# Scottish Marine and Freshwater Science Volume 5 Number 16:The Avoidance Rates of Collision Between Birds and Offshore Turbines

This study reviewed data that have been collected from offshore windfarms and considers how they can be used to derive appropriate avoidance rates for use in the offshore environment.

### 8. Transferability of Avoidance Rates Between Models

There are various collision risk models currently available within the scientific literature to estimate likely collision and mortality of birds due to windfarms (Band 2012; Desholm 2006; Eichorn et al. 2012; McAdam 2005; Smales et al. 2013; Tucker 1996; Holstrom 2011). The models vary in numerous ways including whether static components such as the tower are included in calculations, if oblique angles of attack are considered and whether single or multiple turbines are assessed, as well as how avoidance behaviour is incorporated. Although the Band model (Band 2012) is the most widely used collision risk model in the UK, it is not the only one available and therefore any developments in our understanding of avoidance behaviour should consider, where possible, these alternative models.

Although described in the literature, avian collision risk models are often not presented in enough detail to reproduce. The majority of models consider avoidance behaviour as an add-on to the process of estimating the probability of collision, separate from the calculation of collision probability for a single rotor transit. From the information available, however, it would seem that the definitions and avoidance rates presented in our report would generally be suitable for use within a range of collision risk models, not only Band (2012). Here we provide examples of how the definitions and rates may align with some of these alternative models.

Desholm (2006) developed a stochastic model analysis of avian collision which included variability in the input parameters and outputs of the model. Although it was a very specific example from an offshore windfarm in the Baltic Sea, the method could be used elsewhere. The definitions used in our project seem suitable for the model. The method considered the different stages at which birds may avoid a windfarm and uses values for the proportion of birds entering the windfarm (1 - macro-avoidance), the proportion within the horizontal/vertical reach of rotor blades (1 - meso-avoidance) and also the proportion trying to cross the area swept by the rotor blades without showing avoidance (1 - micro-avoidance).

Eichorn et al. (2012) developed an agent-based, spatially-explicit model of red kite foraging behaviour to assess collision risk related to wind turbines. The model is largely stochastic and combines a spatial model with a collision risk model. Although the study was specific to red kite, the methods could be used more widely. The model uses the method from Band (2007) for calculating probability of collision from a single rotor transit therefore it is likely that any definitions for avoidance behaviour provided by our study will be suitable. The model specifically includes the probability of a bird recognising the threat and actively avoiding, and this avoidance rate is taken from the literature. The value ranges from 0.98 - 0.995 and is therefore likely to be a value for overall avoidance, however the definitions within this study for meso- and micro-avoidance would seem to fit more appropriately because it is a single bird avoiding a single turbine within a 100 m x 100 m grid cell.

Smales et al. (2013) describe a collision risk model developed by Biosis Propriety Limited which has been widely used to assess wind energy developments in Australia since 2002. The model uses a deterministic approach and provides a predicted number of collisions between turbines and a local or migrating population of birds. The model uses flight activity data from the windfarm site and applies avoidance rate to the typical number of turbines encountered per flight. Therefore the definitions and rates for within windfarm behaviour should be suitable in this context.

A note of caution when considering avoidance rates and their application within different collision risk models is that although not the intended purpose, avoidance rate may have become a sink for multiple sources of error and uncertainty within a model. Collision risk models rarely state the associated error along with collision estimates. In the process of apportioning overall avoidance into the different components of macro-, meso-, and micro-avoidance, this previous inclusion of model error may need to be considered, and may be model-specific.