This repository contains the code to reproduce the figures and experiments in our paper Wildfire Danger Prediction and Understanding with Deep Learning, published in Geophysical Research Letters.

Authored by Spyros Kondylatos, Ioannis Prapas, Michele Ronco, Ioannis Papoutsis, Gustau Camps-Valls, Maria Piles, Miguel-Angel Fernandez-Torres, Nuno Carvalhais

Now your project can be installed from local files:

pip install -e .

Or directly from git repository:

pip install git+https://github.com/Orion-AI-Lab/wildfire_forecasting --upgrade

So any file can be easily imported into any other file like so:

from wildfire_forecasting.models.greece_fire_models import LSTM_fire_model
from wildfire_forecasting.datamodules.greecefire_datamodule import FireDSDataModule

The code has been tested in Python 3.8

pip install -r requirements.txt

Download the datasets.tar.gz from https://zenodo.org/record/6528394 and decompress it in your filesystem:

tar -xzf datasets.tar.gz

Add the path to the decompressed directory to configs/datamodule/fireds_spatiotemporal_datamodule.yaml and configs/datamodule/fireds_temporal_datamodule.yaml

IMPORTANT NOTE: Make sure to have enough space to decompress the data. At least 250GB are needed!

The code is GPU-ready, and it is recommended to have a cuda-enabled NVIDIA GPU to run the experiments. They can also be run in a CPU, but expect slow training times

The code has been tested in a server with 128GB RAM and an NVIDIA RTX 3080 (10GB).

See notebook notebooks/RF.ipynb.

Training the LSTM with the hyperparameters that were used in the paper:

python run.py experiment=lstm_temporal_cls

Training the convLSTM with the hyperparameters that were used in the paper:

python run.py experiment=clstm_spatiotemporal_cls

Please refer to the README_template.md of the code template to understand the code structure and perform any custom training.

Kondylatos, S., Prapas, I., Ronco, M., Papoutsis, I., Camps-Valls, G., Piles, M., et al. (2022). Wildfire Danger Prediction and Understanding with Deep Learning. Geophysical Research Letters, 49, e2022GL099368. https://doi.org/10.1029/2022GL099368

This repo uses the lightning-hydra-template.