NatureScot Research Report 1280 - Inshore Wintering Waterfowl in Moray Firth Special Protection Area - 2019/20 digital aerial surveys and comparative analyses of aerial and shore-based surveys

The Moray Firth SPA supports non-breeding populations of European importance of 11 species: great northern diver (Gavia immer), red-throated diver (Gavia stellata), Slavonian grebe (Podiceps auritus), greater scaup (Aythya marila), common eider (Somateria mollissima), long-tailed duck (Clangula hyemalis), common scoter (Melanitta nigra), velvet scoter (Melanitta fusca), common goldeneye (Bucephala clangula), red-breasted merganser (Mergus serrator), and European shag (Phalacrocorax aristotelis). The citation population estimates for these qualifying features and underpinning data sets are summarised in Table 1.

Table 1 Moray Firth SPA citation population estimates and summary of underpinning data sets.  The surveys from which the citation population estimates were derived are indicated as: RSPB: shore-based seaduck surveys; VAS: visual aerial survey (JNCC); and BBS: boat-based survey (JNCC, derived from ESAS).  Further details are in Lawson et al (2015).

Species	SPA citation population (individuals)	Survey type	Basis of estimate
Greater scaup	930	RSPB	Mean of Peak 2001/02 – 2005/06
Common eider	1,733	VAS	Mean of Peak 2001/02; 2003/04 – 2006/07
Long-tailed duck	5,001	RSPB	Mean of Peak  2001/02 – 2005/06
Common scoter	5,479	RSPB	Mean of Peak  2001/02 – 2005/06
Velvet scoter	1,488	RSPB	Mean of Peak  2001/02 – 2005/06
Common goldeneye	907	RSPB	Mean of Peak 2001/02 – 2005/06
Red-breasted merganser	151	RSPB	Mean of Peak  2001/02 – 2005/06
Red-throated diver	324	VAS	Mean of Peak 2001/02; 2003/04 – 2006/07
Great northern diver	144	VAS	Mean of Peak 2001/02; 2003/04 – 2006/07
European shag	6,462	BBS	hotspot analyses  of 1980 – 2006 ESAS data
Slavonian grebe	43 (SBS)	RSPB	Mean of Peak 2001/02 – 2005/06

 

The presence of high densities of wintering waterfowl and European shags in the Moray Firth is indicative of the importance and productivity of these waters during the winter months.  The Moray SPA was identified as an ideal site in which to investigate approaches to monitoring such inshore wintering species, due to the size of the site, assemblage of qualifying features and abundance of existing data. These data spanned multiple years and was gathered using a range of methodologies, providing an opportunity for comparative analyses in an important marine site.

Aims and objectives

NatureScot contracted HiDef Aerial Surveying Limited (hereafter HiDef) to undertake two sample digital aerial surveys and subsequent analyses of the populations and distributions of the wintering waterfowl (divers, grebes, seaduck) and European shag qualifying features of the Moray Firth Special Protection Area (SPA) in winter 2019/20. These surveys were coincident with the Citizen Science Pilot shore-based counts

HiDef also undertook additional explorative comparative analyses of contemporary and historic surveys, including the 2019/20 aerial and Citizen Science Pilot counts, Wetland Bird Survey (WeBS) and historic RSPB vantage point counts, of the Moray Firth for three selected abundant species: red-throated diver, common eider and common scoter.

The overall aims of the work commissioned by NatureScot from HiDef and reported on here were to:

1. Design and undertake two digital aerial surveys of wintering waterfowl and European shag across the Moray Firth SPA in winter 2019/20 to coincide with Citizen Science Pilot surveys;
2. Derive robust population estimates and distribution maps for all 11 qualifying features of the Moray Firth SPA from the digital aerial surveys; and,
3. Undertake exploratory comparative analyses at a variety of spatial scales of data derived from contemporary and historic surveys for three of these qualifying features to examine population change and inform future options for monitoring marine populations of these and similar species.

This report is linked to a separate report (Graham and Thompson, 2022) that: details the NatureScot Citizen Science Pilot surveys of the Moray Firth SPA; extends the comparative analyses developed here by HiDef to all 11 wintering qualifying features of the Moray Firth SPA; and, makes recommendations for future monitoring of inshore wintering waterfowl in MPAs (see Figure 1).  

Digital aerial survey flight dates and weather conditions

The survey duration was over six hours, so it was split between two aircraft, each flying for roughly three hours to ensure surveys could be completed in one day. Inverness Airport was used for fuelling and as a general survey base. Flights were targeted to the times of high tide in the area, where possible, or at least to times when mud flats would not be fully exposed. As well as weather windows, there were constraints imposed by the Military of Defence, which meant that flights had to take place at the weekend in order to achieve full coverage of the SPA. However, this fitted well with the availability of Citizen Science Pilot field workers for the concurrent vantage point watches. Weather considerations also limited surveys to days with no precipitation, no low cloud (less than 600m) and wind speeds no higher than Beaufort Force 6.

The first survey was completed successfully on 19th January between 09:45 and 13:00. Poor weather conditions over successive weekends prevented the second survey being attempted during February, as was originally intended, and the survey was flown on Sunday 8th March 2020 between 10:20 and 13:10. In both January and March, NatureScot Citizen Science Pilot surveys were undertaken on the same dates.

The same flight pattern was flown on each flight, following the pre-planned 42 transect pattern spaced at 3000 m intervals (Figures 4 and 5). Minor differences in aircraft flight route and GPS trimming during post flight data review resulted in slightly different surveys areas being sampled in the January and March surveys (Table 2). This is normal in aerial survey as aircraft are unable to follow exact tracks, and are influenced by altitude, weather (see Table 3), and air traffic control factors.

Table 2 Survey effort, by area, across Moray Firth for January and March 2020

Survey date	Number of transects analysed	Total length of transects analysed (km)	Area covered (km²)	% of SPA covered
19 January 2020	42	593.76	296.88	16.84%
08 March 2020	42	597.76	298.88	16.96%

 

The weather conditions encountered during the surveys are described in Table 3.

Table 3 Recorded weather during surveys

Date	Cloud base (feet)	Wind speed (knots)	Wind Direction	Precipitation
19th Jan 2020	2500 +	15 - 30	235° - 270°	None
8th March 2020	2500 +	15 - 20	210°	None

 

Digital aerial survey flight methodology

HiDef used its GEN II survey rig mounted on a fixed winged aircraft. This rig has been deployed since first developed in 2012 and is described in Webb and Nehls (2019). The GEN II camera rig is designed specifically for high quality seabird and marine mammal surveys. The rig contains four extreme high-resolution digital video cameras. At ~550 m (1760 feet) altitude, the cameras and lenses each survey a strip of ca.125 m resulting in a total potential strip width of 500 m with a ground sample distance (GSD) resolution of 2 cm (Figure 6). A gap of ca. 20 m is maintained between the cameras to ensure no overlap between strips, and therefore eliminate the possibility of double counting. The Moray Firth surveys were flown at a ground speed of 220kph (c. 120 knots). This approach has been found to create the best imagery suitable for data collection without negatively impacting on birds because of disturbance, while also flying at a safe and legal height. The cameras are designed so that fast shutter speeds, in excess of 1/10,000th second, are possible in low light, unlike off-the-shelf photogrammetry cameras.

During the survey, while the aircraft turned between transects, the camera rig was rotated to ensure that it is always pointed either forwards or backwards to an angle of 30° from vertical and away from the sun. This eliminates perception bias caused by sun glare on the sea. Digital video imagery from each camera was recorded continuously to two solid state hard drives (one master and one backup). The position of the aircraft at one second intervals was also recorded from a differential GPS device (with two metre positional accuracy). At the end of the survey the two hard drives were separately couriered to HiDef for post‐processing and image analysis.

Digital aerial survey data review and object identification

The raw video data were converted into a format for further analysis on data review stations. The use of digital video technology facilitates the review process, because objects of interest generally persist in shape and size through multiple frames compared to more variable wave shapes. This reduces the risk of missing objects at this stage, particularly for surveys at higher wind speeds.

Surveys using the GEN II system usually provide an average identification rate for birds at sea in excess of 95%. For difficult to differentiate species, such as Atlantic puffin Fratercula arctica, razorbill Alca torda, common guillemot Uria aalge, diver species (Gaviidae), small gulls and even tern species (Sternidae) it is usually possible to achieve 90-95% rates of identification to species level throughout the year.

The survey footage was viewed by trained HiDef reviewers using high resolution viewing screens and an image management software package. Reviewers recorded discreet objects that required further analysis, such as birds or marine mammals, captured by the footage, marking the location on the survey image. A sample of a minimum of 20% of material was subjected to a ‘blind’ re‐review; if the agreement was less than 90% then a further review of the material, and retraining, was initiated as required.

Due to each object being recorded in multiple video frames it is essential to ensure that the object is only recorded once, and not multiple times. As part of the video review process, an object is only recorded where it reaches a reference line (known as ‘the red line’) which defines the true transect width of 125m for each camera. By excluding objects that do not cross the red line, biases to abundance estimates caused by flux (movement of objects in the video footage relative to the aircraft, such as flight and ’wing wobble’) are eliminated.

Those images where objects had been marked, were passed to experienced marine ornithologists, the majority of whom have worked with HiDef for a number of years, for further analysis. Images were managed using software to enhance their appearance and assist in identifying objects. The ornithologists identified objects to species level where possible, using qualitative confidence levels of ‘Possible’, ‘Probable’ or ‘Definite’, defined as:

• Definite: as certain as reasonably possible;
• Probable: very likely to be this species or species group;
• Possible: more likely to be this species or species group than anything else.

These levels of certainty are based on key identification features (e.g. colour, tone, size, shape and overall ‘jizz’) that an object image shows and how clearly and robustly these features show. “Jizz” is a term used to capture the overall identifying physical and behavioural characteristics of a species e.g. 'A combination of characteristics which identify a living creature in the field, but which may not be distinguished individually' (Campbell and Lack, 1985). Any birds that could not be identified to species level were assigned to a species group (e.g. duck species, diver species, grebe species, cormorant /shag) and a category of ‘No ID’ was recorded in the species column. Any other information available from the imagery was recorded, such as behaviour, flight or direction of travel.

A randomly selected sample of at least 20% of material was identified independently by a separate group of expert ornithologists. The outputs of these results were then compared with the first identification and any discrepancies reviewed by a further set of expert ornithologists for adjudication. In the case of any significant discrepancies (i.e. more than 10% disagreement for the whole audit), the images were re‐reviewed by a third ornithologist who acts as a final adjudicator in the process to decide on the correct observations. While tools were utilised to assist in object detection identification, no part of the identification processes are automated.

Digital aerial survey object data extracted

The parameters recorded for each bird image were:

• Date and time of observation;
• Location (latitude and longitude) (and accuracy);
• Reel number (proxy for transect number);
• Species group (e.g. tern species, large auk species);
• Confidence in species group identification (possible, probable and definite);
• Species (and if not possible, No ID);
• Confidence in species identification (possible, probable and definite);
• British Trust for Ornithology (BTO) Code;
• Behaviour and direction of travel (e.g. flying NW, sitting, loafing (on land) and if possible feeding behaviour, fish-carrying behaviour);
• Age class;
• Sex;
• Additional information, including feeding behaviour where visible and association with man‐ made objects/vessels and;
• Additionally, information on the environmental conditions (sea state, sun glare, water turbidity and visibility) was recorded for every minute of video footage.

The presence of anthropogenic features (such as fixed structures, dredgers, construction vessels, ferries, yachts, or recreational vessels, etc.) which might influence birds’ behaviour was also recorded.

All data were geo-referenced, taking into account the offset of the camera view from the transect line of the cameras, and compiled into a single output; Geographical Information System (‘GIS’) files for the Observation and Track data were issued in ArcGIS shapefile format, using UTM30N projection, WGS84 datum.

Digital aerial survey abundance estimates and distribution mapping

Data were processed for estimating abundance and distribution of the target species and species groups using methods described in Buckland et al. (2001). Density (birds/km²) and population estimates and associated 95% confidence intervals, were derived by a blocked bootstrapping technique (Efron & Tibshirani, 1994) to ensure equal transect effort was sampled across each iteration. This was done by using transect ID as the sampling unit with replacement, and then randomly sampling until the total length of the sampled transects equalled approximately the total length of transects surveyed. The mean density and population estimate, the standard deviation of those means and their relative standard error, as defined by the standard deviation divided by the mean, were calculated. Data were processed in the R programming language (version 3.4.3).

The transect data were used to generate smoothed density maps using a deterministic Watson-Nadaraya type kernel density estimation (KDE) technique (Simonoff, 1996). In KDE, a small ‘window’ function (the kernel) is used to calculate a local density at each point in the study area. To evaluate the density at a given point, the kernel is centred on that point and all the observations within the window are summed to obtain a local count. The total area of the transect(s) intersecting the window is then summed to obtain a local measure of effort. By dividing the local count by the local effort, a local density estimate is obtained. To build a density map, the study area was covered with a fine mesh of study points and the density was calculated at each point in the mesh in turn.

Several of the maps generated may effectively be ‘flat’ with little or no variation in estimated densities. These correspond to distributions where the KDE was not effective at predicting missing data. This can happen with evenly distributed birds or where birds are very sparsely distributed. In the case of sparse distributions, a ‘flat’ map does not necessarily reflect the true underlying distribution. Rather, it may indicate that the data doesn’t contain enough evidence to determine what the underlying distribution is. It is therefore useful to refer back to the population estimates for the relevant species when interpreting such ‘flat’ density maps. The characteristics of KDE density mapping are further described below under Modelling digital aerial survey data.

Exploratory comparative analyses – introduction

The focus of this element of the project was on exploratory comparative analyses of data from contemporary and historic surveys of wintering waterfowl in the Moray Firth, specifically:

• 2019/20 digital aerial surveys (as detailed in this report)
• 2019/20 Citizen Science Pilot shore-based surveys (as described in Graham and Thompson, 2022)
• Contemporary and historic WeBS data
• Historic RSPB vantage point surveys (see also Table 1) 

These data sets differed in survey methods, spatial coverage and time scales (see Figure 7) The focus of the analyses was on examining population changes and investigating the potential of the different types of survey data to provide robust and consistent indices of changes in bird populations within and across the Moray Firth SPA. NatureScot also sought recommendations with respect to survey type and sampling design to underpin future monitoring of populations of wintering divers, grebes, seaduck and shags in MPAs in Scotland.

Analyses were initially developed for three of the qualifying features of the Moray Firth SPA.  These three species were selected on the basis of their differing habits and distributions.  Red-throated divers forage on fish in the water column and are widely dispersed across the Moray Firth SPA (see Figures 12 and 13).  Common eiders feed predominantly on benthic bivalves, such as blue mussels, and are moderately dispersed, predominantly in nearshore waters (see Figures 17 and 18).  Common scoters occur in a few large aggregations (see Figures 21 and 22) and exhibit simultaneous diving behaviour to forage on bivalves in soft sediments.

The raw data for the visual aerial surveys underpinning site selection for common eider and divers (Table 1) were not available and so only qualitative comparisons of these with the 2019/20 digital aerial surveys within the timeframe available were possible (see relevant species accounts). 

The datasets used were variously spatially and temporally defined and aggregated and were stored in a range of formats.  In particular, RSPB vantage point data were only available as scanned files. Consequently, considerable initial data preparation and manipulation was required, as detailed below, to enable subsequent statistical analyses and interpretation.  This necessarily constrained the time available for the main analyses within the short contracted time period available. In particular, it was not feasible to impute count densities across all data sets, as is reflected in adoption of estimates of count rates for some comparisons. The consequent limitations of these preliminary exploratory analyses are reflected in the final recommendations for further work. 

Exploratory comparative analyses – data processing

Modelling digital aerial survey data

Because digital aerial surveys did not sample 100% of the study area, the data needed to be extrapolated between transects to compare bird densities with the shore-based surveys within comparable areas. While KDE also extrapolates data, it is a deterministic method which does not use information about the sampling errors in the data and thus does not provide any estimation of errors in the outputs. The KDE approach assumes equal survey effort across the entire study area, which is the reason circular rings are generated around observations (see e.g. Figure 19). In KDE, the mapped densities are simply a function of distance to points and the magnitude of the values of the observations.

Alternative stochastic spatially-explicit modelling exercises are done by associating covariates to observations, and then predicting (extrapolating) into areas where we have information on the covariates, but not on observations. This approach was applied to the digital aerial survey data to support comparative analyses using a point pattern process in a Bayesian framework. The Bayesian point process model offers an advantage over the standard KDE approaches as it takes into account the unequal survey effort in and between transects; weighting those areas differently due to the uncertainly in the predictions in those areas.

A point pattern records the occurrence of events in a study region where the locations of these observations depend on an underlying spatial process. This spatial process can be characterised using the Cox process, which is a Poisson process with intensity λ(s) that varies in space. This intensity function measures the average number of events per unit of space, and it can be modelled to depend on covariates and other effects (Diggle, 2014; Baddeley et al., 2015).

Under the log-Cox point process model assumption, log intensity of the Cox process is modelled with a Gaussian linear predictor. In this case, the log-Cox process is known as a log-Gaussian Cox process (LGCP, Møller et al., 1998), and inference can be made using the Integrated Nested Laplace Approximation approach (INLA; Illian et al. 2012), which was developed as a computationally efficient alternative to Markov Chain Monte Carlo (MCMC) methods (Robert & Casella, 2010, Brooks et al., 2011). The log-Gaussian part of the LGCP name comes from modelling log (λ(s)) as a latent Gaussian (conditional on a set of hyper-parameters), in the typical GLM/GAM framework.

To fit these models in INLA we used the stochastic partial differential equation (SPDE) approach (Simpson et al., 2016). The SPDE approach consists of representing a continuous spatial process (e.g. a latent stationary Gaussian Field) with the Matèrn covariance function as a discretely indexed spatial random process (e.g. a (GMRF); Rue & Held, 2005). This approach is computationally efficient while accounting for spatio-temporal interdependence and autocorrelation in the data and it considers a direct approximation of the log-Cox point process model likelihood, where observations are modelled considering their exact locations instead of binning them into cells. Along with the flexibility for defining a mesh, this approach can handle non-rectangular areas.

The digital aerial surveys were modelled using only the spatial parameters (i.e., x and y coordinates of observations) as covariates under this approach, which characterised the distribution and abundance of birds based on the location of individual observations. Environmental covariates were not used due to time constraints on data acquisition and tuning of hyperparameters that would be required to fit the model.

Model predictions were output to a 1 x 1 km grid resolution with each grid of the output raster containing a predicted density of birds.

Processing RSPB vantage point data

The Royal Society for the Protection of Birds (RSPB) began monitoring inshore waterfowl populations across the wider Moray Firth in 1978, with a focus on seaduck, as part of work for environmental monitoring for the Beatrice oil field development. Counts of divers, grebes and seaducks by the RSPB were provided for the months of November, December, January, and March from 1998 to 2006. The Firth was divided into 12 sub-regions, each of which was counted in a single day. Wherever possible, adjacent sub-regions were counted on the same, or consecutive days. These surveys used standard point count techniques from coastal vantage points; however, no information was available on the spatial effort, and total survey effort had to be inferred.

It was not possible to extract the exact locations of the RSPB vantage points within the time available for this work, so the data were analysed within the coarse sub-regions reported in Butterfield (2001). It was also not feasible to extract the necessary data for all sub-regions due to inconsistencies in the data sheets (e.g. slight differences with spellings of species and vantage point names), which would have required considerable manual manipulation to address. Accordingly, three sub-regions were chosen to focus on: The Outer Dornoch Firth, Nairn and Culbin Bars, and the Inverness and Beauly Firth (Figures 8 and 9). These three sub-regions were selected for the comparative analyses because both the outer Dornoch Firth and the Nairn and Culbin Bars regions are well known for higher numbers of observations and were well covered by RSPB and WeBS counts. Although the Inverness and Beauly Firth sub-region does not typically have as many birds as the other two sub-regions, it is an area that was well covered by both RSPB and WeBS counts for the 1998 – 2006 period.

The mean count rate (birds/hr) was computed by taking the mean number of birds counted in all of the vantage points within each of the Outer Dornoch Firth, Inverness and Beauly Firth, and Nairn and Culbin Bars sub-regions, and then dividing by the approximate mean time per count (about 0.4 hours).

Comparative analyses – data comparisons

The most recent version of the annotated R code used to carry out these analyses is available on Github.

Relationships were characterised between: the HiDef digital aerial survey models and the Citizen Science Pilot counts; the Citizen Science Pilot counts and the contemporary WeBS counts; and, the historic WeBS counts and the RSPB counts.  Where possible, RSPB counts were then projected forward to 2020 to enable broad comparisons among all these data sources in order to draw some inferences for each species regarding population indices and survey techniques.

Comparing digital aerial survey models to Citizen Science Pilot counts

The NatureScot Citizen Science Pilot surveys were designed to target the qualifying features of the Moray Firth SPA from shore-based vantage points along the coastline. The survey was designed using a look-see methodology, at vantage points with a viewshed out to sea (Bibby et al., 2000 and Gilbert et al., 1998). Volunteer observers were provided with a protocol to undertake the survey and a recording form on which to detail each survey. Birds were identified to species level, where possible. However, if the species was unknown, observers were asked to indicate to lowest taxonomic rank possible, for example, ‘diver sp.’ or ‘scoter sp.’. Further details of the Citizen Science Pilot surveys can be found in Graham and Thompson (2022)

To compare the Citizen Science Pilot counts to the HiDef digital aerial survey data, the mean density in the 2km vantage point buffers (see Figure 8) was compared to the mean modelled density of birds from the digital aerial surveys within the same areas. These comparisons were made using linear regression and correlation. The data were also examined on a log-log scale in case there were issues of magnitude differences in counts or estimates. For each of the three species in the two survey months (January and March 2020), the linear relationship (either natural scale or log-log scale) with the highest Spearman’s correlation coefficient was selected and presented in the Results. Spearman’s correlation coefficient was used as a metric as it is more appropriate for small sample sizes.

Comparing Citizen Science Pilot counts to contemporary WeBS counts

A spatial merge was performed on the Citizen Science Pilot vantage point counts of 19^th January 2020 and WeBS counts of 12^th January 2020 using the R programming language. The WeBS data for March 2020 were not available at the time when these analyses were undertaken. All Citizen Science Pilot vantage points that fell within the boundaries of a WeBS sector were combined and a mean density was calculated. The mean density of the WeBS counts for each sector were also calculated using the area of the WeBS sector polygons (Figure 9). These densities were compared using a linear comparison and a Spearman’s correlation coefficient was calculated.

Comparing historic RSPB counts to WeBS counts

WeBS sectors which fell into each of the selected RSPB sub-regions (Outer Dornoch Firth, Nairn and Culbin Bars, and the Inverness and Beauly Firth) were merged. The mean count rate of birds (birds/hour) was calculated for the RSPB sub-regions for the months of January and December for each year, as counts were made in these months in all years.

The count rate was used here to include a measure of effort because raw count data cannot be used to reliably compare between methodologies (e.g., if in one method an observer counts 100 birds for 2 hours, but in another method the observer counts 10 birds over 15 minutes, the difference is due to temporal effort, and not indicative of the actual population). It was not possible to include a spatial component for this comparison (i.e. to calculate densities) because spatial coverage was not described for the RSPB data and these data were only available as scanned files. Manual entry of these into a new database would have been required for additional manipulation within a GIS and this was not feasible within the available timescale.

The time series of rates of counts imputed from the WeBS data and the RSPB data were then compared using a standard linear model to obtain the Spearman’s correlation coefficient.

Comparing contemporary data with reference to RSPB data

The concept here is to derive an estimate of what the RSPB data might be expected to be if they were performed in 2020, in order to make assessments of predicted changes.  This was restricted to situations where the linear fit from past data is high and was based on examination of Spearman’s correlations between the historic RSPB and WeBS counts for the three sub-regions.  Judgements were then made as to whether the WeBS data could act as a proxy for RSPB data.  

There is no objective way to determine what constitutes a “good” Spearman’s correlation coefficient because this can depend on the context of the question being asked. For the purposes of this work, if the Spearman’s correlation was > 0.6 for a month or sub-region combination, the decision was made to predict the RSPB data for the 2019/2020 season, with 95% confidence limits. The prediction was done by way of a standard linear model, which used the WeBS data as the independent variable and RSPB data as the dependent variable.

There was inadequate time to derive a more sophisticated (Bayesian) model due to the large amount of manual data manipulation required to format the datasets for analysis.

Citizen Science Pilot surveys, WeBS data, and models from digital aerial surveys were summarised within the three broad RSPB sub-regions. Mean modelled densities (from digital aerial surveys) were taken using the Citizen Science Pilot vantage point polygons as spatial boundaries and were averaged within the respective RSPB sub-regions.

Although the observation rates for predicted RSPB counts, WeBS counts, and Citizen Science Pilot counts cannot be directly compared to the mean modelled densities from digital aerial surveys (because it is not possible to calculate observation rates from a modelled distribution), the rankings of the values from highest to lowest (sorted against sub-region) could give some indication of which survey method most closely matched the modelled densities.

RESULTS

The results of the digital aerial surveys for the 11 target species are summarised, followed by more detailed narratives by species, including KDE distribution maps.  The original processed data for the surveys is in Annex 1. Information on other species within the survey area can be found in Annex 2 (numbers encountered), Annex 3 (population estimates) and Annex 4 (distribution maps).

The results of the exploratory comparative analyses of abundance estimates from the digital aerial surveys, Citizen Science Pilot surveys, WeBS, and RSPB counts are presented for red-throated diver, common eider and common scoter. Comparisons are also made here between the modelled population estimates for these three species and those derived by blocked bootstrapping.

As reported in Graham and Thompson (2022), the analyses developed here were subsequently applied, with training and technical support  from HiDef, to the remaining eight target qualifying features

Digital aerial survey: overall results

Bird species identification rates

In total, 92.4% of 7,022 birds detected in the January survey were identified to species level and 94.6% of 7,829 in March. These rates are comparable with HiDef’s overall identification rate average of 95% for digital aerial surveys.

Birds which are marked as unidentified (“no ID”) can still be classed as belonging to a similar group or type. The majority of these were auk species, which are notoriously tricky to split between razorbill and common guillemot in winter. These were not target species for this survey and consequently, as detailed in Table 4, the identification rates for the core groups were higher, averaging 97.7%.

Table 4  Survey identification rates in the Moray Firth in January and March 2020

Survey date	Overall ID rate (%)	ID rate for target groups (divers, grebes, seaducks and cormorants/shags)
19 January 2020	92.4	96.7
08 March 2020	94.6	99.0
Average	93.5	97.7

 

Digital aerial survey: count totals and population estimates

The total numbers of each of the target species and unidentified species groups detected in the sample transects of the two digital aerial surveys are presented in Table 5. The bootstrapped mean density, bootstrapped mean population estimate for the SPA, the 95% confidence limits of the mean, the standard deviation and the coefficient of variation (CV) were calculated from the sample counts for the target species in each survey (Table 6 and Table 7).

Table 5 Number of objects detected in samples during each survey assigned to species or group level for Moray Firth SPA qualifying features

Species	19th January 2020	8th March 2020
Greater scaup Aythya marila	1	0
Common eider Somateria mollissima	642	550
Long-tailed duck Clangula hyemalis	280	728
Common scoter Melanitta nigra	1694	433
Velvet scoter Melanitta fusca	13	2
Common goldeneye Bucephala clangula	9	18
Red-breasted merganser Mergus serrator	8	61
Unidentified duck spp.	72	14
Red-throated diver Gavia stellata	43	148
Great northern diver Gavia immer	31	126
Unidentified diver spp.	23	9
European shag Phalacrocorax aristotelis	0	296
Unidentified cormorant/shag spp.	0	0
Slavonian grebe Podiceps auritus	5	4
Unidentified grebe spp.	0	1
Total (unidentified group)	2821 (95)	2390 (24)

 

Red-throated diver

Density estimates for red-throated diver were calculated at 0.15 birds/km² for Survey 1, equating to a population estimate of 259 birds (± 95% CI 139–386). Birds were distributed from Spey Bay east (similar to great northern diver) with pockets also in the inner Moray Firth and off Loch Fleet in Sutherland (Figure 12). Survey 2 generated a higher estimated density of 0.49 birds/km², equating to 880 birds (± 95% CI 552 – 1,315) across the SPA. The distribution was similar to January but numbers moved closer inshore overall with higher densities around the Black Isle coast (Figure 13).

Alternative population estimates, generated using a modelling approach are presented in Table 11.

Site selection estimates: The peak numbers of red-throated divers estimated from winter visual aerial surveys in the larger Moray Firth Area of Search between 2001/02 and 2006/07 ranged from 159 to 565.The five-season mean of peak estimate was 366, of which 324 birds were within the SPA boundary (Lawson et al., 2015; SNH, 2016). As in 2019/20, these surveys also found red-throated divers to be more widely distributed across the site than great northern divers, with particular concentrations in the Outer Dornoch Firth, the Inverness Firth, around Spey Bay and between the Cromarty Firth and Tarbert Ness along the coast of Easter Ross (SNH, 2016).

Slavonian grebe

The digital aerial surveys detected too few observations of Slavonian grebe to generate meaningful density surface maps, but on the first survey low numbers were recorded around Brora/Golspie and in Spey Bay (Figure 14). A similar distribution was recorded on the second survey, with the addition of a record from the inner Moray Firth (Figure 15).

Overall density estimates for Slavonian grebe were calculated at 0.02 birds/km² for Survey 1, equating to a population estimate of 31 birds (± 95% CI 6 –66). Survey 2 generated a similar estimated density of 0.01 birds/km², equating to 24 birds (± 95% CI 6 – 48) across the SPA.

Site selection estimates: The peak numbers of Slavonian grebe in the Moray Firth SPA estimated from land-based counts between 2001/02 and 2005/06 ranged from 26 to 73 with a mean of 43 (Lawson et al., 2015; SNH, 2016). No Slavonian grebe were deterted in the visual aerial surveys carried out in the early to mid 2000s (Lawson et al., 2015). 

Greater scaup

Density estimates for greater scaup were calculated at ~0.00 birds/km² for Survey 1, equating to a population estimate of 6 birds (± 95% CI 0 –18). No birds were recorded on Survey 2. The distribution was concentrated to one site in the inner Moray Firth, close to Arturlie Point (Figure 16).

Site selection estimates: The peak numbers of greater scaup in the Moray Firth SPA estimated from land-based counts between 2001/02 and 2005/06 ranged from 223 to 2,641 with a mean of 930 (Lawson et al., 2015; SNH, 2016). The largest flocks were seen in the inner Firth and off Culbin Bar.  No scaup were detected in visual aerial surveys in the same period.

Common eider

Density estimates for common eider were calculated at 2.30 birds/km² for Survey 1, equating to a population estimate of 4,091 birds (± 95% CI 612 –12,182). Concentrations were mainly around the Nairn and Culbin Bars area. The KDE analysis method was unable to determine a distinct distribution pattern (Figure 17), despite numbers of common eider being similar to that in Survey 2; this was likely because the birds were insufficiently aggregated for a pattern to be determined.

Survey 2 generated similar estimated densities to Survey 1, of 1.86 birds/km² equating to 3,316 birds (± 95% CI 1,105 – 6,148) across the SPA. Concentrations were observed around the Outer Dornoch Firth and Loch Fleet, while densities remained high around the Nairn coast (Figure 18).

Alternative population estimates, generated using a modelling approach, are presented in Table 11.

Site selection estimates: The peak numbers of common eider previously estimated from visual aerial surveys in the larger Moray Firth Area of Search between 2001/02 and 2006/07 ranged from 1,287 to 2,513, with a five-season mean of peak estimate of 1,745, of which 1,733 birds were estimated to be within the SPA boundary (Lawson et al., 2015; SNH, 2016). These surveys mainly observed common eider close inshore along the south coast of the Moray Firth from Buckie to Nairn.

Long-tailed duck

Density estimates for long-tailed duck were calculated at 0.94 birds/km² for Survey 1, equating to a population estimate of 1,671 birds (± 95% CI 876 –2,658). The Moray coast showed concentrations from Nairn, through Culbin Sands and into Spey Bay. Significantly lower numbers were recorded along the Sutherland coast (Figure 19).

Survey 2 generated a higher estimated density of 2.43 birds/km² equating to 4,328 birds (± 95% CI 1,985 – 7,412) across the SPA. The distribution pattern was very similar to the January survey (Figure 20).

Site selection estimates: The peak numbers of long-tailed duck previously estimated from visual aerial surveys in the larger Moray Firth Area of Search between 2001/02 and 2006/07 ranged from 1,095 to 6,155, with a five-season mean of peak estimate of 3,167, the vast majority within the SPA boundary (Lawson et al. 2015).  However, the citation population estimate for the Moray Firth SPA was derived from land-based counts between 2001/02 and 2005/06. Peak counts ranged from 3,214 to 8,630 with a mean of 5,001 (Lawson et al., 2015; SNH, 2016). As found in the digital aerial surveys, long-tailed duck were abundant and widely dispersed throughout with notable concentrations along the Moray coast and off the Dornoch Firth (SNH, 2016).

Common scoter

Density estimates for common scoter were calculated at 5.80 birds/km² for Survey 1, equating to a population estimate of 10,336 birds (± 95% CI 200 –25,284). Distinct hot spots were recorded off Nairn and Culbin Bar and at the mouth of the Dornoch Firth around Embo (Figure 21). Survey 2 generated a lower estimated density of 1.46 birds/km² equating to 2,607 birds (± 95% CI 511 – 5,726) across the SPA. Distribution patterns were very similar to the January survey (Figure 22).

Alternative population estimates, generated using a modelling approach are presented in Table 11.

Site selection estimates: The peak numbers of common scoter previously estimated from visual aerial surveys in the larger Moray Firth Area of Search between 2001/02 and 2006/07 ranged from 1,264 to 3,845, with a five-season mean of peak estimate of 2,544 (Lawson et al., 2015). However, the citation population estimate for the Moray Firth SPA was derived from shore-based counts between 2001/02 and 2005/06. Peak counts ranged from 3,381 to 8,333 with a mean of 5,479 (Lawson et al., 2015; SNH, 2016). As found in the 2019/20 digital aerial surveys, there were notable concentrations around the mouth of the inner Dornoch Firth between Burghead and Nairn and in Spey Bay (SNH, 2016).

Velvet scoter

The digital aerial surveys detected too few observations of velvet scoter to generate meaningful density surface maps, but overall density estimates were calculated at 0.04 birds/km² for Survey 1, equating to a population estimate of 79 birds (± 95% CI 0 –234). Birds were noted mainly off Loch Fleet (Figure 23).

Survey 2 generated a lower estimated density of 0.01 birds/km² equating to 12 birds (± 95% CI 0 – 36) across the SPA. Distribution was again restricted, this time to the Nairn coast (Figure 24).

Site selection estimates: The peak numbers of velvet scoter previously estimated from visual aerial surveys in the larger Moray Firth Area of Search between 2001/02 and 2006/07 ranged from 0 to 747, with a five-season mean of peak estimate of 249 (Lawson et al., 2015). However, the citation population estimate for the Moray Firth SPA was derived from shore-based counts between 2001/02 and 2005/06. Peak counts ranged from 523 to 2,574 with a mean of 1,488 (Lawson et al., 2015; SNH, 2016). These surveys found no discernible difference in distributions between common and velvet scoters.

Common goldeneye

Density estimates for goldeneye were calculated at 0.03 birds/km² for Survey 1, equating to a population estimate of 55 birds (± 95% CI 12 –107). There were too few observations to generate a meaningful density surface map, but low numbers of birds were detected across coastal waters (Figure 25).

Survey 2 generated a higher estimated density of 0.06 birds/km² equating to 109 birds (± 95% CI 6 – 285) across the SPA. Concentrations were noted in the Inner Moray Firth, close to Kessock, Inverness (Figure 26).

Site selection estimates: The peak numbers of goldeneye in the Moray Firth SPA estimated from shore-based counts between 2001/02 and 2005/06 ranged from 222 to 1,547 with a mean of 907 (Lawson et al., 2015; SNH, 2016) Substantially lower numbers (five-season mean of peak 74 birds) were detected in visual aerial surveys across the Moray Firth Area of Search in the same period (Lawson et al., 2015). Goldeneye were mainly distributed either in the most southerly and shallow parts of the Inverness Firth, in the Dornoch Firth or occasionally on the southern shore in the Culbin/ Findhorn area.

Red-breasted merganser

Density estimates for red-breasted merganser were calculated at 0.03 birds/km² for Survey 1, equating to a population estimate of 49 birds (± 95% CI 6 –108). Sightings were restricted to coastal waters, notably in the Outer Dornoch Firth, (Figure 27) but the dispersion pattern of the birds and the low number of sightings resulted in a KDE map with no spatial variation.

Survey 2 generated a significantly higher estimated density of 0.20 birds/km² equating to 362 birds (± 95% CI 134 – 655) across the SPA. Distribution was round the Inner Moray Firth and Outer Dornoch Firth (Figure 28).

Site selection estimates: The peak numbers of red-breasted merganser in the Moray Firth SPA estimated from shore-based counts between 2001/02 and 2005/06 ranged from 108 to 209 with a mean of 151 (Lawson et al., 2015; SNH, 2016) Similar numbers (five-season mean of peak 80 birds) were detected in visual aerial surveys across the Moray Firth Area of Search in the same period (Lawson et al., 2015). Red-breasted merganser were found at a number of localities along the Moray coast and in the inner firth, with a larger aggregation in the Beauly Firth.

European shag

This species was not recorded in Survey 1. Density estimates were calculated at 0.99 birds/km² for Survey 2, equating to 1,762 birds (± 95% CI 76 –4,607). Small numbers were recorded on the survey, with the main concentration around Helmsdale on the Sutherland coast (Figure 29).

Site selection estimates: Analyses of data from predominantly boat based European Seabirds at Sea (ESAS) surveys collected between 1980 and 2006 identified a non-breeding season European shag population of at least 6,462 shags in the Moray Firth SPA. These birds were largely concentrated in two persistent large aggregations.  The larger was off Helmsdale, as also detected in 2019/20, and regularly held at least 3,179 birds. A second aggregation on the outer Moray coast to the west of Portsoy held an average of 1,967 birds.

Comparative analyses: red-throated diver

Bayesian Point Process Modelling

Overall population estimates derived by Bayesian Point Process Modelling for red-throated diver in the Moray Firth SPA in January and March 2020 are presented and compared with those generated by blocked bootstrapping (Table 11).

Figures 30 and 31 show the modelled densities and vantage point counts respectively for red-throated diver in January and March 2020.  Visual comparison indicates regions of relatively high abundance as detected by these two count methods.   

Red-throated diver were much more broadly distributed than common eider or common scoter but were in fewer numbers and further offshore. The predicted modelling output reflects this, with some higher densities being predicted offshore in the eastern part of the SPA. A few hotspots were predicted in the southwest at the mouth of the Cromarty Firth, and the mouth of Loch Fleet to the north in January. In March, the distribution changes with higher densities being predicted close to shore, particularly in the region of the Outer Dornoch Firth (Figure 30).

In January, Citizen Science Pilot counts were highest for red-throated diver in the Outer Dornoch Firth at the mouth of Loch Fleet, and at the mouth of the Cromarty Firth. However, counts of red-throated divers at eastern vantage points were generally low, which might be due to birds in this region being mostly offshore and out of visual range of the vantage points. Interestingly, in March, Citizen Science Pilot counts for the areas around Loch Fleet are low, despite predicted abundance being greater than approximately 2 birds/km². The Dornoch Firth and mouth of the Cromarty Firth were not surveyed in March and so it is impossible to assess model fit in that region. At those southern vantage points where red-throated diver were recorded, only single birds were observed in each survey, while models from digital aerial surveys did not predict high inshore abundance in this area (Figure 31).

Comparison of digital aerial survey and Citizen Science Pilot counts

Linear relationships between the digital aerial survey models and Citizen Science Pilot counts were poor for January and March on both the linear and log scales (Figure 32). This is reflected in a visual comparison of  Figures 30 and 31.

Comparison of Citizen Science and contemporary WeBS data

There was no discernible relationship between mean red-throated diver densities calculated for Citizen Science Pilot and WeBS surveys in January 2020 (Figure 33).

Comparison of historic RSPB and WeBS counts

For red-throated diver, there were moderate correlations (r > 0.6) between WeBS and RSPB rates of observations for December months (Figure 34) in the Nairn and Culbin Bar sub-region and for January (Figure 35) in the Outer Dornoch Firth region. In general, RSPB rates were higher than those for the WeBS counts. For the January surveys, both WeBS and RSPB counts follow similar patterns in the Inverness and Beauly Firth and Nairn and Culbin Bars sub-regions.

Comparisons of 2020 survey data and predicted RSPB counts

The historic Outer Dornoch Firth WeBS and RSPB surveys for January months had a correlation of 0.69, and so a prediction was made for this region for the 2020 rates. Observed rates of red-throated diver in the historic RSPB data ranged from a peak of 7.92 birds per hour in 1999, to a low of 0 in 2006. The predicted observation rate (1.42) falls within the range of the 2003 – 2006 values (0 - ~ 2), but all earlier observation rates (except for in the year 2000) were higher. It is noted, however, that the upper limit of the prediction (4.70) is higher than all except the peak observation rate in 1999. The Citizen Science Pilot observation rate of red-throated divers (4.0) falls within the confidence intervals of this prediction as well, though was high compared to other values. The highest modelled densities of the three sub-regions were predicted in the Outer Dornoch Firth and this was also true for the observed and predicted observation rates derived for the various shore-based surveys (Table 8).

Table 8 Comparison of mean modelled density (birds/km²) from digital aerial surveys, mean rates of observations in Citizen Science Pilot surveys, mean rates of observations in WeBS count, and predicted rate of obervations for RSPB counts (if available) for red-throated diver in January 2020 within three broad RSPB sub-regions. Negative rates can occur in confidence limits because the predictions were done on a linear scale without clipping to 0.

Sector name	Modelled mean (birds/km2)	Citizen Science Pilot rate (birds/hr)	WeBS rate (birds/hr)	RSPB predicted rate (birds/hr) and 95% confidence limits around prediction
Inverness & Beauly Firth	0.06	0.43	0.35	–
Nairn & Culbin Bars	0.04	0.00	0.27	–
Outer Dornoch Firth	0.14	4.00	0.81	1.42 (-1.85/4.70)

 

Comparative analyses: common eider

Bayesian point modelling

Overall population estimates derived by Bayesian Point Process Modelling for common eider in the Moray Firth SPA in January and March 2020 are presented and compared with those generated by blocked bootstrapping (Table 11).

Figures 36 and 37 show the modelled densities and vantage point counts respectively for common eider in January and March 2020.  Visual comparison indicates regions of relatively high abundance as detected by these two count methods.   

The broad modelled distribution patterns of common eider across the SPA were similar for January and March digital aerial surveys with the highest densities being in the Outer Dornoch Firth region and the Nairn/Culbin Bar region. Generally, the distribution was coastal.

Common eider, being a widely distributed species, were observed from most of the Citizen Science Pilot vantage points with the highest numbers being counted in the Outer Dornoch Firth, and the Nairn/Culbin bar regions (Figure 37). When comparing Citizen Science Pilot data to Figure 36 (modelled HiDef data) qualitatively, there is a reasonable amount of match in the southern parts of the SPA. However, there seems to be a level of mismatch between the modelled output and the Citizen Science Pilot counts in the Outer Dornoch Firth during March, with generally higher densities predicted by the digital aerial surveys.

Comparison of digital aerial survey and Citizen Science Pilot counts

A linear comparison of the Citizen Science Pilot counts and modelled data shows that the modelled output and the counts in March were more closely correlated (Figure 38). However, much of the relationship seems to be defined by a single value at the top right of the plot. The Spearman’s correlation for March was 0.61, which is a moderate value and reflects what is often found in noisy relationships in ecology. The log-log relationship for January had a Spearman’s correlations coefficient of 0.54, but visually there is a large amount of variation in that relationship (Figure 38).

Comparison of Citizen Science and contemporary WeBS data

Comparing Citizen Science Pilot data to WeBS counts for common eider yielded no significant linear relationship (r = 0.02; Figure 39).

Comparison of historic RSPB and WeBS counts

The historic RSPB data and WeBS counts show no strongly positive correlations with each other, with the exception of observations in the Outer Dornoch Firth during December months (r = 0.77). No observations of common eider were made in the Inverness and Beauly Firth during WeBS counts, though a few observations were made during RSPB surveys in 2001 (Figure 40 and Figure 41).

Comparisons of 2020 survey data and predicted RSPB counts

Broadly, the abundance metrics were highest in the Nairn and Culbin Bars sub-region for both the WeBS counts and modelled abundances. For the Citizen Science Pilot counts, the abundance metric (rate of encounters) was highest in the Outer Dornoch Firth (Table 9). Due to poor fit between RSPB and WeBS counts in January for common eider, predictions were not made on the expected RSPB rate.

Table 9 Comparison of mean modelled density (birds/km²) from digital aerial surveys, mean rates of observations in Citizen Science Pilot surveys, mean rates of observations in WeBS count, and predicted rate of obervations for RSPB counts (if available) for common eider in January 2020 within three broad RSPB sub-regions.

Sector name	Modelled mean (birds/km2)	Citizen Science rate (birds/hr)	WeBS rate (birds/hr)	RSPB predicted rate (birds/hr) and 95% confidence limits around prediction
Inverness & Beauly Firth	0.42	0.00	0.14	–
Nairn & Culbin Bars	31.19	55.12	183.49	–
Outer Dornoch Firth	0.88	262.80	0.54	–

Comparative analyses: common scoter

Bayesian point modelling

Overall population estimates derived by Bayesian Point Process Modelling for common scoter in the Moray Firth SPA in January and March 2020 are presented and compared with those generated by blocked bootstrapping (Table 11).

Figures 42 and 43 show the modelled densities and vantage point counts respectively for common scoter in January and March 2020.  Visual comparison indicates a number of regions of relatively high abundance as detected by these two count methods.   

Common scoters are a flocking species, commonly found in groups numbering in the hundreds to thousands in certain areas of the Moray Firth SPA. This is reflected in the modelled output with the majority of the SPA predicted to have densities of near zero, with a few intense hot spots. For both January and March surveys, these were predicted to be in the Outer Dornoch Firth, Nairn and Culbin Bars region, and in the Lossiemouth to Speyside area (Figure 42).

Only five Citizen Science Pilot vantage points (after merging adjacent counts to account for possible issues with double counting) had sightings of common scoter in the January and March surveys, however, these were large flocks (Figure 43). Qualitatively, the hotspots identified by Citizen Science observers match the hotspots predicted by the models from the digital aerial surveys. Additionally, vantage points in areas predicted from the digital aerial surveys to have densities of near zero for common scoter did not detect any of these birds (Figure 42 and Figure 43).

Comparison of digital aerial survey and Citizen Science Pilot counts

As with common eider and red-throated diver, there were no discernible relationships on linear or log scales for common scoter between the modelled HiDef data and the Citizen Science Pilot counts (Figure 44). However, because there were so few Citizen Science vantage points with sightings of common scoter, it would be difficult to ascertain any patterns using this approach.

Comparison of Citizen Science and contemporary WeBS data

A comparison between Citizen Science Pilot data and WeBS counts for common scoter yielded a correlation coefficient of 0.69.  However, the very small sample size (five data points) and complete overlap of the confidence intervals at the low and high extremes of the plot mean that the correlation is likely spurious (Figure 45).

Comparison of historic RSPB and WeBS counts

No sightings of common scoter were reported in the Inverness Beauly Firth for December or January months in WeBS or RSPB surveys. Similarly, the rates of sightings of common scoter in the outer Dornoch Firth regions were low for December and January until 2003 (December, RSPB), and 2004 (January, WeBS). The RSPB and WeBS counts were highly correlated in the Nairn and Culbin Bars regions in December, with patterns matching very well and RSPB counts consistently showing higher rates of observations (Figure 46).   However, this was not replicated in the January data(Figure 47).

Comparisons of 2020 survey data and predicted RSPB counts

Given the strength of linear correlations between WeBS and RSPB counts in December months (Figure 46), a prediction of RSPB rates was made to the contemporary data (January 2020) for the Nairn and Culbin Bars region. The confidence limits on the prediction are very wide, but the point prediction of the rate (618 birds/hour) is within the limits of historic rates (ranging from ~0 to ~ 2,300 birds/hour). At face value this might suggest that the population in this region has increased compared to the 1999 counts (i.e. the baseline of the RSPB counts provided). The Citizen Science Pilot observation rates were highest for the Outer Dornoch Firth, which contrasts with the findings from WeBS or the modelled distribution which shows the Nairn and Culbin Bars region to have the highest survey metrics (Table 10).

Table 10. Comparison of mean modelled density (birds/km²) from digital aerial surveys, mean rates of observations in Citizen Science Pilot surveys, mean rates of observations in WeBS count, and predicted rate of observations for RSPB counts (if available) for common scoter in January 2020 within three broad RSPB sub-regions. Negative rates can occur in confidence limits because the predictions were done on a linear scale without clipping to 0.

Sector name	Modelled mean (birds/km2)	Citizen Science rate (birds/hr)	WeBS rate (birds/hr)	RSPB predicted rate (birds/hr) and 95% confidence limits around prediction
Inverness & Beauly Firth	0.00	0.00	0.00	–
Nairn & Culbin Bars	32.98	64.95	343.06	618 (-269/1505)
Outer Dornoch Firth	3.06	94.00	0.54	–

 

Modelled population estimates

Table 11 shows the population estimates derived from the digital aerial survey data using Bayesian point process modelling for red-throated diver, common eider and common scoter for January and March 2020. The mean estimates derived from modelling differed slightly from those derived from the block bootstrap method for all species but fell within the confidence limits of each other. However, there was a clear difference in the width of the confidence limits, with the modelled population estimates much more precise than those derived using the block bootstrap method for all species in both surveys (Table 11).

The population estimates derived from the Bayesian point process for common eider in January and March 2020 were broadly comparable, with the mean estimate from January falling into the upper boundary of the estimate for March. However, modelled output from the digital aerial surveys suggests that common scoter and red-throated diver populations differed greatly between months (Table 11). Notably, the Moray Firth SPA population for common scoter was estimated to be nearly 10,000 birds in January, and just over 2,000 birds in March. This is consistent with Citizen Science Pilot counts, in that fewer vantage points had high counts in March compared to January (Figure 43). For red-throated diver, the population estimate increased from 254 birds in January to 918 birds in March (Table 11).

Table 11 Population estimates plus 95% confidence limits derived from the modelled distribution of red-throated diver, common eider, and common scoter for January and March 2020. (Estimates derived from the same data by blocked bootstrapping are also shown in brackets).

Month	Species	Low estimate	Mean estimate	High estimate
January	Red-throated diver	179 (139)	254 (259)	362 (386)
January	Common eider	3508 (612)	3808 (4,091)	4200 (12,182)
January	Common scoter	9443 (200)	9989 (10,336)	10682 (25,284)
March	Red-throated diver	774 (552)	918 (880)	1064 (1,315)
March	Common eider	2962 (1,105)	3355 (3,316)	4044 (6,148)
March	Common scoter	1844 (511)	2091 (2,607)	2477 (5,726)

 

DISCUSSION

The results of the surveys and the exploratory comparative analyses are discussed below.  This discussion includes an overview of the digital aerial surveys and examines how the results of these surveys compare with the citation population estimates for the Moray Firth SPA.  This is followed by discussion of characteristics of the four different survey types considered and the results of the comparative analyses of these data sets.

Digital aerial survey results and comparison with SPA citation populations

The digital aerial surveys recorded a total of 7,906 birds of 43 species. There were 2,821 observations of the 11 target species in January 2020 and 2,390 in March 2020. An overall identification rate to species level of 93.5% across the survey program was achieved and of 97.7% for the target inshore wintering waterfowl and shag

For all 11 target species, population estimates have been calculated using bootstrapped mean density methods. For three species (red-throated divers, common eider and common scoter) alternative estimates were also calculated, using a Bayesian modelling process (Table 11).

As detailed in the species accounts, the proposed Moray Firth SPA was recommended for classification largely based on data collected between 2001 and 2006 using shore-based and/or visual aerial survey methods (SNH 2016). These used the standard convention of deriving population estimates from the average of the 5-year seasonal peak count. The SPA population estimate for European shag was based on analysis of longer term ESAS data, largely derived from boat-based surveys. Comparison of these mean of peak counts and those of the digital aerial methods are summarised in Table 12.

Caution is required in inferring population change or that some methods are more robust based on these outputs because the 2020 surveys represent snapshots from one season compared to mean values across five or more years some 14–19 years ago. However, the similarities and differences between the 2020 population estimates and the mean of peak estimates underpinning site selection are here identified and briefly discussed. 

Table 12 Comparison between SPA citation populations and 2020 digital aerial survey estimates (derived by blocked bootstrapping; see also Table 11). The type of surveys from which the citation population estimates were derived are detailed in Table 1.

Species	SPA citation population	19 Jan 2020	8 Mar 2020
Greater scaup	930	6 (0–18)	0
Common eider	1,733	4,091 (612–12182)	3,316 (1,105–6,148)
Long-tailed duck	5,001	1,671 (876–2,658)	4,328 (1,985–7412)
Common scoter	5,479	10,336 (200–25,284)	2,607 (511–5,726)
Velvet scoter	1,488	79 (0–234)	12 (0–36)
Common goldeneye	907	55 (12–107)	109 (6–285)
Red-breasted merganser	151	49 (6–108)	362 (134–655)
Red-throated diver	324	259 (139–386)	880 (552–1,315)
Great northern diver	144	187 (59–360)	747 (440–1,090)
European shag	6,462	0	1762 (76–4,607)
Slavonian grebe	43	31 (6–66)	24 (6–48)

 

These comparisons show that for common eider, common scoter and Slavonian grebe, the SPA citation population estimate fell within the design based confidence limits of both the 2020 digital aerial survey population estimates (Table 12). Slavonian grebe (non-breeding) numbers in both Scotland and the UK have increased substantially since 1993/94, but the trend more recently (from 2005/06 to 2016/17) is declining.   However, for common eider the citation population estimate was below the narrower confidence limits derived from the modelled outputs for both surveys in 2020 (Table 11). This is at odds with 22.5% decline in common eider numbers in both Scotland and the UK since at least 1980/81, and may reflect differences in survey methods. .For common scoter, the citation population estimate was below the modelled confidence limits for the January 2020 digital aerial surveys but higher than those for March 2020 (Table 11 ). Common scoter numbers have increased in the UK since 1980, with their population estimated to have increased by over 155%, so the results here may suggest some variability in detection of flocks within the areas sampled between the two 2020 surveys.

For red-throated and great northern divers the citation population estimates fell within the confidence limits for the lower (January) 2020 digital aerial survey design based estimates, but the March 2020 digital aerial survey generated much higher population estimates (Table 12). The same result applies to the 2020 modelled population estimates for red-throated diver (Table 11). Zydelis et al., (2019) found that red-throated divers were recorded in higher numbers from digital aerial survey than traditional aerial methods. This is consistent with the results noted here for both diver species despite differing national trends.  Great northern diver numbers in both Scotland and the UK have been generally increasing since at least 1993/94 but red-throated divers in the non-breeding period have decreased by 27% in the UK between 2005/06 and 2016/17 (BirdLife International, 2019).

For long-tailed duck and red-breasted merganser the citation population estimates fell within the design-based population estimate confidence limits for the March 2020 digital aerial survey (Table 12).  However, the January 2020 digital aerial population estimates were lower than the citation population. Long-tailed duck numbers in both Scotland and the UK have declined by 83.5% since 2005/06 and in the same period red-breasted mergansers have declined by , 21.3%. 

For greater scaup, velvet scoter, common goldeneye and European shag, the SPA citation population estimates were substantially higher than either of the 2020 digital aerial survey estimates; the differences in estimates for scaup and velvet scoter were especially marked (Table 12).  This is in line with wider trends for goldeneye, which in the UK have decreased by over 8% between 1980-2017, with the short-term trend (2005/06-2016/17) showing an even steeper decrease of around 36%. Scaup numbers across the UK increased  by 15% between 1980/81-2016/17, but between 2005/06 and 2016/17 there has been a steep decline of around 52%; this period more closely matches that between the original surveys underpinning site selection and the 2020 surveys. Similalry, velvet scoter (non-breeding) numbers in both Scotland and the UK have increased substantially since 1980, but the trend from 2005/06 to 2016/17 shows a declined of over 50%.    For some species, longer shore-based and boat-based counts may counteract the effects of availability bias (caused by diving species being underwater at the time of potential detection) more effectively than the snap-shot methods of digital aerial survey.

However, for greater scaup and velvet scoter, differences might also be affected by changes in abundance and their highly clumped nature, which may be problematic for sampling surveys where targeted survey design may be required to increase the number of sample points in areas with known hotspots. Although similarities between velvet and common scoter make them somewhat challenging to identify, low numbers being detected (as found in the digital aerial surveys, see Table 5) are not uncommon and may reflect genuinely low abundance in the area.  The proportions of all scoter seen that were positively identified as velvet scoter were below 1% for both the digital aerial surveys and Citizen Science Pilot counts in both months (Table 5 in this report and Table 7 in Graham and Thompson, 2022).  However, while all scoter were identified to species level in the digital aerial surveys (Table 5), only 24% and 8% were in the Citizen Science Pilot surveys in January and March respectively.  This greater confidence in identification of scoter to species level may reflect the survey methods.  For the digital aerial surveys, analysts were able to review multiple video images of individual birds before determining species, whereas the Citizen Science Pilot observers had to determine identification in the field, often for birds in dense and relatively distant flocks.

European shag were only detected in one of the two 2020 digital aerial surveys and the upper confidence limit of the associated population estimate was lower than the citation population figure. Digital aerial surveys of waters around the Isles of Scilly, Cornwall, found large variation in numbers and distribution of feeding European shag (Webb and Irwin, 2016), suggesting that similar activity could have caused the large disparity in numbers recorded between the two surveys in the Moray Firth.  Also for European shag around the Isles of Scilly, the adjustment to account for availability bias in digital aerial surveys resulted in a two-fold increase in density estimate (Webb and Irwin, 2016).

The SPA citation for European shag in the Moray Firth SPA used boat-based methods from opportunistic surveys contained in the ESAS database and modelling methods to derive population estimates (SNH 2016). This means that a systematic, random design for survey of European shag was not carried out which could potentially have biased the population estimate upwards or downwards and requiring caution if comparing the population estimates with those derived from more robust survey designs.

However, shags using the Moray Firth SPA during the breeding season will include those from East Caithness Cliffs SPA where numbers have declined by 53% since classification (1993-2015. This is in line with wider trends with a 45% decline in shag breeding populations from 1986 to 2015.  Similarly, non-breeding populations within the UK have experienced a 24% decrease (1993-2017) (Frost et al. 2018). The reasons for these declines are uncertain but are potentially associated with poor weather conditions and associated mass mortality events, or “wrecks”.  

Characteristics of different survey methodologies

This section presents a broad discussion of the characteristics of the various available survey data. It precedes more detailed discussion of the results of the comparative analyses.

Digital aerial surveys and models

Aerial surveys are always going to be influenced by weather conditions. At the most extreme this will stop flights, due to high wind speeds, icing conditions, low cloud or precipitation. Aerial surveys are however more reliable from a deployment perspective compared to boats. Land based observers will also be negatively influenced by deteriorating weather conditions to undertake surveys.

Winter surveys at northern latitudes will always be impacted by lower light levels. If flown on an overcast or dark day light levels can be particularly low. Conversely bright sunny winter days can be problematic for glare in shoulder daylight hours due to the low sun. Ideally optimum weather conditions of low wind speeds, during a settled high-pressure system would provide the best results.

Closer transect spacing may have improved detection rates of less abundant or more clumped species. This survey was flown at 3km transect spacing giving c.16.7% coverage. Reducing transect spacing to 2km would result in a 50% increase of percentage survey coverage. There is a trade off in terms of time, costs and data requirements in the survey design process. Additional flying time means expending both financial, resource and carbon budgets. Typically, offshore surveys (mainly for the renewable industry) are completed using data sets based on 10-12.5% coverage. However, for conservation data sets it can be argued that greater detail is required to ensure that less frequent species (e.g. Slavonian grebe) or concentrated flocks of species such as scaup are likely to be detected.

Modelling of strip transect data is a commonly used technique to extrapolate information into areas where survey coverage did not reach. For this report, a more recently developed methodology is applied that considers survey effort and uncertainty in ways that more commonly applied algorithms (e.g. GAMs) cannot. In the approach presented in this work, the spatial component (i.e. intensity of observations and autocorrelation) is the only parameter included in the model. This, in a general sense, creates a smoothing parameter rather than a truly predictive model. The advantage of this technique over a standard kernel density estimate (KDE) is that the consideration of survey effort and uncertainty associated with spatial autocorrelation means the extrapolation between transect lines are more representative of what might be observed in the field. However, this leaves the modelled output more vulnerable to issues around sampling bias than if environmental covariates were included. For example, for species that congregate in only a few areas and in large numbers, random sampling might lead to those flocks being missed by chance. Because only the spatial component is modelled, the missing data means the algorithm does not know to predict birds in that region, which can lead to spatial mismatch with on-the-ground observations.

This could be alleviated through a more focused modelling exercise where environmental covariates are included. This would first require a power analysis to determine how many digital aerial surveys would be needed to make a model that could be used to detect any change in population over time. However, the relatively tight confidence limits around the population estimates generated by the current model suggest that it was able to capture much of the variability in the data, despite the lack of environmental covariates.

This report presents both KDE maps and modelled distributions using a Bayesian point process model. Comparing KDE outputs to outputs from the Bayesian point process models in a rigorous sense would require several processing steps to make any assessment scientifically sound. In a broad sense, both methods can be compared visually, however this must be done with caution as visual differences can be deceiving depending on the scale and colours of the map. This can be demonstrated by comparing (for example) the KDE (Figure 12) and modelled distributions (Figure 30) of red-throated diver in January 2020. The artefact concentric circles around two observation points in the northern part of the study area in Figure 12 suggests the KDE has identified a “warm spot”. However, the output from Figure 30 barely shows this in the same region. That is to say that the Bayesian technique has better handled the stochastic effects driven by areas with only a few observations. The broad spatial patterns are similar, but this is not surprising considering both are derived from the same dataset and are interpolation techniques that use the spatial component to create smooth maps.

WeBS counts

See Wetland Birds Survey: Survey Methods, Analysis & Interpretation for full details of WeBS methodology.   The WeBS database is a formidable source of good quality data on numbers of birds using wetland areas in the UK. A lot of the focus in the database is on shorebirds, estuaries and large inland wetland sites, and this is where its main strength lies. The database does however also contain a great deal of valuable historical data for open coastal areas.  Volunteers are required to count all birds of all species they encounter during their survey, from shorebirds to other waterbirds and even crows and other wildlife. The area being counted, and the precise methods are not always well defined in the open coast. A count might consist of observations from a fixed vantage point or require the counter to walk along a section of coast with data summarised for the entire length of the site. There is no definition of the offshore extent of the counts nor their duration, although there is a requirement that counters will avoid duplicate counts of the same birds and that they will avoid duplication or omission of bird concentrations in abutting neighbouring sites.

As with all visually based surveys from coastal vantage points, the count area may simply not extend far enough into the feeding range for some marine species, especially divers and long-tailed ducks. Volunteers will have access to different elevations of count points, so sites will be a mixture of those where it is possible to stand on a cliff top or high dune and those where the volunteer can only access a point at sea level with limited viewing range offshore.

The greatest uncertainty with respect to the value of WeBS data in the open coastline is around data quality. All counts are co-ordinated to take place in the same middle weekend of the month regardless of the weather conditions on the day. This matters less at enclosed estuarine sites, but when attempting to count divers and seaduck in open coast areas, wave height and direction, sun glare and visibility are all critical to the efficiency of the survey. The unlimited duration of the counts can help in this respect, but weather will still limit the viewing range compared to a still day with good visibility and overcast skies. Prolonged survey periods may also bias the results if birds move between adjacent sections during the count.

As for other surveys, data quality will also be affected by the capabilities (e.g. eyesight), experience, training, and availability of volunteers.  The optical equipment available to counters is of particular relevance to survey of offshore divers and seaduck, for which good quality telescopes are required.  

Citizen Science Pilot counts

Full details of the Citizen Science Pilot count methodology are provided in Graham and Thompson (2022). The 2019/20 Citizen Science Pilot counts have several strong aspects, which make them very useful for comparing vantage point surveys to digital aerial counts. First, the records are much more thorough than those available from WeBS or in the RSPB vantage point surveys in that parameters such as distance and total effort (i.e. time performing surveys) were recorded. Other data fields such as sampling device, sea state, and visibility are all parameters that can be used to incorporate measures of uncertainty in the data models. Data were well organised in a format that could easily be ingested by any scripting language. The biggest strength of these data in relation to this study is the fact that the majority of vantage point counts were undertaken on the days of the digital aerial surveys.

The Citizen Science Pilot counts do however suffer from many of the same issues as most observational based surveys, including observer bias. These issues have been known about and discussed for decades in the marine ornithological community (e.g. Tasker et al., 1984, Van der Meer and Camphuysen, 1996, Spear et al., 2004) though primarily in the context of at-sea surveys. For vantage point surveys, much of the key literature on bias in point surveys come from terrestrial bird surveys (e.g. Faanes and Bystrak, 1981). Although the Citizen Science Pilot dataset does capture much of the information that could be used to estimate observer bias, specific information on the efficacy of individual observers is missing (as is almost always the case in such work). Further analyses and discussion of observer bias in these surveys is in Graham and Thompson (2022).

The other weakness of this dataset is the lack of standardised distance information for observations. The dataset contained some distance records but these were not recorded consistently between surveys (e.g. in some surveys, distance to birds was recorded in broad bands; 1000 – 2000m, while other surveys recorded values such as ~500m). This data collection should be standardized into distance bands (perhaps in 500 m increments at the very least). The lack of inclusion of a detectability calculation may lead to biased estimates for smaller birds which are more difficult to detect at distance.

Perhaps a more substantial an issue is that there is no positional information available on the birds recorded from the vantage points, which makes it very difficult to accurately compare these to the digital aerial surveys in a spatially-explicit manner. A problem also arises from overlaps in areas covered by adjacent vantage points, such that the same birds could be seen from more than one count position. This was accounted for here by averaging the counts across overlapping vantage point buffers, but could lead to statistical issues around the comparison. A more thorough sensitivity analysis was not possible in this study but should be undertaken to derive a more sophisticated approach to this problem. 

Historic RSPB surveys

The biggest strength of the RSPB vantage point data is the standardised nature of the collected data (see Kalejta-Summers and Butterfield (2006) for details). Counts were performed at similar timings every year and usually by the same observers, thus reducing overall observer bias. Counts for each sector were all performed within a few days.  This ensured both that summed population counts for a sector were likely to be robust for that time period (assuming observers counted correctly) and that double counting was minimised. The RSPB data also come with several reports which analysed the counts, giving some indication at least of the populations over time in terms of the surveys that were carried out.

The major weaknesses associated with this dataset were two-fold. Firstly, the format in which the data are stored is not relational or in a format for easy ingestion into analytical frameworks (e.g. R). Although some automation of data extraction is possible, there is a high level of manual labour associated with the process that means analysing the entire dataset was not possible within the timeframe of this project. Secondly, these data lack a thorough description of the methods. The reports simply state that birds were counted, but it does not go into details of how they were counted. For example, were birds estimated in large flocks and, if so, how were those estimates completed? Additionally, very poor information is available on effort, although it can be inferred from the gaps in the start times of counts that observer(s) were quickly counting and then moving to the next vantage point. More importantly, there is no information as to how far out to sea birds were counted. Therefore, it is difficult to ascertain what the densities of birds are for the vantage points.

Comparative analyses

Comparing survey methodologies is challenging and often not possible for a variety of reasons. These reasons can include spatial mismatch, temporal mismatch, data gaps and observer bias. Further to this, a species’ ecology and how they congregate in space and time can play a major role in determining which survey methods are most appropriate. In this report, the relationships between four survey methodologies are examined for the Moray Firth SPA for three species: common eider, common scoter, and red-throated diver.

Broadly, and perhaps unexpectedly in some cases (e.g. WeBS versus RSPB counts for common scoter in the Nairn and Culbin Bars sub-region for December counts), the relationships between derived survey metrics (i.e. observation rate or densities) are poor, even when applying some corrections for spatial and temporal survey effort. In cases where relationships are somewhat apparent, it was possible to make some inferences on population change (e.g. red-throated diver population in January 2020 is similar to the SPA citation population, while the comparison between predicted RSPB counts and initial RSPB counts suggested no change in observation rate over time).

The following discussion is structured around a series of topics and associated questions that were identified by NatureScot for the comparative analyses element of this project. Several recommendations are then made, including with respect to survey methodologies and deeper analysis of the available data.

Changes in populations

Have populations of red-throated diver, common eider, and common scoter changed since citation population estimates and what are the limitations of the available data?

Based on the available data and the analysis in this report, the question of population change cannot be properly addressed due to the incompatibility of the varying datasets associated with a lack of detailed information on spatial and temporal effort. The citation population estimates for red-throated diver and common eider were derived from visual aerial surveys, and so commentary on changes from those estimates requires careful consideration of the survey effort in the visual aerial surveys; these were not analysed in this report.

Another way of addressing possible population change was by looking at rates of observations in RSPB data and comparing to rates of observations in WeBS counts.  Although results in this report were inconclusive with respect to attempting to ascertain the actual population change, this initial study does provide a suggested way forward given time for a more detailed analysis.

Using the relationship between the rate of observations in WeBS counts and rate of observations in RSPB counts, the predicted rate of observations of red-throated divers in January 2020 was 1.42 birds / hour in the Outer Dornoch Firth. An examination of Figure 34 shows that in the Outer Dornoch Firth, January rates of observations were declining towards 2006, which could be indicative of a population decline. However, taking the prediction of 1.42 birds/hour at face value suggests that the current observation rates for red-throated diver are similar to those in the older data for that sub-region. Looking at Table 8, this sub-region had the highest predicted mean densities from the modelled data, the Citizen Science Pilot counts and the WeBS counts. However, this must be caveated with the fact that the relationship presented is simplistic and does not take into account the spatial coverage of both shore-based survey types. Here it is only possible to assume that the spatial coverage is similar based on descriptions of the survey methodology found in Butterfield (2001).

For common eider, only the December RSPB counts in the Outer Dornoch Firth had a strongly positive correlation with WeBS counts. It is notable that RSPB counts report no common eider in the Inverness and Beauly Firth, whereas WeBS data report several. This could be indicative of spatial mismatches in the area surveyed, where sites surveyed during WeBS counts were not covered by RSPB surveyors. It may also be indicative of effort put into the counts. The RSPB counts, performed by one or two surveyors, were often characterised by relatively quick surveys (20 – 30 minutes), while WeBS surveys could go for several hours with no indication of where in the sectors those counts were taking place, thus making it impossible to get a true estimate of temporal survey effort. Factors such as time of day or tides were also unable to be standardised between the two datasets due to lack of information, both of which would impact distribution of birds in the survey areas.

The relationship between historic RSPB data and WeBS observation rates were particularly strong in the Nairn and Culbin Bars region for common scoter. This could be due to the biology of the species being observed. Common scoter tends to aggregate in large flocks in spatially restricted areas. Past surveys suggest that the Nairn and Culbin Bars areas consistently have high numbers of observations, and it is a highly accessible site. It is therefore likely that both the RSPB and WeBS counts were able to achieve full coverage of this area during the survey period. Furthermore, there may be the possibility that the surveyors inherently knew where the birds were going to be and were able to focus efforts on counting there. Other species (e.g. common eider and red-throated diver) were more widely dispersed across the survey area according to digital aerial survey data and thus may be more likely to show variability in the counts. The mismatch of data for common scoter in the Outer Dornoch Firth area may be due to spatial biases (i.e., the entire area not covered in RSPB and/or WeBS counts). However, without more details on exact locations of counts for WeBS surveys, it would be difficult to assess this.

The relatively strong relationship between RSPB and WeBS observations in the Nairn and Culbin Bars sub region for common scoter enabled a prediction to be made to contemporary data had the RSPB surveys been repeated for January 2020. The predicted observation rate of common scoter here was 618 birds/hour, which was about 200 birds/hour lower than the rate observed in 2006. However, there were wide confidence intervals around this prediction. Taken at face value, it might be concluded that the common scoter population in this area has not changed significantly. However, there are some caveats to this approach that must be considered.

Most importantly is the very low sample size from which to compare RSPB and WeBS counts within each sub-region (n = 6 – 8). Comparisons show that these patterns differ between sub-regions and therefore combining the datasets would also not yield a reasonable result at this spatial scale. The other important issue it that a linear model is used here to generate predictions without taking into account the uncertainty around issues such as, for example, observer bias. There was not enough time in the scope of this work to build any more sophisticated type of time series model from which to make assessments. Furthermore, there was only the time to look at three sub-regions of the whole SPA, and an analysis of all sites across the area might yield other strong relationships. There was a lack of information on important temporal aspects as well in the WeBS data (e.g. time of day of survey, tide heights, location of surveys and spatial surveying effort), all of which could have large impacts on the abundance of birds in a survey. Although the RSPB and WeBS counts are consistent within themselves, they are not readily consistent with each other in most cases presented here.

It is not likely that a correction factor can be generated without a much more in-depth analysis of these two datasets independent from the contemporary data. Such correction factors would likely be crude and not capture the true sampling errors in the data. Further details are presented under Recommendations.

Value of shore-based counts

Due to a lack of information, it was not possible to assess the WeBS counts directly against the modelled distribution from the digital aerial surveys. Only six WeBS sectors that fell inside the digital aerial survey area were surveyed in January 2020. This is a similar issue facing the comparison of the 2020 vantage point and WeBS counts. It would be impossible to form any statistically valid opinion on these relationships. The comparison of vantage point and WeBS counts is shown in Figure 33, Figure 39, and Figure 45 to demonstrate the issue around lack of data. Comparing the Citizen Science Pilot data or modelled outputs to data from December or October counts would not be reasonable due to the significant temporal mismatch.

There were no discernible relationships between the modelled distributions from digital aerial surveys and the vantage point counts, at least when looking at abundance. Qualitatively, particularly in the case of common scoter, there is overlap in the location of the predicted hotspots from the digital aerial surveys and the observed hotspots from the Citizen Science Pilot counts. However, the densities from the counts and surveys do not show strong relationships.

Mismatch between vantage point counts and modelled distributions are likely due to a number of issues. One source of error comes from observer bias in the Citizen Science Pilot counts, which can have a large impact on count data (Spear et al., 2004). This can be overcome using multiple observers as suggested in Spear et al., (2004). Another source of error comes from the lack of information on distance to observed birds. Some Citizen Science Pilot observers made rough estimates of the distance to flocks or individuals; however, these were not standardised, and there were many data gaps where observers simply did not record any information. The accuracy of such measures could be improved slightly by using clinometers, graticules or possibly laser rangefinders. However, it should be noted that distance effects are likely to be conflated with the effects of environmental gradients such as distance from the shore and water depth, thus making it difficult to disentangle these effects from each other.

Although the aircraft flew through many of the vantage point buffer zones (set at 2km), the areas covered within those vantage point zones was relatively small because the aircraft is only covering a strip with of 500m. In the case of a highly flocking bird like common scoter, if a large flock is missed during an aerial survey, it would not be included in the extrapolation process. This could potentially be remedied by including environmental covariates in the modelling process; this was not feasible in the time allotted for this work and warrants visiting in future work.

Another potential source of error comes in the choice of the 2km buffer for the vantage point as approximating the maximum distance at which observers can accurately identify birds. However, this might depend on the altitude of the vantage point and increase if the observers were higher up. Ways of alleviating this issue would be to train observers on specific distances so that they are not surveying beyond 2km. A sensitivity analysis using different buffer sizes, applied to the available data, could also be informative.

Common eider showed the most promise when comparing densities between Citizen Science Pilot counts and the modelled data. This could be due to the species’ widespread distribution, such that random sampling from the aircraft was able to detect them at a reasonable enough frequency to extrapolate across areas where they were detected in the vantage point counts. Common scoter did not compare well, potentially because of the spatial mismatch between the digital aerial surveys and the vantage point surveys and the fact that the species aggregates in dense flocks which may have simply been missed in the digital aerial surveys. Red-throated diver observations from vantage point counts also showed a poor relationship with the modelled distribution from the digital aerial surveys. Red-throated divers have a higher availability bias due to their diving behaviour. Availability bias differs between digital aerial and vantage point surveys. A ground-based surveyor has a longer period within which to spot birds that may be spending time underwater than an aircraft flying at speed overhead. In addition, red-throated divers are less coastal than the other two species and the digital aerial surveys were detecting divers in the offshore areas which were out of visual range of the vantage point surveyors.

Unfortunately the WeBS data for March 2020 were not available when these analyses were undertaken and so it was not possible to fully compare the WeBS and Citizen Science Pilot data (see Figure 33, Figure 39, and Figure 45). There are also wider concerns with WeBS data relating to the lack of information about where, when and how the surveys were performed. The data are provided as summed counts within polygons that are spread around the Moray Firth. However, it is not clear if the entirety of this area is covered by surveyors. As such, the assumption is made that the entire area is covered equally and the observers are only searching within the boundaries of those polygons and not further out, bearing in mind that some of the polygons have a width from the coast of about 500m.

As discussed above (under Changes in populations), for some species, sub-regions and months, there were apparent similarities in patterns of counts between WeBS and RSPB vantage point surveys over a number of years, but this was not consistent.  However, the RSPB count rates were generally higher than WeBS estimates.  This is not unexpected given the differing focus and methods for these surveys.  The RSPB (and Citizen Science Pilot) vantage point surveys were designed to focus on wintering waterfowl.  However, WeBS counts include dabbling ducks and waders with waterfowl (e.g. seaducks) only being “incidental” observations.

The potential added value of vantage point surveys is discussed further in Graham and Thompson (2022) 

Detectability

Detectability as determined by perception and distance bias to observed birds is an important factor for most survey methods (Buckland et al., 2001). For the RSPB and WeBS counts, this is a crucial bit of information that is missing, thus giving unrealistic densities. As such, the only way currently to examine the relationships between the varying ground-count datasets is to look at the rate of observations which at least considers temporal effort. It may be possible to apply availability bias corrections for some species for which there are data when comparing with digital aerial surveys, but implementation of this should be done with caution given the lack of good quality information on dive rates for many species. Furthermore, the data as they exist do not contain the relevant information from which to develop detectability curves.

This is a challenging issue to handle, because of the resource required to train volunteers in distance sampling techniques, but is something that could be addressed in future work. However, distance sampling carries with it a number of assumptions encompassing correct estimation of distance, group size estimation, the assumption that all animals ‘on the line’ are detected and binning of data (Buckland et al., 2001). It also assumes that there is no environmental gradient that affects the bird distribution which might conflate with distance effects.

CONCLUSIONS AND RECOMMENDATIONS

Although many of the dataset comparisons were inconclusive, the work reported here has identified a series of further steps that could be taken to examine the questions asked more thoroughly. These were not feasible within the time available for this work and also given lack of relevant information about some of the data sets and small sample sizes. A number of recommendations are made on how best to approach these issues in the future.

Taking some of the initial results at face value, there does not seem to be a discernible change in the observation rates of red-throated diver or common scoter for the January surveys when compared to historic RSPB counts. However, it was not possible to compare Citizen Science Pilot counts to WeBS counts due to a small sample size, and therefore also not possible to compare the digital aerial surveys to the WeBS counts. Comparisons between Citizen Science Pilot counts and digital aerial surveys showed mostly weak relationships that varied across species with common eider showing the strongest relationship. Future work on refining the historic relationships and applying that information forward and doing more data collection to compare WeBS to vantage point counts would be important steps to finding the appropriate population index from which to continue long-term monitoring.

Digital aerial survey provided an effective means for surveying inshore waterfowl at this site, especially for species with distributions that extend beyond visual range from land, such as divers and common eider. Some species with highly clumped distributions are detectable in digital aerial surveys but correction for the effects of diving behaviour may be needed to provide absolute estimates of abundance and consideration should be given to whether greater sampling in some coastal areas would improve modelling approaches to deriving population estimates. The trial use of Bayesian point processing for modelling data for three species generated impressively narrow confidence limits in population estimates.

Recommendations

The following recommendations could help to address some of the outstanding issues that have been identified in this pilot study.

1. Consideration should be given to whether greater survey coverage could be applied to future digital aerial surveys, particularly for species with strongly clumped distribution patterns, such as common scoter. It is possible that the transect width of the aircraft led to flocks of birds being missed. Although 10% of the survey area is often quoted as appropriate, it may be that a higher area coverage would be preferable to deal with these tight aggregations. A stratified survey design, in which higher survey coverage is used in areas where these highly clumped species are most likely to be found, could be one option if combined with the Bayesian point process modelling method.
2. The two different modelling methods tested here for three SPA inshore wintering waterfowl species produced different, but similar, population estimates from the same underlying digital aerial survey data set. It is possible that the inclusion of environmental covariates (e.g., bathymetry, distance to shore, or seabed habitat) would improve model predictions into areas not covered by the digital aerial surveys, which may help when comparing with the ground counts.  However, some models require a minimum quantity of data before being utilised and this can be difficult to predict in advance. Consideration should be given to which model should be used at project inception stage, noting that this may alter depending on survey and third-party environmental data quality and availability.
3. A more detailed analysis of the historic RSPB data compared to the WeBS counts could be performed across the entire Moray Firth. This would require further details of the WeBS counts (i.e., number of observers, and/or spatial extent of the surveys, as well as temporal data such as time of day or tides). It would also require creating a spatial layer of the RSPB vantage points with appropriate buffers (e.g., 2km as per the Citizen Science Pilot counts) to determine densities, which would correct for the spatial component.
4. Outputs from the analysis of recommendation 3 could help develop a more sophisticated predictive model using WeBS counts as an index while taking into account other important factors (tides, time of day, etc.). This would generate an index that could be compared to the historical RPSB counts.
5. It was not possible to determine if the Citizen Science Pilot and WeBS counts were comparable due to the lack of temporally and spatially overlapping data. Future vantage point counts should be repeated at approximately the same time as the WeBS to enable comparison between count types. Once the comparison was made and the relationship solidified, citizen science vantage point counts could continue separately from WeBS. However, it would be necessary to get detailed information on the survey effort from WeBS to make this comparison.
6. Distance sampling techniques should be applied to citizen science vantage point counts if possible, to get realistic densities that take into account detectability. It could be possible to train observers to accurately define distances to birds, which would allow at least for a general detectability curve to be applied, assuming there are no confounding effects from environmental gradients.
7. Results from comparing modelled data to Citizen Science Pilot counts suggests that if the data were compared via presence/absence instead of abundance, there might be a more reasonable fit between the datasets. Furthermore, averaging the count data within overlapping Citizen Science Pilot vantage points may lead to underestimation of those abundance estimates, and a more sophisticated approach should be explored. An investigation into this may lead to a way of building an index that could be useful for monitoring specific populations.

REFERENCES

Baddeley, A., E. Rubak, and R. Turner. 2015. Spatial Point Patterns: Methodology and Applications with R. Boca Raton, FL: Chapman & Hall/CRC.

Bibby CJ, Burgess ND, Hill DA, Mustoe S. Bird census techniques. Elsevier; 2000 Aug 29.

BirdLife International. 2019. Species factsheets. Downloaded from BirdLife International website

Buckland, S. T., Anderson, D. R., Burnham, K. P. Laake, J. L. Borchers, D. L. & Thomas, L. 2001. Introduction to Distance Sampling. OUP, Oxford.

Butterfield, D. 2001. Moray Firth monitoring: winter 2000/01. Report. RSPB and Talisman Energy.

Brooks, S., Gelman, A., Jones, G., Meng, X., & Neal, R. M. 2011. MCMC using Hamiltonian dynamics.

Campbell, B. and Lack, E. (1985). A Dictionary of Birds. T and A D Poyser.

Diggle, P. 2014. Statistical Analysis of Spatial and Spatio-Temporal Point Patterns. 3rd ed. Boca Raton, FL: CRC Press, Taylor & Francis Group.

Efron, B.; Tibshirani, R. (1994). An Introduction to the Bootstrap. Boca Raton, FL: Chapman & Hall/CRC.

Faanes, C.A. and Bystrak, D., 1981. The role of observer bias in the North American Breeding Bird Survey. Studies in Avian Biology, 6, pp.353-359.

Frost, T.M., Austin, G.E., Calbrade, N.A., Mellan, H.J., Hearn, R.D., Stroud, D.A., Wotton, S.R. & Balmer, D.E.  2018.  Waterbirds in the UK 2016/17: The Wetland Bird Survey.  BTO, RSPB and JNCC, in association with WWT.  British Trust for Ornithology, Thetford.  40 pp.

Gilbert, G., Gibbons, D.W. and Evans, J., 1998. Bird Monitoring Methods: a manual of techniques for key UK species. Published by the RSPB in association with British Trust for Ornithology.

Graham, J, and Thompson, K. (2022) Approaches to monitoring wintering waterfowl in Marine Protected Areas – Moray Firth pilot study winter 2019/20 NatureScot Research Report No. 1281

Illian, J. B., S. H. Sørbye, and H. Rue. 2012. “A Toolbox for Fitting Complex Spatial Point Process Models Using Integrated Nested Laplace Approximation (INLA).” Annals of Applied Statistics 6 (4): 1499–1530.

Kalejta-Summers, B. and Butterfield, D. 2006. Numbers and distribution of wintering divers, grebes and seaducks in the Moray Firth, Scotland, 1998/99-2003/04. Wildfowl, 56(56), pp.113-128.

Lawson, J., Kober, K., Win, I., Bingham, C., Buxton, N.E., Mudge, G., Webb, A., Reid, J.B., Black, J., Way, L. & O’Brien, S. (2015). An assessment of numbers of wintering divers, seaduck and grebes in inshore marine areas of Scotland. JNCC Report No 567. JNCC, Peterborough.

Møller, J., A. R. Syversveen, and R. P. Waagepetersen. 1998. “Log Gaussian Cox Processes.” Scandinavian Journal of Statistics 25: 451–82.

Robert, C. P., Casella, G., & Casella, G. 2010. Introducing Monte Carlo methods with r (Vol. 18). New York: Springer.

Rue, H. and Held, L., 2005. Gaussian Markov random fields: theory and applications. CRC press.

Simonoff J. S. 1996. Smoothing Methods in Statistics. Springer, London.

Simpson, D. P., J. B. Illian, F. Lindren, S. H Sørbye, and H. Rue. 2016. “Going Off Grid: Computationally Efficient Inference for Log-Gaussian Cox Processes.” Biometrika 103 (1): 49–70.

SNH. 2016. Moray Firth, Proposed Special Protection Area (SPA) No. UK9020313. SPA site selection document: summary of the scientific case for site selection. Version 8.

Spear, L.B., Ainley, D.G., Hardesty, B.D., Howell, S.N. and Webb, S.W., 2004. Reducing biases affecting at-sea surveys of seabirds: use of multiple observer teams. Marine Ornithology, 32, pp.147-157.

Tasker, M.L., Jones, P.H., Dixon, T.I.M. and Blake, B.F., 1984. Counting seabirds at sea from ships: a review of methods employed and a suggestion for a standardized approach. The Auk, 101(3), pp.567-577.

Van der Meer, J. and Camphuysen, C.J., 1996. Effect of observer differences on abundance estimates of seabirds from ship‐based strip transect surveys. Ibis, 138(3), pp.433-437.

Webb, A., Harrison, N.M., Leaper, G.M., Steel, R.D., Tasker, M.L., and Pienkowski, W. 1990. Seabird distribution west of Britain. Final report of phase 3 of the Nature Conservancy Council Seabirds at Sea Project November 1986 – March 1990. Nature Conservancy Council.

Webb, A. and Irwin, C.I. 2016. Digital video aerial survey at sea of European shag Phalacrocorax aristotelis around the Isles of Scilly in 2015. HiDef unpublished report to Natural England.

Webb, A. and Nehls, G. 2019. Surveying Seabirds. In M.R. Perrow (Ed.) Wildlife and Wind Farms, Conflicts and Solutions. Pelagic Publishing, Exeter pp 60 – 95.

Zydelis, R., Dorsch, M., Heinänen, S., Nehls, F., and Weiss, F. 2019. Comparison of digital video surveys with visual aerial surveys for bird monitoring at sea. Journal of Ornithology, published online at https://doi.org/10.1007/s10336-018-1622-4.

Document downloads