivity derived from ground
        calibrations is used.

        The integrating sphere provides sufficient signal for a
        preliminary 'flat-field' correction of IR data, but not
        VNIR data. Pixel-to-pixel variations in VNIR detector
        responsivity are measured monthly using bland regions of Mars.
        Co-temporal IR measurements of the same bland regions are used
        to derive a secondary flat-field correction for IR data; in
        practice this is thought mostly to reduce propagated artifacts
        from ground and flight calibrations.

      RADIOMETRIC CALIBRATION:

      Radiometric calibration to units of radiance involves
      uncompressing data, correcting instrument artifacts, subtracting
      bias and background, dividing by exposure time, and converting
      of the result of these steps to units of radiance by comparing
      against a radiometric reference. This approach explicitly uses
      measurements of the internal integrating sphere. A simplified
      form of the equation to reduce measurements to units of radiance,
      using ground and flight calibration measurements, is:

      RD(x,lambda) = M(x,lambda,Hz)( ( K(x,lambda,Hz)( D14lambda(
      DN(x,lambda,TV,TW,TI,TJ,T2,Hz,t) ) -
      BiaT(x,lambda,TV,TW,TI,TJ,Hz,t) ) / t - Bkgd(x,lambda,TI,T2,Hz) -
      Scat(x,lambda,TV,TI,T2,Hz) ) / RST(x,lambda,TV,TI,T2,T3,S) )

      Subscripts define the variables on which calibration coefficients
      depend, and include the following:

        x is spatial position in a row on the focal plane, in detector
        elements.

        lambda is position in the spectral direction on the focal plane,
        in detector elements.

        Hz is frame rate, and implicitly includes with it compression
        configuration including wavelength table and binning mode

        TI, is IR detector temperature in degrees K.

        TV is VNIR detector temperature in degrees K.

        T2, is spectrometer housing temperature in degrees K.

        T3 is temperature of the integrating sphere in degrees K.

        TJ is IR focal plane board temperature in degrees K.

        TW VNIR focal plane board temperature in degrees K.
        t is integration time in seconds.

        s is choice of sphere bulb, side 1 (controlled by IR focal
        plane electronics) or side 2 (controlled by VNIR focal plane
        electronics).

      Discussion of the various instrument effects being corrected is
      including in the 'Confidence Level Note' below.

      All of the input temperatures come from instrument housekeeping,
      and are monitored by the focal plane electronics. Temperatures are
      corrected for electronics noise by substituting for temperatures,
      currents, and voltages in the image headers the corresponding
      values at the same spacecraft time from the low-speed telemetry
      stored in 'ST' CDR6s. This step is performed because the low-speed
      telemetry maintains a fixed timing relative to instrument current
      variations on 1-second cycles, whereas the image headers do not;
      this makes the electronics noise more easily calibrated in the
      low-speed telemetry. The raw digital values are corrected for
      effects of frame rate and variable current loads, including lamps
      and coolers, using additive and multiplicative coefficients
      maintained in the 'HD' CDR6, and then scaled to physical units
      using other coefficients maintained in the 'HK' CDR6.

      The terms in the equation and their sequential application are as
      follows:

        Data decompression. D14lambda converts from raw 8- or 12-bit DNs
        to 14-bit DNs. This is accomplished by inverting the 12-to-8-bit
        LUT using the 'LI' CDR6, then dividing by the gain and adding
        the offset used onboard, whose values are contained in the 'PP'
        CDR6.

        Bias subtraction. BiaT(x,lambda,TV,TW,TI,TJ,Hz,t) is detector
        bias derived as described above from flight measurements, and
        stored as a 'BI' CDR4. For the VNIR, it is just a decompressed
        shutter-closed measurement. For the IR, it is the zero-exposure
        intercept of the pixel-by-pixel fit of 14-bit DN to exposure
        time in the bias measurements, added to the bias step function
        stored in the 'BS' CDR6.

        Bias is corrected for changes in focal plane electronics and
        detector temperature since the time of bias measurement, using
        telemetered detector and electronics temperature and the 'DB'
        and 'EB' CDR6's respectively.

        Electronics artifacts correction. K(x,lambda,Hz) applies
        detector ghost and detector nonlinearity corrections.
        This is actually a composite of distinct steps.

          The correction for detector ghosts subtracts the scaled, bias-
          removed DN from each quadrant from every other quadrant of the
          detector. Scaling coefficients are stored in the 'GH' CDR6.

          The nonlinearity correction scales bias- and ghost-removed DN
          to account for nonlinearity in detector response. Detector-
          averaged scaling coefficients are stored in the 'LI' CDR6.

          Pixel-dependent nonlinearity at higher frame rates can be
          corrected using images of bland regions of Mars taken at the
          same frame rate. A non-uniformity matrix, or flat-field,
          the 'NU' CDR4, is constructed from several thousand frames of
          different scenes along-track that are averaged to remove
          non-uniform illumination of the surface due to topography,
          and the spatial image at each wavelength is normalized by its
          mean value.

        Background subtraction. Bkgd(x,lambda,TI,T2,Hz) is a 'BK' CDR4
        constructed by applying the D14, BiaT, and K corrections to a
        shutter-closed IR measurement interspersed with Mars
        measurements. The actual background subtracted from a scene
        measurement is a time-weighted average of preceding and
        subsequent 'BK' shutter-closed measurements, to allow for the
        continuous variation of IR thermal background as spectrometer
        housing temperature changes.

        Scattered light subtraction. Scat(x,lambda,TV,TI,T2,Hz) is the
        stray light subtraction, and includes two components. The first
        component is glare from the gratings, which produces a low level
        of light at a distance of tens or more of pixels from a source.
        For the VNIR detector, this component of scattered light is
        measured directly at each row of the detector as the mean level
        in the scattered light columns at a given row. The locations of
        the scattered light and scene pixels are stored in the 'DM' CDR4.
        It is then extrapolated across the detector using a function
        based on signal at the shortest wavelengths (UV), which is
        dominated by scatter. The shape of he function varies as a
        function of wavelength. For the IR detector, transient bad
        pixels render this correction noisy so instead this correction
        was derived using a bland, dusty scene on Mars, applying the
        correction both to the scene and to the accompanying sphere
        measurement. The extrapolation across the field of view uses a
        function resembling that at the longest VNIR wavelengths. The
        correction was then median filtered and multiplied into the
        sphere radiometric model. Given the derivation of the IR
        scattered light correction, it is most accurate for uniformly
        illuminated scenes.

        The second component of scattered light is the second-order
        light leaked through zone 3 of the IR order sorting filter.
        This is removed by scaling and subtracting the measured signal
        at second order wavelengths from the measured first-order signal
        in zone 3. For each detector row (wavelength), which second-
        order rows to use and their weightings are stored in the 'LL'
        CDR4.

        Responsivity correction using sphere data.
        RST(x,lambda,TV,TI,T2,T3,S) is spectral responsivity derived
        from onboard sphere calibration images. It is calculated by
        processing a sphere measurement through the aforementioned
        steps with two exceptions, and dividing by exposure time to
        create an 'SP' CDR4. The exception is that the background image
        is taken looking into the unilluminated sphere instead of with
        the shutter closed, in order to subtract out the blackbody
        radiation of the sphere's structure. The 'SP' product is
        corrected for non-reproducibility in shutter position by
        measuring the 'peak' in sphere DN/ms near VNIR detector row
        232, and scaling a multiplicative correction stored in the
        'SH' CDR4 by the magnitude of that peak.

        The SP CDR4 is divided by the sphere spectral radiance model
        stored in the 'SS' CDR4 to derive a snapshot of instrument
        responsivity. The model uses as an input the choice of sphere
        bulb. Normally the bulb controlled by the IR focal plane
        electronics is used.

        M(x,lambda,Hz) applies the detector mask in the 'DM' CDR4,
        flagging non-scene data (e.g. scattered light and masked pixels)
        with a value of 65535. This is a standard value for missing or
        'bad' (saturated) data.

        Converting radiance to I/F. RD(x,lambda) is the observed
        spectral radiance in W/m2/steradian/um at the instrument
        aperture, and is the output of the preceding steps for a scene
        measurement. That radiance may be converted to I/F by dividing
        by squared solar distance (stored in EDR and TRDR labels) and
        the solar irradiance model stored in the 'SF' CDR4. That model
        is itself derived convolving a predicted solar spectrum with
        the measured center wavelength (stored in the 'WA' CDR4) and
        spectral bandpass (stored in the 'SB' CDR4) of every detector
        element.

    Data
    ====

      DATA DESCRIPTION:

      There is only one data type associated with this volume, the
      Targeted Reduced Data Records or TRDRs.

      The TRDR consists of the output of one of the constituent macros
      associated with a target ID that contains scene data (Mars or
      other). Not all EDRs are processed to TRDR level; those
      containing bias, background, sphere, or focal plane lamp data
      are processed instead to CDRs. Only scene EDRs are processed to
      the TRDR level.

      The TRDR contains one or more multiple-band images (suffix *.IMG).
      One matches the dimensions of the multiple-band image of raw DN in
      an EDR, except that the data are in units of radiance. The size of
      the multiple-band image varies according to the observation mode
      but is deterministic given the ID of the command macro used to
      acquire the data. Appended to the multiple-band image is a binary
      table of the detector rows that were used, as selected by the
      wavelength filter. This is a one-column table, with each row
      containing one detector row number expressed as a 16-bit unsigned
      integer values, most significant bit first.

      Other multiple-band images may contain I/F, Lambert albedo, or
      derived summary products. The I/F and Lambert albedo images, if
      present, parallel the structure of the radiance image except lack
      the list file. The summary products image has the same spatial
      dimensions, but a different dimension in the spectral direction
      and it lacks that table of row numbers. Each of these three
      multiple-band images has its own label.
      There are 45 summary parameters, as follows:

      SURFACE PARAMETERS:  from Lambert albedo

      NAME: R770
      PARAMETER: 0.77 micron reflectance

      FORMULATION *: R770
      RATIONALE: rock/dust ratio

      NAME: RBR
      PARAMETER: red/blue ratio
      FORMULATION *: R770 / R440
      RATIONALE: rock/dust ratio

      NAME: BD530
      PARAMETER: 0.53 micron band depth
      FORMULATION *: 1 - (R530/(a*R709+b*R440))
      RATIONALE: crystalline ferric minerals

      NAME: SH600
      PARAMETER: 0.60 micron shoulder height
      FORMULATION *: R600/(a*R530+b*R709)
      RATIONALE: select ferric minerals

      NAME: BD640
      PARAMETER: 0.64 micron band depth
      FORMULATION *: 1 - (R648/(a*R600+b*R709))
      RATIONALE: select ferric minerals, especially maghemite

      NAME: BD860
      PARAMETER: 0.86 micron band depth
      FORMULATION *: 1 - (R860/(a*R800+b*R984))
      RATIONALE: select ferric minerals

      NAME: BD920
      PARAMETER: 0.92 micron band depth
      FORMULATION *: 1 - ( R920 / (a*R800+b*R984) )
      RATIONALE: select ferric minerals

      NAME: RPEAK1
      PARAMETER: reflectance peak 1
      FORMULATION *: wavelength where 1st derivative=0 of 5th order
        polynomial fit to R600, R648, R680, R710, R740, R770, R800, R830
      RATIONALE: Fe mineralogy

      NAME: BDI1000VIS
      PARAMETER: 1 micron integrated band depth; VIS wavelengths
      FORMULATION *: divide R830, R860, R890, R915 by RPEAK1 then
        integrate over (1 -  normalized radiances)
      RATIONALE: crystalline Fe+2 or Fe+3 minerals

      NAME: BDI1000IR
      PARAMETER: 1 micron integrated band depth; IR wavelengths
      FORMULATION *: divide R1030, R1050, R1080, R1150
        by linear fit from peak R  between 1.3 - 1.87 microns to R2530
        extrapolated backwards, then integrate over (1 -  normalized
        radiances)
      RATIONALE: crystalline Fe+2 minerals; corrected for overlying
        aerosol induced slope

      NAME: IRA
      PARAMETER: 1.3 micron reflectance
      FORMULATION *: R1330
      RATIONALE: IR albedo

      NAME: OLINDEX (prior to TRDR version 3)
      PARAMETER: olivine index
      FORMULATION *: (R1695 / (0.1*R1050 + 0.1*R1210 + 0.4*R1330 +
        0.4*R1470)) - 1
      RATIONALE: olivine will be strongly +; based on fayalite

      NAME: OLINDEX2 (beginning with TRDR version 3)
      PARAMETER: olivine index with less sensitivity to illumination
      FORMULATION *: (((RC1054 ? R1054)/RC1054) * 0.1)
        + (((RC1211 ? R1211)/(RC1211) * 0.1)
        + (((RC1329 ? R1329)/RC1329) * 0.4)
        + (((RC1474 ? R1474)/RC1474) * 0.4)
      RATIONALE: olivine will be strongly positive

      NAME: LCPINDEX
      PARAMETER: pyroxene index
      FORMULATION *: ((R1330-R1050) / (R1330+R1050)) *
        ((R1330-R1815) / (R1330+R1815)
      RATIONALE: pyroxene is strongly +; favors low-Ca pyroxene

      NAME: HCPXINDEX
      PARAMETER: pyroxene index
      FORMULATION *: ((R1470-R1050) / (R1470+R1050)) *
        ((R1470-R2067) / (R1470+R2067)
      RATIONALE: pyroxene is strongly +; favors high-Ca pyroxene

      NAME: VAR
      PARAMETER: spectral variance
      FORMULATION *: find variance from a line fit from 1 - 2.3 micron
        by summing in quadrature over the intervening wavelengths
      RATIONALE: Ol & Px will have high values; Type 2 areas will have
        low values

      NAME: ISLOPE1
      PARAMETER: -1 * spectral slope1
      FORMULATION *: (R1815-R2530) / (2530-1815)
      RATIONALE: ferric coating on dark rock

      NAME: BD1435
      PARAMETER: 1.435 micron band depth
      FORMULATION *: 1 - ( R1430 / (a*R1370+b*R1470) )
      RATIONALE: CO2 surface ice

      NAME: BD1500
      PARAMETER: 1.5 micron band depth
      FORMULATION *: 1 - ( ((R1510+R1558)*0.5) / (a*R1808+b*R1367)
      RATIONALE: H2O surface ice

      NAME: ICER1
      PARAMETER: 1.5 micron and 1.43 micron band ratio
      FORMULATION *: R1510 / R1430
      RATIONALE: CO2, H20 ice mixtures

      NAME: BD1750
      PARAMETER: 1.75 micron band depth
      FORMULATION *: 1 - ( R1750 / (a*R1660+b*R1815) )
      RATIONALE: gypsum

      NAME: BD1900
      PARAMETER: 1.9 micron band depth
      FORMULATION *: 1 - ( ((R1930+R1985)*0.5) / (a*R1857+b*R2067) )
      RATIONALE: H2O, chemically bound or adsorbed

      NAME: BDI2000
      PARAMETER: 2 micron integrated band depth
      FORMULATION *: divide R1660, R1815, R2140, R2210, R2250, R2290,
        R2330, R2350, R2390, R2430, R2460 by linear fit from peak R
        between 1.3 - 1.87 microns to R2530, then integrate over
       (1 -  normalized radiances)
      RATIONALE: pyroxene abundance and particle size

      NAME: BD2100
      PARAMETER: 2.1 micron band depth
      FORMULATION *: 1 - ( ((R2120+R2140)*0.5) / (a*R1930+b*R2250) )
      RATIONALE: monohydrated minerals

      NAME: BD2210
      PARAMETER: 2.21 micron band depth
      FORMULATION *: 1 - ( R2210 / (a*R2140+b*R2250) )
      RATIONALE: Al-OH minerals

      NAME: BD2290
      PARAMETER: 2.29 micron band depth
      FORMULATION *: 1 - ( R2290 / (a*R2250+b*R2350) )
      RATIONALE: Mg,Fe-OH minerals (at 2.3); also CO2 ice
        (at 2.292  microns)

      NAME: D2300
      PARAMETER: 2.3 micron drop
      FORMULATION *: 1 - ( (CR2290+CR2320+CR2330) /
        (CR2140+CR2170+CR2210) ) (CR values are observed R values
        divided by values fit along the slope as determined between 1.8
        and 2.53 microns - essentially continuum corrected))
      RATIONALE: hydrated minerals; particularly clays

      NAME: SINDEX
      PARAMETER: Convexity at 2.29 microns  due to absorptions at
        1.9/2.1 microns and 2.4 microns
      FORMULATION *: 1 - (R2100 + R2400) / (2 * R2290) CR
        values are observed R values divided by values fit along the
        slope as determined between 1.8 - 2.53 microns (essentially
        continuum corrected))
      RATIONALE: hydrated minerals; particularly sulfates

      NAME: ICER2
      PARAMETER: gauge 2.7 micron band
      FORMULATION *: R2530 / R2600
      RATIONALE: CO2 ice will be >>1, H2O ice and soil will be about 1

      NAME: BDCARB
      PARAMETER: overtone band depth
      FORMULATION *: 1 - ( sqrt [ ( R2330 / (a*R2230+b*R2390) ) *
        ( R2530/(c*R2390+d*R2600) ) ] )
      RATIONALE: carbonate overtones

      NAME: BD3000
      PARAMETER: 3 micron band depth
      FORMULATION *: 1 - ( R3000 / (R2530*(R2530/R2210)) )
      RATIONALE: H2O, chemically bound or adsorbed

      NAME: BD3100
      PARAMETER: 3.1 micron band depth
      FORMULATION *: 1 - ( R3120 / (a*R3000+b*R3250) )
      RATIONALE: H2O ice

      NAME: BD3200
      PARAMETER: 3.2 micron band depth
      FORMULATION *: 1 - ( R3320 / (a*R3250+b*R3390) )
      RATIONALE: CO2 ice

      NAME: BD3400
      PARAMETER: 3.4 micron band depth
      FORMULATION *: 1 - ( (a*R3390+b*R3500) / (c*R3250+d*R3630) )
      RATIONALE: carbonates; organics

      NAME: CINDEX
      PARAMETER: gauge 3.9 micron band
      FORMULATION *: ( R3750 + (R3750-R3630) / (3750-3630) *
        (3920-3750) ) / R3920 - 1
      RATIONALE: carbonates

      ATMOSPHERIC PARAMETERS: from I/F

      NAME: R440
      PARAMETER: 0.44 micron reflectance
      FORMULATION *:  R440
      RATIONALE: clouds/hazes

      NAME: IRR1
      PARAMETER: IR ratio 1
      FORMULATION *: R800 / R1020
      RATIONALE: Aphelion ice clouds  vs. seasonal or  dust

      NAME: BD1270O2
      PARAMETER: 1.265 micron band
      FORMULATION *: 1 - ( (a*R1261+b*R1268) / (c*R1250+d*R1280) )
      RATIONALE: O2 emission; inversely correlated with high altitude
        water; signature of ozone

      NAME: BD1400H2O
      PARAMETER: 1.4 micron band depth
      FORMULATION *: 1 - ( (a*R1370+b*R1400) / (c*R1330+d*R1510) )
      RATIONALE: H2O vapor

      NAME: BD2000CO2
      PARAMETER: 2 micron band
      FORMULATION *: 1 - ( R2010 / (a*R1815+b*R2170) )
      RATIONALE: atmospheric CO2

      NAME: BD2350
      PARAMETER: 2.35 micron band depth
      FORMULATION *: 1 - ( (a*R2320+b*R2330+c*R2350) / (d*R2290+e*R2430) )
      RATIONALE: CO

      NAME: IRR2
      PARAMETER: IR ratio 2
      FORMULATION *: R2530 / R2210
      RATIONALE: aphelion ice clouds vs. seasonal or dust

      NAME: BD2600
      PARAMETER: 2.6 micron band depth
      FORMULATION *:  1 - ( R2600 / (a*R2530+ b*R2630) )
      RATIONALE: H2O vapor

      NAME: R2700
      PARAMETER: 2.70 micron reflectance
      FORMULATION *: R2700
      RATIONALE: high aerosols

      NAME: BD2700
      PARAMETER: 2.70 micron band depth
      FORMULATION *: 1 - ( R2700 / (R2530*(R2530/R2350)) )
      RATIONALE: CO2; atmospheric structure (accounts for spectral slope)

      NAME: IRR3
      PARAMETER: IR ratio 3
      FORMULATION *: R3750 / R3500
      RATIONALE: aphelion ice clouds vs. seasonal or dust

      Note *: 'a', 'b', 'c', 'd', 'e' in band depth formulations
      represent fractional distances between wavelengths wavelength;
      for example, given BD(c), a band depth at a central wavelength
      'c' with nearby continuum points defined at shorter and longer
      wavelengths 's' and 'l':  BD(c) = 1 - R(c) / (a*R(s) + b*R(l)),
      where a = 1 - b and b = (lambda(c) - lambda(s)) / (lambda(l) -
      lambda(s))

      DATA DIMENSIONALITY:

      The size of the multiple-band image varies according to the
      observation mode but is deterministic given the ID of the
      onboard macro the generated the data. A typical multiple-band
      image might have XX pixels in the sample (cross-track) dimension,
      YY pixels in the line (along-track) dimension, and
      ZZ pixels in the wavelength dimension, where:

        XX=640/binning, where binning is 1, 2, 5, or 10, and dark is the
          number of masked and scattered light pixels
        YY=the number of frames of data taken by the macro, and
        ZZ=the number of rows (wavelengths) that are retained by
          the instrument.

      The data in a TRDR do not have optical distortions removed. In
      one column, the projection onto Mars' surface may vary by as much
      as +/-0.4 not-binned detector elements in the XX dimension
      depending on position in the FOV (distortions are worst at the
      edges of the VNIR and IR FOVs). For a single wavelength, its
      location in the ZZ direction may vary by as much as +/-1
      not-binned detector elements depending on wavelength and position
      in the XX direction (distortions are worst at the short- and long-
      wavelength ends of the IR detector).

      To correct for optical distortions, multiband images may be
      resampled in the spectral or spatial direction. Three types of
      resampling may have occurred: (a) resampling in the wavelength
      direction, as coded in the PS CDR; (b) resampling in the spatial
      direction, to remove differences in spatial scale with wavelength
      or band, using the CM CDR; and (c) VNIR data may be rescaled to
      match the slightly different magnification of the IR spectrometer,
      also the CM CDR. A resampled TRDR is distinguished by values of
      the keywords MRO:SPATIAL_RESAMPLING_FLAG,
      MRO:SPATIAL_RESCALING_FLAG, and MRO:SPECTRAL_RESAMPLING_FLAG.

    Ancillary Data
    ==============

      There are various types of ancillary data provided with this
      dataset:

      1. Each hyperspectral targeted (gimbaled) observation TRDR IF
      data product has one browse product in a directory structure
      that parallels that of the TRDR directory. See
      BROWSE/BROWINFO.TXT for further details.

      2. The TRDR EXTRAS visualizations are Portable Network Graphics
      (PNG) format files that depict the geometry and structure of
      CRISM hyperspectral targeted observations. See
      EXTRAS/EXTRINFO.TXT for further details.

    Coordinate System
    =================

      The cartographic coordinate system used for the CRISM data
      products conforms to the IAU planetocentric system with East
      longitudes being positive. The IAU2000 reference system for Mars
      cartographic coordinates and rotational elements was used for
      computing latitude and longitude coordinates.


    Media/Format
    ============


      The CRISM archive will be made available online via Web and FTP
      servers.  This will be the primary means of distribution.
      Therefore the archive will be organized as a set of virtual
      volumes, with each data set stored online as a single volume. As
      new data products are released they will be added to the volume's
      data directory, and the volume's index table will be updated
      accordingly. The size of the volume will not be limited by the
      capacity of the physical media on which it is stored; hence the
      term virtual volume.  When it is necessary to transfer all or part
      of a data set to other media such as DVD for distribution or for
      offline storage, the virtual volume's contents will be written to
      the other media according to PDS policy, possibly dividing the
      contents among several physical volumes.

Confidence Level Overview
    =========================

      There is a number of sources of uncertainty in the interpretation
      of TRDRs including:

      (A) Stochastic noise

      Random noise in the data due to statistical uncertainties in
      counting photons. This is manifested as noisy calibrated data.
      Noise is most significant in darker areas. Typically, the signal
      to noise ratio at <2500 nm is 400 in bright areas and 200 in dark
      areas, in the constituent observations, before pixel binning.

      (B) Optical distortions

      Optical distortion can affect spectra of small-scale features.
      In one column, the projection onto Mars' surface may vary by as
      much as +/-0.4 not-binned detector elements in the XX dimension
      depending on position in the FOV (distortions are worst at the
      edges of the VNIR and IR FOVs). For a single wavelength, its
      location in the ZZ direction may vary by as much as +/-1
      not-binned detector elements depending on wavelength and position
      in the XX direction (distortions are worst at the short- and
      long-wavelength ends of the IR detector). In other words,
      different wavelengths include slightly different combinations
      of signal from spatially adjacent pixels, so that compositional
      interpretations of features near the scale of a pixel are weakly
      wavelength-dependent. Also, wavelength drifts across the field
      of view. Compositional interpretations based on exact
      wavelengths of absorptions may thus be weakly spatially dependent.
      The resampling approach outlined above can remove much of this
      uncertainty.

      (C) Variable spectral resolution

      In order to distinguish spectrally similar minerals that
      have different geological implications for their environments
      of formation, adequate spectral resolution is necessary. This
      requires sufficiently high density spectral sampling, as well
      as a sufficiently narrow full width half maximum (FWHM) of the
      instrument response in the spectral direction. This 'slit
      function,' the effective bandpass for a single detector
      element, represents the convolution of spectral sampling and
      the point-spread function in the spectral direction. CRISM's
      benchmark is distinguishing the minerals montmorillonite and
      kaolinite, which form in hydrothermal environments under
      different temperature regimes [SWAYZEETAL2003]. The requirements
      for this are (a) <20 nm FWHM and (b) sampling of the spectrum
      at this or smaller increments. CRISM's spectral sampling
      requirement is <10 nm/channel to provide oversampling, and the
      actual performance is better at 6.55 nm/channel. FWHM is 8 nm
      in the VNIR across the FOV. In the IR it increases from 10 nm
      at short wavelengths to 15 nm at the longest wavelengths at the
      center of the FOV, and broadens by about 2 nm at 0.8 degrees
      from the center of the field of view. Outside +/-0.9 degrees
      from the center of the field of view the telescope is slightly
      vignetted, so further degradation is expected at extreme field
      angles. Although the spectral sampling and resolution meet
      requirements, their variation across the field-of-view must be
      accounted for when comparing with rock and mineral analog
      spectra.

      (D) Calibration artifacts

      There are several instrument artifacts that are corrected in
      the calibration pipeline. Residual errors in the corrections will
      introduce systematic errors into the data.

      Optical Effects:

      (1) The boundary of zones 1 and 2 of the VNIR order sorting
      filter is a joint between two distinct glasses with different
      indices of refraction. When illuminated during detector-level
      tests, it was found to cause significant (>10%)
      scattered light at shorter wavelengths (<670 nm). This was
      correcting by replacing the VNIR focal plane assembly with the
      flight spare, onto which a narrow black stripe was painted to
      shadow the joint. The black stripe attenuates the light from
      610-710 nm and causes a dip in response at those wavelengths.
      In the processed data, the are two major effects: signal to
      noise ratio is decreased, and non-reproducibility of the
      exact position of the shutter when observing the sphere causes
      shifting of the shadow in the wavelength direction. Measurement
      and correction of this effect are discussed below.

      (2) The spectrometer slit - which defines the mapping of
      wavelengths to detector rows as well as the spatial FOV - is
      mounted on a curved surface whose axis of curvature is parallel
      to the wavelength direction. The slit assembly is fixed with
      pins through holes whose diameters are oversized to provide
      margin for fastening the assembly. During instrument-level
      vibration testing, the slit assembly shifted in the wavelength
      direction by the tolerance in the hole diameters, shifting
      wavelength calibration by about 15 nm in both the VNIR and IR.
      Additional shifting of the slit assembly during and after
      launch is thought to have occurred. Wavelength calibration is
      observed to shift with major thermal excursions of the
      optomechanical assembly, at a magnitude of up to +/- 1 nm.
      This shift has been quantified and can be calibrated out using
      measured positions of Martian atmospheric gas absorptions,
      as recorded in the 'WS' CDR.

      (3) Shutter position irreproducibility - To illuminate the
      spectrometer slit's full 2.12 degree field of view, CRISM's
      telescope illuminates a circular region of slightly larger
      diameter surrounding the slit. The base of the shutter, on the
      hinge end, just protrudes into the illuminated area. At
      position 0, originally intended as the 'open' position, the
      reflective rear surface of the shutter provides the detectors
      an unbaffled view of the scene approximately 1 degree from the
      center of the field of view in a cross-slit direction, creating
      an out-of-focus 'ghost' image of that location. Moving the
      shutter through successive steps redirects the angle from which
      the ghost image originates to further from the center of the FOV.
      At position 3, the angle from which the ghost image originates is
      baffled by the telescope, and the ghost disappears. To remediate
      the ghost image, the open position of the shutter is defined in
      software to position 3.

      There is a small (about 0.1 degree) non-reproducibility in the
      angle at which the sphere is viewed and the fact that, unlike the
      external scene, the spectrometer's view of the sphere is vignetted
      by the sphere's aperture. With a slight shift in shutter position,
      the cone of sphere light entering spectrometer optics shifts. The
      filling of the dual zone gratings changes slightly, decreasing
      responsivity at long VNIR wavelengths and short IR wavelengths.
      Also, the shadow of the black strip on the VNIR order-sorting
      filter zone boundary shifts, creating a distinctive trough and
      peak pattern at detector rows 222-235 (approximately 605-690 nm).

      Because this effect is characteristic as a function of
      wavelength, it is correctable. Ratios of different sphere
      observations during ground calibration are used to create a
      multiplicative correction to a sphere image as a function of
      wavelength, that is maintained in a level 4 CDR. In flight data
      to be corrected, the magnitude of the peak near VNIR row 235 is
      measured. The correction is scaled by the magnitude of the peak,
      and it is multiplied by the data. The VNIR row 235 peak is used
      to scale the corrections for both the VNIR and IR. Errors in
      this correction would lead to high or low values especially
      at 600-700 nm, with the error being systematic within
      a group of observations processed using a single sphere
      observation, but random between such groups of observations.
      Put differently, the systematic errors would change every few
      orbits.

      To the limits of measurement error, the small irreproducibility
      of shutter position at the 'open' position has no measurable
      effect on external scene data.

      (4) IR 2nd order leakage: Zone 3 of the IR order sorting filter
      admits up to 3% of the 2nd order light from the grating, at
      wavelengths 1400-1950 nm, that falls at detector rows whose
      nominal wavelengths are 2800-3900 nm. The leakage peaks at a
      nominal wavelength of 3400 nm. Due to the falloff of both the
      solar spectrum and the Martian reflectance spectrum with
      increasing wavelength, the relative magnitude of the leakage
      to the signal in zone 3 is enhanced so that it becomes tens
      of percent of the total signal in that wavelength range.

      Ground testing provided sufficient data for an empirical
      correction for this effect, in which scaled values of signal at
      second-order wavelengths are subtracted from first-order
      (nominal) wavelengths. The correction is maintained in a level 4
      CDR. Errors in this correction would be manifested in processed
      data as a negative positive additive component to the values
      from 2760-3920 nm, centered and strongest at 3400 nm.

      Electronics Ghost:

      Both detectors, but especially the VNIR
      detector, are subject to a weak ghost image of any illuminated
      spot into its corresponding location in every other of the four
      160-column quadrants of the of the 640-column detector. This is
      a small effect at the <1% level, and is removed by scaling the
      image of each quadrant by an empirically determined value that
      is nonlinearly related to signal level, and then subtracting
      the scaled quadrant image from that of every other quadrant. The
      scale factors are maintained in a level 6 CDR. To the
      uncertainties in measurement, each of the four quadrants in a
      detector behaves only slightly differently. There is a minimal
      effect of frame rate, but ghost magnitude is apparently
      unaffected by detector temperature. Errors in this correction
      could be manifested as anomalously dark or bright spots exactly
      one-fourth of the detector width away (160 samples, 80 2x-binned
      samples, 32 5x-binned samples, or 16 10x-binned samples.

      IR 'Bad Pixels':

      The IR detector is operated at cryogenic temperature
      to minimize dark current and bias level of the detector. As the
      MRO mission has progressed, the setpoint for the IR detector has
      been raised to lessen the wear on the mechanical coolers that
      maintain IR detector temperature.

      With increasing detector temperature, not all pixels accrue an
      elevated bias level or dark current - the latter of which adds
      noise due to its electron counting statistics - at the same rate.
      The most susceptible pixels, within which effective SNR or
      available  dynamic range are adversely impacted, are
      'bad pixels.' Beginning with version 3 TRDRs, and improved
      approach to filtering is applied to I/F cubes to interpolate
      over bad pixels are interpolated over.

      VNIR Calibration Versions
      =========================

      Version 0 was the first version of VNIR radiometric calibration
      applied to flight data. Five major sources of inaccuracy and image
      artifacts were identified.

      The first arose from the method by which scattered light from the
      grating was extrapolated from scattered light columns across the
      scene. A simple linear interpolation was applied, and this
      underestimated the total scattered light within the scene. When
      this procedure was applied to ground calibration data to derive
      a model for sphere radiance, results include an unrealistically red
      spectral slope. When applied to flight measurements of Mars and the
      integrating sphere, results include residual cross-track
      (along-slit) color variations.

      The second problem arose from the choice of which sphere bulb to
      use as the primary calibration source. Originally it was the bulb
      controlled by the VNIR focal plane electronics. However,
      for unknown reasons, this bulb yields much greater scatter from the
      grating than does the bulb controlled by the IR focal plane
      electronics. Residuals within the shadow of zones 1 and 2 of the
      wavelengths order sorting filter led to a large negative artifact at
      600-700 nm.

      The third problem arose from low light levels in the sphere at
      <560 nm. That is, measured sphere signal levels are low enough that
      they introduced systematic noise at short wavelengths into
      observations calibrated using them.

      The fourth problem arose from propagated statistical errors in
      sphere measurements used to calibrate the data. That is, corrections
      for all of the artifacts outlined above require a series of
      algebraic operations each of which propagates small errors or
      effects of noise. The sphere data require more corrections, and at
      lower-signal wavelengths the residual errors at each detector
      element are manifested as a spuriously high or low value at the
      corresponding wavelength and spatial position. This can appear as
      wavelength-dependent striping in the along-track direction.

      Fifth, it was discovered that the bias varies image to image,
      leading to striping in the cross-track direction in processed data.

      Version 1 used an arbitrary scaling across of scattered light from
      the grating across the field of view, in an attempt to remediate the
      first two effects. It proved unsuccessful, and was abandoned after
      validation of the first observations to which it was applied.

      Version 2 addresses and largely corrects each of the five major
      problems with version 0.

      First, there is a more sophisticated extrapolation from the scattered
      columns. At short wavelengths, the distribution is measured from
      measured signal at UV wavelengths. The detector is nearly unresponsive
      to UV signal from Mars, so in most regions the nominally UV
      wavelengths instead provide a measure of the scattered light at the
      corresponding position along the slit. This gradually transitions
      with longer wavelength to a linear extrapolation between the scattered
      light columns, as with version 0. This correction is applied both to
      ground calibration data used to derive the sphere radiance model,
      and to flight scene and sphere data.

      Second, the sphere bulb controlled by the IR focal plane electronics
      is used as the primary radiometric reference, because the lower
      scattered light from it is more easily corrected and leaves lesser
      artifacts at low-signal wavelengths.

      Third, calibration of the data is handled differently at <560 nm
      and at >560 nm. At longer wavelengths, sphere measurements and the
      sphere radiance model traceable to ground measurements are used to
      determine a snapshot of detector responsivity, and apply that to scene
      data to derive radiance. This is appropriate because of temperature
      dependence of optical throughput, especially the beamsplitter, at
      >560 nm. At shorter wavelengths, responsivity is derived directly from
      ground calibration measurements, and low-signal sphere data are not
      used. The approach has consistently yielding nearly identical,
      continuous results at the wavelengths at which the two approaches
      overlap, about 530-600 nm.

      Fourth, to eliminate propagation of statistical errors in processing
      of sphere data, the sphere-derived responsivity at each wavelength
      is averaged. Correction for spatial nonuniformity occurs using a Mars
      flat-field calibration and the NU CDR derived from it.

      Fifth, to remove frame-to-frame variations in bias, an additive
      correction is applied to each frame while still in units of DN,
      to make the physically masked columns at the edge of the detector
      zero after dark subtraction.

      There are two alternate versions of VNIR version 2 specifically
      for special types of observations. The bland Mars scenes used to
      measure non-uniformity are processed to version 9 TRDRs.  The
      version 9 processing is the same as version 2 except it doesn't
      include the non-uniformity correction.  Deimos, Phobos, and any
      other pointlike or compact targets like stars are processed to
      version 8 TRDRs.  The major difference in the processing for such
      scenes is that within-scene scattered light from the grating is
      much less, so the correction for intra-scene scattered light is
      skipped.  The VNIR version 8 processing uses special version 8
      'SS' and 'NU' CDR4s.

      Version 3 is the current version of the VNIR calibration, released
      in late 2010. The most substantive change from version 2 is an
      improved correction for shutter mirror irreproducibility. Version 7
      replaces version 9 for bland Mars scenes. Version 6 replaces version
      8 for Phobos and Deimos.

      For I/F cubes - but not radiance cubes -
      the ratio shift correction was introduced to further reduce the
      magnitude of propagated artifacts of ground calibration. (The
      radiance cube is a 'control' in case the ratio shift correction
      introduces its own artifacts.)

      Known Issues with VNIR Radiometric Calibration
      ==============================================

      Version 3 has been validated using observations of the MER Spirit and
      Opportunity landing sites, with PANCAM measurements that were modeled
      at the top of the atmosphere using CRISM viewing geometry and solar
      longitude. Four artifacts or data quality concerns remain.

      (a) Radiance at <410 nm is typically low, because there is not much
      signal at those wavelengths and they are most susceptible to artifacts
      from scattered light subtraction.

      (b) Some artifact at the wavelengths of the filter zone boundary. In
      some parts of the field of view at sharp brightness contrasts it
      extends to 644-684 nm.

      (c) Scenes with large coverage by ice, especially near either edge of
      the field of view, have more significant artifacts from the scattered
      light correction because, unlike typical Martian soils, ice has
      significant UV reflectance and that decreases accuracy of the
      correction. Typical effects may include degradation at wavelengths
      below 480 nm and above 1010 nm.

      (d) Due to spectral smile, mineralogic absorptions located where
      instrument response varies steeply with wavelength exhibit variation
      with field angle. This is most prominent with Fe mineral
      absorptions near 900 nm.

      IR Calibration Versions
      =======================

      Version 0 was the first version of IR radiometric calibration
      applied to flight data. Five major sources of inaccuracy and image
      artifacts in version 0 are known.

      The first originated from inadequacy of the bad pixel correction.
      On ground, three types of bad pixels were observed: hot with high
      dark current, dead with no response, and noisy. Based on that the
      bad pixel correction was planned to use dark measurements to
      identify bad pixels. Inflight, a new type of bad pixel was
      identified, whereby a detector element abruptly develops high dark
      current, then abruptly returns to normal. At least 2 percent of
      pixels, and probably more, display this behavior at some time.
      This was manifested as sharp, wavelength-dependent striping in the
      along-track direction in image data.

      The second problem originated from incomplete correction of bad
      pixels in ground calibration measurements used to derive the sphere
      radiance model. This was manifested as diffuse, wavelength-dependent
      striping in the along-track direction in image data.

      The third problem was spuriously high radiances near the boundaries
      of the order sorting filters, especially at 1630-1680 and 2690-2770.
      A lesser spuriously high radiance occurred at 1800 nm. Spuriously
      high values also occurred at <1050 nm.

      The fourth problem was 'bumps' in radiances at 1370 and 1850 nm,
      water vapor in the beam and adsorbed water on ground calibration
      sources was under-corrected.

      The fifth problem was that the correction for leaked second order
      light at >2700 nm was not implemented in version 0.

      Version 1 partially corrected problems with version 0 but did not
      close out known IR calibration issues.

      First, within-scene bad pixels were identified were interpolated
      over. For each scene, the entire scene was collapsed to one median
      frame matching the layout of the detector. Then a median filter was
      run in the spatial direction. The difference between the median image
      and the filtered median image was calculated, and detector elements
      having more than a 1.4 standard deviation difference were identified
      as bad pixels. These were replaced throughout the scene by the average
      of the adjacent non-bad pixels at the same wavelength.

      Second, artifacts in the sphere radiance model were removed using a
      Mars flat-field calibration and the NU CDR derived from it. However
      this is not done within strong atmospheric absorptions near 1400,
      1600, 2000, and 2700 nm because spectral smile would introduce new
      artifacts.

      Third, systematically high radiances at the filter zone boundaries
      were corrected by smoothing the sphere radiance model.

      Fourth, the 'bumps' in radiances at 1370 and 1850 were corrected by
      performing an approximate atmosphere removal on bland, dusty terrain
      at the summit of Olympus Mons. The magnitude of the 'bumps' was
      estimated, and the appropriate multiplicative correction to the sphere
      model to remove them was applied.

      Version 2 added six further corrections for outstanding issues
      identified in version 1.

      First, in version 1, incomplete correction for water vapor in the
      spectra of ground calibration sources had introduces a 'bump' in
      radiance at 2550-2650 nm. This wavelength is on the edge of a strong
      3-micron absorption in Mars; spectrum, so Olympus Mons observations
      could not be used to estimate a correction. Instead this artifact
      was corrected using observations of Deimos, which lacks a 3-micron
      band. The magnitude of the 'bump' was estimated, and the appropriate
      multiplicative correction to the sphere model to remove them was
      applied.

      Second, smoothing of the sphere radiance model that had been used
      in version 1 was performed in an attempt to correct anomalously
      high derived Mars radiances in the 2 channels closest in wavelength
      to 2700 nm.

      Third, in version 2 leaked second order light at >2700 nm was
      corrected. This was performed by subtracting scaled radiances from
      one-half the wavelength using the 'LL' CDR4s.

      Fourth, in version 1, the bad pixel correction had been set to
      too sensitive a threshold, so that abrupt brightness boundaries
      often triggered the bad pixel correction, leading to aliasing at
      those boundaries. The threshold for identifying a bad pixel in the
      median-scene image was increased from 1.4 to 2.4 standard
      deviations in version 2.

      Fifth, slight misalignment of the instrument aperture with the ground
      calibration source introduced systematic error into the sphere
      radiance model. This preferentially affected wavelengths below
      1500 nm, and in version 1 scene radiances it had introduced a broad,
      sigmoidal-shaped bump centered near 1400 nm. For version 2, this
      effect was modeled using grating theory, and a one-time correction
      for it was applied to the sphere radiance model.

      Sixth, in version 1, no attempt had been made to remove scattered
      light from the grating at IR wavelengths, either in scene data or
      in observations of the internal integrating sphere. The effect was
      overestimation of IR scene radiance at <1900 nm. In version 2, the
      correction was introduced. However, unlike the VNIR correction for
      grating scatter, for the IR detector transient bad pixels render a
      correction of this form noisy. Instead this correction was derived
      using a bland, dusty scene on Mars, applying the correction both to
      the scene and to the accompanying sphere measurement. The
      extrapolation across the field of view uses a function resembling
      that at the longest VNIR wavelengths. The correction was then
      median filtered and multiplied into the sphere radiometric model.
      Given the derivation of the IR scattered light correction, it is
      most accurate for uniformly illuminated scenes.

      There are two alternate versions of IR version 2 specifically
      for special types of observations. The bland Mars scenes used to
      measure non-uniformity are processed to version 9 TRDRs.  The
      version 9 processing is the same as version 2 except it doesn't
      include the non-uniformity correction.  Deimos, Phobos, and any
      other pointlike or compact targets like stars are processed to
      version 8 TRDRs.  The major difference in the processing for such
      scenes is that within-scene scattered light from the grating is
      much less, so the correction for intra-scene scattered light is
      skipped.  The IR version 8 processing uses special version 8 SS
      CDR4s.

      Version 3 is the current version of IR radiometric calibration,
      released in late 2010. There are several improvements over version
      2.

      First, the effects of shutter mirror position irreproducibility
      were found to vary with each sphere measurement, that was in
      turn applied to a group of scene measurements. That group was
      in many cases systematically too high or low in radiance at
      1000-1700 nm. Instead, each sphere measurement is corrected
      using as a measurement of the shutter mirror effect the shape of
      the VNIR zone boundary artifact. This approach was found to yield
      greatly improved reproducibility of different TRDRs covering the
      same scene.

      Second, the estimation of leaked 2nd order light at IR wavelengths
      near 3100-3300 nm was improved. The algorithm used in version 2 had
      propagated an artifact at the zone 1 - zone 2 filter boundary near
      1630 nm, where the leakage was being estimated, to 3180 nm, where
      the leakage was being removed. Interpolation of the leakage correction
      over a wider band at the zone 1 - zone 2 boundary was found to
      greatly reduce the artifact near 3180 nm.

      Third, the correction for atmospheric water vapor in ground
      calibration measurements was modified to further reduce spectral
      artifacts near 1850 and 2550 microns.

      Fourth, the width of the band near 2000 nm within which flat-fielding
      is not applied was increased, to remove distortions in Mars'
      atmospheric CO2 absorption near 1830 nm.

      For I/F cubes - but not radiance cubes - the ratio shift correction
      and iterative kernel filter were introduced to further reduce the
      magnitude of propagated artifacts of ground calibration and noise
      in flight data. (The radiance cube is a 'control' in case the
      ratio shift correction and iterative kernel filter
      introduce their own artifacts.)


      Known Issues with IR Radiometric Calibration
      ============================================

      Due to spectral smile, mineralogic absorptions located where
      instrument response varies steeply with wavelength exhibit variation
      with field angle. This is most prominent at the edge of atmospheric
      gas absorptions near 1400 and 2000 nm.

      The responsivity correction at IR wavelengths 3000-3920 is suspected
      to contain low wavelength frequency errors, perhaps leading to a
      broad 'bump' centered near 3400 nm. This is currently under
      investigation and may be corrected in a future version of the IR
      radiometric calibration.

      Summary of VNIR and IR 'bad channels'
      =====================================

      The following channels can be routinely excluded:
      VNIR: wavelengths less than 410, 644-684, greater than 1023 nm
      IR: wavelengths less than 1021 nm, 2694 and 2701 nm, and greater
      than 3924 nm

      The following channels may be 'degraded' and their quality is
      observation-dependent. Caution is recommended but the data
      may be valid.

      VNIR:

      Wavelengths less than 442 nm (due to artifacts from scattered
      light correction in very contrasty scenes)

      Wavelengths greater than or equal to 970 nm (radiances are
      observed to misalign with IR radiances; the reason is uncertain
      but may be related to uncorrected effects of beamsplitter
      temperature)

      IR:

      Wavelengths less than 1047 nm radiances are
      observed to misalign with IR radiances; the reason is uncertain
      but may be related to uncorrected effects of beamsplitter
      temperature)

      Wavelengths 2660-2800 nm (the reason is uncertain but may be due
      to problems with correction of water vapor in measurements of the
      ground calibration sources)

      The shape of the spectrum at 3100-3800 nm is suspect and there
      may be a broad, low 'bump'.


    Review
    ======
      This archival data set will be examined by a peer review panel
      prior to its acceptance by the Planetary Data System (PDS).  The
      peer review will be conducted in accordance with PDS procedures.


    Data Coverage and Quality
    =========================
      For each observation, every EDR is compared against frame-by-frame
      predictions of commanded instrument state. The results of the
      comparison are written as a data validation report that
      accompanies the EDRs for that observation.

      In the case of a hardware or configuration discrepancy (shutter
      position, lamp status or level, pixel binning, frame rate, channel
      selection, power status of detectors), processing of the image
      data to TRDR level does not occur in order to avoid introducing
      invalid results, and DDRs are not created. Also, missing frames or
      portions of frames are replaced with a value of 65535 (this cannot
      be a valid data value). That portion of the EDR is not further
      processed, and it also is propagated to a value of 65535 in all
      layers of the TRDR.

      Only a subset of instrument configurations represent 'scene' data,
      as indicated by the keyword MRO:ACTIVITY_ID. Only scene data
      aimed at planetary targets have corresponding TRDRs.


    Limitations
    ===========
      None.