Understanding collision risk is an important factor in managing risk in development of individual offshore wind projects and in quantifying the environmental impacts of the UK’s offshore wind industry. Many observers and practitioners in the industry have identified obtaining sound collision risk data as the most critical issue facing the deployment of offshore wind in the UK. Indeed, to date, the lack of sound data regarding collision risk is the leading factor in delaying consenting decisions, with the refusal of consent by the Department of Energy and Climate Change (DECC) for Centrica’s Docking Shoal Offshore Wind Farm (Docking Shoal) project being due to a lack of evidence to support a positive appropriate assessment under The Conservation of Habitats and Species Regulations 2010 (as amended) (the Habitats Regulations).
Unlike onshore wind farms where bird corpses can be used as evidence to quantify collision, collisions offshore are very hard to observe directly. Furthermore, in many situations direct monitoring of collisions presents a catch-22 situation where statistical significance is achieved only after an unacceptable level of mortality has been observed.
Collision risk assessments are derived from well-understood collision risk models. These models initially assume that birds are oblivious to turbines, and then apply a correction factor known as ‘avoidance rate’ to provide the final estimate. At present, determining avoidance rates is a challenge, and often a mixture of direct measurement, prior knowledge and overly precautionary assumptions have to be applied to derive a value.
Ideally, avoidance would be measured directly; although this can be done under some circumstance, e.g. using radar, there is as yet no general solution that can be applied to all species at all sites.
The established approach to deducing avoidance rate is to measure collision rate at a constructed site and deduce avoidance from this and the pre-construction baseline surveys undertaken at the site being assessed. A fundamental difficulty with collision monitoring is the low anticipated collision rates for many at risk species. An example of this is Sandwich tern in the Greater Wash, where the collision rate is such that even if every single collision in a summer were observed the uncertainty in the resulting estimate of collision rate could still be very large. Data which focusses purely on collisions is therefore unlikely to significantly advance the debate on collision risk unless it is gathered over several years and at several locations.
A better alternative is to study avoidance behaviour; by understanding how birds adapt to the immediate presence of a turbine, we can refine the assumptions that feed into a collision risk model leading to more accurate estimates of mortality. In this approach, rather than studying collisions in the context of population size, we propose to study birds at species level that are ‘available for collision’ (birds which cross the swept area of the rotors, plus a buffer to account for mortality due to potential rotor tip turbulence) in the context of all birds in the vicinity of the turbine.
To achieve this functionality, an ideal survey tool would capture the flight path in three dimensions of all birds entering a pre-defined “box” around the turbine. Each flight path can be tested to see whether or not it intersects the swept area of the rotors, enabling birds ‘available for collision’ to be differentiated from those that avoid the turbine. The detection box would be sized such that all birds available for collision could be identified and a significant number of paths not intersecting the rotor could also be detected.
Avoidance behaviour can then be measured by comparing the proportion of birds ‘available for collision’ with the prediction of a model that assumes no avoidance behaviour. Uncertainty can be robustly estimated using a non-parametric bootstrap technique, which can account for behaviour such as flocking by appropriate aggregation of samples.
The system as implemented is based around a pair of sixteen (16) megapixel digital video cameras equipped with fisheye lenses. The fisheye lenses give a very wide field of view enabling a single pair of cameras to cover the space up to and including the turbine blades and down to the water with a horizontal field of view of approximately 100 degrees (°). To achieve this view the cameras are tilted upwards by 45°, as shown in Figure 2. The combination of this viewing angle and the extremely wide field of view of the lenses allows the rotors to be seen clearly, while enabling detection and tracking of birds at a considerable horizontal distance from the turbine. The system can therefore detect collisions as well as monitor avoidance.
The baseline of the cameras was set at 1.4m which gives the system potentially good spatial resolution of up to 1m at a range of 100m and some ability to measure with decreased accuracy out to 500m. These images are then transmitted back to shore via a wireless link where they are analysed to identify the birds species and track the birds flight. This information tells us which species are in the vicinity it also shows us the impact of the turbine on the birds flight path.
The system spent a year on a turbine in a wind farm in the north sea where it was used to monitor the local tern population.
The developed system for avoidance monitoring at offshore wind farms, was successfully deployed for over six months at an operational offshore wind farm, without any significant faults. From the data collected, it has been possible to track individual birds and identify them to species-group or better. The data collected showed evidence of micro-avoidance behaviour, with none of the birds approaching within a blade-length of the turbine.
The evidence gathered as part of this study indicates that the data provided by the system will enable micro-avoidance to be quantified.
The system has shown to offer:
The capability developed here may have significant benefits in other fields, such as UAV detection.
