Is camera-based bird monitoring good enough for wind farms?

Before Vattenfall's Aberdeen Bay wind farm was built, assessments estimated that each turbine might kill between 2.18 and 8.54 seabirds a year. Since then, an integrated radar and camera system has recorded more than 10,000 bird videos there without a single collision (Tjørnløv et al. 2023), and when Spoor’s cameras later watched one turbine for 19 months, all five suspected collisions proved on review to be something else (Cook 2026).

Estimates of that kind have long had to stand in for observation, because a bird struck at sea leaves no carcass to count. Where the models’ inputs are uncertain, regulators lean towards caution, paid for in wider buffers, fewer or lower turbines, or curtailment, which is why developers have turned to radar and cameras to measure what the models only assume.

What can a camera-based approach record at a wind farm?

A camera system turns a passing bird into a video track from which software estimates counts, position, speed, flight height, direction and possible interactions with turbines, while species are identified by AI and confirmed by ornithologists wherever a record is uncertain or consequential. Radar, by comparison, records an echo that yields position, speed and approximate size across a far larger volume of airspace, but it cannot tell a kittiwake from a herring gull without an observer or a second sensor.

How well a camera measures distance depends on its geometry: stereo setups triangulate with two calibrated cameras, whereas mono setups infer distance from a bird’s apparent size. When Vattenfall trialled Spoor’s system at Aberdeen Bay, the British Trust for Ornithology found that out to 500 m, stereo depth errors were about 5% of range, against 10 to 15% for mono (BTO Research Report 781). Just as important, the evidence is the footage itself, so a sceptical consultant or regulator can replay any species call or flagged collision. In a regulatory process, evidence that can be contested is the kind that ends up being trusted.

Are cameras good enough for environmental impact assessment (EIA)?

Yes, cameras can collect the bird data for an environmental impact assessment on their own, and they work best in combination with traditional bird surveys. Cameras record species, flight height and activity at one location day after day, while surveys cover the whole site but only occasionally: before consent, NatureScot expects “two years of monthly surveys” by aircraft or boat, which observe the site on roughly one weather-dependent day a month, and the guidance itself warns that nocturnally active species may be missed.

Radar has long been used to watch the days in between, although it has limitations of its own. At Nysted in Denmark, it tracked waterbird flocks across 11 km by day and night and showed the share entering the wind farm falling from 40.4% before construction to 8.9% afterwards (Desholm & Kahlert 2005). However, it could not identify species, so a subsample had to be checked visually, and data were collected only in calm, dry weather. Detection distance also depends heavily on the size of the bird: one off-the-shelf system detected small birds only to 0.5 to 1 km, even in calm, clear tests (May et al. 2017), and at Aberdeen Bay, the radar’s detection of gulls close to the unit fell off once the sea reached state 4.

Cameras complete the picture over time, because they record one location continuously and every record carries a species, a flight height and a timestamp, which are the inputs collision risk models depend on. They are most valuable when a baseline raises a question it cannot answer on its own.

That is currently the case for a developer Spoor is working with in Poland. Field surveys for its onshore project recorded breeding pairs of Lesser Spotted Eagle, and the developer expects any permit to be appealed. Without continuous flight data, the likely outcome is a fixed curtailment calendar, whereas continuous camera records would let it argue for curtailment triggered by the eagles themselves.

Floating buoys extend the same approach offshore, before a single turbine is built. Because a buoy moves constantly, the horizon, the usual reference point for flight height, can drop out of view, so Spoor’s software estimates the camera’s orientation to keep flight heights measurable. At Sørlige Nordsjø II, up to 185 km off the Norwegian coast, cameras on a metocean buoy recorded nearly 8,800 hours of video over a full year for Ventyr’s environmental impact assessment. Guidance does not yet say how such fixed-point data should be weighed against traditional bird surveys, so the method is best agreed with the regulator before deployment.

How good are cameras for collision detection once turbines operate?

For post-construction monitoring, cameras record collisions and avoidance behaviour at the rotor, which is precisely where radar struggles, since turbines are “a particular clutter challenge” that can create blind zones (May et al. 2017). At Aberdeen Bay, the integrated system’s radar could not detect low-flying seabirds within 30 m of the unit, so the study relied on its cameras for the decisive moments, recording micro-avoidance rates above 0.96 (Tjørnløv et al. 2023).

Cameras alone can now cover that ground. At FLOATGEN, a 2 MW floating turbine off Le Croisic that Spoor monitored for about ten months for Fondation OPEN-C, experts validated 1,690 bird tracks covering 44 bird and 6 bat species; a fifth showed avoidance, mostly at a distance, and two ended in gull collisions (Environment Coastal & Offshore).

Those two collisions matter because absence of evidence counts as evidence of absence only if the instrument could have seen the event. FLOATGEN shows that it can, which allows Aberdeen Bay's result to be read as a finding: recalculated from 19 months of observed flights, the expected toll at the monitored turbine fell to 0.002 collisions (Cook 2026).

Three cautions apply. Collisions are rare, so monitoring at an operational site must last for the years that consent conditions often require; results are per turbine, so scaling them to a whole farm adds uncertainty; and accuracy is local, since FLOATGEN reached 96% precision only after retraining on site data, which makes recall at a comparable site the figure most worth requesting.

How does camera-based monitoring compare with radar?

Radar and cameras divide the problem by scale: radar covers kilometres of airspace, cameras the space around a single rotor. Cameras resolve what radar cannot reliably provide: the species, its flight height to within about 5% of range with stereo, and what the bird does in the rotor-swept zone, with every event kept as video that can be audited. With infrared illumination they also record at night, at shorter detection distances, and a thermal system at Nysted showed that birds can be filmed at turbines in total darkness. Fog affects cameras more than radar, although in the German North Sea dense fog at night occurs only 1.5 to 4% of the time (Cook, Håkansson & Sykes 2026).

Radar, for its part, follows large migrating flocks across open sea by day and night, although even in calm, clear tests one off-the-shelf system detected small birds only to 0.5 to 1 km, and it cannot reliably identify the species it tracks. Combined, as at Aberdeen Bay, radar tells the cameras where to look. When a project can fund only one, however, quantifying migration across a wide area is a radar question, whereas most consent questions turn on what only cameras can answer: which species fly at rotor height, and whether any of them collide.

When is camera-based monitoring the right choice?

Camera-based monitoring is the right choice wherever the question is what happens at the turbine. Cameras are strongest at four things: identifying species, measuring flight height, recording avoidance and collisions in the rotor-swept zone, and doing all of this by day and, with infrared illumination, by night. Every flagged event is reviewed by an ornithologist and kept as auditable video, the standard Spoor works to.

Traditional bird surveys remain the better tool for estimating bird densities across a whole site, and radar for following large migrating flocks over several kilometres, because a camera watches one turbine or buoy at a time. That boundary is moving, however, since a network of cameras on turbines and buoys can sample a site far more often than surveys do, and with a method agreed in advance, those observations can help estimate effects across the wind farm. When regulators lack evidence they default to caution, which carries costs for birds and for the energy transition alike; better evidence still leaves those trade-offs to be weighed, yet it allows them to be argued on what actually happens at the turbine.

Sources

1. Brighton, C.H., Clarke, J.A. & Boersch-Supan, P.H. (2025). Scientific support to the trial of Spoor AI at the European Offshore Wind Deployment Centre. BTO Research Report 781.
2. Cook, A. (2026). Review of Aberdeen Bay collision monitoring data. The Biodiversity Consultancy.
3. Cook, A., Håkansson, J. & Sykes, G. (2026). Comparative analysis: multi-sensor approaches for bird collision monitoring at offshore wind farms. Technical whitepaper, Spoor.
4. Desholm, M. & Kahlert, J. (2005). Avian collision risk at an offshore wind farm. Biology Letters, 1(3).
5. Environment Coastal & Offshore. Spoor cameras capture bird collisions and avoidance at floating offshore wind turbine.
6. May, R. et al. (2017). Performance test and verification of an off-the-shelf automated avian radar tracking system. Ecology and Evolution.
7. NatureScot (2023). Guidance Note 2: Advice for marine ornithology baseline characterisation surveys.
8. Tjørnløv, R.S. et al. (2023). Resolving key uncertainties of seabird flight and avoidance behaviours at offshore wind farms. Final report for Aberdeen Offshore Wind Farm Ltd.