Regional (CORDEX) Historical changes in climate variables (time series, trends) for REF

[schlunma]

Regional (CORDEX) Historical changes in climate variables (time series, trends) diagnostic/recipe

Tasks

• Add shapefile so that extract_shape can extract CORDEX regions (similar to Support IPCC AR6 regions in extract_shape ESMValCore#2008)
• Write recipe
• Variables: e.g. tas, pr, ta, ua, hus

[bouweandela]

Description of the CORDEX domains: https://cordex.org/wp-content/uploads/2012/11/CORDEX-domain-description_231015.pdf

GitHub repository with machine readable info: https://github.com/WCRP-CORDEX/domain-tables

Example Jupyter notebook using py-cordex: https://github.com/WCRP-CORDEX/cordex-tutorials/blob/main/tutorials/domains.ipynb

[katjaweigel]

There is also another,related
issue: #3872

[katjaweigel]

An older issue which could contain an example for CORDEX data: #3337 and CORDEX related issues in the Core: ESMValGroup/ESMValCore#2335

[bouweandela]

While related, those issues are a bit different from the intended use case here. The aim of this issue here would be add a recipe to analyze global climate model data, but on CORDEX domains, to inform the CORDEX community.

[katjaweigel]

@hb326 What is the aim? If we want time series/trends of any CORDEX region we only need the shape file and a recipe including a fitting preprocessor (with extract_shape) using existing diagnostics.
For time series/ annual cycle:
Should the results be shown (like e.g. monitor/recipe_monitor.yml) or compared to something else (like e.g. monitor/recipe_monitor_with_refs.yml) or both?

• monitor/recipe_monitor.yml can already do multi panel plots for different regions (tested some IPCC AR6 reference regions as a start, see example attached)

recipe_monitor_regions_yml.txt

• as far as I can see monitor/recipe_monitor_with_refs.yml cannot do that, yet.
  (There are the IPCC AR5 plots with panels, showing the bias to a reference for many regions like: https://esmvaltool.dkrz.de/shared/esmvaltool/stable_release/recipe_flato13ipcc_figures_938_941_cmip3_20240523_133148/ which gather multiple regions, but I'd rather try to modify some of the new python diagnostics than using these old NCL ones, especially because there the regions are hard coded in the diagnostic).

For trends I still have to look what is available

[katjaweigel]

Update: With certain preprocessor settings one can run several regions in monitor/recipe_monitor_with_refs.yml , but it either puts all in one plot or makes a new file for each region, a multi-panel plot seems not to be possible.

What would be also possible for regions is using the new portrait plot. @lukruh help me to make an example recipe:

recipe_portrait_regions_yml.txt

[katjaweigel]

Update: The new version of monitor/multi_datasets.py in https://github.com/ESMValGroup/ESMValTool/tree/benchmarking_maps4monitoring can do better time series plots. It is no problem to do them for a region from a shp file instead of global, but multi-panel plots for several regions I could not do.

recipe_model_benchmarking_timeseries_regions_yml.txt

[katjaweigel]

Update for trends:
Most diagnostics currently included in the ESMValTool, which can plot trends are made with NCL or do not do the kind of plots I thought about (e.g. climwip). I managed to make trend plots using the preprocessor linear_trend and a modified version of recipe_model_benchmarking_boxplots.yml from monitor/multi_datasets.py in https://github.com/ESMValGroup/ESMValTool/tree/benchmarking_maps4monitoring to plot trends although it is not made for that (therefore it would probably need some changes in the code; including the observational reference and getting the correct title is not working, yet). Again it is possible to do those plots for any region defined through a shape file (N.Europe in the plot below), but I know currently no way to make multi panel plots.

recipe_model_benchmarking_boxplots_region_trend_yml.txt

[hb326]

Update: The new version of monitor/multi_datasets.py in https://github.com/ESMValGroup/ESMValTool/tree/benchmarking_maps4monitoring can do better time series plots. It is no problem to do them for a region from a shp file instead of global, but multi-panel plots for several regions I could not do.

recipe_model_benchmarking_timeseries_regions_yml.txt

Hey @katjaweigel,
I think in a first instance I would think that individual plots for each region would be ok. As far as I understand, the first instance of the REF would anyway be something like the portrait plot with an RMSE value which would be individual for each region anyway. From that level one can then click through to different other plots, like line plots, or map plots.

[hb326]

Update for trends: Most diagnostics currently included in the ESMValTool, which can plot trends are made with NCL or do not do the kind of plots I thought about (e.g. climwip). I managed to make trend plots using the preprocessor linear_trend and a modified version of recipe_model_benchmarking_boxplots.yml from monitor/multi_datasets.py in https://github.com/ESMValGroup/ESMValTool/tree/benchmarking_maps4monitoring to plot trends although it is not made for that (therefore it would probably need some changes in the code; including the observational reference and getting the correct title is not working, yet). Again it is possible to do those plots for any region defined through a shape file (N.Europe in the plot below), but I know currently no way to make multi panel plots.

recipe_model_benchmarking_boxplots_region_trend_yml.txt

Hey @katjaweigel,
looks pretty cool already! I think, as I mentioned before, in a first instance we provide the information for each region separately, and worry about a multi-panel plot later. I still have not gotten my hands on a CORDEX shape file (although I got plenty of info about the CORDEX regions). I will have to follow-up on this again. But if you start with the IPCC regions, that's already fantastic!