Rural Fire Prevention
The plan is frozen
The hazard map in the municipality's current plan was drawn in 2021 and stays fixed until 2030. The vegetation it describes has since burned, grown, or changed species.
The same information at the desk and in the field.
The problem is not a missing map. It is a missing order.
The budget is limited
No municipality clears everything in a year. The decision that counts is not whether there is risk, but where to start when only part of it can be treated.
Burning is not the same as doing damage
A brush-covered slope far from anything and the same slope up against a village do not deserve the same intervention, even at equal risk.
Three questions, multiplied into one order
Where it burns
Seventeen years of burned areas from the ICNF, crossed with slope, sun exposure, vegetation type, and the state of that vegetation as measured from orbit.
What is exposed
18,914 buildings detected by machine learning over satellite imagery, with the size of the cluster weighing more than plain distance.
How hard it is to fight
Water points from the municipal plan, road access, and the real travel time from the nearest fire station, computed over the road network rather than in a straight line.
Prevention is worth more where suppression arrives late. That is why difficulty is not a footnote: it is one of the three factors.
It updates itself
Every week the tool collects the year's ignitions, the burned areas published by the ICNF and fresh Sentinel-2 imagery, and recomputes the state of the vegetation. It compares that against the median for the same month across 2015-2025, to tell an ordinary summer from a dry one.
Sentinel-2 passes over Baião every five days.
What this does not do
A published map informs decisions about public spending. These numbers are here for the same reason they are in the code.
- It does not forecast a day. No weather goes in. What it computes is structural propensity, not tomorrow's risk.
- It is not demonstrably better than the official plan. Tested on years neither of them had seen, the difference fell inside the margin of error. What differs is that the plan is fixed until 2030 and this model retrains.
- It cannot see the management strips. The resolution is 29 meters. The plan's median strip is 0.09 hectares, far too narrow to follow from this imagery.
- It does not know where firefighters stopped the fire. The data records where it burned, never the fire that never grew. No wildfire model escapes this.
Baião, cell by cell
Baião is the pilot municipality: rated T4 by the ICNF, the worst tier, with both many ignitions and a large burned area. That is 210,998 cells of 29 meters, each with its place in the order and the reasoning written out.
Open the mapContact
For more information about the method, the data or the application.
Send an email