AI cameras helped a wind farm reduce eagle deaths

At Wyoming’s Top of the World wind farm, cameras helped turn the sight of an approaching eagle into an automatic instruction to slow or stop selected turbines. A 2021 study estimated an 82% reduction in eagle fatality rate relative to a nearby comparison wind farm after IdentiFlight was introduced.[1]

Researchers compared both locations before and after the change, adjusted for the probability of finding carcasses and accounted for the amount of turbine operation studied. Deaths declined at the treated site while rising at the comparison site. The result demonstrated a practical way to reduce collisions, although fatalities were not eliminated.[1]

Seeing a bird early enough to act

IdentiFlight combines wide-angle cameras with movable, high-resolution stereo cameras. In the current system, eight fixed cameras search the sky for moving objects. A camera head then turns towards a detected bird, and the stereo images help establish its size and distance. Neural networks classify the species while the system builds a three-dimensional account of its movement.[3]

A separate curtailment module relates that flight path to nearby turbines. If the bird meets the site’s risk criteria, it sends a shutdown instruction through the wind farm’s control network. The record connects the intervention to the detected bird, time and location, so operators can examine what prompted a particular pause.[3]

An early detection study helps explain why that chain can work. In 2018, IdentiFlight detected 96% of flights seen by human observers. For birds it classified as eagles, median classification distance was 793 metres and median time from detection to classification was 0.4 seconds. Those were early-system test results, providing evidence that useful identification could occur while a bird was still some distance away.[4]

Recognition was imperfect: photographic checks found that nine of 149 eagles were labelled non-eagles, while 287 of 1,013 non-eagles were labelled eagles. These errors affect different parts of the operation. A missed eagle may go unprotected; an incorrect eagle classification can produce an avoidable turbine pause. Keeping the photographs makes those mistakes inspectable.[4]

The mortality result survived a stricter reanalysis

The 2022 follow-up addressed criticism of the original comparison. It excluded the installation period, matched 66 turbines of the same model at each site and included further observations through December 2020. It estimated an 85% relative reduction, with a wide 95% highest-density interval of 12–100%. This was evidence from one treated wind farm, not a randomised test across the industry.[2]

The later analysis strengthens the case for a real reduction while showing uncertainty about its size. Its authors acknowledged that unmeasured changes in eagle activity, prey or carrion could also affect fatalities. They could not directly test those explanations. The comparison therefore supports a substantial benefit under the studied conditions without establishing one universal percentage for every wind farm.[2]

Protect the flight that is actually at risk

Research published in Ibis in 2022 examined the next engineering question: which flight paths lead into the area swept by turbine blades? Using automated observations, the researchers found that distance, height and movement towards a turbine were informative. Slow-moving birds heading towards the nearest turbine were particularly relevant; a simple compass direction was not.[5]

Their model distinguished risky approaches in Wyoming and transferred well to a California facility, with reported areas under the receiver-operating curve of 0.96 and 0.97. These are measures of prediction performance, not percentages of birds saved. The result supports refining the timing of shutdowns using the bird’s actual trajectory, instead of treating every distant sighting as equally dangerous.[5]

The approach also operates outside the United States. Australia’s Clean Energy Regulator describes IdentiFlight at Tasmania’s Cattle Hill wind farm, where it was installed before turbine commissioning and trained for Tasmanian wedge-tailed eagles and white-bellied sea eagles. Its account describes turbines stopping as birds approached and restarting after they passed.[6]

That is the useful physical achievement: information from an image changes the behaviour of a large machine quickly enough to provide a safer passage. Operators gain a way to respond to individual flights while the rest of the wind farm continues producing electricity.[3]

Report an error or suggest a correction