CRDS Integration in CAL Code

For JWST, the Calibration Reference Data System (CRDS) is directly integrated with calibration steps and pipelines resulting in conventions for how Steps and Pipelines should be written.

Step Attribute .reference_file_types

Each calibration Step is required to define an attribute or property named reference_file_types which defines the CRDS reference file types that are required for running the calibration step. Note that for some Steps the reference file types actually used vary so the minimal list of required types may not be known if no science data is defined.

Note that the Step parameter reference files do not need to be specified. These are automatically requested in the Step architecture.

CRDS Prefetch

To ensure all reference files required by a pipeline are available prior to processing, Pipelines perform a “pre-fetch” of all references required by any Step in the Pipeline. This generic Pipeline behavior is intended to prevent processing which fails due to missing CRDS files after running for lengthy periods.

When CRDS_SERVER_URL and CRDS_PATH are properly configured, CRDS will download and locally cache the minimal set of references required to calibrate specific data using a particular pipeline. This configuration supports remote processing for users with no access or ineffecient access to the STScI local network.

The pre-fetch also enables CRDS to report on all reference file assignment and availability problems a pipeline will encounter in a single CAL run. This is required in I&T scenarios where the total number of pipeline runs is very limited (often weekly) so solving as many reference file issues per run as possible is needed.

While the prefetch runs for onsite users, since the default CRDS configuration points to a complete CRDS cache, no downloads will occur.

Step Method .get_reference_file()

During processing individual Steps make secondary calls to CRDS via the get_reference_file(input_file, reference_file_type) method to fetch the cache paths of individual reference files. If no file is applicable, CRDS returns the value ‘N/A’ which is sometimes used to skip related Step processing. If there is an error determining a reference file, CRDS will raise an exception stopping the calibration; typically these occur due to missing reference files or incorrectly specified dataset parameters.

While get_reference_file() returns the absolute path of CRDS reference files, reference file assignments are recorded in output products using a crds:// URI prefix which translates to roughly “the path of this file in the local cache you’ve defined using CRDS_PATH, or under /grp/crds/cache if you didn’t define CRDS_PATH”.

Best Reference Matching

The Calibration Reference Data System (CRDS) assigns the best reference files needed to process a data set based on the dataset’s metadata (FITS headers) and plain text CRDS rules files.

CRDS rules and references are organized into a 4 tiered hierarchical network of versioned files consisting of:

  • .pmap - The overall context for the pipeline (i.e. all instruments)

  • .imap - The rules for all reference types of one instrument

  • .rmap - The rules for all reference files of one type of one instrument

  • .fits,.asdf,.json - Individual reference files assigned by .rmaps

Based on dataset parameters, CRDS traverses the hierarchy of rules files, generally starting from the .pmap and descending until a particular reference file is assigned.

Visiting the JWST operational website here:

and opening up instrument panes of the Operational References display can rapidly give an idea about how reference files should be assigned.

CRDS Parameter Naming

For the sake of brevity, the CRDS website often refers to matching parameters using truncated names intended to give the gist of a parameter.

Within CRDS rules for JWST, CRDS refers to parameters using jwst datamodels attribute paths converted to capitalized strings analogous to FITS keywords. For instance the datamodels attribute:

corresponds to CRDS rules parameter name:


and FITS keyword:


Using e.g. ‘META.INSTRUMENT.NAME’ permits consistent naming regardless of the underlying file format (.fits vs. .asdf vs. .json).

When creating or accessing reference files, Python code uses the lower case object path to populate an attribute corresponding to the upper case string.

Example .pmap contents

Generally CRDS reference file lookups begin with a .pmap (context) file.

The .pmap’s serial number describes the overall version of rules for a pipeline.

The contents of context jwst_0493.pmap are shown below:

 header = {
    'mapping' : 'PIPELINE',
    'observatory' : 'JWST',
    'name' : 'jwst_0493.pmap',
    'parkey' : ('META.INSTRUMENT.NAME',),

selector = {
    'FGS' : 'jwst_fgs_0073.imap',
    'MIRI' : 'jwst_miri_0158.imap',
    'NIRCAM' : 'jwst_nircam_0112.imap',
    'NIRISS' : 'jwst_niriss_0117.imap',
    'NIRSPEC' : 'jwst_nirspec_0173.imap',
    'SYSTEM' : 'jwst_system_0017.imap',

Based on the parameter META.INSTRUMENT.NAME (INSTRUME) CRDS selects an appropriate .imap for further searching.

In all CRDS rules files, the header’s parkey field defines the parameter names used to select a file. These parkey names correspond to the values shown in the selector’s keys.

Conceptually all CRDS selectors consist of dictionaries which map parameter values to either a file or a sub-selector.

If META.INSTRUMENT.NAME=NIRSPEC, then CRDS would choose jwst_nirspec_0173.imap to continue it’s search.

Example .imap contents

A .imap file defines the appropriate version of .rmap to search for each reference type supported by the corresponding instrument. Below is an example .imap taken from NIRSPEC:

header = {
  'mapping' : 'INSTRUMENT',
  'instrument' : 'NIRSPEC',
  'name' : 'jwst_nirspec_0173.imap',
  'parkey' : ('REFTYPE',),

selector = {
  'AREA' : 'jwst_nirspec_area_0010.rmap',
  'BARSHADOW' : 'jwst_nirspec_barshadow_0002.rmap',
  'CAMERA' : 'jwst_nirspec_camera_0015.rmap',
  'PATHLOSS' : 'jwst_nirspec_pathloss_0003.rmap',
  'WAVECORR' : 'jwst_nirspec_wavecorr_0003.rmap',
  'WAVELENGTHRANGE' : 'jwst_nirspec_wavelengthrange_0015.rmap',
  'WFSSBKG' : 'N/A',

A value of N/A indicates that a particular reference type is not yet used by this instrument and CRDS will return ‘N/A’ instead of a filename.

If the requested REFTYPE was PATHLOSS, CRDS would continue it’s search with jwst_nirspec_pathloss_0003.rmap.

Example .rmap contents

Slightly modified contents of jwst_nirspec_pathloss_0003.rmap are shown below:

header = {
   'mapping' : 'REFERENCE',
   'observatory' : 'JWST',
   'instrument' : 'NIRSPEC',
   'filekind' : 'PATHLOSS',
   'name' : 'jwst_nirspec_pathloss_0003.rmap',
   'classes' : ('Match', 'UseAfter'),

selector = Match({
   'NRS_AUTOWAVE' : 'N/A',
       '1900-01-01 00:00:00' : 'jwst_nirspec_pathloss_0001.fits',
   'NRS_IFU' : UseAfter({
       '1900-01-01 00:00:00' : 'jwst_nirspec_pathloss_0003.fits',
   'NRS_MSASPEC' : UseAfter({
       '1900-01-01 00:00:00' : 'jwst_nirspec_pathloss_0002.fits',
       '2000-01-01 00:00:00' : 'jwst_nirspec_pathloss_0007.fits',

Each class of CRDS rmap selector defines a search algorithm to be used at that stage of the reference file lookup.

Match Selector

In the example shown above, CRDS selects a nested UseAfter selector based on the value of META.EXPOSURE.TYPE (EXP_TYPE). The nested UseAfter is then used for a secondary lookup to determine the assigned reference.

Parameters which contain or-bars, e.g.:


specify groups of values for which a file is equally applicable.

In this case the file jwst_nirspec_pathloss_0001.fits can be used to calibrate either NRS_FIXEDSLIT or NRS_BRIGHTOBJ.

Or'ed parameter combinations shown in rmaps are almost identical to the or’ed parameter combinations taken from P\_ pattern keywords; the only difference is that rmaps do not specify the trailing or-bar required for P\_ keyword values.

If a parameter combination maps to the value N/A, then the reference type is not applicable for that combination and CRDS returns the value N/A instead of a filename.

UseAfter Selector

The UseAfter sub-selector applies a given reference file only to datasets which occur at or after the specified date. For cases where multiple references occur prior to a dataset, CRDS chooses the most recent reference file as best.

Based on the dataset’s values of:


CRDS will choose the appropriate reference file by comparing them to the date+time shown in the .rmap. Conceptually, the date+time shown corresponds to the value of:


from each reference file with the USEAFTER’s T replaced with a space.

  • In the example above, if the dataset defines:


then CRDS will select jwst_nirspec_pathloss_0002.fits as best.

  • In the example above, if the dataset defines:


then CRDS will select jwst_nirspec_pathloss_0007.fits as best.

  • If the dataset defines e.g.:


then no reference match exists because the observation date precedes the USEAFTER of all available reference files.

UseAfter selection is one of the rare cases where CRDS makes an apples-to-oranges match and the dataset and reference file parameters being correlated are not identical. In fact, not even the count of parameters (DATE-OBS, TIME-OBS) vs. USEAFTER is identical.

Defining Reference File Applicability

Almost all reference files supply metadata which defines how CRDS should add the file to its corresponding .rmap, i.e. each reference defines the science data parameters for which it is initially applicable.

When creating reference files, you will need to define a value for every CRDS matching parameter and/or define a pattern using the P_ version of the matching parameter.

When CRDS adds a reference file to a .rmap, it uses literal matching between the value defined in the reference file and the existing values shown in the .rmap. This enables CRDS to:

  • add files to existing categories

  • replace files in existing categories

  • create new categories of files.

Because creating new categories is an unusual event which should be carefully reviewed, CRDS issues a warning when a reference file defines a new category.

Changing .rmaps to Reassign Reference Files

While reference files generally specify their intended use, sometimes different desired uses not specified in the reference file appear over time. In CRDS it is possible to alter only a .rmap to change the category or dates for which a reference file applies.

This is a fundamental CRDS feature which enables changes to reference assignment without forcing the re-delivery of an otherwise serviceable reference file. This feature is very commonly used, and the net consequence is that .rmap categories and dates do not have to match the contents of reference files.

It is better to view CRDS matching as a comparison between dataset parameters and a .rmap. Although references do state “initial intent”, reference file metadata should not be viewed as definitive for how a file is assigned.

More Complex Matching

CRDS matching supports more complex situations than shown in the example above.

Although reference files are generally constructed so that their metadata defines the instrument modes for which they’re applicable, conceptually, the values shown in .rmaps correspond to values in the dataset. Indeed, it is possible to change the values shown in the rmap so that they differ from their corresponding values in the reference file. This makes it possible to reassign reference files rather than redelivering them.

Match Parameter Combinations

For matches using combinations of multiple parameters, the Match selector keys will be shown as tuples, e.g.:

('NRS1|NRS2', 'ANY', 'GENERIC', '1', '1', '2048', '2048')

Because this match category matches either DETECTOR=NRS1 or NRS2, this single rmap entry represents two discrete parameter combinations. With multiple pattern values (not shown here), a single match category can match many different discrete combinations.

The parkey tuple from the NIRSPEC SUPERBIAS rmap which supplied the example match case above looks like:


The first sub-tuple corresponds to the Match cases, and the second sub-tuple corresponds to the nested UseAfters.

Weighted Matching

It’s possible for CRDS to complete it’s search without finding a unique match. To help resolve these situations, the Match algorithm uses a weighting scheme.

Each parameter with an exact match contributes a value of 1 to the weighted sum. e.g. ‘NRS1’ matches ‘NRS1|NRS2’ exactly once patterns are accounted for.

An rmap value of ANY will match any dataset value and also has a weight of 1.

An rmap value of N/A or GENERIC will match any dataset value but have a weight of 0, contributing nothing to the strength of the match.

Conceptually, the match with the highest weighting value is used. It is possible to create rmaps where ambiguity is not resolved by the weighting scheme but it works fairly well when used sparingly and isolated to as few parameters as possible.

Typically the value GENERIC corresponds to a full frame reference file which can support the calibration of any SUBARRAY by performing a cut-out.

More Information

More information about CRDS can be found in the CRDS User’s Guide maintained on the CRDS server here: