Transcript
Increasing ocean temperatures allow tropical fishes tosurvive overwinter in temperate waters WILL F. FIGUEIRA 1 and DAVID J. BOOTH Fish Ecology Laboratory, Department of Environmental Sciences, University of Technology, Sydney, PO Box 123, Broadway, NSW 2007, Australia Abstract The southeast coast of Australia is a global hotspot for increasing ocean temperaturesdue to climate change. The temperate incursion of the East Australian Current (EAC) isincreasing, affording increased connectivity with the Great Barrier Reef. The survival oftropically sourced juveniles over the winter is a significant stumbling block to polewardrange shifts of marine organisms in this region. Here we examine the dependence ofoverwintering on winter severity and prewinter recruitment for eight species of juvenilecoral reef fishes which are carried into temperate SE Australia (30–37 1 S) by the EACduring the austral summer. The probability of persistence was most strongly influencedby average winter temperature and there was no effect of recruitment strength. Long-term (138 years) data indicate that winter water temperatures throughout this region areincreasing at a rate above the global average and predictions indicate a further warmingof 4 2 1 C by the end of the century. Rising ocean temperatures are resulting in a higher frequency of winter temperatures above survival thresholds. Current warming trajec-tories predict 100% of winters will be survivable by at least five of the study species as far south as Sydney (34 1 S) by 2080. The implications for range expansions of these and other species of coral reef fish are discussed. Keywords: butterflyfish, coral reef fish, damselfish, ocean warming, overwinter survival, range expan-sion, SE Australia, tropical expatriates Received 29 January 2009 and accepted 15 March 2009 Introduction The geographic range of a species is determined bycomplex interactions among ecological factors withinevolutionary processes (Holt, 2003). For a species toexpand its present range, the new area must meet aminimum set of biophysical requirements. It must con-tain adequate habitat and food to support survival of alllife stages and reproduction of adults. The environmentmust also fall within the present range of tolerance of all factors affecting the physiology of the organism(Angilletta et al ., 2002) such as temperature, moisture,oxygen, nutrients and in the cases of photosynthesizingorganisms, sunlight. Lastly the species must be capableof surviving within the biological community (preda-tors and competitors) of the new environment (Preston et al ., 2008). Where these basic criteria are met, rangeexpansions are theoretically possible.While range shifts can occur even without strongdirectional environmental change (Holt, 2003), there isevidence that recent global climate change is responsi- ble for shifts of a large variety of taxa in terrestrial andmarine systems (Parmesan & Yohe, 2003; Parmesan,2006). Many observed range shifts are due to the move-ment of juveniles or adults both on land and in the sea(Attrill & Power, 2002; Crozier, 2004; Perry et al ., 2005;Agatsuma & Hoshikawa,2007), however,marineorgan-isms also have the potential for especially rapid anddramatic range shifts due to their highly mobile larvalphase. Survival of larvae which settle (transition frompelagic to benthic environment) in novel habitat has been associated with range shifts for sea urchins on thenorthwest coast of Japan (Agatsuma & Hoshikawa,2007) and sea anemones on the west coast of the UnitedStates (Sagarin et al ., 1999). 1 Present address: Marine Ecology Laboratories, A11, Centre forResearch on Ecological Impacts of Coastal Cities, School of Biolo-gical Sciences, University of Sydney, NSW 2006, Australia.Correspondence: Will F. Figueira, tel. 1 61 2 9351 2039, fax 1 61 29351 6713, e-mail:will.fi
[email protected] Global Change Biology (2010) 16 , 506–516, doi: 10.1111/j.1365-2486.2009.01934.x 506 r 2009 Blackwell Publishing Ltd There is strong potential for the pelagic larval phaseto drive range expansions of marine organismswherever major ocean currents facilitate long-distancedispersal. Previous work along the SE coast of Australiahas demonstrated the important role of the EastAustralian Current (EAC), a western boundary current,in driving larval settlement pulses of many species of tropical coral reef fish each summer (Booth et al ., 2007).The presence and at least short-term (one season)persistence of these species at these sites which are welloutside of their range indicates that the biophysicalenvironment is suitable for at least the juvenile stages.However, the failure of these species to consistentlysurvive their first winter after settlement indicates that,as has been demonstrated for other species of fish(Ludsin & DeVries, 1997; Hurst & Conover, 2001; Pratt& Fox, 2002), winter is a key bottleneck for survival andpopulation establishment of these fishes.Here we present the results of a 6-year study of theoverwinter survival of several commonly summer-occur-ring tropical coral reef fish species along the SE coast of Australia at six locations, from the Solitary Islands (30 1 S)to Merimbula (37 1 S) including Lord Howe Island (Fig. 1).We examine the roles of the number of newly arrived juveniles before winter (hereafter referred to as ‘prewin-ter’ juveniles) as well as several measures of winterseverity based on water temperature in determining thenumberof fish present after the winter(hereafter referredto as ‘winter survivors’). We then evaluate the potentialfor range shifts of these species as well as their rate inlight of ocean temperature increase at different latitudesusing a long-term (138 years) sea surface temperature(SST) data set. Materials and methods Estimating fish abundance Abundance estimates were derived from data seriescollected by the authors (Lord Howe Island, Sydney, Jervis Bay and Merimbula) and others (Solitary Islands,data courtesy of H. Malcolm) as part of ongoing under-water visual censuses for tropical fish at each of thelocations. These locations were areas in which tropicalfish were known to recruit each summer based uponprevious studies (Booth et al ., 2007). During thesesurveys a single individual on snorkel or SCUBA (de-pending on the site) methodically swam a predefinedarea using either transect tape or local landmarks(above and below water) for reference and countedtropical fish species categorizing them as either juve-niles or as adults based upon size. In all cases censuseswere conducted on fixed areas which did not change inlocation or extent from one survey to the next. The totalarea surveyed for these censuses ranged from 150 to1600m 2 depending upon the location. Surveyed habitatwas shallow ( o 9m) nearshore (or near-island for theSolitary Islands and Lord Howe Island) coastal rockyreef (as described in Booth et al ., 2007). At Long Reef,Shelly Beach and Merimbula, surveys were conductedweekly between December and June from 2003 to 2008.At the other locations surveyswereconducted annually,typically in March or April, from 2002 to 2007 for theSolitary Islands, from 2002 to 2006 for Lord HoweIsland and from 2002 to 2005 for Jervis Bay. Wheremultiple sites were surveyed at one location, valueswere averaged. This was done for the Solitary Islands(14 sites), Lord Howe Island (eight sites), Shelly Beach(two sites) and Jervis Bay (two sites).Using these abundance data we identified species forwhich juveniles were present in at least half of thelocations and which occurred in significant numbers( 4 20 individuals per census on average) at those loca-tions. This was done to ensure adequate sample sizesfor analyses of trends across years and locations. Foreach of these species at each location at which juvenilesoccurred we estimated (1) the number of prewinter juveniles and (2) the number of overwinter survivors.Overwinter survivors were easily identifiable (categor-ized as adults) as they were much larger than the juveniles, which were at most 5–6 months old by theonset of winter. Note that exceptionally warm wintersnever occurred back to back in our data set and thusgiven the typically small number of adults occurring ina census it was possible to assure that adults had onlysurvived one winter (i.e. had settled last season) basedthe absence of any adults previous to the winter inquestion. For locations with weekly surveys, the num- ber of overwinter survivors and the number of juvenilespresent before winter was taken as the average numberfrom the first four and last four surveys of the recruit-ment season (December–June), respectively. On-site temperature logger data Daily average water temperatures for each survey sitewere obtained using data loggers (Stowaway Tidbit; À 5to 37 1 C, Æ 0.2 1 C; Onset (R), Pocassett, MA, USA)maintained in fixed locations adjacent to fish habitatat each location (2–9m deep depending on location). Intwo locations, Jervis Bay and Lord Howe Island, loggerdata series wereshorter thanthe survey period. In orderto establish time series for these locations whichspanned the time series of survey data we evaluatedthe relationship between the logger data which wasavailable for these sites (2 years at both sites) and SSTdata obtained from the Pathfinder AVHRR satellite(details discussed below). There was a very strong WARMING OCEANS AND FISH RANGE EXPANSIONS 507 r 2009 Blackwell Publishing Ltd, Global Change Biology , 16 , 506–516 correlation between logger and SST data from whichseasonality had been removed (Jervis Bay: R 2 5 0.80, P o 0.001; Lord Howe Island: R 2 5 0.90, P o 0.001) withthe offshore SST data being consistently warmer thanthat recorded inshore by the logger. Therefore thePathfinder data were used for each of these locationsin the analyses after being adjusted by the averagedifference to the logger (differences were 0.14 and1.6 1 C for Lord Howe Island and Jervis Bay, respec-tively). The larger difference at Jervis Bay is the result of the logger (and fish habitat)being located inside the Bayas compared with Lord Howe Island where the loggerand the habitat were exposed to much more mixing dueto the oceanic nature of this area. Threshold temperature analysis The type and quality of habitat is very likely to influ-ence observed levels of abundance of the variousspecies at each site. While we attempted to accountfor this by surveying similar habitat types, the exactarea surveyed at each site was slightly different. There Fig. 1 Map of study area. Sea surface temperature data for all locations listed were obtained from both data series as explained in‘Materials and methods’. Visual census surveys for tropical fish were conducted at locations marked by the hollow circles (note twolocations in Sydney, Shelly Beach and Long Reef). 508 W. F. FIGUEIRA & D. J. BOOTH r 2009 Blackwell Publishing Ltd, Global Change Biology , 16 , 506–516 may also be nuances of larval supply that lead todifferent levels of possible recruitment at each locationand also among years, habitat issues aside. Thus inorder to make abundance comparisons across sites andyears,abundance data foreach species werenormalized by the maximum seasonal abundance observed at eachlocation over the course of the study. We then usedgeneral linear models (GLM) to examine the effect of prewinter abundance, location, species and each of fivemeasures of winter severity on the abundance of wintersurvivors. Separate models were run for each of the measures of winter severity, (1) the coldest weekof the year, and four metrics based upon the tempera-ture during the period July–August, (2) average, (3)standard deviation, (4) minimum and (5) maximum.Individual analyses were run because many of thevariables describing winter severity were highly corre-lated. We arcsine square root transformed both theindex of winter survivors and that of prewinter juvenileabundances to meet the assumptions of homogeneity of variance for the GLM.As indicated in the results, average winter temperaturehad the most strongly significant effect of all the severitymetrics in the GLM; however, there was a significantlocation  species interaction ( F 11,87 5 2.8, P 5 0.004).Therefore relationships between winter survivors andaverage winter temperature were explored further forindividual species using a nonlinear sigmoidal model.The model had two parameters; x 0 (the inflection point),and d x (the slope) and had its minimum and maximumfixed at 0 and 1, respectively (in accordance with theabundance indices). Because temperature is inherentlyconfounded with location, we included a dummy vari-able in the model fit to test for an effect of location:Survivors ¼ 1 À 11 þ e Temp Àð x 0 þ b 1  Location Þ d x ; where survivors is the number of fish present afterwinter; Temp the average winter temperature (July–Au-gust); Location the code for each location; x 0 the point of inflection for sigmoidal curve; d x the slope of sigmoidalcurve at the point of inflection; b 1 the coefficient for the Location dummy variable.The model was fit with a least squares loss function in STATISTICA (version 7.1, Statsoft Inc., Tulsa, OK, USA)and where the coefficient on the Location parameter ( b 1 )did not differ significantly from zero, the model was fitfor all locations combined. Where it was significant, themodel was fit to data for each location individually. Long-term SST analysis SST data were obtained for nine locations (Fig. 1) fromtwo sources: NOAA Pathfinder and Met Office HadleyCentre’s HadISST1 data set (Rayner et al ., 2003). NOAAPathfinder data were Level 3 mapped, 4km, 7-day reso-lution, 1985–present, value used was the average of a4  4 cell grid located 10km offshore (Fig. 1) of eachlocation. A point 10km offshore was selected to avoidpoor-quality data which can occur at the extent of theremotely sensed data due to land effects. The Met OfficeHadley Centre’s SST data set was HadISST1, 1 1 , monthlyresolution, 1870–present, value used was for nearest 1 1 cell to each location which did not overlap land.The higher spatial resolution Pathfinder data wereused to visualize concordance in coastal patterns of warming (relative to our sampling locations) over theperiod of the data set and to identify the relativelywarm winters (taken as the average temperature from July to August) experienced in recent years. As indi-cated above, it was also used to generate complete timeseries of temperature data for Lord Howe Island and Jervis Bay. We used the HadISST1 data set to examinethe long-term trends in coastal warming at all oursample locations. We extracted the average wintertemperature (July–August) at all locations (Fig. 1) andcalculated the location-specific temperature changes (in 1 C) based on the comparison of the average of the last10 years of data with that of the first 10 years. We thencalculated the frequency of winters warmer than athreshold value at each location over the course of theentire data set. This threshold temperature value was based on the average value observed from the significantsigmoidal model fits for all the mainland species but wascorrected to account for the fact that the offshore watertemperaturesfrom the HadISST1 dataset are,on average,1.5 1 C warmer than those measured with the loggers atthe study sites due both to the influence of the EAC aswell as the very large offshore area represented by theone HadISST1 data point (approximately 150km 2 ). Thefrequency of winters that were warmer than this thresh-old value was calculated using a running 25-year win-dow centered around the target year. We estimated thetime at which the frequency of warm winters wouldequal 100% in Sydney using a linear fit to the Sydneyfrequency data beginning in 1985 (when the frequency became 4 0). Given climate predictions for SE Australia(Poloczanska et al ., 2007) and the current acceleratingnature of the frequency curve for Sydney, this is aconservative approach. Results Eight species of tropical coral reef fish were observed tocommonly ( 4 20fishlocation –1 season –1 ) settle in at leasthalf of the locations (Fig. 1) during the study period. Sixof these species were damselfishes (family Pomacentri-dae) – Abudefduf bengalensis , Abudefduf vaigiensis , WARMING OCEANS AND FISH RANGE EXPANSIONS 509 r 2009 Blackwell Publishing Ltd, Global Change Biology , 16 , 506–516 Abudefduf sexfasciatus , Abudefduf whitleyi , Pomacentruscoelestis and Stegastes gascoyni – and two were butterfly-fishes (family Chaetodontidae) – Chaetodon auriga and Chaetodon flavirostris . All of these species occur on thesouthern Great Barrier Reef and are generally found inrocky/rubble habitat among and peripheral to coralreefs (Randall et al ., 1997). With exception of S. gascoyni all the pomocentrids are planktivorous and generallyoccur in groups of 5–25 individuals. S. gascoyni areomnivorous and typically found alone defending aterritory (Randall et al ., 1997). The two chaetodontidsare facultative corallivores which inhabit rubble androcky structures and feed on a variety of benthic in-vertebrates with coral being only a minor component of their diet (Cole et al ., 2008). In total these speciesaccounted for 4 97% of the total abundance of alltropical fish species at all locations. Some overwinteringoccurred at all locations except the most southern one inMerimbula (37 1 S) and for all species except A. sexfascia-tus . One or more of the other five species of damselfishwere observed to overwinter at all other locations whilethe two butterflyfishes were observed to overwinteronly at Lord Howe Island.Of the five winter severity metrics used in the GLMs,only average, minimum and maximum winter tempera-tures were significant (Table 1). The number of juvenilespresent before the winter was not a significant factor inany of the models but there were significant site  species interactions in all of them. As indicated in themethods section, because of the site  species interac-tion, we used nonlinear sigmoidal models for each of the species separately to more fully explore the relation-ships between winter survivorsand winter severity.Thesignificance of average, minimum and maximum win-ter temperature in the GLMs was not surprizing giventhat these variables are highly correlated. For thisreason and because it showed the strongest effect basedon comparison of mean squared errors (Table 1), weused the average winter temperature in the sigmoidalmodel fits. Model fits were significant for all speciesexcept A. sexfasciatus (Table 2, Fig. 2d). There was noeffect of location for any of the Abudefduf or Chaetodon species, and threshold temperature values ( x 0 para-meter, point of inflection) were about 17 and 19.5 1 Cfor the two groups, respectively (Table 2). Model fitswere different by location for P. coelestis and S. gascoyni .There was no significant fit to the P. coelestis data fromthe Solitary Islands (Fig. 2g, crosses, Table 2) and boththese species had a higher threshold temperature atLord Howe Island ( $ 19.5 1 C) than at the mainlandlocation where they occurred ( $ 17 1 C).The nearshore temperature profiles from the Pathfin-der SST data set indicated that two of the warmestwinters in this series (since 1985) were in 2001 and 2006,and that this signal is strongly evident at the moresouthern locations: Merimbula (37 1 S) to the SolitaryIslands (30 1 S), only weakly so at Byron Bay (29 1 S) andabsent at One Tree Island (23 1 S) on the southern end of the Great Barrier Reef (Fig. 3a). There is little concor-dance with this latitudinal trend for Lord Howe Island,which is consistent with the fact that the EAC separatesfrom the coast near the Solitary Islands before headinginto the Tasman Sea toward Lord Howe Island (Ridg-way & Dunn, 2003).The HadISST1 data set indicates significant globalwarming of about 0.5 1 C between 1870 and 2000 (Rayner et al ., 2003) and a winter (July–August) warming trend of between 0.7 and 1.5 1 C for the locations explored in thisstudy (Fig. 3b) confirming previous observations that thisarea is a hotspot for increasing ocean temperatures Table 1 Summary of general linear model (GLM) analysis results explaining spatial and temporal patterns of reef fish over-wintering as a function of winter severity, prewinter abundance, location, and speciesWinter variableEffects of variablesFocal variable Other variable’s P -valuesMS F 1,87 P Juveniles previouswinter Location Species Site  speciesColdest week 0.1 0.5 0.468 0.686 0.032 0.248 0.029Average temperature 3.6 33.7 o 0.000 0.423 0.445 0.247 0.003SD temperature 0.0 0.2 0.627 0.677 0.030 0.248 0.030Minimum temperature 3.6 34.0 o 0.000 0.670 0.671 0.249 0.003Maximum temperature 2.9 25.6 o 0.000 0.428 0.329 0.247 0.006Each row gives the results of one of the five GLMs, each using a different measure of winter severity (as indicated in column one).Themean squarederror(MS) and resulting F and P -values aregiven foreach of the primary winter severity variables. Also given arethe P -values of each of the other variables in each of the respective models. The degrees of freedom (numerator, denominator) foreach of the other variables in each model were juveniles (1,87), site (5,87), species (7,87), site  species (11,87). 510 W. F. FIGUEIRA & D. J. BOOTH r 2009 Blackwell Publishing Ltd, Global Change Biology , 16 , 506–516
