Skip to content

README.md

Latest commit

 

History

History
157 lines (113 loc) · 7.25 KB

README.md

File metadata and controls

157 lines (113 loc) · 7.25 KB
Raw

The global atmospheric river network

ARnetwork.py

Codebase for constructing, analysing, and visualising Atmospheric River Transport Networks (ARTNs) — directed, weighted complex networks built from global catalogues of atmospheric river (AR) trajectories.

About

Atmospheric rivers (ARs) transport vast amounts of water vapour and are responsible for a substantial share of global precipitation and wind extremes. The global ARTN emerges from individual AR trajectories with edges defined by recurrent AR transport patterns. ARs are localized by a suitable 2D locator, e.g., the AR centroid. The ARTN is a directed, weighted graph. This enables us to study the planetary-scale pattern of AR transport with the tools of complex network science: centralities, shortest paths, communities, random walks and secondary node/edge attributes that carry additional information on AR transport (e.g., integrated water vapour transport).

This repository contains the analysis code accompanying the paper:

Tobias Braun, Sara M. Vallejo-Bernal, Norbert Marwan, Jürgen Kurths, Johannes Quaas, Albert Diaz-Guilera, Luis Gimeno, Miguel Mahecha Atmospheric river trajectories organise along a global transport network. Preprint (2026). https://doi.org/10.21203/rs.3.rs-7482510/v2

The networks are built from two independent global AR catalogues — PIKART-1.0 and tARget-4 — both derived from ERA5 reanalysis. Most results in the paper are reported as the consensus of the two catalogues.

Repository structure

ARnetwork/
├── analysis/      # core Python modules and analysis scripts (figures of the paper)
├── ARnetlab/      # Jupyter notebooks: exploratory analyses and extensions
├── ARTN.png       # repo header image
├── LICENSE        # MIT
└── README.md

The core modules used across the analysis scripts are:

  • ARnet_sub.py — catalogue preprocessing, transport-matrix construction, network generation, and node/edge attribute handling.
  • NETanalysis_sub.py — network analytics: Computes the hubs, highways and basins of the global ARTN. Also includes consensus averaging, predictability estimation and moisture-transport attributes for edges and nodes.
  • Nullmodels_sub.py — random walker null model family and random rewiring for targeted null hypotheses on the ARTN topology.
  • NETplots_sub.py — map-based plotting utilities.

Data

The analysis draws on several publicly available datasets. The core datasets are the two AR catalogs:

These are derived from the following source data:

  • ERA5 reanalysis (Hersbach et al., 2023), Copernicus Climate Data Store: cds.climate.copernicus.eu
  • MERRA-2 reanalysis, NASA Goddard Earth Sciences Data and Information Services Center (GES DISC): disc.gsfc.nasa.gov

Additionally, the following data has been used in the manuscript:

By default, scripts read from ./data/ and write figures to ./figures/. Override either via environment variables:

export ARNET_DATA=/path/to/catalogues
export ARNET_FIGURES=/path/to/figures

Installation

Clone the repository and install the dependencies (Python 3.9+ recommended):

git clone https://github.com/ToBraun/ARnetwork.git
cd ARnetwork

Core dependencies:

  • numpy, pandas, scipy, scikit-learn
  • networkx
  • h3 — H3 hexagonal grid bindings
  • matplotlib, cartopy, cmcrameri — mapping and perceptually uniform colormaps
  • geopandas — vector spatial data
  • tqdm

A minimal conda environment to get started with:

conda create -n arnet python=3.9.20 numpy pandas scipy scikit-learn networkx \
    matplotlib cartopy geopandas tqdm -c conda-forge
conda activate arnet
pip install h3 cmcrameri

Getting started

The analysis pipeline follows a consistent pattern across scripts:

  1. Regrid AR to hexagonal coordinates (or work with a rectangular grid and risk biases).
  2. Load hex-indexed AR catalogue.
  3. Build the transport network with ARnet_sub.preprocess_catalog → generate_transport_matrix → generate_network. A clipped spatiotemporal extent or conditions can be applied.
  4. Optional steps for analysis, e.g. attach moisture-transport classes to edges and/or nodes via NETanalysis_sub.compute_edge_moisture_transport and compute_node_moisture_transport, or form a consensus network by averaging attributes across multiple catalogs.
  5. Analyse and plot — Calculate network measures, e.g., edge-betweenness ("AR highways"), node/edge moisture sinks and sources, trajectory predictability, etc.

A minimal example:

import pandas as pd
import ARnet_sub as artn

# Load catalog (already gridded on hexagonal grid)
ARcat = pd.read_pickle("data/PIKART_hex.pkl")

# Pre-process: set AR locator, grid type, spatial resolution (and possibly clip spatiotemporally & apply conditions)
l_arcats, d_coord_dict = artn.preprocess_catalog(
    ARcat, T=None, loc="centroid", grid_type="hexagonal",
    X="global", res=2, cond=None, LC_cond=None)

# Compute transport matrix
A, t_idx, t_hexidx, t_ivt, t_grid = artn.generate_transport_matrix(
    l_arcats, "hexagonal", d_coord_dict, LC_cond=None)

# Generate networkx graph representation of the ARTN with a certain threshold
G = artn.generate_network(
    A, t_grid, weighted=True, directed=True,
    eps=16, self_links=False, weighing="absolute")

The full set of analyses reproducing the paper figures lives under analysis/.

Citing

If you use this code or the network construction pipeline in your work, please cite the companion paper:

@article{braun2026artn,
  title   = {Atmospheric river trajectories organise along a global transport network},
  author  = {Braun, Tobias and Vallejo-Bernal, Sara and Marwan, Norbert and Kurths, J{\"u}rgen and others},
  journal = {Research Square preprint},
  year    = {2026},
  doi     = {10.21203/rs.3.rs-7482510/v2},
  url     = {https://doi.org/10.21203/rs.3.rs-7482510/v2}
}

License

Released under the MIT License.

Contact

Dr. Tobias Braun — Postdoctoral researcher, University of Leipzig · Potsdam Institute for Climate Impact Research. For questions, issues, and suggestions, please open an issue and feel free to reach out!