We focused on six species or species groups in this study, namely: the harbour porpoise (Phocoena phocoena), auks (Uria aalge and Alca torda), the black-legged kittiwake (Rissa tridactyla), cormorants (Phalacrocorax carbo and Phalacrocorax aristotelis), the northern fulmar (Fulmarus glacialis) and the northern gannet (Morus bassanus; Table 1 ). Only harbour porpoises and the seabirds listed above were present in numbers adequate for analysis. Harbour porpoises, black-legged kittiwakes, northern gannets and northern fulmars could be clearly identified from the aircraft and analyses could be conducted at the species level. Cormorants and auks needed to be considered as groups based on taxonomic and morphological criteria ( Table 1 ).

The study area covered about 9,400 km² and included the Belgian waters and parts of the French and English waters in the North Sea (from 50.75 to 52°N and from 1 to 3.5°E; Figure 1 ). Aerial visual surveys were conducted in 2017 on April 6-7, June 13-14, August 7-8, and December 4-5, and in 2018 on March 5-7 and May, 4-5. Transects were flown using high-wing aircrafts equipped with bubble windows at a target altitude of 182 m (600 feet) and a speed of 167 km.h⁻¹ (90 knots). During the various flight sessions, two offshore wind farms, operational since 2009 and 2010, were regularly flown over, the Thanet wind farm, off the east end of Kent (England) and the Thorntonbank wind farm between the Belgian and Dutch waters ( Figure 1 ).

We used a multi-target protocol (described in detail in Lambert et al., 2019) to simultaneously record harbour porpoise and seabird sightings. Sightings of seabirds were recorded within a 200 m strip on each side of the aircraft (strip transect protocol), while for marine mammals, the perpendicular distance to the transect was recorded for distance sampling analyses (line transect protocol; Buckland et al., 2001). Linear transects were designed using the equal spaced zig-zag option of the software ‘Distance’ version 6.2 (Buckland et al., 2001; Thomas et al., 2010). This design can produce an unequal coverage probability when study regions are non-rectangular (Strindberg and Buckland, 2004) and this is why we used density surface modelling. This model-based approach relaxes the equal sampling probability assumption of conventional distance sampling by incorporating spatially explicit covariates to obtain a fine-grained density surface in the study area (Miller et al., 2013). Sightings and observation conditions were recorded using the ‘SAMMOA 1.0.4’ software (SAMMOA 1.0.4.; http://www.observatoire-pelagis.cnrs.fr/publications/les-outils/article/logiciel-sammoa). Observation conditions (Beaufort sea state, turbidity, cloud cover, glare severity and subjective observation conditions) were recorded at the beginning of each transect and when any condition changed. The last parameter corresponded to a subjective assessment of the detection conditions by the observer by combining all the observation conditions from poor to medium, good and excellent conditions.

Datasets were prepared for analyses using the ‘Marine Geospatial Ecology Tools’ package for ‘ArcGIS 10.3’ (Roberts et al., 2010; ESRI, 2016). Effort data were linearised and segmented into 2.5 km segments of homogenous observation conditions. Only segments recorded with good observation conditions were kept for analysis (i.e. Beaufort sea-state between 0 and 3 and subjective observation conditions from medium to excellent).

Density surface models (DSM; Miller et al., 2013) allowed us to investigate the relationships between species and their environment and predict their distributions. To do this we included both static and dynamic environmental variables, considered as potential drivers of prey availability.

A total of 8,840 km was flown during the six flight sessions ( Table 3 ). Of the total effort, 89% was performed in good conditions, i.e. with a Beaufort sea-state strictly less than four and subjective observation conditions between medium and excellent.

Sightings recorded during the six flight sessions, detailed in Appendix E , included 1,120 sightings of harbour porpoises (1,412 individuals) and 4,344 seabird sightings (5,861 individuals). The seabird sightings included 1,218 sightings of auks, 1,385 of black-legged kittiwakes, 134 of cormorants, 186 of northern fulmars and 1,421 of northern gannets. These sightings included both individuals and groups.

For each species, the model with the lowest WAIC was selected (all models are tanked in Appendix F ). These models were used to predict species distributions. All species except for cormorants were predicted to be distributed offshore. The predicted distribution of cormorants was very coastal. Four seasonal patterns of abundance could be defined post-hoc: some species were abundant in spring (March to May) and almost absent in summer (June to August); others were abundant in winter (December to February), species whose abundance varies annually, and the rest showed little seasonal variations throughout the year.

The relationships between predicted densities and distance to operational OWFs varied between species groups ( Figure 2 ). Predicted densities of northern fulmars were higher at short distances to OWFs while densities of auks were insensitive to distance to OWFs. In contrast, the highest densities of kittiwakes, cormorants, porpoises and gannets were found at distances larger than 40 km from OWFs.

The estimated distribution of all species in the eastern and north-western part of the study area remained unchanged when predictions from the prospective scenario, wherein all planned OWFs were assumed to be operating, were taken into account ( Figure 3B ). For porpoises, kittiwakes, gannets and cormorants, densities would be lower, on average, in the centre of the study area, particularly near the planned OWFs. Densities of fulmars would however be higher in this central area. The predicted change in auk distribution was more contrasted: densities were predicted to be lower in the centre of the study area but higher in the northern and western parts.

The EMDs, which indicated the match between the predicted density distributions and the counterfactual density distributions, were relatively small (less than 0.3; Figure 4 ). Therefore, the counterfactual predictions were relatively close to the original predictions, but the values were not negligible, so the implementation of the OWFs would affect species distribution. This would be particularly noticeable in spring for harbour porpoise, black-legged kittiwake and northern gannet (higher EMDs), in summer for northern fulmar and throughout the year for cormorants. For auks, the EMDs were relatively stable, so the implementation of the OWFs would probably not have a major impact on the distribution of the species.

In contrast, if the planned OWFs were operational, the impact on species abundance would be higher. The abundance of fulmars would increase significantly (between +20% and +70%; Figure 3C ), all else being equal, while the abundance of all other species would decrease (down to -50% for gannets), although the decrease would be less for auks (maximum -20%). There are two months during the time period where abundances of porpoises and gannets slightly increase (+10%; Figure 3C ).

We are aware that we sampled only a few months of the year and we cannot represent fixed monthly or seasonal distribution patterns for the studied species over multiple years. However, our model, which included month-varying relationships and time-varying spatial effects (modelled as random walk; see Appendix C ) allowed to predict species distributions for each month, thereby illustrating intra-annual variation.

Recommendations for minimisation measures critically hinge on knowledge of seasonal abundance and spatial distribution patterns. Carrying out regular surveys every month is very costly, and impractical over large areas. Here, we capitalised on a series of six surveys carried out between 2017 and 2018 to collect data on marine megafauna. These data were then used to calibrate a model that took into account spatio-temporal variations explicitly. In order to study mobile species, monitoring over a broad area is required which in turn limit temporal replicates. With six surveys spread over a whole year, we highlighted intra-annual variations in the species distribution allowing to identify the period and areas where the impact would be the lowest while with an average annual distribution, we would not have been able to do it. This study therefore highlights how critical spatial and seasonal variations in abundance are to marine spatial planning.

The estimated relationships between animal distributions and the planned OWFs must be interpreted with caution: they are not causal, nor do they arise from direct observations of behavioural responses to OWFs. Instead they could reflect the effect of other variables (e.g. environmental) not considered in the models.

Our predictions are borne out of the best available evidence for the study area, and rely on a state-of-the-art modelling approach that takes into account seasonal and spatial variations. Regular monitoring of the study area in the future can ground-truth these predictions. Our prospective approach allows for the cumulative effect of the commissioning of multiple OWFs (either operating or planned) on marine megafauna distribution to be quantified, in contrast to the DEPONS model (Nabe-Nielsen et al., 2018), which assesses the cumulative impact of noise on harbour porpoise distribution during the construction phase of multiple OWFs. The approach and results we put forward in this study suggest additional impacts of OWFs than those determined using the DEPONS model. The DEPONS model considers the cumulative impact of noise from OWF projects built simultaneously or sequentially, while our approach targets various species, not only those impacted by noise, and assesses the long-term impacts on species distributions once planned OWFs are operational, and not only the impacts during the construction phase. These two approaches complement one another and are both crucial to mitigation efforts and marine spatial planning objectives with regards to the impacts of OWFs at the population scale since they individually concern the construction and operational phases.

Harbour porpoises showed an offshore distribution and were absent along the coast line, as previously highlighted by Lambert et al. (2017). They appeared to prefer cool waters, which may explain their decreased abundance in summer compared to winter and spring. It is likely that they migrate to the central and northern parts of the North Sea in summer (Benke et al., 2014; McClellan et al., 2014).

Most seabirds were predicted to be more abundant further offshore with monthly variations in their abundances with the exception of cormorants, which had a coastal distribution, which is consistent with Lambert et al. (2017) and Virgili et al. (2017). This is because non-breeding cormorants feed within a radius of 35 km around their colonies and resting sites (Grémillet, 1997). Auks and fulmars were mostly abundant in spring and winter and absent from the study area in summer. This is because auks start migrating to their wintering grounds between May and October (Heath et al., 2000; Fort et al., 2013) and fulmars between June and November (Cadiou et al., 2004; Lambert et al., 2017). Kittiwakes were abundant in winter, which is when they are concentrated around breeding colonies, and less abundant in summer when they are dispersed far from colonies (Marchant et al., 2002). The highest abundances of northern gannets were estimated at the beginning of spring, the prenuptial period during which many individuals cross the Dover Strait (Fort et al., 2012). Individuals observed between April and September were probably immature birds because mature individuals breed outside of the study area (Kubetzki et al., 2009; Fort et al., 2012; Wakefield et al., 2013).

In this study, we applied the approach proposed by Methratta (2020), consisting in recording species sightings at different distances from wind farms and to include the distance to the wind farms in the models. We thus estimated relationships between the sightings and the distance to the nearest OWF as well as with variables that characterise the environment. The results were consistent with the known ecology of the species (Grémillet, 1997; Heath et al., 2000; Marchant et al., 2002; Cadiou et al., 2004; Fort et al., 2012; 2013; Benke et al., 2014; McClellan et al., 2014; Lambert et al., 2017). The transect design carried out during the surveys does not strictly follow the BAG design because it was not a sampling at increasing distances from the wind farms but rather a sampling that allowed the area to be optimally gridded with an aircraft (limitation of transit time with optimal coverage). We have therefore adapted the approach proposed by Methratta (2020) to the aerial survey methodology. We highlighted the effect of the OWF present in the area and use the obtained relationships to predict the distribution that species would have if these relationships were maintained with the OWFs currently planned in the area.

The models predicted the highest densities of fulmars at short distances to operational OWFs, whereas for all other groups, except auks, the highest densities were at large distances. Predicted densities of auks were unaffected by the distance to OWFs. Furness et al. (2013) and Vanermen et al. (2015) showed that gannets are sensitive to the presence of OWFs and would avoid them. This is supported in our results by the predictions of lower gannet densities near the operating OWFs in the area. Likewise, Vallejo et al. (2017) showed a constant abundance of common guillemot (an auk species) regardless of the development phases of an OWF, which may explain why the implementation of the planned OWFs had little effect on the auk abundance. Garthe and Hüppop (2004) have argued that cormorants, kittiwakes and fulmars were among the least sensitive species to the presence of OWFs. The model predicted higher densities of fulmars near OWFs, so if the planned OWFs were operational, their density in the study area would increase, all else being equal. Harbour porpoises seem to avoid an OWF area during the construction phase but can become as abundant during the operational phase as before construction started (Madsen et al., 2006; Brandt et al., 2011; Vallejo et al., 2017). Here, densities of harbour porpoises were lower in the east, near the Belgian wind farm, than in the centre of the study area, this may be due to the construction of an extension around a Belgian OWF during the surveys.

The prediction of seabird and harbour porpoise distributions based on the hypothetical scenario in which all planned OWFs were fully operational highlighted changes in species distribution. These predictions can be compared to future monitoring data, putting the model’s predictive ability to a true test and ultimately facilitating the improvement of predictive models. Northern fulmar abundance was predicted to increase near the new OWFs while the abundance of the other groups decreased, suggesting avoidance of OWFs and thus displacement of individuals out of the study area. One hypothesis for the unexpected increase in fulmar abundance, proposed for example by Vanermen et al. (2015), would be that feeding opportunities would be enhanced near OWFs, but any attraction effects of OWFs are currently poorly understood. These predicted changes in abundance and distribution clearly suggest that new OWFs in the study area would impact marine megafauna when construction is finalised and they become operational.

The construction phase of an OWF poses the greatest threat to cetaceans because the noise generated by pile driving drives away the animals (Vallejo et al., 2017) and can induce hearing loss (Tougaard et al., 2004; Thomsen et al., 2006; Dähne et al., 2013). During the operational phase, the noise levels generated by the turbines and maintenance activities are lower than those during pile driving and are unlikely to affect cetaceans (Madsen et al., 2006). The operational phase is the most hazardous for seabirds because of fatal collisions with wind turbines, particularly for gannets (Furness et al., 2013) but also because turbines may create physical barriers along migration routes and displace animals from foraging habitats (Drewitt and Langston, 2006; Furness et al., 2013). Displacement occurs when seabirds avoid OWFs and can lead to increased energy costs and reduced food supplies which might affect individual fitness (Masden et al., 2010; Peschko et al., 2020).

In order to meet the mitigation objectives of avoidance and minimisation (Kiesecker et al., 2010), based on our results, we recommend that wind farms be established in areas of low seabird densities to minimise the risk of collisions and the loss of important foraging habitats during the operational phase, and that the construction phase be planned during periods of low porpoise densities. We would mainly recommend that the various OWFs planned for development be constructed between mid-summer and early fall to minimise the impact on harbour porpoises. The construction phase of a wind farm takes between 1.5 and 2 years, depending, in part, on the number of wind turbines constructed (Carstensen et al., 2006). Some stages of the construction phase are noisier than others and last between 3 and 5 months (Carstensen et al., 2006; Brandt et al., 2011). Therefore, the time window proposed in the study would be compatible with the most noise-generating construction stages of the wind farms. In addition, weather conditions in the area are generally most favourable between mid-summer and fall, which could facilitate the construction phase and limit its duration.

Another recommendation could be to avoid building OWFs in offshore areas in the centre of the study area to minimise the impact on seabirds, in particular auks, kittiwakes and gannets. In the light of these recommendations, the two wind farms planned off the Belgian coast (between 2.5 and 2.8° E and 51.25 and 51.7° N; Figure 1 ) could be harmful to seabirds because the southern North Sea is a migration corridor for numerous seabirds (Stienen et al., 2007). In order to apply this recommendation, further research would probably be required, particularly into the distribution of the species’ prey and whether the construction of the OWFs would result in the displacement of this prey, in which case the loss of habitat would probably be limited. In addition, the main impact of the OWFs would be collisions with seabirds. A better understanding of seabird behaviour in the presence of OWFs, particularly through telemetry, would help to fill in the gaps.

To minimise this impact, we also recommend that the number of OWFs constructed in the coming years be limited, and that the new OWFs be spaced more widely apart, avoiding high concentrations of farms in a small area. The construction of OWFs in various areas should also be staggered temporally so that the disruption occurs only in a limited area at a time, allowing animals, especially cetaceans, areas of refuge in which they can continue foraging, breeding, etc. undisturbed. It is essential that the location of OWFs avoids bird migration corridors (Larsen and Madsen, 2000; Garthe and Hüppop, 2004), and the layout of the turbines must avoid the barrier effect if the wind farm is located on a migration corridor. Turbines could be controlled to limit their operation during periods of high bird migration (Fijn et al., 2015). Finally, the definition, at national or even European level, of zones that are favourable or unfavourable for the installation of wind turbines would allow better positioning of wind farms while limiting the impact on biodiversity.

This study predicts the potential impact of wind farms on marine megafauna species, but in a wider context, marine species are subject to anthropogenic pressures beyond those resulting from OWFs, and in the North Sea those pressures are many. The southern North Sea is crossed by a shipping lane frequently used by more than 400 commercial ships (1/4 of the world traffic), many ferries as well as fishing and recreational vessels, which certainly have an impact on marine megafauna (Halpern et al., 2008). For example, Wisniewska and colleagues (Wisniewska et al., 2018) showed that porpoises make deeper dives and interrupt their foraging and echolocation activities due to the noise of boats. Fishing also affects seabird and cetacean populations through the alteration of food supplies and bycatch incidents in fishing gear (Tasker et al., 2000; Read et al., 2006). The cumulative impacts of boat noise, fishing and wind farm activities could have negative long-term consequences on marine megafauna. These impacts are important to consider in the context of marine spatial planning in order to establish and maintain an organised maritime area that takes the interactions between users (both professional and private) into account and balances the need for maritime development and activities with the protection of the environment (Douvere, 2008; European Parliament and Council Directive 2014/89/EU). The better we understand the impacts of anthropogenic activities – individually and cumulatively, on a specific and broad scale – the more effectively we can reconcile development and environmental conservation and make valuable suggestions for marine spatial planning.