<a id="recipe-overview"></a>

# Overview

The recipe is the main control file of ESMValTool.
It is the only required argument for the  `esmvaltool` command line program.
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](#recipe-documentation), [Datasets](),
[Preprocessors](), and [Diagnostics](), respectively.

<a id="recipe-documentation"></a>

## Recipe section: `documentation`

The documentation section includes:

- The recipe’s author’s user name (`authors`, matching the definitions in the
  [References configuration file](../quickstart/configure.html.md#config-ref))
- The recipe’s maintainer’s user name (`maintainer`, matching the definitions in the
  [References configuration file](../quickstart/configure.html.md#config-ref))
- The title of the recipe (`title`)
- A description of the recipe (`description`, written in MarkDown format)
- A list of scientific references (`references`, matching the definitions in
  the [References configuration file](../quickstart/configure.html.md#config-ref))
- the project or projects associated with the recipe (`projects`, matching
  the definitions in the [References configuration file](../quickstart/configure.html.md#config-ref))

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

```yaml
documentation:
  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.

  authors:
    - demo_le

  maintainer:
    - demo_le

  references:
    - demora2018gmd

  projects:
    - ukesm
```

#### NOTE
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](https://github.com/ESMValGroup/ESMValTool/blob/main/esmvaltool/config-references.yml),
see [References configuration file](../quickstart/configure.html.md#config-ref).
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`.

<a id="datasets"></a>

## Recipe section: `datasets`

The `datasets` section includes dictionaries that, via [facets](../reference/facets.html.md#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](https://yaml.org/refcard.html) 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](https://en.wikipedia.org/wiki/ISO_8601), 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](#timerange-examples).
- model grid (native grid `grid: gn` or regridded grid `grid: gr`, for
  CMIP6 data only).

For example, a datasets section could be:

```yaml
datasets:
  - {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: CCLM4-8-17, project: CORDEX, domain: EUR-11, driver: CNRM-CERFACS-CNRM-CM5, exp: historical, ensemble: r1i1p1, institute: CLMcom, rcm_version: v1, timerange: '2001/2004'}
  - {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: '*'}
```

<a id="dataset-wildcards"></a>

### Automatically populating a recipe with all available datasets

It is possible to use [`glob`](https://docs.python.org/3/library/glob.html#module-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](../quickstart/find_data.html.md#cmor-drs) 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, the
[configuration option](../quickstart/configure.html.md#config-options) `search_data` should be set to
`complete` and an [ESGF data source](../quickstart/configure.html.md#config-esgf) needs to be configured.

For more control over which datasets are selected, it is recommended to use
a Python script or [Jupyter notebook](https://jupyter.org/) to compose
the recipe.
See the [Composing recipes](../notebooks/composing-recipes.html.md) notebook 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:

```yaml
datasets:
  - 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](#timerange-examples) 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.

<a id="supplementary-variables"></a>

### 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](#preprocessors) need this information to
work.
For example, the [area_statistics](preprocessor.html.md#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:

```yaml
datasets:
  - dataset: BCC-ESM1
    project: CMIP6
    exp: historical
    ensemble: r1i1p1f1
    grid: gn
    supplementary_variables:
      - 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.

<a id="supplementary-dataset-wildcards"></a>

### 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:

```yaml
datasets:
  - dataset: BCC-ESM1
    project: CMIP6
    exp: historical
    ensemble: r1i1p1f1
    grid: gn
    supplementary_variables:
      - 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](#dataset-wildcards),
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](preprocessor.html.md#preprocessors-using-supplementary-variables),
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)](preprocessor.html.md#preprocessors-using-supplementary-variables) 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

```yaml
datasets:
  - {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:

```yaml
datasets:
  - {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:

```yaml
datasets:
  - {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:

```yaml
datasets:
  - {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:

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

<a id="timerange-examples"></a>

### Time ranges

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

| timerange                           | effect                                                             |
|-------------------------------------|--------------------------------------------------------------------|
| `'1980/1982'`                       | Spans from 01/01/1980 to 31/12/1982                                |
| `'198002/198205'`                   | Spans from 01/02/1980 to 31/05/1982                                |
| `'19800302/19820403'`               | Spans from 02/03/1980 to 03/04/1982                                |
| `'19800504T100000/19800504T110000'` | Spans from 04/05/1980 at 10h to 11h                                |
| `'1980/P5Y'`                        | Starting from 01/01/1980,<br/>spans 5 years                        |
| `'P2Y5M/198202`                     | Ending at 28/02/1982,<br/>spans 2 years and 5 months               |
| `'*'`                               | Finds all available years                                          |
| `'*/1982`                           | Finds first available point,<br/>spans to 31/12/1982               |
| `'*/P6Y`                            | Finds first available point,<br/>spans 6 years from it             |
| `'198003/*`                         | Starting from 01/03/1980, spans until the<br/>last available point |
| `'P5M/*`                            | Finds last available point,<br/>spans 5 months backwards from it   |

#### NOTE
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.

<a id="cordex-datasets"></a>

### CORDEX datasets

The horizontal coordinates of CORDEX data are often not accurate. Therefore,
the tool provides the option to specify that the standard CORDEX grid for the
domain should be used instead of the grid defined in the files. This can be
enabled by adding `use_standard_grid: true` and disabled by adding
`use_standard_grid: false` to the dataset definition of the CORDEX data.

For example:

```yaml
datasets:
  - dataset: CCLM4-8-17
    rcm_version: v1
    driver: CNRM-CERFACS-CNRM-CM5
    institute: CLMcom
    project: CORDEX
    domain: EUR-11
    exp: historical
    ensemble: r1i1p1
    use_standard_grid: true
```

will use the standard CORDEX grid for the EUR-11 domain, as defined by the
[py-cordex](https://py-cordex.readthedocs.io) package for datasets on a
rotated pole grid. For datasets on a Lambert conformal grid, ESMValCore provides
an algorithm to calculate the standard grid, but this only works if the
[grid mapping](https://cfconventions.org/Data/cf-conventions/cf-conventions-1.13/cf-conventions.html#grid-mappings-and-projections)
is correct.

This feature is disabled by default, but enabled by default for models with
known issues through the default
[CORDEX extra facets file](https://github.com/ESMValGroup/ESMValCore/blob/main/esmvalcore/config/configurations/defaults/extra_facets_cordex.yml).

#### WARNING
If `use_standard_grid: true` is used, the tool will overwrite the existing grid
information. It is recommended to visualize the resulting data and compare it
to easily recognizable features (e.g. coastlines) to make sure the grid is
correct before using it in an aggregated manner.

Example Python script to visualize the grid of a CORDEX dataset and compare it
to coastlines:

```python
from esmvalcore.dataset import Dataset
import iris.quickplot
import matplotlib.pyplot as plt

ds = Dataset(
    short_name="sftlf",
    mip="fx",
    project="CORDEX",
    domain="EUR-11",
    dataset="ALADIN63",
    driver="CNRM-CERFACS-CNRM-CM5",
    institute="CNRM",
    exp="historical",
    ensemble="r1i1p1",
    rcm_version="v2",
    use_standard_grid=True,
)
cube = ds.load()

plt.figure()
iris.quickplot.pcolormesh(
    cube,
    coords=["projection_x_coordinate", "projection_y_coordinate"],
)
plt.gca().coastlines()

plt.figure()
iris.quickplot.pcolormesh(
    cube,
    coords=["longitude", "latitude"],
)
plt.gca().coastlines()

plt.show()
```

<a id="preprocessors"></a>

## 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](preprocessor.html.md#horizontal-regridding) with two
arguments, [Time manipulation](preprocessor.html.md#time-operations) with no arguments (i.e., calculating the
average over the time dimension) and [Multi-model statistics](preprocessor.html.md#multi-model-statistics) with two
arguments):

```yaml
preprocessors:
  prep_map:
    regrid:
      target_grid: 1x1
      scheme: linear
    climate_statistics:
      operator: mean
    multi_model_statistics:
      span: overlap
      statistics: [mean]
```

#### NOTE
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](preprocessor.html.md#cmor-check-and-dataset-specific-fixes)) and saving the data to disk.

Preprocessor operations will be applied using the default order
as listed in [Preprocessor functions](../api/esmvalcore.preprocessor.html.md#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.

<a id="diagnostics"></a>

## Recipe section: `diagnostics`

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

- the [variable(s)](../reference/cmor_tables.html.md#cmor-tables) 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.

<a id="tasks"></a>

### 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

```yaml
diagnostics:
  diagnostic_name:
    title: Air temperature tutorial diagnostic
    description: A longer description can be added here.
    themes:
      - phys
    realms:
      - atmos
    variables:
      variable_name:
        short_name: ta
        preprocessor: preprocessor_name
        mip: Amon
    scripts:
      script_name:
        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`, or `.ncl` for Python, R, and NCL diagnostics
respectively, or be executable if it is written in any other language.

<a id="ancestor-tasks"></a>

### 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.

```yaml
diagnostics:
  diagnostic_1:
    variables:
      airtemp:
        short_name: ta
        preprocessor: preprocessor_name
        mip: Amon
    scripts:
      script_a:
        script: diagnostic_a.py
  diagnostic_2:
    variables:
      precip:
        short_name: pr
        preprocessor: preprocessor_name
        mip: Amon
    scripts:
      script_b:
        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`).

<a id="task-priority"></a>

### 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 [configuration option](../quickstart/configure.html.md#config-options)
`max_parallel_tasks`.
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](../reference/cmor_tables.html.md#cmor-tables)/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`.

```yaml
diagnostics:

  diag_map:
    title: Global Ocean Surface regridded temperature map
    description: Add a longer description here.
    variables:
      tos: # Temperature at the ocean surface
        preprocessor: prep_map
        start_year: 2001
        end_year: 2003
    scripts:
      Global_Ocean_Surface_regrid_map:
        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:

```yaml
scripts:
  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:

```python
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:
        main(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:

```yaml
diagnostics:

  myFirstDiag:
    title: Let's do some science!
    description: John Doe wrote a funny diagnostic
    variables:
      tos: # Temperature at the ocean surface
        preprocessor: prep_map
        start_year: 2001
        end_year: 2003
    scripts:
      JoeDiagFunny:
        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.

### Reusing 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:

```yaml
scripts:
  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`.