1. Introduction

  1. This Annex to the Marine Mammal Technical Report provides an analysis of spatial and temporal distribution of marine mammals observed during the Digital Aerial Survey (DAS) campaign for the Ossian Array (hereafter referred to as the ‘site boundary’). The DAS campaign commenced in March 2021, with a total of 24 months of data collected up to and including February 2023.
  2. The extent of the DAS area provides an indication of marine mammal activity over the site boundary and an 8 km buffer, which constitutes the Array marine mammal study area ( Figure 2.1   Open ▸ ), and therefore will be useful to determine where Zones of Influence (ZoIs) for some impacts associated with the Array extend further than the site boundary (although may not cover the full extent of the ZoI for all impacts e.g. piling noise).
  3. Marine mammal data collected during these DAS complement the historic site-specific survey data that were collected for various offshore wind projects between December 2009 and April 2021 as well as other published data sources for the region. A detailed overview of these data sources is provided volume 3, appendix 10.2.

2. Methodology

2.1. Study Area

  1. The study area for the DAS campaign was delineated as the site boundary plus an 8 km buffer. This whole area, including the buffer corresponds with the ‘Array marine mammal study area’ and will inform the baseline for those impacts which may potentially extend beyond the boundaries of the site boundary. The aerial survey area covers a total area of approximately 2,264 km2 ( Figure 2.1   Open ▸ ).

2.2. Survey Approach

  1. Aerial surveys of seabirds and marine mammals commenced in March 2021 and continued monthly up to and including February 2023 to allow 24 months of data collection, including any additional surveys to account for delayed survey flights (refer to section 2.5.2).
  2. The surveys were conducted by HiDef Aerial Surveying Limited (hereafter ‘HiDef’) from an aircraft flying at an operational speed of 220 km/h (equivalent to 120 kn) at a survey height of approximately 550 m Above Sea Level (ASL). The aircraft was equipped with four HiDef ‘GEN 2.5’ cameras with a set resolution of 2 cm ground sample distance. Each camera surveyed a strip width of 125 m and cameras were set such that a gap of approximately 20 m between the strips was maintained, thereby ensuring there would be no overlap between the strips. For four cameras there was therefore a combined survey width of 500 m.
  3. A total of 31 transects were spaced 2.5 km apart across the Array marine mammal study area, aligned in a broadly north-east to south-west orientation, perpendicular to the depth contours along the coast. The transects followed the routes shown in Figure 2.1   Open ▸ . Position data for the aircraft was recorded at 1 s intervals from a Garmin GPSMap 296 differential Global Positioning System (GPS) device with 2 m positional accuracy, and allowed recording updates to match to seabird and marine mammal observations.

Figure 2.1:
Geographical Location of Array Marine Mammal Study Area and DAS Flight Tracks

Figure 2.1: Geographical Location of Array Marine Mammal Study Area and DAS Flight Tracks

 

  1. The total transect length covered by the DAS was 23,150.13 km, with a monthly mean of 964.59 km. Data from two cameras (approximately 0.25 km combined width, although camera coverage at the end of transects was reduced by clipping of data to the boundary of the Array marine mammal study area) were subsampled to provide a monthly mean sampled area of 226.66 km2, which exceeded the minimum target of 10% coverage of the total aerial survey area (which was 226.41 km2).
  2. Imaging and GPS equipment continued to collect data between transects (i.e. as the aircraft turned at the end of one transect to begin the next), as to avoid including additional survey effort outside of the aerial survey area, all DAS data were clipped to the aerial survey area before analysis began.

2.3. Processing of Aerial Data

  1. Digital aerial imagery, collected via the GEN 2.5 cameras, was reviewed by a team of trained and experienced professionals within HiDef, using high resolution viewing screens. Objects were marked, and their location recorded, before being passed to the second stage of species identification. Here, experienced marine surveyors used high definition digital imagery to identify each marked object to species level where possible. Other features including fixed structures, fishing vessels, dredgers, construction vessels, ferries, yachts and recreational vessels were also recorded.
  2. An object was only recorded where it reached a reference line (known as ‘the red line’) which defines the true transect width for each camera. By excluding objects that do not cross the red line, biases in abundance estimates caused by flux (movement of objects in the video footage relative to the aircraft, such as ‘wing wobble’) could be eliminated.
  3. For marine mammals, image analysts assigned the following classifications to each image:
  • 'surfacing at red line': the dorsal fin (cetaceans) or head (pinnipeds) was above the water surface in the middle frame of the video sequence;
  • 'surfacing': part of the animal appeared above the water surface in any of the frames, but not the dorsal fin or head in the middle frame of the sequence;
  • 'submerged': no part of the animal appeared above the surface in any of the frames; or
  • ‘unknown’: it was not fully clear from the footage whether an animal was surfacing or just submerged.
  1. A qualitative measurement of the confidence in the identification was also provided as follows:
  • 'definite': as certain as is reasonably possible;
  • 'probable': very likely to be this species or species group; or
  • 'possible': more likely to be this species or species group than anything else.
  1. An additional ‘blind’ review was undertaken on a subset (20%) of the data as part of HiDef’s Quality Assurance (QA) process. The reviewed data were compared to the original and if there was less than 90% agreement then all the data were re-reviewed.
  2. All data were geo-referenced and compiled into a single output, taking into account the offset from the transect line of the cameras, which gave a higher degree of positional accuracy to each geo-referenced object. Geographical Information System (GIS) files for the ‘Observation’ and ‘Track’ data were provided by HiDef in ArcGIS shapefile format, using UTM30N projection, WGS84 datum.
  3. On receipt of the geo-referenced aerial survey data, an additional QA on the data was carried out by RPS. Track lines for each camera reel were plotted in GIS and the total effort was subsequently calculated for each transect flown and compared with the minimum target of 10.0% coverage of the aerial survey area. Where the minimum coverage was not met, further detail was sought from HiDef to understand why this was the case. In addition, the marine mammal sightings data were reviewed, and any anomalies were highlighted and discussed with HiDef to validate the data. Further detail is provided in section 3.1.1 ( Table 3.1   Open ▸ ).

2.4. Data Analyses

  1. Summary statistics were produced to describe the data for each of the key species or species groups within the DAS dataset. Data were presented to show the survey effort achieved in each month of survey against the minimum target of 10.0% coverage of the aerial survey area, and a description of any remedial action taken to address data gaps from delayed surveys was given.
  2. Raw count data for each of the species or species groups was presented for each month of survey to highlight the frequency of sightings in each identification category. These raw count data were also spatially mapped in GIS to illustrate the distribution of sightings across the aerial survey area.
  3. Further summary data were also produced to describe the number of sightings that fell into the different surfacing classifications (paragraph 12) and the different confidence classifications (paragraph 13).
  4. Sightings data were corrected for effort in each month of the survey to produce counts per unit effort (i.e. number of individuals per km of track line flown) and are referred to as ‘encounter rate’. These effort-corrected data allowed comparisons across months where effort varied; for example, for months which included weather downtime.

2.4.1. Density Estimates with Bootstrapping

  1. For those species where there were sightings in a sufficient number of surveys to allow for temporal trends in observations to be estimated, seasonal relative densities were calculated from the DAS count data. Although there is no definitive minimum threshold, common dolphin Delphinus delphis was identified in one survey, harbour seal Phoca vitulina was identified in two surveys and minke whale Balaenoptera acutorostrata was identified in four surveys, so it was not possible to ascertain temporal trends for these species across the 24-month DAS campaign. For harbour porpoise Phocoena phocoena, white-beaked dolphin Lagenorhynchus albirostris and grey seal Halichoerus grypus, it was possible for temporal trends to be estimated.
  2. Research into temporal patterns of harbour porpoise density identified two broad divisions in distribution, termed by Heinänen and Skov (2015) as ‘summer’ (April to September) and ‘winter’ (October to March). Similarly for grey seal, broad-scale seasonal patterns of density have been determined based upon potential changes in distribution between the breeding season (defined as September to December for this region (Marine Scotland, 2020; Special Committee on Seals (SCOS), 2020)) and the non-breeding season (January to August). This is because most females would be expected to be hauled out with pups during the breeding season, rather than being at sea.
  3. Pooling data further into two bio-seasons allows the robustness of analyses to be improved where sample sizes in seasonal or monthly divisions may be small, while retaining greater resolution than pooling data by year.
  4. To provide estimates of relative density and associated variance, the data were analysed using a non-parametric bootstrap approach, with replacement (Buckland et al., 2001). Bootstrapping is a commonly applied method to produce an approximate distribution of the empirical data, particularly where the sample size is insufficient for straightforward statistical inference. The resampling generates a probability distribution which is subsequently used to produce estimates of accuracy (e.g. standard errors, confidence intervals (CI)). Non-parametric bootstrapping makes no assumptions about the data, in contrast to parametric bootstrapping which assumes that data follow a specific distribution.
  5. Density estimates with bootstrapping were undertaken for harbour porpoise, white-beaked dolphin and grey seal. Monthly densities were resampled with replacement (1,000 times) to generate an estimated value for overall uncorrected density and 95% CIs for the aerial survey area.

2.4.2. Model Based Density Estimates

  1. Data were imported into R v4.2.0 (R Core Team, 2022) and the MRSea package (Scott-Hayward et al., 2013a) was used in the analysis to best predict the density of marine mammals within the Array marine mammal study area. To account for the missing data appropriately, a Spatially Adaptive Local Smoothing Algorithm (SALSA); (Walker et al., 2010) was used within MRSea (Scott-Hayward et al., 2013a; 2013b). This approach allowed adjustment for the presence of missing data by (a) exploiting empirical relationships between abundance and other variables (water depth, terrain ruggedness and distance to coast) and (b) exploiting commonalities between distributions in different months.
  2. Before any analyses could take place, the data required pre-processing to ensure no transect start or end times/locations differed (start and end times/locations were within both 10 s and 600 m of each other). In two cases across the 24 month DAS campaign this deviation occurred and the corresponding data were removed from further analysis.
  3. In total, 793 transects were used in the analysis, covering a total aerial survey area of 5,439.85 km2 and a mean monthly coverage of 226.66 km2 ( Figure 2.2   Open ▸ ). Note that surveys for May 2021 and February 2022 were flown in June 2021 and March 2022, respectively (refer to section 2.5.2 and Table 3.1   Open ▸ ).The spatial coverage of the monthly surveys used in the analyses, also indicating removed transects, is shown in Figure 2.3   Open ▸ . Note that survey effort was greater in the February 2022 survey, and that all variation has been accounted for in subsequent analyses.

Figure 2.2:
Survey Effort Across 24 Month DAS Campaign within Array Marine Mammal Study Area

Figure 2.2: Survey Effort Across 24 Month DAS Campaign within Array Marine Mammal Study Area

Figure 2.3:
Monthly Survey Coverage (Blue Lines) Across 24 Month DAS Campaign within Array Marine Mammal Study Area (Orange)

Figure 2.3: Monthly Survey Coverage (Blue Lines) Across 24 Month DAS Campaign within Array Marine Mammal Study Area (Orange)

 

  1. It was originally intended that months would be modelled separately, however this approach was not possible due to monthly data being too sparse to fit MRSea models. Data were instead pooled across months within seasons (winter: December, January and February; spring: March, April and May; summer: June, July, August; and autumn: September, October and November) to overcome this issue, incorporating the biological assumption that species behave similarly within each season.
  2. To improve the predictive power of MRSea analyses, data were therefore also pooled into bio-seasons where relevant.
  3. The following covariates were used within modelling to predict species distribution:
  • water depth (m);
  • terrain ruggedness index (TRI);
  • distance to coast (km);
  • X and Y coordinates;
  • season; and
  • bio-season (where species-appropriate).
  1. The degree of smoothing for each season/bio-season was determined within the MRSea package using tenfold cross validation, and the best model was used to predict species distribution. Within each of the models, separate maps for mean and associated upper and lower 95% Confidence Limits (CLs) were also produced for each season/bio-season.
  2. For the purposes of MRSea modelling, the transects were split into 1 km sections (with a final section of less than 1 km on each transect, to ensure no data were omitted). The number of records for each species (across cameras) was then summed within each of these sections. To perform this aggregation, each record was mapped on to the nearest point of the transect line (i.e. the straight line between the transect start and transect end locations). Records did not always lie directly on this line and the distribution of distances between records, and the nearest point on the transect line is shown in Figure 2.4   Open ▸ .

Figure 2.4:
Frequency Distribution of Distances from Marine Mammal Records to the Nearest Point on the Straight Line Between Camera-Specific Transect Start- and End-Point

Figure 2.4: Frequency Distribution of Distances from Marine Mammal Records to the Nearest Point on the Straight Line Between Camera-Specific Transect Start- and End-Point


  1. After removing the two transects as described in paragraph 27, a total of 825 records of harbour porpoise were used to predict densities within the aerial survey area.
  2. Mean seasonal abundance estimates were calculated using the summed density estimates within square kilometre grid cells and scaled back up to estimate abundance across the aerial survey area.

2.4.3. Correction Factors

  1. Noting that the density estimates are relative and do not account for availability bias during the aerial surveys (refer to section 2.5.3) a literature review was undertaken to determine appropriate correction factors for the key species. Further detail on the correction factors is provided in section 3.5 on a species-by-species basis.