Fire Danger map for the Mediterranean for a day, produced from a machine learning model trained Track A.
We introduce mesogeos, a large-scale multi-purpose dataset for wildfire modeling in the Mediterranean. Mesogeos integrates variables representing wildfire drivers (meteorology, vegetation, human activity) and historical records of wildfire ignitions and burned areas for 17 years (2006-2022). It is designed as a cloud-friendly datacube, harmonizing all variables in a standard spatio-temporal grid of 1km x 1km x 1-day resolution. The datacube structure offers opportunities to assess machine learning (ML) usage in various wildfire modeling tasks. We extract two ML-ready datasets that establish distinct tracks to demonstrate this potential: (1) short-term wildfire danger forecasting and (2) final burned area estimation given ignition information. We define appropriate metrics and baselines to evaluate the performance of models in each track. By publishing the datacube, along with the code to create the datasets and models, we encourage the community to foster the implementation of additional tracks for mitigating the increasing threat of wildfires in the Mediterranean.
| Method | Precision | Recall | F1 | AUPRC |
|---|---|---|---|---|
| LSTM | 0,798 | 0,788 | 0,793 | 0,856 |
| Transformer | 0,777 | 0,812 | 0,794 | 0,870 |
| GTN | 0,783 | 0,774 | 0,779 | 0,855 |
| Method | CE | AUPRC |
|---|---|---|
| U-Net (all variables) | 0.0177 | 0.394 |
| U-Net (only ignitions) | 0.0166 | 0.418 |
New Model: To contribute a new model for an existing track, your code has to be (i) open, (ii) reproducible (we should be able to easily run your code and get the reported results) and (iii) use the same dataset split defined for the track.
Submit a new issue containing a link to your code.
New ML Track: To contribute a new track, submit a new issue.
We recommend at minimum:
@misc{kondylatos2023mesogeos,
title={Mesogeos: A multi-purpose dataset for data-driven wildfire modeling in the Mediterranean},
author={Spyros Kondylatos and Ioannis Prapas and Gustau Camps-Valls and Ioannis Papoutsis},
year={2023},
eprint={2306.05144},
archivePrefix={arXiv},
primaryClass={cs.CV}}