Upland raptors and the assessment of wind farm impacts
Abstract
Government targets on renewable energy coupled with anthropogenic constraints on development have resulted in a surge in proposals to locate wind farms in upland areas, where they may conflict with the wellbeing of scarce or rare bird species including raptors. European and UK legislation demand that the effects of wind farm developments, both individually and in combination, be assessed to determine the level of impact on these species. The principle adverse effects of wind farms on raptors, as for other terrestrial birds, potentially involve disturbance (displacement or barrier impacts) or collision fatality. Few long-term studies on such effects of wind farms have been undertaken. We review available research results on displacement of raptors, which primarily involve foraging birds, and conclude that most studies indicate that displacement appears to be negligible, although some notable exceptions occur and more research is needed. There is also a need for better understanding of the numbers of birds likely to be killed through collision with turbine rotors at the site level in order to inform planning decisions, although models of bird distribution at several spatial scales can be used to circumvent potential difficulties when locating turbines. Modelling approaches have also been developed that attempt to quantify the theoretical risk of collision. One such approach, the Band model, is a valuable tool for impact assessment and its use is now widespread in the UK. However, there are practical problems associated with gathering the data required to run the model and numerous assumptions must be made concerning bird behaviour. This can lead to deficiencies in the input parameters which potentially have a large effect on the model outputs. Hence, we make recommendations for potential improvements, such as quantifying error in flight height estimation, training of observers in acuity skills, quantifying bird detection-distance functions, and research on factors influencing activity budgets and flight behaviour. In addition, the model outputs are usually adjusted to take account of turbine avoidance by birds and this aspect of birds’ behaviour is poorly understood. As a result of these limitations, collision predictions are only indicative, and more reliable in some situations, and for some species, than others.
INTRODUCTION
A rapid increase in the number of wind farms in the UK is currently in progress (Drewitt & Langston 2006). The constraints on locating terrestrial wind farm developments are many and varied, including adequacy of the wind resource, concerns over impacts on the visual amenity of the landscape, and issues related to civil and military aviation, electro-magnetic transmissions, hydrology, ecology and archaeology. Crucially, locations in areas of relatively low conservation value are often otherwise contentious because of their close proximity to residential property. These factors have inevitably favoured proposals within upland areas that have low human population (SNH 2004), where they may conflict with the conservation of scarce and dispersed birds, including several species of raptors (Ratcliffe 1990).
European and UK legislation require that the impacts on nature conservation interests due to wind farm proposals, individually and in combination, are formally assessed under planning and environmental impact assessment procedures (e.g. Langston & Pullan 2003). However, we currently have limited understanding of the implications of the various potential effects of wind farms on birds (Drewitt & Langston 2006) and as a result there is a tendency for planning decisions to be heavily dependent on subjective judgement. Greater objectivity can only be achieved with better information on the factors influencing wind farm impacts and with better assessment tools to increase predictive capacity. In this paper we first review briefly studies that have aimed to investigate the principal factors affecting the impact of wind farm developments on upland raptors. We next discuss some of the predictive approaches being used to better quantify the likely effects of wind farms, especially regarding bird collisions with turbines. Finally, we draw attention to the many potential pitfalls in making quantitative estimates of collision risk and make suggestions for how methods may be improved.
EFFECTS OF WIND FARMS ON RAPTORS
The effects of a wind farm on birds depend on a wide range of factors including the specification of the development, topography and habitat of the surrounding land, and species present (Johnson et al. 2000a,b, Percival 2000, Erickson et al. 2002, Langston & Pullan 2003, Barrios & Rodriguez 2004, Smallwood & Thelander 2004, Hoover & Morrison 2005). Ultimately, the main effects of terrestrial wind farms on birds are determined by the sensitivity of individuals to disturbance and therefore whether displacement or a barrier effect (if disturbance occurs) or risk of collision with rotating turbine blades (if disturbance does not occur) is the primary impact. Disturbance can occur as a result of the operation of turbine rotors, human and machine noise during site construction activities, maintenance and repair work, and increased access (for agricultural or recreational purposes) following the establishment of site access tracks (Gill et al. 1996, Percival 2000, Langston & Pullan 2003, M. Reichenbach pers. comm.). Disturbance (visual and noise) may displace birds into less suitable habitat and this may reduce their ability to survive and reproduce (i.e. an indirect form of habitat loss through perceived predation risk) (Frid & Dill 2002). If not physically displaced, birds’ foraging success may be impaired and other important life history behaviours disrupted (i.e. a form of habitat degradation) (F. Bergen pers. comm.). Alternatively, birds that are not displaced risk collision with rotating rotor blades (e.g. Erickson et al. 2001).
Most impact assessments to date have focused on displacement and collision risks and ignored the potential effects of behavioural disruption (Johnson et al. 2000a, F. Bergen pers. comm.). Clearly, birds cannot be simultaneously vulnerable to disturbance and collision since they are mutually exclusive spatially. However the relationship between the two effects may be unstable over time as birds may habituate to the presence of the wind farm (although see Stewart et al. 2005) or site faithful individuals may be lost from the population and not replaced. Post-construction impact studies should therefore ideally be conducted over a period relevant to the generation time of the species involved.
Displacement
Broadly, two methods have been used to investigate bird displacement effects at wind farms (Anderson et al. 1999): BACI (Before-After-Control-Impact) studies and those lacking a pre-construction contrast where abundance and distribution are contrasted between an operational wind farm site and a reference site lacking a wind farm. BACI methods provide the most powerful evidence but are probably less common, especially in European studies (Gill et al. 1996, Langston & Pullan 2003, M. Reichenbach pers. comm.). Displacement of breeding raptors may occur by disturbance preventing use of nest sites, and/or by disturbance preventing use of areas for other activities, notably foraging. Due to the low nesting density of breeding raptors (Newton 1979), studies involving potential displacement of raptors from nest sites are far less common than those examining displacement from foraging areas and inevitably involve lower sample sizes.
Potential displacement of raptor foraging has been investigated in several species, and at several wind farms for some species (Table 1). Although clearly more studies are highly desirable and more should be peer-reviewed in the public domain (Whitfield & Coupar, 2006), in general most research tends to show that disturbance of raptors at wind farms is negligible. Interesting differences are evident in some examples, however.
| Species | Sensitivity to displacement | Number of studies | References* |
|---|---|---|---|
| Turkey Vulture Cathartes aura | Low? | 2 | 1, 4 |
| Griffon Vulture Gyps fulvus | Low? | 1 | 2 |
| Red-tailed Hawk Buteo jamaicensis | Low | 4 | 1, 3, 4, 5 |
| Ferruginous Hawk Buteo regalis | Low? | 2 | 1, 4 |
| Rough-legged Buzzard Buteo lagopus | Low? | 2 | 1, 3 |
| Swainson's Hawk Buteo swainsoni | Low? | 2 | 3, 4 |
| Broad-winged Hawk Buteo platypterus | Low? | 2 | 3, 5 |
| Common Buzzard Buteo buteo | Low-Medium | 6 | 6, 9, 10 |
| Sharp-shinned Hawk Accipiter striatus | Low? | 1 | 5 |
| Golden Eagle Aquila chrysaetos | Low-High? | 4 | 1, 4, 7, 8 |
| Short-toed Eagle Circaetus gallicus | Low? | 1 | 2 |
| Common Kestrel Falco tinnunculus | Low | 5 | 2, 6, 9, 10 |
| American Kestrel Falco sparverius | Low | 4 | 1, 3, 4, 5 |
| Prairie Falcon Falco mexicanus | Low? | 2 | 1, 4 |
| Peregrine Falcon Falco peregrinus | Low? | 2 | 1, 6 |
| Red Kite Milvus milvus | Low? | 3 | 6, 9, 10 |
| Black Kite Milvus migrans | Low? | 1 | 2 |
| Hen Harrier Circus cyaneus | Low-Medium? | 5 | 1, 3, 4, 5, 10 |
| Montagu's Harrier Circus pygargus | Low-Medium? | 2 | 9, 11 |
| Marsh Harrier Circus aeroginosus | Low? | 3 | 9, 10 |
- 1 = Schmidt et al. (2003); 2 = de Lucas et al. (2004); 3 = Johnson et al. (2000b); 4 = Johnson et al. (2004a); 5 = Kerlinger (2002); 6 = Phillips (1994); 7 = Hunt et al. (1995); 8 = Walker et al. (2005); 9 = F. Bergen (pers. comm.); 10 = M. Reichenbach (pers. comm.); 11 = M. de Lucas (pers.comm.)
- The review refers to displacement from foraging areas, and not from nest sites, for which too few studies were available. Sensitivity to displacement classed as: Low = no evidence of any displacement effects, Medium = evidence of possible displacement in an area around individual turbines or from parts of a wind farm, High = evidence of complete displacement from a wind farm. A question mark indicates that it was difficult to reach a definitive conclusion because either: the study results were inconclusive or the number of studies was too low and/or studies did not involve a BACI protocol, or that interstudy variation in results clearly warranted further investigations.
Two studies involving the Golden Eagle Aquila chrysaetos in the US have found no evidence of displacement, one being a simple comparison of the wind farm and a reference site (Schmidt et al. 2003) and a second BACI study (Johnson et al. 2000a). From data at a third US wind farm (the Altamont Pass Wind Resource Area in California) presented by Hunt et al. (1995) there was indication of some displacement by comparison with a reference area away from the wind farm. Control of eagle prey in the wind farm may have produced this effect according to Stewart et al. (2005), although prey control was not encouraged by turbine owners until 1997 (Smallwood & Spiegel 2005), and the wind farm otherwise appeared to increase prey abundance (Smallwood & Thelander 2004). A long-term trend towards increasing raptor use of the Altamont Wind Farm may indicate that any earlier displacement no longer occurs (Smallwood & Thelander 2004). A study at a Scottish wind farm in Argyll (Walker et al. 2005) described shifts in the ranging behaviour of territorial Golden Eagles following wind farm construction and these shifts were consistent with avoidance of the entire wind farm area. The only flights observed crossing the wind farm related to nonterritorial birds, or excursions by the resident birds to intercept and aggressively interact with such birds. No change in the size of the range used by the eagles was apparent. However, the confounding effects of habitat change, a common problem in post-construction studies, must be acknowledged. Immediately prior to wind farm construction, 280 ha of plantation forest was cleared in order to mitigate the potential displacement effects. This resulted in a substantially increased foraging area located away from the wind farm.
At several wind farms, there have been no indications of Hen Harrier (Circus cyaneus) displacement, (Johnson et al. 2000b, Kerlinger 2002, M. Reichenbach pers. comm., Schmidt et al. 2003) however Johnson et al. (2000a) found evidence of both small scale (< 100 m from turbines) and larger scale avoidance of turbines by harriers in the year following wind farm operation. Preliminary results at Argyll and Northern Ireland sites suggest foraging may be little affected, but local displacement of nesting attempts may occur in the order of 2–300 m around turbines (Natural Research unpubl. data).
Collision risk
Few wind farm collision casualties have been recorded in the UK, (e.g. Percival 2000). However, it is important to recognize that, to date, not many wind farms have been built in areas where high concentrations of activity make collisions likely. Moreover, the data available thus far are mostly based on casually found victims, with no corrections for corpses that are overlooked or removed by scavengers (Langston & Pullan 2003); hence the studies can not be considered representative (Whitfield & Coupar 2006). Smallwood & Thelander (2004) also point out that several years of study may be necessary before reliable results are obtained. Remote methods of monitoring collision mortality may be the best way forward (Desholm 2003), although these need to be reliable and sufficiently inexpensive to allow installation at a large number of turbines to counter the likelihood that risk can vary substantially between turbines within a wind farm (Smallwood & Thelander 2004).
Better studies are available from abroad, although even here relatively few have countered all the potential pitfalls in methods used to document fatality rates (Whitfield & Coupar in press). Even when such potential shortcomings have been accounted for, simply presenting mortality rates per turbine or per installed MW, in the absence of any contextual information on the abundance of birds ‘available’ to be killed or otherwise at risk of death, does little to inform about the collision risk posed by wind farms (Smallwood & Thelander 2004). Low collision mortality may often be because wind farms have been located to avoid areas where bird concentrations are high or because turbines have been microsited to avoid localized areas of heavy use (e.g. Johnson et al. 2000a).
Despite such difficulties, there is a degree of consensus that raptors may be more vulnerable to collision than several other bird groups (NWCC 2000). This may be due to generally lower displacement or avoidance effects11
For clarification we follow Band et al. (2005) in defining a bird's avoidance of the whole or part of an operational wind farm as ‘displacement’ and it's avoidance of a moving rotor when on a path of potential collision as ‘avoidance’.
, although several other factors have been postulated as influential. The relatively greater fatalities of predatory species compared with scavengers such as the Turkey Vulture Cathartes aura or Raven Corvus corax at early Californian wind farms has contributed to the suggestion that hunting raptors may become ‘fixed’ on prey so that they may not see turbine blades and hence collide (e.g. Orloff & Flannery 1992). Experiments on visual acuity of raptors and painting of turbine blades (e.g. McIsaac 2001) suggest that some painted patterns may increase conspicuousness, but whether the principle is worthwhile practically is debateable (Young et al. 2003a). Moreover, several European studies imply that scavengers (e.g. the Griffon Vulture Gyps rueppellii) may be susceptible to collision in some situations (Langston & Pullan 2003, Barrios & Rodriguez 2004) so a predatory lifestyle may not be too influential. Preliminary studies at Altamont implicated casualties being a result of attraction to lattice towers as perch sites (Orloff & Flannery 1992) but detailed analyses with more data (Smallwood & Thelander 2004) have dismissed its importance.
Topography and weather are increasingly highlighted as being influential in raptor collision risk (Johnson et al. 2000a, Hunt 2002, Langston & Pullan 2003, Young et al. 2003b, Barrios & Rodriguez 2004, Smallwood & Thelander 2004, Hoover & Morrison 2005). Both factors have long been known as important for the movements of migrating raptors (e.g. Smith et al. 1985) although realization that they may also be relevant for the movement of resident or breeding birds (and therefore home range use) has only been relatively recent and has been rarely studied (e.g. Jiménez & Jaksic 1993, Bögel & Eberhardt 1997, McLeod et al. 2002a,b). More research on topography and weather and their interaction affecting space use and local movements of raptors is clearly desirable: allowing safer micrositing of turbines would be an obvious practical benefit (Johnson et al. 2000a, Young et al. 2003b).
PREDICTION MODELS
Spatial models
The location of a wind farm is one of the few certainties known to affect the impact of a wind energy scheme on birds (NWCC 2000, Langston & Pullan 2003, Percival 2005). Therefore spatial models that attempt to predict areas of greatest sensitivity for birds at the landscape scale (e.g. Williams et al. 1996) can be useful design tools, enabling developments to be located so as to minimize the potential effects on identified key species. At a finer scale, spatial models can be used to evaluate the relative effects of individual turbines within particular locations.
Reliable predictive models can be developed based on data from studies elsewhere. For example, McGrady et al. (1997) developed a simple model to predict the ranging behaviour of breeding Golden Eagles based on radio-telemetry studies in western Scotland. This model has been advanced through incorporation of additional parameters, principally topography, by McLeod et al. (2002a,2002b) – the so-called PAT (Predicting Aquila Territory) model. The model has the ability not only to predict potentially problematic turbine locations but also the likely overlap with a given wind farm layout and the extent of eagle range use (and therefore the likely extent of range loss should displacement occur). Additionally, it can take account of existing constraints on eagle ranging, for example due to forestry, and hence by comparing predictions of the existing patterns of ranging behaviour with those in the absence of habitat constraints it is possible to gain a useful insight to the scale and distribution of habitat management required in order to mitigate potential wind farm displacement effects. A similar model for Hen Harriers (Madders 2003) is currently being evaluated.
Apart from assisting with site location, design and mitigation, generic models are a good starting point on which to base impact assessments and should be used more frequently as they are cost-effective and can identify potentially problematic locations early. However their accuracy is improved considerably by the inclusion of site-specific information on bird activity patterns. Moreover, there is insufficient knowledge for most species to develop predictive models, and therefore site-specific observations are essential.
Collision risk models
Raptor activity may be a predictor of raptor collision risk when making inter or intrasite comparisons (Erickson et al. 2001). If raptor use of a site, or an index of exposure to collision risk is related to a measure of fatality rate, gathered at existing wind farms where comparable data has also been collected, a crude predictive method would be to use the data from operational farms to estimate potential collision fatality, from exposure indices at proposed farms. An illustration of this approach, combining data from Golden Eagles and the Red-tailed Hawk Buteo jamaicensis at selected US wind farms, is shown in Fig. 1. Aside of the dubious practice of combining data from different species22
The form of any functional relationship is liable to vary between species, and limited evidence suggests that in, for example, American Kestrel the kill rate may rise more rapidly with increasing risk exposure according to data in Erickson et al. (2001, 2002).
, the main difficulty with this method is that correlation between use and fatality is low at several newer generation wind farms (Erickson et al. 2001), and so any relationship may also be heavily influenced by one or two sites where fatalities are high. For example, the high correlation in the example of Fig. 1 (R2 = 0.96) is lost when the two extreme fatality rates and exposure indices outliers (both referring to Altamont) are removed (R2 = 0.12). The absence of useful correlations and potential sensitivity to results from a few sites where fatality and risk exposure may be relatively high, has meant that predictions of collision fatality (which adhere to the basic premise of this approach) have tended to be crude, applying to proposed sites the average or range of fatality rates observed at other sites (e.g. Erickson et al. 2003).
The relationship between an index of collision risk and collision rate as observed at several USA wind farms for Golden Eagles and Red-tailed Hawks using data from Erickson et al. (2001, 2002). The index of collision risk is given by mean bird use per 20 min of circular 800 m radius observation plots located within the wind farm site multiplied by the proportion of bird observations at rotor swept height. Collision rate was estimated by searches for collision victims calibrated to account for underestimation due to several biases such as scavenging. Data were used from Altamont, Foote Creek Rim, San Gorgonio Pass, Tehachapi Pass and Buffalo Ridge (see Erickson et al. (2001, 2002) for source studies) and to retain clarity for their illustrative purpose have not been transformed for normalization.
It is difficult to question the crudeness of this predictive approach, as it reflects the current uncertainty in the causes of variation in fatalities. However, when a range of potential collision fatality estimates includes a high value, then the scope for disagreement between different stakeholders potentially increases. Cumulative fatality estimates based on such predictions will also exaggerate the range of predicted estimates in the absence of monitoring studies to give ‘true’ fatality rates; a relevant consideration given the rapid pace of wind energy development in some areas. A desire both for more ‘precise’ fatality estimates and the ability to contrast the risk posed by different turbine types or other scenarios has lead to the development of predictive models of collision risk (whilst models do not necessarily improve fatality estimate ‘precision’, they do make the influence of different parameters on collision more explicit: Band et al. 2005, Chamberlain et al. 2005, 2006).
Tucker (1996a,1996b) produced the first such model, but it has not been widely applied. This may be because the model did not account for the extent of the volume swept by rotor blades, which is clearly pertinent (e.g. Smallwood & Thelander 2004) with the assumption that birds did not avoid collision, which is not the case (e.g. Gill et al. 1996). A more recent model (Podolsky 2003, 2005) appears superficially similar, but is more complex than that of Tucker's and incorporate rotor swept volume and avoidance/ ‘attraction’ factors (although published details are incomplete). This model may be gaining some recognition for practical utility (Schwartz 2004, Smallwood & Neher 2004, Podolsky 2005). A further collision prediction model has been developed by Biosis Research and has been applied in several wind farm assessments in Australia (Organ et al. 2002, Timewell & Meredith 2002, Biosis Research 2003). We have not been able to access full details of this model; nevertheless, it appears similar to others but may be more simplistic in some respects (e.g. it does not explicitly acknowledge that not all birds flying through the spinning rotor blades will necessarily be hit).
Initially developed at around the same time as Tucker's model, the ‘Band’ collision risk model (Band et al. 2005) has been used in Environmental Impact Assessments for several years in Scotland, and more recently in the rest of the UK (Chamberlain et al. 2005, 2006). An appraisal of the Band model (Chamberlain et al. 2005, 2006) has noted that whilst it appears generally robust there is a strong influence of avoidance rates on estimated collision risk and that information on avoidance rates is scant, confirming the conclusions of Band et al. (2005). The particular sensitivity of fatality estimates to avoidance is a common issue to collision risk models which incorporate this factor (Biosis Research 2003, Podolsky 2005). Although this problem has to be acknowledged, we do not agree with the notion that it is severe enough to warrant abandoning the use of collision risk models (Chamberlain et al. 2005, 2006) because the alternative methods of estimating risk are no more robust (being subject to the same sources of potential error, but without the explicit quantification of relevant parameters) and at least there appears to be a broad consensus that avoidance rates are generally very high (e.g. Biosis Research 2003, Band et al. 2005, Podolsky 2005) and that there is scope for their estimation.
With currently available data, one method for estimating avoidance rate is the difference between predicted fatality and observed fatality (avoidance rate = 1 –[observed deaths/predicted deaths assuming no avoidance]). (Note that this means, however, that an avoidance rate calculated using one model may not be transferable to another model.) Bias associated with measuring observed deaths through carcass searches is increasingly highlighted in recent reviews and was rarely accounted for in early fatality studies ( Gill et al. 1996, Morrison 2002, Langston & Pullan 2003, Band et al. 2005, Chamberlain et al. 2005, 2006, Whitfield & Coupar 2005) prompting concern that it may affect measures of avoidance rates ( Band et al. 2005, Chamberlain et al. 2005, 2006). Recent studies however, especially in the US, have addressed this problem (e.g. Erickson 2003a, Smallwood & Thelander 2004) although it may be more intractable for offshore sites when remote sensing of avoidance and strike rates may be essential (Desholm 2003). For diurnal raptors, the scope for reliable estimation of avoidance rates using searches for carcasses and other remains may be better than for other bird groups because of their size (rendering carcass detection more likely) and because estimations of utilization rates (relevant to the ‘predicted deaths’ component of the calculation of avoidance rates) should not be biased by a lack of nocturnal observations, which typifies the vast majority of wind farm studies.
Modelling collision risk under the Band model is a two-stage process (Band et al. 2005). Stage 1 estimates the number of birds that fly through the rotor swept disc. Stage 2 predicts the proportion of these birds that will be hit by a rotor blade. Both stages are prone to bias due to the inclusion of relatively simplistic assumptions about bird behaviour. As we noted above, much attention has focused on the adjustment of the output from Stage 2 to allow for a level of turbine avoidance by birds. However, rather less attention has been given to several potentially influential biases associated with Stage 1, although both Band et al. (2005) and Chamberlain et al. (2005, 2006) discuss some of the possible biases in recording bird activity levels and the sensitivity of collision risk models to bird activity levels, respectively. These potential biases are also obviously relevant to the estimation of avoidance rates (under the ‘predicted deaths’ component of the calculation noted above) and other methods which attempt to describe bird utilization rates at wind farms, and are dealt with more fully below.
RECORDING BIRD ACTIVITY
Stage 1 of the Band model is based on field observations that aim to quantify the level of flight activity within the proposed wind farm and immediate environs (typically, the turbine envelope plus 500 m). Under the protocol suggested by Band et al. (2005), this is done using time budget data gathered from vantage points strategically located with respect to the area of interest. The method used depends on the type of flight behaviour exhibited.
In Type I models birds are assumed to move along fairly predictable corridors. Examples include diurnal movements of wildfowl, feeding trips by breeding divers (Gavia spp.), and spring and autumn migratory bird movements. In Type II models bird movements are assumed to be less predictable, for example ranging activity by territorial and non-territorial raptors.
The collection of data for most Type I models is relatively straightforward, since observations are generally focused on a nest location (e.g. divers) or expected flight corridor (e.g. diurnal movements of geese). An exception is the case of ‘broad-front’ migratory birds where several vantage points will probably be necessary, aligned perpendicular to the expected line of flight.
In Type II models a critical data collection consideration is to ensure that there is appropriate spatiotemporal coverage of the proposed wind farm area. Thus, vantage points are selected that maximize the visibility of the survey area, ideally such that all parts of the survey area lie within 2 km of a vantage point. In order to minimize observer effects on bird behaviour this is achieved using the least number of vantage points possible. Where feasible, vantage points are distributed outside the survey area; again, this is to minimize any effects of the observer on bird use of the survey area. It is worth noting that this is in stark contrast to the typical survey method used in most US and Australian wind farm studies where an observer records bird activity per unit time within circular plots of variable radius (often 800 m) located within the study area (e.g. Johnson et al. 2000a, 2000b, Erickson et al. 2003): more detailed descriptions of the method can also be found in Reynolds et al. (1980) and Bibby et al. (1992). Locating observation points within a wind farm study area incurs the possibility of observer effects on measures of bird utilization rates, and was first raised by Orloff & Flannery (1992), but this potential bias does not appear to have been considered subsequently in relevant wind farm studies.
In practice, locating vantage points outside the study area is often incompatible with the requirement for all parts to be within 2 km of at least one vantage point, especially at large sites. This problem is unlikely to be serious, unless data are being collected from a vantage point within the area under observation at that time. It should be noted that overlap in visibility between the various vantage points is inevitable, and whilst this may increase the effective temporal coverage, it may also violate statistical assumptions that the visible areas from each vantage point are effectively independent (e.g. mean values of flight activity from each vantage point are used as input data in the Type II example calculations given by Band et al. 2005). Provided that observations are not collected from different vantage points simultaneously, however, this violation should not be serious. The ‘circular plot’ (or point count: Bibby et al. 1992) method is less likely to incur this violation, if observation points within the study area are located randomly (Reynolds et al. 1980, Anderson et al. 1999) because spatial coverage is not complete, relying on selected observation points to be representative of the study area as a whole. This does, of course, render the circular plot method vulnerable to bias if the selected observation points are not representative through, for example, too few being chosen.
Adequate temporal coverage is achieved through a study design appropriate to the species known/ suspected to be present and the phenology and expected levels of flight activity (Anderson et al. 1999). It is important that observations are stratified across periods corresponding to different seasons (breeding, migratory, other non-breeding) and, within the breeding period, from different behavioural epochs (for example prebreeding territorial activity, incubation, fledging, post-breeding dispersion). This allows flight activity across the year to be determined with greater precision. In addition, observations should be stratified diurnally across the period during which the target species are known to be active in order that observations are efficient (Anderson et al. 1999). This will vary according to species, making it difficult to combine observations for raptors with those for some species of wader, for example. The overall amount of time devoted to vantage point observations should ideally be determined by undertaking a power analysis of data collected over a reconnaissance period. Further work on this aspect is urgently required in order that adequate sample sizes of data are gathered without unnecessary expenditure of effort.
Additional biases in recording bird activity
Missed observations
Unlike data collection for Type I models, which is usually geared towards a single or related group of species, in Type II models, observers often attempt to gather information on a range of disparate species. If too many target species are chosen or the number of individuals is considerable, there is a danger of observers becoming overwhelmed by a large number of simultaneous flying bouts if the observation protocol is designed to collect time budget data (duration and location of flight, flight height). Similarly, if the species involved have very different behaviours, it may be difficult to adequately scan the landscape for potential activity (for example, searching the skyline for soaring raptors whilst looking for local movements by upland waders). Crucially, it is generally possible to track only one bird at a time. This problem is inherent to many observation methods used in wind farm studies and, importantly, the disparity between recorded and actual activity will be nonlinear, becoming greater as actual activity increases. Therefore, bird activity will be increasingly underestimated. This is especially pertinent when many reviews recommend that wind farms should not be located on sites with high bird activity (e.g. Langston & Pullan 2003) and also means, incidentally, that avoidance rates calculated at operational facilities may be underestimated (cf Chamberlain et al. 2005, 2006).
There are several potential solutions that are not mutually exclusive and their efficacy will vary according to the potential severity of the problem. If the problem is considered not severe, so that temporal overlap in bird activity is negligible, time spent watching target species may be deducted from the overall observation time before activity rates are estimated. Observer number can be increased and labour divided (e.g. Smallwood & Neher 2004). Instantaneous records (Altmann 1974) of the number of birds present can be made at set intervals if bird numbers are high (if numbers are low this method will probably generate a large number of zero values which can be statistically awkward), with more detailed time budget observations collected separately, or in intervening periods on a sample basis (e.g. Johnson et al. 2000a).
Observer acuity
A long-standing plea in reviews and methodological examinations of wind farm/bird interaction studies has been for standardization of approach (e.g. Gauthreaux 1995, Anderson et al. 1999, NWCC 2000). Inter-observer variation in acuity skill countermands standardization (Gauthreaux 1995) and although flight activity recording protocols can be made consistent to a degree through training, ultimately the method is dependent on a high level of observer field skill. The detection of some species requires experience. In many upland areas in the UK, an endemic problem is that long periods of time can elapse between periods of (sometimes frenetic) flight activity. As a result, interobserver differences may be a large source of bias when comparing studies.
Unfortunately, whilst substantial improvements in avian studies at wind farms have been made in recent years, the issue of observer variability has not been addressed, despite its influence being apparent from the earliest methodological review (Gauthreaux 1995). Hence, this aspect is rarely acknowledged in environmental impact assessments, and even more rarely accounted for. Consequently, while other sources of bias have been recognized in syntheses or ‘meta-analyses’ of wind farm impacts, observer variability has been ignored (Erickson et al. 2002, Erickson 2003b), most liekly because there was little that the investigators could do about it. The problem remains, nevertheless, and has become more acute in recent times as skilled observers become relatively less available as a result of the accelerating pace of wind energy development. We would urge therefore that greater attention is paid to this issue through, for example, increased use of observer training and calibration of acuity skills, with such practices being incorporated in to assessments, which place greater emphasis on documentation of and accounting for observer experience.
Visible area
Under the Stage 1 observation protocol advocated by Band et al. (2005) measurements of the airspace within which observations are made can be achieved in a variety of ways and the various approaches can yield widely differing values of flight activity per unit area. For example, some assessments estimate the area of ground visible by eye in the field, whilst others use view sheds generated by GIS software: the latter is most likely more effective (Band et al. 2005). When assessing potential collision risk in areas with varied terrain, view sheds usually describe the visible area of an imaginary laminate suspended at the lowermost height passed through by the proposed turbine rotors (typically 30 m), in order that the potential future risk zone is adequately covered. Interestingly, under the circular plot method as applied in wind farm studies, the characteristic assumption that all airspace is visible seems implicit (e.g. Erickson et al. 2002). Given the terrain in which many wind farms are proposed, we doubt if this assumption is always valid even though in some studies plots may be relocated to ensure that at least the majority of the outer limit of the plot can be seen (e.g. Johnson et al. 2000a).
Observers generally scan for birds within an arc of around 180° under the Band et al. (2005) protocol and thus the actual area observed is often much smaller than that potentially visible. The error this potentially generates is minor compared to that generated by the circular plot method because at any one time, the majority of the presumed observation area will not actually be under observation by a single observer. Surprisingly, this has attracted little critical attention (Reynolds et al. 1980, Bibby et al. 1992) and while indices of activity may be little affected for comparative purposes, absolute indices of activity are bound to be more seriously affected (lowering estimates of avoidance rate, for example).
Detection of flying birds
An important consideration in generic flight activity observations (as opposed to those focused on particular individuals or species) is that the probability of detection is likely to vary between species, and within species can probably vary according to sex, behaviour, terrain and habitat. Thus detection functions are probably both species and site specific. For example, large, slower flying birds that soar conspicuously above the skyline (e.g. Golden Eagle) are likely to be detected at far greater distances from the observer than small, fast flying birds that are cryptic either due to their plumage or behaviour (e.g. Merlin Falco columbarius) (Fig. 2). During observation sessions at vantage points conducted at several proposed wind farm sites in Scotland, for Merlins the area around each vantage point within which birds could be reliably detected appeared to be around 176 ha (within a radius of approximately 750 m: Fig. 2). The observations also suggested that there may be a depressed effect on detection of Golden Eagles to a distance of at least 750 m from the observer, presumably due to disturbance (Fig. 2). A further problem arises with Merlin because the disturbance effects due to the observer are likely to be disproportionately large within the area where the species is reliably detected. Therefore the problem of reduced detectability can not be overcome simply by using more, closely spaced, vantage points. In practice, vantage point observations may be unsuitable for the purpose of generating realistic estimates of flight activity by such birds; alternatively, incorporating a distance sampling approach (Reynolds et al. 1980, Buckland et al. 1993) may be necessary.
Distance of Golden Eagles (upper chart: n = 112) and Merlins (lower chart: n = 85) from an observer at point of first detection collected during vantage point observations at several proposed wind farm sites in Scotland. The frequency distribution of observations with respect to detection distance was significantly different between species (Kolmogorov-Smirnov Z = 6.21, P < 0.001).
Similar considerations apply to variation in terrain and habitat, with birds visible over much larger areas in gently undulating open ground than in areas with rugged relief or convex slopes, or where lines of sight are constrained by woodland. This can lead to variation in the detection of species within a single site, although to some extent it may be possible to overcome this through the judicious location of vantage points.
Flight speed
Flight speed is a common input parameter to collision risk models, and whilst it may have only limited influence on fatality estimates and then mainly at low speeds (Podolsky 2003, Chamberlain et al. 2005, 2006) it is typically derived from the literature and based on a relatively small set of data from studies undertaken in situations which may not be relevant to those in which the birds are potentially at risk, for example due to differences in habitat or behaviour (Spaar 1997, Bruderer & Boldt 2001). It is further assumed that birds fly at a constant speed for any given predictive run of a model (e.g. Band et al. 2005). Constant speed is highly unlikely since a bird's flight speed varies according to its mode of behaviour (i.e. whether hunting, transiting between locations, displaying, interacting with other birds, etc.), flight height and also due to wind speed (for example, Hen Harriers often use headwinds to quarter the ground methodically for prey at slow flight speeds, then fly downwind at rapid speed prior to turning and resuming a slow quartering flight mode). Collision risk is not linearly related to flight speed (Podolsky 2003, Chamberlain et al. 2005, 2006).
Air speed is the actual speed of a bird through the air mass before accounting for the effect of wind or other weather conditions. Ground speed is the speed of a bird compared to the ground after factoring in the effect of wind or other weather conditions and in the context of collision risk is the more relevant metric. Variable ground speed due to wind speed and direction, such as in the Hen Harrier example above, can be accounted for under Stage 2 of the Band model which estimates the probability of a bird passing through rotor blades being struck. Stage 2 has upwind and downwind components (Band et al. 2005), and by separately calculating the effect of wind speed decreasing or increasing air speed on upwind or downwind flights, respectively, an average value of the collision probability can be deduced. A further improvement would be to weight the influence of downwind and upwind flights according to their occurrence in the field, but appropriately targeted field observations would be required. Hence, we would make a further recommendation that such observations are collected in wind farm studies, and would encourage more measurements of flight speeds in an appropriate context.
Flight height
The frequency of flights at rotor blade height is clearly relevant to most studies of collision risk at wind farms, whether for computing estimates of activity within a ‘risk volume’ for a collision risk model (Organ et al. 2002, Podolsky 2003, Band et al. 2005) or for computing indices of risk exposure (e.g. Erickson et al. 2002). Activity at rotor blade height is usually estimated by eye at the point of detection and can then be recorded subsequently at regular intervals. Observers must judge the height of a variety of species over differing types of terrain and habitat at distances ranging from a few hundred metres to several kilometres. There are generally few, if any, landscape features against which the accuracy of height estimates can be checked and parallax effects can further confound the determination of the correspondence between a bird and the ground over which it is flying. Thus, for example, it is possible for an observer to erroneously believe that the bird is flying a short height above a summit ridge when in fact the bird is flying at much greater height over a slope some distance in front of, or behind, the ridge.
Such considerations are consequently important not just for ensuring any error in flight height estimates are addressed by estimates of fatality risk, but also in documenting error in recording the flight paths or location of flying birds because topography may be important in the flight behaviour of some raptors and turbine relocation away from intensively used areas is a potential mitigation tool (Johnson et al. 2000a, Mcleod et al. 2002b, Young et al. 2003b, Smallwood & Neher 2004, Hoover & Morrison 2005). Observers involved in wind farm studies would therefore be usefully trained in flight height and distance estimation (range finder binoculars may not work in all field conditions) and any residual recording error, documented in staged exercises with known-height and known-location observation targets, accounted for in assessment or monitoring studies.
Extrapolation from observed data
There are few published studies describing the activity budgets of upland bird species and potentially influential factors (e.g. Collopy & Edwards 1989). Thus the number of hours per day that birds are potentially active, and the influence of factors such as weather, time of year, breeding status, etc. are poorly understood and so such factors are poorly incorporated in estimates of activity levels. Observations of flight activity are typically collected across the diurnal period and in a range of weather conditions. However, in order to maximize survey efficiency most observations have to be gathered in conditions of good visibility (e.g. Smallwood & Neher 2004) and therefore are unlikely to be representative. The flight activity levels of diurnal raptors are expected to be lower during poor visibility than at other times, although some species may exploit specific circumstances (for example, misty conditions may allow some raptors to closely approach their prey without being detected). It is likely that many raptors are least active during precipitation (e.g. Watson 1997).
Sample size is an important consideration in designing representative sample studies (Anderson et al. 1999). In the case of raptors, wind strength and direction are probably important influences on the distribution of foraging effort, given the reliance of several species on wind conditions for energy-efficient flight (Bögel & Eberhardt 1997, Watson 1997, McLeod et al. 2002b, Smallwood & Neher 2004, Hoover & Morrison 2005). It is important therefore that sampling effort is sufficient to document the range of wind conditions which prevail at a site to reduce risk of bias due to stochastic events.
From the above, it is clear that any adjustment made for weather/visibility must be judged on a species by species basis. Because of the difficulty in recording bird behaviour in poor visibility these adjustments are often made in the absence of quantitative information, yet they have a potentially large effect on the predicted flight activity value. Moreover, visibility is likely to strongly influence flying height (Richardson 2000) and this has implications for the proportion of extrapolated flights assumed to be at rotor height. Although it could be argued that raptors hunting in poor visibility would tend to fly below rotor height, their reactions when approaching rotors in such conditions are unknown. These considerations serve to reiterate the suggestion we made earlier in that more research on the factors affecting raptor flight behaviour and movements is sorely needed.
Finally, it must be remembered that although some birds may be less active in poor visibility, the ability of those that continue to fly to detect and avoid operating turbines is likely to be much reduced. In other words, low visibility may have contradictory effects in Stages 1 and 2 of the Band collision model. This is because in Stage 1 (estimating the number of birds that may pass through the rotor blades) without any correction for low bird activity in poor visibility when observations would not be taken, the number of birds at risk will be overestimated, whilst in Stage 2 the estimated probability that a bird passing through the rotor blades will be struck will be underestimated if low visibility decreases avoidance.
CONCLUSIONS
Despite limited evidence that upland raptors are displaced from wind farms in some situations it is apparent that many are at risk of collision with turbine rotors and any overhead lines. The appropriate use of spatial models at the earliest planning stage can help reduce this risk. Modelling collision risk can help to determine the approximate level of mortality likely to result from particular developments, enabling the consequences to be explored for local and regional populations. However, the reliability of collision models is limited by difficulties in gathering appropriate field data and by the large number of assumptions necessary during the modelling process, notably the level of collision avoidance. As a result care must be taken not to over interpret the model outputs, which are probably best used to evaluate different wind farm configurations, for example, rather than taken definitively until further improvements and realistic confidence limits are introduced to model parameterization. We recognize, nevertheless, that as alternative methods for estimating collision risk are less transparent or more subjective and at least as vulnerable to the same potential biases, conceptually robust models are useful tools that can probably incorporate future improvements in knowledge of factors influencing collision and, perhaps more crucially, steer research towards aspects where such knowledge may be most profitably gained.
The method of quantifying flight activity from vantage points is an important element in the assessment of wind farm impacts on birds. However it has a number of limitations, mainly related to observer skills and bird detection, and as a result it probably does not work well for all species and all situations. Alternative methods which survey bird activity also have similar and additional drawbacks. In many cases it may be more appropriate to gather reconnaissance data through vantage point watches before switching to alternative techniques, for example focal observations of individual birds or pairs of birds using visual observations, radio-telemetry or remote sensing by radar.
Footnotes
ACKNOWLEDGEMENTS
Thanks to Rhys Bullman for pointing us to the Biosis Research collision model, to Bill Band for several helpful discussions concerning collision risk modelling, and to James Pearce-Higgins and Jeremy Wilson for constructive comments on an earlier draft.
