Zero collisions in 19 months: what the Aberdeen Bay data tells us about seabird behaviour and offshore wind

The result is notable not because it proves collisions never happen, but because it provides rare real-world evidence of how seabirds behave around offshore turbines and how current collision predictions compare with observed behaviour.

What the study measured

What the study measured

Between June 2023 and December 2024, a high-resolution camera system monitored bird movements around turbine AW05 at the Aberdeen Bay wind farm. The system operated for approximately 95% of daylight hours across the full study period, automatically detecting and tracking birds as they passed through the field of view. Each tracked flight was recorded with its path, speed, distance, and height.

Over 19 months, the system detected over 137,000 birds, of which 2,007 showed changes in flight behaviour near the turbine. Five of these were initially flagged as potential collision events. On detailed review by expert ornithologists, none were confirmed as actual collisions. In each case, the birds were either well clear of the turbine structure or carrying out natural behaviours such as diving.

The prediction gap

The prediction gap

The environmental impact assessment conducted before construction in 2018 estimated that, under recommended avoidance assumptions (98%), each turbine could cause an average of 8.54 bird collisions per year. Over the 19-month monitoring period, that translates to roughly 13.5 predicted collisions at the monitored turbine.

The observed figure tells a different story. Based on the actual bird activity recorded, The Biodiversity Consultancy calculated 0.002 expected collisions for the entire study period. That is several orders of magnitude lower than the pre-construction prediction, and suggests the collision risk model overestimated actual risk by a factor of nearly 7,000.

This gap between predicted and actual impact is not unique to Aberdeen. Collision risk models used in environmental impact assessments are designed to be precautionary. They use conservative assumptions about avoidance rates, the proportion of birds flying at rotor height, and the probability of collision for each transit through the rotor plane. These assumptions are intentional, they are meant to ensure that consented projects do not exceed acceptable impact levels. But when the assumptions are highly conservative, the mitigation measures derived from them may be disproportionate to actual risk.

Collision risk models remain the primary tool for environmental impact assessment in the wind industry, and for good reason. But until now, their inputs have relied heavily on statistical assumptions rather than direct field observations. The Aberdeen study demonstrates what happens when real monitoring data is fed into those same models: the output shifts from conservative estimates to figures that more closely reflect actual bird behaviour. When The Biodiversity Consultancy applied observed activity data from Spoor's monitoring system to the Band collision risk model, the result was 0.002 predicted collisions over the full study period, a figure then corroborated by the absence of any confirmed collisions in the monitoring record. This suggests that collision risk models can deliver realistic, defensible estimates when grounded in empirical data rather than precautionary assumptions alone.

What avoidance looks like in practice

What avoidance looks like in practice

The study's findings align with earlier research at the same site using radar, camera, and GPS tracking, which showed that seabirds typically adjust their flight paths to avoid turbines at distances of 100 to 200 metres. This macro-avoidance behaviour, where birds alter their course well before reaching the rotor zone, is now understood to be a primary factor in keeping collision rates far below theoretical predictions.

Collision risk models account for avoidance through an avoidance rate parameter, typically set at 95% to 99.5% depending on species. The Aberdeen data suggests that for the species observed at this site, actual avoidance rates are at the high end of, or potentially above, the ranges currently used in standard assessments. If validated across additional sites and species, this has direct implications for the collision estimates used in consent applications and for the curtailment regimes derived from those estimates.

The five flagged events in the Aberdeen study illustrate why continuous monitoring and expert review matter in this context. An automated system sensitive enough to flag potential collisions will inevitably flag events that turn out to be something else. The ability to go back to the original footage, examine the bird's trajectory in detail, and confirm or reject the flag is what separates a detection count from verified evidence. It is also what gives regulators confidence in the reported collision rate.

Why 19 months of continuous data matters

Why 19 months of continuous data matters

Periodic surveys, whether boat-based, aerial, or from vantage points, capture snapshots of bird activity. They are effective for characterising broad patterns but cannot provide the temporal coverage needed to measure rare events like collisions. If collisions occur at the rate suggested by the Aberdeen data (fewer than one per turbine per decade at this site), a survey programme would need to be extraordinarily lucky to observe one directly.

Continuous monitoring changes this calculation. By recording nearly every daylight hour over 19 months, the system at Aberdeen captured a dataset large enough to draw statistically meaningful conclusions about collision frequency. The absence of confirmed collisions in a sample of 2,007 tracked flights is not proof that collisions never occur, but it places a credible upper bound on the collision rate that is far tighter than any periodic survey could achieve.

This kind of data has value beyond the individual site. As more continuous monitoring datasets accumulate across different wind farms, species, and geographies, the industry builds a shared evidence base that can inform more accurate collision risk modelling for future projects. Each site that contributes validated, long-duration monitoring data helps reduce the uncertainty that drives conservative assumptions and, by extension, conservative mitigation requirements.

What this means for future projects

What this means for future projects

For developers preparing consent applications for new offshore wind projects, the Aberdeen findings reinforce a practical point: real-world monitoring data improves the accuracy of environmental assessments and supports more proportionate consent conditions.

The original EIA for Aberdeen Bay used the best available methods and data at the time. The monitoring data gathered since then shows that actual bird behaviour at the site differs substantially from the modelled predictions. That is not a criticism of the original assessment. It is evidence of why post-construction monitoring matters, and why the data it produces should feed back into the models and assumptions used for future projects.

For regulators, the study provides further support for the value of continuous, AI-supported monitoring as a tool for verifying predicted impacts. It also adds to the growing evidence base on seabird avoidance behaviour, which is one of the most consequential parameters in collision risk modelling.

The Aberdeen study was a collaboration between Vattenfall and Spoor, with the analysis and reporting carried out by The Biodiversity Consultancy. The monitoring system used a single high-resolution camera with automated detection and tracking, supported by expert ornithological review.

The more sites that contribute this kind of validated, long-duration data, the closer the industry gets to collision risk estimates that reflect what actually happens in the field, not what models assume in the absence of evidence.

Download the full report below.