On July 8, 2025 a fast moving wildfire north of Marseille forced Marseille Provence Airport to suspend operations as thick smoke and ash reduced visibility and created hazardous conditions on and around the airfield. Multiple carriers saw diversions and cancellations while local authorities instructed residents to shelter indoors.

Operationally, two facts matter for pilots and airport operators when smoke from a wildfire reaches an airport: runway visual range and the integrity of airside systems and emergency response. If smoke reduces runway visual range below landing minima, or if smoke and ash threaten aircraft engines and ground personnel, the safe option is to suspend arrivals and departures until conditions improve. That is what happened in Marseille.

So where does AI-based smoke detection fit into that chain? In short it is useful upstream but it is not a silver bullet for avoiding closures in fast, wind-driven wildfire events. There are three distinct detection tiers to consider and each brings different operational value:

1) Space-based early detection and hotspot monitoring. High cadence thermal satellites and emerging small-sat constellations can flag new thermal anomalies and evolving fire perimeters well before visible smoke shows up at the airport. Recent projects and research show AI models applied to raw Sentinel imagery and to dedicated thermal satellite data can speed identification of hotspots and reduce latency to actionable alerts. In practice that gives emergency planners and airport authorities earlier situational awareness to stand up mitigations or prepare for diversions.

2) Local sensors and cameras with edge AI. Fixed cameras, thermal imagers, and aspirating smoke detectors in and around airport property can identify visible smoke plumes or microscopic particulate signatures earlier than a human watching CCTV feeds. Vision models tuned for smoke and flame detection, and edge platforms that triage alarms for operators, are now mature enough to cut detection time on-site. That matters most for hangars, fuel farms, cargo buildings, and terminal spaces where an internal fire or encroaching smoke can require immediate local action. However detection alone still leaves the judgement call about runway operations to controllers and airport management.

3) Integrated multi-source fusion and operational thresholds. The highest operational value comes when satellite, ground sensors, weather, and camera AI are fused into a common operations picture with pre-agreed thresholds that automatically trigger procedures - for example heightened runway inspections, targeted closures of taxiways, or pre-planned diversion advisories to airlines. That requires prior planning, interoperable data links, and rules that align meteorological minima with sensor alerts. Without those connections fast-moving events still produce ad hoc closures even if the smoke was detected earlier. EU research and demonstration projects have been building those multi-layer systems, including drone and balloon layers for close-in detection, but operational rollouts remain fragmented.

Lessons from Marseille for operators and regulators

1) Early detection helps mitigation but not all closures are preventable. AI could have provided earlier notice that a wildfire was approaching and a clearer forecast of likely smoke movement. That would buy time to alert airlines and to execute diversion and passenger-handling plans, reducing passenger disruption and cascading knock-on effects. But when wind speed and direction suddenly drive a plume over an airfield, safety rules about visibility and air quality still justify a suspension of movements.

2) Invest where the biggest operational returns are. Airports gain the most from AI when they apply it to internal and immediate hazards: hangar fires, fuel farm leaks, cargo-hold smoke and terminal contamination. Technologies such as very early aspirating smoke detection systems are proven in critical facilities and can give minutes of lead time in enclosed spaces; pairing those with vision AI for outdoor plume detection addresses both internal and encroaching threats.

3) Build the data pipes and decision triggers. Detecting smoke is only half the battle. The rest is pushing verified alerts into the Airport Operations Centre and to Air Traffic Control with clear, actionable triggers. That means standard operating procedures that say what to do when a satellite hotspot within X kilometers is detected, or when ground cameras report sustained smoke density above a threshold. Without those playbooks, earlier detection will only compress the time available to make the same difficult decisions.

4) Account for false positives and environmental complexity. Vision AI is improving but false alarms remain when sun glint, dust, or controlled burns mimic smoke signatures. Models must be trained on local conditions and operators must maintain human-in-the-loop verification for critical airside decisions. That reduces the risk of unnecessary disruptions based on a single unverified alert.

Practical recommendations for airports and airlines

  • Establish a fused-monitoring feed that ingests satellite hotspots, regional fire service feeds, ground-based cameras, and aspirating detector alarms into the AOC and ATC dashboards. Define automated escalation rules so alerts are triaged before they reach frontline staff.

  • Run tabletop exercises that assume an approaching wildfire with rapid smoke onset. Practice diversion and passenger care scenarios so that earlier detection translates into smoother operational outcomes rather than ad hoc crisis management.

  • Prioritize high-sensitivity aspirating detectors in enclosed, high-consequence spaces and deploy thermal cameras with AI on the airport perimeter where vegetation or infrastructure sits near operational surfaces. These are proven investments to prevent internal losses and to give early warning of encroachment.

  • Work with national and regional authorities to ensure satellite-derived alerts and fire service intelligence are delivered with low latency during high risk seasons. Where possible, link thresholds to pre-agreed actions such as runway inspections or temporary flow restrictions so decisions are not made under time pressure.

Bottom line for operators and pilots

Smoke-detection AI is a force multiplier for situational awareness. It shortens the time between ignition and recognition and helps public safety agencies and airports move from reactive to proactive postures. But in wind-driven wildfire events like the one that closed Marseille the decisive factors for flight suspensions are visibility and operational safety rather than the mere presence or absence of an alert. Invest in multi-source detection and, crucially, in the operational integration that turns those detections into timely, consistent decisions. That is the path to reducing disruption without compromising safety.