Absence of scale dependence in dolphin-habitat models for the eastern tropical Pacific Ocean

Absence of scale dependence in dolphin-habitat models for the eastern tropical Pacific Ocean
of 14
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  MARINE ECOLOGY PROGRESS SERIESMar Ecol Prog SerVol. 363: 1–14, 2008 doi: 10.3354/meps07495 Published July 15 INTRODUCTION The influence of scale on the identification and inter-pretation of ecological patterns has long been a centraltheme in ecological research (Haury et al. 1978, Wiens1989, Levin 1992). Of particular importance is spatialscale dependence in species–habitat relationships,which has been documented for a variety of seabirdspecies (e.g. Logerwell & Hargreaves 1996, Logerwellet al. 1998, Fauchald et al. 2000, Davoren et al. 2002,Pinaud & Weimerskirch 2005). For example, seabirdstudies suggest that the strength of the correlationbetween seabirds and prey generally increases withincreasing spatial resolution (Schneider & Piatt 1986,Heinemann et al. 1989, Erikstad et al. 1990, Piatt 1990,Hunt 1991, Hunt et al. 1992, Logerwell & Hargreaves1996, Fauchald et al. 2000), although there are excep-tions (e.g. Cairns & Schneider 1990). © Inter-Research 2008 · www.int-res.com*Email: jessica.redfern@noaa.gov FEATURE ARTICLE Absence of scale dependence in dolphin–habitatmodels for the eastern tropical Pacific Ocean J. V. Redfern*, J. Barlow, L. T. Ballance, T. Gerrodette, E. A. Becker Southwest Fisheries Science Center, 8604 La Jolla Shores Drive, La Jolla, California 92037, USA ABSTRACT: Research into the effects of scale oncetacean–habitat relationships is limited and hasproduced ambiguous results. We explored the effectsof spatial resolution (a component of scale) on dolphin–habitat models using 4 yr of data collected in theeastern tropical Pacific Ocean (ETP). We developedgeneralized additive models of dolphin–habitat rela-tionships for 4 species at 6 spatial resolutions usingoceanographic and geographic habitat variables. Forall species, the ecological patterns in the models builtat the different resolutions were similar: the same vari-ables frequently occurred at multiple resolutions andhad similar functional forms, and maps of predicteddistributions identified similar high and low densityregions. Additionally, interannual habitat variability,which is most likely related to the El Niño–SouthernOscillation, had a greater impact on the predictivepower of dolphin–habitat models than spatial resolu-tion. Although it is common to find scale dependencein species–habitat relationships, domains of scale existin which ecological patterns do not change. Theabsence of scale dependence in the models for the4dolphin species suggests that resolutions from 2 to120 km occur within a single domain of scale in theETP. This domain of scale may be determined by thephysical oceanography of the ETP, which is generallydefined by large-scale processes. Although resolutionsfrom 2 to 120 km appear to occur within a domain ofscale, building models at the larger resolutions weinvestigated may reduce the noise in the data due tofalse absences.KEY WORDS: Dolphin density · Habitat modeling ·Striped dolphin · Stenella coeruleoalba · Easternspinner dolphin · Stenella longirostris orientalis  ·Short-beaked common dolphin · Delphinus delphis  ·Risso’s dolphin · Grampus griseus  Resale or republication not permitted without written consent of the publisher  Modeled densities indicate scale-independence in dolphin–habitat relationships Image: J.V. Redfern O PEN  N   CCESS  SS  Mar Ecol Prog Ser 363: 1–14, 2008 Research into the effects of scale on cetacean–habitatrelationships is more limited and has producedambiguous results. Jaquet & Whitehead (1996) ana-lyzed a range of resolutions (i.e. the unit of analysis,acomponent of scale) from 80 to 640 nautical miles(nmile) (148 to 1185 km) and found that sperm whaledensity in the South Pacific was correlated with habitatvariables only at resolutions greater than 320 n mile(593 km). Hamazaki (2002) assessed the effect of spa-tial resolutions ranging from 4 × 4 km grid cells to 96 × 96 km grid cells on predictive habitat models for13cetacean species in the mid-western North AtlanticOcean and found no relationship between resolutionand the correct classification rate of logistic regressionmodels. While Hamazaki’s (2002) results suggest thatspatial resolution did not affect the predictive power ofthe models, ecological questions about the effect ofresolution cannot be addressed because the behaviorof the predictor variables at the different resolutions(e.g. the magnitude and sign of the relationship) wasnot examined.We expect scale-dependent habitat relationships forcetaceans and seabirds because both are apex marinepredators that must respond to the hierarchical patchstructure of their prey in which high density, small-scale patches are nested within lower density, large-scale patches (Murphy et al. 1988, Fauchald 1999,Fauchald et al. 2000). Heterogeneity, which may be adriving factor in scale-dependent species–habitat rela-tionships (Wiens 1989), occurs at every stage of thishierarchy and is coupled with biological and physicalprocesses (Haury et al. 1978). Heterogeneity in small-scale patches (100s of m) may be driven by preybehavior or by turbulent diffusion and mixing forweaklyswimming organisms; heterogeneity in meso-scale patches (10s to 100s of km) may be caused byoceanographic features such as fronts and eddies(Murphy et al. 1988, Fauchald 1999, Fauchald et al.2000).A primary difference between cetacean and seabirdhabitat studies is found in the variables used to definehabitat. Cetacean habitat is often defined using oceano-graphic variables (Redfern et al. 2006), while seabirdhabitat is frequently defined using prey density.Cetacean studies may use variables such as sea sur-face temperature and salinity to represent physiologi-cal constraints or identify water masses typical of goodprey habitat as well as variables such as thermoclinedepth and strength, which are expected to influenceprey abundance or availability. Consequently, rela-tionships between apex marine predators and habitatdefined by oceanographic variables should be subjectto scale dependencies because heterogeneity in preypatches is likely to be linked to oceanographic condi-tions.In this paper, we explore the effects of spatial resolu-tion on dolphin–habitat models using 4 yr of data col-lected aboard research vessels in the eastern tropicalPacific Ocean (ETP). We developed generalized addi-tive models (GAMs) of dolphin–habitat relationships at6 spatial resolutions by varying the unit of analysisfrom 2 to 120 km. Habitat was defined using physicaland biological oceanographic and geographic vari-ables. Four dolphin species were selected to representa range of habitat types and selectivity: striped dol-phin Stenella coeruleoalba , eastern spinner dolphin S.longirostris orientalis  , short-beaked common dol-phin Delphinus delphis  , and Risso’s dolphin Grampus griseus  . MATERIALS AND METHODSStudy area and data collection. The ETP (Fig. 1) is alarge (19.6 million km 2 ), oceanographically diversearea that supports at least 29 cetacean species (Wade &Gerrodette 1993). Temporal variability in the ETPexists at a range of scales, but is dominated by inter-annual variability created by the El Niño–SouthernOscillation (ENSO) (Fiedler 2002a). On a macroscale,the ETP is defined by 3 surface currents: the NorthEquatorial Current, the North Equatorial Countercur-rent, and the South Equatorial Current (Fig. 1). Thenorthern and southern regions of the ETP are definedby Subtropical Surface Water (Fig.1), which has thehighest salinity values (greater than 35) and lowestnutrient concentrations in the area (Fiedler & Talley2006). The central region is defined by Tropical Sur-face Water north of the equator, which includes theeastern Pacific warm pool, and by Equatorial SurfaceWater along the equator, which includes the equatorialcold tongue (Fig. 1). Tropical Surface Water has highersurface temperatures (greater than 25°C), lower sal-inity values (less than 34), and lower nutrient concen-trations than the Equatorial Surface Water (Fiedler &Talley 2006). The eastern boundary currents (i.e. theCalifornia Current and the Peru Current) also havehigh nutrient concentrations relative to surroundingregions (Fiedler & Talley 2006).The ETP contains a number of mesoscale featuresthat also define the habitat of this region (Fig. 1). Ther-mal fronts occur between the eastern boundary cur-rents and the warm pool, as well as between the SouthEquatorial Current and the North Equatorial Counter-current (Fiedler & Talley 2006). The Equatorial Front,which occurs between the Peru Current and the NorthEquatorial Countercurrent, is one of the most promi-nent, low latitude, oceanic fronts in the world. It is apermanent feature of the region, although its intensityvaries both spatially and temporally. A strong, shallow 2  Redfern et al.: Scale independence in dolphin–habitat models thermocline predominates throughout theETP, and a thermocline ridge predictablyoccurs between the North Equatorial Cur-rent and the North Equatorial Counter-current. The Costa Rica Dome, a cyclonicgyre that develops at the eastern end ofthis ridge, is an area of regionally highsurface chlorophyll from May to Septem-ber (Fiedler 2002b). High productivity inthe dome is supported by nutrientsbrought tothe surface by wind mixingandupwelling. These 3 features, theEquatorial Front, the thermocline ridge,and the Costa Rica Dome, are significantfor a number of seabirds and cetaceans(Ballance et al. 2006).We used spatially referenced dolphinand oceanographic data that were col-lected by the Southwest Fisheries Sci-ence Center (NOAA Fisheries). The datawere collected from 2comparable oceano-graphic research vessels from late Julyuntil early December in 1992 and 1998 to2000 (athird comparable vessel was alsoused in 1998). The 1992 survey focusedon asubsection of the study area off theMiddle and South American coast fromGuatemala to Colombia. Dolphin datawere collected during daylight hours usingline-transect methods (Buckland et al.2001); field protocols are described indetail in Kinzey et al. (2000), Barlow etal.(2001), and Gerrodette & Forcada (2005).We used approximately 70000 km of on-effort sampling data, which is defined assampling conducted by the full observingteam when Beaufort sea state was lessthan 6. The transects provided dense cov-erage of the study area (Fig. 2). Surveyeffort consisted of 2 observers searchingfor dolphins from the flying bridge of theship (at a height ranging between 10 and15 m above sea level) using pedestal-mounted 25 × 150 binoculars. A thirdobserver served as the data recorder andsearched by naked eye or with 7 × hand-held binoculars. Observers also regularlyrecorded survey conditions, includingtheBeaufort seastate. When dolphinswere detected, the vessel approached thegroup as needed to identify the speciesand obtain group size estimates. Ob-servers independently recorded their best 3 30°N20°N10°N0°10°S20°S160°W P     e    r    u     C     u    r    r    e    n    t      Subtropical Surface WaterEquatorial FrontCosta RicaDomeEastern PacificWarm PoolEquatorial Cold Tongue C     a    l     i     f      o    r    n    i     a     C     u    r    r    e    n    t      Equatorial Surface WaterTropical Surface WaterSubtropical Surface Water 0 750 1500 3000 km South Equatorial CurrentNorth Equatorial CurrentNorth Equatorial Countercurrent 140°W120°W100°W80°W Fig. 1. Primary oceanographic features in the eastern tropical Pacific Ocean(ETP) that influence dolphin habitats. Study area is shown by the thick blackline; water masses are identified by solid boxes; currents are shown with an ar-row indicating the direction of flow, and other important features are identifiedby dashed boxes (adapted from Wyrtki 1966, 1967, Fiedler & Talley 2006) 30°N20°N10°N0°10°S20°S150°W0 625 1250 2500 km140°W130°W120°W110°W100°W90°W80°W Fig. 2. Transect lines ( - - - - ) used to collect dolphin and oceanographic data in theETP. Data were collected aboard 2 research vessels from late July until earlyDecember in 1992 and 1998 to 2000 (3 vessels were used in 1998). Locations of thelargest 10% of temperature fronts at the 120 km resolution are shown for allyearsof data ( d ). Fronts were defined as the difference between the minimum and maximum temperature recorded on a segment  Mar Ecol Prog Ser 363: 1–14, 2008 estimate of group size, as well as high and low groupsize estimates, for each sighting. To obtain a singlegroup size estimate for each sighting, we averaged thebest estimate from each observer. In mixed speciessightings, we averaged the estimates of the percentageof each species from each observer. We used only on-effort sightings that were identified to species in ouranalyses. Additionally, only those sightings for whichat least 1 observer estimated a best group size and(where applicable) species percentages were includedin our analyses.Oceanographic sampling was systematically con-ducted during each survey (see Fiedler & Philbrick2002 for a detailed description). Surface temperatureand salinity were recorded at 2 min intervals usingathermosalinograph. Surface chlorophyll concentra-tions were measured at approximately 55 km inter-vals using a fluorometer. Water column properties,including thermocline depth and strength (i.e. mid-point of the depth interval containing the maximumtemperature gradient and the value of the gradient,respectively), were derived from data collected usingexpendable bathythermographs (XBTs) and CTDcasts. XBT drops to 760 m were conducted at approxi-mately 55 km intervals each day and CTD casts to1000 m were conducted before sunrise and after sun-set. The seafloor depth was derived from a digitalbathymetric map with a resolution of 1 to 12 km,which captures intermediate and large-scale struc-tures of the ocean basins including canyons, ridges,and seamounts (Smith & Sandwell 1997), usingARCGIS version 9.1 (ESRI). Data analyses. The design of the ETP cetacean andecosystem assessment surveys resulted in transectsthat traversed the study area and ran parallel to eachother or intersected at multiple locations. Transects inclose spatial proximity may be temporally separatedsuch that oceanographic conditions differ substan-tially. Consequently, a gridded representation of ourdata would combine temporally separated transectsand was therefore deemed too coarse to adequatelyrepresent concurrent dolphin–habitat relationships.Instead, we summarized the data in 2, 10, 20, 40, 60,and 120 km segments along the transect lines; segmentlengths were calculated using great circle distancesfrom the starting point of effort each day. The 120 kmresolution corresponds to a minimum amount of tran-sect covered in a typical day; effort beyond 120 km wasexcluded from our analyses. The ship generally contin-ued along the transect at night, producing a break insurvey effort that is assumed to reduce dependenceamong daily segments. Consequently, we did not ex-plore resolutions larger than 120 km.Four dolphin species were included in our analyses:striped dolphin, eastern spinner dolphin, short-beakedcommon dolphin, and Risso’s dolphin. Encounter rateand group size models were built for each species. Dol-phin encounter rates were calculated as the number ofsightings of each species on a segment divided by thedistance traveled on effort in that segment. When morethan one sighting of a given species occurred on asegment, group sizes were averaged to obtain a singlegroup size. We did not assess spatial autocorrelation inour dependent dolphin variables because we wereconcerned with the predictive power of the models,which is not affected by spatial autocorrelation.Habitat variables used in our analyses were surfacetemperature and salinity, the natural logarithm of sur-face chlorophyll concentration (this transformation wasused because the minimum and maximum measuredvalues differed by more than an order of magnitude),thermocline depth and strength, and seafloor depth.We also approximated temperature fronts by subtract-ing the minimum from the maximum temperature oneach segment. Large temperature differences tendedto occur along the Equatorial Cold Tongue and inthecoastal waters of our study area (Fig. 2). We alsoincluded Beaufort sea state as a predictor variableinour models. Although Beaufort sea state is notexpected to affect the number or type of dolphins on asegment, it was included as a correction for sightingconditions because the probability of detecting dol-phins decreases with increasing Beaufort sea states(Barlow et al. 2001). The product-moment correlationcoefficients among all pairs of variables at the 2 kmscale were between –0.5 and 0.5.Surface chlorophyll concentration and thermoclinedepth and strength were interpolated to the midpoint ofthe 2 km segments using inverse distance weighting;the degree of smoothing for each of these variables de-pended on the frequency of data collection (Fig. 3).Only 2 oceanographic measurements, the one tempo-rally before and the one temporally after the segmentmidpoint, were used in the interpolation. Segmentswere not used in analyses unless both measurementswere within 370.4 km (approximately 200nmile) of themidpoint. All surface temperature and salinity mea-surements made within a segment were averaged toobtain a single value. The seafloor depth was calcu-lated at the midpoint of each 2 km segment. MeanBeaufort sea state values, weighted by the amount ofon-effort distance traveled in that state, were also cal-culated for each 2 km segment. Only segments forwhich mean Beaufort sea state was <5.5 were includedin our analyses because of the difficulty in detectingdolphins at higher Beaufort sea states.Estimates of surface chlorophyll, thermocline depthand strength, seafloor depth, and Beaufort sea state forthe 2 km segments were averaged to create the cor-responding values for 10, 20, 40, 60, and 120 km seg- 4  Redfern et al.: Scale independence in dolphin–habitat models ments. Temperature, salinity, and temperature frontvalues were recalculated for each segment. Any largersegment containing one or more 2 km segments withamissing value for a predictor variable was alsoassigned a missing value. Because a single missingvalue at the 2 km resolution produces missing values atthe larger resolutions, there are many days containing2 km segments but no segments at the larger resolu-tions. To ensure consistency in the data across resolu-tions, only days for which the 120km segment con-tained complete oceanographic data were used for allresolutions. The total number of segments, number ofdolphin sightings, and mean dolphin group sizes foreach resolution are presented in Table 1.We used GAMs to explore the effects of spatialresolution on dolphin–habitat relationships in theETP. AGAM approach was selected because it can fita non-parametric relationship between the responseand predictor variables viaa scatterplot smoother;consequently, this approach allowsthe data to identify nonlinearitiesin dolphin–habitat relationshipsrather than imposing parametricfits through polynomial terms in alinear regression (Chambers &Hastie 1991). This approach hasbeen applied to model cetacean–habitat relationships for delphinids(Ferguson et al. 2006a) and beakedwhale species (Ferguson et al.2006b) using data collected in theETP from 1986 to 1990. The soft-ware package SPlus (Windows Pro-fessional Developer, version 7.0,Insightful Corp.) was used to fit theGAMs; we chose cubic smoothingsplines (Hastie & Tibshirani 1990)with a maximum of 3 degrees offreedom for all predictor variablestocapture non-linear relationships,while limiting the inclusion of unrealistic detail in theshape of the function (Forney 2000).Dolphin encounter rates are expected to follow anoverdispersed Poisson distribution;therefore,encounterrate models were built using a logarithmic link andquasi-likelihood error distribution in which the vari-ance was proportional to the mean. The distancetraveled on effort for each segment was incorporatedas an offset in the models. Dolphin group size modelswere fit using only those segments in which the spe-cies was present; models were built using a log-normaldistribution. For both dolphin encounter rate andgroup size models, an automated forward/backwardstepwise approach based on Akaike’s information cri-terion (AIC) was used to select the variables for inclu-sion in each model as well as the degrees of freedomfor the cubic smoothing splines (Ferguson et al. 2006a).Each model was fitted 3 times, starting with a nullmodel that included only the intercept. The dispersion 5Fig. 3. Frequencies of the distances between the midpoints of the 2 km segments and the data collection locations used for the interpolation of (A) thermocline depth and strength and (B) surface chlorophyll measurementsTable 1. Number of sightings (Sight.)and mean group size (GS) are shown for the4species and 6 spatial resolutions (including total number of segments for eachresolution) considered in our analyses. The 120 km resolution has the highest num-ber of encounters for several species because segments with Beaufort sea state val-ues >5.5 were excluded from our analyses. In particular, 2 km segments containingan encounter and occurring in Beaufort sea states >5.5 may not contribute to theanalyses at the smaller resolutions but may contribute at the larger resolutions if theaverage Beaufort sea state on the longer segment was ≤ 5.5. Striped dolphin Stenellacoeruleoalba , eastern spinner dolphin S. longirostris orientalis  , short-beaked common dolphin Delphinus delphis, Risso’s dolphin Grampus griseus  Spatial Striped Eastern Short-beaked Risso’s Total resolution dolphinspinner common dolphinno. of (km)dolphindolphinsegmentsSight.GSSight.GSSight.GSSight.GS242049.0316196.87263160.9313421.46378301042049.0916196.76263162.3513421.2386592042148.6116196.87264163.3113421.8644524042148.5416196.10264163.0313521.4722626042146.3016195.93264164.2613519.80151312042147.3416197.64265164.2513519.44763
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks