Introduction

This repository was created to publish the code for the analysis in [1]. It is meant for recreation of the results in that work and a starting point for further functional climate network analysis.

Disclaimer: As the main goal of this code publication is the reproduction of the results in [1], the code is not well-documented. If you wish to use this work as a basis for your own work, please contact me ([email protected]) .

Prerequisites

The Repository

Please clone the repository and then go inside (for the following commands to work).

git clone https://github.com/timkittel/global-structures-enso-volcanoes.git
cd global-structures-enso-volcanoes

Note that detrend.nino34.ascii.txt was downloaded from NCEP (see link in the lower left corner, here is a direct link, accessed Nov 4, 2017) and all rights belong there.

Data

As detailed in [1] this project uses the daily averaged surface air temperature data (SAT at sigma=0.995) of the NCEP/NCAR Reanalysis I Project. (They provide ftp download servers and you basically need all the files air.sig995.1948.nc until air.sig995.2015.nc.)

The easiest is if create a subfolder called data and place all the .nc files inside. If you prefer to keep them somewhere else, you can use the --data-directory flag of fullrun.py

Python & Dependencies

The Code was developed on an 64bit Ubuntu 16.04 LTS using Python 3.6.1. Furthermore, the following additional Python packages are necessary. The version numbers correspond to the ones I used during the development.

• numpy 1.12.1
• scipy 0.19.0
• matplotlib 2.0.2
• basemap 1.1.0
• pandas 0.20.1
• argcomplete 1.8.2
• igraph 0.7.1.post6
• h5py 2.7.0
• netCDF4 1.2.8

Further, please install simple-mpi which can be found here on github. (I recommend to clone it in a different folder, install it with pip there and then return to the global-structures-enso-volcanoes folder.)

Running the Code

Comment: Running the below analysis is computationally very expensive. If the following code detects mpi (e.g. because it was called via mpirun) it automatically parallelizes the computation. I strongly recommend to use this option! I am happy to provide details/help if necessary.

Comment: ./fullrun.py -h gives a help for the usage. Please check it before running the lines below.

Comment: ./fullrun.py creates (possibly many) temporary files when using mpi. If you want these to be save in a different location, use the --scratch-directory flag.

First, create the cache file for the icosahedral grid .icosahedral-grid.cache.npy. In this step, do not use mpi!

./icosahedral_grid.py

Then run the functional climate network analysis (except modularity). A file called Output.FullRun.daily-paper.icosahedral.hdf5 will be created. I strongly recommend the usage of mpi, e.g. using mpirun.!

./fullrun.py daily paper

The computation of the modularity is not done daily, because the computation time would increase drastically. Hence, it is done in a separate step. A file called Output.FullRun.normal-modularity.icosahedral.hdf5 will be created.

./fullrun.py normal modularity

The analysis for the volcanoes is done with the icosahedral grid where the ENSO-big region is removed, using the following line. A file called Output.FullRun.daily-paper.icosahedral_without_ENSO_big.hdf5 will be created.

./fullrun.py daily paper --grid icosahedral_without_ENSO_big

Finally, if you want to recreate the comparison of the community detection algorithms, run the following line. A file called Output.FullRun.normal-cmp-modularity.icosahedral.hdf5 will be created.

./fullrun.py normal comparison-modularity

Plotting the Results

Comment: ./paper-pix.py -h gives a help for the usage. Please check it before running the lines below.

regions of interest (Fig. 1 in [1]):

./paper-pix.py Output.FullRun.daily-paper.icosahedral.hdf5 icosahedral regions-of-interest --save

transitivity and global average link distance (Fig. 2(a) and (c) in [1]):

./paper-pix.py Output.FullRun.daily-paper.icosahedral.hdf5 icosahedral ENSO-global --save

modularity (Fig. 2(b) in [1]):

./paper-pix.py Output.FullRun.normal-modularity.icosahedral.hdf5 icosahedral ENSO-global --save

composites (Fig. 3 in [1]):

./paper-pix.py Output.FullRun.daily-paper.icosahedral.hdf5 icosahedral composites --save

regionalized climate network properties (Fig. 4 in [1]):

./paper-pix.py Output.FullRun.daily-paper.icosahedral.hdf5 icosahedral ENSO-local --save

analysis of volcanoes (Fig. 5 in [1]):

./paper-pix.py Output.FullRun.daily-paper.icosahedral_without_ENSO_big.hdf5 icosahedral_without_ENSO_big volcanoes --save

comparison of the community detection algorithms

./paper-pix.py Output.FullRun.normal-cmp-modularity.icosahedral.hdf5 icosahedral cmp-modularity --save

References

[1] T. Kittel, C. Ciemer, N. Lotfi, T. Peron, F. Rodrigues, J. Kurths, and R. V. Donner. “Global teleconnectivity structures of the El Niño–Southern Oscillation and large volcanic eruptions: An evolving network perspective”. in prep. 2017