Using satellite telemetry and aerial counts to estimate space use by grey seals around the British Isles

Authors

  • JASON MATTHIOPOULOS,

    Corresponding author
    1. NERC, Sea Mammal Research Unit, Gatty Marine Laboratory, University of St Andrews, St Andrews, Fife KY16 8LB, UK
      J. Matthiopoulos, NERC Sea Mammal Research Unit, Gatty Marine Laboratory, University of St Andrews, St Andrews, Fife KY16 8LB, UK (fax +44 1334 462632; e-mail jm37@st-andrews.ac.uk).
    Search for more papers by this author
  • BERNIE MCCONNELL,

    1. NERC, Sea Mammal Research Unit, Gatty Marine Laboratory, University of St Andrews, St Andrews, Fife KY16 8LB, UK
    Search for more papers by this author
  • CALLAN DUCK,

    1. NERC, Sea Mammal Research Unit, Gatty Marine Laboratory, University of St Andrews, St Andrews, Fife KY16 8LB, UK
    Search for more papers by this author
  • MIKE FEDAK

    1. NERC, Sea Mammal Research Unit, Gatty Marine Laboratory, University of St Andrews, St Andrews, Fife KY16 8LB, UK
    Search for more papers by this author

J. Matthiopoulos, NERC Sea Mammal Research Unit, Gatty Marine Laboratory, University of St Andrews, St Andrews, Fife KY16 8LB, UK (fax +44 1334 462632; e-mail jm37@st-andrews.ac.uk).

Summary

  • 1In the UK, resolving conflicts between the conservation of grey seals, the management of fish stocks and marine exploitation requires knowledge of the seals’ use of space. We present a map of grey seal usage around the British Isles based on satellite telemetry data from adult animals and haul-out survey data.
  • 2Our approach combined modelling and interpolation. To model the seals’ association with particular coastal sites (the haul-outs), we divided the population into subpopulations associated with 24 haul-out groups. Haul-out-specific maps of accessibility were used to supervise usage estimation from satellite telemetry. The mean and variance of seal numbers at each haul-out group were obtained from haul-out counts. The aggregate map of usage for the entire population was produced by adding together the haul-out-specific usage maps, weighted by mean number of animals using that haul-out.
  • 3Seal usage was primarily concentrated (i) off the northern coasts of the British Isles, (ii) closer to the coast than might be expected purely on the basis of accessibility from the haul-outs and (iii) in a limited number of marine hot-spots.
  • 4Although our results currently represent the best estimate of how grey seals use the marine environment around Britain, they are neither definitive nor equally precise for all haul-outs. Further data collection should focus in the south-west of the British isles and aerial counts should be repeated for all haul-outs.
  • 5Synthesis and applications. This work provides environmental managers with current estimates of grey seal usage and describes a methodology for maximizing data efficiency. Our results could guide government departments in licensing marine exploitation by the oil industry, in estimating grey seal predation pressure on vulnerable or economically important prey and in delineating marine special areas of conservation (SAC). Our finding that grey seal usage is characterized by a limited number of hot-spots means that the species is particularly suited to localized conservation efforts.

Introduction

The number of grey seals Halichoerus grypus Fabricius, 1791 in UK waters is estimated at c. 130 000 (Sea Mammal Research Unit 2002). This accounts for around 38% of the species’ world-wide abundance and makes British grey seals the focus of considerable conservation interest. Indeed, the species is listed in Annex II of the European Community (EC) Habitats Directive and its management is subject to the Conservation of Seals Act (1970). Conservation efforts have focused on the designation of protected areas. On the other hand, being large and locally abundant piscivores, grey seals are perceived by the fishing industry as a threat to fish stocks and fisheries. This has led to calls for population control. Thus, the demands of seal conservation and marine resource management have, in the past, come into conflict (Harwood & Croxall 1988). To determine the spatial overlap between seal foraging, fish abundance and human activity, and to inform local and national management policy, a map of the distribution of grey seals at sea would clearly be valuable. The objectives of this study were to: (i) derive such a map of the use of the marine environment by grey seals associated with the coasts and waters of the British Isles; (ii) demonstrate its use in answering questions about seal conservation and marine resource management; and (iii) suggest how it might be improved with further data collection.

Our methodology was partly determined by the species’ natural history. During the breeding season grey seals stay on or close to shore. Outside the breeding season (September–December in the UK) individuals aggregate (haul-out) on land for variable periods of time, particularly during their annual moult between January and May. When not hauled-out, they perform foraging trips at sea. Our data primarily consisted of synoptic haul-out counts obtained by aerial and ground surveys, and information on individual movements derived from satellite telemetry. Previous work on a subset of these telemetry data found that trips lasted an average of 2·3 days, with an average maximum range of 40 km from a haul-out site (McConnell et al. 1999). The majority (88%) of trips returned to the haul-out of departure. The remaining 12% were repeated trips between neighbouring haul-outs and occasional trips to distant haul-outs.

The behaviour and life history of grey seals and the available data gave three types of problems relating to bias, precision and temporal variability. The animals that were captured for tagging were mainly from the north-east of the UK and were predominantly older than 2 years. Therefore, problems of bias arose because sampling of individuals was not balanced across space and age. We dealt with the problem of spatial bias by dividing the population by haul-out groups and calculating regional uncertainty in our estimates of usage. We made and, wherever possible, tested certain assumptions on the similarity between animals from different haul-out groups. Namely, that they had the same distribution of trip durations and that they spent the same proportion of time on land. Bias due to undersampling juvenile age classes might occur if their use of space differed from that of adults, but this possibility was not examined.

As with most telemetry studies, due to economic constraints the number of animals that were tagged unavoidably comprised only a small part (approximately 0·1%) of the total population. Also, the aerial survey of haul-outs has, so far, been conducted only once. We addressed these problems of precision in two ways. First, we used data-efficient statistical methodology and modelling to make the best use of the data. Secondly, we produced spatially explicit measures of sampling effort and estimation uncertainty to help target further data collection.

Temporal variability in the spatial distribution of grey seals is characterized by a seasonal and, probably, an interannual component. As a consequence of their annual activity cycle, the spatial distribution of seals during the breeding season will not be the same as during the rest of the year. However, to determine the extent of their interaction with prey and offshore human activity, we focused on their distribution outside the breeding season, when they perform foraging trips at sea. We assumed that the spatial distribution of usage outside the breeding season was invariant for the duration of data collection (pseudo-equilibrium assumption; Guisan & Zimmermann 2000). In the long term, this assumption will be violated, as studies on other marine species (Fauchald, Erikstad & Systad 2002; Broekhuizen, Stahl & Sagar 2003) have shown. However, the first priority is be able to quantify the main features of the grey seals’ use of space, for which this study offers a methodology.

Materials and methods

synoptic haul-out counts

Data on all the Scottish and English haul-out sites were collected during a single aerial survey of haul-outs carried out during 1996–97. Data on Welsh (Countryside Commission for Wales, personal communication), Irish (Coulahan 2000; Arnett 2001), Faeroese (C. Duck, personal communication; Bloch, Mikkelsen & Ofstad 2000) and French (C. Vincent & V. Ridoux, personal communication) haul-outs were obtained from land-based counts. These data gave the positions of all major grey seal haul-out sites and provided a single count of animals at each of these sites on the occasion of the survey (Fig. 1).

Figure 1.

The onshore distribution of seals. The insert shows the positions of all 130 haul-out sites in our records and the groupings into 24 haul-out groups used in this study. The main plate shows the total numbers of animals counted at each haul-out group. More animals exist in, and more data are available for, the northern part (above the dashed line) of the map.

satellite telemetry

The large-scale movement of individual seals during their trips at sea was observed from May to September (outside the breeding and moulting season) from 110 tagged individuals captured at major haul-out sites during 1991–99 (Fig. 2). Data were obtained using Argos satellite relay data loggers (SRDL), developed by the Sea Mammal Research Unit (SMRU; Fedak, Lovell & McConnell 1996; McConnell et al. 1999; Fedak et al. 2002). These records of individual movement are typically long term (average duration 104·3 days) and low frequency (average 6·46 location fixes per day).

Figure 2.

The satellite location fixes colour-coded (a) by individual animal and (b) by haul-out group.

other movement data

Short-term (typical duration of observation 24 h) and fine-scale (typical interval between observations 10 min) data about the animals’ speed of movement were collected by real-time tracking a small number of acoustically tagged individuals (n = 8) (Thompson & Fedak 1993).

analysis and modelling

Our general approach was to first subdivide the total number of haul-outs into a manageable number of haul-out groups and then estimate the number of animals associated with each. We used information on speed and trip duration to calculate the accessibility of points at sea as a function of distance from the centre of the haul-out group. These maps of accessibility were used to aid estimation of space use by each tagged individual. Spatial usage and its associated uncertainty for a haul-out were estimated by combining the estimates of usage from all tagged individuals. Finally, haul-out-specific usage maps were weighted by the estimated number of seals associated with each haul-out and combined into a single map of usage for the seas surrounding the British Isles.

Haul-out groups

Haul-out sites were grouped according to the following rules: (i) haul-out sites in the same group were geographically adjacent; (ii) each haul-out site belonged to a single group; (iii) no haul-out site was further than 120 km from the centre of its group (most haul-outs in each group were generally closer, on average 31 km apart, well within the recorded average trip range of grey seals; McConnell et al. 1999). This resulted in a total of 24 haul-out groups (Fig. 1, insert).

Taking a haul-out-based approach allowed us to account explicitly for differences in usage and sampling effort between different subsets of the population. Grouping the haul-outs offered considerable computational gains and enabled us to treat transitional trips between neighbouring haul-out sites as return trips to the same group.

Haul-out populations

The haul-out surveys provided a single population count, mi, for each haul-out group, i, of animals that were present at the time of the survey (Fig. 1). However, we needed to know the expected total number, Ni, of animals associated with the haul-out group. In the absence of replicate counts, this number and the associated variance could not be obtained directly. We therefore modelled haul-out behaviour as a binomial process and obtained haul-out numbers and their variances using a Bayesian calculation (see Appendix S1, Supplementary material).

The formulae in Appendix S1 require the probability, p, that an animal from the ith haul-out was on the beach and was counted during the survey. This was assumed constant for all haul-outs and its value was estimated as follows. The total number of seals (116 000) using the British coasts (excluding Shetland) during the period of data collection (1996–97) has been estimated from repeated aerial surveys of breeding colonies (Sea Mammal Research Unit 2002). Assuming that, on average, the same number of seals uses the haul-outs along the British coasts (excluding Shetland) then, of this total number, only 24 047 were observed during the haul-out aerial survey. This gives a crude estimate of p = 0·21 for the probability that a seal was ashore at the time of the haul-out survey.

Pre-processing of satellite telemetry data

Argos location data were filtered by the algorithm described by McConnell, Chambers & Fedak (1992), using a ‘maximum speed parameter’ of 2·0 m s−1. The filter rejected outlying locations and retained a total of 13 684 locations, each indexed by individual animal. Locations were further assigned to a trip performed by that animal. A trip was defined as an excursion outside a 10-km range from the haul-out site following a haul-out event. The start of a haul-out event was defined as the SRDL wet–dry sensor continuously reading dry for 10 min. The event was ended when the sensor read wet continuously for 2 min. These haul-out conditions occasionally occurred tens of kilometres out to sea. This was probably due to the seal remaining motionless at the surface at sea; such events were omitted from our analysis. Trips were assigned to the relevant haul-out group, providing us with the set of trips that each tagged individual performed from a particular haul-out group.

Grey seals predominantly perform return trips (McConnell et al. 1999). Due to the rarity of transition events it is currently not possible to estimate the probability of transition of an animal from one haul-out group to another. We therefore assumed that seals only performed return trips. Collecting individual haul-outs into groups reduced the risk of violating this assumption because most trips returned to the haul-out group of origin. Nevertheless, a small number of individuals travelled from one haul-out group to another during the period of observation. To be able to perform the analysis, we treated the migrating animals as different individuals after they switched haul-out group. Individuals that gave us less than 10 satellite locations or less than 3 days of observation in a haul-out group were removed.

Using the temporal records in our telemetry data, we obtained distributions of trip duration from each haul-out group. We tested the null hypothesis that these differed between haul-out groups by performing a Kolmogorov–Smirnov test for every pair of haul-out groups for which we had trip duration data.

Usage from individual haul-out groups

Modelling and estimation of usage was done on a regular 271 × 380 grid (cell width 5 km) of the seas around the British Isles (Fig. 1). Geographically, the grid also included the Faeroes, the western coast of Norway and northern coast of France. The resolution was determined on the basis of computational convenience and the accuracy of the satellite telemetry data (Vincent et al. 2002).

A number of methods exists for the estimation of usage from telemetry data (Silverman 1986; White & Garrott 1990; Cressie 1991). It has been demonstrated by simulation (Matthiopoulos 2003a) that the efficiency (defined as the rate of convergence to the truth with sample size) of these methods can be improved by using auxiliary information in addition to the telemetry data. Auxiliary information can be obtained from known facts about the animals’ natural history, field measurements that relate to movement, or telemetry data obtained from previous studies. We used the method of model-supervised kernel smoothing (MSKS; Matthiopoulos 2003a), which employs additional information in the form of an auxiliary usage surface to supervise kernel smoothing of the available telemetry data. As our objective was to estimate marine usage, we excluded satellite data corresponding to haul-out events.

The aspect of the grey seals’ natural history that is relevant to their spatial distribution and about which we have auxiliary information is their link with haul-outs. This association with points on land implies that, during a return trip of given duration, not all points at sea are equally accessible. We therefore aimed to supervise usage estimation with maps of offshore accessibility for each haul-out group. To quantify the accessibility of each marine grid cell from the centre of each haul-out group we used the simulation framework proposed by Matthiopoulos (2003b), which estimates accessibility based on (i) the position of the haul-out group (as obtained from the survey data); (ii) the seal's speed of movement (as obtained from the accurate, short-range observations made on acoustically tagged animals); (iii) the distribution of trip durations (as extracted from the satellite data); and (iv) the ‘at-sea distance’ between the haul-out group and the marine grid cell (defined as the shortest path that enables animals to get from the haul-out to any grid cell while circumnavigating land). We approximated the model's output by a smooth, power function of at-sea distance from the haul-out. The exponent of this function represents the rate at which accessibility decays with distance from the haul-out.

Accessibility maps were used in MSKS to estimate the use of space by each tagged animal from each haul-out group. Averaging these maps for all individuals in a haul-out group gave us an estimate (ŝx,i) of the per-capita usage at a location x from the ith haul-out group. Usage could not be estimated for the five haul-out groups for which we had no telemetry data (Table 1). Our best estimate of usage for these haul-out groups was the map of accessibility. This was therefore an estimate based solely on auxiliary data (position of haul-out, speed, trip duration and obstacles) that suggested where animals could have gone given the restrictions in their movement.

Table 1.  Notation used in the paper
SymbolDefinition
xA location in space. In this paper, the coordinates of the centre of a grid cell
kiNumber of satellite tags giving us information on trips from the ith haul-out
pProbability that an animal was on the haul-out at the time of the survey
miNumber of animals counted on the ith haul-out during the survey
NiNumber of animals using the ith haul-out
NmaxEstimated total population of grey seals in the UK
sx,iUsage of x by a single animal from the ith haul-out
Wx,iUsage of x by all the animals from the ith haul-out
P(Ni | mi)Probability mass function of the number of animals using the ith haul-out given the number of animals counted there
inline imageProbability density function describing the error in the mean estimate in local usage of x by animals from the ith haul-out conditional on the number of tags placed at that haul-out

For each point in space and each haul-out group, we calculated the variance in the estimates of usage within our sample of tagged individuals. This could only be obtained for 14 haul-out groups, which had two or more tagged individuals. To calculate the standard error in the estimate of local usage we used a normal approximation to the error distribution. This assumption relied on the central limit theorem and therefore required that sampling units were independent. By using the individual, rather than the trip or the satellite location, as our sampling unit, we guaranteed that variability in spatial usage would not be underestimated as a result of serial autocorrelation (Swihart & Slade 1985, 1986; Hansteen, Andreassen & Ims 1997). We followed the advice of Otis & White (1999) that ‘if the objective of the study is to make inferences about the population of animals from a representative sample of radio-tagged individuals then the appropriate sampling or experimental unit is the individual, not the single location’.

Usage by the entire population

The usage map of each haul-out group was weighted by the estimated number of associated animals. Adding these maps together gave the aggregate map of usage for the sea around the British Isles.

The sampling effort at different haul-out groups was not balanced. Consequently, the estimate of usage at different points at sea was made on the basis of a variable number of tags. Therefore, estimates of usage at different points at sea were subject to variable degrees of sampling error. To visualize relative effort spatially, we plotted the following index for different cells on the grid:

image(eqn 1)

where ŝi,x is the estimate of usage at point x by animals from the ith haul-out group, ki is the number of tagged animals using the ith haul-out and N̂i is the estimated total number of animals using the ith haul-out. The index ux represents the estimated number of tagged animals that use the grid cell at x as a proportion of the estimated total number of animals using x.

Local precision of the estimate (Wx) of aggregate usage at x can be measured by combining the uncertainty in per-capita usage at x with the uncertainty in population numbers for all haul-outs using x (see Appendix S2 in Supplementary material). Due to the small numbers of tags in the southern part of our map (the north–south division; Fig. 1), it was only possible to quantify uncertainty in the northern part. Of the 16 haul-out groups in the north, three were represented by only one tagged animal. We were able to estimate local uncertainty for the remaining 13. The standard error in our aggregate mean usage estimates was calculated as:

image(eqn 2)

The value obtained from equation 2 will be an underestimate of local uncertainty because of the omission of three out of 16 haul-out groups. This effect will be particularly pronounced in areas that were used intensively by animals from these three haul-outs. We represented this visually, by setting the standard error as ‘unknown’ in locations in which these three haul-outs comprised more that 50% of our mean estimate of usage.

Results

Assigning return trips to tagged individuals and haul-out groups resulted in a data set of 149 individuals for all of the 24 groups. Removing seals for which we had little data reduced this to 135 individuals (Table 1), with a median of 58 satellite fixes and median observation duration of 64 days animal−1.

Colour-coding these locations by individual (Fig. 2a) and by haul-out group (Fig. 2b) demonstrated the geographical extent of the data within the area of interest. The majority of tagged animals were associated with haul-outs in the northern half of Britain (Fig. 3), making that region suitable for more in-depth analysis.

Figure 3.

The distribution of tagged animals by haul-out group. We plotted the frequency of haul-outs with particular numbers of tagged animals for all 24 haul-out groups and for the 16 northern haul-out groups. Removing the haul-out groups in the south leaves us with the most data-rich haul-out groups.

Among the pooled data on trip durations (Fig. 4), the bulk of the distribution comprised trips that lasted less than 7 days. Obtaining accessibility maps required a distribution of trip durations for each haul-out. To determine whether we could use the pooled data or whether we needed to disaggregate them by haul-out group, we performed a Kolmogorov–Smirnov test on the similarity between trip duration distributions from each pair of haul-outs. Out of a possible 276 pairwise comparisons, there were enough data for the test to be conclusive (at the 99% level) in 144 cases. Out of these, the null hypothesis of similarity was rejected in a minority of 31 cases, indicating that trip duration distributions were mostly similar. Similarity in trip durations was not related to the at-sea distance between haul-out groups and we concluded that the frequency distribution of pooled trip durations adequately described offshore activity.

Figure 4.

The pooled distribution of trip durations as extracted from the satellite tag data.

The formulae derived in Appendix S1 were applied to the synoptic haul-out count data to calculate the number and variability of animals associated with each haul-out group (Table 2). From the simulations of accessibility, we found that, given the seals’ speed and trip duration distribution, the accessibility of points at sea is given as d−1·98, where d is at-sea distance measured in units of 5 km. We weighted the accessibility maps from all haul-out groups by the estimated number of seals associated with each of them and then added them to get a picture of the entire population's range (Fig. 5).

Table 2.  The number of tagged animals (ki) that performed return trips at the ith haul-out group and the estimated number (i) of animals at each haul-out group
ikiiCI (i)ikiiCI (i)
1 23 8393601–407713 1693592–794
2 24 0913846–433614 0470386–553
3123 9913749–423415 6384309–459
41811 93311 514–12 35216 1265203–328
5 514 26213 804–14 72017 04 8044538–5070
6 14 5664307–482518 02 0951919–2271
7 313 97613 523–14 43019 0669570–768
8 59 7949415–10 17420 0503417–589
9 32 6182421–281421 2717614–819
10 11 049925–117422 32 4282239–2617
112010 3649974–10 75523219 4149042–9786
12282 9502742–315924 114 59414 131–15 058
Figure 5.

The compiled output of the accessibility maps for all 24 haul-out groupings. Predictions for each haul-out group have been scaled up by the estimated number of seals associated with that group. The underlying density plot (in colour) represents the number of animals that would use different points in space if usage was proportional to accessibility. To reveal some more of the detail of this map in regions of expected low usage we log-transformed these values to obtain the contours shown.

The haul-out-specific accessibility maps, together with the satellite data and the population numbers, were used in MSKS to produce usage maps for each tagged individual. Cross-validatory parameter estimation within MSKS consistently made use of the accessibility maps, indicating that they were useful in improving usage estimation. Compiling the maps of usage by haul-out and scaling them up by estimated population numbers gave the aggregate map of usage for the entire population (Fig. 6). This map revealed the existence of discrete, offshore hot-spots of usage. When compared with the aggregate accessibility map in Fig. 5, it suggests that grey seals have a highly heterogeneous spatial distribution, intensively using a small number of areas, relatively close to the coast.

Figure 6.

The estimated usage of the marine environment by the grey seal population. The coloured map represents the estimated number of animals using different locations at sea during the population's offshore time. Note that this is higher than total usage per time unit because seals also spend time on shore. The contours have been obtained by log-transforming usage to reveal some of the detail in the less frequently used areas.

Plotting the index of sampling effort in equation 1 revealed that less sampling effort was associated with the south and west of the British Isles (Fig. 7). We therefore expected that the usage estimates made about these groups were less precise compared with estimates for the north and east.

Figure 7.

Offshore sampling effort for this study. We have plotted the number of tagged animals that are estimated to use each marine location as a proportion of the estimated, total number of animals that go there.

A more detailed investigation of precision for the northern part of the map (Fig. 8) revealed the areas where limited sample size and individual variation had increased the uncertainty in our estimates of usage.

Figure 8.

The plate on the left gives an enlarged view of the marine distribution of seals in the north of England and Scotland. The plate on the right gives the standard error in our estimates. Areas with very large standard error and areas where the standard error is unknown due to limited sampling effort are shown in red. Note that the standard errors shown are strictly in reference to local usage because variations in usage between neighbouring grid cells are not independent.

applications

Overlap with marine, industrial activity

As part of its ongoing activity, the UK Department of Trade and Industry (DTI) is undertaking strategic environmental assessments before any UK Continental Shelf (UKCS) license rounds for oil and gas exploration and production. Before further regional development proceeds, the DTI consults with the full range of stakeholders in order to identify areas of concern and establish best environmental practice. Spatially explicit reports on marine mammals (Hammond et al. 2001, 2002, 2003) form an integral part of the scientific advice submitted before these consultations. Information about marine mammal usage within the region shown in Fig. 9a formed part of such a request (SEA 2; Hammond et al. 2001). We calculated that the boundary contained a small percentage, equal to 0·91%, of the entire grey seal population's at-sea usage and was therefore not an area of concern for the species.

Figure 9.

The results of this study can be used to (a) calculate usage in an arbitrary region at seal (b) delineate areas of high usage containing a given proportion of total usage; (c) calculate total seal usage within a certain distance from the coast; and (d) specify regions of fixed number, shape and area that contain as much of the population's usage as possible.

Delineation of marine SAC

Currently, special areas of conservation (SAC) for grey seals are exclusively terrestrial. While the availability of coastal sites for use as haul-outs is probably important to the seals, it is likely to be less so than the availability of feeding grounds. This has prompted proposals for the creation of marine SAC. Although usage by the animals is not a sufficient criterion for the delineation of SAC, in the absence of information on habitat preference it is the preferred choice.

We propose two ways in which SAC could be defined on the basis of usage. If the requirement is to protect an arbitrary number of regions containing a certain percentage of the population's total usage, then we can simply select the isocline in the map of usage that encapsulates the required percentage. For example, in Fig. 9b we have outlined those hot-spots that contain 50% of the population's total at-sea usage. The total area of these regions (16 550 km2) is relatively small compared with the species’ potential range. This is particularly relevant for conservation efforts because it implies that, unlike other marine mammals, grey seals are amenable to protection by SAC.

Alternatively, the requirement may be to define SAC of fixed shape, number and area, as these may be easy to manage, survey and patrol. As an illustration, we identified five square regions, each of area equal to 2500 km2, containing the maximum percentage of at-sea usage possible (Fig. 9c). This was equal to 33%.

Overlap with coastal fisheries

Questions of by-catch risk for grey seals in relation to coastal fisheries could be addressed by estimating the total percentage of grey seal space use enclosed within the range of the fishery. Using our results, we looked at a 35-km zone around the coasts (Fig. 9d). This was found to include 60% of the population's total at-sea usage. The effect of other human activities, such as contaminant outlets and recreational boating, could also be investigated in this way.

Predation pressure on fish stocks

Estimating predation pressure on endangered or commercially important prey is a recurring theme in applied ecology (Brøseth & Pedersen 2000; Stahl et al. 2001; Hindell et al. 2003). Given increased awareness of the global importance of fish stocks (Ormerod 2003 and references therein) and the need for spatially explicit management, the results of the present study could be used to estimate predation by seals on fish stocks. Given a corresponding map for the distribution of a fish species or a species’ size class, the rate of encounter with grey seals can be calculated.

Admittedly, predators are capable of altering their spatial distribution in pursuit of their prey and the results of this study cannot provide dynamic predictions of such changes. However, in the case of generalist predators such as grey seals, whose diet contains more than 25 prey species (Hammond, Hall & Prime 1994a,b), the distribution of usage by the population will be less sensitive to changes in the distribution of any one prey species.

Discussion

In the last decade, satellite telemetry has yielded spatially extensive data sets on animal movement. For many marine animals, this presents the first opportunity to track the movements of individuals in their aquatic environment. As a result, recent studies have provided valuable insights into the offshore behaviour of individual animals (Boyd 1989; Weimerskirch et al. 1993; Folkow, Martensson & Blix 1996; Bost et al. 1997; Campagna, Fedak & McConnell 1999; Le Boeuf et al. 2000; Block et al. 2001; Field et al. 2001; Hays et al. 2001; Harcourt et al. 2002; McConnell et al. 2002; Thompson, Moss & Lovell 2003).

These data sets are now sufficiently large (Fauchald, Erikstad & Systad 2002; Mauritzen et al. 2002; Broekhuizen, Stahl & Sagar 2003; Hindell et al. 2003) to allow inferences about the spatial distribution of entire populations, an endeavour that is receiving considerable attention in the wider ecological literature (Turchin 1991, 1998; Farnsworth & Beecham 1997, 1999; Flierl et al. 1999). Scaling up from the individual to the population satisfies the immediate objective of providing guidelines for conservation and management on the basis of existing data, and gives rise to biological hypotheses that can be tested with more focused data collection. In this study, we have combined data on the terrestrial and marine distribution of individual grey seals in Britain to produce a map of the population's use of space (compare the raw satellite track data of individual seals in Fig. 2 with our map of estimated population usage in Fig. 6).

Our approach has four distinct advantages. First, we exploited the spatial autocorrelation in the satellite data by using kernel smoothing techniques. Such interpolation techniques aim to reduce the impact of sampling error in the estimates (Silverman 1986; White & Garrott 1990) and are routinely used in studies of terrestrial species (Worton 1989, 1995; Wray et al. 1992; Andreassen et al. 1993; Seaman & Powell 1996; DeSolla, Bonduriansky & Brooks 1999; Blundell, Maier & Debevec 2001). Use of kernel smoothing techniques offers improved efficiency compared with previous work where the authors estimated usage by binning the location data in a spatial raster (Hindell et al. 2003).

Secondly, we side-stepped the problem of temporal autocorrelation in the data (Swihart & Slade 1985, 1986; Solow 1989; Cresswell & Smith 1992; Rooney, Wolfe & Hayden 1998; Salvatori et al. 1999) by using the individual as the sampling unit for estimating usage. We therefore avoided underestimating the variance in expected usage (Aebischer, Robertson & Kenward 1993; Otis & White 1999) and retained all the information in our data set (DeSolla, Bonduriansky & Brooks 1999).

Thirdly, we used auxiliary information on accessibility to supplement satellite telemetry data. The fact that cross-validation in MSKS did not reject the accessibility maps implied that they improved usage estimation and therefore allowed us to use the available telemetry data even more efficiently.

Fourthly, wherever sample size allowed, we provided measures of variability in sampling effort and estimation uncertainty. Our measures of local uncertainty (Fig. 8) bring together the two sources of sampling variation in our data, namely uncertainty in haul-out counts and variation in the telemetry observations. Given that further data collection should aim to increase the precision of usage estimates, these spatially explicit measures of uncertainty allow us to target further data collection.

In interpreting our results some important caveats need to be considered. As our primary interest was on the seals’ marine activity, we ignored usage within 10 km of a haul-out site. Our choice of 10 km as the distance threshold for defining the initiation and termination of trips primarily reflects the spatial quality of the Argos location fixes rather than the biology of seals. It is possible that seals performed short trips within the 10 km range that we did not recognize as trip events. This explains why trips of 6 h or less appear with a relatively low frequency in our distribution of trip durations (Fig. 4)

In the absence of comprehensive data on exchange rates between haul-out groups, we included only return trips in our analysis. We do not consider this a serious problem because most of the trips in the data were return trips. However, it would be useful to incorporate transitional movement in the results. The sampling effort required to calculate exchange rates from rare, migration events probably means that Argos SRDL are currently too costly a method. Alternative methods include the use of cheaper tags, such as GSM phone tags (McConnell et al. in press) or passive photo identification mark–recapture studies (Hiby 1994)

The comprehensive aerial survey of haul-outs that provided the bulk of our information on haul-out positions and numbers has, so far, only been conducted once and we therefore have no direct measures of variability in haul-out numbers. To overcome this, we modelled haul-out behaviour as a binomial process (see Appendix S1) and assumed that, in conditions similar to those under which the counts were made, grey seals haul-out independently, with a constant probability. Repeated counts would allow tests of these assumptions in the future.

Partly due to variable catchability and partly through design, our sample of tagged animals is not representative of the population's age and sex structure. For example, no first-year juveniles were represented. To investigate any possible biases, we need to examine the differences in usage patterns between different ages and genders. Given sufficient data, we could estimate separate usage maps for grey seals belonging to different groups. We could then use parametric, non-parametric and resampling methods to test whether any differences in usage between groups were significant. The ability to test such hypotheses makes our methods a powerful tool for analysing usage data.

A final caveat concerns our assumption of pseudo-equilibrium. The movement behaviour of grey seals and, hence, their distribution, varies seasonally and, possibly, annually. The seasonal component is not of great importance to this study as our prime objective was to examine offshore usage. We therefore limited our tagging and estimation to the period May–September, avoiding the seasons of terrestrial activities such as breeding and moulting. Possible interannual trends in the distribution of usage are more relevant to the validity of our results. To achieve the spatial coverage presented here, we used all the available satellite data, which spanned a period of 9 years. Additionally, there are a few replicate tag deployments at different years from the same haul-out sites. Repeated satellite tagging of animals from a smaller region should reveal the speed with which space use changes. Once again, usage and uncertainty estimates obtained with our methods could be used to test for the significance of annual changes in space use.

Further collection of satellite data is required for the west coast of Scotland to reduce uncertainty in the estimates of usage to the same levels as for the east coast. Repeat aerial or land-based surveys of haul-outs would allow us to measure variance in the haul-out counts directly from the data. Targeted tagging of individuals younger than 2 years would enable us to test the null hypothesis that they use space in the same way as adults. Finally, spatially focused, replicated tag deployments over several years would enable us to quantify the time scale over which the pseudo-equilibrium assumption is valid.

A large proportion of the world-wide population of grey seals is concentrated in the north of the British Isles. With this study, we have demonstrated that even within this geographical region the offshore distribution of grey seals is characterized by clearly identifiable hot-spots. This result is a vital input to the ongoing discussion about the establishment of marine SAC. Further, these hot-spots are not entirely the result of restrictions in accessibility, as a comparison of Figs 6 and 5 readily demonstrates. However, this is not to say that accessibility is unimportant. Recommendations for the placement of marine installations such as fish farms and the setting aside of protected areas should take proximity to seal haul-outs into account. Clearly, locations near to major haul-outs are much more likely to be utilized by seals than more distant areas, all else being equal. The wide range of prey types exploited by seals and the seals’ ability to move about rapidly and with direction means that they will encounter accessible areas frequently and are adaptable enough to take advantage of resources where they find them. Therefore both hot spots and terrestrial haul-outs need to be considered for management. Although the objective of the present study was to describe where seals go and not why they go there, relating the use of space by seals with biotic and abiotic environmental covariates should form a high priority for the immediate future.

We have developed this approach for grey seals but it is in no way specific to this species. In fact, our methodology would be ideally suited to most of the data sets on animal movement (see References above) that have, so far, only been used for inferences at the level of the individual.

conclusion

The primary aim of this study was to present our current best estimate of grey seal spatial usage. It has also served the twin purpose of establishing and illustrating suitable methodology for estimating the spatial distribution of animals such as grey seals, and revealing those gaps in our data that must be high priority for future data collection. Its limitations are characteristic of a study that attempts to generate spatially extensive maps of usage from a patchwork of data, but it is important to note that most of these limitations can be readily resolved with further, more focused data collection.

Acknowledgements

We would like to thank Mark Hindell, Edward Gregr, John Harwood, Phil Lovell, Geert Aarts, an anonymous referee and the journal's editorial team for their constructive comments. Data collection for this paper has been funded by a variety of sources: primarily the EU, the UK Natural Environment Research Council and the UK Ministry of Agriculture Fisheries and Food. The analytical work in this paper was carried out under project number MF0311 of the UK Ministry of Agriculture Fisheries and Food and project number MF0320 of the UK Department for the Environment, Food and Rural Affairs.

Supplementary material

The following material is available from http://www.blackwellpublishing.com/products/journals/suppmat/JPE/JPE911/JPE911sm.htm

Appendix S1. Estimating the number of animals associated with each haul-out.

Appendix S2. Scaling spatial usage and local uncertainty by the number of animals in each haul-out.

Ancillary