Global Flood Forecasting (GFF)

A globally distributed dataset for learning flood forecasting featuring both generated and labelled flood maps.

Dataset download here: Zenodo

This repository

Provides code to:

Use GFF: Train and evaluate models on GFF dataset.
Generate data: Generate GFF-like data from original sources.

Use GFF

To simply train and evaluate models on GFF, just build the conda environment. E.g.

conda env create -p envs/flood --file environment.yml

And run train.py or evaluate.py. Look in configs/* for how to provide configuration options. E.g.:

python train.py configs/two_recunet.yml -o data_folder=path/to/gff

Will create a ./runs folder, and save a model into it.

Generate new data

To generate new data using GFF's methodology/code, see ./HOWTO-GENERATE.md

Cite

To cite us, use:

@article{victor2024off,
  title={Off to new Shores: A Dataset \& Benchmark for (near-) coastal Flood Inundation Forecasting},
  author={Victor, Brandon and Letard, Mathilde and Naylor, Peter and Douch, Karim and Long{\'e}p{\'e}, Nicolas and He, Zhen and Ebel, Patrick},
  journal={arXiv preprint arXiv:2409.18591},
  year={2024}
}

Acknowledgements