AI can reduce the spread of wildfires. So why isn’t Europe using it more?

On 9 July 2026, a fallen power-line pole sparked scrubland near Los Gallardos, close to Almería. By evening, flames were closing on the district of Bédar at speeds of up to 100 kilometres an hour. Twelve of the fire’s eventual thirteen victims died within hours, cut off from their planned route out: four in a vehicle and eight more on foot in a dry riverbed that Andalusia’s emergency chief later called a “trap”. Officials said afterwards that the victims had taken a different route to the one authorities had planned for evacuation. 

That riverbed is exactly the kind of failure AI-based evacuation modelling exists to catch. Fed with live fire-spread forecasts, wind data and the region’s actual road and path network, an egress-risk model flags in advance which routes will intersect a fire’s path, including the ones that look like an escape route on a map and turn into a trap once the wind shifts. That capability isn’t experimental. It runs today, in other fire-prone regions of the world. It wasn’t running in Almería. 

That gap between planned route and lethal reality is the story of this summer, repeated across the European continent. Since the start of the year, wildfires have burned through 464,781 hectares across the EU with 80,282 hectares burning in the previous week alone. Blazes have stretched from Spain, France and Greece into Portugal, Croatia, Germany and the United Kingdom, where the Cairngorms fire burned for twelve days, transforming the region into unrecognisable terrain. 

None of the technology mentioned here is speculative. Detection systems can spot a fire within minutes of ignition. Models  forecast where it will go next. And tools  flag which evacuation route will fail before anyone has to find out the hard way. Each technology is proven, funded, and running somewhere in the world right now. Almost none of it reached the places that burned this summer. 

Camera and satellite networks trained to recognise smoke, heat signatures and movement can now catch an ignition within minutes rather than hours. Galicia proved this at scale. After a devastating 2025 fire season, the Spanish region rolled out a €213 million AI camera and satellite network capable of detecting fires from over 15 kilometres away, rising to 25 kilometres in good conditions. The cameras were designed to shrink the gap between ignition and detection from hours to minutes.

Similar systems elsewhere, including Pano AI’s camera networks in the US and Australia, report comparable detection speeds. In tests, 60% of fires were spotted within four minutes of visible smoke, 85% within ten. California’s ALERT California network caught more than a third of its 2025 detections before any 911 call came in. 

The same underlying data, wind, terrain, fuel load, and live satellite feeds can forecast where a fire goes next, including the extreme “firestorm” behaviour seen at Gironde this year, or the kind of shift that turned a two-week “contained” blaze into the Cairngorms’ twelve-day major incident. Fed into evacuation planning, that same forecasting can model which routes will fail before anyone has to find out on foot. This modelling adds a layer of protection that regions can add without years of new development. But this can only happen if the will exists to deploy what’s already proven. 

Both the Almería and Gironde fires are believed to have started with electrical infrastructure faults. By using infrastructure risk-scoring,  and then cross-referencing grid assets against live drought and wind data, high-risk sites could be flagged before ignition rather than after. But this isn’t yet standard practice anywhere in Europe. 

So why hasn’t a system that already works in one Spanish region reached the places that heavily burned this summer? 

Funding is part of it. The European Parliament estimates that €1 spent on prevention saves €4 to €7 in response and recovery costs.  And yet suppression and response funding is six times higher than prevention spending across the bloc, while land and forestry agencies often lack the technical capacity to access the EU funds that do exist. 

Regulation is a bigger constraint than most people realise. Drone operations that fly beyond the pilot’s direct line of sight (the kind needed to cover large, remote wildfire zones) aren’t a default anywhere in the EU. Under EASA rules, an operator must submit a full risk assessment to their own national aviation authority and get individual, national authorisation – there’s no automatic recognition across borders.

The most significant barrier isn’t the most obvious. Europe’s own sovereign cloud spending is forecast at just €10.6 billion in 2026, a rounding error against the more than $700 billion in global hyperscaler capital expenditure. The EU has flagged that the bloc depends on non-EU providers for over 80% of key digital products, services and infrastructure. Wildfire detection systems need continuous data feeds from national cameras, satellites, and sensors to run permanently on someone’s cloud. Public bodies procuring critical wildfire infrastructure have reasons to hesitate before building it on foreign platforms, reasons that have nothing to do with whether the technology works. 

Fifteen years leading go-to-market across AI and technology have shown the same patterns surface consistently. If you don’t have a buyer, you don’t have a business. You can have the finest product in the world, but if your target buyer has the will but lacks the capability to deploy it, product market fit will never be attained. The gap is widest after the fire, and it’s the least talked-about of the three.

Once a blaze is out, the question shifts from where it is burning to where the recovery effort should go first. That’s a genuinely hard problem to solve by eye across a fire scar that can run to tens of thousands of hectares. Machine-learning models trained on satellite imagery can classify burn severity across an entire site within days, distinguishing scorched-but-recoverable vegetation from soil that’s lost its structure entirely. Models can then cross-reference that data against terrain, hydrology and carbon storage potential to rank which hectares would benefit most from active intervention, and which will recover on their own given time.

Copernicus’s Emergency Management Service already runs a version of this continuously across the EU, producing burn-severity and soil-damage maps as fires happen. It’s been running as core EU infrastructure for years. What the mapping rarely triggers is the nature-based response the evidence says works. At Golticlay in Caithness, a 2018 wildfire reaching the edge of rewetted peatland simply stopped, the waterlogged ground acting as a natural firebreak in a way no engineered barrier had managed elsewhere on the same fire.

The same logic extends to fire-refugia seeding and targeted fuel-load reduction, both shown to lower reburn risk precisely in the areas most likely to catch again. The modelling capability already exists. What’s absent is a funding pathway that would connect a prioritisation map, already sitting on a server somewhere in Brussels, to a rewetting contractor with the budget to act on it. The map exists. The bridge from map to intervention on the ground largely doesn’t. 

None of what’s described here is a research gap. Galicia’s cameras exist and work. Copernicus already maps every fire scar in the EU as it happens. The evidence for peatland rewetting as a firebreak has been available for years. What’s missing across detection, response and recovery alike, is the coordination to move a tool proven in one region or one sector or institution into the hands of whoever needs it next. 

The lack of coordination is a familiar story to organisations working across sectors in silos. Our own FloodAction Coalition among them, is built on the same premise that flood risk doesn’t confine itself to one department’s remit either. Wildfire may turn out to be the sharpest version of that argument yet: a technology gap that closed years ago, sitting next to a coordination gap that hasn’t. 

Galicia found the mandate to close that gap. Will Europe’s catastrophic 2026 fire season build the will to convene the regulators, the insurers, the drone operators and the land managers before the next one hits? 

— 

Katie McPhee is The Conduit’s AI for Good Expert in Residence. She is also founder of Expanded Future, advising mission-driven organisations on AI adoption. She has spent over fifteen years leading technology and AI go-to-market strategy, including at the BBC, Eventbrite and VEED, and co-founded Earth AI, a climate and AI hackathon program.