Evaluating causal Arctic-midlatitude teleconnections in CMIP6

This repository presents a sample of the causal model evaluation (CME) of Arctic-midlatitude teleconnections in observational datasets and CMIP6 simulations. This repository is part of the manuscript submitted to JGR: Atmospheres.

It is available on ArXiV: 

Corresponding DOI:

I. Preparation of the data with ESMValTool

Step 1. Install ESMValTool.

To install ESMValTool, please follow the official documentation and/or ESMValTool Tutorial.

Step 2. Run ESMValTool recipe recipe_galytska23jgr.yml.

This step reproduces data used in this study.

II. Preparation for the analysis

Step 3. Download this repository.

git clone https://github.com/EyringMLClimateGroup/galytska23jgr_EvaluatingCausalArcticMidlatTelec

cd galytska23jgr_EvaluatingCausalArcticMidlatTelec

Step 4. Create the environment from the environment.yml file and activate it.

conda env create --name my_env --file environment.yml
conda activate my_env

In case the environment file environment.yml is not working (most likely because some of the dependencies are not available anymore), we suggest creating the environment with the following key dependencies:

conda create -n my_env python=3.9 numpy matplotlib scipy iris pandas seaborn xarray

Install anaconda ipykernel (if needed) and create a new kernel for Jupyter Notebook

conda install -c anaconda ipykernel
python -m ipykernel install --user --name=my_env

Keep this environment activated and proceed with the Step 5.

Step 5. Install Tigramite and use PCMCI+.

In the activated environment from the Step 4 install Tigramite.

To install Tigramite follow the official GitHub repository for the installation instructions. It is the User's responsibility to install the Tigramite package. Please, follow the official Tigramite tutorials to get acquainted with the application of the PCMCI+ algorithm from the Tigramite package.

To reproduce the causal graphs from Galytska et al., 2023 manuscript use the following parameters:

• For a conditional independence test - linear partial correlation (ParCorr) and significance='analytic', then the null distribution is assumed to be Student's t. 
• mask_type = ‘y“ when testing different seasons, for example winter (December-January-February, DJF)
• maximum time delay tau_max = 5 and significance threshold pc_alpha = 0.01.

III. Causal model evaluation

Data_preparation.ipynb

Use this notebook as an example to produce a dictionary with the results based on PCMCI+ calculations.

1. In the first notebook cell uncomment the lines associated with the import of tigramite and related modules (e.g. import tigramite, from tigramite import ...)
2. Modify the second notebook cell based on the models (model_names), actors (actors), seasons (masking_list), maximum lag in months (max_timelag), and the significance threshold (pc_alpha).
3. Provide in the third notebook cell the path to the base folder (base_folder), which serves as the output folder. The results will be structured and saved in this folder.
4. We suggest reading original data from observations and CMIP6 models into a dictionary with the structure as suggested in the forths notebook cell.
5. In the sevenths notebook cell uncomment the code related to Tigramite plotting. Decide whether you would like to save the original causal graphs (plot_Causal_Graphs) and the dictionary with the output from Tigramite (save_orig_dict), insert True or False.

Summaries_for_causal_model_evaluation.ipynb

Use this notebook to reproduce the number of climate models that simulate identical connections as detected from observations. The output is saved in a .txt file and represents the values in hexagons in Fig. 4 and/or Fig. 7. This notebook also reproduces the summary of causal and contemporaneous links, see Fig.5 in Galytska et al., 2023, JGR and Figs. S5, S6, S8-S10 in the supporting information.

1. Modify the second notebook cell based on the models (model_names), variables (variables), seasons (masking_list), maximum lag in months (max_timelag), and the significance threshold (pc_alpha).
2. Load the dictionaries, saved as the output from Data_preparation.ipynb notebook or alternatively the dictionary with the results from the application of Tigramite on specific datasets (see the last code cell in Data_preparation.ipynb notebook for more tips).

F1-score_correlation_heatmap.ipynb

Use this notebook to reproduce Fig. 6 from Galytska et al., 2023, JGR.

1. In the first notebook cell modify the path to the base_folder and read in the dictionary with the output from Tigramite calculations.

F1-score_individual_networks.ipynb

Use this notebook to reproduce Fig. 8 from Galytska et al., 2023, JGR.

1. In the first notebook cell if needed modify the seasons (masking_list) and the significance threshold (pc_alpha).
2. In the second notebook cell modify the path to the base_folder and read in the dictionary with the output from Tigramite calculations.
3. In the fifth notebook cell modify the output path to save the figure.

IV. RESULTS

Results from running Jupyter notebooks will be saved into the corresponding directory, identified at each Notebook. To refer to figures from the paper, please see list_of_figures.txt to locate the notebook needed to reproduce each specific plot.

References

1. Eyring, V., Bock, L., Lauer, A., Righi, M., Schlund, M., Andela, B., and others: Earth System Model Evaluation Tool (ESMValTool) v2.0 – an extended set of large-scale diagnostics for quasi-operational and comprehensive evaluation of Earth system models in CMIP, Geosci. Model Dev., 13, 3383–3438, https://doi.org/10.5194/gmd-13-3383-2020, 2020.
2. Runge. J.: Discovering contemporaneous and lagged causal relations in autocorrelated nonlinear time series datasets. Proceedings of the 36th Conference on Uncertainty in Artificial Intelligence, UAI 2020, Toronto, Canada, 2019, AUAI Press, https://github.com/mlresearch/v124/blob/gh-pages/runge20a/runge20a.pdf, 2020.