More fires, far less at stake per fire.
Corsica ignites five and a half times more often per square kilometre than California and burns one ninth as much per fire. That inversion decides everything: the system's value scales with severity, not frequency, and Corsica has almost none of the severity that pays for it.
Verdict up front
The concept transfers geographically and fails economically at the California design point. Six nodes cover Corsica properly, but the loss model has no megafire tail, so break-even needs an effective rate of 8.4% against California's 0.53%. The base case still returns 3.6:1; the conservative case returns 0.45:1 and does not pay for itself. In California every scenario paid. That difference is one line item: the absence of a $2.3B urban conflagration to prevent.
One mountain chain breaks the arithmetic
Dividing island area by hexagon area gave two nodes. That answer is worthless here, and for three reasons that all come from the same place.
The island is one and a half cells wide
At a 16-minute response budget the lattice spacing is 57 km. Corsica is 183 km long and 83 km wide. Hexagonal packing assumes a compact blob; on a long thin island the edge losses dominate and the formula stops describing anything real.
The spine blocks line of sight
Corsica rises from sea level to 2,706 m within about 25 km, and the chain runs the length of the island. A node on the west coast cannot see the eastern plain at any altitude a tether reaches. Both flanks need their own nodes.
The terrain factor was California's
The 0.92 viewshed factor at 1,000 m in the main study was calibrated on Californian relief. Corsica has sharper relief per unit distance, so that factor is optimistic here and the detection radius should be read as an upper bound.
Six nodes, placed by flank
Sites chosen on high ground with a usable viewshed, three on each flank of the spine, rather than by dropping a grid on the island. Phase 1 is the three that cover the highest-value wildland-urban interface; phase 2 closes the interior.
| Site | Flank | Covers |
|---|
The drone flies over the ridge. The camera does not.
This asymmetry is what sets the layout. A suppression drone climbs over the spine and reaches anywhere within 33 km, so on response alone four nodes would blanket the island. But the thermal channel only reaches 19 km of direct line of sight, so full thermal coverage would need 10 nodes.
Six is the compromise: island-wide response and smoke detection, with the thermal layer concentrated where people and property actually are. Calling that full thermal coverage would be a lie.
| Constraint | Radius | Nodes |
|---|
Rebuilt for maquis, with no megafire tail
The Californian model cannot be reused. Its entire return comes from demoting megafires, and Corsica has none. Calibrated instead to the Prométhée record: 824 fires and 7,119 hectares a year averaged over 48 years.
| Measure | Corsica | California |
|---|
Frequency is not the same as value
A system that intercepts fires early is worth what the fires it intercepts would otherwise have cost. Corsica offers 5.5× the ignition density, which sounds like five times the opportunity, but the mean Corsican fire burns 8.6 ha against California's 76. High frequency and low severity is the worst combination for this architecture: you pay for coverage by area and get paid by severity.
| Final size | Fires / yr | Mean ha | Suppression | Damages | Total / event | Annual |
|---|
Where the loss sits, and how flat it is
It works if it works well. It does not survive pessimism.
Same scenario structure as the California study, so the two are directly comparable. The difference in the bottom line is entirely structural, not a change of assumptions.
| Scenario | Coverage | Downshift | Effective | Avoided / yr | Return | California |
|---|
| Line | EUR |
|---|
What would have to be true
Break-even needs an effective rate of 8.4%, meaning roughly 69 of Corsica's 824 annual fires held one size class smaller. That is a real operational target rather than a rounding error, and unlike California it is not obviously easy. The PoC would have to demonstrate it, not assume it.
Three things that would change the answer
Bundle, do not replace. Corsica already funds SIS 2A/2B and national Canadair sorties. Costed as an add-on to an existing air-attack budget rather than a standalone programme, the comparison changes.
Count what the model omits. Tourism is a large share of Corsican GDP and a burnt August is expensive in ways the hectare count never shows. Natura 2000 habitat, water quality and human life are all outside this model.
Go smaller. Four nodes on response alone cost €2.5M a year and still cover the island for suppression. Dropping to a detection-light, response-heavy configuration is the version most likely to pay.
The honest read
Corsica is a good place to test the technology and a poor place to justify it on avoided loss alone. If the aim is a European pilot, the geography is convenient and the fire frequency is generous. If the aim is a business case that survives a sceptical auditor, California is the argument and Corsica is the demonstration.