ABEIRO
Wildfire protection in Galicia. It does not detect fire, it turns its spread into a concrete decision for each village and each person.
Abeiro means shelter in Galician, and the project starts from a question more modest than putting out a fire. If the fire is moving towards a village, what should the people who live there do?
Satellites and public agencies already detect fires. ABEIRO does not. It is a free, open map that estimates, village by village, who would have the worst of it if a wildfire reached them in Galicia, so as to turn the way the fire spreads into a concrete decision for each village and each person. The thesis is that the oldest and most isolated villages are among those that need it most.
A number from 0 to 100 for each village
I built a vulnerability index from 0 to 100 for each settlement. It adds up the hazard of the terrain, which comes from slope and vegetation measured by satellite, social sensitivity, which looks at the age and size of the population, and response capacity, which is how many exit roads it has. All of it with real open data from IGE, OpenStreetMap and Sentinel 2.
It started with 12 pilot settlements in Valdeorras and Larouco, each with its evacuation route by road to the nearest safe destination. I then scaled it to the 683 settlements of 50 inhabitants or more in the province of Ourense. The smaller ones were left out of that first sweep because of computing cost, not because they matter less, and by the project’s own thesis they are among the most vulnerable.
What real fires said
To test it I used real fires, the 2025 fire perimeter from Copernicus for the pilot and the GlobFire record of large fires between 2001 and 2021 for Ourense. In the pilot, with 12 settlements, the intervals come out too wide to conclude anything.
Against the Ourense fire record, the index separates the settlements that burned from those that did not better than chance, though not by much. Measured with the AUC, which runs from 0.5 (chance) to 1 (perfect separation), it scores 0.69, with a 95 % confidence interval from 0.64 to 0.74.
The surprise came from elsewhere. The best predictor was not the hazard of the terrain but the lack of exit roads, with an AUC of 0.73. Isolation does not cause fire. Settlements far from towns and with few exits are the ones that sit in large fire areas, and the project measured that instead of assuming it.
Why I left the weights alone
It would have been easy to recalibrate the index weights against burned area and get a better score. I did not, on purpose, because that would teach the index where things burn and not who is vulnerable when they do. Calibration against burned area measures exposure and isolation, not social vulnerability, and the weights are still the provisional ones.
Validating the social part needs real evacuation and lockdown data. The system is ready to take it in and does not have it yet. Until then ABEIRO is a demonstrator with real data, not a tool for emergency decisions.
The identity of vulnerable people stays out of the system, which is right where it belongs.
