Sub-surface hotspots in shallow seas: fine-scale limited locations of top predator foraging habitat indicated by tidal mixing and sub-surface chlorophyll
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
MARINE ECOLOGY PROGRESS SERIESMar Ecol Prog Ser
Vol. 408: 207–226, 2010doi: 10.3354/meps08552
Published June 3
INTRODUCTION
The foraging locations of marine piscivorous preda-tors may be extremely limited spatially and temporally.Evidence increasingly indicates that marine mammalsand seabirds use not only a small set of locations, butalso a limited range of tidal conditions in which to cap-
ture their fish prey (Uda 1952, Irons 1998, Simard et al.2002, Johnston et al. 2005a,b, Bertrand et al. 2008, Ste-vick et al. 2008). The ecological implications are thatthe successful capture of prey by apex predatorsrequires specific environmental conditions at a limitednumber of suitable locations; predators must be at ‘theright place at the right time’.
Sub-surface hotspots in shallow seas: fine-scale limitedlocations of top predator foraging habitat indicated by
tidal mixing and sub-surface chlorophyll
B. E. Scott1,*, J. Sharples2, O. N. Ross3, J. Wang1, G. J. Pierce1, 4, C. J. Camphuysen5
1School of Biological Sciences, Institute of Biological and Environmental Sciences University of Aberdeen,Tillydrone Avenue, Aberdeen, AB24 2TZ, UK
2Proudman Oceanographic Laboratory, Joseph Proudman Building, 6 Brownlow Street, University of Liverpool, Liverpool, L3 5DA, UK
3Mediterranean Centre for Marine and Environmental Research, (CMIMA, CSIC), Marine Technology Unit (UTM), Pg. Marítim de la Barceloneta, 37-49, 08003 Barcelona, Spain
4Instituto Español de Oceanografía, Centro Oceanográfico de Vigo, PO Box 1552, 36200 Vigo, España5Royal Netherlands Institute for Sea Research, PO Box 59, 1790 AB Den Burg, Texel, The Netherlands
ABSTRACT: The foraging habitats of 7 species of marine apex predators were observed simultane-ously in a shallow sea, with continuous measurements taken of the detailed bio-physical watercolumn characteristics to determine habitat preferences. We found the occurrence of small-scale‘hotspots’, where 50% of all animals were actively foraging in less than 5% of the 1000 km of tran-sects surveyed. By investigating a contrasting range of foraging strategies across a variety of fish-eating seabirds and marine mammals, we determined which habitat characteristics were consistentlyimportant across species. A static habitat variable, tidal stratification, log10(h/U 3) (h = water depth,U = tidal current amplitude), was found to be the best indicator of the probability of presence andabundance of individual species. All 7 mobile top-predators preferentially foraged within habitatswith small-scale (2 to 10 km) patches having (1) high concentrations of chlorophyll in the sub-surfacechlorophyll maximum (CHLmax) and (2) high variance in bottom topography, with different speciespreferring to forage in different locations within these habitats. Patchiness of CHLmax was not associ-ated with the locations of strong horizontal temperature gradients (fronts) or high surface chlorophyllvalues, but instead may be related to areas of high sub-surface primary production due to locallyincreased vertical mixing. These small-scale areas represent a newly identified class of spatiallyimportant location that may play a critical role within the trophic coupling of shallow seas. Such sub-surface hotspots may represent the limited locations where the majority of predator-prey interactionsoccur, despite making up only a small percentage of the marine environment.
Resale or republication not permitted without written consent of the publisher
Mar Ecol Prog Ser 408: 207–226, 2010
Review studies on the upwelling systems of thePacific Ocean identified spatially and temporally pre-dictable ‘biological hotspots’ with distinct surface sig-natures (Spear et al. 2001, Bakun 2006, Ballance et al.2006, Sydeman et al. 2006). At this large scale, themechanistic evidence behind the creation of hotspotsfor marine predator foraging points directly to bothtopographical features (Genin 2004, Yen et al. 2004)and primary productivity (Ware & Thomson 2005, Bostet al. 2009). The general conclusion of many large-scale studies is that seabirds and marine mammals arefound preferentially foraging within different types offrontal region: areas of intersection between differentwater mass types where there are steep surface (hori-zontal) gradients in water density. Biological reasoninglinks foraging to frontal locations via the elevatedlevels of primary production and aggregation of plank-tonic organisms found at fronts (Pingree et al. 1975,Franks & Chen 1996, Durazo et al. 1998, Russell et al.1999, Lough and Manning 2001). However, some ofthe studies mentioned above stress the point thatnot all marine mammals and seabirds forage preferen-tially at the fronts (Ballance et al. 2006, Spear et al.2001).
Sea surface temperature (SST) fronts and surfacecolour (calibrated for chlorophyll biomass) from satel-lite imagery are now widely used as proxies for loca-tions of productivity and in the identification of poten-tial foraging areas for marine predators over globalscales (Polovina et al. 2001, Worm et al. 2005). How-ever, a study off the west coast of North America hasshown that SST is not always a reliable predictor of for-aging habitat across a wide range of seabird species(Burger 2003). In this paper, we address the possibilitythat some important areas of foraging do not haveoceanographic surface signatures, but are insteadassociated with (less observationally convenient) sub-surface processes.
Understanding the links between a predator and itsprey is challenging. We need to appreciate that preda-tor-prey interactions may occur on a time scale of min-utes and on spatial scales of tens of metres or less. Thefew studies directed at these smaller scales have beenconducted in areas with very strong tidal currents.Tidal forcing has been found to be the main cause ofprey aggregation at such locations (Decker & Hunt1996, Mendes et al. 2002, Cotté & Simard 2005, John-ston et al. 2005a,b) and primary production has alsobeen implicated as playing an important role in preda-tor distribution (Ladd et al. 2005, Sinclair et al. 2005).Two recent studies, not within tidally active areas,have investigated predator-prey interactions across arange of trophic levels at the appropriate spatial andtemporal scales (Bertrand et al. 2008, Stevick et al.2008). In both cases, internal wave activity was indi-
cated as a potential mechanism for increased predator-prey interaction; a mechanism that our study alsopoints towards. However, these studies were based onlocations where there were large changes in topo-graphy (>100 m differences), located near shelf edgesand which had obvious surface features (temperaturegradients, large surface slicks) defining the regions ofpredictable predator foraging.
In this study, we worked in a shallow sea region(<200 m) with limited tidal speeds (<1.0 m s–1) andminimal differences in topographical features (<30 mdifferences in bottom depth). We define the foraginghabitats for a range of top predators by collecting con-tinuous and simultaneous information on both (1) thedistribution of foraging animals and (2) detailed hori-zontal and vertical structure of the water column inwhich animals were actively foraging. Our approachwas to collect and compare information from 7 specieswith contrasting foraging behaviours (common guille-mot, northern gannet, black-legged kittiwake, minkewhale, harbour porpoise, white-beaked dolphin andgrey seal) and to explore not only the expected differ-ences in foraging habitats, but also to search for anysimilarities across species. We present evidence for theextreme patchiness of foraging hotspots, well awayfrom frontal locations or from regions with high surfacechlorophyll biomass. These findings have importantconsequences for defining top-predator foraging habi-tat within shallow seas.
MATERIALS AND METHODS
Study area. The study area, covering approximately100 × 100 km, is in the North Sea off the east coast ofScotland, UK. Eight transects across the area, approxi-mately 10 km apart, were carried out from 8 to 19 June2003 (see Fig. 1) on a 66 m Netherlands Institute forSea Research vessel, the ‘Pelagia’. The direction oftransects alternated between E–W and W–E and dailystarting locations were driven only by the location ofstoppage the day before, such that time of day andlocation were randomly sampled. Trained observerscontinuously recorded the abundance and behaviourof seabirds and marine mammals from a platformapproximately 20 m above sea level. The resultingindices of animal presence, absence and abundanceare used as response variables: i.e. those variables forwhich we wish to explain the variation in distributionin terms of oceanographic factors. The biophysicalcharacteristics of the entire water column were contin-uously sampled via an undulating ScanFish (MKII1250, EIVA) carrying a Sea-Bird 911 CTD and ChelseaInstruments Aquatracka MKIII chlorophyll fluorometer.Oceanographic characteristics are used as the explana-
Scott et al.: Sub-surface hotspots in shallow seas 209
tory variables in determining the defining habitat char-acteristics for individual species, as well as for iden-tifying common habitat variables for a group of 7representative top-predator species.
Census of seabirds and marine mammals. The strip-transect techniques deployed to count seabirds andmarine mammals were those that have been devel-oped as the standard for ship-based seabird surveys inthe North Sea (Tasker et al. 1984, Buckland et al. 2001,Camphuysen et al. 2004). Seabirds and marine mam-mals were counted in 5 min intervals within a strip300 m ahead and 300 m to one side of the vessel (theside being chosen to select the best light conditions).A minimum of 2 observers were active at all times, witha third observer normally accompanying most obser-vations. The use of 5 different distance bands withinand just beyond the strip allowed for corrections formissed birds and cetaceans; ranges are referred to as A(0–50 m), B (50–100 m), C (100–200 m), D (200–300 m)and E (beyond 300 m and outside transect). The shiptravelled at a constant speed of approximately 8 knots(4.1 m s–1), such that each 5 min interval covered anaverage of 1.24 km. In addition to the standard tech-niques, and to be able to discriminate between feedingor foraging birds from non-feeding individuals (e.g.Ashmole 1971), 20 types of feeding behaviour and 16types of non-feeding behaviour were recorded (detailsin Camphuysen & Garthe 2004: see our Table 1). Birdsassociated with, or apparently attracted by, the re-search vessel were not used for calculations.
Seven representative species were chosen from theobserved range of 27 species for use in this study,based on the most abundant seabird and marine mam-mal species and representing a range of differentforaging strategies. The 7 species were the commonguillemot Uria aalge, northern gannet Morus bas-sanus, black-legged kittiwake Rissa tridactyla, minkewhale Balaenoptera acutorostrata, harbour porpoisePhocoena phocoena, white-beaked dolphin Lageno-rhynchus albirostris and the grey seal Halichoerus gry-pus (see Table 1 for a description of their different for-aging methods).
The abundance and presence of animals in each5 min bin was used to explore explanatory habitat vari-ables for each individual species. In order to investi-gate possible common foraging habitat preferences,the 7 species were also grouped together under 2separate definitions of total abundance: the first indexof abundance was simply the total number of animalsseen foraging per 5 min bin; the second index of abun-dance included a transformation of animal counts formore conservative multi-species comparisons. Smaller,more numerous animals can potentially dominate abun-dance estimates. However, larger animals, althoughmuch less numerous, require a proportionally greater
prey abundance and therefore represent a greaterforaging presence. To begin to correct for this size/number bias, the number of individuals per speciesand per observation was multiplied by the body massof that species. The mean species mass used for thistransformation of animal counts is averaged betweenmales and females (where available) and between thelower and higher published values (see Table 1). Dueto the order of magnitude difference in size betweenmarine mammals and seabirds, the log10 of the meanweight (–w ) × the abundance (At +1) of each species wascalculated, summed for each 5 min bin and used as anindex of total animal biomass abundance AT:
(1)
where Δt = 5 min. This index of total biomass abun-dance decreases the influence of numerous small ani-mals of a single species. The transformation alsoincreases the importance of multi-species presence inany one 5 min bin observation.
Physical and biological oceanographic variablesfrom ScanFish data. The sampling of physical featuresof the water column was carried out with the ScanFish,such that continuous vertical and horizontal informa-tion on temperature, salinity, density and fluorescence(a proxy for the abundance of chlorophyll) was col-lected to within 2 to 5 m of both the sea bed and thesurface. Data were sampled at 1.0 s intervals, yieldinga vertical resolution of between 0.5 to 1.0 m. With themaximum depth of the study area being less than 90 mand the speed of towing being a constant 8 knots(4.1 m s–1), the horizontal distance between the mid-point of up and down casts of the ScanFish was nevermore than 400 m (generally ranging between 200 and300 m). In order to compare the continuous physicalwater column characteristics measured by the Scan-Fish to the 5 min bin observations of visible top-preda-tors, summaries of physical and biological characteris-tics of the water column were created for the sameobservational 5 min bins. Most 5 min bins represent ahorizontal travel distance of approximately 1.24 km.There is a small level of variation in the distance trav-elled every 5 min due to slight variations in ship speed,but this does not affect the analysis or results becausethese variations affect both physical data and seabirdcounts in the same way.
Degree of water column stratification: The differ-ence between surface and bottom temperature (ordensity) produces an index of how well the water col-umn is mixed or stratified (see Fig. 1). This assumptionbreaks down as we approach the mouth of the Firth ofForth, where the density stratification is mainly causedby differences in salinity. Stratification is a useful phys-ical proxy for a range of biological characteristics, such
A w AT tt t
t t
log ( )= +⎡⎣ ⎤⎦=
+
∑ 10 10
0 Δ
Mar Ecol Prog Ser 408: 207–226, 2010
as the likely community of phytoplankton and zoo-plankton species and the type of food web (i.e. shortversus microbial food webs: Cushing 1975). Away fromthe coast and regions of freshwater influence, the mainfactors forcing the strength of stratification are sea-sonal and daily variations in solar heating, wind-drivensurface mixing, and tidally-driven bottom mixing. Inorder to filter out much of the diurnal fluctuations instratification, we used the mean temperature and den-sity above and below the pycnocline over each 5 minbin to produce variables for thermal stratification (ΔT)and density stratification (Δρ). The location of the topand bottom of the pycnocline were defined as thedepths at which the vertical density gradient droppedbelow 0.01 kg m–4. We also produced estimates of theaverage temperature and density gradients (
–––ΔT/zp–––
,–––Δρ/zp
–––) across the width of the pycnocline (zp) for each
5 min bin.Chlorophyll concentrations: The voltage output
from the fluorometer on the ScanFish was used forall statistical analyses and generalised additive model(GAM) graphical output. The fluorometer output wascalibrated to chlorophyll concentration (mg m–3) usingchlorophyll samples taken over a representative rangeof depths and locations across the study area in theform of CHL = a × voltageb. This yielded a calibrationequation of: CHL [mg m–3] = 19.85 × voltage1.91 (SE:a = ± 4.677, b = ± 0.187; n = 70). Fluorometer chloro-phyll concentrations were averaged over all data col-lected within each 5 min bin to produce mean chloro-phyll levels per bin (
–––––CHL; see Fig. 2a). Values for
the sub-surface chlorophyll maximum (CHLmax): seeFig. 2b) were defined as the highest concentration ofchlorophyll within each 5 min bin of observations,regardless of where they occurred vertically in thewater column (note that all but 2 values were deeperthan 5 m and the median depth was 15.6 m). All ob-servations were taken during daylight hours (04:00 to20:00 h GMT).
Seastate and time of day: Rough sea state (Seastate)will decrease the chances of visual detection of marineanimals, and time of day (Time) can also affect thepresence, absence and abundance of many animals;both were therefore tested as explanatory variables.
Static oceanographic variables. The oceanographicvariables defined above are those that were collectedsimultaneously with the animal data. They have thepotential to vary due to local and seasonal forcing, andrepresent the state of the environment which hasresulted from recent meteorological and tidal variabil-ity. However, we also need to consider variables un-affected by inter-annual and shorter timescale vari-ability. This will allow the results from this study to becompared to other years and other bird and mammalforaging locations, as well as determining whether
predators are using a long-term geographical decisionrather than responding to short-term events. We there-fore produce static explanatory habitat variables thatare described below.
Potential for seasonal thermal stratification (h/U 33):In shallow seas (<200 m) the tendency of a water col-umn to thermally stratify can be quantified by the ratiobetween the total depth (h) and the cube of a measureof the tidal current amplitude (U ), h/U 3 (Simpson &Hunter 1974, Pingree & Griffiths 1978). Low values ofh/U 3 indicate areas where the water column is likely toremain vertically mixed all year, while high valuesoccur in areas that will thermally stratify duringsummer. Within stratifying regions, higher h/U 3 indi-cates stronger summer stratification. Values of whatwe have coined ‘Tidal stratification’, log10(h/U3), werecalculated over the whole study area using depthdata from the British Geological Society (BGS: www.bgs.ac.uk/products/digbath250/sample.html) and tidalvelocities from the POLPRED tidal prediction model(Proudman Oceanographic Laboratory, NERC, UK).The tidal velocities used are the mean monthly depth-mean tidal speeds for June 2003 and so represent aver-age tidal speeds over 2 spring-neap cycles.
Depth and topographical data: The following poten-tial habitat variables were derived from BGS data andextracted via GIS ArcView (ESRI) software derived atthe 750 m radius scale for the location of each 5 min binobservation: maximum depth of the seabed Hmax; meandepth of the seabed i
–H ; standard deviation of depth
SD(H ); maximum slope, (dH/ds)max (Slope). The BGSdata were used rather than the shipboard depth data,as the variables could be derived from information inall directions rather than just the E–W directions of thetransects.
Analysis. One-way ANOVA and Tukey: Simultane-ous multiple comparisons of the means of habitatvariables, from the presence/absence and abundancedata for the 7 species, were carried out using 1-wayANOVA and Tukey honestly significant difference(HSD) methods using S-Plus, version 7.0 (TIBCO).
Generalised additive models (GAMs): All potentialexplanatory variables were screened using histograms,dot plots (univariate), scatter plots (bivariate) and cal-culation of variance inflation factors (VIF) to determinedistributions, detect outliers and identify co-linearitybetween variables. Where 2 variables were stronglycollinear (r ≥ 0.8), one was excluded from furtheranalysis. This occurred with the stratification indicesfor temperature and density as well as for 2 indices forthermal and density gradients, confirming that stratifi-cation in the study area was dominated by sea surfaceheat fluxes. Therefore only one variable was selectedfor each characteristic. Surface to bottom temperaturedifference (ΔT) was used to represent stratification.
210
Scott et al.: Sub-surface hotspots in shallow seas
The density difference across the pycnocline was cho-sen to represent the indices for gradient (
–––Δρ/zp–––
). Themaximum depth (Hmax) and mean depth (
–H ) were
highly correlated, as were slope and the variation indepth. Maximum depth (Hmax) and depth variationSD(H ) were kept as the explanatory variables. SD(H )was log-transformed (log10[SD(H) +1]). Seastate wastreated as a continuous variable.
All response variables included a high proportionof zero values (absences) and so a 2-stage modellingstrategy was used (Zuur et al. 2007). Presence wasmodelled using binomial GAM. Within a total of 800observations, numbers of positive records (presence)for marine mammals ranged from 25 (white-beakeddolphins) to 35 (grey seal), with figures for seabirdsranging from 54 (gannet) to 363 (guillemot). For allmarine mammals, the proportion of presence records isless than 5% which limited our ability to fit satisfactorymodels. For subsets of non-zero values of responsevariables, abundance (given presence) was also mod-elled with a GAM, fitted using appropriate distribu-tions, normally a so-called quasi-Poisson (i.e. a Poissondistribution in which dispersion is not constrained,since the over-dispersion parameter for most modelswas >1). It should be noted that, for all the marinemammal species and for the gannet,observation sample sizes ranged from25 to 54; since 9 explanatory variableswere available, model results weretreated with caution.
Explanatory variables were selectedby repeated backward and forwardselection; starting with full models,dropping out the least significantterms sequentially and building upfrom models with single explana-tory variables. Cross-validation wasused to estimate degrees of freedomfor smoothers (the partial residual),although in all cases the maximumvalue was set to 3 (k = 4) to avoidover-fitting. The final model selectedin each case was that in which allterms were significant (p < 0.05) andwhich had the lowest value forAkaike’s information criterion (AIC).Usually all criteria (AIC, devianceexplained, significance of individualterms) were in agreement. Occasion-ally an explanatory variable wouldhave a weak but significant effect (p <0.05) but its inclusion would increasethe AIC score and reduce devianceexplained; it was therefore droppedfrom the final model. In some cases
no satisfactory fit could be achieved. All GAM analysiswas carried out using Brodgar 2.5.6 (Highland Statis-tics) linked to R 2.6.0 (R Development Core Team).
RESULTS
Physical and biological oceanographic data
The location of the study area within the North Seaand the strength of stratification are presented inFig. 1. There are obvious regions of higher stratifica-tion farther off the coast; however, the pattern is muchmore complex than a simple onshore/offshore divide.The northwestern corner of the study area, with theweakest stratification and highest tidal mixing, domi-nated as the location with the highest biomass of depth-averaged chlorophyll (Fig. 2a). However, the spatialpatterns of locations of CHLmax (Fig. 2b) only occur invery limited areas within the more stratified region.They range in size from approximately ~2 to 10 kmlinear distance along the transects. The vast majority(91%) of locations with CHLmax > 2.0 mg m–3 werefound well below the surface at depths >10 m (mediandepth of 15.6 m), indicating that their vertical locations
211
Δρ(kg m–3)
Fig. 1. Location of the study area in the North Seaand density stratification (Δ ρ) of the water column,shown here as the difference in density between5 m of the surface and the bottom. The diagramwas created using the continuous vertical temper-ature data collected by an undulating ScanFish(MKII) along 8 E–W transect lines (dashed lines)and is the equivalent of approximately 4000 sin-gle CTD casts. The data are continuous along thetransect lines, but have been linearly interpolated
to fill in the areas between the transect lines
Mar Ecol Prog Ser 408: 207–226, 2010
may be limited to within the thermocline. These loca-tions would not be detected by sea-surface satellitecolour imaging.
Values of log10(h/U3) for a larger region of the east-ern North Sea are shown in Fig. 3, with the study areamarked by the solid rectangle. A value of 2.75 m–2 s3
represents the locations of tidal fronts, separatingpermanently-mixed water from seasonally-stratifiedregions (Sharples 2008). Values between 2.3 and2.75 m–2 s3 indicate regions that can switch betweenbeing mixed and stratified, depending on the phase of
biweekly tidal currents; values between2.75 and 3.5 m–2 s3 are regions likely to seespring-neap impacts on sub-surface pri-mary production within the thermoclineand that always remain stratified insummer (Sharples 2008). This implies thatonly a small fraction of the study site hasareas (the far northwestern corner and asection at the centre of the southern sec-tion) that could be defined as frontal. Thevast majority of the study area was strati-fied (>1.0°C) and will always stratify in thesummer as a result of surface heatingovercoming mixing by tides. The 10 high-est
–––––CHL values were found within a
narrow range of 3.51 < log10(h/U 3) <3.59 whereas locations with CHLmax >3.0 mg m–3 occurred only in areaswith log10(h/U 3) > 3.55 m–2 s3. This sug-gests that these isolated locations of highCHLmax will occur in areas that canstrongly stratify and that their existencehas no link to fronts that are produced bytidal properties.
Spatially limited foraging
In total, 4847 animals of the 7 speciesin this study were observed foraging over993 km of transect line coverage (Table 1shows total numbers of each species ob-served). In 43% of all the 5 min bins sur-veyed, no animals were observed (Table 2).The majority of all animals (88%) werefound foraging in 25% of the 5 min binsalong our transect lines. These figures areinfluenced by the presence of smaller, butmuch more numerous, birds. However,even when the abundance of all animalswas viewed as a percentage of the totalweighted biomass abundance (Eq. 1),59% of foraging effort was still foundwithin 25% of all the 5 min bins surveyed.
The numbers of foraging animals observed werehighly clumped: 50% of them were found in very lim-ited locations, representing only 4.4% of the surveybins and containing at least 25 animals within any one5 min bin (Table 1). However, these same areas of veryhigh abundance (accounting for 50% of foraging ani-mals by number) only make up 13.9% of the totalweighted biomass abundance, suggesting the loca-tions of highest abundances contain mostly single spe-cies and/or have extremely high numbers of smalleranimals.
212
Fig. 2. Chlorophyll biomass values from continuous collection by aScanFish (MKII) along the 8 transects used. As with Fig. 1, the valuesbetween the transects have been linearly interpolated. (a) Averagechlorophyll biomass,
–––––CHL (mg m–2). (b) Maximum level of chlorophyll
biomass, CHLmax (mg m–3) found within the vertical water column within any one surface to bottom undulation of the ScanFish
Scott et al.: Sub-surface hotspots in shallow seas
Foraging habitat
Single variables: differences and similarities betweenspecies
In this section we will first present the foraging habitatfor each of the 7 species separately, then the foraginghabitat commonalities across species. Because thedifferent species of top-predators capture their preyusing different foraging behaviours, one would expectthat they may select different habitat types in which toforage. The mean values of the biological and physicalexplanatory variables at the locations in which each spe-cies was found (Table 3) indicate whether or not differentspecies were targeting different habitats. Six out of 9explanatory variables showed significant differencesbetween the mean values for different species. The 3variables which showed highly significant differences(p < 0.0001) between species were ΔT, log10(h/U 3) and–––––CHL. The 3 variables which did not show differencesbetween the species were CHLmax, SD(H ) and Time.
The location and abundance of each species ismapped on top of ΔT, CHLmax and log10 (h/U 3) for sea-bird (Fig. 4) and mammal species (Fig. 5). In general,
there appears to be a preferred range of thermalstratification, with a dramatic drop of both pres-ence and abundance for almost all species (exceptgannets and white-beaked dolphins: Figs. 4a &5a, respectively), in the highly stratified watersfarther offshore. This distribution might also berelated to the distance from colonies in the case ofseabirds and seals. The vast majority of foragingkittiwakes and minke whales were centred onlongitude 1.5° W, where the levels of CHLmax werehigh (Figs. 4b & 5b, respectively). Guillemotswere the most numerous species, with some obvi-ous clusters found throughout most of the studyarea except in regions with higher values oflog10(h/U 3) (Fig. 4c).
In Fig. 6, we provide an example of the verticalCHLmax characteristics of the water column andcompare the location values of the total weightedbiomass abundances of all species with CHLmax
and –––––CHL along the representative middle transect
(latitude: 56.25° N). Fig. 6 clearly shows that, incombination, species seemed to be targetingareas with high CHLmax where there was little orno difference in
–––––CHL.
We tested which species foraged in significantlydifferent habitats from each other for the 3 vari-ables which showed highly significant differencesbetween species [log10(h/U 3), ΔT,
–––––CHL] using
Tukey’s HSD (Fig. 7). Three pairs of species werefound to forage in areas with significantly differ-ent values for all 3 variables (at the 95% confi-
dence level): white-beaked dolphins versus grey seals,gannet versus guillemot and gannet versus porpoise.An additional 5 pairs of species were found to forage inareas of significantly different values for 2 of the vari-ables: white-beaked dolphins versus guillemot, white-beaked dolphins versus harbour porpoise, gannet ver-sus kittiwake, gannet versus grey seal and kittiwakeversus grey seal. The species pair which was found innearly exactly the same habitat across all 3 variableswas kittiwake and minke whale. Additionally, 2 spe-cies pairs were found to have 2 variables with the samehabitat value: white-beaked dolphins and gannet; por-poise and grey seal.
Comparing the mammals, harbour porpoises andgrey seals were using very similar habitats with signif-icantly lower values of log10 (h/U3) and ΔT but higher–––––CHL than for the minke whales and white-beaked dol-phins. Comparing birds, kittiwakes and guillemotsmainly shared the same habitat, with kittiwakes con-stantly in areas of higher log10(h/U 3) and ΔT but lower–––––CHL than guillemots. However, habitats used by bothkittiwakes and guillemots were significantly differentfrom those used by gannets, suggesting they do notforage in the more stratified water that only gannets
213
Fig. 3. Values of the tidal stratification variable, log10(h/U 3), cre-ated from bottom depth (h) and tidal speeds (U 3). Data for a muchlarger region is presented to put the study area (within the box)
into context
Mar Ecol Prog Ser 408: 207–226, 2010
appear to utilise. Comparing birds and mammals, kitti-wakes shared the same habitat as minke whales, butdid not overlap with harbour porpoises and overlappedonly slightly with grey seals. Guillemots overlappedwith all other species except gannets, suggesting thatthey are generalist in their habitat preference butprefer water that is less strongly stratified. Comparedto most other species except white-beaked dolphins,gannets foraged in water columns with a significantlyhigher log10(h/U 3) and ΔT, but a lower
–––––CHL.
The Tukey test results for ΔT and –––––CHL (Fig. 7b,c)
show very similar patterns of differences and similari-ties between all species, suggesting that all 3 of thesehabitat variables are similarly related to the habitatpreferences of each individual species. This is ex-pected, as the variables are related but occur over dif-ferent time scales, with large changes in
–––––CHL (weekly)
and ΔT (seasonally), whilst log10(h/U3) is a static vari-able. This suggests that log10(h/U3) and ΔT are ‘longterm’ variables (annual and seasonal) that underpinthe broad changes in distributional abundance, whilstlevels of
–––––CHL are predictably linked to these physical
parameters.The biological and physical variables that were not
significantly different across all species were CHLmax,SD(H ) and Time (Table 3). To investigate whether thelack of variation between the means for each specieswas due to a lack of variance in the variable across thestudy area, or if species were indeed targeting loca-tions where that variable was of a similar value, wetested the means for each variable in the locationswhere the species were present versus the locationswhere they were absent. We found that for bothCHLmax and SD(H ), the mean values were highly sig-nificantly different (t = –6.14 and t = –4.05, respec-
214
Table 1. Average weight (and data source), foraging method and number of animals observed during the survey for each of the 7 species
Species Avg. wt Foraging method List of foraging Number of Data source for Avg. wt(kg) behaviours animals seen
Guillemot 0.95 Deep diving bird Holding fish 3356 http://blx1.bto.org/birdfacts/indexa_short.htmUria aalge Pursuit diving
Minke whale 6500 Individual shallow Lunge-feeding (only 34 www.whalecenter.org/species.htmBalaenoptera diving whale surface activities acutorostrata could be logged)
Harbour porpoise 60 Group, fast speed None specifically 87 http://marinebio.org/species.asp?id=364Phocoena phocoena and pursuit
White-beaked dolphin 230 Group, fast speed and Herding (only 104 http://marinebio.org/species.asp?id=347Lagenorhynchus pursuit, but much larger surface activities albirostris than harbor porpoise could be logged)
Grey seal 190 Diving, pursuit with the None specifically 201 http://marinebio.org/species.asp?id=300Halichoerus grypus ability to use flippers to
manipulate fish/sand
Table 2. Percentages of the surveyed area and correspond-ing percentages of the total number of foraging animals andtotal weighted biomass abundance of foraging animals in
each category
Number of Percentage of Percentage of Total animals per 5 min bins foraging weighted 5 min bin surveyed animals biomass observation (%) (%) abundance (%)
Scott et al.: Sub-surface hotspots in shallow seas
tively: df = 968 and p < 0.001 for both relation-ships). Both variables showed higher valuesin the locations with animals present ratherthan absent: 1.56 compared to 1.33 mg m–3 forCHLmax and 0.66 compared to 0.34 m2 for SD(H ).The indication is that all animals were preferen-tially targeting areas with higher levels of CHLmax
and more abrupt changes in bottom topography.There was no significant difference in Time(of day), between locations where animals werepresent and where they were absent; this im-plies that the lack of differences between meansindicates similar variance across species, andthat we did not have a bias in our sampling ofTime (Table 3). In fact, all of the other 6 variablesalso showed no significant difference betweenlocations with animal presence and locationswith animal absence; what appears to be per-fectly good habitat locations in terms of theimportant values of ΔT, log10(h/U3) and
–––––CHL are
not used for foraging when the preferred valuesfor CHLmax and SDH are not present. This alsoimplies that those preferred parameter valuesare good indicators of fish prey being present.
Multiple explanatory variables for speciespresence and absence
Using GAMs, we explored how non-linearrelations for combinations of multiple physicaland biological variables might explain the pres-ence/absence patterns that were observed foreach species and for the combined species vari-able. All 7 species had viable presence/absencemodels. The most parsimonious models with thelowest AIC scores and only those variables withp < 0.05 are shown in Table 4. In general, themodels did not explain much of the variation inthe presence or absence of species. The guille-mot model has the highest percentage of de-viance explained (40.1%),with grey seals (25.7%)and harbour porpoise (22.1%) the next highest.The model for the weighted biomass abundanceof all animals explained 32.6% of deviance, with7 out of 9 explanatory variables being signifi-cant.
The 4 significant variables that were mostcommon across the species were log10(h/U 3),Time, Seastate and SD(H ). Different speciesshowed quite different non-linear relationshipswith log10(h/U 3) as well as Time (Fig. 8). Guille-mots and kittiwakes showed a minimum proba-bility of presence at the same value of log10
(h/U 3) (3.75 m–2 s3), but probability of presence
215
Tab
le 3
. Mea
n (±
SE
) of p
hys
ical
an
d b
iolo
gic
al s
um
mar
y va
riab
les
from
the
loca
tion
s of
eac
h 5
min
bin
for
each
sp
ecie
s. A
con
serv
ativ
e m
ean
is c
alcu
late
d u
sin
g th
e n
um
ber
of o
bse
rvat
ion
s (n
) an
d n
ot w
eig
hte
d b
y th
e n
um
ber
of
ind
ivid
ual
an
imal
s w
ith
in e
ach
ob
serv
atio
n.
Th
e la
st 2
row
s p
rese
nt
the
resu
lts
of m
ean
s te
sts
for
dif
fere
nce
s b
etw
een
eac
h s
pec
ies
as p
rese
nte
d in
th
e ta
ble
an
d b
etw
een
th
e su
rvey
bin
s w
ith
pre
sen
ce v
ersu
s ab
sen
ce f
or a
ll o
bse
rvat
ion
s of
sp
ecie
s co
mb
ined
Sp
ecie
s T
her
mal
T
idal
str
atifi
cati
onD
ensi
ty
Dep
th a
vera
ged
M
axim
um
M
axim
um
V
aria
tion
T
ime
Sea
stat
est
rati
fica
tion
log
10(h
/U3 )
gra
die
nt
chlo
rop
hyl
lch
loro
ph
yll
dep
thof
dep
thof
day
(1–
6)ΔT
(°C
)(m
–2
s3 )Δρ
(kg
m–
4 )––
–––
CH
L(m
g m
–2 )
CH
Lm
ax(m
g m
–3 )
Hm
ax(m
)S
D(H
)(m
)(%
of
24 h
)
Har
bou
r p
orp
oise
2.55
±0.
153.
56 ±
0.03
0.04
8 ±
0.00
40.
45 ±
0.04
1.53
±0.
1463
±1.
951.
03 ±
0.30
0.50
±0.
032.
00 ±
0.21
(n =
30)
Gu
ille
mot
2.97
±0.
043.
61 ±
0.01
0.05
5 ±
0.00
10.
38 ±
0.02
1.60
±0.
1061
±0.
550.
68 ±
0.05
0.49
±0.
012.
88 ±
0.08
(n =
363
)
Gre
y se
al3.
00 ±
0.18
3.55
±0.
010.
053
±0.
004
0.50
±0.
071.
71 ±
0.17
57 ±
1.86
0.79
±0.
250.
46 ±
0.03
2.23
±0.
28(n
= 3
5)
Kit
tiw
ake
3.21
±0.
073.
64 ±
0.01
0.05
9 ±
0.00
20.
27 ±
0.02
1.51
±0.
1662
±1.
030.
61 ±
0.07
0.49
±0.
022.
64 ±
0.13
(n =
89)
Min
ke
wh
ale
3.29
±0.
203.
66 ±
0.03
0.05
2 ±
0.00
30.
26 ±
0.03
1.53
±0.
1565
±2.
260.
55 ±
0.09
0.46
±0.
041.
89 ±
0.22
(n =
27)
Gan
net
3.75
±0.
123.
72 ±
0.02
0.05
7 ±
0.00
20.
11 ±
0.02
1.43
±0.
2267
±1.
620.
52 ±
0.07
0.49
±0.
032.
77 ±
0.18
(n =
54)
Wh
ite
bea
ked
3.
75 ±
0.22
3.67
±0.
030.
060
±0.
003
0.10
±0.
041.
28 ±
0.33
66 ±
2.06
0.39
±0.
110.
49 ±
0.04
2.83
±0.
29d
olp
hin
(n
= 2
5)
AN
OV
A
p <
0.0
001
p <
0.0
001
p <
0.0
1p
< 0
.000
1n
sp
< 0
.001
ns
ns
p <
0.0
01(b
etw
een
sp
ecie
s)
AN
OV
A (
bet
wee
n
ns
ns
ns
ns
p <
0.0
01n
sp
< 0
.001
ns
ns
area
s of
pre
sen
ce
vs. a
bse
nce
)
Mar Ecol Prog Ser 408: 207–226, 2010
for gannets showed a positive linear increase withlog10(h/U 3) and grey seals showed an optimal levelbetween 3.5 and 3.6 m–2 s3. Kittiwakes and gannetswere similar for Time, with probability of presencemaxima centred on 0.35 (08:40 h GMT) and minimumsat 0.6 (14:40 h), indicating a 6-hourly rise andfall of abundance; guillemot presence increased to anasymptote at 0.4 (09:30 h) and harbour porpoiseshowed a clear optimum at 0.5 (midday: 12:00 h).Seastate and SD(H ) are not shown, as the shape of therelationship with Seastate was consistent and essen-tially negative linear, whilst SD(H ) was positive linear.The less commonly represented variables were ΔT,–––––CHL and Hmax; however, all of these variables had sim-ilar relationships across species with viable models(Fig. 8).
Multiple explanatory variables for species abundancegiven presence
Using the second step in the GAM approach, weexplored how the combination of multiple physicaland biological variables might explain the abun-dance of species within those areas where they arepresent. Six of the 7 species had viable models.Again, only the most parsimonious models with thelowest AIC scores and those variables with p < 0.05are shown (Table 5). The grey seal model has thehighest percentage of deviance explained (96.8%).The next best model was for gannets (83.8%) andthen kittiwakes (54.2%). The model with all speciescombined was a poor model overall, explaining only17.1% of deviance, suggesting that high abundances
216
Fig. 4. Numbers of observed forag-ing seabirds on a background of 3habitat characteristics of the watercolumn. The values of the habitatvariables are displayed using a lin-ear interpolation of summary statis-tics of the 5 min bin observations.(a) Thermal stratification (ΔT) (b)sub-surface chlorophyll maximumCHLmax, (c) tidal mixing log10(h/U 3).For clarity, the legend symbols are2 × the magnitude of the symbols
within the Figs
Scott et al.: Sub-surface hotspots in shallow seas
of individual species are independent of the otherspecies.
The variables that were most common across thespecies (present in 4 models) were again log10(h/U 3)and, this time, CHLmax. The same species have sig-nificant relationships with the explanatory variable,log10(h/U 3), however the shape of the relationshipsare quite different from the presence/absence results(Fig. 9). Guillemots show a distinct drop off in abun-dance at a value of 3.75 m–2 s3, with rather stable levelsat lower values suggesting a strong habitat preferencefor areas that are more mixed. Kittiwakes show a moredistinct optimum between 3.5 and 3.6 m–2 s3. Gannetsshow a distinct increase in abundance from values of3.8 m–2 s3 and upwards. Grey seals show a gradualincrease in abundance with increasing values up to
3.7 m–2 s3, but were not found at all above that value.Kittiwakes and white-beaked dolphins show the high-est abundances at lower levels of CHLmax, whereasgannets show a clear increase up to an asymptoticvalue of 0.25 (1.4 mg m–3) and grey seals have a clearoptimum at just over 0.3 (2.0 mg m–3).
–––––CHL, ΔT, Hmax and SD(H ) are all significant in 2 spe-
cies models each (Fig. 9). Both guillemots and kitti-wakes show optimum values for
–––––CHL of 0.20 (0.92 mg
m–3) and 0.16 (0.60 mg m–3), respectively. Gannetsshow a clear optimum for ΔT at 4.0°C and kittiwakesshow higher abundance at ΔT values lower than 4.0°C.Grey seals obviously prefer shallower habitat, showinghigh increases in abundance in water less than 60 mdeep; kittiwakes show optima between 50 and 60 m.The kittiwake SD(H ) model shows increased abun-
217
Fig. 5. Numbers of observed forag-ing mammals on a background of 3habitat characteristics of the water
column. Other details as in Fig. 4
Mar Ecol Prog Ser 408: 207–226, 2010
dance at values of 0.2, which translates to ±3.2 m(±2 SD log transformed data), whereas the gannetmodel shows a linear increase in abundance with in-creasing variation in bottom depth.
DISCUSSION
Spatially limited foraging locations
In this study, we simultaneously analysed a com-bined range of actively foraging marine top-predators,with contrasting foraging behaviours, in relation tosimultaneous and continuous oceanographic data in ashallow sea region (<200 m). The implications of thisstudy are that the foraging locations for a range ofspecies are extremely limited in space. We found 50%of all animals foraging in very patchy distributions(<2 km), but with high foraging abundances (>25 indi-viduals, whereas the mean was only 6 individuals per5 min survey bin) concentrated in only 4.4% of the
area surveyed (Table 2, Figs. 4 & 5). These high-abun-dance locations were found to be species-specific, asthe weighted biomass abundance value for these areaswas only 14% of the total; although there were highnumbers of animals present within any one 1.24 kmobservational 5 min bin unit, they were composedmainly of a single species.
We have also been able to identify both the im-portant differences and similarities between speciesover a range of foraging location habitat values, as the~10 000 km2 study area contained a wide range of thestratification characteristics typical of shallow seas.Due to the time of year of the survey, the entire areawas stratified (Fig. 1). The locations of high integratedchlorophyll biomass and locations of CHLmax werequite dissimilar, with locations of high CHLmax beingextremely patchy at scales of approximately 2 to 10 km(Fig. 2). Interestingly, areas of high CHLmax were onlyfound in locations with values of log10(h/U3) > 3.55 m–2
s3, suggesting that these types of patches will only befound in more strongly stratified water columns,
218
Fig. 6. All foraging animal species, represented as the total weighted biomass abundance as compared to both the sub-surfacechlorophyll maximum, CHLmax, and the depth averaged chlorophyll biomass,
–––––CHL, in the upper panel. This is compared to the
output of the continuous chlorophyll biomass recorded continuously by the undulating ScanFish for a representative transect of the survey
Scott et al.: Sub-surface hotspots in shallow seas
whereas areas of highest –––––CHL are only within
regions where log10(h/U3) is <3.55 m–2 s3, suggest-ing that the value of 3.55 m–2 s3 has some biologicalsignificance. Also, there is a definite lack of bothCHLmax patches and top predators between valuesof 3.70 and 3.80 m–2 s3 even though there are highlevels of both with values between 3.55 and3.70 m–2 s3 and also >3.80 m–2 s3, suggesting thatthis level of stratification is not conducive to thecreation of foraging areas. Alternatively, it couldjust be coincidence within this study region.
Foraging habitats: differences
Static value: tidal stratification as an indicator offoraging location
A main conclusion of this study is that that the toppredators studied here are more likely to forage indifferent locations, defined to some extent by thelevel of stratification. log10 (h/U3) is an inverse mea-sure of tidal mixing normalised by the water depth,and was shown to be significantly different be-tween species (Table 3, Figs. 4, 5 & 6) as well asbeing the most consistent variable in most pres-ence/absence and abundance models (Tables 4 &5, Figs. 8 & 9). The results suggest that gannets andwhite-beaked dolphins prefer to forage in signifi-cantly more stratified regions than other species.Kittiwakes and minke whales seem to prefer mod-erately stratified areas but the whales generallyused different habitat from other mammals. Guille-mots, grey seals and harbour porpoises preferredareas that were much less stratified than the otherspecies.
Tidal stratification, log10(h/U3), is a very usefulstatic variable as not only does it combine depthand tidal current values into one variable, but alsoboth types of data are readily available for mostshallow sea regions. Testing the foraging locationpreference, defined by log10(h/U 3), for differentspecies can readily be accomplished through the
Fig. 7. Simultaneous pair-wise mean difference and 95%confidence limits (Tukey method) between (a) log10(h/U3),(b) ΔT and (c)
–––––CHL values for tidal mixing for each combi-
nation of foraging species. The dot represents the differ-ence between the means of each 2 pairs of species, withpositive differences indicating that the species on the right(on the y-axis) forages in water masses with higher vari-able values than the species on the left (vice versa for neg-ative values). Dotted lines and brackets represent the 95%confidence limits; for those species which are in signifi-cantly different water mass types, the 95% confidence
limits do not intercept the zero axes
Mar Ecol Prog Ser 408: 207–226, 2010
use of long-term survey data held at manynational agencies. The level of stratification a spe-cies prefers may also help to enlighten studies onmechanistic links between any observed changesin foraging behaviours, distributions, climate (sea-sonal stratification and wind mixing effects) andannual survival.
Different species also used foraging habitatswith significant differences in the seasonal vari-ables ΔT and
–––––CHL. The absolute values for ΔT and
–––––CHL will be affected by annual differences in thestrength and timing of winds, heat input andnutrient concentrations, making it more difficultto obtain the spatial values for these variableswithout field sampling. However, both variablesshowed approximately the same pattern betweenspecies as found for log10(h/U3) in the Tukeyanalysis (Fig. 6). This is logical as log10(h/U3)defines the area’s tendency to stratify or verticallymix, which influences
–––––CHL values. Importantly
though, in the additive models where these vari-ables were significant, whether in the presence/absence or the abundance models, the relation-ships were generally one with an optimum value(Figs. 8 & 9). Together these results suggest that,overall, log10(h/U3) underpins the broad differ-ences in species foraging distributions, but thatseasonal differences in ΔT and
–––––CHL drive the
smaller scale factors in foraging locations.
Daily variables: Seastate and Time
The other habitat variables that played signifi-cant roles in explaining the variation in the pres-ence/absence of animals were Seastate and Time.However, neither variable had much of a role inaccounting for abundance. Both of these are vari-ables that change on an hourly scale, with Seast-ate consistently showing a negative linear relationwith presence, as expected (Table 4 & 5). Al-though Time showed no significant linear differ-ences between species (Table 3), there are differ-ences in the non-linear shape of relationships(additive models, Fig. 8) for the presence of 4 spe-cies. The probability of presence for all the seabirdspecies builds from the early morning to a maxi-mum at 08:40 h (GMT). This suggests that theincrease in birds over time is most likely due tobreeders that have flown out from a nest site in the
220
Tab
le 4
. Pre
sen
ce/A
bse
nce
GA
M: T
he
opti
mal
mod
els
wer
e se
lect
ed b
ased
on
ind
ivid
ual
sig
nifi
can
ce o
f ex
pla
nat
ory
vari
able
s. V
alu
es s
how
n in
dic
ate
the
deg
rees
of
free
-d
om (
df)
for
sm
ooth
ers
and
lin
ear
slop
es (
S)
for
lin
ear
term
s, p
lus
the
sig
nifi
can
ce v
alu
es (
p)
for
all e
xpla
nat
ory
vari
able
s re
mai
nin
g in
th
e fi
nal
mod
els.
For
eac
h fi
nal
mod
el,
% d
evia
nce
exp
lain
ed, A
IC a
nd
nu
mb
er o
f 5
min
bin
s (N
) ar
e al
so g
iven
. SD
(H)
was
log
-tra
nsf
orm
ed
Sp
ecie
sΔT
log
10(h
/U3 )
Δρ––
–––
CH
LC
HL
max
Hm
axS
D(H
)T
ime
Sea
stat
eD
evia
nce
A
ICN
(% o
f 24
h)
(1–
6)ex
pla
ined
(%
)
Har
bou
r d
f=
2.19
S=
2.02
df
=2.
25S
=–
0.43
22.1
199.
3880
0p
orp
oise
p<
0.0
01p
=0.
04p
=0.
01p
=0.
02
Gu
ille
mot
df
=2.
10d
f=
2.52
df
=2.
45d
f=
2.66
S=
2.28
df
=2.
7340
.168
9.62
800
p<
0.0
001
p<
0.0
001
p<
0.0
01p
< 0
.001
p<
0.0
01p
< 0
.000
1
Gre
y se
ald
f=
2.25
df
=2.
86d
f=
1.75
25.7
229.
2680
0p
=0.
01p
< 0
.001
p<
0.0
01
dfi
ttiw
ake
df
=2.
31d
f=
2.66
df
=2.
22d
f=
2.98
df
=2.
7816
686.
3680
0p
< 0
.000
1p
=0.
04p
=0.
01p
< 0
.000
1p
< 0
.000
1
Min
ke
S=
–0.
667.
622
2.09
800
wh
ale
p=
0.00
01
Gan
net
S=
3.89
df
=2.
774.
638
6.58
800
p<
0.0
01p
=0.
04
Wh
ite
df
=2.
815.
921
0.17
794
bea
ked
dol
ph
inp
=0.
02
All
an
imal
sd
f=
2.31
df
=2.
52d
f=
1.99
df
=2.
02d
f=
2.44
S=
3.79
df
=2.
8632
.676
9.95
800
p=
0.01
p<
0.0
01p
=0.
01p
=0.
03p
< 0
.001
p<
0.0
001
p<
0.0
001
Scott et al.: Sub-surface hotspots in shallow seas 221
Fig. 8. GAM relationships (smoothing spline of the partial residual and 95% confidence intervals) for the presence/absence of 5 significantexplanatory variables, log10 (h/U3), Time, ΔT,
–––––CHL, Hmax. In each row, the same partial residual for an explanatory variable is presented as
the y-axis. The values on the y-axis indicated either an increased probability of presence (>0) or absence (<0). For each row, the x-axis hasthe same explanatory variable. Each column contains the same marine top-predator species where possible. Note that Seastate and SD(H)
are also significant explanatory variables, but they are not shown as they are linear relationships
Mar Ecol Prog Ser 408: 207–226, 2010
early morning. However, for kittiwakes andgannets, there is also a definite 6 h pattern tothe probability of presence, suggesting thattheir foraging behaviour is linked to changes intidal currents. Kittiwake foraging was indeedfound to be more frequent at certain tidal speedsin this study area (C. E. Embling et al. unpubl.).
These results imply that even though we occa-sionally witnessed spectacular feeding frenzieswith multiple foraging species, the usual forag-ing behaviour for most species is to specialise inseparate locations from other species. Some ofthis degree of separation between species maybe directly linked to negative interaction, suchas aggressive behaviour between dolphins andharbour porpoises (Ross & Wilson 1996). Indeed,our motivation for creating a response variablethat was a weighted index of total biomass abun-dance of all predators was to explore whethercombining species increased the predictabilityof locations of presence or high abundance. Thefact that combining species as if they were a ‘su-per predator’ did not do well at explaining muchof the variation in these factors lends moreweight to the suggestion that predators forageseparately most of the time. This strongly sug-gests that the range of top predators studied areeither more efficient at catching the same preyin slightly different habitats, which can be de-fined by log10(h/U 3), or that they are targetingdifferent prey fish species which occur in thesedifferent habitats. Indeed, studies of minkewhales and harbour porpoise foraging at finespatial scales have also found that different spe-cies may use the same oceanographic feature(e.g. an island wake), but with different speciesusing different aspects of that feature (Johnstonet al. 2005a,b). The times of overlap betweenmultiple species foraging events are spectacularto view, and indeed many a dramatic naturedocumentary has been produced that focusesjust on these events. However, our data stronglysuggest that these events are rare in this geo-graphic region and that most species will haveseparate foraging habitats most of the time, ifonly a few km away from each other.
Foraging habitats: similarities
Similarities across species: CHLmax and SD(H)
Although the 7 species studied forage usingdifferent methods and in different locations,they appear to target 2 common physical and
222
Tab
le 5
. Ab
un
dan
ce g
iven
pre
sen
ce G
AM
: Th
e op
tim
al m
odel
s w
ere
sele
cted
bas
ed o
n in
div
idu
al s
ign
ifica
nce
of
exp
lan
ator
y va
riab
les.
Th
e va
lues
sh
own
ind
icat
e th
e d
e-g
rees
of
free
dom
(K
) fo
r sm
ooth
ers
and
*li
nea
r sl
opes
(+
or
– va
lues
) fo
r li
nea
r te
rms,
plu
s th
e si
gn
ifica
nce
val
ues
(p
) fo
r al
l exp
lan
ator
y va
riab
les
rem
ain
ing
in t
he
fin
al m
od-
els.
For
eac
h fi
nal
mod
el, %
dev
ian
ce e
xpla
ined
, AIC
an
d n
um
ber
of
5 m
in b
ins
(N)
are
also
giv
en. S
D(H
)w
as lo
g-t
ran
sfor
med
. Qu
asi-
Poi
sson
fits
wer
e ap
pli
ed in
all
cas
es(e
xcep
t fo
r g
rey
seal
, in
wh
ich
dat
a w
ere
un
der
-dis
per
sed
an
d a
Poi
sson
dis
trib
uti
on w
as a
ssu
med
). F
or t
his
ou
tpu
t on
ly,
‘All
an
imal
s’ i
s to
tal
nu
mb
er o
f an
imal
s
Sp
ecie
sΔT
log
10 (h
/U3 )
Δρ
––––
–C
HL
CH
Lm
axH
max
SD
(H)
Tim
eS
east
ate
Dev
ian
ce
AIC
N(%
of
24 h
)(1
–6)
exp
lain
ed (
%)
Har
bou
r S
=–
0.38
18.0
140.
0130
por
poi
sep
=0.
03
Gu
ille
mot
K=
2.83
<K
=2.
47<
S=
–0.
1315
.359
48.9
363
p<
0.00
1p
<0.
01p
<0.
01
Gre
y se
alK
=1.
51K
=2.
68K
=2.
92p
K
=2.
4896
.813
1.21
35p
=0.
04<
0.00
01p
<0.
0001
p<
0.00
01
Kit
tiw
ake
K=
3.00
K=
2.79
K=
2.97
K=
2.38
K=
1.74
K=
2.94
K=
2.40
K=
2.26
54.2
1062
.515
3p
<0.
0001
p=
0.01
p<
0.00
1p
=0.
04p
<0.
0001
p<
0.00
01p
<0.
001
p<
0.00
01
Min
ke
wh
ale
27
Gan
net
K=
2.42
K=
2.68
K=
3.00
K=
2.89
S=
1.00
K=
2.62
83.8
535.
0154
p<
0.00
1p
<0.
0001
p<
0.00
01p
< 0
.001
p<
0.00
01p
=0.
01
Wh
ite
bea
ked
K=
2.08
035
.913
2.61
25d
olp
hin
p=
0.01
All
an
imal
sS
=–1
.69
K=
2.92
K=
2.31
K=
2.25
K=
2.90
S=
–0.
1817
.193
24.7
453
p<
0.00
1p
<0.
01p
<0.
001
p=
0.04
p<
0.01
p<
0.00
1
Scott et al.: Sub-surface hotspots in shallow seas 223
Fig. 9. GAM relationships (smoothing spline of the partial residual and 95% confidence intervals) for abundance of all the signif-icant explanatory variables, log10 (h/U 3), CHLmax,
–––––CHL, Hmax, SD(H). In each row, the same partial residual for an explanatory
variable is presented as the y-axis (except row 3, where 2 explanatory variables are on the same row). The values on the y-axis in-dicated either an increased probability of an increase in abundance (>0) or an increased probability of a decrease in abundance(<0). For each row, the x-axis has the same explanatory variable (except row 3, where 2 explanatory variables are on the same
row). Each column contains the same top predator species where possible
Mar Ecol Prog Ser 408: 207–226, 2010
biological oceanographic features: higher levels ofsubsurface chlorophyll biomass (CHLmax) and highervariation in bottom depth (SD(H )). These were theonly single variables that had significant differencesbetween the regions of species presence and absenceand yet showed no difference in mean values betweenall the species (Table 3). This suggests that all 7 speciesare targeting similar values for these variables and thatthese are therefore highly important general factors inidentifying foraging locations across species.
Subsurface chlorophyll maximum (CHLmax)
CHLmax is a significant variable in 4 of the abun-dance models but shows a different shaped relation-ship for those species (kittiwake, gannet, white-beakeddolphins and grey seal: Fig. 9). We hypothesise that toppredator species are using these areas of CHLmax forforaging as they may be predictable in both space andtime. Their patchiness also suggests the possibility thatthey are limited point-source locations of primary pro-duction in the stratified sections of shallow seas.
One limitation of this study is that it is a ‘snapshot’ intime, and we suggest that caution is exercised in inter-preting the results. However, information from in-dependent multiple-year studies mapping grey seal for-aging locations using satellite tags (Matthiopoulos et al.2004) reveal that seals are consistently found within thesame areas that show high sub-surface chlorophyll in ourdata (Fig. 2b). Also, seasonal effects of chlorophyll pro-duction based on site specific mooring studies within thisstudy area (Scott et al. 2006, Sharples et al. 2006) havebeen used in longer-term analysis. A fisheries study,focused on a subsection of this study area since 1997, hasshown that incorporating seasonal chlorophyll produc-tion does explain sampling variance in fish abundance(Greenstreet et al. 2006). We therefore suggest that thepredictability of these locations may allow highly mobilemarine predators to remember and re-visit these areas,and that the persistent aggregations of phytoplanktonwill attract a range of zooplankton and fish species (sim-ilar to frontal regions: Durazo et al. 1998, Sims & Quayle1998, Russell et al. 1999, Lough & Manning 2001). Preyspecies will move in and out of these areas depending ontheir species-specific foraging strategies, predator avoid-ance behaviours and migration patterns. However, anychanges to primary production (i.e. phenology, abun-dance, species composition), and hence to trophic en-ergy transfer, will be quickly reflected by changes in mo-bile predator foraging behaviour, distribution, andannual reproductive success. For instance, the timing ofthe spring bloom in this region significantly affects thereproductive success of kittiwakes (Scott et al. 2006,Sharples et al. 2006).
Variability in depth
In general, the shape of the relationship betweenSD(H ) and presence/absence or abundance is one ofpositive linear increase, with slopes ranging from 1 to2.28 (Tables 4 & 5); this confirms that increasing varia-tion in depth, if only by meters in a shallow sea, leadsto the probability of increases in predator presence.Therefore, the combination of these results suggeststhat all 7 species forage within, or very near, locationswith higher levels of sub-surface chlorophyll and thatmore animals were found where there was higher vari-ation in depth. However, there was no significant rela-tionship between just CHLmax and SD(H ), suggestingthat additional factors such as stratification must alsohave the necessary values within those locations be-fore they provide an environment for high concentra-tions of top predators.
What defines critical foraging habitat?
Firstly, the findings of this study point to a need tounderstand the mechanism(s) for the creation of patchylocations of sub-surface chlorophyll maximum (CHLmax)before trying to predict the exact locations of these crit-ical areas at the spatial scale at which they occur. Wesurveyed some of these hotspots in circuits repeatedevery 2 h over whole tidal cycles and showed that tidalspeed both affects the behaviour of the main prey fishspecies, sandeel Ammodytes marinus, via a changein schooling behaviour, and produces increases inthe foraging activity of surface feeding birds (C. E.Embling et al. unpubl.).
Secondly, we hypothesise that increased variation inbottom depth (SD(H )) leads to an increased probabil-ity of presence and abundance for most of our studyspecies. This indicates that slope-generated mixingmay be playing an important role in creating thesecritical areas, as this factor may be responsible for thegeneration of internal waves. Briefly, the formation ofinternal waves, with typical wavelengths in the rangeof 1 to 2 km and amplitudes of 10 to 20 m, are createdby flow over non-uniform topography during high andfalling tidal current speeds in areas where the watercolumn is stratified (Moum & Nash 2000, Sharples2008). The increase in vertical mixing due to internalwaves may increase primary production and may alsoassist in aggregating smaller prey items, making forag-ing more likely in these areas. To date there have beenonly a few studies which look at trophic interactions atthese fine spatial and temporal scales (Moore & Lien2007, Bertrand et al. 2008, Stevik et al. 2008) and theyhave all found internal waves to be a possible mecha-nism for enhancing trophic coupling.
224
Scott et al.: Sub-surface hotspots in shallow seas
CONCLUSIONS
Most fish species are both predators and prey, and assuch their behaviour will switch between evasion ofpredators and their own feeding. In contrast, apexpredators such as seabirds and marine mammals aremainly concerned with feeding, such that their forag-ing times and locations will represent a foraging strat-egy focused at productive locations and constrainedonly by breeding behaviour. Encouragingly, we foundthat the fine scales characteristic of this study (~2 to10 km) were the same habitat variables (topography,primary production and water column mixing char-acteristics) that are found to be important in determin-ing top predator distributions at much larger spatial(~100 km) and temporal scales (Genin 2004, Yen et al.2004, Bakun 2006, Ballance et al. 2006), and may helpto explain why others have found an absence of scaledependence for dolphin habitat (Redfern et al. 2008).However, this study uniquely focused this questionwithin a representative 10 000 km2 region of shallowsea, where there are changes of only tens of metres indepth and no frontal areas (as defined by Simpson &Hunter 1974 or Sharples 2008). Our conclusion is that,in shallow seas, log10(h/U 3) is a very important predic-tive habitat variable and that values of log10(h/U3)beyond frontal areas are also biologically important.
Understanding exactly why subsurface chlorophyllpatches are important to predators is particularly chal-lenging but necessary to predict where these poten-tially transient but critical habitats will occur. As thesepatches appear to be associated with topographicallydriven internal waves, there are 2 alternative hypothe-ses which could explain the localised concentration ofpredators in such areas: (1) they result from complextrophic interactions via ‘bottom-up forcing’, with moreprey available due to higher primary productivitywhere internal waves have caused vertical mixing ofnutrients; or (2) they result from just topographicalforcing, with prey in those areas simply easier to catchas the internal waves bring them closer to the surfaceand/or aggregate them. Testing these alternate hypo-theses and understanding exactly where, when andwhy larger predators move into, and actively foragewithin, specific locations will enhance the decisionmaking process on a wide range of marine conserva-tion issues.
These extremely patchy areas represent a newlyidentified class of spatially important location, with sub-surface chlorophyll characteristics that are not readilyidentifiable from surface characteristics (Weston et al.2005), yet they may play a critical role in trophic cou-pling within regions of stratified water in shallow seasand need to be understood in detail. This level ofunderstanding is needed for the spatial planning of the
marine environment, such as decisions on the designand locations of marine protected areas (MPA), spa-tially explicit fishing management and providing confi-dence that placement of offshore renewable energydevices will not overlap or interfere with these areas ofcritical marine habitat.
Acknowledgements. Funding for original fieldwork was viaEU Q5RS 2000-30864, IMPRESS (Interactions between theMarine Environment, Predators and prey: implications forSustainable Sandeel fisheries). Funding for extended workwas provided by NERC Sustainable Marine Bioresources pro-gramme NE/F001983/1; G.J.P. was funded under the EUMarie Curie programme (MEXC-CT-2006-042337). Specialthanks to the crew of the ‘Pelagia’, especially Martin Laanand Santiago Gonsalez. Bird/sea mammal observers: Suzanvan Lieshout, Luc Meeuwisse, Phillip Schwemmer and NicoleSonntag. Volunteers for oceanography: Jackie Smith, DamionNixon and Dr. Patrick Holligan (University of Southampton)for the analysis of the chlorophyll samples. Very specialthanks for great efforts from FRS Marine Lab, Aberdeen: JohnDunn and Eric Armstrong. We thank Dr. Lisa Ballance and 2anonymous reviewers whose suggestions enhanced the work.We also acknowledge C. D. Macleod for extracting and con-figuring the BGS data and Proudman Oceanographic Labora-tories for the use of POLPRED.
LITERATURE CITED
Ashmole NP (1971) Sea bird ecology and the marine environ-ment. In: Farner DS, King JR (eds) Avian Biology, Vol 1.Academic Press, New York, NY p 224–286
Bakun A (2006) Fronts and eddies as key structures in thehabitats of marine fish larvae: opportunity, adaptiveresponse and competitive advantage. Sci Mar 70:105–122
Ballance LT, Pitman RL, Fiedler PC (2006) Oceanographicinfluences on seabirds and cetaceans of the eastern tropi-cal Pacific. A review. Prog Oceanogr 69:360–390
Bertrand A, Gerlotto F, Bertrand S, Gutiérrez M, and others(2008) Schooling behaviour and environmental forcingin relation to anchoveta distribution: An analysis acrossmultiple spatial scales. Prog Oceanogr 79:264–277
Bost CA, Cotté C, Bailleul F, Cherel Y and others (2009) Theimportance of oceanographic fronts to marine birds andmammals of the southern oceans. J Mar Syst 78:363–376
Buckland ST, Anderson DR, Burnham KP, Laake JL, BorchersDL, Thomas L (2001) Introduction to distance sampling:Estimating abundance of biological populations. OxfordUniversity Press, Oxford
Burger AE (2003) Effects of the Juan de Fuca Eddy andupwelling on densities and distributions of seabirdsoff southwest Vancouver Island, British Columbia. MarOrnithol 31:113–122
Camphuysen CJ, Garthe S (2004) Recording foragingseabirds at sea: standardised recording and coding of for-aging behaviour and multi-species foraging associations.Atlantic Seabirds 6:1–32
Camphuysen CJ, Fox AD, Leopold MF, Petersen IK (2004)Towards standardised seabirds at sea census techniquesin connection with environmental impact assessments foroffshore wind farms in the UK. Report commissioned byCOWRIE for the Crown Estate, London. Royal Nether-lands Institute for Sea Research, Texel
Cotté C, Simard Y (2005) Formation of dense krill patches
225
Mar Ecol Prog Ser 408: 207–226, 2010
under tidal forcing at whale feeding hot spots in the St.Lawrence Estuary. Mar Ecol Prog Ser 288:199–210
Cushing DH (1975) Marine ecology and fisheries. CambridgeUniversity, Cambridge
Decker MB, Hunt GL Jr (1996) Foraging by murres (Uria spp.)at tidal fronts surrounding the Pribilof Islands, Alaska,USA. Mar Ecol Prog Ser 139:1–10
Durazo R, Harrison NM, Hill AE (1998) Seabird observationsat a tidal mixing front in the Irish sea. Estuar Coast ShelfSci 47:153–164
Franks PJS, Chen C (1996) Plankton production in tidal fronts:A model of Georges Bank in summer. J Mar Res 54:63l–651
Genin A (2004) Bio-physical coupling in the formation of zoo-plankton and fish aggregations over abrupt topographies.J Mar Syst 50:3–20
Greenstreet SPR, Armstrong E, Mosegaard H, Jensen H andothers (2006) Variation in the abundance of sandeelsAmmodytes marinus off southeast Scotland: an evaluationof area-closure fisheries management and stock abun-dance assessment methods. ICES J Mar Sci 63:1530–1550
Irons DB (1998) Foraging area fidelity of individual seabirdsin relation to tidal cycles and flock feeding. Ecol 79:647–655
Johnston DW, Westgate AJ, Read AJ (2005a) Effects of fine-scale oceanographic features on the distribution andmovements of harbour porpoises Phocoena phocoena inthe Bay of Fundy. Mar Ecol Prog Ser 295:279–293
Johnston DW, Thorne LH, Read AJ (2005b) Fin whales Bal-aenoptera physalus and minke whales Balaenoptera acu-torostrata exploit a tidally driven island wake ecosystem inthe Bay of Fundy. Mar Ecol Prog Ser 305:287–295
Ladd C, Jahncke J, Hunt GL Jr, Coyle KO, Stabeno PJ (2005)Hydrographic features and seabird foraging in AleutianPasses. Fish Oceanogr 14(Suppl. 1):178–195
Lough RG, Manning JP (2001) Tidal-front entrainment andretention of fish larvae on the southern flank of GeorgesBank. Deep-Sea Res II 48:631–644
Matthiopoulos J, McConnell B, Duck C, Fedak M (2004)Using satellite telemetry and aerial counts to estimatespace use by grey seals around the British Isles. J ApplEcol 41:476–491
Mendes S, Turrell W, Lütkebohle T, Thompson P (2002) Influ-ence of the tidal cycle and a tidal intrusion front on thespatio-temporal distribution of coastal bottlenose dolphins.Mar Ecol Prog Ser 239:221–229
Moore SE, Lien RC (2007) Pilot whales follow internal solitarywaves in the south China Sea. Mar Mamm Sci 23:193–196
Moum JN, Nash JD (2000) Topographically induced drag andmixing at a small bank on the continental shelf. J PhysOceanogr 30:2049–2054
Pingree RD, Griffiths DK (1978) Tidal fronts on shelf seasaround the British Isles. J Geophys Res 83:4615–4622
Pingree RD, Holligan PM, Mardell GT, Head RN (1975) Sum-mer phytoplankton blooms and red tides along tidal frontsin approaches to English Channel. Nature 258:672–677
Polovina JJ, Howell E, Kobayashi DR, Seki MP (2001) Thetransition zone chlorophyll front, a dynamic global featuredefining migration and forage habitat for marine resources.Prog Oceanogr 49:469–483
Redfern JV, Barlow J, Bailance LT, Gerrodette T, Becker EA(2008) Absence of scale dependence in dolphin-habitatmodels for the eastern tropical pacific ocean. Mar EcolProg Ser 363:1–14
Ross HM, Wilson B (1996) Violent interactions between bot-tlenose dolphins and harbour porpoises. Proc Biol Sci 263:283–286
Russell RW, Harrison NM, Hunt GL Jr (1999) Foraging at a
front: Hydrography, zooplankton, and avian planktivoryin the northern Bering Sea. Mar Ecol Prog Ser 182:77–93
Scott BE, Sharples J, Wanless S, Ross ON, Frederiksen M,Daunt F (2006) The use of biologically meaningful oceano-graphic indices to separate the effects of climate and fish-eries on seabird breeding success. In: Boyd IL, Wanless S,Camphusen CJ (eds) Top predators in marine ecosystems.Their role in monitoring and management. ConservationBiology No 12, Cambridge University Press, Cambridge
Sharples J (2008) Potential impacts of the spring-neap tidalcycle on shelf sea primary production. J Plankton Res30:183–197
Sharples J, Ross ON, Scott BE, Greenstreet SPR and 1 other(2006) Inter-annual variability in the timing of stratifica-tion and the spring bloom in the North-western North Sea.Cont Shelf Res 26:733–751
Simard Y, Lavoie D, Saucier FJ (2002) Channel head dynam-ics: capelin (Mallotus villosus) aggregation in the tidallydriven upwelling system of the Saguenay-St. LawrenceMarine Park’s whale feeding ground. Can J Fish Aquat Sci59:197–210
Simpson JH, Hunter JR (1974) Fronts in the Irish Sea. Nature250:404–406
Sims DW, Quayle VA (1998) Selective foraging behaviour ofbasking sharks on zooplankton in a small-scale front.Nature 393:460–464
Sinclair EH, Moore SE, Friday NA, Zeppelin TK, Waite JM(2005) Do patterns of Steller sea lion (Eumetopias jubatus)diet, population trend and cetacean occurrence reflectoceanographic domains from the Alaska Peninsula tothe central Aleutian Islands? Fish Oceanogr 14(Suppl 1):223–242
Spear LB, Balance LT, Ainley DG (2001) Response of seabirdsto thermal boundaries in the tropical Pacific: the thermo-cline versus the Equatorial Front. Mar Ecol Prog Ser 219:275–289
Stevick PT, Incze LS, Kraus SD, Rosen S, Wolff N, Baukus A(2008) Trophic relationships and oceanography on andaround a small offshore bank. Mar Ecol Prog Ser 363:15–28
Sydeman WJ, Brodeur RD, Grimes CB, Bychkov AS, Mc-Kinnell S (2006) Marine habitat ‘hotspots’ and their use bymigratory species and top predators in the North PacificOcean: Introduction. Deep-Sea Res II 53:247–249
Tasker ML, Jones PH, Dixon TJ, Blake BF (1984) Countingseabirds at sea from ships: a review of methods employedand a suggestion for a standardized approach. Auk 101:567–577
Uda M (1952) On the relation between the variation of theimportant fisheries conditions and the oceanographic con-ditions in the adjacent waters of Japan. J Tokyo Univ Fish38:363–389
Ware DM, Thomson RE (2005) Ecology: bottom-up ecosystemtrophic dynamics determine fish production in the north-east pacific. Science 308:1280–1284
Weston K, Fernand L, Mills DK, Delahunty R, Brown J (2005)Primary production in the deep chlorophyll maximum ofthe central north sea. J Plankton Res 27:909–922
Worm B, Sandow M, Oschlies A, Lotze HK, Myers RA (2005)Global patterns of predator diversity in the open ocean.Science 309:1365–1369
Yen PPW, Sydeman WJ, Hyrenbach KD (2004) Marine birdand cetacean associations with bathymetric habitats andshallow-water topographies: implications for trophic trans-fer and conservation. J Mar Syst 50:79–99
Zuur AF, Ieno EN, Smith GM (2007) Analysing ecologicaldata. Springer, New York, NY
226
Editorial responsibility: Hans Heinrich Janssen,Oldendorf/Luhe, Germany
Submitted: October 28, 2009; Accepted: February 19, 2010Proofs received from author(s): May 30, 2010