Defence · Field Note

EN FI SV

Fire brigades as drone defence units — three reasons to build the network now

15 May 2026 · by Risto Anton Paarni

The news, in one line

Finland's national drone strategy (2026–2030) names dual-use Drone-as-a-Service as the model for civilian and defence authorities. The Ukraine front teaches the same lesson in faster time. The people closest to the sky are already organised, trained, and on call tonight — they wear yellow helmets.

Source: Finnish national UxS strategy, 2026; DWS field notes 15 Apr 2026 and 01 Apr 2026.

Good news comes already organised.

Finland has 33 regional rescue services, more than 700 volunteer fire brigades (VPK), and roughly 13,000 active volunteers. They are dispatched tonight. Turning them into the country's first-tier drone defence network is not a new build. It is a small extension on top of an existing one.

And because cheering for a network is not the same as standing it up, here is the honest part.

1. Rapid action — minutes, not hours

A drone over a substation has minutes of decision time, not hours. A national C-UAS unit, however well equipped, cannot be in 700 places at once. The VPK already is. Average alarm to scene: 6–10 minutes. That is the metric the threat actually has.

What the clock looks like

Small commercial UAS at 60 km/h covers 5 km in five minutes.
National C-UAS rotor flight from a single base: 20–40 minutes.
VPK alarm-to-scene in its own kunta: 6–10 minutes, on average.
The first eyes on a hostile airframe are not the eyes that arrive after takeoff. They are the ones already at the station.

2. Local conditions — the wind, the woods, the wires

A drone defence unit needs to read the sky the way a local reads the weather. Where will a damaged airframe fall? Which forest road is passable in April? Which substation feeds the hospital, and which feeds the dairy? The VPK already answers these every weekend.

A national operator flying in for the first time needs an hour just to learn what the local knows by heart. In a drone incident, that hour is the incident.

What “local knowledge” actually means

Microclimate: aurora-driven GPS drift, sea-fog windows, lake-effect snow corridors.
Terrain: which forest cuts let a low-altitude airframe through, which clearings make safe net intercept points.
Civilian load: school routes, ambulance corridors, ice-fishing spots that put people under any debris cone.
Critical infrastructure: which substation, mast, or pump-house is on which feeder, and what fails if it goes down.

3. Fail-over — 700 nodes beats one

Centralised systems fail centrally. Distributed systems degrade gracefully. If a single jamming event takes down regional comms, or a single command node is contested, a 700-node mesh keeps operating.

The Ukraine lesson is the same lesson the Internet learned in 1969: survivability is a property of topology, not of any single node. The strongest counter-drone system is the one that has no single point of failure to attack.

Why redundancy is the actual product

33 regional rescue services + 700+ VPKs = 700+ stations with independent power, independent comms fallback, independent dispatch.
Lose one node, lose 0.14 % of national coverage. Lose one battalion, lose all coverage in its sector.
Local mesh keeps detecting and cordoning even when the national backhaul is offline. When the link returns, the detections sync upward.
This is how the grid survived 1939–1944. It is how the sky survives 2026.

What the VPK model does not promise — and why that matters

VPKs are first response, not endpoint solution. The model assumes the national level still owns long-range identification, intent inference, kinetic engagement authority above small commercial UAS, and cross-border deconfliction with Russia, NATO and EASA U-space.

The VPK layer is the eyes, the cordon, the first minutes. It is not the kinetic solution to a fixed-wing intruder. We will not over-claim the volunteer's mandate.

Where DWS fits on top of the rescue layer

The heroes are already on shift. We add the agent layer above them, not the unit below.

One tablet UI — Fleet Commander triage screen, KYA-bound to a verified operator. Same workflow the duty officer already uses for fire and SAR.
Local-first compute — detection and classification run on the station's edge node. Works without backhaul. Syncs when the link returns.
EASA U-space integration — so the national level sees what the local sees, with the local timestamp and the local provenance.
EU AI Act Article 12 reasoning lineage — every detection carries a tamper-evident trace. Admissible in a court, defensible in an audit.

A worked example — a Karelian evening, 19:42

A drone, commercial class, 60 km/h, low altitude. Crosses the eastern border north of Imatra. Unidentified. Heading along the Saimaa shore toward a hydroelectric plant.

Timeline: alarm to case closed

Second 0. The Imatra Border Guard radar paints the target. The alarm goes out.
Second 12 — Control Room (field). The Joutseno VPK duty officer sees the alert on a tablet. The Fleet Commander triage screen shows three things at once: target track, local wind (northwest, 4 m/s), and the nearest unit's 6-minute ETA. KYA binds the duty officer's identity to this case — everything from here forward carries her name.
Second 18 — SiteSense agent. Overlays the track on the local infrastructure map. Hydro plant. 220 kV line. School yard 800 m off-track. The duty officer sees the same screen — same chain of reasoning, same evidence.
Minute 1:30 — Situation Room (strategic). Helsinki looks at the same target in a bigger frame. Any others right now? Yes, two: Kotka and Lappeenranta. This is a swarm, not a single. The Themis agent opens a case and spins up three parallel field tasks.
Minute 4. The Joutseno unit is on scene. Cordon up. Local eyes, local mesh, local decision.
Minute 6:30. The national C-UAS unit (Karelia Air Wing) receives the picture Themis assembled: three targets, three local cordons, GPS, local chain of evidence. No duplicated work. No “who saw what, when.”
Minute 9. The Air Wing takes the kinetic responsibility. Joutseno holds the cordon. The traffic flows — no autonomous unit would have managed alone, no national unit would have arrived alone.
Minute 22. Case closed. EU AI Act Article 12 reasoning trace stored from twelve agents and three humans. Auditable a month, a year, or a decade later.

What Control Room and Situation Room each deliver

Control Room (local) does one thing well: turns the alarm into a decision in minutes, with local knowledge in the loop and the decision traceable.

Situation Room (strategic) does another thing well: sees the swarm, not the single, and stitches 700 local nodes into one picture without freezing any of them while it does.

Together they do a third thing neither does alone: keep local agency and national overview in the same workflow — under the same KYA root, the same reasoning chain, the same shared picture.

Why we will not centralise this

The temptation is always to centralise. A single C-UAS battalion, a single command node, a single procurement. It looks tidy on a Powerpoint. It loses on day one.

Defence-in-depth is not a slogan. It is what survived 2022–2026 on the eastern front. Finland's strategy already names dual-use DaaS. The institutional shape — palokunnat plus VPK — is in place. Add the agent layer above, not the unit below.

The short version

Finland already has 33 rescue services, 700+ VPKs, ~13,000 trained volunteers.
Drone defence is a minutes-game. Only distributed units arrive in minutes.
Local knowledge of terrain, wind, infrastructure is unbuyable.
A 700-node mesh degrades gracefully where one battalion fails completely.
Fleet Commander adds the agent and identity layer. The heroes are already on shift.