After config-user.yml, the recipe.yml is the second file the user needs to pass to esmvaltool as command line option, at each run time point. Recipes contain the data and data analysis information and instructions needed to run the diagnostic(s), as well as specific diagnostic-related instructions.

Broadly, recipes contain a general section summarizing the provenance and functionality of the diagnostics, the datasets which need to be run, the preprocessors that need to be applied, and the diagnostics which need to be run over the preprocessed data. This information is provided to ESMValTool in four main recipe sections: Documentation, Datasets, Preprocessors, and Diagnostics, respectively.

Recipe section: documentation#

The documentation section includes:

For example, the documentation section of recipes/recipe_ocean_amoc.yml is the following:

  title: Atlantic Meridional Overturning Circulation (AMOC) and the drake passage current
  description: |
    Recipe to produce time series figures of the derived variable, the
    Atlantic meridional overturning circulation (AMOC).
    This recipe also produces transect figures of the stream functions for
    the years 2001-2004.

    - demo_le

    - demo_le

    - demora2018gmd

    - ukesm


Note that all authors, projects, and references mentioned in the description section of the recipe need to be included in the (locally installed copy of the) file esmvaltool/config-references.yml, see References configuration file. The author name uses the format: surname_name. For instance, John Doe would be: doe_john. This information can be omitted by new users whose name is not yet included in config-references.yml.

Recipe section: datasets#

The datasets section includes dictionaries that, via key-value pairs or “facets”, define standardized data specifications:

  • dataset name (key dataset, value e.g. MPI-ESM-LR or UKESM1-0-LL).

  • project (key project, value CMIP5 or CMIP6 for CMIP data, OBS for observational data, ana4mips for ana4mips data, obs4MIPs for obs4MIPs data, ICON for ICON data).

  • experiment (key exp, value e.g. historical, amip, piControl, rcp85).

  • mip (for CMIP data, key mip, value e.g. Amon, Omon, LImon).

  • ensemble member (key ensemble, value e.g. r1i1p1, r1i1p1f1).

  • sub-experiment id (key sub_experiment, value e.g. s2000, s(2000:2002), for DCPP data only).

  • time range (e.g. key-value start_year: 1982, end_year: 1990). Please note that yaml interprets numbers with a leading 0 as octal numbers, so we recommend to avoid them. For example, use 128 to specify the year 128 instead of 0128. Alternatively, the time range can be specified in ISO 8601 format, for both dates and periods. In addition, wildcards ('*') are accepted, which allow the selection of the first available year for each individual dataset (when used as a starting point) or the last available year (when used as an ending point). The starting point and end point must be separated with / (e.g. key-value timerange: '1982/1990'). More examples are given here.

  • model grid (native grid grid: gn or regridded grid grid: gr, for CMIP6 data only).

For example, a datasets section could be:

  - {dataset: CanESM2, project: CMIP5, exp: historical, ensemble: r1i1p1, start_year: 2001, end_year: 2004}
  - {dataset: UKESM1-0-LL, project: CMIP6, exp: historical, ensemble: r1i1p1f2, start_year: 2001, end_year: 2004, grid: gn}
  - {dataset: ACCESS-CM2, project: CMIP6, exp: historical, ensemble: r1i1p1f2, timerange: 'P5Y/*', grid: gn}
  - {dataset: EC-EARTH3, alias: custom_alias, project: CMIP6, exp: historical, ensemble: r1i1p1f1, start_year: 2001, end_year: 2004, grid: gn}
  - {dataset: CMCC-CM2-SR5, project: CMIP6, exp: historical, ensemble: r1i1p1f1, timerange: '2001/P10Y', grid: gn}
  - {dataset: HadGEM3-GC31-MM, project: CMIP6, exp: dcppA-hindcast, ensemble: r1i1p1f1, sub_experiment: s2000, grid: gn, start_year: 2000, end_year, 2002}
  - {dataset: BCC-CSM2-MR, project: CMIP6, exp: dcppA-hindcast, ensemble: r1i1p1f1, sub_experiment: s2000, grid: gn, timerange: '*'}

Automatically populating a recipe with all available datasets#

It is possible to use glob patterns or wildcards for certain facet values, to make it easy to find all available datasets locally and/or on ESGF. Note that project cannot be a wildcard.

The facet values for local files are retrieved from the directory tree where the directories represent the facets values. Reading facet values from file names is not yet supported. See CMIP data for more information on this kind of file organization.

When (some) files are available locally, the tool will not automatically look for more files on ESGF. To populate a recipe with all available datasets from ESGF, search_esgf should be set to always in the user configuration file.

For more control over which datasets are selected, it is recommended to use a Python script or Jupyter notebook to compose the recipe. See Composing recipes for an example. This is particularly useful when specific relations are required between datasets, e.g. when a dataset needs to be available for multiple variables or experiments.

An example recipe that will use all CMIP6 datasets and all ensemble members which have a 'historical' experiment could look like this:

  - project: CMIP6
    exp: historical
    dataset: '*'
    institute: '*'
    ensemble: '*'
    grid: '*'

After running the recipe, a copy specifying exactly which datasets were used is available in the output directory in the run subdirectory. The filename of this recipe will end with _filled.yml.

For the timerange facet, special syntax is available. See Time ranges for more information.

If populating a recipe using wildcards does not work, this is because there were either no files found that match those facets, or the facets could not be read from the directory name or ESGF.

Defining supplementary variables (ancillary variables and cell measures)#

It is common practice to store ancillary variables (e.g. land/sea/ice masks) and cell measures (e.g. cell area, cell volume) in separate datasets that are described by slightly different facets. In ESMValCore, we call ancillary variables and cell measures “supplementary variables”. Some preprocessor functions need this information to work. For example, the area_statistics preprocessor function needs to know area of each grid cell in order to compute a correctly weighted statistic.

To attach these variables to a dataset, the supplementary_variables keyword can be used. For example, to add cell area to a dataset, it can be specified as follows:

  - dataset: BCC-ESM1
    project: CMIP6
    exp: historical
    ensemble: r1i1p1f1
    grid: gn
      - short_name: areacella
        mip: fx
        exp: 1pctCO2

Note that the supplementary variable will inherit the facet values from the main dataset, so only those facet values that differ need to be specified.

Automatically selecting the supplementary dataset#

When using many datasets, it may be quite a bit of work to find out which facet values are required to find the corresponding supplementary data. The tool can automatically guess the best matching supplementary dataset. To use this feature, the supplementary dataset can be specified as:

  - dataset: BCC-ESM1
    project: CMIP6
    exp: historical
    ensemble: r1i1p1f1
    grid: gn
      - short_name: areacella
        mip: fx
        exp: '*'
        activity: '*'
        ensemble: '*'

With this syntax, the tool will search all available values of exp, activity, and ensemble and use the supplementary dataset that shares the most facet values with the main dataset. Note that this behaviour is different from using wildcards in the main dataset, where they will be expanded to generate all matching datasets. The available datasets are shown in the debug log messages when running a recipe with wildcards, so if a different supplementary dataset is preferred, these messages can be used to see what facet values are available. The facet values for local files are retrieved from the directory tree where the directories represent the facets values. Reading facet values from file names is not yet supported. If wildcard expansion fails, this is because there were either no files found that match those facets, or the facets could not be read from the directory name or ESGF.

Automatic definition of supplementary variables#

If an ancillary variable or cell measure is needed by a preprocessor function, but it is not specified in the recipe, the tool will automatically make a best guess using the syntax above. Usually this will work fine, but if it does not, it is recommended to explicitly define the supplementary variables in the recipe.

To disable this automatic addition, define the supplementary variable as usual, but add the special facet skip with value true. See Supplementary variables (ancillary variables and cell measures) for an example recipe.

Saving ancillary variables and cell measures#

By default, ancillary variables and cell measures will be removed from the main variable before saving it to file because they can be as big as the main variable. To keep the supplementary variables, disable the preprocessor function that removes them by setting remove_supplementary_variables: false in the preprocessor profile in the recipe.

Concatenating data corresponding to multiple facets#

It is possible to define the experiment as a list to concatenate two experiments. Here it is an example concatenating the historical experiment with rcp85

  - {dataset: CanESM2, project: CMIP5, exp: [historical, rcp85], ensemble: r1i1p1, start_year: 2001, end_year: 2004}

It is also possible to define the ensemble as a list when the two experiments have different ensemble names. In this case, the specified datasets are concatenated into a single cube:

  - {dataset: CanESM2, project: CMIP5, exp: [historical, rcp85], ensemble: [r1i1p1, r1i2p1], start_year: 2001, end_year: 2004}

Short notation of ensemble members and sub-experiments#

ESMValTool also supports a simplified syntax to add multiple ensemble members from the same dataset. In the ensemble key, any element in the form (x:y) will be replaced with all numbers from x to y (both inclusive), adding a dataset entry for each replacement. For example, to add ensemble members r1i1p1 to r10i1p1 you can use the following abbreviated syntax:

  - {dataset: CanESM2, project: CMIP5, exp: historical, ensemble: "r(1:10)i1p1", start_year: 2001, end_year: 2004}

It can be included multiple times in one definition. For example, to generate the datasets definitions for the ensemble members r1i1p1 to r5i1p1 and from r1i2p1 to r5i1p1 you can use:

  - {dataset: CanESM2, project: CMIP5, exp: historical, ensemble: "r(1:5)i(1:2)p1", start_year: 2001, end_year: 2004}

Please, bear in mind that this syntax can only be used in the ensemble tag. Also, note that the combination of multiple experiments and ensembles, like exp: [historical, rcp85], ensemble: [r1i1p1, “r(2:3)i1p1”] is not supported and will raise an error.

The same simplified syntax can be used to add multiple sub-experiments:

  - {dataset: MIROC6, project: CMIP6, exp: dcppA-hindcast, ensemble: r1i1p1f1, sub_experiment: s(2000:2002), grid: gn, start_year: 2003, end_year: 2004}

Time ranges#

When using the timerange tag to specify the start and end points, possible values can be as follows:




Spans from 01/01/1980 to 31/12/1982


Spans from 01/02/1980 to 31/05/1982


Spans from 02/03/1980 to 03/04/1982


Spans from 04/05/1980 at 10h to 11h


Starting from 01/01/1980, spans 5 years


Ending at 28/02/1982, spans 2 years and 5 months


Finds all available years


Finds first available point, spans to 31/12/1982


Finds first available point, spans 6 years from it


Starting from 01/03/1980, spans until the last available point


Finds last available point, spans 5 months backwards from it


Please make sure to use a consistent number of digits for the start and end point when using timerange, e.g., instead of 198005/2000, use 198005/200012. Otherwise, it might happen that ESMValTool does not find your data even though the corresponding years are available. This also applies to wildcards: Wildcards are usually resolved using the timerange in the file name. If this is given in the form YYYYMM, then the other time point in timerange needs to be in the same format, e.g., use */200012 instead of */2000 in this case. If you use wildcards and get an unexpected error about missing data, have a look at the resolved timerange in the error message (ERROR No input files found for variable {'timerange': '197901/2000', ...}) and make sure that the number of digits in it is consistent.

Note that this section is not required, as datasets can also be provided in the Diagnostics section.

Recipe section: preprocessors#

The preprocessor section of the recipe includes one or more preprocessors, each of which may call the execution of one or several preprocessor functions.

Each preprocessor section includes:

  • A preprocessor name (any name, under preprocessors);

  • A list of preprocessor steps to be executed (choose from the API);

  • Any or none arguments given to the preprocessor steps;

  • The order that the preprocessor steps are applied can also be specified using the custom_order preprocessor function.

The following snippet is an example of a preprocessor named prep_map that contains multiple preprocessing steps (Horizontal regridding with two arguments, Time manipulation with no arguments (i.e., calculating the average over the time dimension) and Multi-model statistics with two arguments):

      target_grid: 1x1
      scheme: linear
      operator: mean
      span: overlap
      statistics: [mean]


In this case no preprocessors section is needed the workflow will apply a default preprocessor consisting of only basic operations like: loading data, applying CMOR checks and fixes (CMORization and dataset-specific fixes) and saving the data to disk.

Preprocessor operations will be applied using the default order as listed in Preprocessor functions. Preprocessor tasks can be set to run in the order they are listed in the recipe by adding custom_order: true to the preprocessor definition.

Recipe section: diagnostics#

The diagnostics section includes one or more diagnostics. Each diagnostic section will include:

  • the variable(s) to preprocess, including the preprocessor to be applied to each variable;

  • the diagnostic script(s) to be run;

  • a description of the diagnostic and lists of themes and realms that it applies to;

  • an optional additional_datasets section.

  • an optional title and description, used to generate the title and description in the index.html output file.

The diagnostics section defines tasks#

The diagnostic section(s) define the tasks that will be executed when running the recipe. For each variable a preprocessing task will be defined and for each diagnostic script a diagnostic task will be defined. These tasks can be viewed in the main_log_debug.txt file that is produced every run. Each task has a unique name that defines the subdirectory where the results of that task are stored. Task names start with the name of the diagnostic section followed by a ‘/’ and then the name of the variable section for a preprocessing task or the name of the diagnostic script section for a diagnostic task.

A (simplified) example diagnostics section could look like

    title: Air temperature tutorial diagnostic
    description: A longer description can be added here.
      - phys
      - atmos
        short_name: ta
        preprocessor: preprocessor_name
        mip: Amon
        script: examples/diagnostic.py

Note that the example recipe above contains a single diagnostic section called diagnostic_name and will result in two tasks:

  • a preprocessing task called diagnostic_name/variable_name that will preprocess air temperature data for each dataset in the Datasets section of the recipe (not shown).

  • a diagnostic task called diagnostic_name/script_name

The path to the script provided in the script option should be either the absolute path to the script, or the path relative to the esmvaltool/diag_scripts directory.

Depending on the installation configuration, you may get an error of “file does not exist” when the system tries to run the diagnostic script using relative paths. If this happens, use an absolute path instead.

Note that the script should either have the extension for a supported language, i.e. .py, .R, .ncl, or .jl for Python, R, NCL, and Julia diagnostics respectively, or be executable if it is written in any other language.

Ancestor tasks#

Some tasks require the result of other tasks to be ready before they can start, e.g. a diagnostic script needs the preprocessed variable data to start. Thus each tasks has zero or more ancestor tasks. By default, each diagnostic task in a diagnostic section has all variable preprocessing tasks in that same section as ancestors. However, this can be changed using the ancestors keyword. Note that wildcard expansion can be used to define ancestors.

        short_name: ta
        preprocessor: preprocessor_name
        mip: Amon
        script: diagnostic_a.py
        short_name: pr
        preprocessor: preprocessor_name
        mip: Amon
        script: diagnostic_b.py
        ancestors: [diagnostic_1/script_a, precip]

The example recipe above will result in four tasks:

  • a preprocessing task called diagnostic_1/airtemp

  • a diagnostic task called diagnostic_1/script_a

  • a preprocessing task called diagnostic_2/precip

  • a diagnostic task called diagnostic_2/script_b

the preprocessing tasks do not have any ancestors, while the diagnostic_a.py script will receive the preprocessed air temperature data (has ancestor diagnostic_1/airtemp) and the diagnostic_b.py script will receive the results of diagnostic_a.py and the preprocessed precipitation data (has ancestors diagnostic_1/script_a and diagnostic_2/precip).

Task priority#

Tasks are assigned a priority, with tasks appearing earlier on in the recipe getting higher priority. The tasks will be executed sequentially or in parallel, depending on the setting of max_parallel_tasks in the User configuration file. When there are fewer than max_parallel_tasks running, tasks will be started according to their priority. For obvious reasons, only tasks that are not waiting for ancestor tasks can be started. This feature makes it possible to reduce the processing time of recipes with many tasks, by placing tasks that take relatively long near the top of the recipe. Of course this only works when settings max_parallel_tasks to a value larger than 1. The current priority and run time of individual tasks can be seen in the log messages shown when running the tool (a lower number means higher priority).

Variable and dataset definitions#

To define a variable/dataset combination that corresponds to an actual variable from a dataset, the keys in each variable section are combined with the keys of each dataset definition. If two versions of the same key are provided, then the key in the datasets section will take precedence over the keys in variables section. For many recipes it makes more sense to define the start_year and end_year items in the variable section, because the diagnostic script assumes that all the data has the same time range.

Variable short names usually do not change between datasets supported by ESMValCore, as they are usually changed to match CMIP. Nevertheless, there are small changes in variable names in CMIP6 with respect to CMIP5 (i.e. sea ice concentration changed from sic to siconc). ESMValCore is aware of some of them and can do the automatic translation when needed. It will even do the translation in the preprocessed file so the diagnostic does not have to deal with this complexity, setting the short name in all files to match the one used by the recipe. For example, if sic is requested, ESMValCore will find sic or siconc depending on the project, but all preprocessed files while use sic as their short_name. If the recipe requested siconc, the preprocessed files will be identical except that they will use the short_name siconc instead.

Diagnostic and variable specific datasets#

The additional_datasets option can be used to add datasets beyond those listed in the Datasets section. This is useful if specific datasets need to be used only by a specific diagnostic or variable, i.e. it can be added both at diagnostic level, where it will apply to all variables in that diagnostic section or at individual variable level. For example, this can be a good way to add observational datasets, which are usually variable-specific.

Running a simple diagnostic#

The following example, taken from recipe_ocean_example.yml, shows a diagnostic named diag_map, which loads the temperature at the ocean surface between the years 2001 and 2003 and then passes it to the prep_map preprocessor. The result of this process is then passed to the ocean diagnostic map script, ocean/diagnostic_maps.py.


    title: Global Ocean Surface regridded temperature map
    description: Add a longer description here.
      tos: # Temperature at the ocean surface
        preprocessor: prep_map
        start_year: 2001
        end_year: 2003
        script: ocean/diagnostic_maps.py

Passing arguments to a diagnostic script#

The diagnostic script section(s) may include custom arguments that can be used by the diagnostic script; these arguments are stored at runtime in a dictionary that is then made available to the diagnostic script via the interface link, independent of the language the diagnostic script is written in. Here is an example of such groups of arguments:

  autoassess_strato_test_1: &autoassess_strato_test_1_settings
    script: autoassess/autoassess_area_base.py
    title: "Autoassess Stratosphere Diagnostic Metric MPI-MPI"
    area: stratosphere
    control_model: MPI-ESM-LR
    exp_model: MPI-ESM-MR
    obs_models: [ERA-Interim]  # list to hold models that are NOT for metrics but for obs operations
    additional_metrics: [ERA-Interim, inmcm4]  # list to hold additional datasets for metrics

In this example, apart from specifying the diagnostic script script: autoassess/autoassess_area_base.py, we pass a suite of parameters to be used by the script (area, control_model etc). These parameters are stored in key-value pairs in the diagnostic configuration file, an interface file that can be used by importing the run_diagnostic utility:

from esmvaltool.diag_scripts.shared import run_diagnostic

# write the diagnostic code here e.g.
def run_some_diagnostic(my_area, my_control_model, my_exp_model):
    """Diagnostic to be run."""
    if my_area == 'stratosphere':
        diag = my_control_model / my_exp_model
        return diag

def main(cfg):
    """Main diagnostic run function."""
    my_area = cfg['area']
    my_control_model = cfg['control_model']
    my_exp_model = cfg['exp_model']
    run_some_diagnostic(my_area, my_control_model, my_exp_model)

if __name__ == '__main__':

    with run_diagnostic() as config:

This way a lot of the optional arguments necessary to a diagnostic are at the user’s control via the recipe.

Running your own diagnostic#

If the user wants to test a newly-developed my_first_diagnostic.py which is not yet part of the ESMValTool diagnostics library, he/she do it by passing the absolute path to the diagnostic:


    title: Let's do some science!
    description: John Doe wrote a funny diagnostic
      tos: # Temperature at the ocean surface
        preprocessor: prep_map
        start_year: 2001
        end_year: 2003
        script: /home/users/john_doe/esmvaltool_testing/my_first_diagnostic.py

This way the user may test a new diagnostic thoroughly before committing to the GitHub repository and including it in the ESMValTool diagnostics library.

Re-using parameters from one script to another#

Due to yaml features it is possible to recycle entire diagnostics sections for use with other diagnostics. Here is an example:

  cycle: &cycle_settings
    script: perfmetrics/main.ncl
    plot_type: cycle
    time_avg: monthlyclim
  grading: &grading_settings
    <<: *cycle_settings
    plot_type: cycle_latlon
    calc_grading: true
    normalization: [centered_median, none]

In this example the hook &cycle_settings can be used to pass the cycle: parameters to grading: via the shortcut <<: *cycle_settings.