Scottish Marine and Freshwater Science Volume 5 Number 9: Strategic Surveys of Seabirds off the West Coast of Lewis to Determine Use of Seaspace in Areas of Potential Marine Renewable Energy Developments

3. Methods

Aerial survey desk study

3.1 A desk study was undertaken in September 2011 to identify previous seabird survey datasets available from the west coast of Lewis. Historical aerial survey data and reports from WWT Consulting and Joint Nature Conservation Committee ( JNCC) were reviewed.

Digital aerial survey strip transect methodology

Survey dates

3.2 Seven surveys between April 2012 and February 2013 were planned in the survey programme, focussing on key ecologically relevant seabird periods through the year:

• One in late April;
• One in May;
• One in June
• One in mid July/mid August;
• One late August/early September;
• One in January; and
• One in February

3.3 Actual dates of surveys accomplished are presented in the Results section below.

Aerial survey design

3.4 To achieve a robust sampling regime from which seabird population estimates could be calculated with relatively high confidence, a series of 32 transects were used running perpendicular to the main coastal depth gradient covering the study area ( Figure 1). The area covers the areas of wet renewables development identified in the Regional Locational Guidance and includes a 2km buffer area around these.

3.5 For an optimum combination of coverage and species identification the area was surveyed with a sensor array comprised of four 2cm Ground Sample Distance super high-definition video cameras each with 50m strip width on the ground, giving a total strip width of 200m, and overall 10% coverage of the study area.

Strip based analysis

3.6 The HiDef approach is based on the continuous capture of super high definition video which offers considerable flexibility for subsequent application of strip analysis. Running in excess of fourteen frames per second the camera array (comprised of four cameras for this project) produced a set of four continuous imagery strips for each transect. The strips were separated by 50m to ensure that there is no duplication of counts as a result of flying birds crossing into the view of one camera from another.

3.7 These video strips form a rich data source for analysis immediately following capture and may subsequently be revisited should areas or species prove to be of particular interest at a later date.

3.8 Surveys were planned to be conducted in conditions of sea states of calm to slight, cloud cover over 2,000 feet and good visibility, using a range of long and short range forecasts for the region and further offshore. All surveys were carried out adhering to WWT Consulting/HiDef's strict health and safety policy.

Figure 1 - Digital aerial survey line transect and vantage point ( VP) locations in relation to wave/tidal areas of interest. VP numbers refer to those used in Table 2 and 3

3.9 On completion of surveys, data were backed up and returned to HiDef for processing, storage and analysis. Reviewers at HiDef reviewed 100% of the video footage and identified objects that were potentially birds or marine mammals. The times and frame numbers of these were logged on a spreadsheet subsequently sent with copies of the video data to trained observers at WWT Consulting for identification. Twenty percent of footage reviewed was blind-reviewed a second time as part of HiDef's Quality Assurance process to ensure 90% or better match between reviewers in detecting objects. This Quality Assurance Process is outlined in Figure 2 (Appendix I).

3.10 On receiving the marked spreadsheets and video data, WWT Consulting ornithologists recorded the species (where possible), species group and behaviour of those objects identified by the reviewers. WWT Consulting's identification system allows a confidence category to be assigned to each record at the species and species group level, which can be 'definite' (99+% confident - diagnostic features clear), 'probable' (more than 50% confident - diagnostic features indicated but not clear, ) and 'possible' (less than 50% confident - no diagnostic features but GISS (general impression of size and shape ) is indicative). Where identification to species was not possible, for example due to a sitting gull or auk presenting insufficient features for an identification in the given light conditions, 'No ID' was recorded as the species but the lowest taxonomic level identification possible given as the species group. Occasionally, bird behaviour combined with ephemeral light and weather conditions at sea create images which are difficult to identify. In these cases broad species group categories are used for species which would ordinarily be easy to separate, such as 'auk/small gull'. Species groupings used in the analysis are detailed in Table 4 (Appendix II).

3.11 Records were then geo-referenced to produce locations of sightings.

3.12 For Quality Assurance purposes, 20% of all object records were selected at random and analysed by a second observer without reference to the original analysis. The data sheets were compared for species and species group identification and confidence levels and the original sheet accepted if there was 90% or better agreement, or a feedback system using a 3 ^rd observer instigated. This process is summarised in Figure 3 (Appendix I).

3.13 For the purposes of calculating abundance estimates for species all levels of confidence of identification were combined. There are arguments for and against this approach, with alternatives including to only use probable or definite identifications. However including all levels (except 'No ID') represents the best identifications available by experienced ornithologists and not including them is more likely to result in bias caused by under-estimates for those species. Identifications to different confidence levels are presented in Results below and the data spreadsheets supplied to the client have the confidence levels against each record, so they are available for further scrutiny and different analysis approaches if required at a later stage. Obviously records for 'No ID' will lead to some underestimation of species abundances and where these are frequent, it is more appropriate to analyse these records at the species group level, for example for 'auk species.'

3.14 Note that correction factors have not been applied for diving species to account for a proportion of birds that may have been underwater and thus unavailable for counting. Density estimates for these species referred to in the results should thus be interpreted as minimum uncorrected densities.

Ground-based point count methodology

3.15 To provide ground count data that complement and 'ground truth' the aerial survey data, surveys were undertaken from vantage points ( VP) located at approximately equal intervals along the NW coast of Lewis (Port Ness southwest to Carloway). The exact locations were chosen on the basis of the view they afforded of the sea, the height above sea level and the desire to include area with breeding seabirds.

3.16 For the Loch Roag survey area, vantage points were positioned around the coasts of Loch Roag as frequently as required to give good coverage of the whole of the survey area but mediated by access considerations. Apart from the outermost parts, Loch Roag was not covered by the aerial survey ( Figure 1).

3.17 Survey work was undertaken by an experienced ornithologist working from a series of carefully selected shore vantage points ( VP). VPs were selected on the basis of ease of access and the view they afforded. Binoculars (10 x magnification) and a zoom spotting scope (20-60 x magnification) were used to help find and identify species out to a range of 2km (in good conditions).

NW coast - Port Ness via Butt of Lewis southwest to Carloway

3.18 This is a long, very exposed stretch of coast. Access from roads is mostly good but intermittent, and there is mostly adequate elevation for VPs. At the north end, round the Butt of Lewis east to Port Ness, there are cliffs with moderate numbers of breeding seabirds.

3.19 Initially, fourteen VPs were used along this stretch though some of these were later discontinued due to either poor elevation or restricted view ( Figure 1). This stretch of coast includes the locations of the two VPs used to collect data for the baseline seabird and marine mammals studies for the proposed Lewis Wave Array (Aquamarine Power Ltd) and these VP locations were also used for this study.

3.20 The counting method involved undertaking a single 'snapshot scan' from each VP on each survey date. The aim was to measure the instantaneous distribution of birds, seals, cetaceans and basking sharks using the area of coverage visible from each VP ; typically (depending on the adjacent coastline geography) comprising a semicircle of sea up 2 km offshore. A snapshot scan took about 30 minutes to complete, depending on how many birds were present. Except for scarce species ( e.g. divers, Arctic skuas Stercorarius parasiticus), flying birds that were transiting through the area of coverage were not recorded. Searches were undertaken by systematically searching the arc of the search area from one side to other using a combination of telescope and binoculars, going sufficiently slowly so as to reduce the likelihood of overlooking actively diving birds or cetaceans because they are underwater (dives by birds typically last <1 minute).

3.21 For seabirds the species; age/size; group size; behaviour; activity when first seen; location; and travel direction were recorded. In the case of marine mammals and basking sharks the following was recorded: time; species; age/size; group size; activity when first seen; location and travel direction.

3.22 An animal's location was recorded in terms of a compass bearing (measured using compass binoculars) and an angle of declination from the VP. Trigonometry was later used to calculate the grid reference of locations from these field measurements and the height above sea level and grid reference of VPs.

3.23 The angle of declination was measured using a digital level attached to a spotting scope firmly mounted on a tripod fitted with a levelling head. The digital level measured angles to a precision of 0.05 degrees. The angle of declination of the horizon was also measured to provide a consistent reference.

3.24 VP watches were undertaken in sea state conditions of moderate or less and not in continuous heavy rain or when good visibility was less than 2km.

Loch Roag

3.25 Twenty seven VPs were used to cover Loch Roag. Figure 1 shows the location of the VPs, Table 3 (Appendix II) the count location names, grid references and coverage . There was some variation between survey visits in the VPs used; the choice of VPs depended on light and wind conditions at the time and the amount of day light time available to count.

3.26 As far as possible survey work was restricted to fair weather and relatively calm sea conditions (below sea state 3). Each visit was planned to be completed in a single day, though in practice the July visit was spread over several days due to unsettled weather and the two mid-winter visits were spread over two days due to daylight constraints.

3.27 The surveyor searched all areas of water and shoreline visible from the vantage point, typically spending about 15-25 minutes at each point; the actual time depended on the sea conditions, numbers of animals present and the extent of the visible area, and to some extent how much time was available (see Future Recommendations).

3.28 All target species (seabirds, marine mammals and basking sharks) seen were recorded in terms of species identity, age/size, plumage, behaviour and location. Time, weather and sea conditions were also recorded.

3.29 Inner Loch Roag extends over 103 individual 1km squares, though most of these contain some land. The whole area has complex coastlines with many small islands and skerries ( Figure 1). The 1km square as apparent from a 1:50,000 scale OS map (Landranger series Sheet 13) was recorded for all target species. This was considered to give adequate spatial resolution and proved to be a highly satisfactory and easy method for recording location.

3.30 Observers were vigilant to the possibility of double recording, i.e. the possibility that the same individual was recorded from adjacent vantage points. This problem was minimised by appropriate spacing of vantage points and observers only recorded individuals that they believed were different to those previously recorded on the same visit. Because of these measures the potential for double recording to inflate total counts is believed to be negligible.

3.31 The potential for under recording was much greater because on all visits some areas of Loch Roag were not visible from all vantage points or were too distant (>2km) from the vantage point, and so complete coverage was not achieved. Furthermore, it is possible that some actively diving species were overlooked even if they were within the visible areas searched. The problems of under recording were reduced by choosing vantage points that maximised coverage, surveying in low sea states and good light, and searching from each vantage point for as long as time permitted, (but bearing in mind the aim to complete each visit in a single day - (see Future Recommendations).

3.32 Data were entered into an Excel spread sheet and exported to ArcMap GIS to produce maps. Pivot table routines within Excel were used to tabulate results.

3.33 The inner Loch Roag survey area was arbitrarily divided into five survey sectors to facilitate the reporting of results ( Figure 4, Appendix I).

3.34 No attempt has been made in this report to account for under recording caused by either overlooking birds in the areas searched or because some parts of the survey area were out of view from all vantage points. Attempt was also not made to correct density estimates for decreasing detectability of birds further from the observer, due to the confounding and for this study (given small sample sizes) inseparable effects of detectability and the ecological variable distance from coast. However see below for comparison of aerial survey and ground count methods where the relative effect of these has been studied. Thus VP density estimates should be considered as minimum uncorrected densities particular for distances of over 1km from the observer.

Abundance estimates

3.35 Point abundance estimates in the study area for each species recorded from aerial surveys were calculated from design-based analysis of encounter rates within the strip transects following Buckland et al. (2001).

3.36 For divers and auks, abundances of each species were estimated using encounter rates of records as above, but also by applying the proportions of species identified in each aerial survey to the total estimates for those groups.

Density surface modelling

3.37 Density surface modelling was applied to the digital aerial survey data only.

3.38 The count model of Hedley and Buckland (2004) was used to model trends in spatial distribution of birds in the study region. In this approach, the realised survey tracklines were divided into small segments each 1km long and the response variable in the statistical model was the estimated number of birds in the segment. In contrast to line transect surveys, during the video surveys birds were recorded in strip transects, so there was no adjustment for detection probabilities.

3.39 The British National Grid (Ordnance Survey) easting ( x) and northing ( y) coordinates of the birds were fitted as a smoothed two-dimensional term in a Generalised Additive Model ( GAM) with a negative binomial error distribution. To reduce the likelihood of overfitting the maximum number of knots k was set to 10 and gamma 1.4 was used to penalise models with higher degrees of freedom (Wood 2006). Automatic smoothness selection through Restricted Maximum Likelihood Estimation ( REML) was used in the GAM function of the 'R' package 'dsm' (R Development Core Team 2008, Miller et al. 2013). For this project the use of mixed models ( GAMMs) to explore autocorrelation in the data was not undertaken.

3.40 The GAM was used to predict density of birds in a 1km x 1km 'prediction grid' populated with the explanatory variables x and y covering the study region and abundance obtained by integrating under this surface.

3.41 Thus the model used was: N ~ s(x, y, k = 10) + offset, Family = negative binomial, Theta = 0.166, gamma=1.4.

3.42 The variance of the abundance estimates derived through density surface modelling was calculated using the variance estimation via Bayesian results method of Wood (2006) as implemented in package 'dsm'.

Comparison of aerial survey and ground count methods

3.43 Locations of records from aerial and vantage point surveys were overlaid as layers in ArcGIS for comparison where spatial coverage overlapped and temporal coverage overlapped or was similar.

3.44 To compare densities for auks data were split into those Vantage Points (Port of Ness, Butt of Lewis E, Butt of Lewis W and Swainbost) and aerial survey transects (transects 1-4) around the Butt of Lewis and the remainder of the west coast vantage points and their respective aerial survey transects (transects 5-21). Aerial survey transects south of transect 21 (transects 22-32) and thereby more than 2km away from the area covered by Vantage Points were excluded from these analyses. Using ArcGIS arcs were drawn around each Vantage Point representing the area surveyed. These were intersected by 500m distance bands from each VP to derive areas and thence densities in each band. The resulting intersected polygons were subsequently intersected by 500m distance bands from the coast so that densities in these bands could also be calculated. Data beyond 2km and any birds on land or flying were removed from analyses. The aerial survey transects were intersected by 500m distance bands from the coast and densities derived for records of sitting birds of each species or species group within these. Densities were combined from each sub-set of samples (around Butt of Lewis or west coast of Lewis) and by season: April to July or September to February.