The research was funded by the US Department of Energy (DOE), supported by multiple stakeholders in the renewable energy industry. The report delivers a comprehensive evaluation of these DTBird models' detection and deterrence systems, something that is of crucial importance for environmental regulators when establishing requirements in wind farm environmental permissions in order to comply with environmental regulations such as the Bald and Golden Eagle Protection Act and the Migratory Bird Treaty Act.
At Manzana wind farm, in a 9-month pilot study, the DTBird V4D4 model (2016) was installed on 7 wind turbine generators with 65 metre tower height and 82.5 metre rotor diameter.
At Goodnoe Hills wind farm, in a 2-year study, the DTBird V4D8 model (2019) was installed on 14 wind turbine generators with 87 metre tower height and 110 metre rotor diameter.
After three years of data collection, the analysis found significant reduction in eagles and raptors collision risk, with the DTBirdV4D4 and DTBirdV4D8 (models 2019 and 2021, respectively) reducing the likelihood of golden eagles and other large raptors entering the rotor-swept zone (RSZ) by 20-30 percent, with an even higher deterrence rate (over 40 percent) for eagles approaching or flying directly toward the RSZ.
Unmanned aerial vehicles (UAVs) resembling eagles were used in flights to assess detection efficacy across multiple sites. After analysis, the data confirmed that the overall probability of detecting large raptors was 65 percent within 240 metres of the cameras, with the highest detection rate being around 75 percent at 50-75 metres and the detection probability at 240 metres being around 50 percent. At 380 metres, the detection rate reduced to around 30 percent.
The likelihood of missing a detectable flight was found to be generally less than 20 percent when the minimum line of sight distance to the camera was approximately 30 to 120 metres. This increased to less than 30 percent at distances below 20 metres and between 120 and 160 metres. However, it exceeded 50 percent at distances greater than 200 metres.
With regard to deterrence efficacy across multiple sites, for large raptors, 73 percent of the case4s at the Manzana wind farm and 63 percent at Goodnoe Hills were confirmed or potentially effective responses.
With regard to specific species or groups, the response rates were 79 percent for golden eagles at the Manzana site compared to 60 percent at Goodnoe Hills. For turkey vultures, they were 81 percent at Manzana versus 61 percent at Goodnoe Hills. For buteos, they were 72 percent at Manzana compared to 56 percent at Goodnoe Hills.
The rate of False Positive (FPs) triggering deterrence in Manzana was 1.2 – 1.8 FP/turbine/day and 0.8 minutes/turbine/day. In Goodnoe Hills, it was 0.8 FP/turbine/day and 0.96 minutes/turbine/day, after fine-tuning (starting at 3.9 FP/turbine/day).
The rate of deterrence triggered by Non-targeting Avian False Positive (NTAFP) was 40 percent in Manzana and 36 percent in Goodnoe Hills.
There was some concern among researchers that eagles could become less responsive to deterrent signals, but this did not actually occur. Eagles spent less time near turbines when the deterrence signal was activated frequently, even by FPs.
At Goodnoe Hills wind farm, the effect of deterrence in avoidance behaviour was compared between DTBird systems with sound muted versus systems with sound activated. In muted systems, deterrence was reduced 24 to 27 percent. The dwell time that eagles spent in the vicinity of the wind turbines (WTGs) in comparison with muted systems, with an average dwell time reduced to approximately 25 to 19 s per event.
In the case of golden eagles, deterrence can reduce the dwell time by 29 percent, with an average dwell time reduction of 26 to 17 s per event.
Birds responded better to deterrence signals at Manzana wind farm than at Goodnoe. This is possibly due to differences in resident/migratory birds, landscape, and climate conditions. However, the eagle’s response to the deterrence signal was higher in Goodnoe Hills than in Manzana.
Eagles and vultures tended to respond better to deterrence signals as wind speed increased, but the opposite occurs with buteos.
The probability of effective deterrence with wind above approximately 4 m/s was higher for eagles and vultures than for buteos and increased as the wind speed increased. For buteos, the probability decreased as the wind speed increased.
For all raptors, the rate of effective deterrence was the highest in flights categorised as moderate risk, possibly because birds had enough time to maneuver effectively.
While the REWI report underscores DTBird’s success, it also highlights opportunities for further refinement.
Improving camera resolution and further optimising the AI algorithms used to detect birds against complex backdrops (such as clouds and moving turbine blades) could enhance the system's detection accuracy.
In addition, AI can reduce the rate of FP and NTAFPs.
The report also recommends regular camera lens maintenance to avoid solar degradation, which can adversely impact detection capabilities.
For additional information: