useofencounterdatatomodelspatio ... · nelson bighorn sheep (ovis canadensis nelson; turner et al.,...

Use of encounter data tomodel spatio-temporal distribution patterns ofendangered smalltooth sawfish,Pristis pectinata, in the westernAtlantic

JOHN D. WATERSa,*, RUI COELHOb, JOANA FERNANDEZ-CARVALHOa,b, AMY A. TIMMERSc, TONYA WILEYd,JASON C. SEITZe, MATTHEW T. MCDAVITTf, GEORGE H. BURGESSa and GREGG R. POULAKISc

aFlorida Program for Shark Research, Florida Museum of Natural History, University of Florida, Gainesville, FL, USAbCentro de Ciencias do Mar, University of Algarve, Campus de Gambelas, Faro, Portugal

cFlorida Fish and Wildlife Conservation Commission, Fish and Wildlife Research Institute, Charlotte Harbor Field Laboratory, PortCharlotte, FL, USA

dHaven Worth Consulting, Palmetto, FL, USAeANAMAR Environmental Consulting, Inc., Gainesville, FL, USAfNational Legal Research Group, Inc., Charlottesville, VA, USA

ABSTRACT

1. Sawfishes are among themost threatenedfish species globally,with only the smalltooth sawfish (Pristis pectinata) currentlyregularly observed in thewesternAtlantic. TheNational SawfishEncounterDatabase (NSED)documents reported encounterswith sawfishes in the western Atlantic and contains 4945 reports of 8773 individual P. pectinata (1782–2011).2. Statistical modelling (generalized linear models and generalized additive models) and kernel density analyses

were used to (1) identify spatio-temporal patterns among encounter reports, including range reduction in the westernAtlantic; (2) determine current distribution to identify areas and time periods where conservation and recovery effortscould be focused; and (3) identify and describe spatio-temporal distribution patterns of large juveniles and adults.3. Pristis pectinata were found to be year-round residents of Florida but showed relatively consistent spatial and

temporal trends by life stage throughout the year. Although the historical range in the western Atlantic includedcoastal waters from North Carolina to Brazil, the current geographic range of the species was limited to Floridafrom 2001 through 2011, with occasional reports in neighbouring states, the Bahamas, and Cuba.4. Seasonally, encounters of all life stages peaked fromMarch through July and annual recruitment of juveniles was

apparent during the study period. Spatial hotspots based on increased numbers of encounters of large juveniles(201–340 cm) and adults (>340 cm) were identified in southern Charlotte Harbor, the Ten Thousand Islands, FloridaBay, the Atlantic side of the FloridaKeys, and off St. Lucie in south-east Florida. The analyses presented herein provideevidence of range reduction in the western Atlantic, provide an important tool for resource managers to focus research,monitoring, and conservation efforts, and may provide a framework to model and predict habitat use of other species.Copyright # 2014 John Wiley & Sons, Ltd.

Received 23 July 2013; Revised 17 February 2014; Accepted 16March 2014

KEY WORDS: brackish; estuary; ocean; endangered species; distribution; modeling; fish; fishing; recreation; urban development

*Correspondence to: J.D. Waters, Florida Program for Shark Research, Florida Museum of Natural History, University of Florida, Gainesville, FL32611, USA. E-mail: [email protected]

Copyright # 2014 John Wiley & Sons, Ltd.

AQUATIC CONSERVATION: MARINE AND FRESHWATER ECOSYSTEMS

Aquatic Conserv: Mar. Freshw. Ecosyst. 24: 760–776 (2014)

Published online 24 April 2014 in Wiley Online Library(wileyonlinelibrary.com). DOI: 10.1002/aqc.2461

INTRODUCTION

Encounter data, reports of species occurrence fromscientists, managers, and the general public, areoften more readily available and encompass a largergeographic area than conventional field research;these data can provide an important long-termmeans to identify gross changes in populations overtime, especially when studying rare or endangeredspecies. Large-scale surveys for rare species areoften prohibitively expensive (Odom et al., 2001),while the public can contribute to the researchquicker and more cost-effectively (Scott andHerman, 1995). Sampling effort by the public maybias results, however, since sampling is oftendisproportionately higher in some areas and subjectto other biases, including the public’s willingness toreport encounters and ability to identify the speciescorrectly. As with any other form of ecologicalsurvey, there are limitations and caveats to theiruse; however, encounter data have been usedsuccessfully to evaluate habitat use and range inother distinctive species, such as leatherback seaturtles (Dermochelys coriacea; Eckert, 2002),Nelson bighorn sheep (Ovis canadensis nelson;Turner et al., 2004), and Iberian lynx (Lynxpardinus; Palma et al., 1999). Encounter data havebeen used to help conserve and manage thesmalltooth sawfish (Pristis pectinata) populationin the USA, including producing the USSmalltooth Sawfish Recovery Plan (NMFS,2009b) and designating juvenile critical habitat(NMFS, 2009a; Norton et al., 2012).

Sawfishes (family Pristidae) are currently among themost threatened fishes in the world, with decliningnumbers and reduced distributions worldwide(Wueringer et al., 2009; Faria et al., 2013). Allsawfish species are categorized as ‘CriticallyEndangered’ with decreasing populations by theInternational Union for Conservation of Nature(IUCN) (Adams et al., 2006; IUCN, 2006). Onlytwo species of sawfish inhabit Atlantic waters: thelargetooth sawfish (Pristis pristis) and P. pectinata(Faria et al., 2013). Like its congener, P. pectinatahas been observed in tropical and subtropicalwaters in both the western and eastern Atlantic,with northern and southern range extremes largelydictated by seasonal water temperature regimes.

Early faunal surveys indicated healthy populationsof sawfishes across the western Atlantic in areas suchas Brazil, Nicaragua, and the USA (Evermann andBean, 1898; Thorson, 1982; Faria et al., 2013),although most of these populations had declined bythe mid-1980s (Faria et al., 2013). In the USA,P. pectinata are the only resident sawfish speciescurrently reported (NMFS, 2009b). Pristis pectinatawere historically commonly encountered fromNorth Carolina to Texas (Bigelow and Schroeder,1953), however, more recently they have beenobserved primarily in south and south-west Floridamarine and estuarine waters from Charlotte Harborto the Dry Tortugas (Seitz and Poulakis, 2002;Poulakis and Seitz, 2004; Wiley and Simpfendorfer,2010). In general, sawfish populations declined dueto many factors including entanglement of theirelongated toothed rostrum, bycatch in commercialgill net and trawl net fisheries, retention of sawfishrostra as trophies by anglers, and habitatdegradation (e.g. Seitz and Poulakis, 2006; NMFS,2009b). In addition, their life history strategy,characterized by slow growth, late sexual maturity,and low fecundity, corresponds to the species’ lowcapacity for recovery after population depletion(Simpfendorfer, 2000). As a result of large-scalepopulation losses and range reduction over the lastcentury, P. pectinata has been protected in Floridasince 1992 (FWC, 1999), in the USA under theEndangered Species Act (ESA) since 2003 (NMFS,2003), and international trade has been prohibitedunder the Convention on International Trade inEndangered Species of Wild Fauna and FloraAppendix I since 2007 (CITES, 2007).

Over the last decade, research has focusedprimarily on juvenile life stages of P. pectinata inFlorida waters. Pristis pectinata are born at67.1–81.2 cm stretched total lengths (STL)(Poulakis et al., 2011). Length–frequency andtag–recapture data in Florida nurseries indicatethat growth is rapid during the first two years afterbirth, with young doubling in length during thefirst year (to 135–155 cm STL) and continuedrapid growth over the second year (Simpfendorferet al., 2008). Recent research has shown thatneonates are found in estuarine nurseries wherethey remain for up to three years (Poulakis et al.,2011, 2013). While in the nurseries juveniles use

PRISTIS PECTINATA DISTRIBUTION IN THE WESTERN ATLANTIC 761

Copyright # 2014 John Wiley & Sons, Ltd. Aquatic Conserv: Mar. Freshw. Ecosyst. 24: 760–776 (2014)

spatial hotspots, react to changing environmentalconditions (e.g. freshwater inflow-related effects),and exhibit ontogenetic niche shifts (Simpfendorferet al., 2010, 2011; Poulakis et al., 2011, 2013).Juvenile P. pectinata have been observed inbrackish, estuarine, and marine habitats insouth-western Florida, while adults have typicallybeen observed in open-water marine habitats(Poulakis and Seitz, 2004; Poulakis et al., 2011)and are presumed to be wider ranging in theirmovements and habitat use. Specific habitat use oflarge juveniles and adults is less well known.

Since previous studies have focused primarilyon small juveniles, required management actions(e.g. designation of critical habitat) have only beentaken on these early life stages (NMFS, 2009a;Norton et al., 2012). Although juvenile P. pectinatahave been studied scientifically in localized regionsof Florida, the broader scope of the NationalSawfish Encounter Database (NSED) can provideinformation regarding all life stages over the entirerange of the species. The combined database, whichhas up to five times the number of reports used inprevious analyses, provided an opportunity to morefully analyse spatio-temporal distribution patternsby life stages as identified in the Smalltooth SawfishRecovery Plan (NMFS, 2009b), especially forlarge juveniles and adults for which requiredmanagement actions are still needed. Wiley andSimpfendorfer (2010) used the NSED to illustratethe inverse relationship between sawfish size andextent of their northern distribution, describeimportant habitat for the species, and illustrate therelationship between sawfish size and water depththey were observed in. The current study aims toexpand the findings of Wiley and Simpfendorfer(2010) by statistically modelling the observedrelationships, test for other factors that may affectP. pectinata distributions from a larger and morecurrent dataset, and predict critical habitat for largejuveniles and adults.

Specifically, the aims of this study were to modelP. pectinata encounter data to facilitate futureanalysis of the US Smalltooth Sawfish RecoveryPlan delisting/downlisting criteria regarding threebroad categories of data: (1) identify spatio-temporalpatterns among encounter reports, including rangereduction in the western Atlantic; (2) determine

current distribution to identify areas and timeperiods where conservation and recovery effortscould be focused; and (3) identify and describespatio-temporal distribution patterns of largejuveniles and adults.

METHODS

National Sawfish Encounter Database (NSED)

Initially, the NSED was maintained by researchersat Mote Marine Laboratory to gather informationregarding P. pectinata encounters, both recent andhistorical. In September 2008 the NSED wasformally transferred to the Florida Program forShark Research at the Florida Museum of NaturalHistory, University of Florida, where additionalscientific and private databases were subsequentlymerged including the Florida Museum ofNatural History database (primarily historicaland museum-based records), two databases fromthe Florida Fish and Wildlife ConservationCommission (research and encounter reports),and two databases from private sawfish researchersM.T. McDavitt (historical, recent, and museumrecords) and J.C. Seitz (historical and recent records).Bottom longline and trawl fishery observer recordsprovided by NMFS collaborators were alsoentered into the NSED. As a result, it is believedthat most of the existing US sawfish records havebeen incorporated into the NSED. Further,encounters have been researched globally since thedatabase’s inception (Wiley and Simpfendorfer,2010), including analyses of both Atlantic species(NMFS, 2010). Largetooth sawfish encounters wereresearched extensively throughout the AtlanticOcean (Fernandez-Carvalho et al., 2013), andencounters of both species were incorporated intothe database. Consequently, the database is nowreferred to as the International Sawfish EncounterDatabase (ISED).

Encounter data collection is ongoing withencounters voluntarily reported by researchers andthe general public or extracted from scientificliterature, print, and online resources. Encounterreporters are asked a series of questions about theirencounter(s) which include the date and location ofthe encounter, estimated total length (ETL), and

J. D. WATERS ET AL.762


habitat characteristics. Interview questions are alsoused to verify that the animal reported was asawfish, with many encounters includingphotographic evidence (see Seitz and Poulakis,2002; Poulakis and Seitz, 2004; and Wiley andSimpfendorfer, 2010 for details on interviewmethods). In addition, the NSED is regularlychecked for potential duplicate encounters; whenidentified, duplicates are removed and the data areintegrated into one record.

To provide a contemporary perspective of theP. pectinata population and recovery, recordsfrom January 2001 (when the NSED was formallyinitiated) to December 2011 were used in theseanalyses over the scale of the western AtlanticOcean. The data used in this study were obtainedfrom the NSED on 14 January 2012. Life stagedefinitions and spatial recovery units (i.e. recoveryregions) defined in the Smalltooth SawfishRecovery Plan were used in this analysis tomaximize direct application to specific recoverycriteria and ongoing recovery planning (NMFS,2009b). The following database fields were used asparameters in the models: month, year, county,

recovery region, latitude and longitude in decimaldegrees based on a geolocation qualitativeconfidence scale (see below), water depth in metres,estimated or measured stretched total length incentimetres, and life stage as defined in theSmalltooth Sawfish Recovery Plan (NMFS, 2009b).

The Smalltooth Sawfish Recovery Plandesignated recovery regions throughout thehistorical range of the species in the USA to assistin management of the species, provide a spatialreference in which to determine the extent ofrecovery, and provide guidelines required fordelisting or downlisting the species under the ESA(NMFS, 2009b). Designation of recovery regionsaccounted for biogeographic boundaries andinformation concerning the historical and currentdistribution of P. pectinata, extending offshore tothe United States’ Exclusive Economic Zoneboundary (Figure 1). All assigned encounterlocations were based on a qualitative GeolocationConfidence Scale (GCS), ranging from zero whenno geographical information was provided to sixwhen an observer provided the exact latitude andlongitude of the encounter from a GPS. Seasons

Figure 1. Study area map including the eight recovery regions in Florida (E to L; NMFS, 2009b) and four areas of interest occupied by large juvenileand adult Pristis pectinata in Florida as identified using kernel density analyses. Area 1 includes Charlotte and Lee counties, Charlotte Harbor, theCaloosahatchee River, and Sanibel Island (Figure 5); area 2 includes Collier and Monroe counties and the Ten Thousand Islands (Figure 6); area 3includes Monroe county, Everglades National Park, Florida Bay, and the Florida Keys (Figure 7); and area 4 includes Broward, Palm Beach,

Martin and St. Lucie counties (Figure 8).



were defined as spring (March–May), summer(June–August), autumn (September–November),and winter (December–February). Estimated totallengths of individuals were used to assign lifestages, as defined in the Recovery Plan, wherein:very small juveniles=< 100 cm, smalljuveniles= 100–200 cm, large juveniles= 201–340cm, and adults => 340 cm (NMFS, 2009b). Datawere also aggregated by ‘Water Body’ as occurringin either the Gulf of Mexico or Atlantic Ocean. Theseparation of Atlantic and Gulf of Mexico watersfollowed US Highway 1 from Key Largo throughKey West, with points north and west of US 1considered ‘Gulf of Mexico’ and south and eastconsidered ‘Atlantic Ocean’. At Key West, theseparation occurred from Boca Grande Key (24.55°N, 81.81°W) with points north considered ‘Gulf ofMexico’ and south considered ‘Atlantic Ocean’,continuing to the southern Marquesas Keys and thesouthern point of Garden Key (Dry Tortugas)(24.55°N, 82.16°W).

Data analysis

Generalized linear models (GLMs) and generalizedadditive models (GAMs) were run to determine whichmodelling framework best described NSED datapredicting the relationship between P. pectinata lengthin relation to predictor variables. GLMs were usedfor analyses in this study and GAMs were usedprimarily as a descriptive tool to assess the linearityof continuous explanatory variables. Due tolimitations of encounter data, namely that noabsence data are included, presence/absence couldnot be modelled. Instead, geographic locations wereused to indicate current geographic range.

GLMs and GAMs are statistical models used torelate responses to combinations of predictorvariables. While GLMs fit parameters assuming alinear relationship, GAMs provide a frameworkfor many commonly encountered statistical modelsusing a flexible generalization of ordinary linearregressions that allows for the response variablesto have non-Gaussian distributions. GAMs caneasily become overfitted; therefore the use ofGLMs is preferable unless GAMs substantiallyimprove the predictive ability of the models. In thecase of continuous variables, GLMs assume that

the variables are linear with predictor variables(in this case the logit) and such linearity wasassessed by categorizing continuous variables bytheir quartiles, as described by Hosmer andLemeshow (2000), and by analysing GAM plots.If evidence of non-linearity was observed,multivariate fractional polynomial transformationswere calculated and used in models instead oforiginal values, following the methods developedby Royston and Altman (1994) and recommendedby Hosmer and Lemeshow (2000). Fractionalpolynomial transformations complicate directinterpretation but provide a mathematicalequation that can be used for prediction and canbe back-transformed to allow interpretation and topredict parameters directly, similar to usingGAMs. Regarding categorical variables, GLMsassume all levels of covariates have sufficientinformation to allow contrasts in the data andachieve model convergence. These assumptionsfollow contingency table and chi-square testassumptions, in which contingency tables shouldnot have cells with zero values or more than 25%of the cells with less than five predicted values.These assumptions were validated by buildingcontingency tables for all categorical variables.

To determine if the number of P. pectinatareported differed from the null hypothesis thatthe number of individuals encountered did notdiffer within levels of the parameters assessed,contingency tables were constructed and chi-squaretests were analysed. Contingency tables with thenumber of individuals reported by Year/Season/Month and Water Body/Recovery Region/Countywere analysed. When testing temporal covariates,expected values tested were equal to the nullhypothesis that parameter level values did notdiffer. When testing the Recovery Region spatialcovariate, expected values were tested relative to thearea (km2) of regions, as calculated using a shapefilein ArcGIS (ESRI, 2011), to determine if valuesvaried in proportion to the area of each region.

For variable selection in the models, a manualmethod was used. For all models, relativegoodness-of-fit was evaluated by comparing thecorresponding Nagelkerke coefficient ofdetermination value (R2) (Nagelkerke, 1991) andAkaike Information Criterion (AIC) values



(Akaike, 1973). In nested models, likelihood ratiotests were used to determine if models weresignificantly different, and if so, the modelpossessing the lowest AIC value was selected,which incorporates parameter degrees of freedomto test for overfitting. First, a null model wasconstructed and subsequent nested models weretested. Univariate significance of each explanatoryvariable was determined by the Wald statistic andlikelihood ratio tests, comparing each univariatemodel with the null. Significant variables wereused to construct a simple effect multivariateGLM, with non-significant variables (at the25% level) eliminated consecutively. Next, simpleeffects multivariate models were tested using thesignificant covariates from the univariate modelswith covariates significant at 10% in thismultivariate approach retained. Covariates thatwere removed from univariate models were thenre-tested in the multivariate model framework, asrecommended by Hosmer and Lemeshow (2000),in order to determine significance within theframework of a multivariate model. Once a finalmultivariate simple effects model was obtained,each pair of possible first degree interactions wastested. First degree interactions were retained ifsignificant at the 5% level.

Four candidate models were run for predictingthe expected P. pectinata lengths, using differentcombinations of candidate explanatory variables.For all models, the error was assumed to follow agamma distribution and the log-link function wasused. A Gaussian distribution was tested but wasdiscarded due to the non-normality of the responsevariable. Only data with GCS values of 2 orgreater were used, representing data that could beassigned coordinates by averaging distancesbetween two extremes and could be placed withina county or Recovery Region with confidence.Generally, these encounters could be assigned to aspecified area, but exact placement was not alwayspossible. Data with GCS values of 2 or greaterwere selected to provide confidence regarding thelocations of encounters to model at county andrecovery region scales in GLMs and GAMs. Seasonand Recovery Region variables were collinear withMonth and Latitude/Longitude, respectively. Thesimplified Season variable was used instead of

Month since some months had few encounters,causing convergence problems in the models. Twoapproaches were possible at this initial modellingstage regarding location, namely using RecoveryRegion versus using Latitude and Longitude in thecandidate models, which were formulated as:

Model 1: Size ~ Season +Year +Depth +RecoveryRegion

Model 2: Size ~ Season+Year +Depth+Latitude +Longitude

The simple effects model (Model 1) used threediscrete (Season, Year, and Recovery Region) andone continuous (Depth) explanatory variables,which according to the GAM plot was non-linear(Figure 2). Because the effects of the Depthvariable were non-linear, the continuous fractionalpolynomial transformation of Depth was modelledwithin a GLM approach and compared with theresults in terms of candidate models' goodness-of-fit to both original Depth values and Depth withina GAM modelling approach using non-parametricsplines. Fractional polynomial transformed Depthwas found to best fit the data and was notsignificantly different than the GAM predicted Depthvariable. The fractional polynomial transformationof Depth resulted in two terms where:

Depth1 = ((Depth/10)^-1)

Depth2 = ((Depth/10)^-1 * log (Depth/10))

The depth transformations were then applied tothe original simple effect GLM model (Model 1)that was formulated as:

Model 3: Size ~ Season+Year +Recovery Region+Depth1 +Depth2

In this simple effects model, possible firstdegree interactions were tested for significance. Byusing the Wald statistic and likelihood ratiotests, it was determined that the significant firstdegree interactions that should be incorporatedinto the model were between Recovery Regionand Season and between Recovery Region andDepth. A final GLM with significant interactionswas formulated as:



Model 4: Size ~ Season+Year +Recovery Region+Depth1 +Depth2 +Recovery Region:Season +Recovery Region: Depth1 +Recovery Region: Depth2

Statistical analyses were performed using the Rlanguage for statistical computing version 2.14.1(R Development Core Team, 2011). Contingencytables were constructed and tested using thelibrary ‘gmodels’ (Warnes, 2007). In GLMs,fractional polynomials were created for the Depthfield using the ‘mfp’ library (Ambler and Benner,2008), while for GAMs non-parametric splineswere used from the ‘mgcv’ library (Wood, 2004).

To identify patterns in spatio-temporaldistributions of large juveniles and adults,encounters were plotted on maps using ArcGIS(ESRI, 2011). Areas of disproportionately largenumbers of encounters of large juvenile and adultindividuals were identified through kernel densityestimation (Silverman, 1986; Worton, 1989) usingthe Spatial Analyst’s kernel density tool inArcMap (ESRI, 2011). Hotspots, as referred to inthis paper, refer to areas of higher numbers of

encounters in relatively small areas. This term doesnot refer to an index of abundance as no effort datacould be incorporated into the analyses. TheArcMap kernel density tool produces a PercentageVolume Contour (PVC), in which 100% representsa minimum convex polygon encompassing 100% ofthe data points and decreasing percentages highlightrelatively small areas with higher numbers ofencounters than average areas. The specific PVCvalues used were selected to highlight the greatestnumber of localized regions, or ‘hotspots’, inwhich high numbers of encounters were reported insouth-west Florida. Encounters within thesehotspots were then selected in GIS and the numberof individuals was analysed by Season to test fortemporal patterns in habitat use.

RESULTS

On 14 January 2012, 5939 reports of 9780 individualswere in NSED records from the western Atlantic(1782–2011), of which 4945 reports of 8773individuals were recorded as P. pectinata by picture,

Figure 2. Effects of the explanatory variables (Season, Year, Recovery Region, and Depth) in a multivariable generalized additive model, for analysingthe effects of the categorical and continuous response variables. The continuous variable Depth, using a smoothing spline transformation, shows a non-

linear effect on the expected smalltooth sawfish sizes.



rostral tooth count, or other described characteristics.Western Atlantic encounters recorded as P. pectinatawere included from the Bahamas, Bermuda, Brazil,Colombia, Cuba, Guyana, Mexico, Nicaragua,Panama, Trinidad, the USA, and Venezuela.Within the USA, a total of 4898 reports of 8711individual P. pectinata were encountered fromTexas (73 individuals encountered), Louisiana (11),Mississippi (8), Alabama (16), Florida (8537),Georgia (3), South Carolina (19), North Carolina(23), Virginia (3), Maryland (2), New Jersey (4),New York (1), and unknown localities withinthe US (11). In the western Atlantic, excludingFlorida, from 2001 to 2011, encounters withphotographic evidence of P. pectinata were reportedfrom Georgia (one individual encountered) inthe USA and internationally in the Bahamas (8),and Cuba (2). However, very small and smallP. pectinata (with affirmed photographs) wereobserved only in Florida waters in the USA andthe Bahamas over the study period, with encountersof very small and small P. pectinata withoutphotographic evidence reported from Alabama(one individual encountered), Louisiana (3), andTexas (2).

In total, 2846 Florida records were found tocontain the fields used in the models occurring fromJanuary 2001 to December 2011, representing 3574individuals encountered. Encounters occurred whilereporters were conducting a variety of activities,with the number of encounters nearly evenlydistributed among captures (50.3%) and sightings(49.7%). Most encounters occurred while reporterswere recreational fishing (60.9%). Other activitiesthat generated encounters included scientificresearch (18.5%), undetermined activities (9.5%),diving (5.0%), boating (3.1%), kayaking (1.7%),commercial fishing (1.1%), and snorkelling (0.2%).

The number of P. pectinata reported differed byYear, Season, Water Body, Recovery Region, andCounty. The majority of encounters in Floridaoccurred in the Gulf of Mexico (n= 3056; 85.5%)in the spring and summer seasons (69.0%). Thenumber of individuals reported varied by Year(ranging from 177 encountered in 2001 to 457 in2009 (Figure 3)), Season (ranging from 459encountered in autumn to 1352 in spring), WaterBody (518 encountered in the Atlantic Ocean and

3056 encountered in the Gulf of Mexico), RecoveryRegion (ranging from six encountered in region Lto 1485 in region G), and County (ranging fromzero encountered in four coastal counties to 1369encountered in Monroe County). Monroe Countyrepresented 38.3% of all P. pectinata encountered(Recovery Region I and part of H), followed by29.8% in Lee County (Recovery Region G), 14.3%in Collier County (Recovery Region H), and 9.8%in Charlotte County (Recovery Region G), withthe remaining 7.8% of individuals encountered in27 coastal counties throughout Florida.

Spatio-temporal covariates differed among lifestages encountered. The number of individualsencountered differed relative to Life Stage byYear, Season, and Water Body. Recovery Regionand County could not be tested due to zero valuesor too many cells with less than five individualsencountered. The following factors were notindependent of Life Stage observed: Year, Season,and Water Body (chi-squared tests for independence,χ2= 148.39–589.94, P≤ 0.05, df=3–33). Whether anindividual was juvenile or adult was not independentof Water Body or Recovery Region (chi-squaredtests for independence, χ2=42.25–569.74, P≤0.05,df=1–11). Recovery Region F was found to be thelargest recovery region in Florida encompassingapproximately 20% of Florida waters, followed byG (17.5%), H (16.4%), L (16.1%), I (10.4%),

Figure 3. Annual number of Pristis pectinata reported to the NationalSawfish Encounter Database in Florida by life stage (as defined by theRecovery Plan, NMFS 2009b) from January 2001 to December 2011(n=number of sawfish). Very small juveniles< 100 cm, small juveniles100–200 cm, large juveniles 201–340 cm, and adults >340 cm estimatedtotal length. Individuals were reported from all life stages every yearindicating consistent occurrence and recruitment in Florida over the

last decade.



K(10.2%), E (7.4%), and J (1.9%). Season ofoccurrence was not independent of the Water Bodythe animal was observed in (chi-squared tests forindependence, χ2 = 25.05, P≤ 0.05, df= 3).

The results of the candidate models tested in thisstudy are presented in Table 1. Model 4, the mostcomplete model using the fractional polynomialtransformed Depth variable and significant firstdegree interactions, was found to have the bestgoodness-of-fit. In this final model, and in terms ofsimple effects, Recovery Region and Depth werefound to be the most important explanatoryvariables, while Year and Season contributed lessto the model but were still significant (Table 2). Aresidual analysis was completed to validate thefinal model. Plots of the residuals versus fittedvalues, as well as the quantile–quantile (Q-Q) plot,showed that the final model was adequate in termsof residuals (without any major outliers) and therewas relatively homogeneous variance along thedata (Supplementary Materials, Figure 1).

Year was found to be a significant factor inpredicting the length of P. pectinata reported.Expected estimated total length remained nearlyconstant between 2001 and 2005 and decreased after2005. Overall, from 2001 to 2011 the number ofindividuals reported to the NSED steadily increaseddespite a few years of decreased reporting. Thenumber of small and very small juveniles encounteredremained constant or increased over this decade andthe number of large juveniles and adults declined orremained constant (Figure 3). During the spring andsummer seasons, expected lengths were observed tobe greater than in the autumn and winter seasons.

From the models, shorter individuals wereexpected in western Florida in regions E, F, G, andH, with an increase in expected lengths observed innorth-east Florida between regions I, J, K and L.There was an increasing average individual lengthfrom western Florida to north-eastern Florida fromregion E through L for all seasons, except duringthe summer when a peak of larger individuals wasreported in north-west Florida in Recovery Region F(Figure 4(a)). Estimated length increased withincreasing depth for all recovery regions, except inRecovery Region J (off south-east Florida) wherelarger specimenswere found at all depths (Figure 4(b)).

Based on the data, the current range of thespecies was found to be limited to Florida waters,which contains Recovery Regions E to L (NMFS,2009b) and 35 coastal counties (Figure 1), adjacentwaters, and the Bahamas. Maps of encounters bylife stage indicate five spatial hotspots with highnumbers of encounters of large sawfish (Figure 1),

Table 1. Candidate generalized linear models for predicting Pristispectinata expected sizes as a function of various candidate explanatoryvariables, with a comparison of the model’s goodness-of-fit in terms ofcoefficient of determination value (R2) and Akaike InformationCriterion (AIC) values. Model 4 was the most complete (final) candidatemodel.

Model R2 AIC

Model 1 27.9% 33499Model 2 23.6% 33639Model 3 51.2% 32433Model 4 53.6% 32366

Table 2. Deviance table analysis of the final generalized linear model (Model 4) estimating Pristis pectinata total length as a function of Season, Year,Recovery Region, Depth (transformed using fractional polynomials), and their significant interactions. Data in this model represent 3574 P. pectinatareported in Florida from January 2001 to December 2011.

Model parameter df Deviance rdf rdev P-value

Null Model 2845 1006.15Season 3 4.801 2842 1001.35 < 0.001Year 10 19.022 2832 982.32 < 0.001Recovery Region 7 172.587 2825 809.74 < 0.001Depth1 1 36.825 2824 772.91 < 0.001Depth2 1 237.843 2823 535.07 < 0.001Season:Recovery Region 18 10.786 2805 524.28 < 0.001Recovery Region:Depth1 7 9.144 2798 515.14 < 0.001Recovery Region:Depth2 7 3.545 2791 511.6 0.013

df = degrees of freedom used for each variable.rdf = residual degrees of freedom.rdev= residual deviance.



including southern Charlotte Harbor (Figure 5), theTen Thousand Islands area (Figure 6), Florida Bayand the Atlantic side of the Florida Keys (Figure 7),and off St. Lucie in south-eastern Florida(Figure 8). In the southern Charlotte Harbor area,both large juveniles and adults were typicallyobserved off Sanibel Island in summer. In the TenThousand Islands, large juveniles and adults wereobserved from March to August, with a few largeindividuals present throughout the year. In westernFlorida Bay, most large juveniles and adults wereobserved in March, while in the Florida Keys, largejuveniles and adults were observed year-round witha peak in April. Off south-eastern Florida, largejuveniles and adults exhibited two peaks inencounters with the first peak occurring in June andthe second in September.

DISCUSSION

The current study analysed the occurrence ofP. pectinata over the entire known range of thespecies, utilizing the largest number of sawfishencounters of any published study to-date. Whilethe species has been observed historicallythroughout much of the western Atlantic, thespecies was recently observed almost exclusively inthe north-west Atlantic off Florida and theBahamas. The results presented here are stronglytied to US recovery plan objectives, providingmodel analysis and prediction potential for all eightrecovery regions in Florida (representing 57% of allrecovery regions in the USA), and represent animportant step towards the identification of criticalhabitat for large juveniles and adults in the USA.

Figure 4. Interaction plots between Recovery Region and Season (a) and Depth (b). (a) Increasing average estimated total length from region E to L(from west Florida to north-east Florida; Figure 1) for all seasons except during the summer when a peak of reports of larger animals occurs along thewest coast of Florida (region F). (b) Smalltooth sawfish length increasing with increasing depth (using 10% percentile depth classes) for all recovery

regions, except Recovery Region J (off south-east Florida) where larger specimens were found at all depths.



In this study, the number of P. pectinata reportedand encountered differed according to spatial andtemporal factors. The majority of encounters(92.4%) were juveniles reported off south andsouth-west Florida (Recovery Regions G, H,and I), corroborating the results of previous studiesand supporting previously identified nursery areas(Seitz and Poulakis, 2002; Poulakis and Seitz, 2004;NMFS, 2009a; Wiley and Simpfendorfer, 2010;Norton et al., 2012). In addition, very smalljuveniles were observed every year over the studyperiod in Florida waters in the current study(Figure 3). Although the encounter data are not

standardized, the presence of this life stage indicatesthat consistent juvenile recruitment occurredannually along the Gulf coast of Florida (RecoveryRegions G and H) from 2001 through 2011.

An increase in expected lengths was observedfrom south to north on the east coast of Florida(Recovery Regions I, J, K and L). Wiley andSimpfendorfer (2010) found that within Floridathere was an inverse relationship between size andnorthern distribution, with the smallest animalsoccurring the farthest north on the east and westcoasts and having the broadest distribution range.The current results support the finding that smaller

Figure 5. Area of interest 1 (see Figure 1 for reference) in western Florida (Recovery Region G). Kernel density analyses indicated that large numbersof large juvenile and adult Pristis pectinata were observed in southern Charlotte Harbor, including the Caloosahatchee River and waters off Sanibel

Island.



Figure 6. Area of interest 2 (see Figure 1 for reference) in south-west Florida (Recovery Region H). The Ten Thousand Islands area was identifiedthrough kernel density analyses to contain large numbers of large juvenile and adult Pristis pectinata. Large juveniles and adults were reported in

this region throughout the year, suggesting that this area has a resident population of these life stages.

Figure 7. Area of interest 3 (see Figure 1 for reference) in southern Florida (Recovery Regions H and I). Florida Bay and the Atlantic side of theFlorida Keys were identified through kernel density analyses to contain large numbers of large juvenile and adult Pristis pectinata. The Florida Bayhotspot is used primarily by adults in the spring. The hotspot along the Florida Keys consists of large juveniles and adults throughout the year,

suggesting that this area has a resident population of these life stages.



animals have a broader distribution range,however, the likelihood of encountering a largerindividual was found to be greater farther norththan was expected in the south. This discrepancymay have been an effect of increased sample sizefrom Florida waters (2846 and 1004 encounters,respectively) and an increased proportion of largerlife stages in the current study than during theWiley and Simpfendorfer (2010) reporting period(1998–2008).

Pristis pectinata were not reported uniformlythroughout Florida; they were observed primarilyin south and south-west Florida (from Recovery

Regions G to I). Records illustrate a gap inP. pectinata reported from Biscayne Bay north toPalm Beach County (Recovery Region J), withmany encounters reported south in the FloridaKeys (Recovery Region I) and in north-eastFlorida (Recovery Region K). This discrepancy,including the presence of the large juvenile andadult hotspot near St. Lucie, may be due to thepresence of a mosaic of habitats such as coastalreefs, a productive estuary, and the St. Lucie Riverwhich is a major freshwater source in the regionflowing into Lake Okeechobee, the largest lake inFlorida (Sime, 2005). However, more information

Figure 8. Area of interest 4 (see Figure 1 for reference) in south-eastern Florida (Recovery Region K). The St. Lucie River area (Martin County) wasidentified through kernel density analyses to contain large numbers of large juvenile and adult Pristis pectinata. Large juveniles and adults were

reported in this region throughout much of the year, suggesting that this area has a resident population of these life stages.



on adult movements and habitat use is needed. Thishas management implications as the Recovery Plandictates that encounters in historical regions mustincrease for delisting or downlisting to occur.Delisting/downlisting may be delayed if outreach,encounter reporting, and research are not promoted.

The majority of very small and small juvenileP. pectinata were encountered in shallow coastalhabitats, while large juveniles and adults tended tobe encountered more frequently in coastal habitatsor open water (Seitz and Poulakis, 2002; Poulakisand Seitz, 2004; Wiley and Simpfendorfer, 2010;this study). Recent research has shown localizedconsistent annual recruitment and use of specificportions of nurseries or ‘hotspots’ for long timeperiods (Poulakis et al., 2011). Movementsbetween nursery hotspots have occurred whenenvironmental conditions change (Poulakis et al.,2013). The increased sample size and expandedtime series presented here illustrates that consistentannual recruitment occurred more broadly insouth and south-west Florida nurseries over thelast decade. Collectively, these data suggestthat: (1) although the population is threatened,consistent recruitment may have played animportant role in the high genetic diversityidentified in this species (Chapman et al., 2011);and (2) small areas within the current range of thespecies were disproportionally more importantthan other areas for multiple life stages. Verysmall and small juvenile hotspots have been shownto be associated with freshwater inflow-relatedparameters such as salinity in their nurseries(Simpfendorfer et al., 2011; Poulakis et al., 2013).More research is needed regarding the largejuvenile and adult hotspots identified in this study.However, the hotspots identified in this study maybe used to increase large juvenile and adultresearch initiative success and better identify areasand factors important for habitat use in thispoorly understood portion of the population.

In the final model (Model 4), the significantinteraction between Recovery Region and Seasonmay be related to the hypothesized northwardmovements of large sawfish in the warmer months(Bigelow and Schroeder, 1953). The analyses inthe present study showed that average lengthincreased from south-west Florida through

north-east Florida (Recovery Regions E through L)for all seasons, except during the summer when apeak of larger individuals was expected fromSarasota Bay to Cedar Key (Recovery Region F).In the Ten Thousand Islands, the occurrence oflarge juveniles and adults coincided with peakrecruitment of very small and small juveniles alsoobserved during the spring (Poulakis et al., 2011).The presence of a few large individuals observedthroughout the year also suggests possiblelong-term residence in the Ten Thousand Islands. Inthe Florida Keys, the occurrence of large juvenilesand adults was greatest in the spring whenparturition peaks (Poulakis et al., 2011). In FloridaBay, researchers from the Florida Program ofShark Research have caught adults in shallowwaters (< 3 m) with recently attained elongatedscratch marks and healed scars, presumed to behave been acquired during copulatory orpre-copulatory activities or male–male interactionsassociated with mating. Preliminary telemetrydata show that these adults remain in Florida Bayfrom June through August, after which they leavethe area (G. Burgess and Y. Papastamatiouunpublished data). Off the south-east coast ofFlorida, the occurrence of large juveniles andadults exhibited two peaks in encounterssuggesting that P. pectinata either may not movefar once they arrive at the area or they move toother regions during mid-summer and return tothe area in the autumn. Although foraging mayalso play a role in the occurrence patterns oflarger sawfish, reproduction-related factors (suchas mating and parturition) likely play a majorrole. The results of the present study providepredictions of the locations and timing of largejuvenile and adults, however, since the databasecould not be standardized it was not possible todefinitively illustrate hotspot use and otherhotspots may be present. Further research,including tagging studies (i.e. acoustic tags andsatellite tags), is needed to test these hypothesesand determine the spatial and temporal scales onwhich these movement patterns occur.

The present study was a collaborative effortamong a number of researchers and institutionsthat collected encounter data on this speciesthroughout its range as well as voluntary reporting



from citizen scientists. Although the analysis ofencounter datasets can be useful, particularly inthe establishment of range and distribution, caremust be taken not to interpret the data beyondappropriate expectations. The models used in thisstudy rely on the assumption that the data areindependent, which was largely not the case forthis dataset. However, the NSED is the mostcomplete dataset in the world regarding thedistribution of P. pectinata, and is the mostcomprehensive treatment of this species’ westernAtlantic range. As a voluntary reporting network,there are many factors that affect whether asawfish is encountered and whether it is reportedfor inclusion into the NSED. In the first case,more people are on the water in certain locationsas a result of accessibility, nearby populations, andenvironmental factors such as precipitation,temperature, and sea state. Secondly, encounterreporting is a function of education and outreachefforts, social and political views of encounterreporting, laws regarding P. pectinata, as well asother factors including people’s willingness toreport encounters. As a result, these data areuseful only as a guide and cannot be used toestimate relative abundance.

The models presented here can be used to predictthe length of P. pectinata likely to be encounteredby recovery region, season, and depth. Analyseshave shown that fewer individuals were observedin Recovery Regions E through F on the westcoast and regions J and L on the east coast ofFlorida. These models can also be used toinfluence the timing and location of researchactivities. For example, the final model (Model 4)estimates that in Recovery Region K during thesummer, and when sampling at a depth of 50 m,individuals of 429 cm ETL are predicted to beencountered, with a 95% confidence interval of351–508 cm ETL. These analyses provide importanttools for scientists and resource managers to focusresearch, monitoring, and conservation efforts.

ACKNOWLEDGEMENTS

We gratefully thank everyone that has providedinformation regarding their sawfish encounters and

the continued support of the public to help betterunderstand and conserve this species. We appreciatethe continued forwarding of commercial fisheryobserver data, recreational fishery encounterreports, and fishery independent data by the FloridaFish and Wildlife Conservation Commission, JohnCarlson and Dana Bethea (National MarineFisheries Service, Panama City), and Jason Osborne(Everglades National Park, Homestead). We thankColin Simpfendorfer for his work while at MoteMarine Laboratory. Thanks to Cathy Bester(Florida Museum of Natural History) for herassistance with graphic design of figures andAmanda Frick (NMFS) for her assistanceproducing the recovery region shapefile. We thankDr Brett Presnell (University of Florida, StatisticsDepartment) and Felipe C. Carvalho (Program ofFisheries and Aquatic Sciences, University ofFlorida) for their assistance with the statistics usedin this study. Dr Philip Stevens improved earlierversions of the manuscript. Funding for the salaryof John D. Waters and the maintenance andintegration of the NSED was provided by a grantfrom the NMFS (WC133F-11-SE-3008). Duringpart of this work, Rui Coelho and JoanaFernandez-Carvalho were supported by grants fromFCT, the Portuguese Science and TechnologyFoundation (Ref. SFRH/BPD/40523/2007 andSFRH/BD/60624/2009, respectively).

REFERENCES

Adams WF, Fowler SL, Charvet-Almeida P, Faria V, Soto J,Furtado M. 2006. Pristis pectinata. In IUCN 2012. IUCNRed List of Threatened Species. Version 2012.1. http://www.iucnredlist.org [17 June 2013].

Akaike H. 1973. Information theory and an extension of themaximum likelihood principle. In International Symposiumon Information Theory, Petran BN, Csaaki F (eds).Akademiai Kiado, Budapest; 267–281.

Ambler G, Benner A. 2008. mfp: Multivariable FractionalPolynomials. R package version 1.4.4.

Bigelow HB, Schroeder WC. 1953. Fishes of the western NorthAtlantic, Part 2. Sawfishes, guitarfishes, skates, rays, andchimaeroids. In Fishes of the Western North Atlantic.Memoirs of the Sears Foundation for Marine Research,Yale University: New Haven, CT.

Chapman DD, Simpfendorfer CA, Wiley TR, Poulakis GR,Curtis C, Tringali M, Carlson JK, Feldheim KA. 2011.Genetic diversity despite population collapse in a criticallyendangered marine fish: the smalltooth sawfish (Pristispectinata). Journal of Heredity 102: 643–652.



CITES (Convention on International Trade in EndangeredSpecies of Wild Fauna and Flora). 2007. Fourteenthmeeting of the Conference of the Parties The Hague(Netherlands), 3–15 June 2007. CoP14 Prop. 17.

Eckert SA. 2002. Distribution of juvenile leatherback sea turtleDermochelys coriacea sightings. Marine Ecology ProgressSeries 230: 289–293.

ESRI (Environmental Systems Research Institute). 2011.ArcGIS Desktop 10.0. ESRI: Redlands, CA.

Evermann BW, Bean BA. 1898. Indian River and its fishes. USCommission on Fish and Fisheries 22: 227–248.

Faria VV,McDavittMT,Charvet P,Wiley TR, Simpfendorfer CA,Naylor GJP. 2013. Species delineation and global populationstructure of critically endangered sawfishes (Pristidae). ZoologicalJournal of the Linnean Society 167: 136–164.

Fernandez-Carvalho J, Imhoff JL, Faria VV, Carlson JK,Burgess GH. 2013. Status and the potential for extinction ofthe largetooth sawfish Pristis pristis in the Atlantic Ocean.Aquatic Conservation: Marine and Freshwater EcosystemsDOI: 10.1002/aqc.2394.

FWC (Florida Fish and Wildlife Conservation Commission).1999. Protected species: sawfishes, basking shark, whiteshark, sand tiger shark, bigeye sand tiger shark, manta rayand spotted eagle ray; prohibition of harvest, landing, orsale. Chapter 68B–44.008 (formerly 46–44.008) FloridaAdministrative Code Rule 11: 1222.

Hosmer DW, Lemeshow S. 2000. Applied Logistic Regression,2nd edition. Wiley: New York.

IUCN (International Union for the Conservation of Nature).2006. IUCN Red List of Threatened Species. http://www.iucnredlist.org [17 June 2013].

Nagelkerke N. 1991. A note on a general definition of thecoefficient of determination. Biometrika 78: 691–692.

NMFS (National Marine Fisheries Service). 2003. Endangeredand threatened species; final endangered status for a distinctpopulation segment of smalltooth sawfish (Pristis pectinata)in the United States. Federal Register 68: 15674–15680.

NMFS. 2009a. Critical habitat for the endangered distinctpopulation segment of smalltooth sawfish (Pristis pectinata)in the United States. Federal Register 74: 45353–45378.

NMFS. 2009b. Recovery Plan for Smalltooth Sawfish (Pristispectinata). Prepared by the Smalltooth Sawfish RecoveryTeam for NMFS: Silver Spring, MD.

NMFS. 2010. Status Review of the Largetooth Sawfish (Pristisperotteti). NMFS: St. Petersburg, FL.

Norton SL, Wiley TR, Carlson JK, Frick AL, Poulakis GR,Simpfendorfer CA. 2012. Designating critical habitat forjuvenile endangered smalltooth sawfish in the United States.Marine and Coastal Fisheries: Dynamics, Management, andEcosystem Science 4: 473–480.

Odom RH, Ford WM, Edwards JW, Stihler CW, Menzel JM.2001. Developing a habitat model for the endangeredVirginia northern flying squirrel (Glaucomys sabrinus fuscus)in the Allegheny Mountains of West Virginia. BiologicalConservation 99: 245–252.

Palma L, Beja P, Rodrigues M. 1999. The use of sighting datato analyse Iberian lynx habitat and distribution. Journal ofApplied Ecology 36: 812–824.

PoulakisGR, Seitz JC. 2004. Recent occurrence of the smalltoothsawfish, Pristis pectinata (Elasmobranchiomorphi: Pristidae),in Florida Bay and the Florida Keys, with comments onsawfish ecology. Florida Scientist 67: 27–35.

Poulakis GR, Stevens PW, Timmers AA, Wiley TR,Simpfendorfer CA. 2011. Abiotic affinities andspatiotemporal distribution of the endangered smalltoothsawfish, Pristis pectinata, in a south-western Floridanursery. Marine and Freshwater Research 62: 1165–1177.

Poulakis GR, Stevens PW, Timmers AA, Stafford CJ,Simpfendorfer CA. 2013. Movements of juvenileendangered smalltooth sawfish, Pristis pectinata, in anestuarine river system: use of non-main-stem river habitatsand lagged responses to freshwater inflow-related changes.Environmental Biology of Fishes 96: 763–778.

R Development Core Team. 2011. R: a language andenvironment for statistical computing. R Foundation forStatistical Computing: Vienna, Austria.

Royston P, Altman DG. 1994. Regression using fractionalpolynomials of continuous covariates: parsimoniousparametric modeling (with discussion). Applied Statistics 43:429–467.

Scott FW, Herman TB. 1995. The use of volunteers inecological monitoring. In Ecosystem Monitoring andProtected Areas. Proceedings of the Second InternationalConference on Science and the Management of ProtectedAreas, Herman TB, Bondrup-Nielsen S, Munro N, WillisonJ (eds). Science and the Management of Protected AreasAssociation: Wolfville, NS; 416–423.

Seitz JC, Poulakis GR. 2002. Recent occurrence of sawfishes(Elasmobranchiomorphi: Pristidae) along the southwestcoast of Florida (USA). Florida Scientist 65: 256–266.

Seitz JC, Poulakis GR. 2006. Anthropogenic effects on thesmalltooth sawfish (Pristis pectinata) in the United States.Marine Pollution Bulletin 52: 1533–1540.

Silverman BW. 1986.Density Estimation for Statistics and DataAnalysis. Chapman and Hall/CRC: New York.

Sime P. 2005. St. Lucie estuary and Indian River Lagoonconceptual ecological model. Wetlands 25: 898–907.

Simpfendorfer CA. 2000. Predicting population recoveryrates for endangered western Atlantic sawfishes usingdemographic analysis. Environmental Biology of Fishes 58:371–377.

Simpfendorfer CA, Poulakis GR, O’Donnell PM, Wiley TR.2008. Growth rates of juvenile smalltooth sawfish Pristispectinata Latham in the western Atlantic. Journal of FishBiology 72: 711–723.

Simpfendorfer CA, Wiley TR, Yeiser BG. 2010. Improvingconservation planning for an endangered sawfish usingdata from acoustic telemetry. Biological Conservation 143:1460–1469.

Simpfendorfer CA, Yeiser BG, Wiley TR, Poulakis GR, StevensPW,HeupelMR. 2011. Environmental influences on the spatialecology of juvenile smalltooth sawfish (Pristis pectinata): resultsfrom acoustic monitoring. PLoS ONE 6: e16918.

Thorson TB. 1982. The impact of commercial exploitation onsawfish and shark populations in Lake Nicaragua. Fisheries7: 2–10.

Turner JC, Douglas CL, Hallum CR, Krausman PR, RameyRR. 2004. Determination of critical habitat for theendangered Nelson’s bighorn sheep in southern California.Wildlife Society Bulletin 32: 427–448.

Warnes GR. 2007. gmodels: Various R programming tools formodel fitting. Includes R source code and/or documentationcontributed by Bolker B, Lumley T, Johnson RC. R packageversion 2.14.1.



Wiley TR, Simpfendorfer CA. 2010. Using public encounterdata to direct recovery efforts for the endangeredsmalltooth sawfish Pristis pectinata. Endangered SpeciesResearch 12: 179–191.

Wood SN. 2004. Stable and efficient multiple smoothingparameter estimation for generalized additive models.Journal of the American Statistical Association 99: 673–686.

Worton BJ. 1989. Kernel methods for estimating theutilization distribution in home-range studies. Ecology 70:164–168.

Wueringer BE, Squire L, Collin SP. 2009. The biology ofextinct and extant sawfish (Batoidea: Sclerorhynchidae andPristidae). Reviews in Fish Biology and Fisheries 19: 445–464.

SUPPORTING INFORMATION

Additional supporting informationmay be found in theonline version of this article at the publisher’s web site.



useofencounterdatatomodelspatio ... · nelson bighorn sheep (ovis canadensis nelson; turner et al.,...

Documents