Writing your own recipe

Last updated on 2024-11-26 | Edit this page

Estimated time: 45 minutes

Overview

Questions

How do I create a new recipe?
Can I use different preprocessors for different variables?
Can I use different datasets for different variables?
How can I combine different preprocessor functions?
Can I run the same recipe for multiple ensemble members?

Objectives

Create a recipe with multiple preprocessors
Use different preprocessors for different variables
Run a recipe with variables from different datasets

Introduction

One of the key strengths of ESMValTool is in making complex analyses reusable and reproducible. But that doesn’t mean everything in ESMValTool needs to be complex. Sometimes, the biggest challenge is in keeping things simple. You probably know the ‘warming stripes’ visualization by Professor Ed Hawkins. On the site https://showyourstripes.info{:target=“_blank”} you can find the same visualization for many regions in the world.

Warming stripes Shared by Ed Hawkins under a Creative Commons 4.0 Attribution International licence. Source: https://showyourstripes.info

In this episode, we will reproduce and extend this functionality with ESMValTool. We have prepared a small Python script that takes a NetCDF file with timeseries data, and visualizes it in the form of our desired warming stripes figure.

The diagnostic script that we will use is called warming_stripes.py and can be downloaded here.

Download the file and store it in your working directory. If you want, you may also have a look at the contents, but it is not necessary to do so for this lesson.

We will write an ESMValTool recipe that takes some data, performs the necessary preprocessing, and then runs this Python script.

Drawing up a plan

Previously, we saw that running ESMValTool executes a number of tasks. What tasks do you think we will need to execute and what should each of these tasks do to generate the warming stripes?

Answer

In this episode, we will need to do the following two tasks:

A preprocessing task that converts the gridded temperature data to a timeseries of global temperature anomalies
A diagnostic tasks that calls our Python script, taking our preprocessed timeseries data as input.

Building a recipe from scratch

The easiest way to make a new recipe is to start from an existing one, and modify it until it does exactly what you need. However, in this episode we will start from scratch. This forces us to think about all the steps involved in processing the data. We will also deal with commonly occurring errors through the development of the recipe.

Remember the basic structure of a recipe, and notice that each component is extensively described in the documentation under the section, [“Overview”][recipe-overview]{:target=“_blank”}:

[documentation][recipe-section-documentation]{:target=“_blank”}
[datasets][recipe-section-datasets]{:target=“_blank”}
[preprocessors][recipe-section-preprocessors]{:target=“_blank”}
[diagnostics][recipe-section-diagnostics]{:target=“_blank”}

This is the first place to look for help if you get stuck.

Open a new file called recipe_warming_stripes.yml:

BASH

nano recipe_warming_stripes.yml

Let’s add the standard header comments (these do not do anything), and a first description.

YAML

# ESMValTool
# recipe_warming_stripes.yml
---
documentation:
  description: Reproducing Ed Hawkins' warming stripes visualization
  title: Reproducing Ed Hawkins' warming stripes visualization.

Notice that yaml always requires two spaces indentation between the different levels. Pressing ctrl+o will save the file. Verify the filename at the bottom and press enter. Then use ctrl+x to exit the editor.

We will try to run the recipe after every modification we make, to see if it (still) works!

BASH

esmvaltool run recipe_warming_stripes.yml

In this case, it gives an error. Below you see the last few lines of the error message.

ERROR

...
yamale.yamale_error.YamaleError:
Error validating data '/home/users/username/esmvaltool_tutorial/recipe_warming_stripes.yml'
with schema
'/apps/jasmin/community/esmvaltool/miniconda3_py311_23.11.0-2/envs/esmvaltool/lib/python3.11/
site-packages/esmvalcore/_recipe/recipe_schema.yml'
	documentation.authors: Required field missing
2024-05-27 13:21:23,805 UTC [41924] INFO
If you have a question or need help, please start a new discussion on
https://github.com/ESMValGroup/ESMValTool/discussions
If you suspect this is a bug, please open an issue on
https://github.com/ESMValGroup/ESMValTool/issues
To make it easier to find out what the problem is, please consider attaching the
files run/recipe_*.yml and run/main_log_debug.txt from the output directory.

We can use the the log message above, to understand why ESMValTool failed. Here, this is because we missed a required field with author names. The text documentation.authors: Required field missing tells us that. We see that ESMValTool always tries to validate the recipe at an early stage. Note also the suggestion to open a GitHub issue if you need help debugging the error message. This is something most users do when they cannot understand the error or are not able to fix it on their own.

Let’s add some additional information to the recipe. Open the recipe file again, and add an authors section below the description. ESMValTool expects the authors as a list, like so:

YAML

authors:
  - lastname_firstname

To bypass a number of similar error messages, add a minimal diagnostics section below the documentation. The file should now look like:

YAML

# ESMValTool
# recipe_warming_stripes.yml
---
documentation:
  description: Reproducing Ed Hawkins' warming stripes visualization
  title: Reproducing Ed Hawkins' warming stripes visualization.

  authors:
    - doe_john
diagnostics:
  dummy_diagnostic_1:
    scripts: null

This is the minimal recipe layout that is required by ESMValTool. If we now run the recipe again, you will probably see the following error:

ERROR

ValueError: Tag 'doe_john' does not exist in section
'authors' of /apps/jasmin/community/esmvaltool/ESMValTool_2.10.0/esmvaltool/config-references.yml

Pro tip: config-references.yml

The error message above points to a file named [config-references.yml][config-references] This is where ESMValTool stores all its citation information. To add yourself as an author, add your name in the form lastname_firstname in alphabetical order following the existing entries, under the # Development team section. See the [List of authors][list-of-authors]{:target=“_blank”} section in the ESMValTool documentation for more information.

For now, let’s just use one of the existing references. Change the author field to righi_mattia, who cannot receive enough credit for all the effort he put into ESMValTool. If you now run the recipe again, you should see the final message

OUTPUT

ERROR   No tasks to run!

Although there is no actual error in the recipe, ESMValTool assumes you mistakenly left out a variable name to process and alerts you with this error message.

Adding a dataset entry

Let’s add a datasets section.

Filling in the dataset keys

Use the paths specified in the configuration file to explore the data directory, and look at the explanation of the dataset entry in the [ESMValTool documentation][recipe-section-datasets]{:target=“_blank”}. For both the datasets, write down the following properties:

project
variable (short name)
CMIP table
dataset (model name or obs/reanalysis dataset)
experiment
ensemble member
grid
start year
end year

Answers

Note that the grid key is only required for CMIP6 data, and that the extent of the historical period has changed between CMIP5 and CMIP6.

Let us start with the BCC-ESM1 dataset and add a datasets section to the recipe, listing this single dataset, as shown below. Note that key fields such as mip or start_year are included in the datasets section here but are part of the diagnostic section in the recipe example seen in Running your first recipe.

YAML

# ESMValTool
# recipe_warming_stripes.yml
---
documentation:
  description: Reproducing Ed Hawkins' warming stripes visualization
  title: Reproducing Ed Hawkins' warming stripes visualization.

  authors:
    - doe_john
datasets:
  - {dataset: BCC-ESM1, project: CMIP6, mip: Amon, exp: historical,
     ensemble: r1i1p1f1, grid: gn, start_year: 1850, end_year: 2014}

diagnostics:
  dummy_diagnostic_1:
    scripts: null

The recipe should run but produce the same message as in the previous case since we still have not included a variable to actually process. We have not included the short name of the variable in this dataset section because this allows us to reuse this dataset entry with different variable names later on. This is not really necessary for our simple use case, but it is common practice in ESMValTool.

Pro-tip: Automatically populating a recipe with all available datasets

You can select all available models for processing using glob patterns or wildcards. An example datasets section that uses all available CMIP6 models and ensemble members for the historical experiment is available [here] [include-all-datasets]{:target=“_blank”}. Note that you will have to set the search_esgf option in the config_file to always so that you can download data from ESGF nodes as needed.

Adding the preprocessor section

Above, we already described the preprocessing task that needs to convert the standard, gridded temperature data to a timeseries of temperature anomalies.

Defining the preprocessor

Have a look at the available preprocessors in the [documentation][preprocessor]{:target=“_blank”}. Write down

Which preprocessor functions do you think we should use?
What are the parameters that we can pass to these functions?
What do you think should be the order of the preprocessors?
A suitable name for the overall preprocessor

Show me the solution

We need to calculate anomalies and global means. There is an anomalies preprocessor which takes in as arguments, a time period, a reference period, and whether or not to standardize the data. The global means can be calculated with the area_statistics preprocessor, which takes an operator as argument (in our case we want to compute the mean).

The default order in which these preprocessors are applied can be seen [here][preprocessor-functions]{:target=“_blank”}: area_statistics comes before anomalies. If you want to change this, you can use the custom_order preprocessor as described [here][recipe-section-preprocessors]{:target=“_blank”}. For this example, we will keep the default order..

Let’s name our preprocessor global_anomalies.

Add the following block to your recipe file between the datasets and diagnostics block:

YAML

preprocessors:
  global_anomalies:
    area_statistics:
      operator: mean
    anomalies:
        period: month
        reference:
          start_year: 1981
          start_month: 1
          start_day: 1
          end_year: 2010
          end_month: 12
          end_day: 31
        standardize: false

Completing the diagnostics section

We are now ready to finish our diagnostics section. Remember that we want to create two tasks: a preprocessor task, and a diagnostic task. To illustrate that we can also pass settings to the diagnostic script, we add the option to specify a custom colormap.

Fill in the blanks

Extend the diagnostics section in your recipe by filling in the blanks in the following template:

YAML

diagnostics:
  <... (suitable name for our diagnostic)>:
    description: <...>
    variables:
      <... (suitable name for the preprocessed variable)>:
        short_name: <...>
        preprocessor: <...>
    scripts:
      <... (suitable name for our python script)>:
        script: <full path to python script>
        colormap: <... choose from matplotlib colormaps>

Show me the solution

YAML

diagnostics:
  diagnostic_warming_stripes:
    description: visualize global temperature anomalies as warming stripes
    variables:
      global_temperature_anomalies_global:
        short_name: tas
        preprocessor: global_anomalies
    scripts:
      warming_stripes_script:
        script: ~/esmvaltool_tutorial/warming_stripes.py
        colormap: 'bwr'

You should now be able to run the recipe to get your own warming stripes.

Note: for the purpose of simplicity in this episode, we have not added logging or provenance tracking in the diagnostic script. Once you start to develop your own diagnostic scripts and want to add them to the ESMValTool repositories, this will be required. Writing your own diagnostic script is discussed in a later episode.

Bonus exercises

Below are a few exercises to practice modifying an ESMValTool recipe. For your reference, here’s a copy of the recipe at this point. This will be the point of departure for each of the modifications we’ll make below.

Specific location selection

On showyourstripes.org, you can download stripes for specific locations. Here we show how this can be done with ESMValTool. Instead of the global mean, we can pick a location to plot the stripes for. Can you find a suitable preprocessor to do this?

Show me the solution

You can use extract_point or extract_region to select a location. We used extract_point. Here’s a copy of the recipe at this point and this is the difference from the previous recipe:

DIFF

--- recipe_warming_stripes.yml
+++ recipe_warming_stripes_local.yml
@@ -10,9 +10,11 @@
   - {dataset: BCC-ESM1, project: CMIP6, mip: Amon, exp: historical,
      ensemble: r1i1p1f1, grid: gn, start_year: 1850, end_year: 2014}

 preprocessors:
-  global_anomalies:
-    area_statistics:
-      operator: mean
+  anomalies_amsterdam:
+    extract_point:
+      latitude: 52.379189
+      longitude: 4.899431
+      scheme: linear
     anomalies:
       period: month
       reference:
@@ -27,9 +29,9 @@
 diagnostics:
   diagnostic_warming_stripes:
     variables:
-      global_temperature_anomalies:
+      temperature_anomalies_amsterdam:
         short_name: tas
-        preprocessor: global_anomalies
+        preprocessor: anomalies_amsterdam
     scripts:
       warming_stripes_script:
         script: ~/esmvaltool_tutorial/warming_stripes.py

Different time periods

Split the diagnostic in two with two different time periods for the same variable. You can choose the time periods yourself. In the example below, we have chosen the recent past and the 20th century and have used variable grouping.

Show me the solution

Here’s a copy of the recipe at this point and this is the difference with the previous recipe:

DIFF

--- recipe_warming_stripes_local.yml
+++ recipe_warming_stripes_periods.yml
@@ -7,7 +7,7 @@

 datasets:
-  - {dataset: BCC-ESM1, project: CMIP6, mip: Amon, exp: historical,
-  	  ensemble: r1i1p1f1, grid: gn, start_year: 1850, end_year: 2014}
+  - {dataset: BCC-ESM1, project: CMIP6, mip: Amon, exp: historical,
+     ensemble: r1i1p1f1, grid: gn}

 preprocessors:
   anomalies_amsterdam:
@@ -29,9 +29,16 @@
 diagnostics:
   diagnostic_warming_stripes:
     variables:
-      temperature_anomalies_amsterdam:
+      temperature_anomalies_recent:
         short_name: tas
         preprocessor: anomalies_amsterdam
+        start_year: 1950
+        end_year: 2014
+      temperature_anomalies_20th_century:
+        short_name: tas
+        preprocessor: anomalies_amsterdam
+        start_year: 1900
+        end_year: 1999
     scripts:
       warming_stripes_script:
         script: ~/esmvaltool_tutorial/warming_stripes.py

Different preprocessors

Now that you have different variable groups, we can also use different preprocessors. Add a second preprocessor to add another location of your choosing.

Solution

Here’s a copy of the recipe at this point and this is the difference with the previous recipe:

DIFF

--- recipe_warming_stripes_periods.yml
+++ recipe_warming_stripes_multiple_locations.yml
@@ -15,7 +15,7 @@
       latitude: 52.379189
       longitude: 4.899431
       scheme: linear
-    anomalies:
+    anomalies: &anomalies
       period: month
       reference:
         start_year: 1981
@@ -25,18 +25,24 @@
         end_month: 12
         end_day: 31
       standardize: false
+  anomalies_london:
+    extract_point:
+      latitude: 51.5074
+      longitude: 0.1278
+      scheme: linear
+    anomalies: *anomalies

 diagnostics:
   diagnostic_warming_stripes:
     variables:
-      temperature_anomalies_recent:
+      temperature_anomalies_recent_amsterdam:
         short_name: tas
         preprocessor: anomalies_amsterdam
         start_year: 1950
         end_year: 2014
-      temperature_anomalies_20th_century:
+      temperature_anomalies_20th_century_london:
         short_name: tas
-        preprocessor: anomalies_amsterdam
+        preprocessor: anomalies_london
         start_year: 1900
         end_year: 1999
     scripts:

Pro-tip: YAML anchors

If you want to avoid retyping the arguments used in your preprocessor, you can use YAML anchors as seen in the anomalies preprocessor specifications in the recipe above.

Additional datasets

So far we have defined the datasets in the datasets section of the recipe. However, it’s also possible to add specific datasets only for specific variables or variable groups. Take a look at the documentation to learn about the additional_datasets keyword [here][additional-datasets]{:target=“_blank”}, and add a second dataset only for one of the variable groups.

Show me the solution

Here’s a copy of the recipe at this point and this is the difference with the previous recipe:

DIFF

--- recipe_warming_stripes_multiple_locations.yml
+++ recipe_warming_stripes_additional_datasets.yml
@@ -45,6 +45,8 @@
         preprocessor: anomalies_london
         start_year: 1900
         end_year: 1999
+        additional_datasets:
+          - {dataset: CanESM2, project: CMIP5, mip: Amon, exp: historical, ensemble: r1i1p1}
     scripts:
       warming_stripes_script:
         script: ~/esmvaltool_tutorial/warming_stripes.py

Multiple ensemble members

You can choose data from multiple ensemble members for a model in a single line.

Show me the solution

The dataset section allows you to choose more than one ensemble member Here’s a copy of the changed recipe to do that. Changes made are shown in the diff output below:

DIFF

--- recipe_warming_stripes.yml	2024-05-27 15:37:52.340358967 +0100
+++ recipe_warming_stripes_multiens.yml	2024-05-27 22:18:42.035558837 +0100
@@ -10,7 +10,7 @@
-     ensemble: r1i1p1f1, grid: gn, start_year: 1850, end_year: 2014}
+     ensemble: "r(1:2)i1p1f1", grid: gn, start_year: 1850, end_year: 2014}

Pro-tip: Concatenating datasets

Check out the section on a different way to use multiple ensemble members or even multiple experiments at [Concatenating data corresponding to multiple facets] [concatenating-datasets]{:target=“_blank”}.

Key Points

A recipe can work with different preprocessors at the same time.
The setting additional_datasets can be used to add a different dataset.
Variable groups are useful for defining different settings for different variables.
Multiple ensemble members and experiments can be analysed in a single recipe through concatenation.