From logistic regression to deep learning : machine learning modeling of lightnings in reanalysis data

This repository is linked to the paper [INSERT PAPER LINK TO PREPRINT ONCE SUBMITTED].

---

Abstract

[Insert abstract here once submitted.]

---

Structure of the repo

----| data        samples of the most extreme days from the test dataset
----| models      models described in the paper
----| notebooks   3 notebooks to reproduce and expand the paper's results
----| src         source files used in the notebooks

• data/ — Samples of the most extreme days from the test dataset. See data/README.md for details on the data format and channel descriptions.
• models/ — The models described in the paper. See the paper and the README.md in the notebooks folder for more details.
• notebooks/ — Three notebooks to reproduce and expand some of the paper's results. See notebooks/README.md for a description of each notebook and how to run them.
• src/ — Source files and utility functions used across the notebooks.

---

Installation

uv build (fastest)

git clone git@github.com:AdrienBq/lightning_modelling.git
cd lightning_modelling
uv sync

pip build

git clone git@github.com:AdrienBq/lightning_modelling.git
cd lightning_modelling

# Portable install (uses the loose version ranges in pyproject.toml; CPU-friendly)
pip install .

# OR, to reproduce the authors' exact environment (a fully pinned freeze that
# includes GPU/CUDA wheels — heavier, and intended for a CUDA machine):
pip install -r requirements.txt

Note: pip install . / uv sync and requirements.txt are not equivalent. The first resolves the loose ranges from pyproject.toml and works on CPU-only machines; requirements.txt is a full pinned freeze of the authors' GPU environment. The demo notebooks run on CPU, so the portable install is recommended for most users.

---

Data sources & acknowledgements

Atmospheric convective parameters — thundeR

Channels 1–5 of the input tensors are derived from atmospheric convective parameters computed with the thundeR R package. This postprocessed data was provided by Mateusz Taszarek under contribution from a grant from the Polish National Science Centre (2020/39/D/ST10/00768) and can be made available. Contact him (mateusz.taszarek@amu.edu.pl) for usage information.

Lightning location data — Met Office ATDnet

Channel 6 contains lightning location data from the Met Office ATDnet (Arrival Time Difference network):

ATDnet data provided by the Met Office (2026) under a CC-BY-SA 4.0 license.

---

License

The code in this repository is released under the MIT License.

The data samples in data/ include a channel derived from Met Office ATDnet data, provided under a CC-BY-SA 4.0 license. Any derivative work incorporating that channel must comply with the terms of CC-BY-SA 4.0.

---

Citation

If you use this code or data in your work, please cite:

[Insert BibTeX citation once the preprint is available.]

@manual{thunder2025,
  title  = {thunder: Computation and Visualisation of Atmospheric Convective Parameters},
  author = {Bartosz Czernecki and Mateusz Taszarek and Piotr Szuster},
  year   = {2025},
  note   = {R package version 1.1.5},
  url    = {https://bczernecki.github.io/thundeR/},
}

@article{atdnet2006,
author = {Gaffard, Catherine and Nash, John and Atkinson, N. and Bennett, Alec and Callaghan, Greg and Hibbett, Eric and Turp, Myles and Schulz, Wolfgang},
year = {2008},
month = {01},
pages = {},
title = {Observing Lightning Around the Globe from the Surface},
journal = {The Preprints, 20th Interna-tional Lightning Detection Conference}
}