This repository tracks the xsearch utility, which allows PCMDI users to query CMIP5 and CMIP6 data. The utility returns dataset paths for a given search criteria.

These dataset directories can be used with xarray and xcdat to open and read the data (e.g., via xcdat.open_mfdataset()). The x in xsearch denotes that the search is supposed to complement xarray and xcdat.

xsearch makes use of json files that store dataset directories and associated metadata. The json files are organized by experiment and variable: jsonDirectory/experiment/variable.json.

You should be able to add the xsearch package to your PYTHONPATH. In your .bashrc file add:

export PYTHONPATH='/PATH/TO/XSEARCH/'

If the xsearch version is incremented you will receive a warning on import:

Note that xsearch has been updated to version 0.0.2.

You were last using version 0.0.1.

xsearch will create a file in your home directory to support version checking.

Alternatively, you can download the contents of this repository and import the package (or the search.py file). In this case, we suggest you watch the repository (in GitHub) so that you know when there are version changes. You can either add export PYTHONPATH='/PATH/TO/YOUR/XSEARCH/' to your ~/.bashrc file or use the following command to include xsearch in your current conda environment:

python setup.py install

xsearch is built around metadata logic documented in the CMIP6 Global Attributes, DRS, Filenames, Directory Structure, and CV’s document. These logic underpin the filename and directory creation, written by CMOR and other CMIP-writing libraries. The CMIP6 CMOR Tables (e.g. Amon - designate A=atmospheric realm, and mon=monthly frequency), similar logic underpinned the CMIP5 and CMIP3 phases.

Note that this software library is designed to help PCMDI users search for CMIP data. If the search works correctly, but points to a dataset that has problems (e.g., a corrupted or incomplete dataset) these issues should not be logged as an issue / bug in this repository.

Contributions from users of this utility are critical (issue reports and contributed code to improve the package or address bugs).

import xsearch as xs

dpaths = xs.findPaths('historical', 'tas', 'mon')  # always need to specify an experiment / variable / frequency

Multiple values for cmipTable. Consider filtering by cmipTable.
Available values: Amon, ImonGre, ImonAnt

Multiple values for realm. Consider filtering by realm.
Available values: atmos, landIce

The search warns that the returned paths include data that spans multiple realms and multiple tables. You can add optional facets to select one realm/table:

dpaths = xs.findPaths('historical', 'tas', 'mon', realm='atmos', cmipTable='Amon')
print(dpaths.keys())

['/p/css03/cmip5_css01/data/cmip5/output1/BCC/bcc-csm1-1-m/historical/mon/atmos/Amon/r1i1p1/v20120709/tas/',
'/p/css03/cmip5_css01/data/cmip5/output1/BCC/bcc-csm1-1-m/historical/mon/atmos/Amon/r2i1p1/v20120709/tas/',
'/p/css03/cmip5_css01/data/cmip5/output1/BCC/bcc-csm1-1-m/historical/mon/atmos/Amon/r3i1p1/v20120709/tas/',
'/p/css03/cmip5_css01/data/cmip5/output1/BCC/bcc-csm1-1/historical/mon/atmos/Amon/r1i1p1/v1/tas/',
'/p/css03/cmip5_css01/data/cmip5/output1/BCC/bcc-csm1-1/historical/mon/atmos/Amon/r2i1p1/v1/tas/',
'/p/css03/cmip5_css01/data/cmip5/output1/BCC/bcc-csm1-1/historical/mon/atmos/Amon/r3i1p1/v1/tas/',
'/p/css03/cmip5_css01/data/cmip5/output1/CNRM-CERFACS/CNRM-CM5/historical/mon/atmos/Amon/r10i1p1/v20110901/tas/',
'/p/css03/cmip5_css01/data/cmip5/output1/CNRM-CERFACS/CNRM-CM5/historical/mon/atmos/Amon/r1i1p1/v20110901/tas/',
'/p/css03/cmip5_css01/data/cmip5/output1/CNRM-CERFACS/CNRM-CM5/historical/mon/atmos/Amon/r2i1p1/v20110901/tas/',
...

Note that, by default, dpaths is a dictionary and that each key is a dataset path. Each dictionary entry contains dataset metadata:

key = list(dpaths.keys())[0]  # get the first key
print(key)  # print dataset
print(dpaths[key])  # print metadata for this dataset

/p/css03/cmip5_css01/data/cmip5/output1/CNRM-CERFACS/CNRM-CM5/historical/mon/atmos/Amon/r10i1p1/v20110901/tas/

{'keyid': 'CMIP5.CMIP.CNRM-CERFACS.CNRM-CM5.historical.r10i1p1.Amon.atmos.mon.tas.gu.glb-z1-gu.v20110901',
'mip_era': 'CMIP5',
'activity': 'CMIP',
'institute': 'CNRM-CERFACS',
'model': 'CNRM-CM5',
'experiment': 'historical',
'member': 'r10i1p1',
...

Each entry in the returned dictionary is a nested dictionary in which the dataset path is the key with the following entries: keyid, mip_era, activity, institute, model, experiment, member, cmipTable, realm, frequency, variable, grid, gridLabel, version, timepoints, nfiles, nc_creation_date, sumfilesize_bytes, has_zero_size_files, created, modified, accessed, retired, retire_datetime, retracted, retracted_datetime, ignored, ignored_datetime, comment, unique, alternate_paths.

For example, to get the model of one dataset directory, p, we would use:

print(dpaths[p]['model'])

CNRM-CM5

Several functions in xsearch can further use the full metadata dictionary returned by xsearch. One function allows the user to get all values for a given search facet:

xs.getValuesForFacet(dpaths, 'model', 'E3SM*')

['/p/css03/esgf_publish/CMIP6/CMIP/E3SM-Project/E3SM-1-0/historical/r1i1p1f1/Amon/tas/gr/v20190913/',
'/p/css03/esgf_publish/CMIP6/CMIP/E3SM-Project/E3SM-1-0/historical/r2i1p1f1/Amon/tas/gr/v20190729/',
'/p/css03/esgf_publish/CMIP6/CMIP/E3SM-Project/E3SM-1-0/historical/r3i1p1f1/Amon/tas/gr/v20190729/',
...

This function supports wildcard searches. Another search function allows the users to get all values for a given facet:

 xs.getGroupValues(dpaths, 'model')

['CNRM-CM5', 'bcc-csm1-1', 'bcc-csm1-1-m', 'GFDL-CM2p1', 'GFDL-ESM2G', 'GFDL-ESM2M', ...

Other search facets include: mip_era (CMIP5 or CMIP6), activity (e.g., CMIP or ScenarioMIP), institute (e.g., E3SM-Project), model (e.g., E3SM-1-1), member (e.g., r1i1p1f1), grid (e.g., gn or gr), or gridLabel (e.g., glb-z1-gr).

Searches can also include a wildcard (*) in required search terms or for additional facets:

dpaths = xs.findPaths('ssp*', 'tas', 'mon', cmipTable='Amon', realm='atmos', activity='Scenario*')
print(xs.getGroupValues(dpaths, 'experiment'))

['ssp119', 'ssp370', 'ssp245', 'ssp126', 'ssp434', 'ssp534-over', 'ssp585', 'ssp460']

The user can also specify a number of optional arguments:

• deduplicate (default True): flag whether to de-duplicate so that a single dataset path is returned for each unique combination of model and member. If one of multiple possible datasets are chosen, the alternates are listed in alternate_paths (see fullMetadata option). Make sure to read the note on de-duplication below.
  • criteria: The ordering of criteria used to deduplicate. The default is ["version", "timepoints", "nc_creation_date", "esgf_publish", "gr"], which orders by a) newest version, b) number of timesteps, the c) creation_date embedded in netCDF file metadata, and a prioritization for data that has been d) published on ESGF and e) has the grid label gr.
• printDuplicates (default False): If True, all potential paths will be printed and paths that are chosen will be denoted with an asterisk
• verbose (default True): flag to print out information during the search
• lcpath (default False): flag to adjust the data directories to reflect the LC (Livermore Computing/LLNL HPC) mountpoint
• filterRetired (default True): flag to filter out datasets that are retired (datasets are retired if they are moved)
• filterRetracted (default True): flag to filter out datasets that are retracted
• filterIgnored (default True): flag to filter out datasets that have been marked as ignored
• jsonDir (str): This defaults to the current location for the json metadata files, but the user can modify this
• fullMetadata (default True): flag to specify the datatype returned. If true, xsearch will simply return a dictionary with complete metadata for selected paths. xsearch can also return a list of paths by setting fullMetadata=False:

Note that xsearch is configured to de-duplicate for each set of paths related to each model and model realization. For example, if you have several versions of data for Model A / r1i1p1f1, xsearch will select the most recent and complete version for you.

But for a given model and realization, there may be more aspects of the data than the dataset version. For example, if we search for all variables beginning with ta for E3SM-1-0:

dpaths = xs.findPaths('historical', 'ta*', 'mon', realm='atmos', cmipTable='Amon', member='r1i1p1f1', model='E3SM-1-0', printDuplicates=True)

We get a message noting that we are spanning multiple variables:

Multiple values for variable. Consider filtering by variable.
Available values: tauu, tauv, tasmin, tasmax, ta, tas

xsearch chooses one dataset (of the datasets spanning several possible variables) for a given member (r1i1p1f1):

* /p/user_pub/work/CMIP6/CMIP/E3SM-Project/E3SM-1-0/historical/r1i1p1f1/Amon/ta/gr/v20220108/

duplicate: /p/user_pub/work/CMIP6/CMIP/E3SM-Project/E3SM-1-0/historical/r1i1p1f1/Amon/tauu/gr/v20190913/
duplicate: /p/user_pub/work/CMIP6/CMIP/E3SM-Project/E3SM-1-0/historical/r1i1p1f1/Amon/tauv/gr/v20190913/
duplicate: /p/user_pub/work/CMIP6/CMIP/E3SM-Project/E3SM-1-0/historical/r1i1p1f1/Amon/ta/gr/v20191220/
duplicate: /p/css03/esgf_publish/CMIP6/CMIP/E3SM-Project/E3SM-1-0/historical/r1i1p1f1/Amon/tas/gr/v20190913/
...

This isn't typically a problem if you specify an exact experiment, variable, and frequency.

Much of the de-duplication logic and ideas from this are from the durolib library, created by Paul Durack @durack1 to search for CMIP data. durolib handled CDAT xml files, which could be used to read in multi-file datasets. Logic to produce the xml files was refactored in xagg. xagg xmls are being phased out in favor the current approach: rapidly searchable json files, which allow the users to locate datasets and read them in with xarray based tools. Stephen Po-Chedley @pochedls produced the xsearch initial version.