Scottish Marine and Freshwater Science Volume 5 Number 13: Population consequences of displacement from proposed offshore wind energy developments for seabirds breeding at Scottish SPAs

4 Discussion

4.1 Summary of results

This study represents, to our knowledge, the most comprehensive assessment of the effects of displacement and barrier effects from wind farms on breeding seabirds yet undertaken. Using the best available empirical data and advanced modelling approaches across five species, we have demonstrated how these two factors may alter adult survival and breeding success mediated via changes in foraging energetics and body condition.

These results suggest the potential for declines in adult survival of more than 1% for Forth Island kittiwakes and Forth Island puffins, and for declines of more than 0.5% for Fowlsheugh kittiwakes and Forth Island razorbills. The results do not indicate any potential for declines of 0.5% or more for kittiwakes at St. Abbs, or for gannets or guillemots at any SPA.

Forth Island puffins show the largest estimated declines, but only if the distribution of prey is assumed to be homogeneous. Inch Cape and Alpha make the largest contributions to declines for this species- SPA combination (both have declines of more than 1% individually, but, again, only under an assumption of homogeneous prey distribution - if the prey is heterogeneous then the overall effect is much smaller and the main contribution is from Neart na Gaoithe). One possibility for this difference is due to the relative densities of birds in the wind farm footprint plus 1km buffer (Zone 4) compared to those in the surrounding 5km wide zone into which birds are displaced (Zones 3+5). Under heterogeneous prey, the 5km buffer area (Zones 3+5) may necessarily have quite different prey densities than the wind farm plus 1km exclusion area (Zone 4). For puffins at Alpha, the estimated density of birds within the wind farm footprint is much lower than that in the surrounding 5km area, and there is a large hotspot of predicted bird density (and therefore also of predicted prey in the heterogeneous prey simulations) just to the west of the wind farm. This means that under heterogeneous prey conditions displaced birds forage in a neighbouring location with a comparatively high density of prey and so little effect of the wind farm is felt. However, under homogeneous prey conditions displaced birds forage in an adjacent location with a comparatively lower prey level, which is unable to compensate for the increased density of birds and flight costs. Therefore, the effect of the wind farm in homogeneous prey conditions is much greater than that resulting from heterogeneous prey conditions where simulated prey much more closely matches the simulated bird distribution. The same is true for puffins at Inch Cape, although to a lesser extent because although the density of birds in the immediate vicinity of Inch Cape is relatively high compared to several other areas, the difference is not as great as that in the 5km buffer zone for Alpha. In summary, this means that birds displaced from wind farms under heterogeneous prey can move into areas with richer prey and so incur an advantage over their initial choice of foraging location that in part offsets the cost incurred. However, under homogeneous prey all areas are equivalent with respect to prey density, so this offset does not occur.

This result necessarily provokes the question as to which prey method, homogeneous or heterogeneous, is the most reliable. This is not a question that can easily be answered. Both methods rely on assumptions that are unlikely to be realistic in practice, but we do not know which of the two scenarios is likely to be closer to reality. Specifically,

1) the heterogeneous prey results assume that the density of prey can be directly inferred from the density of observed seabird foraging locations (within relatively small datasets), but in reality the GPS data may not give a complete picture of the density of foraging birds, and, further, the density of foraging birds is unlikely to be related solely to the density of prey.

2) the homogeneous prey results assume that prey is uniformly distributed across the Forth/Tay area, but this is clearly not true in reality.

We therefore recommend that the results from both methods should be considered, and that considerable caution should be applied to interpretation of all results. The greatest caution is needed in cases where bird distributions were inferred from GPS data for small numbers of birds, such as puffins, and in these situations the heterogeneous prey distributions are likely to be of particular concern.

Forth Island kittiwakes show cumulative declines of almost 2%, under both heterogeneous and homogeneous prey scenarios. Neart na Gaoithe appears to be the biggest contributor to this, with an estimated decline of more than 1%. Fowlsheugh kittiwakes show a cumulative decline of just under 0.5% (with either homogeneous or heterogeneous prey). This seems to be primarily driven by Alpha, which has an individual effect (under both homogeneous and heterogeneous prey scenarios) of between 0.5 and 1%. In all cases, the cumulative effect of all four wind farms is broadly similar to the sum of the effects of the individual wind farms.

Results for breeding success are qualitatively similar, but are of smaller magnitude (assuming that a 1% decline in adult survival decline corresponds to a 5% decline in breeding success; Freeman et al. 2014) and are more affected by stochastic noise (as shown by the assessment of reliability using additional runs; Tables H1 and H2). The only decline that is greater than 2.5% is for the cumulative impact on Forth Island puffins under homogeneous prey, and, although this decline is actually much closer to 5% than 2.5%, there are no declines of more than 5%. The effect of ~5% on Forth Island puffins appears to approximately decompose into a 2% effect of Inch Cape, a 1% effect of Neart na Gaoithe, a 1% effect of Alpha, and a 1% interaction effect.

The species- SPA-wind farm combinations with the largest declines in adult and chick survival generally correspond to those for which birds spend a substantial proportion of time in the zones (4, 5 and 6) that are affected by the wind farm. Forth Island puffins (for all wind farms except Bravo), Forth Island kittiwakes (for all four wind farms) and Fowlsheugh kittiwakes (for Alpha and Bravo) all have more than 2.5% of their foraging destinations in these zones. However, there are species- SPA-wind farm combinations with birds spending a substantial amount of time in these zones that do not have large estimated effects (Forth Island guillemots with Neart na Gaoithe; Forth Island razorbills with Neart na Gaoithe, and Forth Island gannets with all wind farms).

The interaction of Forth Island puffins with the Alpha wind farm entirely results from displacement effects, whereas the interaction with the Neart na Gaoithe wind farm is almost entirely in terms of barrier effects. Other important species- SPA-wind farm combinations involve a mix of barrier and displacement effects, but the percentage of time spent in barrier-related areas (zones 5 and 6) is often substantially larger than that spent in displacement-related areas (zone 4).

The relative effects on different study species result from variation in foraging ecology. Guillemots and razorbills typically have more restricted foraging ranges during chick-rearing than the other species (Daunt et al. 2011), resulting in limited interaction with wind farm footprints. The higher effects in kittiwakes and puffins are primarily because they have a greater foraging range than guillemots and razorbills, resulting in greater overlap with wind farms. Gannets from Bass Rock have foraging ranges that extend hundreds of kilometres beyond the wind farms (Hamer et al. 2007). For this species, the proportion of birds interacting with wind farms is comparatively high, but associated costs are small relative to the overall cost of foraging trips, so overall effects are negligible.

4.2 Uncertainty

4.2.1 Sources of uncertainty

There are a number of different sources of uncertainty associated within our results:

1) stochastic uncertainty associated with using a single run of the (full or fast) foraging model which involves a particular sample of birds ("sampling uncertainty");

2) uncertainty associated with the values of the parameters within the model ("parametric uncertainty");

3) uncertainty associated with the structure of the model that we use ("structural uncertainty").

Within the timescale of the project it has not been possible to perform a full quantification of uncertainty. Within the exploratory runs we were able to quantify sampling uncertainty and one particular component of parametric uncertainty. The results suggested that sampling uncertainty was substantial, and this motivated us to reduce this uncertainty by using a much larger sample of birds (20000 rather than 1000) when generating the final results. The additional computational effort required to run the larger samples uncertainty meant that it was not feasible to perform the additional model runs that would have been needed to properly quantify uncertainty within the final results, although we did use some additional runs to provide a rough quantification of sampling uncertainty. We can, nonetheless, make some general comments regarding the three different sources of uncertainty.

4.2.2 Sampling uncertainty

Additional simulations from the fast model ( Section 2.6.3 and Appendix H) suggest the level of sampling uncertainty within the final results is low for adult survival: i.e. the results obtained by simulating one set of 20000 birds are similar to those obtained by simulating a different set of 20000 birds. The level of sampling uncertainty for chick survival is considerably higher. The higher level of sampling uncertainty for chick survival is likely to stem from certain threshold effects in the foraging model that determine chick survival based on the amount of time nests are unattended by adults.

These results suggest that it was valuable to run the final analysis using larger samples of birds than those which were used for the exploratory analysis ( i.e. 20000 rather than 1000), but that uncertainty for adult survival would not be substantially reduced - and the results would not be qualitatively altered - if we were to re-run the analyses using more than 20000 birds. More precise results for chick survival could, however, be obtained through re-running the simulations with larger numbers of birds.

4.2.3 Parametric uncertainty

We used a sensitivity analysis to examine the effect of changing the values of specific parameters on the resulting estimates of chick and adult survival. The sensitivity analysis suggested that changes to the values of four of the parameters considered (adult body mass below which adult leaves chick unattended, chick body mass below which chick dies, adult priority of resourcing between self and chick, and intra-specific competition) have substantial - and in some cases very substantial - impacts upon chick survival. The impacts of changing parameter values on adult survival are generally much more modest, but the effects of wind farms on adult survival are highly sensitive to the values of the intra-specific competition parameter. Adult and chick survival both seem to be insensitive to the value of the fifth parameter that we considered (unattendance duration at the breakpoint).

The sensitivity analysis is of use in telling us which parameters are influential in the model, but should be interpreted cautiously. In some cases it will over-estimate uncertainty, because it will include parameter sets which would be associated with data characteristics (adult mass and chick survival in the baseline run) which would have led them to be rejected. In other cases it may under-estimate uncertainty, because a wider range than that considered would have led to data characteristics that would have been accepted.

The ideal way to avoid these problems would be through a full quantification of parametric uncertainty. Established methods for quantifying uncertainty within contexts such as this do exist ( e.g. Approximate Bayesian Computation), but are computationally intensive because they require thousands, or tens of thousands, of simulations and so could not feasibly be used within the timescale of this project.

We know that model outputs are very sensitive to some parameters that were not explored within the sensitivity analysis - the total amount of prey is the most prominent of these, and we know that small changes in this value can have very substantial effects on the model output. The barrier and displacement rates, which were agreed by the Steering Committee, are also likely to be important parameters in determining the magnitude of the response to the wind farm (and our exploratory analyses, which used different scenarios for barrier and displacement rates, suggest that this is indeed the case). The parameters associated with the adult mass-survival relationship are also likely to be influential: the large standard errors given by Erikstad et al. (2009) suggest that there is considerable uncertainty regarding the magnitude of the mass-survival relationship, but for species other than kittiwake the standard deviation of adult masses is also likely to be a key parameter (because this determines the magnitude of change in standardized mass that results from a change in absolute mass).

4.2.4 Structural uncertainty

Inevitably, due to a lack of data for some of the key foraging behaviours and processes involved in determining seabird response to wind farms, there are a number of structural uncertainties in our model that will have had a bearing on model results and conclusions. One of the most important is the uncertainty about the form of the adult mass-survival relationships, and the lack of data on this relationship for three of the species, and for any species based on local data. We have attempted to quantify this uncertainty to the best of ability given the available published data, but we are only able to do so within the bounds of the two published studies that are currently available. The only way to better account for this would be to analyse local data (available for kittiwakes and guillemots) or collect new data (required for razorbills, puffins and gannets).

Some key behavioural responses are simply unknown: for example, how birds would balance the number of foraging trips taken against additional barrier flight costs imposed by wind farms. Our model has been structured to include behavioural processes that we believe are likely to result from the addition of a wind farm, but there is no way to assess the legitimacy of these processes without additional data. Various exploratory attempts to improve/amend the behavioural assumptions within the models did suggest that the magnitude of the wind farm effect was strongly related to the assumptions that we made about how birds determine the number of trips they will do in a day. However, we were unable to fully assess the consequences of alternative formulations of bird trip behaviour within the time constraints of this project.

We have made crude assumptions regarding the spatial distribution of prey: assuming that it is either uniform, or else proportional to the density of birds that were found within an area using the GPS data (after accounting for the effect of distance to colony). These scenarios are likely to correspond to two extreme cases (bird distributions do not reflect prey distributions at all, or bird distributions perfectly reflect prey distributions), and reality is likely to lie somewhere between these two extremes.

There is considerable uncertainty regarding the precise behaviours that birds will adopt during avoidance or displacement. In terms of barrier effects we have assumed that birds will fly right up to the edge of the 1km buffer zone around the wind farm before they begin to modify their flight path; this may be overly-conservative, because birds may in reality learn to avoid the wind farm by following a shorter route ( e.g. flying directly from the colony to one edge of the wind farm footprint, and then flying directly from there to their destination). We have also assumed that birds do not habituate to the wind farm over the course of the breeding season, which is likely to be a conservative assumption.

The representativeness of the GPS tracking data is a key consideration when interpreting the results of the model. Confidence comes with larger sample sizes and consistent results across situations ( e.g. among years within SPAs, or across SPAs). The most restricted sample sizes were apparent for guillemots away from Forth Islands and for puffins. For guillemots, we made an expert judgement on the representativeness of these data largely from our knowledge and experience of the at-sea range of Isle of May individuals estimated across many years (reviewed in Daunt et al. 2011). Whilst it is not possible to test the validity of this approach, it is probably reasonable that space use recorded at one SPA provides an indication of likely space use at adjacent SPAs, because of expected correlations in environmental conditions across the region (Frederiksen et al. 2007; Cook et al. 2011). On this basis, we discarded Buchan Ness data because sample sizes were very low and birds foraged in a very restricted area. In contrast, we considered that the data from Fowlsheugh and St Abbs Head guillemots were more representative because sample sizes were higher and foraging range and distribution were more in keeping with data from the Isle of May. Ultimately, the decision to exclude Buchan Ness on the basis of the quality of the data is unlikely to be a factor in assessments because of the distance of this colony from proposed wind farms. There is also increasing evidence that species from this colony preferentially forage to the north ( RSPB unpublished data). For puffins, we considered the data from the 7 study individuals was plausible with respect to mean maximum foraging range and direction. However, there is a concern that shorter trips were under-represented (Harris et al. 2012). Thus, the true distribution may be concentrated closer inshore than we recorded, and overlap with the more distant proposed wind farm developments could be lower. However, without further data it is impossible to assess the extent of this potential under-representation. Further insights could be gained by combining tracking with at-sea survey data, although the latter would include non-breeding Puffins and those from other colonies, and much of it is comparatively old.

4.2.5 Reducing uncertainty: further work

This project has highlighted the need for more data regarding several crucial aspects of seabird behavioural response to wind farms, as well as more basic data on life history. In particular, data for displacement rates from wind farms by foraging birds, levels of barrier effects and width of buffer zones is required to better understand how individuals adjust their behaviour in response to wind farm development. Existing empirical data are primarily based on non-breeding birds that are not under the same spatial or physiological constraints. It is not known whether these estimates are relevant to breeding seabirds that have restricted foraging ranges and the requirement to repeatedly return to a central place; in other words, whether these behavioural responses are generic or state dependent. Furthermore, these data need to be collected over long timescales such that behavioural mechanisms such as habituation can be included in future modelling efforts.

One of the largest sources of uncertainty in this project has been the translation of adult body mass into subsequent survival over the remainder of the year. There is an urgent need for more local studies that attempt to determine the functional relationship between end of breeding season adult body mass and subsequent survival for these species. Furthermore, we have also not been able to include any effect of fledging mass of chicks on post-fledging survival; it is likely that chicks fledging at a lighter mass have lower over-winter survival prospects, but quantification of this relationship is currently lacking in the literature. In addition, this model did not consider other periods of the breeding cycle that could also be affected, including the probability of breeding and survival rates of eggs during incubation.

Finally, the addition of accurate data regarding prey distribution and density would greatly enhance the ability to better estimate impacts of wind farms. Moreover, collecting data on the prey response to wind farm development will also be crucial to better understand the impacts of wind farms on seabirds.

4.3 Conclusions

This analysis is the first, to our knowledge, to quantify consequences of displacement and barrier effects on seabird demographic rates. Displacement effects have been considered to potentially affect chick survival, but what has been less widely appreciated is that alterations to adult survival are also possible, mediated via changes in body condition. This model is readily adaptable to other locations, in particular in situations where GPS tracking data are available.

We have shown that there is considerable variation in the potential effects of SPA/species/wind farm combinations, with the greatest effects apparent with Forth Island kittiwakes, Fowlsheugh kittiwakes and Forth Island puffins. Within the scope of this project it has not been possible to conduct a full quantitative assessment of uncertainty; however, all of the qualitative indications are that the uncertainty in the magnitude of the wind farm effect is likely to be large. The outputs from this work should therefore be interpreted with considerable caution. Parameterisation with local data, in particular prey distribution, behaviour of seabirds in response to wind farms (including habituation) and influence of adult body mass change on subsequent survival, would be an important step for the future.