Discussion
Wind regimes are changing, in terms of the mean strength, and the frequency of extreme weather events35,36. Yet research on how wind affects seabirds has focused on their at-sea behaviour (though see37,34). Through our novel application of CFD, we demonstrate that airflows are also critical in the selection of breeding habitat. Areas of coastline vary predictably in their exposure to prevailing wind conditions, and we show this is an important driver of habitat preference in cliff-nesting auks. Interestingly, while colony location was predicted by low exposure to the prevailing wind, it was not predicted by low wind speed. This apparent contradiction is explained by the fact that windward cliffs can block the oncoming flow, with the blocking effect increasing with cliff height and slope, producing low wind speeds and high pressures over large parts of windward cliffs (bar the top, where flow is accelerated, Fig 3a,b). Areas of low wind speed therefore occur on both windward and leeward cliffs (Supplementary Fig. 3), but guillemots select the latter. This suggests habitat selection is driven as much by the need to shelter young from the impact of rain or wave action (both of which should increase with exposure), as it is to shelter from high wind speeds, which can affect wind chill38 and flight capacity (either of the guillemots or their aerial predators31,38). Nonetheless, flight capacity may be more critical for species such as large albatrosses, which require relatively high winds to take-off and therefore may be constrained to nest in exposed areas, despite the intuitive benefits of shelter for chicks across species.
Our models confirm the role of slope angle in colony selection, with the densest and largest colonies on Skomer being associated with the steepest cliffs. Steep slopes offer the possibility of breeding in high densities with better protection from predators39, as well as easier access to the sea when chicks jump from their nests40,41. Yet here we show that steep cliffs with a south-westerly orientation are avoided on Skomer, even though they are widely available. This trend was not significant in a previous assessment of whether colonies varied in aspect31, confirming our prediction that cliff aspect alone is not a good proxy for exposure. Furthermore, the fact that slope angle had a lower contribution in our models than pressure and turbulence, suggests that colonies are better tuned to wind rather than topographical features.
While guillemots preferentially breed in areas that are not exposed to the prevailing wind, they cannot shelter from all wind directions. Winds diametrically opposed to the prevailing direction (here NE winds) will be problematic for any species breeding in sheltered sites. The penalty of exposure to NE winds for the 10 largest colonies on Skomer, was a ~10% increase in mean wind speed compared to the same at-sea wind speeds from the SW. How this might impact birds will depend on the factors driving the need for shelter and the magnitude of the wind when it comes from a different direction. Nonetheless, our results highlight that colonies experience increased exposure from changes in wind direction, independent of rising wind speeds. Increases in wind speed, as already observed in the North Atlantic and other areas35,36, are also likely to be most detrimental to birds at the nest when accompanied by a change in wind direction37.
A further challenge potentially facing birds on Skomer in NE and NW winds is increased turbulence. The absolute levels of turbulence that birds experience in SW winds are low because the wind speeds themselves are low. However, in NE winds of the same magnitude, birds experience both stronger winds and increased turbulence. Wind speed has been shown to reduce the probability of guillemots landing successfully at their breeding cliffs31 and turbulence is likely to present further difficulties for flight control in stronger winds42,43.
The fact that our models performed better in correctly predicting the densest colonies, compared to the presence of any breeding birds, suggests that they work best in predicting high quality habitat. Previous studies have shown that breeding success increases with the density of breeding pairs7,44. Appropriate areas that can support larger numbers are therefore of higher quality. Such areas have previously been described in terms of the number of walls, slope and width of the ledge where the egg was incubated, and distance from the top of the cliff6,7. The ability to predict high quality breeding habitat without such fine-scale topographical information is advantageous, as it allows habitat quality to be predicted in remote and inaccessible sites.
Models of absence should be interpreted with more caution than models of presence, as cliffs that are unoccupied now may have been occupied in the past. Indeed, photographs of the breeding cliffs on Skomer from the 1930s provide evidence that numbers were much higher historically45, and whole island counts undertaken since 1963 demonstrate that numbers have been increasing since then46. The relative abundance of common guillemots makes this less of an issue than for many species where current breeding activity occurs in a small fraction of the former range. In cases where populations are increasing, our approach could be extended to see whether airflow characteristics can predict colony growth rates, or which areas most likely to be expanded into.
Overall, the fact that 90% of the densest colonies on Skomer could be predicted solely from variation in pressure values i.e. without the need for slope angle, is testament to the predictive power of our approach. CFD is particularly well-suited to modelling habitat selection in seabirds, as marine and coastal environments experience some of the most extreme wind conditions47, and wind fields also tend to be reasonably laminar ahead of islands. A key future challenge will be to test this approach over larger areas. Combining airflow modelling with data on rainfall and breeding success will also provide new mechanistic insight into the basis for habitat selection and how global change may impact birds at their nesting sites.
Methods
Our approach centres around the estimation of airflow parameters around Skomer Island (51° 44.271’N, 5° 17.668’W) and the use of these parameters, in combination with slope angle, from a highly resolved LiDAR digital elevation model, to predict the distribution of breeding guillemots on Skomer and then on the neighbouring island of Skokholm. The 2015 Skomer guillemot breeding bird survey48 was digitized in ArcMap 10.5.1 (ESRI, Redlands, California) and used to delineate sections on the island’s cliffs that were occupied by breeding birds. It was also used to identify the 10 largest colonies (count ≥ 592 individuals) and 11 densest colonies (density ≥ 0.835, birds per sq. m), with thresholds being selected by visually identifying clear breakpoints.
A “digital elevation model” (DEM) (50 cm resolution retrieved from Lle Geo-Portal http://lle.gov.wales) was used to identify cliff habitat by selecting slopes ≥ 20o (initial trials showed this value performed well in isolating cliff habitat). The resultant area was divided into sections according to those used in the breeding survey. These same sections (71 in total) were used in all further analyses, 38 of which were occupied by breeding birds (Fig. 2 a). The minimum height of each section was taken as 10 m to account for variation in tide height (maximum tide height ~ 5 m on the day the DEM was produced), the maximum wave height (taken to be 3 m), as well as a minimum distance above water that birds tend to nest, taken to be 2 metres7. The maximum height of each section was reduced using a minimum distance of 15 m from the top of the cliff. This distance was the mean proximity of nests from the top of the cliffs for three major colonies, based on highly resolved theodolite measurements49.
A similar approach was taken to digitize the distribution of breeding guillemots on Skokholm from the 2018 breeding bird survey50 (Supplementary Fig. 1a). However, because the elevation of Skokholm’s cliffs is much lower, the minimum distance from the top of the cliffs, was set at 7 m (this value was arrived at in consultation with the wardens). The small proportion of occupied cliffs that did not satisfy this threshold were not mapped. In the cases where estimates of bird numbers were given in relation to a single point on the map, we used a minimum section length of 30 m of coastline, unless ascribing this width to adjacent colonies would have resulted in unoccupied sections of < 30 m, in which case we assigned a section of 30-50 m in length. This approach resulted in 35 colonised areas from a total of 91 (Supplementary Fig. 1b).