SPECIAL FEATURE

Latitudinal Gradients1

In 1772, J. R. Forster sailed with Captain Cook in search of the great southern continent hypothesized by geographers to be necessary to ``balance'' the world. During the three-year trip they ranged from below the Antarctic Circle to the southern tropics, and Forster sampled the ¯ora throughout the Paci®c. On his return, he published his account of the voyage in which he noted that different islands support a variable number of species, which he attributed primarily to climate (tropical islands support more species than Antarctic islands), and secondarily to island ``circumference'' (large islands support more species than small ones). Thus, we have what is probably the ®rst documented geographical pattern in ecology, as well as an explanation for it that differs from some modern explanations only by lacking an evolutionary component. Of course, we now refer to the ®rst part of the pattern as the ``latitudinal diversity gradient,'' and the second as a ``species±area relationship.'' The study of geographic patterns faces two fundamental challenges. First, ecological data across very large scales are not easy to generate. For example, to understand diversity gradients, we need global estimates of species richness for particular taxonomic groups. Given our poor state of knowledge about the diversity of most taxa, we are mostly restricted to the study of a handful of groups or, alternatively, to the geographic regions where we have reasonable databases of species distributions. If this were not problem enough, to explain gradients we also need a wide range of quantitative explanatory variables. Historically, the dominant explanatory variable in- cluded in geographical analyses was latitude, simply because it was the only variable that could be measured (hence the term ``latitudinal gradients''). Although better than nothing, a problem with latitude and other purely geographical variables is that they can only describe a pattern; they cannot explain it. Thus, the literature is replete with possible explanations based on what workers believed to underlie correlations between space and ecology. This leads to the second fundamental challenge in geographical ecology. When faced with a collection of correlations, how do we test hypotheses when it is essentially impossible to conduct controlled experiments? Nearly all studies of broadscale patterns must use statistical techniques to tease apart complex patterns of covariation among suites of variables. Given the challenges geographical ecologists faceЮnancial, methodological, and epistemologicalÐprogress is per- haps slow. Nonetheless, recent major advances have been made. This Special Feature comprises six cutting-edge studies, illustrating where we stand with respect to explanations for broadscale patterns in ecology, highlighting both how much, and in some cases how little, we know. The ®rst four papers focus on the original question in geographical ecology; that is, what factors determine the number of species across large distances. Arguably, we know more about what drives tree diversity than any other group, and Field et al. present a powerful model for under- standing the species richness of woody plants in regions for which we have data and for predicting richness patterns in areas where we do not. They also compare and contrast their model against an alternative generated to explain the global pattern of angiosperm family richness. What is most telling is that both models are based on the realization that elements of climate related to energy and water inputs represent the ®rst-order driving forces of plant diversity. On the other hand, both models are designed to understand how currently existing taxa distribute themselves at broad scales, but they do not explicitly include speciation/extinction dynamics. The interplay between ecological and evolutionary processes represents the largest remaining challenge to a full explanation of diversity patterns, and Cardillo et al. take on this issue. They ®nd, as many workers have suspected, that diversi®cation rates (of birds) are generally higher in tropical com-

1 Reprints of this 68-page Special Feature are available for $10.25 each, either as PDF ®les or as hard copy. Prepayment is required. Order reprints from the Ecological Society of America, Attention: Reprint Department, 1707 H Street, N.W., Suite 400, Washington, DC 20006.

2261 Special Feature swl nobel le h aeo clgcldvriyaddnmc.W a etunderstand best future. had the We predict dynamics. to and now diversity have ecological we of abstract progress. face patterns an the the rapid just alter making than chang- undoubtedly global fact more anthropogenic will other clearly in es and is change, is climate patterns introductions, ecology Species large-scale exercise. geographical generate intellectual that that factors shows the them, Understanding answer understanding to an in used to patterns closer being step geographical one populations. are indeed, of we dynamics If, that be the diversity. may in it than variation rule, other the of how be indicates patterns to this clearly prove global species That dependence that large-bodied about populations. density any arguing for northern know issue dynamics, to we the two compared address population little that to southern in attempt demonstrates in ®rst variability Post the pervasive represents cycles. latitudinal more population is show dependence their also density and species mammals bioge- mammal small and large by has ecology famous this of made general, theory was If global others. simple the or a from diversity generate predicted of to be independently attempts ography. cannot vary for interaction pattern strengths implications that case one the interaction important appears In so herbivores, it Silliman). and densities, its problem, (Pennings and population the predicted as grass of analyses tropics such saltmarsh complexity the of of and toward two span are increase ®nal scope geographic indeed The the dynamics the do climates. Although strengths by population cold issues. limited in allied consequence, effects necessarily these partial abiotic address is and feature a climates the as warm in in papers and effects tropics, biotic by the interactions dominated biotic toward that surmised been stronger long has become It scales. geographic broad across pattern patterns ecological human-in¯uenced intriguing on an based is research This additional diversity. stimulate plant diversity. plants to native of biotic invasive likely that and of one conclude exotic they and diversity both species; conclusion, the native constrain for that forces pattern ®nd environmental the factors al. follows and than closely et richness actually expect Stohlgren America species might However, North exotic in we of activities. humans, determinants human to weaker to due be indirectly related to or effects ecology. directly biotic geographical in are or get introductions environmental can we modern as most experiment effects. that manipulative a Given their to are integrating close gradients) as by comes diversity perhaps made which drive be present can are the progress there only more that Although indicate and or oceans. again wrong, past far well-documented certainly the so less almost results only the the (e.g., data, into explanations deep-sea question single-factor quantitative this the generating answering take in challenges but who extreme analysis, al., their et gradients. in Rex diversity clear of on not research is of be future might the represents this undoubtedly Why question clades. temperate to pared aieseis aiuia rdet;mcoclg;sailpplto yais pce±nrytheory; species±energy dynamics; population spatial macroecology; richness. gradients; species latitudinal species; vasive eepc htti apeo aes lutaigcretqetosadtedt n methods and data the and questions current illustrating papers, of sample this that expect We Holarctic) the (in south to north from systematically vary dynamics population that notion The only the not certainly is richness species attention, of share lion's the captures it Although species, exotic using conducted be can gradients diversity of drivers the of test independent An focus main the also is diversity of drivers as present the and past the between interaction The e od:boiest;gorpia clg;goa omnt clg;itrcinsrnts in- strengths; interaction ecology; community global ecology; geographical biodiversity; words: Key ÐB ᭧ ÐA 05b h clgclSceyo America of Society Ecological the by 2005 onl University Cornell Editor Features Special California±Irvine of University Editor Guest RADFORD NURAG .A A. .H A. GRAWAL AWKINS Ecology, 86(9), 2005, pp. 2263±2277 ᭧ 2005 by the Ecological Society of America

GLOBAL MODELS FOR PREDICTING WOODY PLANT RICHNESS FROM CLIMATE: DEVELOPMENT AND EVALUATION

RICHARD FIELD,1,4 EILEEN M. O'BRIEN,2 AND ROBERT J. WHITTAKER3 1School of Geography, University of Nottingham, NG7 2RD UK 2Gainesville College, Watkinsville, Georgia 30677 USA, and Institute of Ecology, University of Georgia, Athens, Georgia 30602 USA 3Biodiversity Research Group, School of Geography and the Environment, University of Oxford, Mans®eld Road, Oxford OX1 3TB UK

Abstract. There have been few attempts to generate global models of climate±richness relationships, and fewer still that aim to predict richness rather than ®tting a model to data. One such model, grounded on theory (biological relativity to water±energy dynamics) is the interim general model (IGM1) of the climatic potential for woody plant richness. Here we present a second-generation model (IGM2), and genus and family versions of both models. IGM1 describes horizontal climate±richness relationships based on climate station data and systematic species range maps, with IGM2 additionally incorporating vertical changes in climate due to topographic relief. The IGMs are mathematical transformations of empirical relationships obtained for the southern subcontinent of Africa, whereby the re-described regression models apply to the full range of global variation in all independent climate parameters. We undertake preliminary validation of the new IGMs, ®rst by mapping the distribution and relative spatial variation in forecasted richness (per 25 000 km2) across

the continent of Africa, then by evaluating the precision of forecasted values (actual vs. Feature Special predicted) for an independent study system, the woody plants of Kenya. We also compare the IGMs with a recent example of purely statistical regression models of climate±richness relationships; namely, the ``global'' model of A. P. Francis and D. J. Currie for angiosperm family richness. We conclude that the IGMs are globally applicable and can provide a fundamental baseline for systematically estimating differences in (woody) plant richness and for exploring the hierarchy of subordinate relationships that should also contribute to differences in realized richness (mostly at more discrete scales of analysis). Further, we found that the model of Francis and Currie is useful for predicting angiosperm richness in Africa, on a conditional basis (somewhere, sometime); we examined the relationship that it describes between climate and richness. Lastly, we found that indices of available soil water used in ``water-budget'' or ``water-balance'' analyses are not proxies for available liquid water as a function of climatological dynamics. Key words: Africa; climate; climatic potential for richness; diversity gradients; interim general model; Kenya; model evaluation; species richness; water±energy dynamics; woody plant diversity.

INTRODUCTION but have tended not to result in globally applicable models (e.g., Currie and Paquin 1987). Such work tends It has long been known that global gradients in rich- to focus on mid to high latitudes; regions for which ness covary with global gradients in climate. The de- data availability is better, but which contain relatively velopment of statistical models of this relationship that few of the world's species. An exception is an interim apply globally is an important but elusive goal for ecol- general model (IGM) of the ``climatic potential for ogists. In addition to improving our understanding of (woody) plant richness'' (O'Brien 1998). Rather than diversity patterns, such models could prove useful for using a purely statistical approach, O'Brien worked predicting reasonable values of plant or animal richness from ®rst principles to develop a ®rst-order mechanistic where actual values are unknown, and for modeling explanation for covariation between climate and rich- how changes in climate could alter the richness (and ness globally: biological relativity to water±energy dy- vegetation) patterns we see today. namics (O'Brien 1989, 1993, 1998). This idea effec- Attempts to apply models, developed in particular tively links water±energy dynamics (work done by wa- regions, to other regions have produced some success ter) to fundamental parameters of both climatological and biological dynamics, at all scales of analysis: liquid Manuscript received and accepted 16 December 2004; ®nal water and solar energy (e.g., hydrologic cycle and pho- version received 1 February 2005. Corresponding Editor: B. A. Hawkins. For reprints of this Special Feature, see footnote tosynthesis, respectively). In accord with energy's dy- 1, p. 2261. namic relationship with the state of water, the model 4 E-mail: richard.®[email protected] describes the relationship of climate to woody plant 2263 Special Feature aeadrcns aafrtesuhr subcontinent 15 southern (from the Africa for of data richness and mate 1993, (O'Brien insolation of optimal 1998). an and function rainfall of (parabolic) function linear a as richness 2264 iin eeana anal( con- rainfall describing energy'' annual for were ``optimal ditions and variables water'' climate liquid best ``available the that gested sug- analyses Regression vegetation. and climate perate sihrn nIM,tevria etri o.The not. is vector vertical the IGM1, in vector inherent horizontal is the Although oro- and spots'' respectively). (e.g., ``hot refugia, and area rainshadows, an the and of rainfall on graphic richness topography and of climate effects prevailing idiosyncratic the vicariance, in taxonomic and and disjunctions of continental today vectors evident in are vertical effects and The dynamics. horizontal climatological the sea both and land to of relative location the These alter dy- diastrophism). necessarily and geological processes tectonics independent plate to (e.g., due namics change to been began, subject have life dynamics biological terrestrial and Since climatological both also 2004). (see Ricklefs reasons and following Qian the for unsurprising is The last re- richness. genus family the for and weaker of species signi®cantly but strength at levels, identical the almost Africa; this being southern found lationship for (1998) true al. be et to O'Brien to families. apply climate±rich- should and species same genera to the today, applying so, relationships space If ness over time. apply over apply that should dynamics same to fundamental the are model life, dynamics a water±energy improve if as should First, relief ®t. dynamics topographic climatological of in function inclusion changes (2) vertical and time, of over same apply the (1) should dynamics: relationship water±energy biolog- to of relativity implications ical other two investigated have we (for values States). United actual the and both (gradients) of terms differences in relative (United China) world Africa, America, the South in States, elsewhere richness woody species for potential estimates plant climatic reasonable generates the in- richness of resulting for (IGM1) the model that general showed terim southern in (1998) sampled O'Brien range the Africa. outside en- fall where values climates ergy to parabolic extrapolation energy's prevents because function necessary was formation cl 2 0 km 000 (25 scale yptnileaornprto (PET evapotranspiration potential ly pce richness species omds hti ple otefl lblrneof range global full the trans- to applies mathematically in it variation was that southern the so (SAF1) B), formed Appendix model (see data of African such Because of coverage. lack global the ``ac- with similar data de- qualitatively richness relationship of tual'' availability this the of on model pends global based pirically miial,mdldvlpetwsbsdo cli- on based was development model Empirically, ic eeomn fteIM o pce richness, species for IGM1 the of development Since R an n PET and ϰ 2 ri) hc pn rpclt tem- to tropical spans which grain), R Њ an Sto35 ϩ min (PET OBin19) h trans- The 1998). (O'Brien Њ R aiue ttemacro the at latitude) S an min n iiu month- minimum and ) Ϫ (PET min ,respectively: ), min ) IHR IL TAL. ET FIELD RICHARD 2 .A em- An ). consfragetrprino h aito nfamily in variation model ( the their of richness portion that greater a found for accounts They model). F&C (2003; angiosperm Currie hereafter, and with statistical Francis by ``global'' climate developed of richness a family relationship is the shift of this model of One example relationships. considered recent climate±richness are energy describing and when water one liquid to richness, both with whereby energy emphasis of con- traditional relationship has a the on from 2003) only shift al. growing et a to Hawkins tributed 2001, al. et Whittaker ®ndings. richness these we from family Herein developed and were 2000). genus, that species, al. for et IGMs (O'Brien present richness family being in improvement (SAF2 greatest 81.5% the and respectively; 86.8, models), and 85.6, to 79.8, models) 78.8, for (SAF1 69.7% from models increased climate± family relationships of precision and richness and strength genus, the Africa, species, southern the topographic adding to On range richness. greatest family be to should regard increase with mod- the of species, evolution and recent genera relation- more ern the the climate±richness given increase (2) of and should ships; precision (1) vector parts: and vertical two strength the has of tested we inclusion that implication second nt.Arc sas utbefrgnrtn global a generating for suitable also is dom- Africa to con- seems inate. rainfall of where and paucity tropics, the richness the within between energy with relationships keeping empirical in vincing is fo- Africa The on vegetation. or and cus expected richness re¯ect in differences should known pattern Af- predicted of continent The the var- rica. across relative richness the predicted describe in to iation (2) spe- and Africa, of Kenya, southern values in of ci®cally absolute outside estimate richness ex- to plant We woody (1) richness. unpub- ability family their and hitherto amine genus the for with IGM1s along lished genus, richness, species, family for and (IGM2) IGMs second-generation tr r erse sn hi aa( data their using regressed are eters tion. rtcladeprclciei,wt h atrtaking latter the ( strength with statistical over criteria, precedence the- empirical meet to and more had meeting oretical IGMs the to criteria, addition statistical In stringent models. ex- statis- regression are simply tical than IGMs rather that models, and given regression planatory families) especially in and angiosperm data, differences vs. climate given plant expected (woody is richness discrepancy a such oe ecie odtoa n smwee some- (See (somewhere, one time). a conditional F&C a the the variable, describes water model given its of comparison, nature conditional by priori always); the relationship that (everywhere, climate±richness general mean a differences describe important IGMs other and model. statistical ``best-®t'' These the explan- be best not the may model consequence, atory a As model. ``best'' the hsadohreiec eiwdeswee(e.g., elsewhere reviewed evidence other and This u isaeteeoetood is,w rsn the present we First, twofold. therefore are aims Our ) ϳ 4)ta osteIM hnisparam- its when IGM1 the does than 84%) eut:Mdltrsadgoa applica- global and terms Model Results: clg,Vl 6 o 9 No. 86, Vol. Ecology, R ϳ 2 3) However, 63%). hnselecting when ) September 2005 LATITUDINAL GRADIENTS 2265 model since virtually the full range of variation in IGM and water availability, reaching maximum capacity climate parameters occurs there. Secondly, we compare where surface water from precipitation remains in the IGMs with the F&C model. We illustrate the sim- a liquid state year-round and its amount exceeds the ilarities, but also the difference between explanatory climatological demand (evaporation). and purely statistical regression models of climate± In terms of ecological ®rst principles, the direct richness relationships. We focus attention on theoret- and primary relationship between climate and ter- ical and empirical discrepancies between models, rath- restrial life should be its relationship with plants, er than purely statistical ones. whereby raw energy, water and essential abiotic For reasons of limited space, and to save duplication, matter are transformed into biotic energy and mat- we do not attempt to review relevant literature. This ter. There should be a strong secondary relationship has burgeoned in recent years, and contains richly con- between climate and animal richness, via trophic trasting views of controls on diversity (notably Huston plant±animal exchange, as found by Andrews and 1994, Whittaker et al. 2001, 2003, Blackburn and Gas- O'Brien (2000) for the distribution and richness of ton 2003, Hawkins et al. 2003, Colwell et al. 2004, mammals in southern Africa. Finally, climatological Currie and Francis 2004, Qian and Ricklefs 2004). dynamics are independent of life. Although second- ary and tertiary feedbacks develop if life exists (e.g., Conceptual basis: biological relativity vegetation cover decreasing albedo), life per se is to water±energy dynamics not necessary to the operation of climatological dy- The following is drawn from E. M. O'Brien (un- namics. Crucially, however, climatological dynam- published manuscript) and brie¯y summarizes what is ics are necessary for terrestrial biological dynamics. meant by biological relativity to water±energy dynam- Thus realized climate limits the capacity for terres- ics and how it applies to terrestrial life. trial biological dynamicsÐand over time we expect richness to respond to this. Theoretically, the idea follows from ®rst principles Other independent and dynamic parameters (e.g., governing climatological, biological and ecological geomorphological water±energy dynamics) also lim- Feature Special dynamics that can be gleaned from standard refer- it the capacity for biological dynamics (e.g., via dis- ences and texts. The key is liquid water. solved nutrients), and must form part of a complete In terms of climatological ®rst principles, given explanation for spatial richness patterns. However, the Earth's energy regime, water is the only matter given both the smaller distances over which these at the Earth's surface that is fully dynamic as a parameters exhibit measurable heterogeneity and function of energy±matter exchange. Like all other their dependence on climatological water±energy matter at the Earth's surface, water is subject to dynamics, their inclusion must await the develop- changes in form and location. In addition, however, ment of trans-scalar modelling (O'Brien 1989, water occurs in and moves through all three states, O'Brien et al. 2000, Whittaker et al. 2001). In the primarily via climatological (atmospheric) dynam- interim the IGMs describe only the ®rst-order ``cli- icsÐthe hydrologic cycle. When this is combined matic potential for richness'' and assume all else to with water's physical properties, the resulting wa- be equal or non-limiting. ter±energy dynamics (work done by water) account for almost all work done at the planet's surface MATERIALS AND METHODS throughout geological time. In terms of terrestrial The thrust of regression analysis can be explanation life, climatological water±energy dynamics deter- or simply statistical description. In either case, rela- mine the very existence of water on landmasses, as tionships need to be empirically plausible and based well as its state, amount, duration and, in conjunc- on analyses carried out at an appropriate scale (grain tion with topography, its distribution. size). Sampling area should be held constant. If glob- In terms of biological ®rst principles, liquid water ally applicable predictive models are a goal, then the is the essential matter and matrix of life (see details sampled variation in independent model parameters in Franks 2000). Its physical properties make it the should be representative of (or reasonably extrapolated key agent of biological dynamics, driving all bio- to represent) their full range of variation globally, and logical processes and functions, everywhere and al- spatial autocorrelation in richness values should be ways. Since the state of water varies as a function minimized, if not eliminated, to avoid biasing model of ambient energy conditions, this a priori means development, a priori, towards particular richness and that biological dynamics are restricted to optimal associated climate conditions. Unlike purely statistical energy conditionsÐthe range in which water is liq- models, explanatory models explicitly test potential ex- uid and energy (light/heat) is still available for work. planations for phenomena rather than simply docu- Outside this envelope biological dynamics cease menting their existence. They should be both empiri- (plant dormancy, aestivation, hibernation, death, cally and theoretically plausible in terms of how ex- etc.). Within this range the capacity for biological planatory variables relate to each other and to the re- dynamics should vary as a function of both energy sponse variable. Correlation between explanatory Special Feature te.Temnmmmnhyaon fpotential of amount monthly (PET each evapotranspiration minimum of and The independent general), temporally in other. Africa are in they and empirically Africa, southern in 0.5 misleading (1) avoid (in¯ated) to minimized be should variables 2266 alsue r swal orltda osbe( possible as correlated weakly as are climato- used var- ®rst-order water iables dynamic and energy be the Secondly, to parameters. logical variable, had energy which one of and both variable water one to models time. and ter- of space (distribution) over location life, or restrial (richness) form changes the for capacity in relative the is, that water±en- dynamics; to ergy relativity biological of outcome fun- damental one atmospheric describe models of the the off Theoretically evapotranspiration landmasses. exceeds and water or for demand meets energy avail- water as long liquid as only able but oceans), off insolation evaporation with (and increase to tend should (e.g., photosynthesis) dynamics water±energy biological water±energy and atmospheric dynamics biological for capacity and The climatological dynamics. both to water±energy relate how dynamics they redundancy minimal Empirically with models. describe regression explanatory are pasblt,gnrlt,smlct,prioy and strength. statistical criteria parsimony) on theoretical lastly simplicity, on based generality, be (plausibility, should explan- ``best'' models the atory of selection Thereafter, variables. environmental ``missing'' important of role can the which obscure estimates, parameter unstable of problem the ti ekycreae ihteohrepaaoyvar- explanatory other the ( with iables correlated rates. weakly lapse cli- is adiabatic in It and variation environmental vertical as such of mate, richness po- on the effects models tential range), topog- ln(topographic The variable, with globally. consistent raphy climate is in This changes least. seasonal be most to tends rainfall when utcliert a iiie,®s yrestricting by ®rst minimized, was Multicollinearity IGMs and (SAFs) models African southern the of All r ϭ eeomn fsecond-generation of Development R nei eea oes(IGM2s) Models General Interim 2 .0 and 0.409 ausrsligfo eudny n (2) and redundancy, from resulting values r ϭϪ min sal cusi winter, in occurs usually ) .3 ntsgicn]with signi®cant] [not 0.032 IHR IL TAL. ET FIELD RICHARD r ഠ R Ͼ ob esnbe hr hudb togcreain( correlation strong be should there reasonable, be transformation to mathematical IGM2s the in For B). included Appendix was (see range (1998), O'Brien topographic in that detailed except those as methods same exactly followed the IGMs into for family) SAF2 and family; genus, and species, genus for (SAF1 models African (e.g., expected is richness Antarctica). any and/ before water needed in increase energy of or can degree They the indicate taxa. to zero taken of be values predictions as predicted treated Negative be variable. model, should the missing of a inter- de®ciency as Positive indicate such to water. taken liquid be without can cepts exist since can desert), coastal life condi- Peruvian no energy (e.g., when optimal even are water, tions in liquid zero of be absence should the Richness variables. explanatory from negative originates the relationship A the empirically. that indicates and intercept theoretically both ative, 1). SAF1 (Fig. re- SAF2 for autocorrelation of residuals spatial residuals the No in of 2000). mains al. distribution et indicated spatial (O'Brien as variable, the topography by the by eliminated of then addition and the (SAF1), stations associated climate samples with richness only using by reduced ®rst increases relief topographic 2000). al. as habitats) et occupied (and (O'Brien area be surface can function in linear that increase simpler the the of but than because function; reasonable more linear is a it than only statistically is range better topographic slightly of logarithm natural the that nteuepandvrac.Ti a h aefor case the was richness: This species for variance. IGM1 unexplained change the minimal with in IGMs, using generated models those SAF and empirical three the using generated values hsi sraoal oasm httesrnt ( strength the that assume to reasonable is it Thus an n soitderrtr ( term error associated and . n ideally and 0.7 ahmtcltasomto ftebs southern- best the of transformation Mathematical neg- be should intercept the that argue we Finally, were autocorrelation spatial of effects potential The n PET and oe IM eeomn ( General development Interim (IGM) in Model used data (SAF) African ern min A1rsdas(anal[ (rainfall residuals SAF1 itnecassbigsgicn at largest signi®cant two being and classes smallest distance two the with mation), P ls.TeSF orlga sntsigni®cant not is ( correlogram distance SAF2 smallest The the class. is this on point signi®cant ah2 0 km 000 25 each n A2rsdas(rainfall residuals SAF2 and ssgicn ( signi®cant is correlogram richness species The model). raphy fe ofroicreto.TeSF corre- ( SAF1 signi®cant is The logram correction. Bonferroni after F epciey nsuhr fia.Note Africa). southern in respectively, , ϭ IG .Sailcreorm o h south- the for correlograms Spatial 1. . 0.597). r Ͼ .)btenfrcse richness forecasted between 0.9) P 2 :woypatseisrichness; species plant woody ): RMSE Ͻ r .0,Bnern approxi- Bonferroni 0.001, ഠ P fterltosi of relationship the of ) .7('re 1998). (O'Brien 0.97 clg,Vl 6 o 9 No. 86, Vol. Ecology, ϭ R .1) u h only the but 0.011), an N ] ϩ ϭ ϩ PET 5gi cells, grid 65 E model); PET P ϩ Ͻ topog- 0.05 R 2 r ) September 2005 LATITUDINAL GRADIENTS 2267

FIG. 2. Maps of Kenya showing: (a) the locations and identi®cation numbers of the climate stations and the 25 000-km2 circles surrounding them; (b) woody plant family richness per circle; and (c) ``residuals'' from IGM2 family level (i.e., forecast minus observed family richness). Large lakes (Turkana and Victoria) are indicated (stippled ®ll). Circles that have Ͻ80% of their area as land in Kenya, although shown here, are excluded from consideration in the paper; these are numbers 13, 17, 19, 21, 22, 25, 26, 27, and 28. climate with richness in southern Africa also apply with resampled to 0.1Њ resolution, giving Ͼ200 elevation respect to IGMs. As demonstrated for species richness points/25 000-km2 grid cell. The minimum value was by O'Brien (1998), this assumption is supported math- subtracted from the maximum to give the ``topographic Feature Special ematically if there is little difference between the ideal range'' (in meters), for each grid cell. Values were as-

SAF model (based on PETmin) and the corresponding signed to the 980 climate stations according to the grid IGM in terms of the coef®cients for annual rainfall and, cell that they occupied in the case of IGM2, topographic relief. (Compare ideal SAF models in O'Brien et al. [1998] with IGMs re- Examination of the spatial pattern of model predictions ported here.) Given the increased range in PETmin val- ues, IGM coef®cients for PETmin should markedly de- All IGMs were used to predict the climatic potential crease relative to ideal SAF model coef®cients. And, for richness across the continent of Africa, based on given the greater range of positive (potential) richness data from the 980 climate stations. The same was done values, the intercept value should increase but remain for the F&C model at the family level (using the negative. Thornthwaite climate data). Examination of the re- sulting patterns in forecasted richness, and how they African climate and topography data relate to variation in vegetation and physiography, was We used Thornthwaite and Mather (1962±1965) for undertaken following O'Brien (1998). climate data from 980 stations in Africa (i.e., mean annual rainfall and potential evapotranspiration, both Actual vs. predicted richness for Kenya of which are dynamic climate variables). Thornth- All the IGMs forecast richness for a circular area of waite's PET is a proxy for the intensity of insolation 25 000 km2 (i.e., within a radius of 90 km of a climate at the Earth's surface. It measures the energy demand station), the same area as the grid cells used in model for liquid water (mm), and thus the role of energy in development (e.g., O'Brien 1993). Actual woody plant climatological dynamics (evaporation and transpira- richness data for Kenya were calculated accordingly, tion). The data were calculated using a formula he de- based on presence±absence data per circle, with each rived from experimental data on the amount of water circle centered over a climate station (Fig. 2). Pres- evaporated and transpired from samples of grass-cov- ence±absence data were obtained using a comprehen- ered land never suffering from lack of water. The for- sive set of distribution maps and site location data for mula requires data on prevailing temperature and in- Kenyan woody plants (Beentje 1994). Following the tensity of insolation at a given time (date) and place same criteria as in O'Brien (1993) to determine which (latitude). Unlike many measures of PET, his is not species to include, 1417 out of 1862 species were re- adjusted to sea level and thus measures the energy tained; these represented the largest and longest lived (heat/light) regime actually in¯uencing vegetation at of plant species, and thus those most likely to be robust the Earth's surface. indicators of climate conditions. Those eliminated were Following O'Brien et al. (2000), topography data non-native species, plants Յ2.5 m in height, and/or were extracted from the USGS DEM of Africa, and plants that are not truly woody. The distributional rang- Special Feature Ͻ communication. personal ϳ (30 degree a resolution pixel half with of pixel, per data presence/absence and total), in families). genera (126 (635 families genera within within species rang- aggregating of by es determined were taxa higher of es 2268 n h & oe hudbe should model us- F&C values the richness ing family predicted general, in Africa rae hnata od ln aiyrcns but richness family plant woody actual than greater aepooo sfrte2 000-km 25 the for as protocol same qao,tesailrslto ftedsrbto data distribution the of is resolution spatial the equator, isemfml ihesi 4900-km 34 in an- maximum richness global family the giosperm maps) (not range data from richness derived Currie all a and Francis as the circle. to 211 any According in leaves richness family this angiosperm for families, maximum (31) pteridophyte and h n aesi(94.Ana ae ect(WD de®cit of water that Annual is (1994). used Tateishi they and that Ahn PET of formulation The interpreted/interpolated data). (i.e., databases climate global informative. forecasts be IGM with could comparison but bound, lower a give eslewti ahcrl.Crlswith Circles circle. each within cen- whose lie pixels ters those from data species amal- the Pro- gamating by MapInfo UK) Berkshire, using Windsor, (MapInfo, circle fessional per richness actual culated hr ob 4 ln aiisi ey H Beentje, (H. Kenya in families plant 245 F&C know be We the to exist. and implications there testable IGMs some the but model, prevents between This maps comparison angiosperms. range Kenyan direct for species exist of not series does comprehensive fam- A pteridophyte ilies. and gymnosperm excludes fam- and nonwoody ilies includes which richness, family sperm this at criterion equal-area the size. for cli- met grain Kenyan 1b 27 on Table centered stations, circles (see mate 37 model the Of their terms). for error Francis (2003) by used Currie area and sampling the to corresponds This esfr3 900-km 2a). 34 (Table for circles ness 28 of total a leaving set, data from removed the were borders Kenya's within land as area a et rdce nisemfml ihesper richness family Ken- angiosperm the 900-km 34 in predicted So, 1). test, Fig. yan 2003: Currie and (Francis 201 hnrseie notesm 30 same the into were rasterized locations then These booklet. districts postal of and Kenyan a CD-ROM, checklist Encarta (1988) Microsoft localities, Polhill's collecting including var- sources, using coordinates ious were longitude places and named col- latitude the named assigned all as these, only For given localities. is lecting (1994) Beentje in species a acltda PET as calculated was 1 n rfrbyn oeta 0.Smlry for Similarly, 201. than more no preferably and 211 km 3080 itiuinmp rmBete(94 oss of consist (1994) Beentje from maps Distribution ϳ h lmt aaue yFacsadCri r from are Currie and Francis by used data climate The od ln aiyrcns ifr rmangio- from differs richness family plant Woody eas acltdata od ln aiyrich- family plant woody actual calculated also We 55k n ie rai fetvl osatat constant effectively is area pixel and km 55.5 2 2 itiuinifrainfor information Distribution . icecnb ena esnbei tis it if reasonable as seen be can circle 2 ice 15k ais sn the using radius) (105-km circles Ј .Gvnta ey iso the on lies Kenya that Given ). an .Ecuiggmopr (3) gymnosperm Excluding ). iu enana actual annual mean minus Ͻ Ј 0.I sdfcl to dif®cult is It 202. rdsse.W cal- We system. grid 2 ice Tbe2b). (Table circles Ͻ 2 ϳ 0 ftheir of 80% rdclsis cells grid IHR IL TAL. ET FIELD RICHARD 0 fthe of 30% an ) vptasiain(AET evapotranspiration inwsasge WD sta- climate assigned Kenyan was Each Tateishi. tion and Ahn of values RMSE Ͼ l n eottema results. mean the report sam- each and within ple values richness correlated forecasted produced We and nine. this actual to seven overlapped; from no ranging sample sizes that any sample such in circles, circles the two of samples 30 random pseudorep- generated different we such correlation, avoid performing To a when at lication correlation). circle by one (e.g., than more time considering with predicted by compare values to actual try we if however, problem, a hc logvsthe Table gives in also reported which are 1a, IGMs new all and 1998) (O'Brien PET otenArcnmdl nogoa oes(IGMs). ( synonymy models statistical cases, global all into In models African the of southern transformation mathematical the for coef®cients hne curd uprigti supin(see assumption mathematical this expected supporting the occurred, changes cases all In IGMs. spective hswsncsaybcuetetoPET two Currie. the and because climate Francis necessary the by was used used This we formulations) Kenya (and in data richness forecast to el RMSE ( niae ht ossetwt oprsn between comparisons with coef®cients consistent model that, IGM2 indicates and IGM1 com- between Furthermore, 2000). parison al. et O'Brien in coef®cients hrtwiesPET Thornthwaite's in thge lvtos n ntemutisand mountains the ( Africa in eastern and sta- of climate elevations, uplands For coastal higher similar. at and measures tions two basins the (xeric) are plains, interior low-elevation in y12errtrs esnbe®;adtoeotby out those and out ®t; those reasonable ®t; a close terms, a the error as of 1±2 classed error) by be square can mean model (root original term error forecasts one intervals, within con®dence ac- to the Similar with richness. richness tual predicted individually: the circle compare simply each we consider ir- we not is also when is but important It development, richness. forecasted model generating in in relevant problem a be would J. D. and Francis P. A. by Currie). provided kindly were data aigi esnbet sueta empirical that assume to reasonable it making aesidt eekie vra0.5 a over and kriged Ahn were (the data Tateishi location geographic its to closest location nyi em ftehge PET higher the A1). of and Fig. terms mm in A: Only 668±1938 (Appendix respectively of mm, ranges 1177±1727 with different, very are ro em,apo t vrapn ice r a are circles Overlapping ®t. poor a terms, error 2 eil n ehd:Dvlpeto second-genera- Models General of Interim tion Development Methods: and terials oe ofcet o G1seisrichness species IGM1 for coef®cients Model ayo h eyncrlsoelp(i.2.This 2). (Fig. overlap circles Kenyan the of Many min ausfrteeprclsuhr African southern empirical the for values ) o da PET ideal for oes nadto,Tbe1 ie correlation gives 1a Table addition, In models. an min e IGMs New smrel oe hntePET the than lower markedly is R an oesas pl otere- the to apply also models ESULTS an R .We sn h & mod- F&C the using When ). 2 n PET and Ͼ n otma qaeerror square mean root and opr ihSFmodel SAF with compare ; 00maoesalevel), sea above m 1000 an clg,Vl 6 o 9 No. 86, Vol. Ecology, r aus hc occur which values, Њ an Ͼ rd hs climate these grid; .)i indicated, is 0.9) ausfo the from values an measures R 2 Ma- and an September 2005 LATITUDINAL GRADIENTS 2269 empirical SAF1 and SAF2 models (O'Brien et al. variables, especially at the lowest predicted values. 2000), the addition of topographic range independently However, the fact that IGM1 predicts substantially increases precision and strength of the model. In all more woody plant families in rich cells than the F&C cases, this is shown by an increase in predictive power, model predicts angiosperm families is a notable dif- reduction in unexplained variance, and the more neg- ference. IGM2 produces lower predicted family rich- Ͼ ative intercept, while the Ran and PETmin coef®cients ness at the top end (only six 175, maximum 226.0), remain nearly the same. and also in the least rich cells (e.g., 12 predictions of zero), but not overall. Mean predictions for the 980 Statistical results climate stations across Africa are 134.0 (F&C model), Spatial pattern of model predictions.ÐAs expected 62.6 (IGM1), and 63.0 (IGM2). (given the strong correlations between species, genus, Kenyan test.ÐPredictions of actual woody plant and family richness in southern Africa), the patterns of richness values using the two generations of IGMs are predicted genus (especially) and family richness across mostly reasonable or close ®ts [within 2 or 1 error Africa were very similar to those for species richness term(s), respectively], with a slight increase in preci- (Figs. 3, 4a, and Appendix A: Fig. A2). Visual com- sion being found among IGM2 predictions (Table 2a). parison of these maps with topographic maps of Africa Using the random sample protocol (see Materials and indicate, consistent with the changes in coef®cients Methods: Actual vs. predicted richness for Kenya)to (see Results: New IGMs and Table 1), that IGM2 pre- eliminate pseudoreplication caused by circle overlap, dictions result in an increase in forecasted richness val- the mean values for ``residuals'' (forecast minus actual ues in areas with high topographic relief and a decrease values) were very small. Mean observed species rich- for areas with relatively ¯at terrain, especially for fam- ness was 307.2 Ϯ 3.0 (mean Ϯ SE); the mean residual ily richness. For example, Mt. Cameroon, the African of the species forecasts for IGM1 was Ϫ4.3 Ϯ 3.7 (SD Rift Valley, the Ethiopian Highlands, and the Tibesti of residual sizes ϭ 71.5 Ϯ 2.4 [SD Ϯ SE]), and for Dome all exhibit increases in predicted richness, while IGM2 it was 23.1 Ϯ 3.3 (75.2 Ϯ 2.2). Mean genus there are notable decreases in portions of the interior richness was 172.5 Ϯ 1.4; IGM1 mean residual Feature Special plateaus of eastern and southern Africa, and portions wasϪ1.4 Ϯ 1.8 (34.0 Ϯ 1.2), and IGM2 mean residual of the Congo and Chad Basins. In between these areas, was 10.6 Ϯ 1.6 (34.7 Ϯ 1.0). Mean family richness predicted values remain similar to those generated by was 65.4 Ϯ 0.25, IGM1 mean residual was Ϫ9.7 Ϯ IGM1. This emphasizes the independent and idiosyn- 0.45 (10.9 Ϯ 0.19), and IGM2 mean residual was Ϫ2.7 cratic nature of vertical changes in climate and re¯ects Ϯ 0.34 (6.2 Ϯ 0.22). the increased strength and precision of the models. The mean Pearson correlation coef®cients between Again in line with the greater contribution that topo- actual and forecast values were high: IGM1 species r graphic relief makes to the strength and precision of ϭ 0.71 Ϯ 0.008; IGM2 species r ϭ 0.78 Ϯ 0.007; IGM1 IGM2 for family richness, the most pronounced chang- genera r ϭ 0.72 Ϯ 0.007; IGM2 genera r ϭ 0.80 Ϯ es occurred for family richness. 0.006; IGM1 families r ϭ 0.68 Ϯ 0.006, IGM2 families The pattern generated by the F&C model (using the r ϭ 0.81 Ϯ 0.005. Signi®cance values based on these climate station data) is similar to that generated by the mean r values and a sample size of eight are marginal IGMs for family richness in that the ordering of pre- for IGM1 (P ഠ 0.05) and signi®cant for IGM2 (P ഠ dictions is similar (compare Fig. 4a with Fig. 4b). When 0.02). Further, the best-®t lines between observed and the 980 climate stations are ranked from lowest to high- forecast values for IGM2 approximated the ideal 1:1 est predicted value for each model, the mean difference line in the cases of species and genera, suggesting no in ranks between the two models is only 65 (although systematic errors in relation to richness levels: mean this difference is signi®cant; Wilcoxon signed rank test slope did not differ signi®cantly from 1 nor mean in- Z ϭ 3.6, P Ͻ 0.001). The main distinction between the tercept from 0. For families, and for all IGM1 forecasts, two sets of predictions is in the spread of values. The the slope (observed values arbitrarily on the y-axis) F&C model predicts high numbers of families under was ¯atter than 1. (For IGM2, species: mean slope ϭ extremely arid conditions (see also Fig. 5). Indeed, all 0.95 Ϯ 0.03, mean intercept ϭϪ6.3 Ϯ 7.3; genera: F&C model predictions for Africa are Ͼ50 angiosperm mean slope ϭ 0.99 Ϯ 0.03, mean intercept ϭϪ7.8 Ϯ families and only 11 points are Ͻ70. The maximum 4.3; families: mean slope ϭ 0.82 Ϯ 0.02, mean inter- predicted value is 186.4, which is only just below the cept ϭ 14.0 Ϯ 0.9.) Thus the correspondence between maximum possible from the model (186.8; Fig. 5); 55 observed and forecast richness values was good. In- predicted values are Ͼ175 (there would have been more terestingly, ln(topographic range) correlated very had we used Ahn and Tateishi PET data in this exer- strongly with richness values in the random samples: cise). In contrast, IGM1 produces 255 predictions be- r ϭ 0.80 Ϯ 0.005, 0.83 Ϯ 0.005, 0.86 Ϯ 0.003, re- low 35 woody plant families, the lowest being 5.8. The spectively, for species, genera, and families (P ഠ 0.01). maximum predicted value for Africa is 263.9, but only Geographically there was a broad pattern of over- 12 predictions are Ͼ175. Some of these differences prediction in west and northeast Kenya and under- between the two models relate to the different response prediction in south and central Kenya (Fig. 2). Possible Special Feature atdvle r rae hntenme factual of 900-km 34 number every the in families than plant greater woody are Fore- values 2b). (Table casted Kenya in richness family angiosperm were maps. which Beentje's Kenya), for of basis parts the remote spec- from botanical (e.g., of undercollection imens of could function underground a overprediction be or cases, also some rivers In exotic resources). of water lack leaching, moisture soil (high low and soils poor relatively as- with being sociated overprediction and re- elsewhere]), water rainfall bringing water rivers from [i.e., underground rivers exotic lakes, and/or sources, of soil (presence higher-than-normal and moisture derived) (basalt soils mal richer-than-nor- with with ones, associated edaphic being be underprediction to likely are factors contributing R taxa plant woody of number predicting richness: for potential climatic the of models general Interim A) 2270 nana analadPET and rainfall annual in h ulrneo aito lblyi PET in globally variation of range full the 21 fagopr aiispr3 0 km 900 34 number per maximum families global angiosperm the Kenya of than in (201) less occur also to and known (211) families F&C num- angiosperm total of the the than ber less with are values associated forecasted all 187 model, of prediction possible aiis( families 2 ABLE rae hnteIMpeitdwoypatfamily plant woody 000-km 25 (for IGM-predicted richness the than greater yFacsadCurrie. and Francis by iey WD tively; and , oeta vptasiain(E)mdl ihvle ae rmFacsadCri 20)adrsda ensquare mean residual and (2003) Currie and Francis from taken values with ( model, errors (PET) evapotranspiration potential IGM1 ࿣ Notes: h & oe per o ogosyunderpredict grossly to not appears model F&C The r retained. are ntrso lblapiain h ag fvariation of range the application, global of terms In fia SF oesadtoegnrtduigteIM,mkn traoal oasm htec G saglobally a is (see IGM values zero each to that converted southern are assume the predictions For to negative 1998). when reasonable O'Brien values (see it the Africa making are southern for parentheses IGMs, obtained in the relationship numbers empirical models, using ideal African generated the of those redescription and applicable models (SAF) African eotdfrteIM r ohteeprclrltosisotie o h otensbotnn fArc ( Africa of subcontinent southern the for obtained relationships empirical the both are IGMs the for Reported ² )FacsadCri (2 Currie and Francis B) au eotdi (PET is reported Value ¶ T au eotdi PET is reported Value au eotdi WD is reported Value § (1998). O'Brien See ³ n ehd:Dvlpeto eodgnrto nei eea Models General Interim second-generation of Development Methods: and Species³ IGM2 aiis8.8 Families Model Families Genera Species Families Genera .Peitv lblmdlseictosfr()ItrmGnrlMdl IM n G2,ad()FC2 F&C (B) and IGM2), and (IGM1 Models General Interim (A) for speci®cations model global Predictive 1. RMSE h orlto ofcet niaesaitclsnuaiy( singularity statistical indicate coef®cients correlation The N RMSE oe em n lblapplication global and terms Model an ϭ R sana ae ect(e 're 1998). O'Brien (see de®cit water annual is 're ta.[98) n hs bandfo ahmtcltasomto nogoa oes(Gsfor (IGMs models global into transformation mathematical from obtained those and [1998]), al. et O'Brien ; an 4224) rvddb .J Currie. J. D. by provided ) sana anal PET rainfall; annual is (constant) Intercept Ϫ Ϫ Ϫ Ϫ Ϫ Њ 7 .572.6250 5.1186 0.1597 0.2987 5.6294 170 371 0.3494 150 an an 2 Ϫ 7007 0.6455 0.0372 2.9008 57 0.1836 70 an . aiuelniuesailrslto)goa E oe:peitn h ubro angiosperm of number the predicting model: PET global resolution) spatial latitude/longitude ice) ie h maximum the Given circles). m) not (mm), .430.7197 0.0473 1 ) 2 . min cosArc srepre- is Africa across Ϫ R R min an .61 0.2199 0.0641§ an . n PET and min (mm) ). 2 G ofcet ( coef®cients IGM IHR IL TAL. ET FIELD RICHARD ice and circle, 2 reported an r iiu otl n nulptnileaornprto,respec- evapotranspiration, potential annual and monthly minimum are PET (mm) min ettv ftegoa ag fvraini hs pa- these PET in where variation of Even range rameters. global the of sentative ࿣ where ela ln rwh ne hs odtos(see conditions these under PET where growth, Thus, 1998). plant O'Brien as energy, as optimal of well duration and/or amount the sures E) tesaeata vptasiain(E)and (AET) evapotranspiration actual are Others minus PET). (precipitation available climate of from indices derived basic moisture the soil vari- of the water-budget one of de®cit, a water sign use able, the Currie in and and Francis used intercept. variable water the of terms plausible. empirically (e.g., is expected is which richness Antarctic), plant woody zero case this In eueto reduce saiyeita oetm()drn h er(e.g., year the during time(s) Thus some is summer). there at nec- if exist water liquid words, essarily for conditions other energy the In then water. rainfall pre- measure liquid the than available as rather water) of rainfall solid of includes use (which cipitation the given relates richness, climate how to describe plausibly IGMs the poles) oe ae±nrydnmc,dsierdcn to reducing despite dynamics, water±energy model nulrifl.Where rainfall. annual h GsadteFCmdldfe rmrl in primarily differ model F&C the and IGMs The N r richness ϭ Ϫ Ͼ Ϫ richness a ;te®s ubr eutwe h eaievalues negative the when result numbers ®rst the ); 980) .)btenpeitosgnrtduigsouthern using generated predictions between 0.9) 6.79 (PET ste(eaie necp and intercept (negative) the is Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ (mm .041.100.966 0.971 0.963 0.971 0.974 0.973 10.1120 19.6916 42.7155 0.0034 0.0127 0.0257 0.0039 0.0141 0.0284 ϰϪ ϫ min 2 ϰϪ ) R 10 ) a 2 an Ϫ ϩ 5 mlcty(n ttsial)mea- statistically) (and implicitly a ¶ [ R ϩ R an ntpgahcrange)]. ln(topographic an ln(topographic [ ϩ ag)(m) range) loeul eo h IGMs the zero, equals also ntpgahcrange)] ln(topographic min ϭ min clg,Vl 6 o 9 No. 86, Vol. Ecology, mdlttdsto latitudes (mid 0 ϭ ,teIM still IGMs the 0, Transformation correlation R an (Pearson stemean the is R 2 Materials adjusted , r )² Њ September 2005 LATITUDINAL GRADIENTS 2271

TABLE 1. Extended. to each other in southern Africa or Africa in general, see O'Brien (1993, 1998). Since WD ϭ 0 when PET ϭ 0 or when precipitation Ն PET, the F&C model is a priori conditional. And Southern African model ®ts ( ϭ 65) ϭ N when WDan 0, the model reduces to an optimal en- R2² adj. R2² RMSE² ergy model of the relationship of climate with richness, with no water component. For Condition 1, the soil

water-budget model, when Pan (mean annual precipi- Ͻ ϭ Ͼ 0.788 (0.804) 0.778 (0.794) 73.7 (70.8) tation) PETan, then AETan Pan and WDan 0: 0.798 (0.808) 0.788 (0.798) 36.7 (35.8) ϰ Ϫ Ϫ ϩ Ϫ 2 0.697 (0.697) 0.683 (0.683) 11.8 (11.8) richness a (PETanP an) [PET an(PET an ) ] ϭ a Ϫ (PET Ϫ AET ) ϩ [PET Ϫ (PET )2 ] 0.856 (0.868) 0.846 (0.859) 61.2 (58.6) an an an an 0.868 (0.874) 0.859 (0.865) 29.9 (29.3) ϭ a Ϫ WD ϩ [PET Ϫ (PET )2 ]. 0.815 (0.815) 0.803 (0.803) 9.3 (9.3) an an an For Condition 2, the optimal energy-only model, when Ն ϭ ϭ Pan PETan or when PETan 0, then AETan PETan 0.837 0.837 17.40 ϭ and WDan 0: ϰ Ϫ Ϫ ϩ Ϫ 2 richness a (PETanPET an ) [PET an(PET an ) ]

ϭ Ϫ Ϫ ϩ Ϫ 2 water surplus. All are conditional indices of available a (PETanAET an ) [PET an(PET an ) ] soil moisture for plant growth; none is a dynamic cli- ϭ a ϩ [PET Ϫ (PET )2 ]. mate variable. AET ϭ precipitation only when precip- an an itation Ͻ PET. Otherwise, when precipitation Ն PET, Thus the F&C model is a soil moisture±energy model

AET ϭ PET. WD ϭ PET Ϫ AET when precipitation where soil moisture de®cits limit plant growth and is Feature Special Ͻ PET. When precipitation Ն PET, WD ϭ 0, because purely an energy model where water de®cits do not AET ϭ PET. Water surplus ϭ precipitation Ϫ PET occur. Where there is no available water, it predicts when precipitation Ͼ PET (i.e., when AET ϭ PET and between nine and 98 angiosperm families (within the ϭ WD 0). For a detailed analysis of how these indices globally observed range of PETan; Fig. 5). This aspect statistically relate to richness, to climate variables, and of the model, which derives from both the positive

FIG. 3. Predicted species richness for Africa based on (a) IGM1 and (b) IGM2. Values were calculated using the Thornth- waite climate data for all 980 African climate stations (source of climate data: Thornthwaite and Mather 1962±1965). This ®gure is reproduced in color in Appendix A. Special Feature ttsia tt h aa n stu o l h Francis 2 the (from all models for PET true global is Currie and data, and the the within to subsumed ®t are sug- statistical effects coef®cients, other unmodeled the that of gests values the and intercept 2272 iia ai o h Gscnb osdrdreason- considered be em- can the IGMs Thus able. the latitudes. for high to basis mid pirical in the unreasonable of is parts Plio- many It during oscillations. species ice-age of Pleistocene migration major of to absence barriers the given physical with Africa, of reasonable ¯ora the seems to regard assumption to This likely environmental most complete. is its be example, with for and eruption, par volcanic deglaciation major a after recolonization on where be potential; to spe- assume richness reasonably can cies or know we where requires regions also development Africa, (southern model However, them Australia). between transition latitudes the they the for high And exist to States). of United mid Europe, regions Western of of parts parts tropical for (e.g., humid exist do range for They species exist world. exhaustive not and do Systematic maps which based. on data is the of it quality the on depends richness for erdcdi oo nApni .As,seFg 2i pedxAfrIM n G2peitosa h eu and genus the at predictions IGM2 and IGM1 is for ®gure A This Appendix 1962±1965). in Mather A2 and Fig. using (Thornthwaite calculated see stations were Also, range.Values climate speci®ed A. African the Appendix levels. within 980 in family richness all color predicted for in have data Africa reproduced in climate stations Thornthwaite climate the no i.e., empty, is eut rmtecliert ewe PET between collinearity that the Simi- redundancy from the 2). results by Table hindered is 2003: interpretation Currie larly, and Francis resolution; F neprclgoa oe fteciai potential climatic the of model global empirical An IG .Peitdfml ihesfrArc ae n()IM n b & oe.Prnhssidct httecategory the that indicate Parentheses model. F&C (b) and IGM1 (a) on based Africa for richness family Predicted 4. . D ISCUSSION Њ o10 to an IHR IL TAL. ET FIELD RICHARD n WD and Њ spatial an . es ie htteIM pl otefl ag of range full the R to in apply variation IGMs global rich- the realized that co- and Given the climate ness. realized cause between could bio- variation dynamics for capacity water±energy climatic logical the in variation geographic consequent that dy- and dynamics, biological water±energy on and namics climatological re- both the of the from that dependence stems richness idea and climate the between support lationship They three at levels. richness for taxonomic potential climatic the of IGMs new ain fpeitv models predictive of cations alsta edicuini oecmlt expla- complete more a var- (see in other nation inclusion highlight need should that models iables underprediction the or by over- richness of gross climate richness, the for describe only potential IGMs positive the ap- a Since should globally. in same ply The relief, respectively. topographic fashion, low negative or or in high IGM2s exclusively of in almost areas values climate richness in forecasted changes alter vertical of addition predicted the in Empirically, 1998). O'Brien variation (see Africa relative across of richness does as pattern this, realistic supports the Kenya actual in between richness ®t predicted good and The world. the possible in is elsewhere richness for for potential climatic the of casting h eut upr h lblapiaiiyo the of applicability global the support results The miia efrac ftemodels the of performance Empirical icsin oeapiain n impli- and applications Some Discussion: an n PET and ). clg,Vl 6 o 9 No. 86, Vol. Ecology, min esnbefore- reasonable , September 2005 LATITUDINAL GRADIENTS 2273

to observed or expected differences in richness, es- pecially the latitudinal gradient in richness. All work to date suggests that the IGMs meet these and other empirical and theoretical criteria. Our results, together with those of Francis and Currie, suggest that the F&C model also tends to produce a good ®t to empirical data in a range of climatic conditions. However, its upper prediction limit of 187 angiosperm families and its pos-

itive, often high lower limit (depending on PETan; Fig. 5) suggest that its predictive usefulness is best in mod- erate (e.g., temperate), rather than harsh (e.g., very arid) or benign (e.g., equatorial humid), climatic con- ditions. The F&C model is a purely statistical regression model, which its authors interpret via the importance of secondary or tertiary biotic relationships to climate, such as the correlation of productivity with richness. We have only compared the F&C model with IGM1, since it applies only to the horizontal relationship of climate with richness. Like the other models, the F&C model was limited to a single water variable and a single energy variable (Table 1). In both cases the en- ergy variables are similar (aspects of PET) and related to richness in a functionally similar fashion: an optimal (parabolic curve) function, modeled as PET Ϫ PET2. Feature Special Thus according to both sets of models, richness initially increases and then decreases as energy increases. In terms of climate, such an optimal relationship is em- pirically plausible with regard to rainfall: as we move from the poles to the tropics, rainfall increases as re- FIG. 5. (a) Envelope of possible predictions from the F&C alized evaporation and the capacity for atmospheric model (Francis and Currie 2003). The solid line shows the saturation (dew point) increase up to some global op- condition of no water de®cit (WD) and represents the upper timum. At this point the capacity for evaporation con- limit of predictions for this model. The dashed line shows tinues to increase but subsequent condensation, satu- the condition of no precipitation and thus represents the lower limit of predictions for this model. All model predictions fall ration, and rainfall decrease as dew point increases. within the area bounded by the two lines. A plot showing the Instead, more and more (and eventually all) evaporated range of predictions of IGM1 can be found in Whittaker et water is held in the atmosphere: the same amount of al. (2003: Fig. 7.2b). (b) Predicted values from IGM2 plotted atmospheric moisture can keep Holland green, but the against minimum monthly potential evapotranspiration Sahara a desert. (PETmin) when there is no rainfall and no topography (i.e., completely ¯at terrain), at the species, genus, and family lev- These similarities between the F&C model and the els. Note that all are negative, which means that no woody IGM1 model are considerable and, along with other plants are predicted to exist under such conditions, even when recent ®ndings (e.g., Hawkins et al. 2003), could sug- optimal energy conditions exist (see Introduction: Conceptual gest that we are close to determining which climatic basis: biological relativity to water±energy dynamics)Ðin contrast to the F&C model. variables tend to be responsible for constraining plant richness. If so, the differences between the models may be even more instructive and important for advancing Annual water de®cits are common in Kenya and our our knowledge. In the F&C model, a conditional, non-

®ndings support the predictive usefulness of the F&C dynamic soil water-budget variable (WDan) represents model where annual water de®cits occur, even in low available liquid water, as opposed to the dynamic cli- latitudes. mate variable (Ran) in the IGMs. The result is a model of how energy, coupled sometimes with an insuf®- Comparison of IGMs and the F&C model ciency of soil moisture, relates to richness. The con- To be a general, globally applicable description of ditional rather than general nature of the F&C model the ®rst-order (and ideally causal) relationship of cli- can be attributed to at least two factors; (1) the lack of mate with richness, a model should generate reasonable a theoretical or empirical explanatory framework in predicted richness values (without regional re®tting), model development, and (2) statistical issues including and the resulting distribution of relative predicted rich- multicollinearity between the variables chosen to de- ness should be similar in pattern and relative magnitude ®ne climate and how it relates to richness. Special Feature 2274 tttn h uko n lbldtbs nrichness. on database WD between global redundancy any priori A of con- bulk values the richness asso- stituting associated vegetation with the them, with does dom- ciated as conditions de®cit systems, water terrestrial that inate fact the by biased is statistically. powerful explanatory even less rejection, are an alternatives model if Within suggest are would families present). this angiosperm framework, being nine as energy, modeled no no is there and (when intercept on water the in¯uence in explanatory subsumed is missing richness pos- some the of that effect suggests model itive F&C pos- the the of Empirically intercept criteria. itive statistical model to ``best'' restricted the is of other discrimination each and to generated, relate data are input statis- how only of descriptions framework, tical explanatory cri- statistical an on Without as teria. well as observation, empirical and simplicity) parsimony, on (generality, based grounds modi®ed or theoretical rejected, accepted, be can models Climate tto ra(%) Area station ABLE hs isscudacutfrtehg statistical high the for account could biases Together relationships. possible these relationship other particular of expense this the at of precision statistical the Notes: n aiyrcns/500km 000 richness/25 family and ntescn ae ecnie htteFCmodel F&C the that consider we case, second the In framework, explanatory an given case, ®rst the In 7107 02 8 1 8151139 65 80 111 73 188 71 76 185 226 81 199 344 69 194 79 422 206 38 70 372 237 69 37 360 185 58 71 386 229 65 441 110 212 59 84 343 186 92 93 175 199 58 427 91 60 181 389 197 70 82 345 171 257 314 368 145 282 55 168 112 295 69 359 175 102 68 24 310 477 216 239 528 56 304 157 378 60 23 86 545 318 209 334 87 394 40 445 208 78 81 284 704 160 82 277 178 66 77 383 621 89 88 379 270 70 71 78 289 202 248 529 322 264 114 81 220 88 295 101 87 477 272 58 100 368 456 201 218 90 486 387 72 391 232 89 542 337 100 486 274 56 37 100 350 400 124 720 36 82 100 404 184 81 625 244 35 100 525 79 101 34 100 208 33 87 100 313 249 81 460 227 32 100 84 143 252 31 100 30 100 480 274 98 408 239 29 88 411 269 24 23 100 525 82 413 20 88 469 18 16 100 96 15 95 14 12 11 10 T nisemfml ihes3 0 km 900 richness/34 family angiosperm 0 3 4 439276 2 3 80 80 82 77 230 37 223 96 240 85 429 213 112 84 417 88 446 292 260 397 182 68 252 63 547 273 483 72 59 468 207 38 507 191 84 221 80 379 179 115 76 347 83 404 272 251 321 187 84 238 92 501 265 462 87 93 437 247 47 488 291 91 274 83 435 286 93 87 560 83 525 291 237 100 526 100 160 279 100 561 236 414 9 100 8 100 508 7 100 410 99 6 5 100 4 100 3 2 1 .()ItrmGnrlMdl IM) cul rdce,ad`rsda' ausfrwoypatseis genus, species, plant woody for values ``residual'' and predicted, actual, (IGMs): Models General Interim (A) 2. Area ϭ ecnaeo h rao iceta sln ihnKna ``Residuals'' Kenya. within land is that circle a of area the of percentage pce eu Family Genus Species od ln richness plant Woody culvalues Actual 2 ( N an ϭ n PET and 2 8ciaesain) B & oe:ata od ln ihesadpredicted and richness plant woody actual model: F&C (B) stations). climate 28 ( N ϭ IHR IL TAL. ET FIELD RICHARD an 8ciaestations). climate 28 increases pce eu Family Genus Species )IGMs A) rdciiya huhi eeadnmcprmtrof parameter dynamic a primary were treating it to though as similar productivity is var- assumption climate This are iables. variables with water-budget climate of that relationship richness; the of studies many to mon year. a seasonally of and day, course a the pho- of during in course decrease the as and during dynamics, increase tosynthesis empirical biological the to by regard shown with applies The same decreases). saturation atmospheric for while potential increases the point dew increases, for energy potential op- (as the rainfall the and with so energy consistent between does empirically relationship It is timal richness. that with fashion a relationship in its in effects energy negative of the emphasizes rich- model with their climate First, ness. of better relationship a the to of ways understanding important several in contributes climate. study of function a as water mea- liquid always available rainfall sures phenomena; biological climate how to modeling relates when considered be to need mean- and ings perfor- these Climatic and poor meaningful conditions. are its variables climatic water-budget despite other model, under F&C mance the of strength eod hi td ihihsa supincom- assumption an highlights study their Second, (2003) Currie and Francis the issues, these Despite IGM1 oe predictions Model ϭ pce eu Family Genus Species rdce iu culrichness. actual minus predicted clg,Vl 6 o 9 No. 86, Vol. Ecology, IGM2 September 2005 LATITUDINAL GRADIENTS 2275

TABLE 2. Extended.

B) F&C model

A) IGMs Model Actual values predictions ``Residuals'' Woody plant Angiosperm IGM1 IGM2 richness richness Species Genus Family Species Genus Family Area (%) Family Family 78 29 0 97 37 5 100 85 175 Ϫ71 Ϫ41 Ϫ11 Ϫ40 Ϫ27 Ϫ3 98 88 179 48 14 Ϫ3 69 23 2 97 87 172 Ϫ60 Ϫ19 Ϫ7 Ϫ14 1 5 100 91 156 27 22 Ϫ92219Ϫ10 100 56 77 Ϫ205 Ϫ107 Ϫ34 Ϫ129 Ϫ73 Ϫ16 100 94 158 Ϫ121 Ϫ53 Ϫ15 Ϫ79 Ϫ34 Ϫ5 100 87 152 Ϫ213 Ϫ100 Ϫ29 Ϫ143 Ϫ68 Ϫ12 100 94 150 Ϫ56 Ϫ40 Ϫ16 Ϫ6 Ϫ17 Ϫ4 100 85 153 156 68 17 152 65 18 92 87 183 307 148 33 291 139 31 95 86 185 Ϫ125 Ϫ56 Ϫ16 Ϫ80 Ϫ35 Ϫ5 100 87 152 131 43 10 134 43 12 60 NA NA 78 37 1 120 55 12 85 84 186 Ϫ24 Ϫ1 Ϫ4 Ϫ3 8 2 100 89 145 Ϫ55 Ϫ35 Ϫ33 2 Ϫ9 Ϫ19 88 59 87 Ϫ138 Ϫ66 Ϫ29 Ϫ92 Ϫ45 Ϫ18 95 89 130 Ϫ24 Ϫ24 Ϫ16 32 2 Ϫ3 100 79 151

171 84 10 181 88 12 100 69 155 Feature Special Ϫ142 Ϫ65 Ϫ18 Ϫ98 Ϫ45 Ϫ8 100 87 152 Ϫ120 Ϫ75 Ϫ26 Ϫ61 Ϫ47 Ϫ12 100 85 151 44 15 Ϫ8 91 36 3 100 82 158 Ϫ168 Ϫ97 Ϫ28 Ϫ100 Ϫ66 Ϫ12 100 90 159 Ϫ87 Ϫ52 Ϫ19 Ϫ31 Ϫ26 Ϫ6 95 83 152 Ϫ58 Ϫ31 Ϫ22 4 Ϫ3 Ϫ8 100 87 150 Ϫ118 Ϫ73 Ϫ22 Ϫ55 Ϫ44 Ϫ7 100 90 157 Ϫ215 Ϫ102 Ϫ28 Ϫ185 Ϫ89 Ϫ21 85 87 141 111 70 14 115 71 15 99 25 95

photosynthesis. The fundamental, dynamic parameters only recently been documented because few studies of climate are atmospheric moisture, energy, wind, and sample a suf®cient range in climate and richness to pressure. Since derived water-budget variables are sub- obtain the relationship empirically. Most only examine ordinate and conditional corollaries of realized cli- portions of the curve, documenting positive, negative, mate±soil interactions, it follows that they should be or insigni®cant (humid tropics) statistical relationships conditional correlates of associated richness. But be- between energy and richness. These results tend to be cause they are not parameters of climatological dy- treated as competing hypotheses, but our ®ndings and namics, it does not follow that they should describe those of Francis and Currie show that they can be rec- how climate per se relates to richness. onciled when energy is described as an optimal func- Third, the Francis and Currie (2003) study docu- tion. ments the fact that, when the global variation in PETan Fourth, the Francis and Currie study highlights the is sampled, which is not possible using only African issue of using only statistical strength to refute a hy- data, an optimal relationship between richness and en- pothesis. To test the global applicability of IGM1 in ergy pertains. Thus, Francis and Currie provide the ®rst principle (see Appendix B), they regressed their family independent empirical corroboration of the optimal re- richness data accordingly, as a function of increasing lationship between richness and energy ®rst docu- annual rainfall and optimal energy (PETmin) conditions. mented by O'Brien (1989, 1993). It is therefore puz- (Note that Francis and Currie cite Legates and Willmott zling that Francis and Currie (2003) argue that their [1992] and Ahn and Tateishi [1994] as the sources of ®ndings refute the existence of such a relationship. Vir- their data; the IGM was derived using climate data from tually all plant growth depends on liquid water, and Thornwaite and Mather [1962±1965]). Despite the dis- since liquid water availability depends on ambient en- crepancies in climate and richness data, they found that ergy conditions, some form of optimal relationship be- this relationship accounted for 63% of the global var- tween energy and richness should be expected (to ac- iation in angiosperm family richness (Francis and Cur- count for the solid and gaseous states of water). It has rie 2003:530). This is close to the 68% accounted for Special Feature al n miial luil nprinciple. in appli- plausible globally in- empirically are ®rst and the IGMs cable providing the this as that seen in evidence be model dependent can F&C study richness stronger their statistically case, family in the IGM1 rejected plant of they favor Although woody 1998). al. et for (O'Brien SAF1/IGM1 by 2276 owoypatrcns nArc oa,a ntepast the 1999 in Peters as today, and Africa (O'Brien in richness plant woody to contribution edaphic material important parent most the soil make and probably exotic reserves, these, water soil Of ground and content. reserves, rivers, nutrient water and permanent material, ground parent rivers, (pans), lakes exotic ephemeral especially a plant in in richness, variations included global of be explanation to complete to more need points that next parameters Kenya the as independent for factors hydrological maps associated and soil edaphic and hydrological, topograph- of ic, examination visual Africa 2000), southern al. et for (O'Brien ®ndings with Consistent in¯uencing factor richness. only the not Kenya is in climate that values emphasize richness actual and predicted between not climate. variation by residual the explained the doing, on analyses active so as focusing eliminated In factors, be can 2003). climate of 2001, effects edaphics, al. ®rst-order al. (e.g., et et O'Brien Whittaker factors 1989, 2000, (O'Brien and richness to variables relate shade) other how incor- analyses be of in richness'') can of (``potential scale constants scales as macro discrete porated the more at at predictions operational. working analysis, theory when hierarchy essence, making In trans-scalar to of and development importantly the models, to for most contribute used Perhaps can be they predictions. could GCM they GCMs, testing climate of in independent changes being Alternatively, patterns. future richness present-day to and/or alter could used past and di- how (GCMs) linked Models examine be Climate cli- Global can dynamic to IGMs rectly invoke the they the parameters, in that elsewhere matological case given eval- the Second, for is this evidence world. not of po- or line climatic whether a uating its provide with IGMs equilibrium the tential, near is be ¯ora Africa's to if assumed First, Other possible. unknown. also are are values applications actual where richness dict analaoe(.. u omeltwater). of to function due a as (e.g., expected alone that rainfall soil than or greater normal, is than richer moisture are (especially soils rivers exist, rivers) and is exotic lakes underprediction perennial perennial Gross where scarce). where a expected are as especially rivers expected and and that lakes rainfall, than than of moisture poorer in function are lower soils or where normal (e.g., its potential below is climatic richness extant where richness overpredict ntrso uuets mlctos rs differences gross implications, test future of terms In pre- to is models predictive of use obvious most The oeapiain n implications and applications Some fpeitv models predictive of ,b a, .IM hudgrossly should IGMs ). IHR IL TAL. ET FIELD RICHARD akn,B . .Fed .V onl,D .Cri,J.-F. Currie, J. D. Cornell, V. H. Field, R. A., B. Hawkins, Royal edition. Second life. of matrix a Water: 2000. F. Franks, rni,A . n .J ure 03 lblyconsistent globally A 2003. Currie. J. D. and P., A. Francis, ure .J,adV aun 97 ag-cl biogeograph- Large-scale 1987. Paquin. V. and J., cli- D. Currie, versus Regional 2004. Francis. P. A. and J., D. Currie, nrw,P,adE .OBin 00 lmt,vegetation, Climate, 2000. O'Brien. M. E. and P., Andrews, monthly Tateishi and Ahn 1994. Tateishi. R. and H., C. Ahn, owl,R . .Rhe,adN .Gtli 04 h mid- The 2004. Gotelli. J. N. and Rahbek, C. K., R. Colwell, Macroe- 2003. editors. Gaston, J. K. National and lianas. M., T. and Blackburn, shrubs trees, Kenya 1994. H. Beentje, omnso ale esoso hspaper. this of their versions for earlier reviewer Ole on anonymous Hawkins, one comments producing and A. in Currie, Bradford assistance David thank Vetaas, for we Hawkins Finally, climate±richness A. correlograms. on Bradford with collaboration assistance and for our GIS Stevens RJW Lucy of thank to also funding stage We for relationships. early Oxford of an We University at parameters. sup- the model for to and grateful and data are models climate their unpublished regarding plying questions our swering nrcns tmr iceesae fanalysis. of scales it discrete (hold more factor differences at for active richness causes an in other as analyzing in it when now eliminate constant) are to we richness, position plant a woody in to differences relate ®rst-order high climate how glob- of of and model systematic spots'' applicable a ally ``hot providing idiosyncratic by re- Lastly, as diversity. mountainous well in as richness gions, in gradients observed of elevational understanding better a and to climate contribute in should change of vectors in vertical gradient'' the and both horizontal ``latitudinal re¯ect that a parameters invokes be IGM2 richness. should out- one which globally, of richness come covari- and the rela- climate explains between biological dynamics ation that water±energy idea to the tivity support topographic also without They or three with data. use at for rich- versions levels, plant operational taxonomic woody providing for thereby potential ness, climatic the of model general interim (1998) O'Brien's of applicability global brof,E .OBin .E otr n .R .Turner. G. T. R. Mittelbach, J. and G. Porter, E. G. E. Kerr, O'Brien, T. M. J. E. Oberdorff, Kaufman, M. D. Guegan, UK. Cambridge, Chemistry, of Society ihesciaerltosi o nisem.American angiosperms. Naturalist for relationship richness±climate 327. clpten fseisrcns ftes Nature trees. of richness species of patterns ical Naturalist American Qian to Ricklefs. reply angiosperms: and in richness taxon on effect matic n rdcal rdet nmma pce ihesin richness London species Zoology, of mammal Journal Africa. in southern gradients predictable and GNV183. set Data Database, Resource Information Programme/Global balance. Environment water Nations and United evapotranspiration actual and potential ere ofr mrcnNaturalist we American have far? what so patterns: richness learned species and effect domain UK. Cambridge, University Press, Cambridge consequences. and concepts cology: Kenya. Nairobi, Kenya. of Museums aytak oDvdCri n nhn rni o an- for Francis Anthony and Currie David to thanks Many h eut fti td ohspotadetn the extend and support both study this of results The grid.unep.ch/ 161 ͘ :523±536. A L C CKNOWLEDGMENTS ITERATURE ONCLUSIONS C ITED 163 clg,Vl 6 o 9 No. 86, Vol. Ecology, 163 :780±785. :E1±E23[Epub]. 251 ͗ http://geodata. :205±231. 329 :326± September 2005 LATITUDINAL GRADIENTS 2277

2003. Energy, water, and broad-scale geographic patterns and human evolution. Oxford University Press, Oxford, of species richness. Ecology 84:3105±3117. UK. Huston, M. A. 1994. Biological diversity: the coexistence of O'Brien, E. M., and C. R. Peters. 1999b. Climatic perspec- species on changing landscapes. Cambridge University tives for Neogene environmental reconstructions. Pages Press, Cambridge, UK. 53±78 in J. Agusti, L. Rook, and P. J. Andrews, editors. Legates, D. R., and C. J. Willmott. 1992. Monthly average Hominoid evolution and climatic change in Europe. Vol- surface air temperature and precipitation: digital raster data ume 1. The evolution of Neogene terrestrial ecosystems in on a 30-minute geographic (lat/long) 360 ϫ 720 grid. In Europe. Cambridge University Press, Cambridge, UK. Global ecosystems database. Version 1.0. Disc A. National O'Brien, E. M., R. J. Whittaker, and R. Field. 1998. Climate Oceanic and Atmospheric Administration, National Geo- and woody plant diversity in southern Africa: relationships physical Data Center, Boulder, Colorado, USA. at species, genus and family levels. Ecography 21:495±509. O'Brien, E. M. 1989. Climate and woody plant species rich- Polhill, D. 1988. Flora of East Africa: index of collecting ness: analyses based upon southern Africa's native ¯ora localities. Royal Botanic Gardens, Kew, Richmond, Surrey, with extrapolations to sub-Saharan Africa. Dissertation. UK. University of Oxford, UK. Qian, H., and R. E. Ricklefs. 2004. Taxon richness and cli- O'Brien, E. M. 1993. Climatic gradients in woody plant spe- mate in angiosperms: Is there a globally consistent rela- cies richness: towards an explanation based on an analysis tionship that precludes region effects? American Naturalist of southern Africa's woody ¯ora. Journal of Biogeography 163:773±779. 20:181±198. Thornthwaite, C. W., and J. R. Mather. 1962±1965. Average O'Brien, E. M. 1998. Water±energy dynamics, climate, and climatic water balance data of the continents: Africa. Pub- prediction of woody plant species richness: an interim gen- lications in Climatology 15±18, Laboratory of Climatology, eral model. Journal of Biogeography 25:379±398. Centerton, New Jersey, USA. O'Brien, E. M., R. Field, and R. J. Whittaker. 2000. Climatic Whittaker, R. J., K. J. Willis, and R. Field. 2001. Scale and gradients in woody plant (tree and shrub) diversity: water± species richness: towards a general, hierarchical theory of energy dynamics, residual variation, and topography. Oikos species diversity. Journal of Biogeography 28:453±470. 89:588±600. Whittaker, R. J., K. J. Willis, and R. Field. 2003. Climatic- O'Brien, E. M., and C. R. Peters. 1999a. Landforms, climate, energetic explanations of diversity: a macroscopic per- ecogeographic mosaics, and the potential for Early Hom- spective. Pages 107±129 in T. M. Blackburn and K. J. Gas- inid biodiversity. Pages 115±137 in T. G. Bromage and F. ton, editors. Macroecology: concepts and consequences.

Schrenk, editors. African biogeography, climate change, Cambridge University Press, Cambridge, UK. Feature Special

APPENDIX A Figures showing the relationship between PET data used for IGMs and PET data used for the F&C model for Kenyan climate stations (Fig. A1), the predicted genus richness for Africa based on (a) IGM1 and (b) IGM2 (Fig. A2), color reproductions of Figs. 3 and 4, and associated literature citations are available in ESA's Electronic Data Archive: Ecological Archives E086-120-A1.

APPENDIX B A description of the general methodological constraints, a summary of transformation of southern African models into IGMs, and associated literature citations are available in ESA's Electronic Data Archive: Ecological Archives E086-120-A2. Ecology, 86(9), 2005, pp. 2278±2287 ᭧ 2005 by the Ecological Society of America

TESTING FOR LATITUDINAL BIAS IN DIVERSIFICATION RATES: AN EXAMPLE USING NEW WORLD BIRDS

MARCEL CARDILLO,1,2,4 C. DAVID L. ORME,1 AND IAN P. F. O WENS1,3 1Division of Biology, Imperial College London, Silwood Park, Ascot SL5 7PY UK 2Institute of Zoology, Zoological Society of London, Regents Park, London NW1 4RY UK 3NERC Centre for Population Biology, Imperial College London, Silwood Park, Ascot SL5 7PY UK

Abstract. Study of the latitudinal diversity gradient to date has focused largely on pattern description, with relatively little work on the possible mechanisms underlying the pattern. One proximate mechanism is a latitudinal bias in the discrepancy between speciation and extinction rates, leading to higher rates of species diversi®cation toward lower latitudes. Despite being central to many explanations for high tropical diversity, this mechanism is tested very rarely. We discuss some of the problems involved in testing for latitudinal bias in diversi®cation rates and present an example phylogenetic analysis for endemic bird genera of the New World. The results provide evidence for higher diversi®cation rates in clades inhabiting lower latitudes, both when genera are considered independent and when phy- logeny is controlled for using independent contrasts. High rates of diversi®cation are also associated with larger geographic area and higher net primary productivity, although these do not fully account for the latitudinal effect. The latitudinal pattern is stronger in younger clades, supporting the prediction of a simple model in which the signal of latitudinal bias in diversi®cation rates diminishes as clades age and become saturated with species. Our study demonstrates that a clade-based approach can help answer important questions that a geographic approach cannot, but large phylogenies and geographic databases are needed to cope with the large amount of noise inherent in this type of analysis. Key words: diversi®cation; extinction; latitudinal diversity gradient; New World birds; speciation; species richness.

INTRODUCTION has been spent on empirical testing of possible mech- anisms than on pattern description and correlative tests. Explaining why there are more species in the tropics In this paper, we focus on the idea that the proximate Special Feature compared to higher latitudes is one of the great con- mechanism for the latitudinal diversity gradient is a temporary challenges of ecology. The continuing higher rate of species diversi®cation at lower latitudes, strong interest in the latitudinal diversity gradient is where diversi®cation is de®ned as the discrepancy be- demonstrated by the large number of papers recently tween speciation and extinction rates. A latitudinal bias published on the topic, the frequent publication of re- in diversi®cation rate is central to a number of high- views (Rohde 1992, 1999, Rosenzweig 1995, Gaston pro®le explanations for high tropical species richness. 2000, Willig et al. 2003), and the regular emergence Rosenzweig's (1992, 1995) area hypothesis states that of novel hypotheses to explain the pattern (Stevens the larger geographic distributions attainable by trop- 1989, Colwell and Hurtt 1994, Dynesius and Jansson ical species serve both to elevate speciation rates (by 2000, Guernier et al. 2004). Although the pattern re- increasing the likelihood of allopatric subdivision) and mains famously poorly explained, slow progress to- reduce extinction rates (by providing a buffer against wards a solution is being made. For example, recent range or population decline). The species±energy hy- analyses of large data sets point increasingly towards pothesis (Wright 1983, Wright et al. 1993) proposes climate rather than geographic area as the major driving that extinction rates are reduced in lower latitude spe- force of higher tropical species richness (Hawkins and cies as a result of the higher populations sustainable Porter 2001, Macpherson 2002, Francis and Currie by the greater amount of available energy. Rohde 2003, H-Acevedo and Currie 2003, Hawkins et al. (1992, 1999) suggests that speciation rates at lower 2003a, b, Hawkins and Porter 2003). However, much latitudes are elevated by shorter generation times driv- uncertainty remains. In particular, the mechanisms by en by higher temperatures. Other authors have pre- which latitudinal variation in climate determines spe- dicted higher extinction rates at high latitudes due to cies numbers largely remain a mystery: far less effort ``harsher'' or more variable climatic conditions (Cra- craft 1985, Dynesius and Jansson 2000). Manuscript received 25 January 2005; accepted 31 January 2005. Corresponding Editor: B. A. Hawkins. For reprints of this Despite being central to many explanatory hypoth- Special Feature, see footnote 1, p. 2261. eses, faster diversi®cation is not a necessary precon- 4 E-mail: [email protected] dition for higher species richness at lower latitudes. 2278 September 2005 LATITUDINAL GRADIENTS 2279

that direct empirical tests for latitudinal bias in diver- si®cation rates are relatively scarce. Previous tests are almost entirely con®ned to the paleontological litera- ture, and these have tended to support higher diversi- ®cation rates in the tropics. Faster tropical diversi®- cation has been implied by lower average taxon ages or higher rates of ®rst appearances in the tropics (Stehli et al. 1969, Durazzi and Stehli 1972, Hecht and Agan 1972, Jablonski 1993, Flessa and Jablonski 1996), or from a higher rate of diversity accumulation through time in tropical groups (Crame 2000, 2001, Buzas et al. 2002). FIG. 1. Simple model of the growth in the number of An alternative approach is to infer diversi®cation species in tropical and temperate regions or clades, from zero rates using the information contained within phylog- to saturation levels (SÃ and SÃ ). The solid curve represents trop temp enies of extant taxa. Using phylogenetic methods al- diversi®cation of tropical groups, the dotted curve represents diversi®cation of temperate groups when a latitudinal bias in lows tests to be done for groups with a poor or unre- diversi®cation rate exists, and the dashed curve represents liable fossil record (e.g., the majority of terrestrial ver- diversi®cation of temperate groups when there is no latitu- tebrates). Whereas paleontological analyses are usually dinal bias. geographic (involving the comparison of assemblages across different geographic regions or grid cells), phy- This can be visualized using the simple model shown logenetic analyses are explicitly clade based. The mean in Fig. 1. The number of species in a region increases diversi®cation rate of a clade throughout its history is from zero to a saturation level where rates of speciation a function of its age and number of extant species. In and extinction are in equilibrium. The inherent diver- the absence of dated phylogenies, sister-clade com- si®cation rate of the temperate region may be equal to parisons are often used to infer relative diversi®cation Feature Special (dashed line) or lower than (dotted line) that in the rates, an approach that is now standard practice in com- tropics, but a lower saturation level keeps temperate parative tests of the effects of life history or ecological species richness below that in the tropics. This model traits on diversi®cation rates (see Barraclough et al. can be applied equally to the growth of individual 1998). clades. A regional assemblage of species consists of However, testing for geographic variation in diver- numerous clades, at different stages along this trajec- si®cation rates by this method presents additional dif- tory. Below saturation (e.g., at time t1), any inherent ®culties and has received very little attention. As far variation in diversi®cation rate between tropical and as we are aware, only three times has latitude been temperate clades should be easily detectable as lati- tested explicitly as a predictor of diversi®cation rates tudinal differences in mean diversi®cation rates over using phylogenetic methods. Farrell and Mitter (1993) the lifetime of clades (the period from t0 to t1). When attempted the ®rst such test by identifying ®ve tropical± saturation is reached (time t2), the diversi®cation rate temperate pairs of sister clades of phytophagous in- of temperate and tropical clades at that point in time sects, but with such a small sample their result was will be equal (speciation minus extinction ϭ 0), but inconclusive. Cardillo (1999) extended this approach comparisons of mean diversi®cation rates over the life- to passerine birds and swallowtail butter¯ies, and in- time of clades (the period from t0 to t2) should still creased the number of useful sister-clade comparisons carry the signal of any inherent latitudinal bias. This by treating latitude as a continuous variable, measured signal will diminish in strength after long periods at as the midpoint between the northern and southern lim- saturation (time t3). This model therefore predicts that its of the geographic range of each species. This study any latitudinal bias in diversi®cation rates should be suggested that lower latitudes were indeed associated most strong among younger clades, which have yet to with higher rates of diversi®cation. Most recently, Da- become, or have only recently become, saturated with vies et al. (2004b) used a similar approach on a much species. It also demonstrates that latitudinal variation larger data set for angiosperm families (see also Davies in species numbers can arise without any latitudinal et al. 2004a). Again, this analysis suggested a link be- bias in diversi®cation rates. Establishing whether or tween latitude and diversi®cation rates, although the not diversi®cation rates show latitudinal bias, there- test for latitudinal variation of Davies et al (2004b) was fore, is an important step in understanding the origin incidental to an analysis of the effects of environmental of the latitudinal diversity gradient. energy. Taken together these three studies serve to highlight a number of challenges in doing comparative TESTING FOR LATITUDINALLY BIASED tests that are both phylogenetically and geographically DIVERSIFICATION RATES explicit. These challenges may be summarized as fol- Given its central position in many hypotheses to ex- lows. (1) Phylogenetic accuracy. As with any phylo- plain the latitudinal diversity gradient, it is surprising genetic comparative analysis, the reliability of sister- Special Feature plcto frbs hlgntcreconstruction phylogenetic robust (2) routine methods. the of with changing is application This accepted. and acknowl- be and phylogeny, must edged used a phylogeny of the no of selection or limitations the little the is in there available means choice usually num- comparisons suf®cient of the a identify ber However, to phylogenies results. large for con¯icting recon- need phylogeny give of may methods Dif- and struction used. data phylogeny of the types are of ferent used) quality (where the on lengths contingent branch and groupings clade 2280 itiuin fteioae ouain aebecome have historical populations re¯ect isolated may the maps if occurrence distributions of Even extent true. be not so, may which boundaries, polygon with- the distribution in continuous a from inhabits species implies records the This that occurrence locations). separated, of widely polygons often number (i.e., single, a occurrence of around extent drawn the as often presented are distributions world Geographic the surveyed. of poorly parts are some taxa and other records world, few the relatively of have parts some in taxa well-studied mates. d ocntutlrepyoeis(.. aise al. et Davies (e.g., phylogenies 2004 meth- large supertree construct of to use ods increasing the with this changing However, is available. the currently with phylogenies even test, largest statistical com- acceptable enough an produce Farrell for to parisons unlikely by is attempted (1993), Mitter as and clades, an sister Restricting ical±temperate sizes trop- low. non-overlapping sample between power comparisons phylogeny, to statistical analysis large and a small small with a were even only permis- so were result, contrasts sible, a sister-clade available As of clades. subset sister over- and between separation latitudinal imposi- lap of the criteria required strict of comparisons tion pairwise of use the ilb oe hr egahcrne r otyrel- mostly (5) are small. ranges atively geographic (Cardillo where error lower overlap be statistical latitudinal will substantial 1 of likelihood Type elevates The than 2002). probably rather this 2 Type although clades, ranges, their two in overlap geographic latitudinal between substantial be differ still may there estimates point itudinal umr fweetebl fteae fterneof range the of (4) area lies. the species of a bulk misleading the lati- a where high is of midpoint into summary latitudinal but extending case, range, this band circular In coastal tudes. a narrow occupies entirely a it almost with where found tropics be the latitude. may in to species respect a with example, shaped For irregularly are low that a es with rang- have circular, species and many However, ratio. small perimeter:area is where range species geographic a for its estimate all point of is accurate midpoint midpoints relatively Latitudinal a latitudinal clade. a this the within did of species member (1999) mean the Cardillo using estimate. by point a sum- be as must clade marized each by inhabited the latitudes models, of range statistical in variable predictor continuous b .(3) ). lhuhdsrbtosaewl-nw o some for well-known are distributions Although umrzn ld latitude. clade Summarizing oso data. of Loss aiuia overlap Latitudinal cuayo egahcrneesti- range geographic of Accuracy nCril' 19)analysis, (1999) Cardillo's In vnweelat- where Even . ACLCRIL TAL. ET CARDILLO MARCEL o s sa as use For euto aia rgetto) (6) fragmentation). habitat as of (e.g., result distribution a continuous once a from separated rpi n itrcli¯ecs(akn tal. et (Hawkins biogeo- in¯uences region possible historical single for 2003 a control and of of graphic endemics measure the a to provided analysis restrict- the in and well-known, ing birds relatively New of are World the distributions New of the geographic clades the bird since sec- endemic previous World, on the of in focused some shown We 7) with tion. and dealing 3, (2, by challenges the (1999) the on Cardillo improve of phy- to analysis using attempts which rates methods, diversi®cation logenetic in bias latitudinal for N ewe ar ftx ( taxa of hybridized pairs of between temperatures DNA (1990). melting Ahlquist in differences and The Sibley hy- of DNA the phylogeny using bridization ages clade estimated We independent latitude contrasts. phylogenetically we and using so analyses independence, rates of repeated assumption diversi®cation this phy- violate any in would course, bias Of analysis. indepen- logenetic the an in as clade observation each dent treat to this us Using allowed diver- phylogeny. approach dated species a using of clades rates of si®cation absolute estimated phy- we a from logeny, clades sister of pairs separated latitudinally are variation clade severe. a less of and be latitude to the clades likely of estimate among point the overlap around problems the latitudinal lati- so taxa, of of higher range within restricted species more than a tudes within occupy species to addition, tend In genera levels. taxonomic available endemic lower size) sample at of larger number (hence clades larger World New the for primarily analysis, siaeo h ag fevrnetlcniin ex- species. conditions a environmental by of perienced point range a the around of variance error estimate substantial be latitude, may with in there As 2004). interest Diniz-Filho rich- and of species (Hawkins of ness are distribution that geographic the factors explaining historical environmental underlying or the but gradients it itself, Ultimately, latitude pattern. not this is of causes possible further no testing went in it rate, diversi®cation in bias latitudinal h ag htteseishsocpe hogotits throughout represents occupied it has (7) species well history. the how that know range with to the mapped is dif®cult species is a accuracy, of range contemporary of Even the 1998). if result Gaston 1997, the al. in et as (Price contract events or speciation and changes expand environmental but to time, response of periods long over terns. boundaries. epeethr neapeo napoc otesting to approach an of example an here present We oaodtels fdt novdwt choosing with involved data of loss the avoid To A b .W hs eu stetxnmclvlo the of level taxonomic the as genus chose We ). lhuhCril' 19)aayi upre a supported analysis (1999) Cardillo's Although N E APEUSING XAMPLE egahcrne r nieyt estatic be to unlikely are ranges Geographic niomna rvr flttdnlpat- latitudinal of drivers Environmental B IRD Methods C N LADES EW ⌬ T W 50 )wr ovre to converted were H) clg,Vl 6 o 9 No. 86, Vol. Ecology, ORLD yai range Dynamic E NDEMIC September 2005 LATITUDINAL GRADIENTS 2281 millions of years (106 yr) using the calibration factors centroid of the clade range. The area of the reprojected ⌬ ϭ ϫ 2 suggested by Sibley and Ahlquist: T50H 1.0 2.3 range was then recorded in km . We included measures 6 ⌬ ϭ ϫ 6 10 yr for passerines and T50H 1.0 4.2 10 yr for of three climatic factors, in order to test possible un- nonpasserines. Species diversi®cation rate was then es- derlying causes of any latitudinal bias in diversi®cation timated for each clade as ln(number of species)/age rates: mean annual temperature, annual actual evapo- (Purvis 1996). We only accepted an age estimate for a transpiration (AET), and annual net primary produc- clade from the Sibley and Ahlquist phylogeny if its tivity (NPP), all of which have previously been shown sister clade was also present in the phylogeny, as the to be correlated with species richness at large scales absence of the sister clade means that the age of a clade (Currie 1991, Wright et al. 1993, H-Acevedo and Currie will appear arti®cially high, giving misleadingly low 2003, Hawkins et al. 2003a, b) or suggested to in¯u- estimates of diversi®cation rates. We established ence rates of speciation or extinction (Currie 1991, whether the sister of each clade was present in the Rohde 1992, 1999, Wright et al. 1993). All three cli- Sibley and Ahlquist phylogeny by using the Sibley and matic variables were recorded as raster data on geo- Monroe (1990) classi®cation of birds to determine the desic grids. The temperature and AET data (0.5Њ res- full list of genera within the next most inclusive taxon olution) were from the United Nations Environment (usually a tribe or subfamily). If all other genera within Program GRID databases (UNEP 2003), and the NPP the next most inclusive taxon were represented in the data (1Њ resolution) were from Woodward et al. (1995). Sibley and Ahlquist phylogeny, we accepted the age Estimates of climatic variables for each species were estimate of the clade because its sister clade must be obtained by converting the breeding range for the spe- present. In some cases, if not all genera were present cies and each climatic variable onto a 0.05Њ grid and in Sibley and Ahlquist, we were able to refer to in- then calculating the mean value across the cells oc- dependent phylogenies to determine the sister of a cupied by the species. We then summarized the variable clade. In this way, we were able to estimate ages for for each clade as the mean of the member species val- 37 New World endemic genera. We are conscious of ues. All the above analyses were performed using the uncertainty of date estimates derived from DNA ArcMap 8.2 (ESRI 2002). Feature Special hybridization data, and of criticisms of some of the We tested for associations between diversi®cation methods used by Sibley and Ahlquist (see O'Hara rate and predictor variables using multiple linear re- 1991). However, this phylogeny is currently the largest gressions. We ®rst examined histograms of all variables available for birds built using a single method, and thus and univariate plots against diversi®cation rate to de- the only one with which a test such as this can be termine the most suitable transformations to stabilize carried out (see Bennett and Owens 2002). We em- variances and normalize error distributions, and to test phasize that the analysis we present here is illustrative, for second- and third-order polynomial effects. A min- and that the same approach can be applied with new imum adequate model (MAM) was found by ®tting all and perhaps more rigorous dated phylogenies as they predictors, then sequentially removing each predictor become available. that contributed the smallest amount of marginal var- All geographic data were processed within a geo- iance to the model until all remaining predictors were graphic information system. This had several important signi®cant at P Յ 0.05. We then reintroduced predic- advantages over the analysis of Cardillo (1999): it al- tors, and interaction terms, one at a time and tested the lowed species latitudes to be measured more accurate- model for signi®cance each time. ly, and it allowed us to include geographic area and Because of the possibility of phylogenetic bias in climatic factors as covariates in the analysis. The geo- diversi®cation rates, we repeated these tests using phy- graphic range characteristics for the clades were mea- logenetically independent contrasts, calculated using sured as follows, using polygon range data from Ridg- software written in R by A. Purvis (available online).5 ley et al. (2003). First, we measured the latitude of the All regressions on independent contrasts were forced centroid of the breeding range of each species, a more through the origin (Garland et al. 1992). Because we accurate characterization of the latitude of a species did not use a complete phylogeny, we could not test than latitudinal midpoint when ranges are irregularly clade area as a predictor of diversi®cation rate using shaped. The latitude of each clade was then summa- independent contrasts calculated for internal nodes. rized as the mean of the species latitudinal centroids; Our independent-contrasts tests using clade area are effectively providing a measure that is weighted by therefore restricted to contrasts calculated for terminal species richness. We then measured the total geograph- nodes only. ic area occupied by each clade, a factor shown to be a A prediction of the model presented in the Intro- signi®cant predictor of diversi®cation rates among bird duction is that the strength of any latitudinal bias in families (Owens et al. 1999, Bennett and Owens 2002). diversi®cation rates, measured as the mean rate over The non-overlapping breeding ranges of species were the lifetime of a clade, will depend on the point at which combined into a single range for each clade, which was most clades appear along the trajectory shown in Fig. then reprojected into a Lambert Azimuthal Equal Area projection with central meridian and parallel set to the 5 ͗http://www.r-project.org/͘ Special Feature rsswr sd( used con- were independent when trasts signi®cant but NPP), for 0.062 ϭ 10 and age clade included we effects, age latitude test such a To clades. for any younger levels for among saturation stronger at and been time, di- have long to in that clades likely bias older more latitudinal among are weaker any be of to rate signal versi®cation the expect We 1. results Cross-taxon A) 2282 oe aiuia etod,bt ncostxntss( tests with cross-taxon clades Diver- in both in 3). centroids, higher 2, latitudinal lower signi®cantly Figs. was 1, rate (Table si®cation predictors single as rate diversi®cation with associated signi®cantly were (NPP) f39.48 of e n le hntemda eu g f23.52 of age genus median the than older youn- and clades lat- ger for of separately effect rate the diversi®cation on tested itude also We latitude. with together )Idpnetcnrssresults Independent-contrasts B) Fg ) oeaietedge owihteegenera these which to degree the have examine to To genera of tend 2). majority (Fig. which the than species, centroids one latitudinal higher only with genera of tests. in independent-contrasts predictors in or single tests as cross-taxon latitude signi®- with not associated were cantly temperature and evapo- Actual (AET) low. was nodes transpiration test this terminal of power only the used, because were were although contrasts in (44.7%), independent variance used when the of rate proportion diversi®cation substantial a explained ABLE rs-ao et ( in tests nonsigni®cant marginally cross-taxon were with rate associations positive diversi®cation NPP, and area clade both For aiue(ere rmequator) from (degrees Latitude Notes: ra enana eprtr,ana culeaornprto AT,adntpiaypoutvt NP nbird on (NPP) productivity primary net and (AET), evapotranspiration actual annual temperature, rates. diversi®cation annual mean area, aiue ld ra n e rmr productivity primary net and area, clade Latitude, E m)0115 .3 .507 0 0.77 0.05 0 0.98 0.039 0.08 1.55 0.035 0.1 1.52 0.09 C/m (kg NPP (mm) AET C/m (kg NPP (mm) AET 6 aiue(ere rmequator) from (degrees Latitude .2)adidpnetcnrsstss( tests independent-contrasts and 0.027) ² ep( Temp ( Temp a o o o h etwt oopc® ldsecue) l rdcosecp ld raael-rnfre;tediver- the ln-transformed; are area clade except predictors All excluded). clades transformed. monospeci®c square-root with is test rate the si®cation for low too was rs-ao n needn-otat et,rsetvl) n ihmnseiccae xldd( excluded clades monospeci®c with and respectively), tests, independent-contrasts and cross-taxon T otat) needn-otat eut o ld raaefrtria oe ny( only nodes terminal for are area clade for results Independent-contrasts contrasts). n etr fordt e stelreproportion large the is set data our of feature One P P rfrcostxntss n h einndlage nodal median the and tests, cross-taxon for yr * ld ra(km area Clade ld ra(km area Clade Յ Յ .()Costxnrslsad()idpnetcnrssrslsfrsprt fet flttdnlcnri,clade centroid, latitudinal of effects separate for results independent-contrasts (B) and results Cross-taxon (A) 1. 0.1. ** 0.05; ϫ Њ Њ eut r hw o ersin fec aibeuigteetr aast( set data entire the using variable each of regressions for shown are Results )00 .7000 .90 0.59 0 0.04 0.56 0 0.05 0.57 0 0.05 0.61 0.05 C) C) ϫ Variable ld g neato em ntemodels the in terms interaction age clade 10 2 2 6 .628* .6 .710 0.001 1.01 0.018 0.07 1.19 0.167 0.11 2.83** 0.07 0.16 1.93² 0.11 ) ) 2 2 rfridpnetcnrsstests. independent-contrasts for yr P ϫ ϫ Յ P P 10 10 0.01. ϭ ϭ 6 6 .0 .²00900716 0.077 1.68 0.447 0.007 3.02* 0.059 0.02 1.8² 0.008 ) ) Results .1 and 0.015 and area clade for 0.08 P Coef®cient ϭ Ϫ Ϫ 0.06 0.06 ACLCRIL TAL. ET CARDILLO MARCEL .0) Area 0.008). P ϭ niedt set data Entire 0.023). P Ϫ Ϫ ϭ ϫ P .8 0.118 2.38* 0.108 2.31* tr ld g,adapstv latitude positive a and latitude, were age, the (MAM) clade in model included adequate minimum predictors ®nal independent anal- only cross-taxon the In yses, 2). mul- (Table in models variables regression other tiple with included when rate cation (Table substantially 1). sig- declined detect to also tests associations the ni®cant dis- of power rate the although diversi®cation appeared, the with and associations ¯atter For 14 slightly signi®cant analysis. became the slopes the with predictors, from most tests dropped the genera repeated monospeci®c we di- rate, and latitude versi®cation between association negative the drive onst atrdvricto ncae fNwWorld New of clades in diversi®cation analysis faster our diversi®cation to set, data points species the of in limitations the Within trends rates. geographic latitudinal other for test or to used be can methods phylogenetic lds u osgicn soito mn l clades. young old among among association signi®cant rate no be- but diversi®cation association clades, and positive latitude signi®cant a tween was independent-contrasts there and tests, tests Fig. cross-taxon 3, (Table both clades In old 2). and young on tests separate corroborated by was association rate tude±diversi®cation diver- on NPP of effect rate. nonlinear si®cation lati- positive, included a MAM and tude the analyses contrasts, In 1. independent Fig. in using shown supporting model the clades, of younger prediction is the for rate negative diversi®cation on steeply effect more latitudinal the the that of indicates interaction slope positive The term. action aiuewssgicnl soitdwt diversi®- with associated signi®cantly was Latitude h nlssw rsn eei neapeo how of example an is here present we analysis The h feto ld g ntesrnt ftelati- the of strength the on age clade of effect The 2 Coef®cient n Ϫ Ϫ D ϭ n ISCUSSION 0.04 0.04 ϭ 0cnrss ere ffreedom of degrees contrasts; 10 oopc®sexcluded Monospeci®cs 7cae n 5cnrssfor contrasts 35 and clades 37 clg,Vl 6 o 9 No. 86, Vol. Ecology, n Ϫ Ϫ ϫ ϭ 0.074 1.64 . 0.054 1.5 tr ld g inter- age clade 3cae n 21 and clades 23 2 September 2005 LATITUDINAL GRADIENTS 2283

parent diversi®cation rate of zero, since any speciation events have been balanced by extinctions. However, because these clades do vary in age and we have no way of knowing how imminent the next bifurcation event might be, any real variation in diversi®cation rate among monospeci®c clades may be obscured. Never- theless, weakening of the latitude±diversi®cation rate association when monospeci®c clades are dropped does not, in our view, necessarily nullify the result obtained when all clades are included in the analysis. The ten- dency for monospeci®c clades to be found at higher latitudes than most other clades in our data set is con- sistent with latitudinal variation in diversi®cation rates. Furthermore, there is still a positive latitude±diversi- ®cation rate slope with monospeci®c clades excluded, and the fact that this slope is not signi®cantly different from zero may result simply from the substantially re- duced power of the test. This underscores the impor- tance of large phylogenetic and geographic data sets in testing for geographic variation in diversi®cation rate: with larger data sets, the in¯uence of monospeci®c clades on patterns of diversi®cation rate will be testable with more power. A potential problem of using molecular data to es- timate the ages of bird clades is the possibility of lat- Feature Special itudinal bias in rates of molecular evolution, and thus in estimates of diversi®cation rate. For example, higher temperatures at low latitudes may lead to faster indi- vidual growth rates and shorter generation times, which could speed the rate of molecular evolution (Rohde 1992). However, Bromham and Cardillo (2003) have recently examined this possibility and found no evi- dence for latitudinal bias in rates of molecular evolu- tion in birds, using both DNA hybridization data and FIG. 2. Association between latitude (mean latitude of molecular sequence data. In any case, any such bias species geographic centroids) and diversi®cation rate would be likely to be conservative, as a higher rate of (ln[species]/106 yr), among New World endemic bird clades: (a) the association in arithmetic space; (b) the association molecular evolution at lower latitudes would increase with axes square-root transformed as in the analysis, with estimates of node ages, making diversi®cation rates of regression lines for young clades (solid line) and old clades lower latitude clades appear arti®cially low. Therefore, (dashed line); (c) the association using phylogenetically in- we do not believe that our result could have been an dependent contrasts, with regression lines for young and old clades. artifact of latitudinal bias in rates of molecular evo- lution. One assumption that many phylogenetically based endemic birds that inhabit lower latitudes. This pattern analyses of geographic patterns must make is that con- is found regardless of whether diversi®cation rates are temporary geographic ranges of species represent, at considered to be independent of phylogeny, or phy- least approximately, the distributions species have oc- logeny is controlled for using independent contrasts. It cupied throughout their histories. Unfortunately, we do also appears that the association between latitude and not know the extent to which this is true, given that diversi®cation rate is stronger among younger clades, geographic ranges expand, contract, and shift over time supporting the prediction that the association should (Price et al. 1997, Gaston 1998). Furthermore, climatic diminish as clades age beyond the point at which they zones are also nonstatic: within the lifetimes of many become saturated with species. of the genera included in our analyses, there have been The pattern of higher rates of diversi®cation at low periods when climatic conditions we now consider latitudes that we have detected here does appear to be ``tropical'' extended to far higher latitudes than they driven, at least in part, by the tendency for monospe- do today (Brown and Lomolino 1998). Ideally, an anal- ci®c clades to inhabit higher latitudes. Although clades ysis such as ours would make use of reliably recon- that are currently monospeci®c may well have under- structed paleodistributions and paleoclimates to sum- gone periods of rapid cladogenesis, they have an ap- marize the mean latitudinal position and climatic con- Special Feature ieyta egahcrne rc ogtr expan- seems long-term it track 2004), ranges Wiens geographic sug- 1999, that been al. likely has et as (Peterson niches, gested phylogenetic ecological for of tendency a conservatism is pos- there be never if will However, species sible. of numbers paleodistri- large reconstructing for and butions poor, relatively birds re- is fossil For the cord however, be 1996). vertebrates, terrestrial indeed Jablonski other may and and this Flessa bivalves, (e.g., marine possible as fossil such high-quality a record, have its that throughout groups species some each For history. of range the within ditions 2284 otat.Alpeitr xetae r ntasomd ies®ainrt ssur-ottransformed. square-root independent is phylogenetically rate using diversi®cation associations ln-transformed; column: are Right-hand area observations. except predictors independent All as contrasts. treated genera bird with F AT m,adntpiaypoutvt NP gC/m kg (NPP, productivity primary net and mm), (AET, IG .Ascain ewe ld ra(iloso km of (millions area clade between Associations 3. . ACLCRIL TAL. ET CARDILLO MARCEL 2 ,addvricto aeo idcae.Lf-adclm:associations column: Left-hand clades. bird of rate diversi®cation and ), 2 ,ma nultmeaue( temperature annual mean ), egahcae n lmt.Rsnwi 19,1995) (1992, are Rosenzweig contenders climate. and major area two geographic The mechanism? diversity un- this that latitudinal gradients derlie environmental the the are of what origin gradient, the for mechanism lifetimes. their experienced throughout conditions rea- species a climatic by provide of least summary at sonable may distributions lat- there- current if dynamic, been fore, Even have 1997). species of al. positions et itudinal been (Price has birds this in zones; demonstrated climatic of contractions and sions fvraini ies®ainrtsi proximate a is rates diversi®cation in variation If Њ ) nulata evapotranspiration actual annual C), clg,Vl 6 o 9 No. 86, Vol. Ecology, September 2005 LATITUDINAL GRADIENTS 2285

TABLE 2. Cross-taxon tests (n ϭ 37 clades) and independent-contrasts tests (n ϭ 35 contrasts) for combined effects of latitudinal centroid, mean annual temperature, annual actual evapotranspiration (AET), and net primary productivity (NPP) on bird diversi®cation rates.

Cross-taxon results Independent-contrasts results Predictor Coef®cient t Partial r2 Coef®cient t Partial r2 Intercept 0.51 6.23*** Latitude (degrees from equator) Ϫ0.14 Ϫ3.91*** 0.133 Ϫ0.06 Ϫ1.84² 0.143 Clade area (km2 ϫ 106) Temp (ЊC) AET (mm) NPP (kg C/m2) 0.55 2.7* 0.065 (NPP)2 Ϫ0.18 Ϫ2.42* 0.108 Clade age Ϫ0.007 Ϫ3.48** 0.089 Latitude ϫ AET Latitude ϫ clade age 0.002 2.74** 0.145 Notes: Results shown are for minimum adequate models (see An example using New World endemic bird clades; Methods). Clade area was not ®tted in the model using independent contrasts. * P Յ 0.05; ** P Յ 0.01; *** P Յ 0.001. ² P Յ 0.1. has argued that because the tropics are the largest of 2002), and may re¯ect an increased potential for pop- the world's major bioclimatic zones, tropical clades are ulation divergence or a greater buffer against extinc- able to occupy larger areas, which promotes a higher tion. The in¯uence of climate on diversi®cation rates speciation rate and a lower extinction rate. The area is supported by a positive association with net primary hypothesis has been controversial: for example, it has productivity (NPP), but not with actual evapotranspi- been argued that land area does not even increase to- ration (AET) or temperature. Compared to AET or tem- Feature Special wards low latitudes (Rohde 1997). The balance of em- perature, NPP may be a closer indicator of the habitat pirical evidence seems not to favor the area hypothesis, features that promote speciation in birds, such as the with recent large geographic analyses offering little structural complexity of tropical forests. It should be support for area as a major driver of the latitudinal kept in mind, however, that NPP and AET are highly diversity gradient (Hawkins and Porter 2001, Mac- correlated across bird clades, that diversi®cation rate pherson 2002). Climate, on the other hand, varies un- also showed a positive trend against AET (Fig. 3), and arguably across latitudes, and there is strong support that the overall power of the data set to detect signif- for associations between climatic factors and species icant patterns was relatively low. richness of birds and other major taxonomic groups If large clade area and high NPP are associated with across latitudes (Currie and Paquin 1987, Currie 1991, high rates of diversi®cation among bird clades, are Francis and Currie 2003, H-Acevedo and Currie 2003, these the factors driving the increase in diversi®cation Hawkins et al. 2003a, b, Hawkins and Porter 2003). rates towards lower latitudes? Both clade area and NPP Our study has taken a different approach from the increase strongly towards lower latitudes (clade area: majority of previous studies by examining area, cli- P ϭ 0.008, r2 ϭ 0.16; NPP: P Ͻ 0.0001, r2 ϭ 0.82), mate, and latitude as properties of clades rather than so these two factors must certainly account for at least of geographic entities such as grid cells. Although our some of the latitudinal bias in diversi®cation rates. data set is small and subject to a number of limitations, However, the association between latitude and diver- our results suggest that diversi®cation rates may be si®cation rate is independent of clade area (Table 2), in¯uenced by both area and climate. The tendency for and largely independent of NPP, although when inde- clades occupying larger areas to have higher diversi- pendent contrasts were used, the addition of NPP re- ®cation rates is consistent with previous ®ndings for duced the slope of the latitudinal effect considerably bird families (Owens et al. 1999, Bennett and Owens (Table 2). Clearly, there is a component of latitudinal

TABLE 3. Effects of clade age on the latitudinal bias in diversi®cation rate, for young and old clades, using cross-taxon tests (n ϭ 18 and 19 clades, respectively) and independent-contrasts tests (n ϭ 17 and 18 contrasts, respectively).

Cross-taxon results Independent-contrasts results Young clades Old clades Young clades Old clades Predictor Coef®cient t Coef®cient t Coef®cient t Coef®cient t Intercept 0.41 3.99** 0.18 4.41*** Latitude Ϫ0.1 Ϫ2.31* Ϫ0.02 Ϫ1.05 Ϫ0.09 Ϫ2.52* 0.003 0.12 * P Յ 0.05; ** P Յ 0.01; *** P Յ 0.001. Special Feature hnlre hlgne n egahcdt esbe- sets data available. geographic come and phylogenies investigation larger for when matter another be older should among per This stronger clades). be richness should species clades among in area unit variation latitudinal applies rate that diversi®cation (i.e., for op- that the to Hence, prediction deter- gradient. posite that diversity latitudinal rates, species the diversi®cation mine of than is levels rather it equilibrium clades, richness, in saturated older, variation among latitudinal that own implies there- model its fore The has not. have have each some some and that saturation that reached and present, richness, species are of ages level clades saturation of these that range unites simply a 1 assuming of Fig. by in viewpoints than model opposing rather the clades entities, on focusing clade geographic By or 1992). rate (Rohde from diversi®cation ages arise must in it differences latter, the latitudinal if equilib- level; geographic (saturation) the rium arise in must differences gradient latitudinal the regional former, from nonequilibrium the If or richness. species equilibrium which either at sume ages) of range saturation. reach (or to age tend provid- the clades perhaps into age, clade insight with ing slope in function decline the phy- estimate the to of dated possible be robustly even and may it large logeny, suf®ciently In result. a similar with a fact, and produce unimportant, should relatively division is any clades almost old de®nition and precise young decline the of Hence, should age. any clade rate with of diversi®cation smoothly strength in the bias that predicts latitudinal it and Rather, young of clades. the division old discrete arbitrary, a was predict division ``old'' not does our and model ``young'' Although into 1. clades from Fig. of derived in prediction presented the model the with consistent younger is in steeper clades was rate diversi®cation and latitude and (Dynesius periods time 2000). longer Jansson either over variation, temperature or ele- of annually be degree high may a rates by vated extinction not example, extinction have For we or features examined. environmental speciation by is driven in are possibility rates variation Another clade. latitudinal each that a for to estimate from conditions set resulting point data climatic error statistical our heterogeneous 2 because Type summarizing simply high is to prone this is that possible examined. have accounted is we not It measures environmental is the that by rate for diversi®cation in variation 2286 en fdvriy okt aehsbe oiae by dominated been has date to Work diversity. pat- of geographic terns other and gradient the diversity studying latitudinal in approaches, approaches, there clade-based geographic complement First, on to emphasis presented. more have be should we analysis illustrative yohssfrtelttdnldvriygain as- gradient diversity latitudinal the for Hypotheses between association the of slope the that ®nding Our hr r w eta esgsta rs rmthe from arise that messages central two are There C ONCLUSION ACLCRIL TAL. ET CARDILLO MARCEL eetevrnetlfcos sats etsie to suited best task a databases. is large dif- factors, by environmental for de- accounted ferent further components into then this latitude, composing diver- by in explained variation rates of si®cation component innovations. the key detecting of evolution Hence, the or events such dispersal factors as unpredictable as well as factors, are and are biological environmental study numerous rates by of determined diversi®cation be type to clade this likely in since used noisy, sets inherently geographic data and The is phylogenetic sets. there large Second, data for analysis. need of urgent units an rather the clades as use the regions so present, than doing for At methods demonstrated. richness, developed explicitly best species be in must gradient and the be underlie cannot to rates assumed diversi®cation shown, in have variation we address As latitudinal can questions. important recognized, and are different limitations analyses, the phylogenetic provided whereas analyses, geographic ure .J,adV aun 97 ag-cl biogeograph- Large-scale 1987. Paquin. V. and J., D. Currie, rma,L,adM adlo 03 etn h ikbe- link the Testing 2003. Cardillo. M. and L., Bromham, ecol- Evolutionary 2002. Owens. F. P. I. and M., P. Bennett, Sister- 1998. Harvey. H. P. and Nee, S. G., T. Barraclough, ua,M . .S oln,adS .Cle.20.Latitu- 2002. Culver. J. S. and Collins, Biogeography. S. L. 1998. A., Lomolino. M. V. Buzas, M. and H., J. Brown, owl,R . n .C ut.19.Nnilgclgra- Nonbiological 1994. Hurtt. C. G. and K., di- R. latitudinal Colwell, of basis history life The 2002. M. in Cardillo, diversi®cation of rates and Latitude 1999. M. Cardillo, rcat .18.Booia ies®ainadiscauses. its and diversi®cation Biological 1985. J. Cracraft, ure .J 91 nryadlresaepten fanimal of patterns large-scale and Energy 1991. through J. gradients D. Currie, diversity Taxonomic 2001. A. J. Crame, gradi- diversity taxonomic of Evolution 2000. A. J. Crame, oRbr igl o loigu ouehsdt nthe on data (U.K.). his Council by use Research supported was Environment to work Natural The us the birds. World allowing New for of distribution Ridgely and Robert suggestions, and to comments helpful many for Diniz-Filho 327. clpten fseisrcns ftes Nature trees. of richness species of patterns ical 49. fmlclreouin ora fEouinr Biology Evolutionary of Journal rates 16 and evolution. richness molecular species in of gradient latitudinal the tween UK. extinction. Oxford, and systems Press, mating University histories, Oxford life birds: in ogy diversi®cation± Ecology of Evolutionary correlates comment. identifying in analysis group aeo nrae rceig fteNtoa cdm of Academy National the tropical USA of Science, higher Proceedings by increase. caused of biodiversity rate in difference dinal USA. Massachusetts, Sunderland, Associates, Sinauer insi pce ihesadasuiu aooteffect. Rapoport spurious a Naturalist and American richness species in dients Ecology Animal of Journal poles the equator? from the vary traits to species do How gradients: versity of Society Sciences Royal B±Biological the Series of London Proceedings butter¯ies. and birds naso h isuiBtnclGardens Botanical Missouri the of Annals n ln pce ihes mrcnNaturalist American richness. species plant and Distributions and Diversity time. geological composition the Paleobiology from faunas. evidence bivalve realm: recent marine of the in ents hnst nyPri,Bafr akn,adAlexandre and Hawkins, Bradford Purvis, Andy to Thanks :200±207. 99 :7841±7843. A L CKNOWLEDGMENTS ITERATURE 144 :570±595. C 12 ITED :751±754. clg,Vl 6 o 9 No. 86, Vol. Ecology, 266 26 :188±214. :1221±1225. 72 71 7 :794±822. :175±189. :79±87. 329 137 :326± :27± September 2005 LATITUDINAL GRADIENTS 2287

Davies, T. J., T. G. Barraclough, V. Savolainen, and M. W. Macpherson, E. 2002. Large-scale species-richness gradients Chase. 2004a. Environmental causes for plant biodiversity in the Atlantic Ocean. Proceedings of the Royal Society of gradients. Philosophical Transactions of the Royal Society London Series B±Biological Sciences 269:1715±1720. of London Series B±Biological Sciences 359:1645±1656. O'Hara, R. J. 1991. A book review of ``Phylogeny and clas- Davies, T. J., V. Savolainen, M. W. Chase, J. Moat, and T. si®cation of birds: a study in molecular evolution''. The G. Barraclough. 2004b. Environmental energy and evo- Auk 108:990±994. lutionary rates in ¯owering plants. Proceedings of the Roy- Owens, I. P.F., P.M. Bennett, and P.H. Harvey. 1999. Species al Society of London Series B±Biological Sciences 271: richness among birds: body size, life history, sexual selec- 2195±2200. tion or ecology? Proceedings of the Royal Society of Lon- Durazzi, J. T., and F. G. Stehli. 1972. Average generic age, don Series B±Biological Sciences 266:933±939. the planetary temperature gradient, and pole location. Sys- Peterson, A. T., J. Soberon, and V. Sanchez-Cordero. 1999. tematic Zoology 21:384±389. Conservatism of ecological niches in evolutionary time. Dynesius, M., and R. Jansson. 2000. Evolutionary conse- Science 285:1265±1267. quences of changes in species' geographical distributions Price, T. D., A. J. Helbig, and A. D. Richman. 1997. Evo- driven by Milankovitch climate oscillations. Proceedings lution of breeding distributions in the Old World leaf war- of the National Academy of Sciences, USA 97:9115±9120. blers (genus Phylloscopus). Evolution 51:552±561. ESRI. 2002. ArcMap. Version 8.2. Environmental Systems Purvis, A. 1996. Using interspecies phylogenies to test mac- Research Institute (ESRI). Redlands, California, USA. roevolutionary hypotheses. Pages 153±168 in P.H. Harvey, Farrell, B. D., and C. Mitter. 1993. Phylogenetic determinants A. J. Leigh Brown, J. Maynard-Smith, and S. Nee, editors. of insect/plant community diversity. Pages 253±266 in R. New uses for new phylogenies. Oxford University Press, E. Ricklefs and D. Schluter, editors. Species diversity in Oxford, UK. ecological communities. University of Chicago Press, Chi- Ridgley, R. S., T. F. Allnutt, T. Brooks, D. K. McNicol, D. cago, Illinois, USA. W. Mehlman, B. E. Young, and J. R. Zook. 2003. Digital Flessa, K. W., and D. Jablonski. 1996. The geography of distribution maps of the birds of the Western Hemisphere. evolutionary turnover: a global analysis of extant bivalves. Version 1.0. NatureServe, Arlington, Virginia, USA. Pages 376±397 in D. Jablonski, D. H. Erwin, and J. H. Rohde, K. 1992. Latitudinal gradients in species diversity: Lipps, editors. Evolutionary paleobiology. University of the search for the primary cause. Oikos 65:514±527. Chicago Press, Chicago, Illinois, USA. Rohde, K. 1997. The larger area of the tropics does not ex- Francis, A. P., and D. J. Currie. 2003. A globally consistent plain latitudinal gradients in species diversity. Oikos 79:

richness±climate relationship for angiosperms. American 169±172. Feature Special Naturalist 161:523±536. Rohde, K. 1999. Latitudinal gradients in species diversity Garland, T., P. H. Harvey, and A. R. Ives. 1992. Procedures and Rapoport's rule revisited: a review of recent work and for the analysis of comparative data using phylogenetically what can parasites teach us about the causes of the gra- independent contrasts. Systematic Biology 41:18±32. dients? Ecography 22:593±613. Gaston, K. J. 1998. Species±range size distributions: prod- Rosenzweig, M. L. 1992. Species diversity gradients: we ucts of speciation, extinction and transformation. Philo- know more and less than we thought. Journal of Mam- sophical Transactions of the Royal Society of London Se- malogy 73:715±730. ries B±Biological Sciences 353:219±230. Rosenzweig, M. L. 1995. Species diversity in space and time. Gaston, K. J. 2000. Global patterns in biodiversity. Nature Cambridge University Press, Cambridge, UK. 405:220±227. Sibley, C. G., and J. E. Ahlquist. 1990. Phylogeny and clas- Guernier, V., M. E. Hochberg, and J.-F. Guegan. 2004. Ecol- si®cation of birds. Yale University Press, New Haven, Con- ogy drives the worldwide distribution of human diseases. necticut, USA. PLoS Biology 2:740±746. Sibley, C. G., and B. L. Monroe. 1990. Distribution and H-Acevedo, D., and D. J. Currie. 2003. Does climate deter- taxonomy of birds of the World. Yale University Press, New Haven, Connecticut, USA. mine broad-scale patterns of species richness? A test of the Stehli, F. G., R. G. Douglas, and N. G. Newell. 1969. Gen- causal link by natural experiment. Global Ecology and Bio- eration and maintenance of gradients in taxonomic diver- geography 12:461±473. sity. Science 164:947±949. Hawkins, B. A., and J. A. F. Diniz-Filho. 2004. `Latitude' Stevens, G. C. 1989. The latitudinal gradient in geographic and geographic patterns. Ecography 27:268±272. range: how so many species coexist in the tropics. Amer- Hawkins, B. A., R. Field, H. V. Cornell, D. J. Currie, J.-F. ican Naturalist 133:240±256. Guegan, D. M. Kaufman, J. T. Kerr, G. G. Mittelbach, T. UNEP (United Nations Environment Program). 2003. GRID Oberdorff, E. M. O'Brien, E. E. Porter, and J. R. G. Turner. datasets. ͗http://www.grid.unep.ch/data͘ 2003a. Energy, water, and broad-scale geographic patterns Wiens, J. J. 2004. Speciation and ecology revisited: phylo- of species richness. Ecology 84:3105±3117. genetic niche conservatism and the origin of species. Evo- Hawkins, B. A., and E. E. Porter. 2001. Area and the lati- lution 58:193±197. tudinal diversity gradient for terrestrial birds. Ecology Let- Willig, M. R., D. M. Kaufman, and R. D. Stevens. 2003. ters 4:595±601. Latitudinal gradients of biodiversity: pattern, process, Hawkins, B. A., and E. E. Porter. 2003. Relative in¯uences scale, and synthesis. Annual Review of Ecology and Sys- of current and historical factors on mammal and bird di- tematics 34:273±309. versity patterns in deglaciated North America. Global Ecol- Woodward, F. I., T. M. Smith, and W. R. Emanuel. 1995. A ogy and Biogeography 12:475±481. global land primary productivity and phytogeography mod- Hawkins, B. A., E. E. Porter, and J. A. F. Diniz-Filho. 2003b. el. Global Biogeochemical Cycles 9:471±490. Productivity and history as predictors of the latitudinal di- Wright, D. H. 1983. Species±energy theory: an extension of versity gradient of terrestrial birds. Ecology 84:1608±1623. species±area theory. Oikos 41:496±506. Hecht, A. D., and B. Agan. 1972. Diversity and age rela- Wright, D. H., D. J. Currie, and B. A. Maurer. 1993. Energy tionships in Recent and Miocene bivalves. Systematic Zo- supply and patterns of species richness on local and re- ology 21:308±312. gional scales. Pages 66±72 in R. E. Ricklefs and D. Schlu- Jablonski, D. 1993. The tropics as a source of evolutionary ter, editors. Species diversity in ecological communities. novelty through geological time. Nature 364:142±144. University of Chicago Press, Chicago, Illinois, USA. Ecology, 86(9), 2005, pp. 2288±2297 ᭧ 2005 by the Ecological Society of America

LARGE-SCALE BIOGEOGRAPHIC PATTERNS IN MARINE MOLLUSKS: A CONFLUENCE OF HISTORY AND PRODUCTIVITY?

MICHAEL A. REX,1,3 J. ALISTAIR CRAME,2 CAROL T. S TUART,1 AND ANDREW CLARKE2 1Department of Biology, University of Massachusetts, 100 Morrissey Boulevard, Boston, Massachusetts 02125 USA 2British Antarctic Survey, High Cross, Madingley Road, Cambridge CB3 OET UK

Abstract. Large-scale biogeographic patterns in marine systems are considerably less well documented and understood than those in terrestrial systems. Here, we synthesize recent evidence on latitudinal and bathymetric gradients of species diversity in benthic mollusks, one of the most diverse and intensively studied marine taxa. Latitudinal gradients in coastal faunas show poleward declines in diversity, but the patterns are highly asym- metrical between hemispheres, and irregular both within and among regions. The extensive fossil record of mollusks reveals that latitudinal gradients have become steeper during the Neogene, partly because of a rapid diversi®cation in tropical coral reefs and their associated biotas. Much of the inter-regional variation in contemporary latitudinal trends depends on the longitudinal distribution of reefs and major Neogene vicariant events. Thus, coastal faunas reveal a strong evolutionary±historical legacy. Bathymetric and latitudinal gradients in the deep ocean suggest that molluscan diversity is a function of the rate of nutrient input from surface production. Diversity may be depressed at abyssal depths because of extremely low rates of organic carbon ¯ux, and at upper bathyal depths and high latitudes by pulsed nutrient loading. While the deep-sea environment is not conducive to fossilization, rela- tionships between local and regional diversity, and the distribution and age of higher taxa indicate an evolutionary signal in present-day diversity gradients. Marine invertebrate com- munities offer tremendous potential to determine the relative importance of history and ecological opportunity in shaping large-scale patterns of species diversity. Key words: bathymetric gradients; history; latitudinal gradients; marine mollusks; productivity; species diversity.

INTRODUCTION to help reveal the potential causes of variation in di- versity at large horizontal scales. We stress from the Special Feature Making sense of large-scale marine biogeography outset that latitude and depth per se are not environ- remains a dif®cult challenge. The oceans represent a mental variables; they are merely geographic gradients huge and complex ecosystem that supports Earth's rich- that de®ne macroecological patterns (Hawkins and est diversity of animal phyla. Much of the marine realm Diniz-Filho 2004). However, as we will see, large-scale has been inaccessible to exploration until quite re- marine biogeography is still very much in a descriptive cently, particularly polar regions and the deep sea. Yet phase. Establishing pattern is a primary objective. In the very distinctiveness of the marine fauna and en- the case studies that we present, we do not consider vironment, when compared to terrestrial systems, holds species richness to be a response variable, or depth and the promise of discovering the common underlying latitude to be explanatory variables. Rather we docu- causes of global biogeographic patterns. ment geographic gradients of diversity in a variety of In this paper, we take a fresh look at two major environmental circumstances, and attempt to identify biogeographic trends in the oceans: latitudinal and factors that begin to provide consistent explanations. bathymetric gradients of species diversity. We focus on Patterns of species diversity in time and space pre- gastropod and bivalve mollusks, important faunal com- sent a multivariate problem. Of the welter of hypoth- ponents of virtually all major marine habitats. We ex- eses proposed to explain them, recent attention has cen- amine past and present latitudinal gradients of species tered on energy and history as potential unifying diversity in coastal assemblages, and contemporary lat- themes (Ricklefs and Schluter 1993, Hawkins et al. itudinal gradients in the deep sea. Coastal and deep- 2003). It is mainly in this context that we explore ma- sea systems have fundamentally different ecologies. rine molluscan diversity. Whittaker and Field (2000) The two worlds are linked by depth gradients of di- point out that the energy±diversity theory makes spe- versity in which pattern and scale become compressed ci®c predictions about patterns of diversity, but that history cannot; the evolution and subsequent geograph- Manuscript received 30 June 2004; revised 9 December 2004; accepted 10 December 2004. Corresponding Editor: B. A. Haw- ic spread of taxa are inherently stochastic processes kins. For reprints of this Special Feature, see footnote 1, p. 2261. that can only be inferred a posteriori. They propose 3 E-mail: [email protected] reconciling the two explanations by testing contem- 2288 September 2005 LATITUDINAL GRADIENTS 2289 porary trends against expectations of the energy±di- versity theory, and then determining if residual vari- ation can be attributed to historical contingency. Of course, given suf®cient information about history, the opposite approach seems equally plausible. Historical development may be modulated by present-day ecol- ogy. One disadvantage of working with terrestrial or- ganisms is their relatively poor fossil record, especially for plants that have ®gured so importantly in energy± diversity studies. By contrast, marine invertebrates have an impressive fossil record that includes proxy variables for climate reconstruction. The theory that energy limits diversity has two basic versions (Hawkins et al. 2003). One is the reductionist FIG. 1. Latitudinal variation in species richness of gas- view that ambient energy in the form of heat from solar tropod molluscs from the Paci®c continental shelf of North radiation regulates the biochemical kinetics of metab- and South America and Antarctica. Data are from Roy et al. olism, and consequently organism abundance and evo- (1994, 1998) and Valdovinos et al. (2003), with Antarctic data from British Antarctic Survey records. Data are pooled lutionary rates (e.g., Allen et al. 2002, Vermeij 2005). by bins of 5Њ. Negative latitudes are southern. This ®gure was This theory predicts that diversity should be a positive modi®ed with permission from the journal Ecography. function of temperature. A second version is the pro- ductivity hypothesis, which is more concerned with the rate of energy ®xation for food and its effect on pop- arctic records from the British Antarctic Survey's ulation growth (Huston 1994, Rosenzweig 1995). The SOMBASE database. This is the most well-docu- relationship between diversity and productivity is ex- mented and geographically controlled latitudinal pat- pected to be either unimodal or positive and linear (Mit- tern known for marine benthic organisms. Apart from Feature Special telbach et al. 2001). Presumably, water availability, an the fact that diversity is markedly higher within tropical integral part of the energy±diversity theory as it per- and subtropical latitudes than at higher latitudes, there tains to terrestrial biotas, is not relevant in the oceans. are a number of unexpected features. The pattern is A comparison of molluscan diversity patterns suggests asymmetrical with peak diversity occurring at northern strong roles for both history and productivity in the subtropical latitudes rather than near the equator. One development of large-scale marine biogeographic pat- of the most striking discoveries to emerge from recent terns. studies is the high variability of diversity within the tropics (Flessa and Jablonski 1995, Crame 2000a, b). LATITUDINAL GRADIENTS OF DIVERSITY Tropical latitudes contain among the highest and lowest IN THE SHALLOW SEAS values of diversity shown in Fig. 1. In parts of the vast Indo-Paci®c province there is no obvious latitudinal Contemporary spatial patterns gradient between the equator and approximately 30Њ N Studies of both bivalves and gastropods have been and 30Њ S, and this is particularly so in the western fundamental in developing our concept of taxonomic Indian Ocean for both benthic mollusks and hermatypic diversity gradients in the shallow marine realm. Ini- corals (Sheppard 1998, Mackie et al. 2005). tially, the patterns they revealed seemed to be simple Much of the variation in tropical shallow diversity, and regular (Fischer 1960, Stehli et al. 1967, Schopf and hence the overall shape of latitudinal gradients, is 1970), but as the scale of investigations expanded, and controlled by the distribution of coral reefs. Reefs are more comprehensive databases were assembled, it be- the sites of highest diversity in a wide range of marine came apparent that regional gradients were much more invertebrate, vertebrate, and plant taxa (Briggs 1996, complicated than originally envisaged (Roy et al. 1994, Ellison et al. 1999, Bellwood and Hughes 2001, Rob- 1998, Flessa and Jablonski 1995, Crame 2000a, b, Val- erts et al. 2000), and wherever they are absent or only dovinos et al. 2003). In particular, the decline in tax- poorly developed tropical diversity is much reduced. onomic diversity values from low to high latitudes is The northern and southern limits of hermatypic coral much less regular than is often portrayed and there are growth are sharply de®ned (Veron 1995) and have in signi®cant inter-regional differences in gradients on a turn served to delimit the latitudinal ranges of many variety of spatial scales. reef-associated bivalves, gastropods, and other taxa. It Fig. 1 shows the contemporary latitudinal gradient is likely that these very sensitive outer boundaries to of species richness in gastropods of the East Paci®c reef growth were accentuated by intense Late Neogene along the continental shelves of North, Central, and (i.e., the last 10±15 ϫ 106 yr) glacioeustatic sea-level South America. It combines data for the northern hemi- cycles that left the reef tract repeatedly and abruptly sphere from Roy et al. (1994, 1998), the southern hemi- stranded (Rosen 1981, Paulay 1990). This is particu- sphere from Valdovinos et al. (2003), and new Ant- larly so in regions such as the western Atlantic and Special Feature ossadlrervroto ntoia atr coasts eastern 2000 tropical (Crame on the out¯ow river to western large tropical and part, on coasts currents in African boundary least cold and at of presence American attributed, been South has the continents around reefs in reduction coral marked The 1995). (Veron Paci®c western 2290 advnse l 20)nt htlwdvriybetween diversity 10 low that note (2003) unclear. al. remains et Valdovinos scales spatial on large productivity at diversity of coastal effect been the seems evidence not Similarly, current have contradictory. the basis, diversity global tem- a marine of on effects on evaluated potential area the and while perature (Valdovinos Thus, occurs 2003). al. opposite south- et the the In just 1998). al. hemisphere, et ern (Roy shelf continental correlated the of area positively with uncorrelated but is temperature sea-surface with diversity molluscan sphere, strongest hemisphere. the northern in the seen in that gradients to comparable gradient this sys- marine deep and shallow (2004 Crame the Hillebrand tems, both 1996, from set Gaston data (e.g., 2000 hemispheres and northern southern the in gradients latitudinal between ilarity notehgetlttd ein Rye l 98 Crame 1998, al. 2000 et (Roy regions latitude highest the into nanme fohrtxnmcgop Cak and (Clarke groups taxonomic 2000 other Crame 2000, of Lidgard mirrored number is pattern a This 1). in (Fig. tropics the of edge ern rdeti iav iest nteesencatof coast eastern the 2000 (Crame on America diversity South bivalve in gradient aeHr,adrmi ihi h naci Peninsula Antarctic (55±65 at the in region species high 300 remain and reaching Horn, south, Cape the to sharply increase os iest srltvl o n osatat constant and gastro- low Paci®c relatively East is signi®cantly in diversity be example, pods, For may form. they in there that different studied, indications intensively strong less are much gra- southern been the have though even dients this for respects result, surprising many a is In the of hemispheres. slope between or strength gradient the either in difference ni®cant r eipeegain eoe uhsalwrat shallower much becomes 35±40 gradient north- The hemisphere 2003). al. ern et Valdovinos gra- 1981, temperature (Rosen latitudinal dient accompanying the than er utai n e eln oAtrtc (Crame Antarctica to Zealand New 2000 and that Australia those South are through Territory/Queensland hemisphere Northern de- southern from gradients the run molluscan in far strong so only to tected The appear ¯at. Antarctica very East to be or Africa, West incom- South either through is from Africa, diversity gradients regional bivalve but tropical knowl- plete, Our African counterparts. of northern edge its of any matches pce rm10 from species Њ ln h atPc® os ftenrhr hemi- northern the of coast Paci®c East the Along sim- the about debate considerable been has There atoo iest al rcptul ttenorth- the at precipitously falls diversity Gastropod n 40 and b b b .Teei vnahn fasapi¯cinin in¯ection sharp a of hint a even is There ). large a on based meta-analysis weighted a In ). 1). Fig. , Њ hr tfrsadsic ec-iefeature bench-like distinct a forms it where N Њ Fg )cicdswt n ftemost the of one with coincides 1) (Fig. S b Њ i.1). Fig. ) hr r niain fasteeper a of indications are There S). Њ o40 to ,b a, Њ Fg ) hratrvalues Thereafter 1). (Fig. S b a nbet eetasig- a detect to unable was ) n ssgicnl steep- signi®cantly is and ) b ,btti tl nn way no in still this but ), IHE .RXE AL. ET REX A. MICHAEL ϳ 100 ftesuhatr aiccata ue continental outer at coast depths. Paci®c shelf regions southeastern some benthic in the occurs which of as in reduced, greatly zone is minimum diversity oxygen can an productivity produce high Exceptionally marine interactions. biolog- and in ical growth diversity population accelerating depress by can systems suggest loading we nutrient Later, productivity. that seasonal and sup- also high regions port Polar ocean. the in regions productive ae ag rmatrefl oa to global esti- three-fold in Era; a rise from Cenozoic range marked the mates the through is diversity years taxonomic recent in emerge we can so, or if and, form when? gradient the pinpoint latitudinal in the shifts of signi®cant strength any time. of been effects there the consider Have to need also undoubtedly we are important, processes Lom- such and Although 1998). Brown olino species±energy 1995, of Rosenzweig form (e.g., some hypothesis or tropics, com- the habitat or of area the plexity attribute greater of to the as feature possible such be time-invariant surface, Earth's may some it to then generation case, their the is 2001). this Crame If in (summarized features time persistent geological been through have gradients diversity dinal eeomn nteCnzi curdi h al to Early the ( in reef Miocene occurred Middle coral Cenozoic of the phase in main development The 1987). (Vermeij ternary ntels 23 concentrated last was not the it but and in largely nature, was in it tropical emerge: exclusively to beginning are cation Fes n alnk 96 rm 2000 Crame 1996, bivalves Jablonski 1998), and Rosen (Flessa and scleractinian Wilson 1995, of (Veron diversi®cation corals pantropical of period a ies®aini h niePaeooc(Valentine 2003 Phanerozoic al. et of entire Jablonski phases 2001, the spectacular most in alone, the biases diversi®cation of other one or represents cannot preservational and et it to that (Alroy recognized attributed unchallenged generally be now gone is not it has 2001), al. trend a existence al. the such et While of (Valentine 2001). Johnson time and this Jackson 1978, during diversity species in ic(CP ihdvriyfc CaeadRosen and (Crame foci high-diversity Indo- Pa- (ACEP) discrete Atlantic±Caribbean±East the ci®c and of (IWP) formation Paci®c that the West events to vicariant of eventually series led a tropical by homogeneous disrupted was once biota the Neogene, the through sub- and tropical of diversity. tropical development rapid through Neo- gene the marine during pronounced the more in became words, diversity environment species other other of of In gradients range 1987). latitudinal a (Vermeij and taxa 1990), Kohn reef-associated 1980, al. et (Taylor oeeiec ugssta osa aielatitu- marine coastal that suggests evidence Some n ftems tiigplotlgclted to trends paleontological striking most the of One w atclrfaue fteCnzi diversi®- Cenozoic the of features particular Two stelttdnltmeauegain steepened gradient temperature latitudinal the As ϫ itrcldevelopment Historical 10 ϳ 6 23±17 rdrn h egn n Qua- and Neogene the during yr ϫ b ). 10 6 clg,Vl 6 o 9 No. 86, Vol. Ecology, er g) hswas This ago). years Ͼ 0fl increase 10-fold a ,gastropods ), September 2005 LATITUDINAL GRADIENTS 2291

2002). The split between these two realms can be traced back to no earlier than the collision between Africa/ Arabia and Eurasia 20 ϫ 106 years ago. Prior to that, in the Paleogene (ϳ65±30 ϫ 106 years ago), there is evidence of coral, mollusk, and mangrove taxa that had essentially pantropical distributions (Vermeij 1987, Wilson and Rosen 1998, Ellison et al. 1999). In the Eocene±Early Oligocene (ϳ50±30 ϫ 106 years ago) coral reef diversity was centered in the Mediterranean and Caribbean, and IWP faunas were comparatively sparse. However, in the Early Miocene there was a four- fold increase in IWP coral genera, and reefs prolifer- ated in the shallow-water habitats created by the col- lision of Australia/New Guinea with Southeast Asia (Crame and Rosen 2002). Although the tropical marine fauna is of great an- tiquity, it is apparent that the distinctive IWP and ACEP FIG. 2. The bathymetric pattern of species richness of high-diversity foci of the present day may be no more deep-sea neogastropods in the eastern North Atlantic. Di- versity is estimated as the number of coexisting species ranges ϫ 6 than 15±20 10 years old. Tropical biotas retracted in 500 m depth increments. Data are from Bouchet and WareÂn signi®cantly in areal extent through the early part of (1980, 1985). the Cenozoic Era, but from approximately 20 ϫ 106 years ago the two high-diversity faunas developed on discrete clusters of continental and oceanic islands (Chase et al. 1998, Rex and Etter 1998). They are best (Crame and Rosen 2002). As the steepest latitudinal documented in the North Atlantic Ocean where benthic Feature Special gradients of the present day are all associated with either the IWP or ACEP foci, it can be concluded that communities have been sampled extensively from the they are no more than 15±20 ϫ 106 years in age. Some continental shelves to abyssal depths. Diversity appears ϳ may be substantially younger (Vermeij 1987, Crame to increase from the shelf±slope transition ( 200 m) and Rosen 2002). Clearly, the development of latitu- to a peak at upper- to mid-bathyal depths on the con- dinal gradients in coastal marine mollusks and their tinental margin, and then decrease toward the abyssal highly variable expression between hemispheres and plain (Sibuet 1977, Rex 1981, Etter and Grassle 1992, oceans can only be fully understood in an historical Paterson and Lambshead 1995, Cosson-Sarradin et al. context. 1998). The exact shape of the curves varies among taxa and ocean basins (Stuart et al. 2003). BATHYMETRIC GRADIENTS OF DIVERSITY Our knowledge of these unimodal diversity±depth Bathymetric gradients of community structure on the patterns is based almost entirely on estimates of alpha ␣ sea ¯oor are counterparts to elevation gradients in ter- ( ) diversity along sampling transects. Here we present restrial systems. Both exhibit strong environmental an alternative way of illustrating them that is similar changes over relatively short distances (10s to100s of to the use of long-term biotic surveys to detect bio- km). Species diversity along vertical gradients is easier geographic trends in terrestrial and coastal systems. We to measure and interpret than global-scale latitudinal compiled data on ranges of neogastropods, the largest trends that are often interrupted by major geographic taxon of deep-sea snails, from two comprehensive sys- barriers and confounded by complicated inter-regional tematic monographs that cover the eastern North At- differences in basic ecology. Because of the smaller lantic fauna (Bouchet and WareÂn 1980, 1985). The data spatial scales involved, and consequently greater dis- represent 189 species in over 1300 samples collected persal potential, vertical gradients of diversity are apt through a century of deep-sea dredging in a region east to be caused more by contemporary ecological forces of the Mid-Atlantic Ridge from the Azores to the Nor- than by the evolutionary±historical processes of di- wegian Sea. To assess the diversity±depth pattern, we versi®cation and subsequent geographic spread of taxa; summed the number of coexisting species within 500 although as we point out later, it is regional-scale ra- m depth increments. The results indicate that diversity diations that create the species pools from which ver- has a unimodal relationship with depth on regional as tical diversity gradients are formed. well as local scales (Fig. 2). The curve is asymmetrical Diversity±depth gradients in the deep-sea benthos with a maximum at upper bathyal depths (1000±1500 coincide with a wide range of biotic and abiotic en- m), and is skewed to the right with a long tapering vironmental gradients (reviewed in Etter and Mulli- decline to very low levels in the abyss (Ͼ4000 m). A neaux 2001, Levin et al. 2001, Stuart et al. 2003), and conservative interpretation might be that diversity sim- are attended by rapid faunal turnover (Gage and Tyler ply decreases with depth; but depressed diversity at 1991) and clinal effects within individual species upper continental slope depths, while less pronounced, Special Feature epsadcessb ± reso antd from magnitude individuals/m of orders of the 2±3 in 000s 10 by stock Standing decreases 1997). sea al. to deep et ¯ux carbon (Smith organic benthos of 1983). the rate average indicator (Rowe the accurate for most Ocean available the probably World is is that stock the Standing depth with throughout stock found standing benthic exponential in distance the for decrease with accounts This decreases depth. and benthos shore from the of rate to average input partially the nutrient column, is water sinks phytodetritus the sinking in and remineralized and coasts, production high- near is production est surface Because column. water as the through ecosystem. deep-sea originates the in Food no production is primary there situ communities, in important. chemosynthetic be factors to from seem other Apart does that Productivity or involved. in temperature are drop to is diversity sensitive the either acutely while that implying dramatic, minute, the is is temperature thermocline, diversity permanent in the bathyal decrease Below upper 2). the (Fig. across zone 1968) deep (Sanders the decreasing in temperature and depth way with consistent increases a diversity di- where in sea affects out borne temperature not et that is Rex versity prediction 2004, al. The et 2005). avail- McClain al. energy 2003, has al. of attention et importance (Stuart recent potential ability much the different space, on on and focused operate time that in¯u- of surely factors scales is numerous diversity by deep-sea enced While 2001). Levin al. 1981, (Rex et trends diversity±depth for account to causes. their that about curves clue important diversity±depth an of provide feature may persistent a is 2292 pce r osas htte a ereproductively be may they that sparse many so of are densities population species 1973); (Rex effect Allee abys- and depths. bathyal upper sal at depressed is most diversity explaining why requires is What found depths. environment intermediate benign at physically relatively mollusks, and the in deep-sea productive accumulated has of community rich ability a dispersal that spe- many, large and not the pool given and are surprising, cies species, not `Why is of but It more?''' number species?' many particular many so a there there are becomes `Why of question not the ``Thus number diversity. per- high any permit of can sistence that processes out nonequilibrial or point equilibrial Tilman (1993:19) uncertain. Pacala remains produc- and diversity which in in¯uences through universal) mechanism tivity not actual (though the common nature, be to appear ships North deep uni- the a in has productivity Atlantic. diversity carbon to species organic relationship then for modal benthos, proxy the a to as ¯ux accepted be can stock aha etst 0±0so individuals/m of 10s±100s to depths bathyal g/m ayeooia xlntoshv enproposed been have explanations ecological Many o iest nteayshsbe trbtdt an to attributed been has abyss the in diversity Low relation- diversity±productivity unimodal Although 2 nteays(aeadTlr19) fstanding If 1991). Tyler and (Gage abyss the in 2 n 0 fg/m of 10s and IHE .RXE AL. ET REX A. MICHAEL 2 2 tupper at and Ͻ 1 htspot h oetdvriyi yfrtelargest the far by 1981). is (Rex diversity region plain lowest abyssal the supports the that indeed, is diversity; basins within with major sea uncorrelated of deep of extent the effects in areal the features The modify physiographic versa! may vice these of or All productivity, 2001). al. et Lev- in in (reviewed disturbance or constraints, boundary dispersal, environmental and as dynamics such metapopulation heterogeneity, causes the potential deny other not of does action benthos deep-sea the in productivity marine in values extreme systems. most the productivity productivity± of including range spanned, the large the exploring is relationship for diversity diversity high of at bathymetric patterns productivity of advantage One of systems. marine variance be- in relationship levels and positive mean a the from tween result may curves diversity±productiv- ity unimodal that reasons, suggested, deep-sea similar (1999) for Gaston in and diversity Chown habitats. depressed benthic with and associated abundance seems high always later, loading out no point nutrient is we as pulsed There although, diet. this; in for evidence specialization preda- direct by of diversify ability to the tors limit might productivity because variable and de- displacement, competitive is popula- and growth accelerates diversity tion productivity Rex bathyal high because upper blooms. pressed that close phytoplankton proposed the seasonal (1983) of because to variable proximity presum- temporally and also more but depths, (Rosenzweig only ably bathyal not upper for is at input higher nutrient account much of rate to The 1993). dif®cult productivity Abramsky of more levels high is very at diversity in The and decline energy. available productivity, more with of increases levels diversity that low extremely low explains at phenomenon fauna of diversity kind molluscan this ecol- Most that abyssal sources. agree bathyal ogists the from effect of mass a as much exists and that bathyal and sys- that source±sink tem, suggests a as This function may 2005). populations al. abyssal abil- et dispersal high (Rex with species ity extensions bathyal range of subset deeper a represent for Atlantic in North mollusks abyssal the Most dependence. from extinction density local inverse chronic suffer and unsustainable tatcadNreinSa(e ta.19,2000, 1993, al. et (Rex Sea Norwegian North and the in Atlantic depths bathyal gra- at diversity latitudinal Gas- species of of 1998). dients evidence interbasin show Wilson bivalves large 1996, and at Sanders tropods and makeup (Allen species scales level and the both diversity in of differences that shows clear fauna now deep-sea is thick the it However, the column. by water variation overlying climatic and surface throughout from uniform be insulated to assumed was vironment h ossetrltosi ewe iest and diversity between relationship consistent The ni h atfwdcdstede-e eti en- benthic deep-sea the decades few last the Until L ATITUDINAL NTHE IN G AINSOF RADIENTS D EEP S clg,Vl 6 o 9 No. 86, Vol. Ecology, EA D IVERSITY September 2005 LATITUDINAL GRADIENTS 2293

a whole responds to environmental gradients that par- allel latitude (Stuart et al. 2003). Why is deep-sea species diversity depressed toward the poles? Rex et al. (1993) suggested that it might be related to the higher and more seasonally variable sur- face production at high latitudes (Campbell and Aarup 1992). The resulting pulsed nutrient loading to the ben- thos may reduce diversity for the same reasons invoked in the section on bathymetric gradients of diversity for upper bathyal depths. High organic carbon ¯ux to the seabed from intense upwelling (Sanders 1969), lateral advection (Blake and Hilbig 1994), oxygen minimum zones (Levin and Gage 1998), concentrating bottom topography (Vetter and Dayton 1998), and reactive sed- iments introduced by disturbance (Aller 1997), is typ- ically associated with elevated standing stock and low- er diversity. Chemosynthesis at hydrothermal vents and cold seeps, the only source of in situ production in the deep sea, also results in much higher biomass and lower diversity than in adjacent soft-sediment habitats (Van Dover 2000). There is no immediate indication that temperature is related to variation in diversity at these very large scales, but this has never been examined statistically. Several indirect lines of evidence suggest that evo- Feature Special lutionary±historical factors also come into play at in- terbasin and global scales. Because of low temperatures and high pressure, the deep-sea environment is not con- FIG. 3. The relationships of the residuals of diversity vs. depth against the residuals of latitude vs. depth for gastropods ducive to fossilization. However, Stuart and Rex (1994) and bivalves in the deep North Atlantic Ocean. The analyses demonstrated a positive relationship between local and show the relationship between diversity and latitude with the regional diversity in contemporary deep-sea gastropod effect of depth statistically removed. Diversity was measured assemblages that is also related to the dispersal poten- as the Sanders-Hurlbert Expected Number of Species. The ®gure has been adapted and modi®ed with permission from tial of constituent species. This suggests that local di- Nature (Rex et al. 1993). Regression statistics: gastropods Y versity re¯ects enrichment from the regional species ϭϪ0.211X, r ϭϪ0.865, F ϭ 104.4, P ϭ 0.0001, N ϭ 37; pool, which is generated by basin-scale speciation and ϭϪ ϭϪ ϭ ϭ bivalves Y 0.101X, r 0.672, F 42.9, P 0.0001, adaptive radiation as in other marine environments N ϭ 54. (Karlson et al. 2004, Witman et al. 2004). Allen and Sanders (1996) argued that the modern distribution of Stuart and Rex 1994). Since diversity varies with depth deep-sea protobranch bivalves in the Atlantic shows a (Fig. 2), it is necessary to statistically remove the effect Tethyan in¯uence. The distribution and diversity of of depth by partial regression to detect an independent other taxa also seem to bear the imprint of history. For latitudinal relationship (Fig. 3). Rex et al. (2000) example, deep-sea foraminifera show latitudinal gra- showed that turrid gastropods in the deep eastern North dients in the North and South Atlantic (Culver and Atlantic also show a latitudinal gradient of species di- Buzas 1998), and the gradual development of fora- versity when the number of coexisting species ranges miniferan latitudinal gradients since the Eocene is ev- is summed over latitudinal bands within depth incre- ident in seabed cores (Thomas and Gooday 1996). Wil- ments. It must be stressed that the apparent latitudinal son (1998) proposed that the higher diversity of deep- gradients illustrated in Fig. 3 are based on limited sam- sea isopods in the South Atlantic and the poleward pling in both the eastern and western corridors of the decrease of diversity in the North Atlantic is partly an deep North Atlantic (see Rex et al. 1993, 2000, for historical consequence of a fairly recent invasion of maps of sampling localities). Both taxa also show the deep sea in the southern hemisphere superimposed weak, but statistically signi®cant, latitudinal gradients on a much earlier in situ radiation. in the South Atlantic where geographic coverage is Area may also be implicated at very large scales, even more restricted and inter-regional variation pro- but it is dif®cult to separate its potential role from nounced (Rex et al. 1993). Also, unlike terrestrial and historical events and unusual ecological circumstances. coastal habitats where latitudinal gradients are well es- For example, the deep Mediterranean (Bouchet and tablished and general (Hillebrand 2004a, b), it remains Taviani 1992) and Norwegian (Bouchet and WareÂn to be seen whether diversity of the deep-sea fauna as 1979) Seas support much lower molluscan diversity Special Feature esna aiiycii fteLt icn (5±6 the Miocene in Late out the dried of 10 and crisis salinity Atlantic Messinian the from Mediterranean separated The was Neogene. the during physical catastrophic disturbances experienced have shallow also by They isolated sills. partially are and Atlantic the than small- considerably er are Both Atlantic. adjacent the than 2294 d oitrrtps lmts n aelt mgr to imagery satellite and climates, past interpret meth- to paleoceanographic ods of fos- availability and the existing faunas, compilation on sil databases the extensive with more pre- rapidly much of and very past developing oceans, is complicated sent, the and of speci®c Knowledge more explanation. a ob- require regions and will hemispheres viously between observed terns 2005). intense (Vermeij more becomes the process base, this resource special- the and richer differentiation The functional ization. to that labor leads of turn division in a promotes diversity. then for systems these within tropical material resources limiting raw locally essential for Competition the is and var- that the proli®c create iation regions more tropical and of his resources temperatures In predictable diversity. higher taxonomic the compe- of of view, generation role the potential in (1978, the tition Vermeij stressed syntheses, has major 2004) of no 1987, series ex- in For a This factors. in other ecosystems. ample, of other importance the in as precludes have way just diversity, to of appear patterns they inter- large-scale have form availability to energy ev- acted and provide history mollusks that general, marine idence In dominant mollusks. on and into the widespread databases insight taxon, one new provide of comparing diversity is by the here 2003). process do al. and to et pattern Hawkins attempt (e.g., we biotas it What as continental biogeography, for marine is of analysis global hensive ito n xeine uesdmn lmsa re- as 6±8 slumps as gla- sediment cently huge Quaternary experienced by and affected 1998). ciation was al. Sea a et Norwegian by (Rothwell The buried landslide the was During seabed submarine deep 2003). massive its al. of et much (Duggen Pleistocene, Pliocene Early the iest±rdciiycreo naogbsnscale. unimodal among-basin the an of on extremes curve near diversity±productivity the two The represent 1976). may al. seas et (Dahl stock and standing production, benthic surface seasonal high Norwegian (Tse- very frigid and low the high contrast, very has Sea By is 2000). benthos al. deep et Consequently lepides the of phytodetritus. stock bio- sinking standing intensify of that temperatures degradation bottom im- elevated production be surface and low also has may Mediterranean The input portant. nutrient di- in limit from certainly differences recovery could versity, events and extinction size mass basin recent small relatively While h upiigvraiiyo aooi iest pat- diversity taxonomic of variability surprising The compre- a undertake to possible not is it Currently, 6 er g) n a nnae rmteAlni in Atlantic the from inundated was and ago), years C NLSO AND ONCLUSION ϫ FOR 10 F 3 UTURE er g Bgee l 1988). al. et (Bugge ago years R ESEARCH P ROSPECTS IHE .RXE AL. ET REX A. MICHAEL ϫ iest Jbosie l 2003 al. functional et or to (Jablonski morphological diversity be either might of gradients patterns ex- latitudinal analyze of of ways nature investigated, the little ploring yet as taxa. but of number Alternative, the variable, response single a for ac- count to variables on environmental quest exclusively independent almost identifying the centered has First, diversity gradients. explain to diversity marine on search systems. discern other to hard in con- are that forces especially fun- driving historical biogeography, most cerning the in of problems some answer tech- damental help and information to New promise scales. nology large very vari- on environmental ables with directly trends to possible biodiversity it correlate make will temperature This productivity. as and such variables environmental measure oigrvesta aeudutdyhle oipoethe improve to helped undoubtedly thought-pro- have provided that referee reviews anonymous voking an and Vermeij J. eug w e prahst rae uuere- future broaden to approaches new two urge We yjs n tenoatoos fnn component nine of neogastropods) 2000 (Crame (the gastropods clades the one of just that and by clades component heteroconchs) seven (the gradients of one bivalve just by in largely underpinned expansion was Cenozoic marked the rpi nlssutmtl a etekyt under- to patterns. key biogeographic the large-scale be standing may Phylogeo- ultimately diversity. analyses of graphic patterns tropicality just modern than generating other factors in to points areas reef diversi®cation equal in climate of concentration an The or had clades. have all energy on to effect as expected be such might factor which change, extrinsic an than rath- 1987), er global (Vermeij promoting escalation as in such some interaction diversi®cation of evolutionary importance of the to form pointing be may expansion flttdnldvriygains(o ta.20,Ja- 2000, 2003 al. al. et et (Roy blonski gradients composition diversity taxonomic latitudinal the on of attention more focus a to provide 2004). also (McClain research size of body avenue promising species and between abundance, relationships (Taylor richness, The gastropods 1977). Taylor prosobranch and predatory At- for northeastern the lantic in complex demonstrated similar been very has A gradient 2002). al. et ¯at-bench (Valentine the the pro®le into from rise departure and marked regions a high-latitude shows total clearly the of fauna, proportion in gastropod a these, as of reanalyzed plotted largest when the being carnivores, of the now One groups. is functional of America terms west North the lati- along striking of diversity very coast gastropod The in true. gradient was and tudinal reverse disparity, the morphological others low in with linked was dis- diversity clearly taxonomic morphological high regions, of certain In those parity. gastropod with Indo-Paci®c Strombidae the family within patterns regional richness in species variation of the compared (2001) al. etakA e o edn h aucit rfso G. Professor manuscript. the reading for Rex A. thank We eod twl eiprati uueinvestigations future in important be will it Second, A a CKNOWLEDGMENTS a, .Freape ti paetthat apparent is it example, For ). 01 02.Aymti clade Asymmetric 2002). 2001, a .Freape o et Roy example, For ). clg,Vl 6 o 9 No. 86, Vol. Ecology, September 2005 LATITUDINAL GRADIENTS 2295 clarity of the views presented here. This research was sup- Crame, J. A. 2000b. The nature and origin of taxonomic ported by National Science Foundation Grant OCE-0135949 diversity gradients in marine bivalves. Pages 347±360 in to MAR. Maria Mahoney helped prepare the manuscript, and E. M. Harper, J. D. Taylor, and J. A. Crame, editors. The H. J. Grif®ths and K. Linse (BAS) supplied information from evolutionary biology of the Bivalvia. Geological Society, the SOMBASE molluscan database. This paper is a contri- London, Special Publications, 177. bution to the British Antarctic Survey core research project Crame, J. A. 2001. Taxonomic diversity gradients through ``Antarctic marine biodiversity: an historical perspective.'' geological time. Diversity and Distributions 7:175±189. Crame, J. A. 2002. Evolution of taxonomic diversity gradi- LITERATURE CITED ents in the marine realm: a comparison of Late Jurassic and Allen, A. P., J. H. Brown, and J. F. Gillooly. 2002. Global Recent bivalve faunas. Paleobiology 28:184±207. biodiversity, biochemical kinetics, and the energy-equiv- Crame, J. A., and B. R. Rosen. 2002. Cenozoic palaeogeog- alence rule. Science 297:1545±1548. raphy and the rise of modern biodiversity patterns. Pages Allen, J. A., and H. L. Sanders. 1996. The zoogeography and 153±168 in J. A. Crame and A. W. Owen, editors. Palaeo- diversity of deep-sea protobranch bivalves of the Atlantic: biogeography and biodiversity change: the Ordovician and the epilogue. Progress in Oceanography 38:95±153. Mesozoic±Cenozoic radiations. Geological Society, Lon- Aller, J. Y. 1997. Benthic community response to temporal don, Special Publications 194. and spatial gradients in physical disturbance within a deep- Culver, S. J., and M. A. Buzas. 1998. Global latitudinal spe- sea western boundary region. Deep-Sea Research I 44:39± cies diversity gradient in deep-sea foraminifera. Deep-Sea 69. Research I 47:259±275. Alroy, J., et al. 2001. Effects of sampling standardization on Dahl, E., L. Laubier, M. Sibuet, and J.-O. StroÈmberg. 1976. estimates of Phanerozoic marine diversi®cation. Proceed- Some quantitative results on benthic communities of the ings of the National Academy of Sciences, USA 98:6261± deep Norwegian Sea. Astarte 9:61±79. 6266. Duggen, S., K. Hoernle, P. van den Bogaard, L. RuÈpke, and Bellwood, D. R., and T. P. Hughes. 2001. Regional-scale J. Phipps Morgan. 2003. Deep roots of the Messinian sa- assembly rules and biodiversity of coral reefs. Science 292: linity crisis. Nature 422:602±605. 1532±1534. Ellison, A. M., E. J. Farnsworth, and R. E. Merkt. 1999. Blake, J. A., and B. Hilbig. 1994. Dense infaunal assem- Origins of mangrove ecosystems and the mangrove bio- blages on the continental slope off Cape Hatteras, North diversity anomaly. Global Ecology and Biogeography 8: Carolina. Deep-Sea Research II 41:875±899. 95±115. Bouchet, P., and M. Taviani. 1992. The Mediterranean deep- Etter, R. J., and J. F. Grassle. 1992. Patterns of species di- Feature Special sea fauna: pseudopopulations of Atlantic species? Deep- versity in the deep sea as a function of sediment particle Sea Research 39:169±184. size diversity. Nature 360:576±578. Bouchet, P., and A. WareÂn. 1979. The abyssal molluscan Etter, R. J., and L. S. Mullineaux. 2001. Deep-sea commu- fauna of the Norwegian Sea and its relation to other faunas. nities. Pages 367±393 in M. D. Bertness, S. D. Gaines, and Sarsia 64:211±243. M. E. Hay, editors. Marine community ecology. Sinauer Bouchet, P., and A. WareÂn. 1980. Revision of the north-east Associates, Sunderland Massachusetts, USA. Atlantic bathyal and abyssal Turridae (Mollusca, Gastro- Fischer, A. G. 1960. Latitudinal variation in organic diversity. poda). Journal of Molluscan Studies 8(Supplement):1±119. Evolution 14:64±81. Bouchet, P., and A. WareÂn. 1985. Revision of the northeast Flessa, K. W., and D. Jablonski. 1995. Biogeography of Re- Atlantic bathyal and abyssal Neogastropoda excluding Tur- cent marine bivalve molluscs and its implications for pa- ridae (Mollusca, Gastropoda). Bollettino Malacologico, leobiogeography and the geography of extinction: a pro- Supplemento 1:123±296. gress report. Historical Biology 10:25±47. Briggs, J. C. 1996. Tropical diversity and conservation. Con- Flessa, K. W., and D. Jablonski. 1996. Geography of evo- servation Biology 10:713±718. lutionary turnover. Pages 376±397 in D. Jablonski, D. H. Brown, J. H., and M. V. Lomolino. 1998. Biogeography. Erwin, and J. Lipps, editors. Evolutionary paleobiology. Sinauer Associates, Sunderland, Massachusetts, USA. Chicago University Press, Chicago, Illinois, USA. Bugge, T., R. H. Belderson, and N. H. Kenyon. 1988. The Gage, J. D., and P.A. Tyler. 1991. Deep-sea biology: a natural Storegga Slide. Philosophical Transactions of the Royal history of organisms at the deep-sea ¯oor. Cambridge Uni- Society of London, Series A 325:357±388. versity Press, Cambridge, UK. Campbell, J. W., and T. Aarup. 1992. New production in the Gaston, K. J. 1996. BiodiversityÐlatitudinal gradients. Pro- North Atlantic derived from seasonal patterns of surface gress in Physical Geography 20:466±476. chlorophyll. Deep-Sea Research 39:1669±1694. Hawkins, B. A., and J. A. F. Diniz-Filho. 2004. `Latitude' Chase, M. R., R. J. Etter, M. A. Rex, and J. M. Quattro. 1998. and geographic patterns in species richness. Ecogeography Bathymetric patterns of genetic variation in a deep-sea pro- 27:268±271. tobranch bivalve, Deminucula atacellana. Marine Biology Hawkins, B. A., E. E. Porter, and J. A. F. Diniz-Filho. 2003. 131:301±308. Productivity and history as predictors of the latitudinal di- Chown, S. L., and K. J. Gaston. 1999. Patterns in procel- versity gradient for terrestrial birds. Ecology 84:1608± lariiform diversity as a test of the species±energy theory 1623. in marine systems. Evolutionary Ecology Research 1:365± Hillebrand, H. 2004a. On the generality of the latitudinal 373. diversity gradient. American Naturalist 163:192±211. Clarke, A., and S. Lidgard. 2000. Spatial patterns of diversity Hillebrand, H. 2004b. Strength, slope and variability of ma- in the sea: bryozoan species richness in the North Atlantic. rine latitudinal gradients. Marine Ecology Progress Series Journal of Animal Ecology 69:799±814. 273:251±267. Cosson-Sarradin, N., M. Sibuet, G. L. J. Paterson, and A. Huston, M. A. 1994. Biological diversity: the coexistence of Vangriesheim. 1998. Polychaete diversity at tropical At- species on changing landscapes. Cambridge University lantic deep-sea sites: environmental effects. Marine Ecol- Press, Cambridge, UK. ogy Progress Series 165:173±185. Jablonski, D., K. Roy, and J. W. Valentine. 2003a. Evolu- Crame, J. A. 2000a. Evolution of taxonomic diversity gra- tionary macroecology and the fossil record. Pages 368±390 dients in the marine realm: evidence from the composition in T. M. Blackburn and K. J. Gaston, editors. Macroecol- of Recent bivalve faunas. Paleobiology 26:188±214. ogy: concepts and consequences. Blackwell, Oxford, UK. Special Feature 2296 e,M .18.Gorpi atrso pce diversity species of patterns Geographic 1983. A. M. Rex, e,M .18.Cmuiysrcuei h epsaben- deep-sea the in structure Community 1981. A. M. Rex, e,M .17.De-e pce iest:dcesdgas- decreased diversity: species Deep-sea 1973. A. M. Rex, ¯uctua- sea-level Cenozoic late of Effects 1990. G. Paulay, Bathymetric 1995. Lambshead. D. J. P. and J., L. G. Paterson, Gross, L. K. Scheiner, M. S. Steiner, F. C. G., G. Mittelbach, cli,C . .A ono,adM .Rx 04 Mor- 2004. Rex. A. M. and Johnson, A. N. R., C. McClain, abun- richness, species Connecting 2004. R. C. McClain, Mortimer. K. and Darbyshire, T. Oliver, G. P. Y., S. A. between Mackie, Relationships 1998. Gage. D. J. and A., L. Levin, ei,L . .J te,M .Rx .J ody .R. C. Gooday, J. A. Rex, A. M. Etter, J. R. A., Conidae. L. in Levin, evolution of mode and Tempo 1990. J. A. Kohn, Coral 2004. Hughes. P. T. and Cornell, V. past H. Measuring H., R. 2001. Karlson, Johnson. G. K. and C., B. J. Jackson, oet,C . .J cla,J .N eo,J .Hawkins, P. J. Veron, N. E. J. McClean, J. C. M., C. di- Roberts, Species 1993. editors. Schluter, D. and E., R. L. Ricklefs, H. Allen, A. J. Hessler, R. R. Stuart, T. C. A., M. Rex, Latitudinal 2000. Coyne. G. and Stuart, T. C. A., M. Rex, A. J. Etter, J. R. Johnson, A. N. McClain, R. C. A., of M. patterns Rex, Bathymetric 1998. Etter. J. R. and A., M. Rex, ntede-e eto.Pgs453±472 Pages benthos. deep-sea the in 353. hs nulRve fEooyadSystematics and Ecology of Review Annual thos. rpddvriya bsa ets Science depths. abyssal at diversity tropod Pa- islands. oceanic tropical of faunas leobiology bivalve the on Trough, tions Rockall the I Research in Deep-Sea diversity Atlantic. northeast polychaete of patterns relationship Ecology observed productivity? the and is Dodson, 2381±2396. richness I. What S. species Willig, 2001. between R. Gough. M. Waide, L. B. and R. Reynolds, L. H. 348. ladayslgsrpdasmlgs Evolution assemblages. gastropod abyssal bathy- and lower in al metric biodiversity a as disparity phological Ecol- Global gastropods. Biogeography deep-sea and in ogy size body and dance Royal B Series the London, di- of Society, Transactions latitudinal Sey- Philosophical the the gradient? for of versity evidence invertebrates providing benthic plateau: chelles marine Shallow 2005. II macrofau- Research bathyal Deep-Sea of diversity na. and matter organic oxygen, isdvriy nulRve fEooyadSystematics and Ecology 32 of Review spe- Annual deep-sea diversity. regional cies on Pawson. in¯uences D. and Environmental Hessler, R. 2001. R. Stuart, T. C. Pineda, J. Smith, Malacologia bio- Nature oceanic gradient. an diversity along enriched regionally are communities Science biodiversity. ntehsoyo aiedvriy Science diversity. marine of history the on alnk,D . .Ry .W aetn,R .Pie n P. and Price, M. R. Valentine, W. J. Roy, K. J., D. Jablonski, .R le,D .MAlse,C .Mtemir .W. F. Mittermeier, G. C. McAllister, E. D. Allen, R. G. Chicago USA. of Illinois, Chicago, University Press, communities. ecological in versity benthos. latitu- deep-sea the Global-scale in diversity Nature 1993. species Wilson. of F. patterns D. dinal G. and Sanders, of Academy National the USA of Sciences, the of Proceedings benthos Atlantic. deep-sea the North in richness species of gradients Naturalist American biodiversity. 165 Ware abyssal A. for and hypothesis Bouchet, P. Allen, II Deep-Sea Research biodiversity. deep-sea for implications size: body New Wiley, USA. sea. York, The New 8. York, Volume biology. Deep-sea editor. :51±93. .Adro.2003 Anderson. S. :163±178. 365 16 :636±639. 32 45 :415±434. :55±67. :103±127. 97 b. :4082±4085. h mato h ulo h Recent the of pull the of impact The 293 13 :2402±2404. :327±334. 45 429 363 :129±163. :867±870. :203±228. n 05 source±sink A 2005. Ân. 42 181 300 in :1199±1214. :1051±1053. :1133±1135. IHE .RXE AL. ET REX A. MICHAEL .T Rowe, T. G. 12 58 :331± :338± 82 : oe,B .18.Tetoia ihdvriyenigmaÐthe diversity high tropical The 1981. R. B. Rosen, o,K,D alnk,adJ .Vlnie 94 Eastern 1994. Valentine. W. J. and Jablonski, D. K., Roy, pat- Spatial 2001. Hellberg. E. M. and Balch, P. D. K., Roy, adr,H .16.Mrn eti iest:acomparative a diversity: benthic Marine 1968. L. H. Sanders, Rosenberg. G. and Valentine, W. J. Jablonski, D. K., Roy, Dissecting 2000. Valentine. W. J. and Jablonski, D. K., Roy, adr,H .16.Bnhcmrn iest n h sta- the and diversity marine Benthic 1969. L. H. Sanders, tat .T,adM .Rx 94 h eainhpbetween relationship The 1994. Rex. A. M. and T., C. Stuart, Tax- 1967. Helsley. E. C. and McAlester, L. A. G., F. Stehli, D. Dobbs, C. F. DeMaster, J. D. Berelson, W. R., C. Smith, diversite et Repartition 1977. Indian M. in Sibuet, patterns Biodiversity 1998. C. R. C. Sheppard, ec- in gradients diversity Taxonomic 1970. M. J. T. Schopf, tat .T,M .Rx n .J te.20.Large-scale 2003. Etter. J. R. and Rex, A. M. T., C. Stuart, oezeg .L,adZ basy 93 o r di- are How 1993. Abramsky. Z. and L., M. Rosenzweig, time. and space in diversity Species 1995. L. M. Rosenzweig, roiisfrtoia ef.Science conservation reefs. and tropical hotspots for Wer- biodiversity priorities B. T. Marine and Vynne, 2002. C. ner. Wells, F. Spalding, M. Schueler, oas-y iw ae 103±129 Pages view. corals'-eye oe .T 93 ims n rdcino h deep-sea the of production and Biomass 1983. T. G. Rowe, Low-sea- 1998. Kahler. G. and Thomson, J. G., R. Rothwell, h ainlAaeyo cecs USA Sciences, of of Proceedings Academy rule.'' National ``Rapoport's the for gra- diversity evidence latitudinal no dient: and provinces molluscan Paci®c the of B Series Proceedings London, gastropods. Society, Indo-Paci®c: Royal the strombid across using diversity analyses morphological of terns New York, New Wiley, sea. The USA. York, 8. Volume biology. sea td.Aeia Naturalist American study. Sci- of Academy USA National ences, the of causal of Proceedings tests hypotheses. gradients: diversity latitudinal Marine 1998. Lon- B Society, Series Royal the don, of Proceedings bivalves. marine clades of and groups functional gradients: diversity latitudinal iiytm yohss rohvnSmoi nBiology in 22 Symposia Brookhaven hypothesis. bility±time bac nis ae 119±136 Pages snails. pros- deep-sea obranch in diversity species and pattern development Bulletin America of Society 455±466. Geological implications geology. some and for bivalves Recent of diversity onomic the in II Stephens. ¯ux. Research processes particle Deep-Sea M. biogenic benthic by control and in Paci®c: Pope, equatorial abyssal variations H. Latitudinal R. Hoover, 1997. J. D. Hammond, Research Deep-Sea Gascogne. de (Holothurides±Aste Bio- data. in error Conservation taxonomic and of diversity effects and corals, Ocean Bulletin America of Geo- Society implications. geologic logical their and bivalves and toprocts pta n eprlpten fde-e eti species benthic deep-sea 297±313 of Pages diversity. patterns temporal and spatial University USA. York, Columbia New benthos. York, New deep-sea Press, the re- in and biology cruitment larval Reproduction, editors. Eckelbarger, est n rdciiyrltd ae 52±65 Pages related? productivity and versity UK. Cambridge, Press, University Cambridge History), UK. Cambridge, (Natural Press, University Museum Cambridge and London, British biosphere. evolving arbnhs ae 97±122 Pages macrobenthos. me- Pleistocene Nature Sea. Late Mediterranean 377±380. large western the very in a gaturbidite of emplacement level USA. Illinois, perspectives. Chicago, Press, geographical Chicago of and University historical ecological in diversity communities: Species editors. Schluter, D. and lefs :71±81. 95 267 :3699±3702. :293±299. rds nzn rfnedn eGolfe le dans profonde zone en Ârides) 44 :2295±2317. in 102 7 .A ye,eio.Ecosystems editor. Tyler, A. P. :847±868. in :243±282. in .T oe dtr Deep- editor. Rowe, T. G. in 268 24 295 clg,Vl 6 o 9 No. 86, Vol. Ecology, .M on n .J. K. and Young M. C. .L oe,eio.The editor. Forey, L. P. 81 :549±563. :2503±2508. e Echinodermes des  :1280±1284. :3765±3768. 91 :8871±8874. in .E Rick- E. R. 392 78 : : September 2005 LATITUDINAL GRADIENTS 2297

of the world. Volume 28. Ecosystems of the deep oceans. Van Dover, C. L. 2000. The ecology of deep-sea hydrother- Elsevier, Amsterdam, The Netherlands. mal vents. Princeton University Press, Princeton, New Jer- Taylor, J. D., N. J. Morris, and C. N. Taylor. 1980. Food sey, USA. specialization and the evolution of predatory prosobranch Vermeij, G. J. 1978. Biogeography and adaptation: patterns gastropods. Palaeontology 23:375±409. of marine life. Harvard University Press, Cambridge, Mas- Taylor, J. D., and C. N. Taylor. 1977. Latitudinal distribution sachusetts, USA. of predatory gastropods on the eastern Atlantic shelf. Jour- Vermeij, G. J. 1987. Evolution and escalation. Princeton Uni- nal of Biogeography 4:73±81. versity Press, Princeton, New Jersey, USA. Thomas, E., and A. J. Gooday. 1996. Cenozoic deep-sea Vermeij, G. J. 2004. Economies evolving: how the struggle benthic foraminifers: tracers for changes in oceanic pro- for life made the past and creates the future. Princeton ductivity? Geology 24:355±358. University Press, Princeton, New Jersey, USA. Tilman, D., and S. Pacala. 1993. The maintenance of species Vermeij, G. J. 2005. From phenomenology to ®rst principles: richness in plant communities. Pages 13±25 in R. E. Rick- toward a theory of diversity. Proceedings of the California lefs and D. Schluter, editors. Species diversity in ecological Academy of Sciences, in press. communities: historical and geographical perspectives. Veron, J. E. N. 1995. Corals in space and time. The bioge- University of Chicago Press, Chicago, Illinois, USA. ography and evolution of the Scleractinia. Comstock/Cor- Tselepides, A., K.-N. Papadopoulou, D. Podaras, W. Plaiti, nell, Ithaca, New York, USA. and D. Kontsoubas. 2000. Macrobenthic community struc- Vetter, E. W., and P. K. Dayton. 1998. Macrofaunal com- ture over the continental margin of Crete (South Aegean munities within and adjacent to a detritus-rich submarine Sea, NE Mediterranean). Progress in Oceanography 46: canyon system. Deep-Sea Research II 45:25±54. 401±428. Whittaker, R. J., and R. Field. 2000. Tree species modelling: Valdovinos, C., S. A. Navarette, and P. A. Marquet. 2003. an approach of global applicability? Oikos 89:399±402. Mollusk species diversity in the Southeastern Paci®c: why Wilson, G. D. F. 1998. Historical in¯uences on deep-sea iso- are there more species towards the pole? Ecography 26: pod diversity in the Atlantic Ocean. Deep-Sea Research II 139±144. 45:279±301. Valentine, J. W. 2001. Geologic time, history of biodiversity. Wilson, M. E. J., and B. R. Rosen. 1998. Implications of Pages 215±231 in S. A. Levin, editor. Encyclopedia of bio- paucity of corals in the Paleogene of SE Asia: plate tec- diversity. Volume 3. Academic Press, New York, New tonics or centre of origin? Pages 165±195 in R. Hall and York, USA. J. D. Holloway, editors. Biogeography and geological evo-

Valentine, J. W., T. C. Foin, and D. Peart. 1978. Provincial lution of SE Asia. Backhuys, Leiden, The Netherlands. Feature Special model of Phanerozoic diversity. Paleobiology 4:55±66. Witman, J. D., R. J. Etter, and F.Smith. 2004. The relationship Valentine, J. W., K. Roy, and D. Jablonski. 2002. Carnivore/ between regional and local species diversity in marine ben- non-carnivore ratios in northeastern Paci®c marine gastro- thic communities: a global perspective. Proceedings of the pods. Marine Ecology Progress Series 228:153±163. National Academy of Sciences, USA 101:15664±15669. Ecology, 86(9), 2005, pp. 2298±2309 ᭧ 2005 by the Ecological Society of America

PLANT SPECIES INVASIONS ALONG THE LATITUDINAL GRADIENT IN THE UNITED STATES

THOMAS J. STOHLGREN,1,6 DAVID BARNETT,2 CURTIS FLATHER,3 JOHN KARTESZ,4 AND BRUCE PETERJOHN5 1National Institute of Invasive Species Science, U.S. Geological Survey, Fort Collins Science Center, Fort Collins, Colorado 80526 USA 2Natural Resource Ecology Laboratory, Colorado State University, Fort Collins, Colorado 80523 USA 3USDA Forest Service, Rocky Mountain Research Station, Fort Collins, Colorado 80526 USA 4Biota of North America Program, University of North Carolina, Chapel Hill, North Carolina 27599 USA 5Patuxent Wildlife Research Center, U.S. Geological Survey, 12100 Beech Forest Road, Laurel, Maryland 20708 USA

Abstract. It has been long established that the richness of vascular plant species and many animal taxa decreases with increasing latitude, a pattern that very generally follows declines in actual and potential evapotranspiration, solar radiation, temperature, and thus, total productivity. Using county-level data on vascular plants from the United States (3000 counties in the conterminous 48 states), we used the Akaike Information Criterion (AIC) to evaluate competing models predicting native and nonnative plant species density (number of species per square kilometer in a county) from various combinations of biotic variables (e.g., native bird species density, vegetation carbon, normalized difference vegetation in- dex), environmental/topographic variables (elevation, variation in elevation, the number of land cover classes in the county, radiation, mean precipitation, actual evapotranspiration, and potential evapotranspiration), and human variables (human population density, crop- land, and percentage of disturbed lands in a county). We found no evidence of a latitudinal gradient for the density of native plant species and a signi®cant, slightly positive latitudinal gradient for the density of nonnative plant species. We found stronger evidence of a sig- ni®cant, positive productivity gradient (vegetation carbon) for the density of native plant species and nonnative plant species. We found much stronger signi®cant relationships when biotic, environmental/topographic, and human variables were used to predict native plant species density and nonnative plant species density. Biotic variables generally had far greater in¯uence in multivariate models than human or environmental/topographic variables. Later, we found that the best, single, positive predictor of the density of nonnative plant species in a county was the density of native plant species in a county. While further study is

Special Feature needed, it may be that, while humans facilitate the initial establishment invasions of non- native plant species, the spread and subsequent distributions of nonnative species are con- trolled largely by biotic and environmental factors. Key words: native plant species; non-indigenous plant species; patterns of invasion; species densities.

INTRODUCTION [1999]). The theory that species richness is largely de- termined by environmental factors is sometimes called General ecological theories about the patterns of bi- ``local determinism'' (Ricklefs 2004). It logically fol- ological diversity provided competing hypotheses for lows that these relationships might be similar for both this study. On one hand, von Humboltd (1808), Currie native and nonnative plant species, assuming they re- (1991), Huston (1979, 1994), Rosenzweig (1995), and spond similarly to energy and resource availability. We Hawkins et al. (2003) described a fairly consistent trend refer to this as the ``environment-matching hypothe- of decreasing richness of vascular plants, birds, mam- sis.'' mals, amphibians, and reptiles with increasing latitude On the other hand, it also follows from theories of and co-varying environmental factors. The pattern fol- plant invasions that competition for available resources lowed declines in actual and potential evapotranspi- ration (AET, PET) and solar radiation, with optimal and subsequent resource limitations could constrain conditions for growth and productivity associated with species coexistence, especially if newly arriving small the peaks in richness (Currie 1991; but see Huston seeds of nonnative species must compete with well- established, mature, native plants (Stohlgren et al. 2003). If available niches are ®lled with many native Manuscript received 29 July 2004; revised 23 November species, and if competition for resources were a strong, 2004; accepted 29 November 2004. Corresponding Editor: B. A. Hawkins. For reprints of this Special Feature, see footnote broad-scale force in nature, we might expect to see 1, p. 2261. negative relationships between native and nonnative 6 E-mail: tom [email protected] species richness (or densities) in local or regional ¯o- 2298 September 2005 LATITUDINAL GRADIENTS 2299 ras. This has been broadly termed ``competitive exclu- non-indigenous species is largely human-induced (via sion theory'' (Grime 1973). trade, modern transportation, and urbanization pat- Humans increase the complexity of the issue and, terns), we included human population density and road potentially, the uncertainty associated with any anal- density in the list of potential drivers of the patterns yses of patterns of native and nonnative species plant of species introductions. Approximately 60% of the richness. Humans may have settled in native species- invasive nonnative plant species were escaped horti- rich areas or may have introduced nonnative species cultural or agricultural products (Reichard and White to those areas or adjacent species-poor areas. Further 2001). Since many nonnative plant seeds arrived as establishment and spread of invasive nonnative species contaminants with forage crops and are thought to be- may be a complex function of human-caused and nat- come initially established on disturbed sites (Reichard ural disturbances, land-use change, and environmental and White 2001), we included percentage of cropland conditions, along with the complex traits and autecol- area in a county and percentage of disturbed lands (e.g., ogy of the invading and resident species. cultivated land, mining, urban areas) as ``human'' var- Understanding the patterns of native species richness iables in regression analyses. Currie (1991), Huston at local, landscape, and regional scales may further our (1994), Rosenzweig (1995), and Hawkins et al. (2003) understanding of invasion by nonnative species. Some do not directly discuss modern patterns of diversity landscape-scale surveys of plant diversity in the United resulting from the exchange of species among conti- States have shown signi®cant positive relationships be- nents and habitats. Our analysis may complement these tween native and nonnative plant species richness in works by evaluating native and nonnative plant species 0.1-ha plots in the Central Grasslands, Rocky Moun- densities at a large spatial scale. tains of Colorado, and arid ecosystems in southern Utah We assessed preliminary trends in native and non- (Stohlgren et al. 1997, 1998, 1999a, 2001, 2002). In native plant species densities with a large county-level many cases, native and nonnative species richness was data set on vascular plant distributions in the United positively correlated to soil fertility and water avail- States. Our objectives were to: (1) evaluate simple pat- ability. Preliminary county-level data showed that na- terns of native and nonnative plant species density Feature Special tive plant species richness was positively correlated to (number of species per square kilometer in a county) nonnative species richness in 45 of 46 states (Stohlgren related to latitude and productivity (total vegetation et al. 2003). Still, little is known about the abiotic and carbon) gradients in the continental United States; and biotic factors associated with national-scale patterns of (2) use the Akaike Information Criterion and infor- plant species distributions. mation-theoretical model selection to provide a more It is intuitive that factors other than latitude more detailed evaluation of competing models explaining na- directly in¯uence plant species richness and that lati- tive and nonnative plant species density relative to var- tude is cross-correlated with AET, PET, precipitation, ious combinations of biotic, environmental/topograph- mean annual temperature, solar radiation, and other ic, and human variables. factors (Currie 1991, Rosenzweig 1995, Hawkins et al. 2003). Given the complex topography of the United METHODS AND STATISTICAL APPROACH States, predominantly north-south mountain ranges, The plant data set was gathered over the past 35 years and elevation and rain shadow effects, productivity by the Biota of North America Program (BONAP). The may be a better surrogate of optimum growth condi- taxonomic accuracy and completeness of the collection tions than latitude. We anticipate that native and non- have made it the standard plant data set for many gov- native plant species densities may be affected by op- ernment and nongovernment agencies. The data set in- timal combinations of warm temperatures, high light, cluded the occurrence of over 22 800 native and 3726 water, and nutrient availability (for carbon accumula- nonnative plant taxa in 3114 counties in the 48 con- tion) and by avoiding high stress, extreme environ- terminous states. Nonnative plant species were de®ned ments such as very high latitudes and elevations, or as those plant species with origins in other countries. extremely arid or low-light environments. Habitat het- This distinction is nonambiguous and well accepted. erogeneity and moderate levels of disturbance also may Nonnative plant species records re¯ect all past intro- increase species richness (Huston 1994, 1999), so we ductions, including some species that have not been included the variation in elevation and the number of recently reported. Due to incomplete data on current land cover classes in a county as simple measures of species distributions, we assumed that the patterns of habitat variation. past introductions are only a ®rst approximation of cur- The role of modern humans in altering species dis- rent introduction patterns. To further protect against tributions cannot be denied. Direct habitat loss and the in¯uence of incomplete data in some counties, all intentional invasive species introductions have been counties with fewer than 100 native plant species re- directly linked to human habitation (Soule 1991a, b, corded were removed from analyses, leaving 3000 Wilcove et al. 1998), and approximately 60% of hu- counties in the continental United States. mans live within approximately 170 km of the ocean Because counties vary greatly in area (from 59 km2 (or sea; Hindrichson 1997). Since the introduction of for New York County, New York, to over 52 000 km2 Special Feature lie iteo h aito nntv pce richness species native in ( variation the of little ex- plained ``area'' Pre- that showed densities. analyses species species±area liminary We of pri- analyses. models as regression several species in investigated variables plant the dependent nonnative used and mary we native California), of County, density Bernadino San for 2300 Ϯ say pedx.Freape aiueadmean ( and PET to latitude related highly example, were temperature For Appendix). essary; erle ntosroae o rdciiy(potential productivity for surrogates scales. two spatial on large relied We at pro- measure is to and expensive annually, hibitively and index seasonally com- belowground changes and ponents, be- vegetation aboveground measure involves to it difference dif®cult cause is productivity normalized Plant (NDVI). the and pendix). r r a ihycreae ihhmnpplto density population human ( with total density correlated road to highly since was However, mines] Appendix). disturbed surface area; area county an and of and (ratio cultivated, county, disturbance [developed, a habitat in of cropland index of percentage density, environmental/topographic (Appendix). were seven variables degree-days leaving growing and excluded, days, frost humidity, temperature, minimum di- Likewise, plant analysis. dropped further of were from temperature driver mean and direct latitude more so versity, a considered was PET u aibe htwr ihycoscreae ( cross-correlated highly were screened we that 2002), variables Anderson out and et Burnham (Neter 1990, multicollinearity al. to related frost-free problems sta- of reduce tistical to However, number degree-days. the growing and and days humidity, radiation, so- evapotranspiration, lar potential pre- tem- and annual minimum actual mean cipitation, mean temperature, county, annual a mean in perature, elevation in var- and iation elevation, mean latitude, including variables ic analyses. the such but for preferred, exist been not (below have do analyses data would resolution level) ®ner county smaller that the considerably and were average realize counties on we U.S. but area eastern of counties, that effect among the data reduced comparing This density when of scales. models spatial complex these more at than work density better would or kilometer) species well square as of per species measures of We (number simple ignored. be not that could it concluded sense), statistical a of (in effects signi®cant area were there species because in¯uencing Still, in richness. area than important history more land-use evolutionary be may and heterogeneity, history, habitat disturbance change, environment, the related factors to other many Thus, respectively). 0.0001, hl qa rlse mut fvrainwr ex- were log variation to of relationships amounts by lesser plained or equal while 2 .0 ofroitss rnfre aaweenec- where data transformed tests, Bonferroni 0.80; itcvralsiiilyicue eeaincarbon vegetation included initially variables Biotic ϭ ϭ ua atr nlddhmnpplto,road population, human included factors Human eiiilycniee 3environmental/topograph- 13 considered initially We .5,i a xlddfo ute nlss(Ap- analysis further from excluded was it 0.85), .6 n ontv pce ihes( richness species nonnative and 0.16) 10 ae)( (area) HMSJ THGE TAL. ET STOHLGREN J. THOMAS r r 2 Ͼ r ϭ 2 .1,and 0.91), ϭ .1and 0.11 0.001), r Ͼ eefud( cross-correlations found egregious were no and USA), Richmond, California, Port Software, SYSTAT signi®cance 10; of version tests (Systat pair- Bonferroni using and comparisons variables environ- wise the human and and density, mental/topographic species bird NDVI, carbon, variables. at human densities environmental/topographic and to species compared nonnative scales and county describe native better of would primary patterns variables our test to biotic to analysis that designed models the hypothesis of expanded suite and priori so a set 2005), an al. data et the (Stohlgren updated data models we of two set only preliminary tested a We on variables. vari- human densities environmental/topographic or species ables either plant to nonnative compared and native strongly predict more would variables hab- biotic heterogeneity, and itat may environments ®ne-scale biota and because coarse- that integrate hypothesized gradi- further latitude We than ents. rather gradients conter- productivity associated the closely with more for be would counties States United within minous density species plant atlases. published lacking states and for distribu- state information county-level Various tion provided 1990s. late publications the regional and between conducted mid-1970s projects the atlas bird pub- breeding from obtained lished were data distribution states, states. 50 36 across For counties 3079 for producing data richness USA, species Patuxent Maryland, USGS Center, the Native Research at Wildlife densities. compiled were species data species plant bird nonnative predict a to in and models species in native bird variable (Currie independent native scales an of as density global county the at used we environment many 1991), integrates the also Be- of species 1995). bird aspects of (Rosenzweig density studies the other cause in found has high as been birds productivity, disturbance moderate (nesting high and county heterogeneity, of habitat a elements in integrated species only) den- bird the native that of demonstrated we sity paper 2005), another al. In et density. (Stohlgren species bird native included variables. other and latitude productivity with a compare of to estimate carbon gradient used Vegetation commonly details). a for and provided 2004) Appendix al. Re- (see et NDVI Hof (VEMAP2 Atmospheric USA; for centroids Colorado, Boulder, Center search, county National to 2000, kriged DATA, of United then the grams in locations States, as 3168 at meter measured square 30- per (1961±1990) carbon a mean on annual based yr carbon) vegetation total aboveground ihtlrnelvl ( levels tolerance high ersin ormv nin®atpeitr rmthe ( from models predictors insigni®cant remove to regressions easse rs-orltosaogvegetation among cross-correlations assessed We nonnative and native mean that hypothesized We heterogeneity habitat of measure biotic additional An aty oe eiul eeisetdfrunderlying for ignored. inspected were were predictors residuals signi®cant model Lastly, no that assure to P Ͼ r .5,t eoeadtoa aibe with variables additional remove to 0.15), Ͼ .0.W odce tpiemultiple stepwise conducted We 0.80). Ͼ .5 oewr on) and found), were none 0.95; clg,Vl 6 o 9 No. 86, Vol. Ecology, September 2005 LATITUDINAL GRADIENTS 2301 patterns (Zar 1974), and none were found. Signi®cant values in all cases were determined using the SYSTAT statistical software (version 10). We used the Akaike Information Criterion (AIC) and information-theoretical model selection (Burnham and Anderson 2002) to evaluate multiple regression models related to the speci®c hypotheses above. While this approach may not produce the ``best'' model as com- pared to other computationally intensive approaches, it does allow for direct comparisons among multiple models. For least-squares regressions, we assumed normally distributed errors (a fair ®rst approximation in our case), and AIC was computed as AIC ϭ n log(RSS/n) ϩ 2K (1) where n was the sample size, RSS was the residual sum of squares in the model such that RSS/n is the maxi- mum likelihood estimator, and K was the number of estimable parameters in the model (including the in- tercept and residual variance; Burnham and Anderson 2002:63). The differences in AIC values between the best model (lowest AIC) and other models were used to rank the models in relation to their support based on the data. Feature Special

RESULTS Latitudinal gradients of plant species density Regression analyses showed no relationship between latitude and native plant species density (r2 ϭ 0.0001, P ϭ 0.60; Fig. 1). There was a statistically signi®cant but biologically meaningless positive relationship be- tween latitude and nonnative plant species density (r2 ϭ ϭ FIG. 1. Regressions of species density in a county on 0.01, P 0.0001; Fig. 1). latitude for (a) native plant species (r2 ϭ 0.0001, P ϭ 0.60) and (b) nonnative plant species (r2 ϭ 0.01, P ϭ 0.0001) (n Productivity gradients of plant species density ϭ 3000 counties in 48 states of the United States). Data were Plant species density was more strongly correlated log transformed. to total vegetation carbon than to latitude. Higher na- tive and nonnative plant species densities were asso- ciated with areas of high total vegetation carbon (Fig. cipitation and negatively correlated to radiation, ele- 2). Total vegetation carbon explained approximately vation, and the variation in elevation in a county. Po- 22% of the variation in native species densities and 7% tential evapotranspiration, which was strongly corre- of the variation in nonnative species densities. There lated to latitude, and AET were positively correlated were several counties moderately high in vegetation to one another and to precipitation and negatively cor- carbon with high values of native and nonnative plant related to elevation and variation in elevation (Table species density, but these were offset by a large number 1). of counties with low densities of species (Fig. 2). Solar radiation was negatively correlated to vege- tation carbon. This shows the potentially confounding Cross-correlations among variables geographic patterns in the United States where there We found many cross-correlations among factors re- are signi®cant positive correlations between latitude lated to latitude and total vegetation carbon that can and elevation and between latitude and the variation in confound or help interpret patterns of plant species elevation in a county. This describes the geography and density (Table 1). Native and nonnative plant species topography of the United States due to north-south- density were strongly positively correlated to bird spe- oriented mountain ranges (e.g., Rocky Mountains, Si- cies density, and these three biotic variables were pos- erra Nevada, Cascade, and Appalachian ranges), which itively correlated to vegetation carbon and NDVI. All intercept precipitation and cause rain shadow effects the biotic variables were positively correlated to pre- with west-to-east-moving jet streams. Special Feature 2302 iepatseisicue l itc environmental/ biotic, all included species plant tive the follow. for that PET models with multivariate replaced were and variables latitude However, temperature latitude. with neg- correlated temperature, signi®cantly atively all annual were AET mean and PET, precipitation, and minimum gree-days, 1). (Table species county bird a with in and and density density native species with plant associated county nonnative positively a but vari- in habitat weakly these disturbed was with of percentage correlated The negatively ables. was area native crop- density, land while carbon, species vegetation and plant density, species nonnative bird and native to rdciiy(oa eeaincro)fr()ntv n (b) and native (a) ( species for plant nonnative carbon) vegetation (total productivity eelgtransformed. log were F h etsnl oe eciigtedniyo na- of density the describing model single best The de- growing variables, the of screening initial the In correlated positively strongly was population Human IG .Rgesoso pce est nacut on county a in density species of Regressions 2. . nomto-hoeia models Information-theoretical fpatseisdensity species plant of n ϭ 00cute n4 tts.Data states). 48 in counties 3000 HMSJ THGE TAL. ET STOHLGREN J. THOMAS nsadrie ata ersince®ins( coef®cients regression based partial predictors variables important standardized individual most on the 15 model, the the in Of used native density. in variation species the plant of 69% value explained AIC and negative) 691 of±16 (most 1). smallest the model had 2, model (Table This variables human and topographic, Ϫ .6.Ti sasrne oe i em of terms (in model stronger a is This 0.66). ain( vation h togs igepeitrvral a aiebird native ( was density 1±4), variable species predictor (models single factors strongest biotic the included In that [2002:70]). Anderson models and the Burnham (see 10 exceed h mlet(otngtv)ACvleof value AIC had negative) 8 of (most Model arrival smallest countries. the the other to from prior an species millennia nonnative as were for density species entrenched species native well the plant because native variable the independent use could density, species we plant nonnative predicting In 3, 8). envi- (Table model biotic, variables included human and also ronmental/topographic, species plant nonnative of environ- or 2). variables (Table human variables mental/topographic including models better than den- performed for variables species biotic general, plant including models native In sity, describing diversity. models species seven plant predic- the strong native were of elevation tors and rela- PET with negative tionships variables, environmental/topographic ontv ln pce est ( density species the plant 3), relationship native positive Table the to 8±12; was predictor (models important factors most biotic used that icn oiie olna eainhpbtennative between relationship nonlinear positive, ni®cant other these support not do models. data the that again suggest modeling would However, information-theory 3). to (Table the adherence model strict linear 9±12, simple models a density, being species latter plant nonnative in ol ugs httedt ontspotmdl 2 models ( differences support AIC not the do because data 7 through the that suggest modeling would information-theory to adherence Table strict 2±4; but (models 2), density species plant native in ation itude). ntetptrepeitr ( predictors three species top bird native the of in density the included also models eainhpwt aiepatseisdniy( density positive species the by plant largely native driven with was relationship and spe- plant density, non-native cies in variation the of 86% explained xlddNV,ma rcptto,ln oe,and ( cover, elevation 8 land in Model precipitation, variation impossible. mean is NDVI, models excluded a of the so comparison but variables, density, direct independent species different plant included native models for 1 model than lddapstv eainhpt aiebr species bird native to ( relationship density positive a cluded AIC ntal,tebs igemdldsrbn h density the describing model single best the Initially, aty naps o oe,w on ihysig- highly a found we model, hoc post a in Lastly, variation the of 75±85% explained models other Four he te oesepand6±7 ftevari- the of 63±67% explained models other Three min S S b where , b ϭϪ ϭ .4adPET and 0.24 .0 n eaierltosist ele- to relationships negative and 0.50) S b i Ն nee optn oes greatly models) competing indexes .0.I h oesta included that models the In 0.50). P Ͼ S .5.I h v models ®ve the In 0.15). S b b Ն ϭϪ S clg,Vl 6 o 9 No. 86, Vol. Ecology, 0.21). b .4 lora lat- read also 0.24; Ն .6,adfour and 0.66), ⌬ AIC R Ϫ 2 i ϭ 4055, 24 values) S b S AIC in- ) b ϭ i September 2005 LATITUDINAL GRADIENTS 2303 and nonnative plant species density, with 90% of the radiation, negative relationship between mean annual variation in nonnative plant species density explained temperature and elevation (r ϭϪ0.62; P Ͻ 0.0001), solely by the density of native plant species in a county and positive relationship between mean annual tem- (Fig. 3). Had we included this model in our original perature and vegetation carbon (r ϭ 0.31; P Ͻ 0.0001) suite (Table 3), its AIC ϭϪ25 125 would have made suggest that optimal conditions for productivity involve it the ``best model.'' We tested the model regionally in the interactions of energy and water (Hawkins et al. selected areas of the United States and found that be- 2003). There is growing evidence that diversity peaks tween 64% and 98% of the variation in nonnative spe- at those places with optimal conditions for growth and cies densities could be explained solely by the density productivity (Currie 1991, Rosenzweig 1992, Badgley of native plant species (Table 4). and Fox 2000, Hawkins et al. 2003; but see Huston It is worth noting that model rank order based on [1979, 1994]). AIC from best to least supported given our data is identical to the ranking based on adjusted R2 (Tables Multiple models and multiple variables associated 2, 3). In fact, the correlation between those two statis- with plant species density tics exceeds 0.96 for both native and nonnative species density response variables (unreported data). Although Multiple models of native plant species diversity in- we were initially surprised by this pattern, it is clear cluded all the variables inserted in the models (Table from Eq. 1 that when dealing with consistently large 2, model 1), initially complicating our understanding sample sizes (n ϭ 3000) and Ks that vary little (3±15) of the factors predicting patterns of native plant species most of the variation in AIC will be captured by the diversity. All the subsequent models (Table 2, models ⌬ residual sum of squares. Since that term is shared by 2±7) would be dismissed based on the AIC criterion, AIC and adjusted R2 the common rank orders should and we did not expect the model based on environ- have been anticipated. mental/topographic variables (model 6) to perform the worst of the seven models. It was instructive that mod- pca Feature Special DISCUSSION els that included biotic factors performed better than models based strictly on environmental/topographic Understanding optimal conditions for plant and human variables. The importance of humans might species density be dismissed on the grounds that humans settled in The weak relationship between latitude and native species-rich areas rather than ``causing'' species-rich species density in the conterminous United States (Fig. areas (Stohlgren et al. 2005). In this study, the best 1) might be expected due to north-south-oriented model of nonnative plant species density ultimately mountain ranges, which affect precipitation patterns supported by the data (Fig. 3) also excluded human and cause rain shadow effects with west-to-east-mov- variables. It was more informative to recognize that the ing jet streams. That is, sites of similar latitude on the primary predictor variables in the best model of native western and eastern sides of a mountain range will have plant species density (Table 2, model 1) were native very different environments, species compositions, and bird species density, elevation, and PET (biotic and likely species densities. Several authors discuss the environmental variables). limitations of latitude as an environmental predictor of Biotic factors may integrate complex and coarse- species richness (Huston 1999, Hawkins et al. 2003, resolution environmental/topographic data. In loosely Hawkins and Diniz-Filho 2004). However, the ob- comparing the models for native plant species density served productivity gradient with native and nonnative (models 1±7; Table 2) and the models for nonnative plant species densities more strongly suggests that na- species density (models 8±15, excluding model 12; Ta- tive and nonnative plant species may be responding to ble 3), the AIC rankings of two sets of models were similar environmental constraints (Fig. 2), providing a unexpectedly identical, suggesting: (1) generally biotic link to similar global-scale energy patterns (Rosen- variables k human variables ഠ environmental/topo- zweig 1995) and water-driven patterns of plant diver- graphic variables in describing patterns of native and sity (see Hawkins et al. [2003] for a review). Because nonnative plant species density, and (2) the dominant vegetation carbon was correlated more strongly with predictor biotic variable for native plant species density precipitation (r ϭ 0.78) than with radiation (r ϭ was always native bird species density, while the dom- Ϫ0.20), mean annual temperature (r ϭ 0.31), PET (r inant predictor biotic variable for nonnative plant spe- ϭ 0.24), or AET (r ϭ 0.61; all P Ͻ 0.0001), our data cies density was always native plant species density. corroborate the Hawkins et al. (2003) conclusion that It may be that native plant and bird species densities water may be a major driver of species richness in the summarize species richness data from rare and common continental United States. habitats in a county and, as such, integrate micro- and However, energy (solar radiation and temperature) macroenvironments, and disturbed and undisturbed also may play an important role in plant species density. habitats that are not adequately represented by mean The consistent signi®cant negative relationships be- environmental/topographic or human variables at coun- tween all the biotic variables and both elevation and ty scales. Special Feature DI02 .401 .81.00 0.38 0.18 1.00 0.14 0.34 1.00 NS 0.26 0.26 0.77 0.46 1.00 0.76 1.00 0.86 Elevation NDVI carbon Vegetation density bird Native density plant Nonnative density plant Native 2304 01,cnietladrgoa cls(thge tal. et al. (Stohlgren 2005 scales et regional Chong and 2002, continental 1997, land- 2001), al. at et (Stohlgren patterns scales richness scape species explain may factors adjusted and AIC ex- model both best of new terms the in 3), amined (Fig. density species plant ok onanNtoa ak ooao mesic in Colorado, ( example, aspen Park, For and 2003). meadows National al. mountain et Mountain Hawkins Allen Rocky 2002, 2000, al. Mooney and et Kleidon 1995, Rosenzweig piration. ieie ei eeaintpsi h ei Grand xeric the in types sites. low-elevation vegetation than mesic species Likewise, nonnative fewer far ABLE est nslce ein Tbe4,adi h con- native the of species. density in plant the is and 3), (Fig. 4), States (Table United terminous regions selected species in nonnative of density predictor strong the all single, of that regions a fact various country, in the enormously despite vary factors And, these variables. same these cross-correlated to signi®cantly plant are nonnative also and densities native species that precipita- surprising not and it's temperature, tion, minimum mean tem- annual perature, mean PET, AET, radiation, solar pro- ductivity, including sig- latitude, are with factors cross-correlated many ni®cantly Since establish- species. nonnative the of to ment conducive create conditions may densities environmental species con- plant conditions native high environmental to ducive total the that suggests apo)adoealrcns ae nmlil plots multiple on 1999 based 1997, al. richness et (Stohlgren overall 0.1- and (per plot) densities ha species non-plant had water and and native nitrogen, higher light, in high were that habitats Notes: ³ hr sgoigeiec htafwenvironmental few a that evidence growing is There nonnative to native of model nonlinear post-hoc The ² rcptto .501 .607 0.41 0.78 0.26 0.18 0.35 Precipitation cover Land elevation in Variation ϭ T Radiation itrac .201 0.12 0.23 0.15 0.55 0.12 0.24 0.70 0.59 Disturbance area Crop density population Human PET AET P P a P ϭ ϭ ,adgoa cls(ure19,Hso 1994, Huston 1991, (Currie scales global and ), .Coscreain Pasnce®ins mn eetdboi,evrnetltpgahc n ua variables. human and environmental/topographic, biotic, selected among coef®cients) (Pearson Cross-correlations 1. Ͼ Variable 0.09. 0.06. rnfre aawr sdweei a prpit.Creain r in®atat signi®cant are Correlations appropriate. was it where used were data Transformed ..NV,nraie ifrnevgtto ne;AT culeaornprto;PT oeta evapotrans- potential PET, evapotranspiration; actual AET, index; vegetation difference normalized NDVI, 0.1. atrsta rs pta scales spatial cross that Patterns a .Hg-lvto ie had sites High-elevation ). density Native Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ plant 0.37 0.06³ 0.28 0.15 0.20 0.33 ouu tremuloides Populus HMSJ THGE TAL. ET STOHLGREN J. THOMAS Nonnative density plant Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ R NS 0.25 0.10 0.26 0.11 0.14 0.27 2 further , density Native ) bird Ϫ Ϫ Ϫ Ϫ Ϫ NS .806 0.26 0.61 0.34 0.18 0.25 0.20 0.23 0.29 esyadVnsaa2003). Veneskaia Ve- 2003, and al. nevsky et Hawkins of 1995, pre- function (Rosenzweig and/or complex cipitation temperatures, a warm be heterogeneity, may spe- (Huston region habitat plant a factors native in that density human report cies also and studies Other disturbance, 1999). varying of endemism, of levels patterns pools, species and local regional to despite due potentially determinism,'' variation unexplained ``local some called (2004) Ricklefs coun- regions. in and 2005) al. states, et ties, (Stohlgren densities et and (Stohlgren 2003) richness al. species na- plant nonnative between and relationship there tive positive scales, overwhelming national an At mi- is scales. important county describe at suf®ciently crohabitats elevation not in may variation 1) and (Table classes cover land mea- like coarse-scale heteroge- sures but this habitat 1995), in high 1992, (Rosenzweig with density neity associated species often is bird study) other (e.g., 2000; many and al. groups species et plant biological Stein of richness 1997, The al. 4). Table et (Dobson of zones mix climate and paleobotany, and large topography heterogeneity, complex habitat area, spe- high plant with has nonnative coinciding California and cies, native scales, of statewide number At (Stohl- greatest the 2001). sites al. species-poor et than rich- gren cover) species foliar plant more and nonnative ness were in (i.e., areas invaded native-species-rich the heavily of grasslands States, central plots the United in multiple sites prairie on sites tallgrass steppe shortgrass to based the From 2002). richness al. et 0.1-ha (Stohlgren overall (per densities and na- species plot) higher plant had nonnative phosphorus, and and tive nitrogen with in Utah, high soils Monument, National Staircase-Escalante u td em ospotsm set fwhat of aspects some support to seems study Our Vegetation ab DIElevation NDVI carb. Ϫ Ϫ Ϫ Ϫ Ϫ NS .000 0.20 0.37 0.09 0.07 0.50 0.24 0.20 0.29 0.49 Ϫ Ϫ Ϫ Ϫ Ϫ NS .201 0.21 1.00 0.15 0.27 0.76 1.00 0.72 0.44 0.25 0.14 P Ͻ .5ecp hr noted; where except 0.05 clg,Vl 6 o 9 No. 86, Vol. Ecology, Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ 0.11 0.28 0.61 0.13 0.70 0.57 0.11 aito in Variation elevation Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ 0.46 0.11 0.13 0.47 0.66 0.33 September 2005 LATITUDINAL GRADIENTS 2305

TABLE 1. Extended.

Human population Land cover Radiation Precipitation AET PET density Crop area Disturbance

1.00 0.38 1.00 Ϫ0.10 Ϫ0.23 1.00 Ϫ0.06² NS 0.73 1.00 0.24 0.49 0.38 0.67 1.00 NS Ϫ0.14 0.17 0.08 NS 1.00 Ϫ0.12 Ϫ0.16 Ϫ0.42 Ϫ0.23 Ϫ0.15 Ϫ0.23 1.00 Ϫ0.29 Ϫ0.38 Ϫ0.10 0.08 Ϫ0.10 0.16 0.64 1.00

Local determinism has notable exceptions, typically and they may be less common (in terms of area) at where unique evolutionary history plays a signi®cant regional scales. role (Gentry 1986, Venevsky and Veneskaia 2003). Realizing that exceptions exist, this present study Some arid landscapes with infertile soils, such as the now provides a framework to evaluate general patterns fynbos region of South Africa (Bond 1983, Cowling et of plant density. The relatively high predictability of pca Feature Special al. 1998), southwest Australia (Abbott 1977, Abbott native and nonnative species densities at subcontinental and Black 1980), and the southwestern United States scales (R2 ϭ 0.68 and 0.90, respectively; Table 2 and (Dobson et al. 1997), are hotspots of high plant en- Fig. 3) strongly supports similar global patterns and demism and diversity. However, more of the globe con- proposed mechanisms reported where declining diver- forms to the general ``hotspots'' pattern closely asso- sity predictably coincides with harsh environments ciated with tropical and subtropical areas with abundant (Currie 1991, Rosenzweig 1995, O'Brian 1998, Francis solar radiation and rainfall (Rosenzweig 1995, Myers and Currie 2003, Hawkins et al. 2003, Venevsky and et al. 2000). At landscape scales, some endemic species Veneskaia 2003, Stohlgren et al. 2005). may be associated with arid environments such as de- sert habitats in southern Utah (Stohlgren et al. 2004), Predictable patterns of invasions but in these and other areas throughout the United of nonnative plant species States, species-rich habitats also are common in highly The patterns and relationships presented here are pre- productive areas with fertile soils, mesic habitats, and liminary. Additional data on native and nonnative plant low-elevation areas. In the continental United States distributions may improve our ®ndings. Most counties and at county scales, this pattern is corroborated by are incompletely surveyed, and the additions of veg- our ®nding that native species densities are positively etation plot data, species lists from parks and refuges, correlated with potential vegetation carbon (Tables 1 and other point locations of plant species will improve and 2, Fig. 2). However, highly productive sites that sparsely surveyed counties (Crosier and Stohlgren have not been recently disturbed and where trees gain 2004). We would also bene®t from new systematic and dominance may have low understory plant and bird periodic plant surveys, especially for invasive plant diversity. These sites may rarely cover entire counties, species, and better information on plant species mi-

TABLE 2. Models evaluated for native plant species density (NatSpDen).

Model Percentage Adjusted no. Native plant species density models AIC ⌬AIC difference R2 K 1 NatSpDen ϭ biotic, environmental/topographic, human variables Ϫ16 691 0 0 0.69 15 2 NatSpDen ϭ biotic, human variables Ϫ16 525 165 1.0 0.67 8 3 NatSpDen ϭ biotic, environmental/topographic variables² Ϫ16 320 370 2.2 0.65 9 4 NatSpDen ϭ biotic variables Ϫ16 174 517 3.1 0.63 5 5 NatSpDen ϭ environmental/topographic, human variables Ϫ15 622 1069 6.4 0.56 11 6 NatSpDen ϭ human variables Ϫ14 553 2138 12.8 0.37 5 7 NatSpDen ϭ environmental/topographic variables Ϫ14 215 2475 14.8 0.29 9 Notes: AIC is the Akaike Information Criterion; K is the number of estimable parameters in the model. ² The normalized difference vegetation index, radiation, and land cover were removed from model 3 (P Ͼ 0.15). Special Feature este trgoa cls( scales regional at densities relation- hy- working and reasonable potheses. patterns as serve the below im- set, discussed needed ships data the the Despite county in resolutions. the provements ®ner at at and extirpations scale and persistence, grations, 2306 Model n ontv ln pce est n30 onisi the in transformed. counties log 3000 were in Data States. density United species plant nonnative and eg,Tla 99,i togfreihbtn inva- inhibiting force strong a is 1999), Tilman neighborhoods plant (e.g., of scale the is force at strong invasions It a inhibiting heterogeneity). be to habitat imagined of competition, that indicator unlikely good a den- is species sity bird (assuming heterogeneity ``the high productiv- and with high ity, stress, associated environmental low factors life!'': good temperatures, warm low-latitude and in in elevations, low greatest at be areas, high-precipitation to di- in areas, expected species be plant can nonnative con- versity the States, In 2004). United Ricklefs terminous (Huston 2003, al. diversity et plant Hawkins 1999, native mech- to proposed contributing and anisms patterns global same the supports n ucnietlsae ( scales subcontinental and ABLE o ontv ln pce est oesAIC models density species plant Nonnative no. rmmdl8 DI rpae,addsubnewr eoe rmmdl9 adcvradpeiiainwr removed were precipitation and ( cover 13 land model 9; from model removed from were removed evapotranspiration were actual disturbance and and radiation area, 10; crop model from NDVI, 8; model from F Notes: h eyhg rdcaiiyo ontv species nonnative of predictability high very The NISpDen 8 T IG aito neeain adcvr rcptto,adtenraie ifrnevgtto ne,(DI eeremoved were (NDVI) index, vegetation difference normalized the and precipitation, cover, land elevation, in Variation ² 5NISpDen NISpDen NISpDen 15 NISpDen 14 NISpDen 13 NISpDen 12 11 10 .Rltosi ewe aiepatseisdensity species plant native between Relationship 3. . NISpDen 9 .Mdl vlae o ontv o oidgnu)patseisdniy(NISpDen). density species plant nonindigenous) (or nonnative for evaluated Models 3. I steAak nomto Criterion; Information Akaike the is AIC ϭ ϭ ϭ ϭ ϭ ϭ ϭ ϭ itc niomna/oorpi,hmnvariables² human environmental/topographic, biotic, niomna/oorpi variables environmental/topographic variables human variables² human environmental/topographic, model) (linear NatSpDen variables biotic variables² environmental/topographic biotic, variables² human biotic, R 2 R ϭ 2 ϭ .0 i.3 strongly 3) Fig. 0.90; .409;Tbe4) Table 0.64±0.98; HMSJ THGE TAL. ET STOHLGREN J. THOMAS K stenme fetmbeprmtr ntemodel. the in parameters estimable of number the is T ieteutmt itiuin n este fnon- of densities and species. distributions native ultimate deter- the establishment may mine factors initial environmental via However, diversity locations. plant patterns nonnative observed the in of role important an fac- play may Human tors environments. productive pro- less in of and invasions ductive pool future species facilitating regional species, the nonnative to add continually would environments Ta- productive 3, in invasions (Fig. Increasing invasions 4). to ble and prone richness particularly species of are native in terms density high in regions widespread that is (and gen- area) ecology more in The scales. principal regional eral and landscape at sions A R D09 118 195 155 125 58 0.96 140 0.72 0.95 0.64 0.98 0.64 ID OR, WA, SD ND, WY, MT, ME PA, NH, NY, LA MS, GA, FL, CO UT, NM, AZ, itself) (by CA vnsrne hnw eotderirfrterichness 3). not the (Table for earlier AIC were reported using we factors than data stronger human the Even by included supported and 3); that strongly and 2 models (Tables way (3) consistent the a tracked in generally native factors (2) same densities 4); Table species 3, nonnative (Fig. and cohorts den- native the their was of species plant sity nonnative of densities predictor the strongest of the (1) non- because: of species invasion the indigenous in role important extremely an play oa A ahntn R rgn D Idaho. ID, Oregon; Da- MT, OR, South Washington; Maine; SD, WA, Dakota; ME, kota; North York; ND, Pennsylvania; Wyoming; New WY, PA, NY, Montana; Hampshire; Louisiana; New LA, NH, Mississippi; GA, Florida; MS, FL, Colorado; Georgia; CO, Utah; UT, Mexico; New NM, ABLE ϩ Note: tt rusAdjusted groups State oa Rclf 04 n einldtriimmay determinism regional and 2004) (Ricklefs Local C rmntv ln pce est ( density species plant native from x AUTIONS ϩ .Rgoa et fteplnma oe ( model polynomial the of tests Regional 4. tt brvain r:C,Clfri;A,Arizona; AZ, California; CA, are: abbreviations State c rdcignnaiepatseisdniy( density species plant nonnative predicting ) Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ Ϫ 89259 1202 9 5 11 3 0.21 5 0.52 11 0.57 12 6 0.75 21.2 0.80 15.5 0.82 0.86 13.5 0.85 7.0 5093 3.8 3720 2.6 962 18 3248 0 0.7 335 20 1677 913 806 20 619 378 22 166 0 142 23 435 23 889 23 055 24 ,C ONCLUSIONS P ⌬ AIC Ͼ 0.15). Percentage difference , AND clg,Vl 6 o 9 No. 86, Vol. Ecology, x R si i.3. Fig. in as ) 2 N EXT Adjusted o counties No. S R TEPS 2 y ϭ K y x ) 2 September 2005 LATITUDINAL GRADIENTS 2307 of vascular plants in counties (Stohlgren et al. 2003) grassland sites, tree fall gaps, small-mammal mounds, and a linear model comparing native and nonnative and throughout many natural areas (Stohlgren et al. plant species densities (Table 3, model 12), we found 1999a, b). additional evidence to support ``the rich get richer'' In many areas, the richness and foliar cover of non- pattern (Fig. 3, Table 4). There is little evidence of a native plant species are strongly positively correlated saturation of nonnative plant species densities at coun- (Stohlgren et al. 1998, 1999a, 2003, 2005b), so habitats ty, regional, or continental scales. But because the den- vulnerable to establishment may also be vulnerable to sity of nonnative plant species generally tracks hotspots successful invasion by nonnative species. However, de- of native plant and bird species diversities and a subset spite the generalized patterns shown here where non- of environmental factors, areas of high native plant and native plant species have successfully established in bird species diversity near current invasions should be native species-rich counties, many nonnative plant spe- speci®cally targeted for early detection and rapid re- cies also have been successful in species-poor areas sponse programs. We need to carefully monitor inva- (e.g., cheatgrass in some arid shrublands). Additional sions in species-rich areas. research is needed on the multiple factors associated One next step will be to include information on spe- with successful species invasions. Predictive models cies identity and characteristics. For example, Brazilian may be improved by using many of the factors iden- pepper tree, Australian melaleuca, and Asian cogon- ti®ed here (Tables 1±3, Fig. 3) to begin forecasting grass were pre-adapted to the climatic conditions they where species will move in space and time. Theories found in Florida. Climate matching (Huston 1979, pertaining to rates of species invasions will have to 1994, Currie 1991, Rosenzweig 1995) and habitat accommodate extensive coexistence of native and non- matching, where nonnative annual plant species target native species at large spatial scales, lag times, and habitats high in native annual plant species (Stohlgren highly dominant native species or superinvaders in cer- et al. 2005), may support increasing invasions by non- tain habitats (not all species are equivalents [Hubbell native species facilitated by trade and transportation. 2001]). We need to better predict the interactions of species Feature Special ACKNOWLEDGMENTS traits and the vulnerability of habitats to invasion. Sev- eral nonnative plant species may bene®t from continued We would like to thank Wanda Manning for supplying avi- an data, assisting with data management, and obtaining lit- introduction into many habitats each year by trade and erature sources. The U.S. Geological Survey thanks the many transportation (Reichard and White 2001). This will people who contribute data to the program. We thank Greg complicate the development of site-speci®c predictive Newman, Jim Graham, Misako Nishino, and Catherine Cro- models of invasion. sier for assisting with data management. Helpful comments There are many additional factors that could affect were provided by Bradford Hawkins, Michael Huston, Ken Burnham, and two anonymous reviewers. Funding was pro- the success of nonnative species in species-rich areas vided by the US Fish and Wildlife Service, NASA Goddard including weak competition for resources from native Space Flight Center, and the US Geological Survey Biological species, use of previously underused resources, or es- Discipline and the National Biological Information Infra- cape from natural predators (see Mack et al. [2000]). structure. Logistical support was provided by the USGS Fort Collins Science Center and the Natural Resource Ecology Native and nonnative plant species may simply be re- Laboratory at Colorado State University. To all we are grate- sponding to similarly inviting habitats and resources ful. (Stohlgren et al. 2005a, b), coexisting due to habitat heterogeneity (Huston 1994), high species turnover, in- LITERATURE CITED creased pulses of available resources from continued Abbott, I. 1977. Species richness, turnover and equilibrium large and small disturbances (D'Antonio et al. 1999), in insular ¯oras near Perth, Western Australia. Australian Journal of Ecology 1:275±280. and thus, more opportunities for the establishment of Abbott, I., and R. Black. 1980. Changes in species compo- nonnative plants (Abbott 1977, Abbott and Black 1980, sition of ¯oras on islets near Perth, Western Australia. Jour- Rosenzweig 1995, Stohlgren et al. 1997, 2003). This nal of Biogeography 7:399±410. is the case for plants in Rocky Mountain National Park, Allen, A. P., J. H. Brown, and J. F. Gillooly. 2002. Global biodiversity, biochemical kenetics, and the energetic- Colorado, and throughout the Central Grasslands of the equivalence rule. Science 297:1545±1548. United States (Stohlgren et al. 1999a, 2002). Identi- Badgley, C., and D. L. Fox. 2000. Ecological biogeography fying speci®c mechanisms and processes associated of North American mammals: species density and ecolog- with invasion is beyond the scope of this correlative ical structure in relation to environmental gradients. Journal study. However, the highly signi®cant nonlinear pat- of Biogeography 27:1437±1467. Bond, W. 1983. On alpha diversity and the richness of the terns that we observed (Fig. 3, Table 4) and the highly Cape Flora: a study in southern cape fynbos. Pages 337± signi®cant cross-correlations and models are reason for 356 in F. J. Kruger, D. T. Mitchell, and J. U. M. Jarvis, concern. editors. Mediterranean-type ecosystems: the role of nutri- Disturbances such as wild®re and hurricanes, land ents. Springer-Verlag, New York, New York, USA. Burnham, K. P., and D. R. Anderson. 2002. Model selection use change, and altered disturbance regimes may fa- and multimodel inference: a practical information-theoret- cilitate invasions. However, many fairly undisturbed ical approach. Second edition. Springer, New York, New areas have been successfully invaded, such as ungrazed York, USA. Special Feature 2308 ety .H 96 neimi rpclvru temperate versus tropical in Endemism 1986. H. A. Gentry, rni,A . n .J ure 03 lblyconsistent globally A 2003. Currie. J. D. S. and P., D. A. and Francis, Roberts, M. W. Rodriguez, P. J. P., A. Dobson, ee,J,W asra,adM .Kte.19.Applied 1990. Kutner. H. M. and Wasserman, W. J., Neter, da B. A. G. Mittermeier, G. C. Mittermeier, A. R. N., Myers, Clout, M. Evans, H. Lonsdale, M. Simberloff, D. N., R. Mack, distribution global A 2000. Mooney. A. H. and A., Kleidon, patterns: regional and processes Local 1999. A. M. Huston, of coexistence the diversity: Biological 1994. diver- A. species M. of Huston, hypothesis general A 1979. A. M. Huston, biodiversity of theory neutral uni®ed The 2001. P. S. Hubbell, and Forest 2004. King. R. and Baltic, T. Flather, a C. costs: J., world's Hof, the and Humanities 1997. D. Hindrichson, F. J. Currie, J. D. Cornell, V. H. Field, R. A., B. Hawkins, ``Latitude'' 2004. Diniz-Filho. F. A. J. and A., B. Hawkins, rm,J .17.Cnrlo pce est nherbaceous in density species of Control 1973. P. J. Grime, ln omnte.Pgs153±181 Pages communities. plant ihesciaerltosi o nisem.American angiosperms. Naturalist for relationship richness±climate Science States. United the in spe- cies endangered of distribution Geographic 1997. Wilcove. USA. York, New York, New Elsevier, ground. disturbed of 'noi,C . .L uly n .Mc.19.Dis- 1999. Mack. M. and Dudley, L. T. M., C. D'Antonio, ure .J 91 nryadlresaepten fanimal- of patterns large-scale and Energy 1991. J. D. Currie, biodiversity Improving 2004. Stohlgren. J. T. and C., Crosier, Esler. G. K. and Desmet, G. P. Rundel, M. P. M., R. Cowling, Stohl- J. T. and Kalkhan, A. M. Reich, M. R. W., G. Chong, iersaitclmdl:rgeso,aayi fvariance, of analysis regression, models: statistical linear con- for Nature hotspots Biodiversity priorities. servation 2000. Kent. J. and Fonseca, Ecology epidemi- in causes, Issues 5 control. invasions: and Biotic consequences global 2000. ology, Bazzaz. F. and Bi- Change Global study. results ology modeling constraints: process-based climatic a from from inferred biodiversity of di- Oikos the animals. in and plants variation of understanding versity for scales appropriate University USA. Cambridge York, New landscapes. York, New changing Press, in species Naturalist American sity. Princeton, USA. Press, Jersey, New University Princeton biogeography. and anal- envelopment data Science using Forest opportunity ysis. of na- areas identifying tional indicators: condition ecosystem rangeland Journal Amicus report. status Ecology patterns richness. geographic species Turner. broad-scale G. of T. and R. water, Mittelbach, J. Energy, and G. Porter, E. G. 2003. E. Kerr, O'Brien, T. M. J. E. Oberdorff, Kaufman, M. D. Guegan, Ecography richness. 268±272. species in patterns geographic and 167. eeain ora fEvrnetlManagement Environmental of Journal vegetation. USA. Massachusetts, diversity. Sunderland, and scarcity Sinauer, of science the biology: Conservation :1±20. ubneadbooia nain:drc fet n feed- and 413±452 effects Pages direct backs. invasions: biological and turbance 49. n ln-pce ihes mrcnNaturalist American richness. plant-species and S Supplement Technology Weed 1441±1444. Colorado. non-native in of study plants case a synergy: dataset with knowledge global and subcontinental Distributions lands: and diversity Diversity arid comparisons. plant Africa scale southern regional in high Extraordinarily 1998. North Western richness. species modeling Naturalist plant and American sampling exotic for and approaches native New 2001. gren. 6 :507±523. 161 :523±536. 50 in :473±494. 61 .R akr dtr Ecosystems editor. Walker, R. L. :328±335. 113 403 84 :81±101. 18 :853±858. :3105±3117. :16±20. 275 in 86 :550±553. .E Soule E. M. :393±401. HMSJ THGE TAL. ET STOHLGREN J. THOMAS 4 :27±36. ,editor. Â, 1 137 :151± 27 :27± 18 : : 'ra,E .19.Wtreeg yais lmt,and climate, dynamics, Water-energy 1998. M. E. O'Brian, ikes .E 04 opeesv rmwr o global for framework comprehensive A a 2004. as E. R. Horticulture Ricklefs, 2001. White. S. P. and H., S. Reichard, Soule time. and space in diversity Species 1995. L. M. we Rosenzweig, gradients: diversity Species 1992. L. M. Rosenzweig, thge,T . .D cel n .Vne evl 1999 Heuvel. Vanden B. and Schell, D. L. J., T. Stohlgren, Belnap. J. and Lee, M. Villa, C. Otsuki, Y. J., T. Stohlgren, N. and Evangelista, H. P. Guenther, A. D. J., T. Stohlgren, rich The 2003. Kartesz. J. and Barnett, D. J., T. Stohlgren, thge,T . .Bnly .W hn,M .Kalkhan, A. M. Chong, W. G. Binkley, D. J., T. Stohlgren, imn .19.Teeooia osqecso changes of consequences ecological The 1999. D. Tilman, thge,T . .T ant,C .Fahr .L ulr B. Fuller, L. P. Flather, H. C. Barnett, T. D. J., T. Stohlgren, Soule ti,B . .S unr .A amro,L .Master, L. L. Hammerson, A. G. Kutner, S. L. A., B. Stein, thge,T . .A ul .Osk,C .Vla n M. and Villa, A. C. Otsuki, Y. Bull, A. K. J., T. Stohlgren, thge,T . .W hn,L .Shl,K .Rmr Y. Rimar, A. K. Schell, D. L. Chong, W. G. J., T. Stohlgren, D. L. and Kalkhan, A. M. Chong, W. G. J., T. Stohlgren, thge,T . .Coir .Cog .Gete,adP. and Guenther, D. Chong, G. Crosier, C. J., T. Stohlgren, rlmdl ora fBiogeography of gen- Journal interim model. an richness: eral species plant woody of prediction Home- Irwin, edition. USA. Third Illinois, wood, designs. experimental and atrsi idvriy clg Letters Ecology biodiversity. United in the patterns in introductions plant BioScience invasive States. of pathway abig nvriyPes abig,UK. Cambridge, Press, Mam- University of Cambridge Journal thought. we than less mology and more know pce ososadrr aias ilgclInvasions Biological habitats. native 37±50. in rare example and case hotspots a invasions: species plant of Patterns 2001. 2005 Alley. Environment States. the United and the Ecology in in invasions plant Frontiers of patterns richer: get .D cel .A ul .Osk,G emn .Bash- M. 1999 Newman, Son. G. Y. Otsuki, Y. and Bull, kin, A. K. Schell, D. L. in iest nRcyMuti rslns clgclAp- Ecological grasslands. Mountain plant plications exotic Rocky and native in affect diversity quality soil and grazing How nboiest:asac o eea rnils Ecology principles. 1455±1474. general for search a biodiversity: in eejh,J ats,adL .Mse.2005 Master. L. L. and Kartesz, J. Peterjohn, York, New York, New United Press, the USA. University in biodiversity Oxford of States. status the heritage: Precious eim n nqeesi nai adcp.Ecological landscape. arid an 15, Applications in uniqueness and demism, Science n .E os.20.Saeo h tts geographic states: the of 119±157 Pages State endemism. and 2000. rarity, diversity, Morse. of patterns E. L. and landscape. urban an Association Planning in American the wildlife of Journal conserving for tenance: aignnntv ln pce.PatadSoil, and Plant species. plant non-native vading e.19.Rpra oe shvn o xtcpatspe- plant exotic for Ecology havens Plant as cies. zones Riparian 1998. Lee. Monographs Ecological 25±46. diversity. plant native of spots tui .Le .A aka,adC .Vla 2002. Villa. Management Environmental 29 A. scales. spatial C. multiple spe- at and plant nonnative cies Kalkhan, by invasion A. to vulnerability M. Assessing Lee, M. Otsuki, Monitoring Assessment Environmental and landscapes. patterns: for diversity methodology plant of a assessment Rapid 1997. Schell. vneit.2005 Evangelista. ,M .1991 E. M. Â, nteUie tts ilgclInvasions, Biological States. ®shes United and birds, the plants, in in invasion of patterns and richness ,M .1991 E. M. Â, .A ti,L .Kte,adJ .Aas editors. Adams, S. J. and Kutner, S. L. Stein, A. B. :566±577. 73 253 :715±730. 9 :45±64. c. :744±750. atrso ln pce ihes aiy en- rarity, richness, species plant of Patterns a. b. 48 npress. in 51 osraintcisfracntn crisis. constant a for tactics Conservation b. :25±43. :103±113. ads lnigadwllf main- wildlife and planning Landuse 138 iehsoyhbttmthn nin- in matching habitat Life-history a. :113±125. xtcpatseisivd hot invade species plant Exotic clg,Vl 6 o 9 No. 86, Vol. Ecology, 25 :379±398. 7 :1±15. 1 :11±14. npress. in 57 :313±323. a. npress. in Species 80 69 3 : : b. : September 2005 LATITUDINAL GRADIENTS 2309

Venevsky, S., and I. Veneskaia. 2003. Large-scale energetic Wilcove, D. S., D. Rothstein, J. Dubow, A. Phillips, and E. and landscape factors of vegetation diveristy. Ecology Let- Losos. 1998. Quantifying threats to imperiled species in ters 6:1004±1016. the United States. Bioscience 48:607±615. von Humboltd, A. 1808. Ansichten der Natur mit wissenschaft- Zar, J. H. 1974. Biostatistical analysis. Prentice-Hall, Engle- lichen Erlauterungen. J. G. Cotta, TuÈbingen, Germany. wood Cliffs, New Jersey, USA.

APPENDIX A table presenting species richness and ancillary data used in this analysis is available in ESA's Electronic Data Archive: Ecological Archives E086-121-A1. pca Feature Special Ecology, 86(9), 2005, pp. 2310±2319 ᭧ 2005 by the Ecological Society of America

LINKING BIOGEOGRAPHY AND COMMUNITY ECOLOGY: LATITUDINAL VARIATION IN PLANT±HERBIVORE INTERACTION STRENGTH

STEVEN C. PENNINGS1,3 AND BRIAN R. SILLIMAN2,4 1Department of Biology and Biochemistry, University of Houston, Houston, Texas 77204 USA 2Department of Ecology and Evolutionary Biology, Brown University, Providence, Rhode Island 02906 USA

Abstract. Ecological interactions may vary geographically as a function of diversity, density, or per capita interaction strengths, but we know little about the relative importance of these three mechanisms. We examined variation in species richness, abundance, and interactions among leaf-chewing herbivores and the dominant salt-marsh plant Spartina alterni¯ora along the Atlantic Coast of the United States. High-latitude S. alterni¯ora plants are more palatable to herbivores than are low-latitude plants. Within this range of latitude, diversity and density of the dominant leaf-chewing consumers, snails and grasshoppers, in Spartina-dominated portions of the marsh varied little. Low-latitude plants, however, ex- perienced much greater levels of leaf damage from consumers than did high-latitude plants. Per capita feeding rates of low-latitude snails (Littoraria irrorata) and grasshoppers (Or- chelimum ®dicinum) in the laboratory were greater than feeding rates of high-latitude snails (Melampus bidentatus) and grasshoppers (Conocephalus spartinae). In ®eld experiments, low-latitude snails strongly suppressed S. alterni¯ora growth, but high-latitude snails had no effect on primary production. Thus, latitudinal differences in the effect of herbivores on plants (i.e., interaction strength), driven by differences in per capita effects among species, rather than differences in diversity or density, may contribute to selection for latitudinal differences in plant palatability. Because geographical differences in interaction strength can occur in the absence of differences in diversity or density, linking biogeography with community ecology will require experimental studies that explicitly measure inter- action strength at multiple geographic locations. Key words: gastropods; grasshoppers; interaction strength; latitude; plant±herbivore interac- tions; salt marsh; Spartina; top-down effects.

INTRODUCTION latitude. Ecologists have examined latitudinal variation Special Feature Over the past several decades, ecologists have in- in the strength and nature of interactions between pred- creasingly appreciated the insights that can be gained ators and prey (Vermeij 1978, Jeanne 1979, Fawcett by examining ecological questions from a geographic 1984, Stachowicz and Hay 2000, Sanford et al. 2003), perspective (MacArthur 1972, Brown and Maurer between herbivores and plants (Coley and Aide 1991, 1989, Brown 1995, Gaston and Blackburn 2000). At Bolser and Hay 1996, Pennings et al. 2001, Sotka and the same time, there has been increasing interest in the Hay 2002, Post 2005), and among competitors (James variable nature of interactions among species (Thomp- et al. 1997, Bertness and Ewanchuk 2002, Pennings et son 1988, 1994, Dunson and Travis 1991, Bertness and al. 2003). This growing list of studies is laying the Callaway 1994, Travis 1996, Callaway and Aschehoug foundation for a general theory of latitudinal variation 2000, Callaway et al. 2002). Together, these two lines in interaction strength, which generally is supporting of inquiry have stimulated studies that have examined the intuition of early naturalists that biological inter- geographic variation in interaction strength. actions are more intense towards the tropics. The strength of biological interactions is not uniform Because many factors can vary across latitude, a across space and time. Rather, interactions vary as a major task for this developing theory will be to dis- function of abiotic conditions, population densities, ge- entangle the relative contributions of different causal netic structure, and a host of other factors (Thompson factors to variation in interaction strength. Consider the 1988, Thompson and Cunningham 2002, Harley 2003, case of herbivores and plants. The nature of the inter- Menge 2003), all of which are likely to vary across action between these two groups of organisms may vary across latitude as a function of plant and herbivore Manuscript received 28 June 2004; revised 17 September species richness, population density, and/or intra- or 2004; accepted 27 September 2004. Corresponding Editor: A. A. interspeci®c variation in per capita interaction strength. Agrawal. For reprints of this Special Feature, see footnote 1, p. To the best of our knowledge, the relative importance 2261. of these factors in mediating latitudinal variation in 3 E-mail: [email protected] 4 Present address: Department of Zoology, University of plant±herbivore interactions has not been explored for Florida, Gainesville, Florida 32611 USA. any plant±herbivore system. Here we consider these 2310 September 2005 LATITUDINAL GRADIENTS 2311 issues for salt marshes along the Atlantic Coast of the latitudinal patterns in palatability. In past work, 19 of United States, focusing on the impact that herbivores 21 trials, conducted both early and late in the growing have on plants. season with ®ve different herbivores from Rhode Island Salt marshes are the dominant intertidal habitat along and Georgia, indicated signi®cant and strong feeding the Atlantic Coast of the United States, from central preferences for high- vs. low-latitude S. alterni¯ora Florida through Maine. Throughout this range, the most leaves (Pennings et al. 2001). Both early and late in abundant marsh plant is the grass Spartina alterni¯ora, the growing season, high-latitude S. alterni¯ora leaves which dominates the middle to lower intertidal zones, were softer and contained fewer phenolics than low- often in monospeci®c stands. The wide range of S. latitude leaves (Siska et al. 2002). Leaves from high- alterni¯ora provides a rare opportunity to make intra- latitude plants were also higher in nitrogen, consistent speci®c comparisons across latitude, thus minimizing with a widespread tendency for leaf nitrogen to increase artifacts that might be associated with comparing dis- with latitude (Reich and Oleksyn 2004). Latitudinal tantly related taxa. In Atlantic Coast marshes, S. al- differences in palatability and plant traits persisted over terni¯ora is attacked by a variety of consumers, in- ®ve clonal generations in a greenhouse, suggesting ge- cluding stem-boring insects (Stiling and Strong 1984), netic control of ®xed differences (Pennings and Sal- sap-sucking leafhoppers (Denno 1977), tettigoniid gado 2005). Across intertidal elevation within a single grasshoppers (Smalley 1960), snails (Silliman and Zie- marsh, S. alterni¯ora varies in height as a function of man 2001), and mammals (Pfeiffer and Wiegert 1981). edaphic conditions: plants immediately adjacent to Here, we will focus on the two most common groups creekbanks are taller and more palatable than plants of leaf-chewing invertebrates, snails and grasshoppers, throughout the marsh platform (the short-Spartina because these are geographically widespread and rel- zone) (Valiela et al. 1978, Goranson et al. 2004, Rich- atively easy to count in the ®eld, leave easily quanti- ards et al. 2005). Here we focus on the short-Spartina ®able feeding marks, and account for the vast majority zone because this comprises the vast majority of the of chewing damage to S. alterni¯ora leaves on the At- acreage of S. alterni¯ora. To explore latitudinal vari- lantic Coast of the United States (S. C. Pennings and ation in the effects of consumers on S. alterni¯ora and Feature Special B. R. Silliman, personal observations). the mechanisms underlying these effects we ask three Biogeographic theory predicts that plant±herbivore questions: (1) How do grasshopper and snail diversity, interactions will be more intense and thus plant de- density, and feeding damage vary across latitude? (2) fenses better developed at low vs. high latitudes (Mac- Do northern and southern consumers differ in per capita Arthur 1972, Gaines and Lubchenco 1982, Coley and feeding rates in the laboratory? (3) Does the impact of Aide 1991). Studies in both terrestrial (Coley and Aide consumers on S. alterni¯ora vary across latitude? 1991) and marine (Bolser and Hay 1996) systems have generally supported this hypothesis, although many ex- METHODS ceptions exist (Targett et al. 1992, Swihart et al. 1994). Study species Some early studies suffered from limitations in exper- imental design and scope, including limited replication Spartina alterni¯ora (henceforth, Spartina)isa of sites within regions, working with processed rather wind-pollinated, perennial grass. One of the most wide- than fresh plant material, and comparing between dis- spread and locally abundant plants in salt marshes tantly related plant taxa (reviewed in Pennings et al. along the Atlantic and Gulf Coasts of the United States, [2001]). These limitations have been largely overcome it spreads rapidly and extensively via belowground rhi- by recent work in salt marshes along the Atlantic Coast zomes. Seedlings are rarely observed in undisturbed of the Unites States, which has documented that, across vegetation, but new sites are colonized both by seed most of the plant community (the 10 most common and and by drifting rhizome fragments (Travis et al. 2002). widespread species), plants from replicated low-lati- The tettigoniid grasshoppers Conocephalus sparti- tude sites are less palatable to (i.e., less readily eaten nae and Orchelimum ®dicinum are the most abundant by) 13 species of leaf-chewing herbivores than are con- Orthoptera found on Spartina in high- and low-latitude speci®cs from high-latitude sites (Pennings et al. 2001). Atlantic Coast marshes, respectively (Pfeiffer and Wie- The proximal factors governing this variation in pal- gert 1981, Bertness et al. 1987, Bertness and Shumway atability include latitudinal variation in nitrogen con- 1992). Like most tettigoniids, they are omnivores. In tent, toughness, and chemistry (Siska et al. 2002), and addition to feeding on leaves, both also eat insect prey, these latitudinal differences appear to be constitutive and Conocephalus is known to feed on ¯owers and seed rather than induced (Salgado and Pennings 2005). The heads. Adults are abundant in the months of July and ultimate drivers creating variation in these plant traits August. remain unexplored. The snails Littoraria irrorata (salt marsh periwinkle) Here we focus on the most abundant of these 10 and Melampus bidentatus (coffee bean snail) are the plants, S. alterni¯ora, and examine whether latitudinal most abundant gastropods in Atlantic and Gulf Coast variation in herbivore richness, herbivore abundance, U.S. salt marshes. Both species may live for several and/or herbivore impacts on plants might be driving years. Melampus, the smaller species (up to 1.3 cm in Special Feature Littoraria Silliman, R. B. and Lee C. (S. sites low-latitude well at common as latitudes, locally higher be can at East it abundant U.S. although particularly entire is the and along Coast distributed is length), shell 2312 edn aaecudb eeae ydfeecsin differences by generated be could damage feeding 94 .R Silliman, R. B. 1984; n/rmnh oyedasnl au o ahvariable each for site. value each single at a yield to months transects, and/or leaves, across averaged the were data feed- to and damage density ing estimated Grasshopper area. visually leaf of 10% was nearest youngest, leaves two expanded the plant to fully damage single grazing A grasshopper transect, 2002. and each following of selected and August haphazardly was and Georgia, stem July Carolina, in (South Florida) Bight Atlantic the South and Connecticut) and Island, Rhode Massachusetts, lrcnatadcnan ohdtia n green and detrital both contains and contact ular n u otn nlsssgetthat suggest analysis observations content Field gut 2003). and Newell Zieman and and Silliman (Silliman 2001, reproduction and growth grass eeecsteen.A ihdensities, ascend high and they At [1982] (Daiber therein). as tide references high fungi during stems associated grass marsh the and on material standing-dead algae consuming plant and and tide low matter at surface organic marsh (New on sites feeding detritivore-omnivore latitude grazers, are lower Both Florida). at through only Jersey occurs and length) n ing igevlefrec aibea ahsite. a each yield at to variable each months averaged for and/or were value sites, single plants, data or damage plots across of radulation Snail ruler. centimeters, and a in with density area, measured was total stem) (i.e., per stems stem radulations plant whole ®ve the sites, damage to radulation the snail of and selected subset haphazardly were a At 2001). through Zie- man and cut (Silliman ``radulations'' termed that leaf, entire scars the grazing longitudinal acteristic ln,Goga n lrd)i h umr f2000± of on summers feeding the Snail in 2002. Florida) and Car- Georgia, South olina, Con- Carolina, North and Virginia, Atlantic Island (Maryland, South the Bight and (Rhode mid-Atlantic the England in and New necticut) in each sites aua rzn by grazing radular ( rnet ( transects rzslive grazes manuscript tina tina ots h yohssta aiuia ifrne in differences latitudinal that hypothesis the test To 0.25-m or 0.0625- in counted were Snails 10 on visually counted were Grasshoppers ϭ Spartina rshpe bnac n edn damage feeding and abundance Grasshopper aeili t u Huea 92 Thompson 1932, (Hauseman gut its in material oea 0stsec nNwEgad(Maine, England New in each sites 10 at zone ±0sal/ie nteshort- the in snails/site) 8±20 e aiafeigrtsi h laboratory the in rates feeding capita Per ni bnac n edn damage feeding and abundance Snail rzslv as rs hog ietrad- direct through grass marsh live grazes , n ). Spartina ϭ oee,hsntbe evaluated. been not has however, , Littoraria ± rshpesst)i h short- the in grasshoppers/site) 6±8 Melampus n a togysprs marsh suppress strongly can and esnlobservation personal Spartina slre u o32c nshell in cm 3.2 to (up larger is fet h rwho liv- of growth the affects TVNC ENNSADBINR SILLIMAN R. BRIAN AND PENNINGS C. STEVEN lnsrslsi char- in results plants Spartina Melampus Littoraria unpublished oea nine at zone .Whether ). 2 quadrats ϫ Spar- Spar- like , 2m also niiulgasopr rmtenrh(hd Island, (Rhode mostly north the (2001). from al. grasshoppers et Individual Pennings in reported pref- experiments feeding erence from rates data reanalyzed feeding we estimate grasshoppers, To of diets. exper- common laboratory using in iments of consumers rates southern feeding and compared northern we rates, feeding capita per Littoraria Spartina Melampus fcgn nvgtto,sal,o h ni±ln in- snail±plant ( the Snails or site. any snails, at teraction vegetation, on caging of fprle ni nlsr xeiet odce in conducted experiments short- enclosure the snail parallel of orpeettelws,tpcl n ihs densities highest and typical, lowest, densities the high represent and moderate, to (zero), low at enclosures a upre yte®l xeiet ecie be- described experiments ®eld low. the by supported was rmteshort- the from rae hnta ftevgtto 63-mms,0.25 mesh, (6.35-mm height or1m vegetation a the of to that than cloth greater were hardware cages zinc-coated roo¯ess of Brie¯y, constructed 2001). for Zieman methods and (Sil- and liman published Virginia previously in been have sites experiments two these the of one Results sites. from latitude Georgia) (Virginia, low- four and r n n efo southern a of leaf north- one a of and leaf one ern offered each were assays) different eoe n auain esrd hsexperiment that were assumption This leaves the measured. h, involved 72 radulations After and 2004). removed al. et Goranson 2002, ntredfeetasy)adtesuh(eri,most- (Georgia, south the ly and assays) different three in pce,w fee naae evsof leaves undamaged offered we species, and northern between mass/area potential leaves. any southern leaf for corrected in et not were (Pennings differences data leaves so northern 2001), on ma- al. was vast feeding The of herbivory. of jority absence the controls shrank nor in Autogenic grew appreciably neither leaves hours. cut because elapsed run not of were du- number eaten leaves for the both standardize of by amount To total elapsed. the divided had we ration, h leaf) 72 one of or 30% occurred (ca. feeding substantial until ran cates ol aeasmlraiiyt edon feed to ability similar a have would sasi ohRoeIln n eri eerun were Georgia 22±24 and approximately Island at water. Rhode of indoors vial both a in in bases Assays their with upright placed were h tall- the 0c ics,fo ihrteshort- the either from pieces), 10-cm ohznst o®mpeiu eut htpat from plants tall- that results the previous from con®rm plants to used zones We zone). both per species snail per leaves four ots h yohssta h feto nison snails of effect the that hypothesis the test To rhlmm®dicinum Orchelimum oetmt e aiafeigrtso h w snail two the of rates feeding capita per estimate To Snail± Melampus 2 nae) ata aecnrl niae oeffect no indicated controls cage Partial area). in ooehlsspartinae Conocephalus Spartina Spartina aisars aiue ecmae h results the compared we latitude, across varies Spartina Spartina nVrii n eri)wr tce in stocked were Georgia) and Virginia in rmhg-aiuests nasmto that assumption an sites, high-latitude from Spartina bt olce rmGeorgia; from collected (both oe ogop ffour of groups to zone, oeaemr aaal hnthose than palatable more are zone neato teghi h ®eld the in strength interaction oeo w ih RoeIsland) (Rhode high- two of zone 1 elct nml neight in animals replicate 113 , oe(ilmnadBertness and (Silliman zone Melampus Melampus 6rpiaeanimals replicate 76 , Њ Spartina clg,Vl 6 o 9 No. 86, Vol. Ecology, .Idvda repli- Individual C. Spartina nRoeIsland, Rhode in Spartina rmGeorgia from Littoraria u leaves Cut . Spartina oeor zone n ϭ (two 16 as or September 2005 LATITUDINAL GRADIENTS 2313

hoppers were acridids (mostly Dicromorpha elegans, Orphulella pelidna, and Paroxya clavuliger), which were transients from other marsh zones. Although grasshopper densities were similar between geographic regions on the dates we sampled, feeding damage was almost 10 times greater in the South Atlantic Bight than in New England (Fig. 1A), suggesting that high- latitude grasshoppers have a smaller per capita effect on Spartina foliage than low-latitude grasshoppers. Two snail species were present in quadrats. Litto- raria irrorata was absent in New England but common at most mid- and south Atlantic sites, with mean den- sities ranging from zero to over 800 snails/m2 at dif- ferent sites (Fig. 1B). (Littorina littorea and L. saxatilis occur in some New England salt marsh sites in small numbers but were not present in our quadrats.) Melam- pus bidentatus showed the opposite distribution pat- tern: it was common at many sites in southern New England, with mean densities ranging from zero to over 400 snails/m2, but rare at mid- and south Atlantic sites. In these low-latitude sites, Melampus can potentially reach high densities but is typically excluded by ex- ploitative competition and habitat modi®cation by Lit- toraria (S. C. Lee and B. R. Silliman, unpublished manuscript). Radulation damage to Spartina was ob- Feature Special served only at mid- and south Atlantic sites (Fig. 1B), where it positively correlates with snail density (Sil- liman and Zieman 2001). Thus, as with the grasshop- FIG. 1. (A) Density of grasshoppers (mostly Conoce- pers, the geographical differences in damage were phalus and Orchelimum) and grasshopper grazing damage to leaves of Spartina alterni¯ora in the short-Spartina zone of much more striking than the differences in snail abun- marshes in New England (northern sites: Maine, Massachu- dance or diversity, suggesting that high-latitude snails setts, Rhode Island, Connecticut) vs. the South Atlantic Bight have a smaller per capita effect on Spartina than low- (southern sites: South Carolina, Georgia, Florida), USA. (B) latitude snails. Density of snails (Melampus, Littoraria) and radulation dam- age to leaves of Spartina alterni¯ora in the short-Spartina In the laboratory, southern grasshoppers ate more zone of marshes in southern New England (Rhode Island, than three times as much leaf material per hour as did Connecticut) vs. the mid-Atlantic and South Atlantic Bight northern grasshoppers (Fig. 2). Littoraria fed on Spar- (Maryland, Virginia, North Carolina, South Carolina, Geor- tina leaves from both marsh zones, producing visible gia, Florida). Data are means ϩ SE, and sample sizes and P values are given above bars. Data were analyzed with two- sample t tests or with one-sample t tests vs. zero, treating sites as replicates. found at each site. Each snail density treatment was replicated three or six times. After 6 mo (May±Octo- ber), live vegetation was harvested from a 0.0625-m2 quadrat centered in each plot. Total radulations were measured on 10 haphazardly chosen stems from each plot.

RESULTS Grasshopper density was similar among geographic regions (Fig. 1A). Most individuals observed were tet- tigoniids, primarily Conocephalus spartinae in New England and Orchelimum ®dicinum in the South At- FIG. 2. Per capita grazing rates of northern and southern lantic Bight (these two species cannot be distinguished grasshoppers (Conocephalus and Orchelimum, respectively) and snails (Melampus and Littoraria, respectively) feeding on visual transects, but relative abundance was deter- on Spartina alterni¯ora. Data are means ϩ SE; sample sizes mined by representative collections and accords with and P values are given above bars. Data were analyzed with previous studies). A small proportion (Ͻ10%) of grass- two-sample t tests or with one-sample t tests vs. zero. Special Feature ni este Fg E F). with 3E, increased (Fig. radulations densities densities sites, snail these (zero) At low C). and to 3B, (Fig. compared moderate densities at snail several-fold high reduced often biomass 2). (Fig. zone tall leaves Bertness on the heavily and most from fed 2004), (Silliman al. et expected Goranson 2002, as but radulations, 2314 lnsfo ihlttd ie r oepltbeto palatable more are sites high-latitude from plants ni este Fg D.I otat manipulating contrast, In 3D). (Fig. densities snail Melampus hd sad(i.3) ordltoswr observed were the radulations in No 3A). (Fig. Island Rhode pnlB rawtrmrh aso et eepeiul ulse Slia n imn2001). Zieman and (Silliman published previously were left) on bars marsh, Broadwater B, (panel feto live on effect iil auain nlae rmete zone. either from leaves on radulations visible ttefu ie nVrii n eri,with Georgia, and Virginia in sites four the at F toraria sad E igna n F eri dmg a eoi l ramnsi hd sad.Saldniisaegvnbelow given are densities Snail Island). Rhode in treatments all in zero was (damage means Georgia are Data (F) bars. and Virginia, (E) Island, ( nteAlni os fteUie tts atmarsh salt States, United the of Coast Atlantic the On IG Manipulating .Ipcso niso ims of biomass on snails of Impacts 3. . Melampus est a togefcson effects strong had density ,()Vrii ( Virginia (B) ), Spartina Melampus aiuain,ee ttehighest the at even manipulations, ϩ D SE ISCUSSION n eeaaye ihAOA raigstsa lcs eut foeo h ignaexperiments Virginia the of one of Results blocks. as sites treating ANOVA, with analyzed were and Littoraria ims ttetostsin sites two the at biomass est nte®l a no had ®eld the in density Melampus TVNC ENNSADBINR SILLIMAN R. BRIAN AND PENNINGS C. STEVEN ,ad()Goga( Georgia (C) and ), Spartina rdcdno produced priaalterni¯ora Spartina Spartina biomass Lit- Littoraria o h otaudn atmrhplant, here marsh presented salt abundant Data most 2001). the al. low-lati- for et from (Pennings plants sites conspeci®c tude are than consumers 91.T h xetta hsdfeec ngaigpres- grazing in difference this that Aide extent and the To (Coley 1991). forests broad-leaved for been documented have damage Sim- herbivore England. in gradients New latitudinal in ilar plants conspeci®c than herbivore more damage far accumulate Bight Atlantic South the ut aeipratipiain o tepst link biogeography. to and attempts ecology for community implications important re- by These have than density. sults or effects diversity be- capita herbivore differences per in differences by in more species driven herbivore tween be her- to of herbivore appear in intensity damage differences the latitudinal in Moreover, differences bivory. latitudinal selected by be for could palatability in that differences suggesting latitudinal herbivores, from damage also more plants receive low-latitude less-palatable the that indicate ;adsaldmg rdltos npat n()Rhode (D) in plants on (radulations) damage snail and ); u apigdt niae that indicated data sampling Our neprmnsa w ie ahi A hd Island Rhode (A) in each sites two at experiments in clg,Vl 6 o 9 No. 86, Vol. Ecology, Spartina .alterni¯ora S. lnsin plants , September 2005 LATITUDINAL GRADIENTS 2315 sure selects for plant defenses against herbivores, it snails on leaves of Spartina were several times those may contribute to the latitudinal differences in plant of high-latitude consumers. In ®eld experiments, low- traits (Seliskar et al. 2002, Siska et al. 2002) and pal- latitude snails (Littoraria) strongly suppressed live bio- atability (Pennings et al. 2001) that have been docu- mass of Spartina, but high-latitude snails (Melampus) mented previously. Other factors, such as differences had no effect on live biomass, despite being stocked in soil type, porewater chemistry, or the length of the at higher densities. It is likely that these differences in growing season, may also select for differences in plant interaction strength are linked to interspeci®c differ- traits (Reich and Oleksyn 2004, Wright et al. 2004), ences in feeding ecology. Conocephalus spartinae, the perhaps with indirect effects on palatability, and the grasshopper most abundant at high latitudes, feeds relative importance of these different, putative selec- heavily on ¯owers and seeds of marsh plants (Bertness tive forces has yet to be determined. Nevertheless, the et al. 1987, Bertness and Shumway 1992) and may be striking difference in herbivore damage across latitude poorly adapted to feeding on leaves. It is also smaller argues that variation in herbivore damage could be one than the grasshopper that is most abundant at low lat- important factor selecting for different plant pheno- itudes, Orchelimum ®dicinum. Similarly, the ability of types across latitude. Consistent with this suggestion, snails to feed on the tough leaves of Spartina is size- Spartina introduced to San Francisco Bay appears to dependent. Very small Littoraria irrorata individuals have rapidly lost defenses to herbivores in the absence cannot feed on live Spartina (B. R. Silliman, personal of herbivore pressure (Daehler and Strong 1997). observation), probably because their radulas are too The latitudinal differences in damage that we doc- small and weak. Because Melampus is smaller than umented were much larger than latitudinal differences Littoraria and possesses a weaker radula at a given size in consumer density. Moreover, for the leaf-chewing (B. R. Silliman, personal observation), it may be un- guild of herbivores on Spartina, there are not striking able to feed on living leaves and be limited to a diet latitudinal differences in diversity. Conocephalus and of detritus (Rietsma et al. 1988), fresh Spartina frag- Orchelimum are both widely distributed along the East mented by other mechanisms (e.g., other grazers), and Coast of the United States, although Conocephalus is microalgae (Thompson 1984). Feature Special more common at high latitudes and Orchelimum at low Why herbivore morphology and per capita impacts latitudes. The richness of common marsh snails in- should vary across latitude is an interesting question creases from zero in northern New England to one (Me- deserving of attention. It is possible that these herbi- lampus) in southern New England to two (Melampus vore traits are driven by the impacts of temperature on and Littoraria) at lower latitudes. At low latitudes, physiology, growth rate, and activity levels. Alterna- however, Melampus is functionally absent from the tively, herbivore traits may be determined by plant short-Spartina zone at many sites (S. C. Lee and B. R. traits, in an evolutionary arms race (Vermeij 1987) that Silliman, unpublished manuscript), so it may be more is proceeding at different rates in different geographic correct to think of this as a species replacement at low regions. Finally, phylogenetic history undoubtedly latitudes than as an increase in diversity. Moreover, both predisposes and constrains species traits. For ex- Littorina littorea and L. saxatilis are occasionally pres- ample, most species of Littorara can graze on vascular ent in high-latitude marshes in New England (Bertness plants (typically mangroves), but to our knowledge no 1984, Barlocher and Pitcher 1999), although we did species of Melampus does so, suggesting that the dif- not record them in our quadrats. In addition, we have ferences in interaction intensity for snails that are re- observed lepidopteran larvae feeding on S. alterni¯ora ported here are at least in part due to phylogenetic only at high-latitude sites in northern New England, constraints. although these were also rare (Pennings et al. 2001). At ®rst glance, the strong effects of snails on live In sum, the diversity of the leaf-chewing guild on Spar- biomass of Spartina at low-latitude sites (Fig. 3B, C) tina is low and varies little across latitude within the might appear dif®cult to reconcile with the fact that range of latitude considered here. We have not exam- low-latitude marshes are dominated by Spartina. Four ined latitudinal patterns in the sap-sucking or stem- points likely explain how Spartina persists as a dom- boring guilds, however, and it is possible that diversity inant plant in the face of such strong consumer pres- and/or density of these herbivores changes across lat- sure. First, the highest snail density treatment repre- itude. We have also not examined latitudinal patterns sents the high end of densities observed in the ®eld. in feeding on seeds and ¯owers; latitudinal variation Although we do observe areas in which snails reach in these processes might affect the ability of plants to these high densities and Spartina is completely or al- colonize new sites, but would likely not affect vege- most completely suppressed, these areas are patchy and tation dynamics within a site where growth is primarily represent a small percentage of the total marsh surface. vegetative. Second, although the intermediate snail density treat- Although we did not ®nd strong differences in her- ment, which represents more typical snail densities, bivore abundance or diversity across latitude, we did also produced moderate to strong suppression of Spar- ®nd strong differences in per capita impacts. Labora- tina, the plants were not completely eradicated. We tory feeding rates of low-latitude grasshoppers and argue that this treatment represents typical conditions Special Feature Biotic cosvs ra of areas vast across Abiotic 2316 eas fdfeecsaogidvda lnsi stress in likely plants individual South, among differences the of in because replaced interactions are competitive England strong New by in found often facilitative the interactions however, lat- coast, lower entire the at plants Across stress itudes. among physical increasing competition with England, decreases (Table New latitudes Coast lower Within at Atlantic 1). intense in more are types marshes all salt of interactions biological acdtruhfriiain(ilmnadZieman and Silliman, (Silliman R. B. fertilization 2001; through hanced n otls20) hr,lwiogncnitrogen inorganic short- low the Third, in availability 2003). Bortolus and eanana-ooutr hogotmc fthe of much can throughout and marsh. plants near-monoculture of a species other remain by replaced be not will aaet lnsi edl bevd and radulation observed, densities, readily is moderate plants at to damage present are Snails netdlzn nteesl ase.Tu,regardless Thus, much marshes. how salt of lower these in the zone occupying intertidal of capable physiologically plant hc a cu titreit ni este (200± extirpation, densities snail local snails/m intermediate 600 at and occur can overgrazing which from cordgrass ABLE n ®nally, and upesd( suppressed ebvr ebvr aaet lnsicessat increases plants to damage herbivore increase may mussels on predation crab Herbivory Predation eleva- among marsh facilitation intermediate England, at New stress within salt Competition/facilitation stress Salt itrac c itrac otitnea higher at intense most disturbance ice Disturbance uvyo h ieauesget ht ngeneral, in that, suggests literature the of survey A T aiuia aito ntesrnt fspecies of strength the in variation Latitudinal neatosi tatcCatsl marshes salt Coast Atlantic in interactions .Lttdnlvraini clgclpoessi atmrhso h tatcCato h ntdStates. United the of Coast Atlantic the on marshes salt in processes ecological in variation Latitudinal 1. rcs atr References Pattern Process priaalterni¯ora Spartina 2 ϳ hnntoe viaiiyi ral en- greatly is availability nitrogen when ) Spartina 07% u o rdctd(Silliman eradicated not but 30±70%) Spartina nulse manuscript unpublished ssprse yhrioe,it herbivores, by suppressed is Spartina TVNC ENNSADBINR SILLIMAN R. BRIAN AND PENNINGS C. STEVEN as tlwlatitudes. low at marsh steol pce of species only the is cosetr os,cmeiinamong competition coast, entire across rc itrac otitnea high- at intense most disturbance wrack oelkl shields likely zone yiprata ihlatitudes) high local- at be important can ly geese (but per effects increased capita to due latitudes lower activity and increased densities to crab due latitudes lower at ability tol- competitive stress or in erance differences capita due per latitudes to lower at increases plants stress salt due increased latitudes to lower at increases plants increased evapotranspiration temperatures, higher latitudes to lower due at intense most tions rlttdsdet esnlt of growth seasonality plant to due latitudes er latitudes lnsls aaal ohrioe at herbivores to palatable less plants et nrae ogns n polar chemistry and con- toughness N increased lower tent, to due latitudes lower Spartina .Fourth ). is rlyspot ihsm xetosadcvas the caveats, and exceptions some with support, erally di- species not effects, capita differ- per versity. to or due densities primarily in latitude ences with interactions change biological to cases, coastal appear these in be all latitudes In to marshes. lower interactions salt at biological negative) (more and increase stronger latitudes to disturbance higher for pattern at general a Table suggest in low- cited 1 in publications the higher sum, sa- In somewhat marshes. and are latitude ¯ooding levels from salinity but stress linity, abiotic chronic marsh- suffer more salt all is es de®nition, stems) By plant latitudes. higher dead at ¯oating intense of (mats ice wrack from to and disturbance contrast physical In interactions, biotic marshes. nesting these goose high-latitude near some herbivory at heavy to grounds the exception is notable plant pattern A in this latitudes. decrease more lower a at is despite palatability plants latitudes, test lower higher to at on done intense interactions Herbivory been hypothesis. not crab±mussel have this however, of eleva- latitude, marsh studies across higher Parallel to them tions. marsh limiting lower and at elevations mussels in- eradicating more sites, essentially correspondingly southern tense, appears warmer, predation at crab active and most more are and more mussels appear abundant Crabs and elevations. edge, marsh lower marsh at abundant the to modest is limited crabs In by and mussels effects). on capita predation England, per New (i.e., competition and tolerance eut rmohrnasoemrn aiasgen- habitats marine nearshore other from Results hsppr .C Pennings, C. S. paper; this Lin (1985), Grosholz and Bertness (2003) al. et Pennings (2002) Ewanchuk and Bertness 2001) (1999, Bertness and Pennings etesadElsn(97,Pennings (1987), Ellison and Bertness Hardwick-Witman (1972), Red®eld ennsadBrns (2001) Bertness and (1992), Pennings Gardner and Stiven (1989), etes(2001) and Bertness Pennings (1998), Richards and (2001) Bertness and Pennings (1985), ebse l 19) rvsaaadJef- and Srivastava (1990), al. et Kerbes ennse l 20) ik tal. et Siska (2001), al. et Pennings eis(1996) feries (2002) data clg,Vl 6 o 9 No. 86, Vol. Ecology, unpublished September 2005 LATITUDINAL GRADIENTS 2317 hypothesis that both predation (Vermeij 1978, Bertness account for much of the variation in interaction strength et al. 1981, Menge and Lubchenco 1981, Fawcett 1984) between plants and herbivores (Bigger and Marvier and herbivory (Horn 1989, Bolser and Hay 1996, Cro- 1998, Cebrian 1999). We would argue that latitude (or nin et al. 1997) are more intense at lower latitudes. In more speci®cally some underlying environmental var- contrast to salt marsh studies, however, work on rocky iable[s] that correlate with latitude) may explain much shores has emphasized the importance of upwelling of the remaining variation in interaction strength be- (Connolly and Roughgarden 1998, Sanford 1999, Con- tween plants and herbivores. This is not a new idea nolly et al. 2001, Menge 2003), which affects delivery (MacArthur 1972, Coley and Aide 1991), but there are of nutrients, larvae, and food items. This likely re¯ects few rigorous tests. Incorporating latitude into meta- a fundamental difference between rocky coasts and salt analytical approaches (Post 2005) and conducting ex- marshes, where much of the food web, even for ®lter plicit geographical comparisons of plant±herbivore in- feeders (Stiven and Kuenzler 1979, Kreeger et al. 1988, teraction strength, as others have done with predation Langdon and Newell 1990), is derived from marsh or (Harley 2003, Menge 2003), would both be pro®table estuarine sources, rather than from oceanic sources. approaches. Linking biogeography and community ecology In general, ecologists who are interested in bioge- ography document patterns but not interactions. Com- Our results have important implications for any at- munity ecologists, in contrast, often study interactions tempt to link the ®elds of biogeography and community but usually do not know how these vary geographically. ecology. One major focus of biogeography is docu- When examined, however, geographic differences in menting and understanding geographical patterns in interactions often turn out to be large, indicating that biodiversity (Hillebrand 2004). All else being equal, they cannot be ignored if we hope to develop gener- variation in diversity is likely to affect community alities (Thompson 1994, Travis 1996, Bertness and function (Loreau et al. 2002); however, all else will not be equal across latitudinal gradients. Across latitudinal Ewanchuk 2002, Thompson and Cunningham 2002, gradients, the large differences in climate and produc- Sanford et al. 2003). Merging the approaches of bio- Feature Special tivity that produce patterns in diversity are likely to geography and community ecology is logistically chal- also directly affect ecological processes, perhaps more lenging, but we argue it will lead to a new level of strongly than variation in diversity does. In the case of understanding and generality. Because geographic var- Spartina along the Atlantic Coast of the United States, iation in interactions may be driven either by ecological we argue that there are strong latitudinal differences factors (e.g., resource availability, species richness or in interaction strength, driven by the traits of particular densities, and temperature) or by evolutionary factors herbivore species, in the absence of strong differences (traits of species that alter per capita effects or re- in diversity. In other cases, the reverse might be true: sponses to ecological factors), understanding geo- herbivore diversity or density might matter more than graphic variation in interaction strength may be a chal- per capita interaction strength. It would be interesting lenging task. Our understanding of geographical var- to conduct studies similar to ours but in higher-diver- iation in community ecology has grown to the point, sity systems. At this point, however, we simply are not however, that it is no longer satisfying either to un- yet capable of inferring community process from di- critically generalize from single studies or to treat each versity patterns alone at the geographic spatial scale. study as unique. Rather, we must seek to develop an Even at the local scale, interaction strength cannot be explicit understanding of how ecological interactions inferred from a knowledge of feeding ecology or abun- vary geographically, one that incorporates not only dance, but must be measured directly (Paine 1980, geographic mosaics (Thompson 1994, 1999) but also 1992). Thus, forging a robust link between biogeog- predictable large-scale clines such as those discussed raphy and community ecology will require studies that here and that will allow us to integrate results obtained explicitly measure interaction strength at a geographic from different parts of the world. Developing and test- scale. ing such an understanding will be a critical step towards Because plants make up so much of the world's bio- more robust generalizations in ecology. mass and herbivores so much of the world's biodiver- sity, understanding interactions between plants and her- ACKNOWLEDGMENTS bivores is central to understanding both ecosystems and S. C. Pennings thanks NSF and the Georgia Coastal Eco- community ecology. Interactions between herbivores systems LTER (DEB-0296160, GCE-LTER OCE 99-82133) and plants, however, have a wide variety of mecha- and B. R. Silliman thanks NSF (dissertation improvement nisms and outcomes. In order to make sense of the rich grant and BIO OCE grant to M. D. Bertness), EPA (STAR diversity of interactions between plants and herbivores, Fellowship), and NOAA (NERR fellowship) for funding. We we must understand the variables that mediate the na- thank M. Allan, T. Buck, N. DaveÂ, K. Geyer, C.-K. Ho, C. McFarlin, and C. Salgado for assistance in the ®eld, and Brad- ture and strength of these interactions. Meta-analyses ford Hawkins, Chuan-Kai Ho, Eric Post, and anonymous re- comparing different types of herbivores and different viewers for helpful comments on the manuscript. This is habitats have shown that herbivore type and habitat UGAMI contribution number 947. Special Feature alce,F,adP .Pthr 1999. Pitcher. A. P. and F., Barlocher, 2318 etes .D,adE rsoz 95 ouaindynamics Population 1981. 1985. Grosholz. E. Levings. and D., C. M. Bertness, S. and Garrity, D. S. D., M. Bertness, and Latitudinal 2002. Ewanchuk. J. P. and D., M. Bertness, of Determinants 1987. Ellison. M. A. and D., M. Bertness, etes .D,adR alwy 94 oiieinteractions Positive 1994. Callaway. R. and D., modi®cation M. Bertness, community and Habitat 1984. D. M. Bertness, rnn . .J al .E a,adW eia.19.Are 1997. Fenical. W. and Hay, E. M. Paul, J. V. G., Cronin, latitudinal A 1998. Roughgarden. J. and R., S. Connolly, A 2001. Roughgarden. J. and Menge, A. B. R., S. Connolly, ahe,C . n .R tog 97 eue herbivore Reduced 1997. Strong. R. D. and C., C. Daehler, oe,P . n .M ie 91 oprsno herbivory of Comparison 1991. plant Aide. M. in T. and production D., of P. Coley, fate the in among Patterns interactions 1999. Positive J. Cebrian, 2002. al. et M., R. Callaway, plants Invasive 2000. Aschehoug. T. E. and M., the R. Callaway, Macroecology: 1989. Maurer. A. B. and Chicago H., of J. University Brown, Macroecology. 1995. H. J. Brown, better plants tropical Are 1996. Hay. E. M. and C., R. Bolser, would different How 1998. Marvier. A. M. and S., D. Bigger, Consumer 1987. Ellison. M. A. and Wise, C. D., M. Bertness, driven Consumer 1992. Shumway. W. S. and D., M. Bertness, ntoNwBusiksl ase.I.Ptnilueby use Potential II. marshes. salt saxatalis Brunswick Littorina New two in ologia fterbe mussel, ribbed the of Evolution tropical±tem- comparison. a perate foraging: gastropod and pressure Predation Oec- marshes. salt bi- England of New nature ologia in and interactions strength the ological in community. variation climate-driven plant marsh salt England New Monographs Ecological a in pattern 193. ncmuiis rnsi clg n Evolution and Ecology in Trends communities. in Ecology snail. herbivorous introduced an by oe osaedceia eess iepni metab- Diterpenoid herbi- from defenses? temperate chemical olites than seaweed to resistant vores more herbivores tropical ma- Naturalist of American theory. synthesis community based rine oceanographically an for evidence structure: community intertidal Paci®c northeast in gradient Ecology Ocean. Paci®c northeast the in invertebrates intertidal of recruitment in gradient latitudinal USA. York, New York, New Sons, and Plant±animal Wiley John editors. regions. temperate and tropical Benson, in W. ecology evolutionary W. interactions: and Fernandes, W. eitnei nitoue mohcrgas( cordgrass smooth introduced an in resistance of Journal urchins. and Ecology ®shes Chemical temperate versus tropical n ln eessi eprt n rpclbroad-leaved tropical 25±49 and Pages temperate forests. in defenses plant and Naturalist American communities. Nature stress. with increase plants alpine exotic for mechanism Science a neighbors: invasion. old and new their versus continents. on species among space and Science food of division USA. Illinois, Chicago, Press, trop- vs. Ecology temperate seaweeds. of ical defenses and Palatability defended? in generality for search A Biology be? Integrative ecology. herbivory without world a com- plant perennial marsh salt a Oecologia in munity. set seed grasses. and pressure marsh in production Botany of seed Journal of American limitation pollen 204. eet fa grgtddsrbto.Oecologia distribution. aggregated an of bene®ts terni¯ora 110 132 fe etr fhrioefe rwh Oec- growth. herbivore-free of century a after ) 243 :99±108. :392±401. itoaacutiloba Dictyota :1145±1150. ultno aieScience Marine of Bulletin . 290 23 ITERATURE 71 L in :289±302. :190±200. :521±523. ekni demissa Geukensia 57 .W rc,T .Lwnon G. Lewinsohn, M. T. Price, W. P. 77 :129±147. :2269±2286. 79 35 1 :60±67. C TVNC ENNSADBINR SILLIMAN R. BRIAN AND PENNINGS C. STEVEN :288±293. :995±1007. sfeigdtret for deterrents feeding as ITED 154 priaalterni¯ora Spartina :449±468. 82 417 :1799±1813. h ot and costs the : 151 65 64 :844±848. priaal- Spartina :370±381. :307±313. :311±326. 67 9 :191± :192± en,R .17.Cmaio fteasmlgso sap- of assemblages the of Comparison 1977. F. R. Nostrand Denno, Van marsh. tidal the of Animals 1982. C. F. Daiber, aly .D .20.Seisiprac n otx:spa- context: and importance Species 2003. G. D. C. Harley, of consequences Biological 1985. N. En- M. Hardwick-Witman, 2004. Pennings. C. S. and Ho, C.-K. E., C. Goranson, process and Pattern 2000. Blackburn. M. T. and J., K. Gaston, approach uni®ed A 1982. Lubchenco. J. and D., S. Gaines, pre- in variation latitudinal and Local 1984. H. M. Fawcett, factors abiotic of role The 1991. Travis. J. and A., W. Dunson, ago,C . n .I .Nwl.19.Uiiainof Utilization 1990. Newell. E. I. R. and J., C. Langdon, tutrlydfeetsl as rse ntegnsSpar- genus the Entomology in Environmental two grasses tina. marsh inhabiting salt different (Homoptera-Hemiptera) structurally insects feeding USA. York, New York, New Reinold, iladtmoa aito nseisitrcin.Pages interactions. species in variation 44±68 temporal and tial Ecology and Biology Journal Marine community. marsh Experimental salt of England New a in rafting ice Oecologia marshes. salt in preferences coastal feeding herbivore and gradients vironmental UK. Oxford, Science, Blackwell macroecology. in Systematics and Ecology Biogeography. of II. Review interactions. Annual plant±herbivore marine to Ecology snail. marine 1230. herbivorous an on dation Naturalist American organization. 1067±1091. community in asmn .A 92 otiuint h clg fthe of ecology the to contribution A 1932. A. S. Hausemen, ree,D . .J ago,adR .E eel 1988. Newell. E. I. R. and Langdon, J. C. A., D. 1990. Kreeger, Jefferies. L. R. and ant Kotanen, of M. rates P. in H., gradient R. latitudinal Kerbes, A 1979. L. R. Jeanne, eg,B .20.Teoerdn motneo environ- of importance overriding The 2003. A. B. Menge, in patterns ecology: Geographical 1972. H. R. MacArthur, Bio- 2002. editors. Inchausti, P. and sur- Naeem, on S. marsh M., salt Loreau, a in location of In¯uence 1989. J. Lin, ae,A . .B .Aeeo n .Prrde 1997. Partridge. L. and Azevedo, R. B. R. C., A. ®shes. James, herbivorous marine of Biology 1989. H. di- latitudinal M. the Horn, of generality the On 2004. H. Hillebrand, priaalterni¯ora Spartina melanogaster Drosophila ertsadbcei sfo ore ytobvlesus- bivalve two by oyster sources the food pension-feeders, as bacteria and detritus atmrhsnail marsh salt USA. Jersey, New Princeton, triage. and Press, expendability University on Princeton perspectives species: of tance 146 tlzto frfatr ellsccro eie from derived carbon cellulosic refractory of Utilization a Journal Bay. geese: Hudson Ecology of snow Applied coast of west lesser the by on species habitats keystone wetland of Destruction Ecology predation. 58 eineprmns ae 16±43 species-de- of Pages outcome experiments. the letion determining in context mental York, New Row, and USA. Harper York, species. New of distribution the per- UK. Oxford, and Press, synthesis University Oxford functioning: spectives. ecosystem and diversity Se- Progress Ecology Marine ries mussels. ribbed of vivorship xedblt n rae rneo nvriyPress, University Princeton USA. Jersey, triage. New Princeton, on and perspectives species: of expendability importance The editors. Levin, uralist eei n niomna epne otmeaueof temperature to responses environmental and Genetic Reviews Annual Biology 167±272. Marine and Oceanography Naturalist American gradient. versity missa mussel :299±310. :881±890. 56 aieEooyPors Series Progress Ecology Marine . in ekni demissa Geukensia :105±110. 66 .KriaadS .Lvn dtr.Teimpor- The editors. Levin, A. S. and Kareiva P. :541±545. eapsbidentatus Melampus 60 27 yterbe mussel ribbed the by :1211±1224. :242±258. rmalttdnlcie Genetics cline. latitudinal a from aieEooyPors Series Progress Ecology Marine . rsote virginica Crassostrea 140 6 :359±372. :591±600. in clg,Vl 6 o 9 No. 86, Vol. Ecology, .KriaadS A. S. and Kareiva P. 163 a.Aeia Nat- American Say. 42 :171±179. 13 :192±211. 87 ekni de- Geukensia :111±138. :283±298. 65 :1214± 138 n the and : 27 : September 2005 LATITUDINAL GRADIENTS 2319

Menge, B. A., and J. Lubchenco. 1981. Community orga- Silliman, B. R., and J. C. Zieman. 2001. Top-down control nization in temperate and tropical rocky intertidal habitats: of Spartina alterni¯ora production by periwinkle grazing prey refuges in relation to consumer pressure gradients. in a Virginia salt marsh. Ecology 82:2830±2845. Ecological Monographs 51:429±450. Siska, E. L., S. C. Pennings, T. L. Buck, and M. D. Hanisak. Paine, R. T. 1980. Food webs: linkage, interaction strength 2002. Latitudinal variation in palatability of salt-marsh and community infrastructure. Journal of Animal Ecology plants: which traits are responsible? Ecology 83:3369± 49:667±685. 3381. Paine, R. T. 1992. Food-web analysis through ®eld mea- Smalley, A. E. 1960. Energy ¯ow of a salt marsh grasshopper surement of per capita interaction strength. Nature 355:73± population. Ecology 41:785±790. 75. Sotka, E. E., and M. E. Hay. 2002. Geographic variation Pennings, S. C., and M. D. Bertness. 1999. Using latitudinal among herbivore populations in tolerance for a chemically variation to examine effects of climate on coastal salt marsh rich seaweed. Ecology 83:2721±2735. pattern and process. Current Topics in Wetland Biogeo- Srivastava, D. S., and R. L. Jefferies. 1996. A positive feed- chemistry 3:100±111. back: herbivory, plant growth, salinity, and the deserti®- Pennings, S. C., and M. D. Bertness. 2001. Salt marsh com- cation of an Arctic salt-marsh. Journal of Ecology 84:31± munities. Pages 289±316 in M. D. Bertness, S. D. Gaines, 42. and M. E. Hay, editors. Marine community ecology. Sin- Stachowicz, J. J., and M. E. Hay. 2000. Geographic variation auer, Sunderland, Massachusetts, USA. in camou¯age specialization by a decorator crab. American Pennings, S. C., and C. L. Richards. 1998. Effects of wrack Naturalist 156:59±71. burial in salt-stressed habitats: Batis maritima in a south- Stiling, P. D., and D. R. Strong. 1984. Weak competition west Atlantic salt marsh. Ecography 21:630±638. among Spartina stem borers by means of murder. Ecology Pennings, S. C., E. R. Selig, L. T. Houser, and M. D. Bertness. 64:770±778. 2003. Geographic variation in positive and negative inter- Stiven, A. E., and S. A. Gardner. 1992. Population processes actions among salt marsh plants. Ecology 84:1527±1538. in the ribbed mussel Geukensia demissa (Dillwyn) in a Pennings, S. C., E. L. Siska, and M. D. Bertness. 2001. Lat- North Carolina salt marsh tidal gradient: spatial pattern, itudinal differences in plant palatability in Atlantic Coast predation, growth and mortality. Journal of Experimental salt marshes. Ecology 82:1344±1359. Marine Biology and Ecology 160:81±102. Pfeiffer, W. J., and R. G. Wiegert. 1981. Grazers on Spartina Stiven, A. E., and E. J. Kuenzler. 1979. The response of two and their predators. Pages 87±112 in L. R. Pomeroy and salt marsh molluscs, Littorina irrorata and Geukensia de-

R. G. Wiegert, editors. The ecology of a salt marsh. Spring- missa, to ®eld manipulations of density and Spartina litter. Feature Special er-Verlag, New York, New York, USA. Ecological Monographs 49:151±171. Post, E. 2005. Large-scale spatial gradients in herbivore pop- Swihart, R. K., J. P. Bryant, and L. Newton. 1994. Latitudinal ulation dynamics. Ecology 86:2320±2328. patterns in consumption of woody plants by snowshoe hares Red®eld, A. C. 1972. Development of a New England salt in the eastern United States. Oikos 70:427±434. marsh. Ecological Monographs 42:201±237. Targett, N. M., L. D. Coen, A. A. Boettcher, and C. E. Tanner. Reich, P. B., and J. Oleksyn. 2004. Global patterns of plant 1992. Biogeographic comparisons of marine algal poly- leaf N and P in relation to temperature and latitude. Pro- phenolics: evidence against a latitudinal trend. Oecologia ceedings of the National Academy of Science, USA 101: 89:464±470. 11 001±11 006. Thompson, J. N. 1988. Variation in interspeci®c interactions. Richards, C. L., S. C. Pennings, and L. A. Donovan. 2005. Annual Review of Ecology and Systematics 19:65±87. Community-wide patterns of phenotypic variation in salt Thompson, J. N. 1994. The coevolutionary process. The Uni- marsh plants. Plant Ecology 176:263±273. versity of Chicago Press, Chicago, Illinois, USA. Rietsma, C. S., I. Valiela, and R. Buchsbaum. 1988. Detrital Thompson, J. N. 1999. Speci®c hypotheses on the geographic chemistry, growth, and food choice in the salt-marsh snail mosaic of coevolution. American Naturalist 153:S1±S14. (Melampus bidentatus). Ecology 69:261±266. Thompson, J. N., and B. M. Cunningham. 2002. Geographic Salgado, C. S., and S. C. Pennings. 2005. Latitudinal vari- structure and dynamics of coevolutionary selection. Nature ation in palatability of salt-marsh plants: Are differences 417:735±738. constitutive? Ecology 86:1571±1579. Thompson, L. S. 1984. Comparison of the diets of the tidal Sanford, E. 1999. Regulation of keystone predation by small marsh snail, Melampus bidentatus, and the amphipod, Or- changes in ocean temperature. Science 283:2095±2097. chestia grillus. Nautilus 98:44±53. Sanford, E., M. S. Roth, G. C. Johns, J. P. Wares, and G. N. Travis, J. 1996. The signi®cance of geographical variation Somero. 2003. Local selection and latitudinal variation in in species interactions. American Naturalist 148:S1±S8. a marine predator±prey interaction. Science 300:1135± Travis, S. E., C. E. Prof®tt, R. C. Lowenfeld, and T. W. Mitch- 1137. ell. 2002. A comparative assessment of genetic diversity Seliskar, D. M., J. L. Gallagher, D. M. Burdick, and L. A. among differently-aged populations of Spartina alterni¯ora Mutz. 2002. The regulation of ecosystem functions by eco- on restored versus natural wetlands. Restoration Ecology typic variation in a dominant plant: a Spartina alterni¯ora 10:37±42. salt-marsh case study. Journal of Ecology 90:1±11. Valiela, I., J. M. Teal, and W. G. Deuser. 1978. The nature Silliman, B. R., and M. D. Bertness. 2002. A trophic cascade of growth forms in the salt marsh grass Spartina alterni- regulates salt marsh primary production. Proceedings of ¯ora. American Naturalist 112:461±470. the National Academy of Science, USA 99:10 500±10 505. Vermeij, G. J. 1978. Biogeography and adaptation. Harvard Silliman, B. R., and A. Bortolus. 2003. Underestimation of University Press, Cambridge, Massachusetts, USA. Spartina alterni¯ora production in western Atlantic salt Vermeij, G. J. 1987. Evolution and escalation: an ecological marshes. Oikos 143:549±555. history of life. Princeton University Press, Princeton, New Silliman, B. R., and S. Y. Newell. 2003. Fungal farming in Jersey, USA. a snail. Proceedings of the National Academy of Science, Wright, I. J., et al. 2004. The worldwide leaf economics USA 100:15 643±15 648. spectrum. Nature 428:821±827. Ecology, 86(9), 2005, pp. 2320±2328 ᭧ 2005 by the Ecological Society of America

LARGE-SCALE SPATIAL GRADIENTS IN HERBIVORE POPULATION DYNAMICS

ERIC POST1 Department of Biology, Penn State University, 208 Mueller Laboratory, University Park, Pennsylvania 16802 USA

Abstract. Spatial gradients in density dependence and cyclicity are familiar features of the population dynamics of small mammals, particularly Fennoscandian rodents. The most well-documented of such gradients is a weakening of direct density dependence and an increase in the tendency of populations to cycle the farther north they occur, a phenom- enon that has been attributed to gradients in predation and seasonality. Among large mam- mals, however, for which evidence of cyclicity is less clear, geographic gradients in pop- ulation dynamics are limited to spatial variation in the strength of density independence. The population dynamics of caribou and muskoxen in Greenland, for example, display latitudinal gradients in the response of populations to large-scale climatic ¯uctuation. To my knowledge, the existence of spatial gradients in density dependence has not been ex- plicitly investigated in large mammals. Here I present an analysis of the dynamics of 27 populations of caribou and reindeer in Greenland, Finland, and Russia, spanning 21 degrees of latitude (51.7Њ±72.7Њ N) and 215 degrees of longitude (56.4Њ W±159.5Њ E), to identify spatial gradients in density dependence and independence. Results of autoregressive time series analysis show a clear gradient in the strength of direct density dependence exhibited by these populations that declines from southern to northern latitudes. Although this pattern mirrors the latitudinal gradient evident in Fen- noscandian rodent dynamics, an analysis of the dimensionality of these time series suggests that few, if any, of the populations are limited by predators. The existence of an inverse latitudinal gradient in the magnitude of the in¯uence of large-scale climate on the dynamics of these populations suggests there may be a tension in the strength of density-dependent vs. density-independent limitation experienced by them. Key words: caribou; density dependence; density independence; latitude; longitude; Rangifer tarandus; reindeer.

Special Feature INTRODUCTION dynamics that was comparable in authority to the den- sity-dependent models of Lotka (1925) and Nicholson The earliest body of ecological theory concerning and Bailey (1935). animal population dynamics focused on competition as An elegantly unifying treatment of the roles of den- the primary, if not sole, factor regulating populations sity dependence and independence in population lim- (Elton 1927, Nicholson 1933). The exemplar of pop- itation has emerged over the past decade in the study ulation limitation by intraspeci®c competition for food of spatial gradients in population dynamics. Although (Lack 1954) persisted nearly unchallenged for decades, until the publication of a paper demonstrating the in- many examples exist, by far the most well-studied and ¯uence of temperature on population growth in the rice familiar gradients in population dynamics are those ev- weevil (Calandra oryzae) (Birch 1948) signaled a ident in the spatial heterogeneity of the small mammal change in thinking among ecologists about density de- cycles in northern environments. In Europe, for in- pendence and population dynamics. Birch's paper was stance, there is a clear tendency for the degree of cy- followed that year, in the same journal and volume, by clicity displayed by populations of rodents to increase a seminal paper quantifying the population dynamics from southern to northern latitudes (Hansson and Hent- of apple blossom thrips (Thrips imagines) solely in tonen 1985). As well, the dynamics of small mammals terms of density-independent, environmental factors in Fennoscandia display an increase in the length of (Davidson and Andrewartha 1948). With the subse- the population cycle with increasing latitude (Hanski quent publication of the classic work by Andrewartha et al. 1991) that is associated with a weakening of the and Birch (1954), ecologists were provided a frame- strength of direct density dependence along the same work for quantifying density-independent population gradient from south to north (Bjùrnstad et al. 1995). Among the many hypotheses that have been pro- posed to explain the latitudinal gradient in density de- Manuscript received 18 May 2004; revised and accepted 26 pendence and cyclicity apparent in the dynamics of July 2004. Corresponding Editor: A. A. Agrawal. For reprints of this Special Feature, see footnote 1, p. 2261. small mammals in northern Europe, the two most com- 1 E-mail: [email protected] monly invoked are the ``generalist predator'' hypoth- 2320 September 2005 LATITUDINAL GRADIENTS 2321 esis (Hanski et al. 1991) and the ``seasonality'' or ``snow cover'' hypotheses (Hansson and Henttonen 1985). According to the former, the abundance of gen- eralist predators at southern latitudes and the domi- nance of specialist predators at northern latitudes ex- plain the increase in cyclicity and the weakening of direct density dependence as one moves from south to north (Hanski et al. 1991, Bjùrnstad et al. 1995, Hanski and KorpimaÈki 1995). By contrast, the latter hypoth- eses predict that the greater persistence of snow cover at northern latitudes limits the vulnerability of small mammals to predation, thereby producing dynamics that are dominated more by reproduction (and, hence, delayed density dependence) than by mortality due to predation (and, hence, direct density dependence) (Hansson and Henttonen 1985, Steen 1995). Although the Fennoscandian rodent cycles have been by far the most closely scrutinized (Stenseth 1999), they constitute by no means the only such examples of latitudinal gradients in density dependence and cyclic- ity. On the Japanese island of Hokkaido, for example, the dynamics of the grey-sided vole (Clethrionomys rufocanus) are characterized by a transition from sta- bility in the southwestern populations to cyclicity in the northeastern populations (Saitoh et al. 1998). No- FIG. 1. (a) Hypothetical tension between the strength of Feature Special density dependence and density independence in population tably, however, in contrast to the pattern seen in Fen- dynamics. Populations limited mainly by abiotic, environ- noscandian rodent cycles, the gradient on Hokkaido is mental, or other density-independent factors may be pre- due to a cline in delayed rather than direct density vented from reaching densities at which density-dependent dependence (Stenseth et al. 1996, Saitoh et al. 1998) factors limit population growth. Conversely, populations un- der favorable environmental conditions may be limited more that has been attributed to spatial variation in the length by density-dependent processes. If these conditions hold, then of winter across the island (Stenseth et al. 1998). spatial gradients in population dynamics may be attributable Despite clear evidence for spatial gradients in the to similar gradients in biotic interactions or abiotic conditions. strength of density dependence and cyclic behavior of For example, (b) gradients in population dynamics attribut- rodent dynamics, similar patterns have not, to my able to spatial variation in the strength or importance of biotic interactions should be discernible in latitudinal and/or lon- knowledge, been explicitly investigated or clearly doc- gitudinal gradients in the order of density dependence (solid umented in large mammals. This may lead some to line) but not in the order of density independence (dashed question whether there is something inherently differ- line). Conversely, (c) gradients in population dynamics at- ent about the biology, ecology, and dynamics of small tributable to spatial variation in abiotic conditions in¯uencing dynamics should be characterized by latitudinal and/or lon- vs. large mammals that would explain the presence of gitudinal variation in the order of density independence strong dynamical gradients in rodents but their absence (dashed line) but not density dependence (solid line). in, for example, ruminants (Caughley and Krebs 1983, Sñther 1997). If so, are the hypotheses developed in the study of rodent population dynamics unsuited to dynamics may be visualized as a continuum along the study of the dynamics of large mammals? It has which the relative in¯uence of each varies inversely been suggested, for example, that large mammal pop- with that of the other (Fig. 1a). According to this hy- ulations may be limited more by extrinsic than by in- pothesis, populations inhabiting favorable environ- trinsic factors, while the reverse may often be the case ments with, for instance, mild or relatively stable for small mammals (Caughley and Krebs 1983). In weather conditions may be more likely to increase to northern environments, in particular, harsh environ- the carrying capacity of the environment and experi- mental conditions and extreme weather during winter ence density-dependent regulation than would popu- may result in stronger limitation of large herbivore pop- lations limited by abiotic conditions in harsh environ- ulations by density-independent than by density-de- ments (Elliott 1987, Albon and Clutton-Brock 1988). pendent factors (Albon and Clutton-Brock 1988), Such a tension would mean that spatial gradients in though the two are often likely to operate in concert the dynamics of large herbivores, if present, would be (Forchhammer et al. 1998, Milner et al. 1999, Post and dif®cult to attribute to biological interactions vs. spatial Stenseth 1999). variation in abiotic conditions. Inferences might be The potential for tension between density-indepen- drawn, however, on the basis of the signatures that dent and density-dependent in¯uences on population biotic and abiotic in¯uences leave on the autoregressive Special Feature h ouaindnmc fcrbu( caribou of dynamics on population 1995) Hurrell the (NAO; system, Oscillation climate Atlantic large-scale North a the of in¯uence the of mag- herbi- The nitude recently. large documented been of indeed, dynamics has, vores the in independence density 1c). (Fig. dependence density of such order of the lack in the gradient but a and dependence density strength in of gradients strength the inverse the and both independence in density of be gradients order would spatial conditions re- in abiotic the evident to in populations variation gra- of to spatial sponse attributable contrast, dynamics In in 1b). dients den- (Fig. of order independence the sity in gradient such density no of but strength independence, and, the dependence, in density gradients of inverse potentially, order both the in and gradients strength for the longitudinal in, evident or be latitudinal would to example, dependence attributable density dynamics in in density clines gradients of ex- spatial can order that we the pect de®nitions, as these of Applying thought independence. population be in¯uence may conditions of dynamics abiotic number which the at Similarly, (Royama 2001). lags repro- al. dynamics et in Bjùrnstad population lags 1992, time at in¯uencing or time interactions duction in trophic com- backward of the with plexity steps increases correlated, of are densities number which depen- the density of or lag) dence, temporal (i.e., order Forchhammer The and 2001). (Post data series time the of structure 2322 ta.20) nlzdtepplto yaisof dynamics population the analyzed I 2002), al. et atr usa(i.2b). (Fig. Russia eastern ogtd (56.4 longitude re flttd (51.7 latitude of grees ouain fcrbu eidmsi ener and reindeer, (all reindeer semi-domestic wild caribou, 27 of of dynamics large the populations of on focuses dynamics analysis population My herbivores. the in large-scale for gradients potential spatial the investigate Forch- to attempt of (2002) an Forchhammer analyses in and Post the and (2002) upon al. expand et hammer I spa- Here same popu- the scale. over the tial herbivores in large of gradients dynamics similar lation con- is for there potential 2a), Northern Fig. siderable 2003; the al. et throughout (Hurrell Hemisphere gradients the of longitudinal in¯uence latitudinal both and the along conditions of weather local strength on tremen- NAO the the in considering Forch- variation and 2002, dous 2004), al. Post et Walther and 2001, hammer al. spe- et numerous of (Ottersen dynamics cies population of the role in a NAO for the evidence widespread Forchhammer given and Hence, Post 2004). 2002, al. spe- et each (Forchhammer in cies populations the among latitude with varies dus h xsec fsailgainsi h teghof strength the in gradients spatial of existence The olwn ehd ecie ale (Forchhammer earlier described methods Following n ukxn( muskoxen and ) ouaindt n iesre analysis series time and data Population M Њ TRA AND ATERIAL W±159.5 agfrtarandus Rangifer vbsmoschatus Ovibos Њ ±72.7 Њ ) rmws reln to Greenland west from E), Њ M )ad25dgesof degrees 215 and N) ETHODS pnig2 de- 21 spanning ) agfrtaran- Rangifer nGreenland in ) RCPOST ERIC utn n aePeev ftefre SR(Syroech- USSR former the 1995). of of Administration kovskii Preserve Main Game California, the and by Redlands, denoted Hunting regions Institute, the (Environ- of Research USA), 3.0 version Systems ArcView mental using located represent dots centroids, the populations, the Russian the Ten- For Knoxville, Tennessee, USA). nessee, of University Ecol- Evolution, of Synthesis, Department and the and ogy and Analysis USA; California, Ecological Barbara, for Santa UK; Center Ascot, Research Biology, National Environment Population the for Natural Centre the [NERC] Data- of Council Dynamics collaboration Population (a populations Global base Finnish the the from of extracted Ap- Locations were (b) paper. this [2003]). reindeer in al. and caribou analyzed the of et populations for the Hurrell index of NAO locations through also proximate the (see in December change 1900±2003 months unit period a the to degrees response for of in March tenths winter (in during the temperatures of Celsius) surface magnitudes mean the in are changes Shown Hemisphere. throughout Northern temperatures the local on (NAO) Oscillation Atlantic F IG .()Sailpten ftei¯ec fteNorth the of in¯uence the of patterns Spatial (a) 2. . clg,Vl 6 o 9 No. 86, Vol. Ecology, September 2005 LATITUDINAL GRADIENTS 2323 caribou and reindeer in seven populations in Greenland focal population is as follows (Forchhammer et al. (Post and Forchhammer 2002), four populations in Fin- 2002): land, and 16 populations in Russia (Syroechkovskii iϭdjϭk 1995; Fig. 2b). To my knowledge, this comprises the ϭ ␤ϩ͸͸ ␤ ϩ␻ NttN Ϫ10exp΂΃itX ϪijtU Ϫj (1) most spatially extensive analysis of the population dy- 11 namics of a species of large mammal. Data from the in which abundance at time t is represented by N , and Greenlandic populations consist of numbers of caribou t in which X ϭ log (N ), U represents the in¯uence of shot annually from 1908 through 1981 in each popu- t e t t-j climate, and ␤ is the intrinsic growth rate of the pop- lation (Meldgaard 1986), before hunting quotas were 0 ulation (Forchhammer et al. 1998, Post and Stenseth introduced there (Linnell et al. 2000). As such, these 1999). After taking the natural logarithm of both sides data do not constitute actual counts of individuals in of Eq. 1, a generalized autoregressive model with the populations, but interannual variation in the num- summed climatic components is obtained (Forchham- bers shot have not been determined or constrained by mer et al. 2002): quotas and thus are assumed to match interannual var- ϭ␤ ϩ ϩ␤ ϩ␤ ϩ␤ ϩ iation in actual population sizes (Forchhammer et al. Xt 01(1 )XtϪ12XtϪ23XtϪ3 ´´´ 2002). Moreover, our previous analyses of these data, jϭk using a subset of the time series for which independent ϩ␤ ϩ␻͸ ϩ␧ dtX ϪdjtU Ϫjt (2) estimates of population size are available in the form 1 of aerial counts of caribou in two of the focal popu- ϩ␤ ␤ in which the coef®cients (1 1) and 2 quantify direct lations, have revealed a positive, linear correlation be- ␻ and delayed density dependence, respectively, j quan- tween numbers of caribou shot and population counts ti®es the direct and/or delayed climatic in¯uences on (Post and Forchhammer 2004). The Finnish popula- ␧ population dynamics, and t represents the remaining tions comprise semi-domesticated reindeer (Helle and variance not quanti®ed by the model parameters.

Kojola 1994), while the Russian populations are wild It is important to note that Eq. 2 is a candidate model Feature Special reindeer (Syroechkovskii 1995). Data from the Finnish specifying all potential direct and delayed density-de- and Russian populations represent population estimates pendent and climatic in¯uences on population dynam- based on actual counts of animals (Helle and Kojola ics, but not necessarily the form of the most parsi- 1994, Syroechkovskii 1995). The Finnish populations monious AR model for any particular population. Be- have been counted annually during roundups in Sep- cause the strength of trophic interactions and, hence, tember±January each year, and variation in accuracy of direct and delayed density dependence, may vary along the counts among years is reportedly small (Helle and large-scale spatial gradients (Pennings and Silliman Kojola 1993). Data on annual population densities of 2005), the form of the most parsimonious model will all four Finnish populations cover the period 1968± likely vary among the focal populations. 1990. Annual counts of wild reindeer in the Russian Applying the model in Eq. 2 to the caribou data from populations span 1961±1984, but are continuous tem- West Greenland covering 1980±1981, the semi-do- porally only from 1975 to 1984. Although data were mestic reindeer data from Finland covering 1968±1990, available for an additional six populations of wild rein- and the wild reindeer data from Russia covering 1975± deer in Russia, I restricted my analysis of the Russian 1984, I identi®ed the most parsimonious (i.e., best ®t) populations to those for which annual estimates of pop- model for each population. The most parsimonious or- ulation size were temporally continuous from 1975 to der or dimension of each time series was identi®ed by 1984. Use of continuous data was necessary to avoid choosing the AR model with the lowest corrected Akai- spurious estimates of statistical density dependence in ke Information Criterion (AICc) score (Sakamoto et al. the autoregressive approach described below (Royama 1986) among all possible models for each population 1992). and values of d from 0 to 3, but no covariates. Because Using an established approach (Forchhammer et al. estimation of the AR coef®cients may be biased if the 1998, 2002), I applied autoregressive (AR) time series time series are not stationary (i.e., if they possess a analysis to identify or con®rm previous reports (Forch- signi®cant trend over time; Royama 1992), the term hammer et al. 2002, Post and Forchhammer 2002) of ``year'' was included in analyses of all time series with the best ®t AR models including the NAO as a covar- a signi®cant temporal trend (Forchhammer et al. 1998). iate. I used the same approach that has been applied After identifying the most parsimonious order of each recently in the development of models of plant±her- time series, further improvement of the ®t of each mod- bivore±climate (Forchhammer et al. 1998) and preda- el was then tested for by inclusion of a term quantifying tor±herbivore±plant±climate (Post and Forchhammer the in¯uence of the NAO on dynamics at lags of 0±3 2001) interactions that have focused on the dynamics yr; reduction of the AICc score of any model by a value of red deer (Cervus elaphus); and wolves (Canis lupus), of 1 or more was considered signi®cant improvement moose (Alces alces), and balsam ®r (Abies balsamaea), of the model (Bjùrnstad et al. 1995). None of the best- respectively. The general form of this model for any ®t models contained more than one NAO term, and the Special Feature 0.05) reported. the not are of did both and analysis data differ, Finnish the nontransformed from and coef®cients transformed AR trans- using Re- the comparison. sults for both states in nontransformed data and total formed the those of analyzed 13% I in observations, zeros produced series time Finnish time Russian the among 2.06 series. the and among series, 2.8 time series, was time Finnish models Greenlandic ®t the best the among in 2.6 parameters of number mean 2324 ihflCroain20) n nec tpo the ( of nonsigni®cant lowest step the (In- each with 6.0 in term the version and analysis S-plus 2001), in Corporation function sightful model linear the using estimated were variables predictor of population. Coef®cients each for den- independence of and dependence order sity potential the quantifying as terms included variables predictor I the above, to mentioned addition In coordinates elim- covariates. backward nonsigni®cant with of regression ination linear stepwise used populations, I these of dynamics dependence the density in of independence and strength the in gradients spatial 2002). al. et narrow of (Forchhammer be- area variation an 2002) to longitudinal con®ned Forchhammer are populations and the cause (Post lat- a gradient primarily along itudinal NAO local the in caribou, with whereas by vary the 2003), inhabited temperatures Greenland al. in West et of deviations area Hurrell the unit 2a; (Fig. to index response longitudinal NAO their and in latitudinal variation strong display tem- (Fig. local particular, peratures gradients in Russia, and both Finland Across along 2a). Hemisphere throughout Northern of temperatures local strength the on the NAO in the of variation effect clear the anal- is the variables. there in because included predictor were yses as longitude and populations latitude Both the co- of latitude/longitude with ordinates and models variables AR dependent statistical the as the with from regression the coef®cients linear indepen- respective using in density analyzed and were gradients dependence dence Spatial density of respectively. strength 2002, 2004), al. 1995) et Post al. (Forchhammer et independence density (Bjùrnstad and dependence density lagged of parsimonious highest most orders the the indicate terms while 1992, NAO and 2002), density (Royama al. dynamics et climatic Forchhammer population and dependence on density in¯uences direct statistical of civdcnann nysgicn ( signi®cant only containing achieved rlniuia rdet ntesrnt fdniyde- density of strength the in gradients longitudinal or composition models. the best-®t for- con®rm these using of to analyses regression these stepwise repeated ward I variables. dictor ial,bcuetentrllgtasomto fthe of transformation log natural the because Finally, ncnutn ttsia et ftesgicneof signi®cance the of tests statistical conducting In strength the quantify models AR the of Coef®cients odtriewehrayptnillttdnland/ latitudinal potential any whether determine To t au a lmntdutlardcdmdlwas model reduced a until eliminated was value pta analyses Spatial P Յ .5 pre- 0.05) P RCPOST ERIC Ͼ eso oh oee,nn fteGMrslswere results GAM the (all of none signi®cant however, both; or- of in the ders and gradients independence, nonlinear density dependence, signi®cant density potentially to for 1999) Tibshirani ad- test and generalized (Hastie (GAMs) using models repeated ditive were linear All analyses analysis. stepwise spatial forward a pre- likewise with were signi®cant con®rmed models These with achieved. was model only reduced dictors a regression predictor until linear nonsigni®cant variables indepen- of stepwise density elimination used backward of I with well order as the Here in dence. variation analysis the spatial in dependence a of density included of order likewise the for and as term variable independence density predictor of in- potential order I a the dependence, for density term of a order cluded the in spa- variation of analysis tial dependence the in density Additionally, of independence. latitude, and coef®cients variables: the predictor and following vari- longitude, dependent the the and as ables orders those dependence with quantifying regression density linear terms of using independence order density for the and tested I in Next gradients 1a). Fig. spatial (sensu populations 27 models all parsimonious for most den- the and from dependence independence density sity direct between of correlation popu- a coef®cients on for the tested in¯uences ®rst I abiotic dynamics, vs. variation lation spatial interactions to biotic attributable in be focal the might of dynamics populations the in independence or pendence ednedsrbn h yaiso h oa popu- focal the of ( inde- lations dynamics density the and describing dependence pendence density of coef®cients populations. eight pop- remaining nine for the 1 for populations, 2 10 and for ulations, direct) (i.e., ®t independence, 0 best density the was of entered order pop- NAO the the one quantifying which model, only at of lag dynamics The ulation. the de- third-order a described best and best them, model of model four second-order of dynamics a the scribed while models, autoregressive ®rst-order (AR) by described best were lyzed edne( pendence est eednears ouain nlddlati- included of ( populations strength tude the across in dependence variation density describing The 1995). model al. et best-®t (Bjùrnstad depen- latitudes density northern weaker at depen- and dence latitudes density southern stronger at indicating dence with 3b), positively (Fig. covaried latitude dependence density direct of fdniyidpnec ( independence density of omduigSpu eso 6.0. version S-plus using formed eeue nta frwdt,tebs-tmdlcon- model best-®t (latitude, the variables same data, data the raw tained Finnish of instead log-transformed used were natural When dictors. tegho est independence, density of strength h yaiso 2(1)o h ouain ana- populations the of (81%) 22 of dynamics The hr a la,ivrerltosi ewe the between relationship inverse clear, a was There t ϭ r ϭϪ 3.67, t ϭϪ 0.54, P P 2.91, values ϭ P Ͻ .0) tegho est inde- density of strength 0.001), P R .0;Fg a.Tecoef®cient The 3a). Fig. 0.005; ESULTS ϭ Ͼ t .0) n h dimension the and 0.008), ϭ .0.Aaye eeper- were Analyses 0.10). 2.36, clg,Vl 6 o 9 No. 86, Vol. Ecology, t ϭ P t 3.80, ϭ ϭϪ .3 spre- as 0.03) 2.96, P Ͻ 0.001; P ϭ September 2005 LATITUDINAL GRADIENTS 2325

general pattern of inverse correlation between the strength of density dependence and density indepen- dence in Greenland (r ϭϪ0.64, df ϭ 5, P Ͼ 0.10), Finland (r ϭϪ0.58, df ϭ 2, P Ͼ 0.20), and Russia (r ϭϪ0.55, df ϭ 14, P Ͻ 0.05), although with consid- erably lower degrees of freedom. Combining the Greenlandic and Finnish data reveals the same inverse association (r ϭϪ0.60, df ϭ 9, P ϭ 0.05). Furthermore, a generalized linear model with populations assigned to one of three categories (Greenland, Finland, Russia) revealed no difference among them in the strength of the correlation between density dependence and in- dependence (F ϭ 0.75, df ϭ 2, 24, P ϭ 0.48). In accordance with the patterns of spatial variation in the strength of the in¯uence of the NAO on local temperatures throughout the regions inhabited by the focal populations (Fig. 2a), the coef®cient of density independence varied along gradients of both latitude (r ϭϪ0.62, P Ͻ 0.001; Fig. 3c) and longitude (r ϭϪ0.64, P Ͻ 0.001; Fig. 3d). The best-®t model de- scribing variation in the strength of density indepen- dence across populations included longitude (t ϭϪ2.64, P ϭ 0.01), latitude (t ϭϪ2.59, P ϭ 0.02), and the dimension of density independence (t ϭϪ2.21, P ϭ 0.04) as predictors. When natural log-transformed Feature Special Finnish data were used instead of raw data, the best- ®t model contained the same variables (longitude, t ϭϪ2.40, P ϭ 0.03; latitude, t ϭϪ2.49, P ϭ 0.02; di- mension of density independence, t ϭϪ2.04, P ϭ 0.053). The dimension of density dependence did not vary signi®cantly with latitude (t ϭ 0.52, P ϭ 0.61; Fig. 3e), but did show weak variation with longitude (t ϭϪ2.30, P ϭ 0.03; Fig. 3f). The dimension of density independence displayed signi®cant variation with both latitude (r ϭϪ0.72, P Ͻ 0.001; Fig. 3g) and longitude (r ϭϪ0.66, P ϭ 0.01; Fig. 3h).

FIG. 3. Partial residual correlation plots from multivariate DISCUSSION regression analyses of (a) the strength of density dependence (DD) vs. density independence (DI) in the population dy- Although time series on the population ¯uctuations namics of caribou and reindeer in Fig. 2b, (b) density de- of lynx (Lynx canadensis) throughout the boreal forest pendence vs. latitude, (c) density independence vs. latitude, of Canada display large-scale spatial structure (Sten- (d) density independence vs. longitude (negative values in- seth et al. 1999), the analysis reported here constitutes, dicate longitudes west of the Prime Meridian), (e) order of density dependence vs. latitude, (f) order of density depen- to my knowledge, the ®rst documentation of spatial dence vs. longitude, (g) order of density independence vs. gradients of density dependence and independence in latitude, and (h) order of density independence vs. longitude. the population dynamics of large mammals. The lati- tudinal gradient in density dependence of caribou and reindeer dynamics documented here matches the pat- 0.007; dimension of density independence, t ϭ 2.53, tern reported for microtine rodents in Fennoscandia P ϭ 0.02). (Bjùrnstad et al. 1995): direct density dependence is Although these partial correlations are fairly strong, stronger among populations inhabiting southern lati- a certain degree of caution may be warranted because tudes than among those at more northern latitudes. the time series derive from three distinct sets of pop- While strikingly similar, the relationships generating ulations under different harvest and management re- this pattern in microtine rodents vs. caribou and rein- gimes. If the hypothesized tension in Fig. 1a is a gen- deer are likely distinct. Among Fennoscandian popu- eral phenomenon, however, it should be apparent in lations of microtine rodents, the weakening of density data from each of the three regions separately. Con- dependence from southern to northern populations ap- sidering these regions individually reveals the same pears to relate to a transition from limitation by gen- Special Feature eedne oal,hwvr ovs( wolves however, Notably, dependence. density of second-order one displayed only populations dynamics; Russian be higher-order 16 may detect reindeer to wild short quan- Russian too series of time dynamics the the note, tifying cautionary a On de- density pendence. direct au- ana- only ®rst-order containing populations by models, the described toregressive best of were Most however, 1992, 2001). lyzed, (Royama step al. time et one Bjùrnstad than more by densities among separated dependence au- or predation. higher-order processes by as toregressive apparent limited be are should populations limitation few, Such these that however, of indicates, any, if reindeer and caribou of Korpima and (Hanski cycles such contributing major to by a factor ¯uctuations is populations density the predator rodent specialist presumably, of 1995); tracking en- al. numerical southern et in not (Bjùrnstad but vironments environments cy- northern Fen- multi-annual in clear of cles undergo populations to for rodents tendency noscandian consistent increasing also an is hypothesis with populations This 1991). northern al. in et (Hanski predators limi- specialist to by populations tation southern among predators eralist 2326 h aiuia rdeti est eedneapparent dependence density in so, gradient If latitudinal limited 1987). the be (Elliott processes to population-intrinsic ca- densities by high carrying suf®ciently reaching and/or from pacity prevented density-independent be may by factors strongly limited populations orlto ossget swt e er( deer this red with relationship, as causal suggest, a does correlation for does it evidence While constitute 1a. not Fig. in depicted the support tension cross- to hypothetical appears the 3a) in (Fig. comparison evident population independence density and dence 2005). and (Pennings Silliman herbivores insect in- and systems gastropod inter- salt-marsh volving in trophic documented herbivore±plant been of has actions vari- strength spatial the Such in 1999). ation pop- 1968, density-dependent (Klein crashes experience ulation to likely there- less and beds fore lichen winter less fragile be overgraze instance, to for likely may, persistent where more latitudes is cover higher Cari- snow at forage. populations reindeer winter and of bou overexploitation buffer instead, may population predation; it in¯uence from refuge to providing likely by not how- dynamics is mammals, large cover of snow case and ever, the (Hansson In hy- dynamics 1985). cover rodent snow Henttonen the microtine with of more agree pothesis to appear document- here patterns ed the therefore, hypothesis, predator density in 2002). gradient al. et latitudinal (Forchhammer clear dependence a them- display populations Greenlandic selves the (Bùving yet 1997), Greenland Post West caribou and from limiting absent of are capable populations predators large other and phus yaayi ftedmninlt ftedynamics the of dimensionality the of analysis My h nes eainhpbtendniydepen- density between relationship inverse The specialist-generalist the with concurring than Rather nSoln AbnadCutnBok18) that 1988), Clutton-Brock and (Albon Scotland in ) k 1995). Èki ai lupus Canis evsela- Cervus RCPOST ERIC ) h ouaindnmc fahd ( aphids on of work dynamics population empirical the recent on in¯uences Notably, their populations. and local conditions abiotic spatial in re¯ect Hemisphere variation Northern the of reindeer and much caribou across of dynamics gradi- population density large-scale the hypothesis in the of ents the underlying order support mechanisms the the to that the in appear and gradients would independence clear dependence density such of of order lack the Furthermore, 3c). gra- in longitudinal (Fig. and dients case latitudinal strong the of be existence the to appears fact which independence, inverse in density an of to strength the attributable in be gradient actually may data these in ompltnseist uuecagsi lmt.The climate. in changes quasi- future any to species of hope cosmopolitan response can distribution-wide we the which to predict extent to the which (Fig. question NAO to into the calls extent in 2a) changes the with vary in temperatures Hemisphere local Northern the through- out variation of degree change? tremendous the climate Observing of consequences about indicate ecological here revealed potential patterns the 1995), might what (Ives so destabilizing considered is independence oetal tblzn o ouaindnmc fi is 1973 it (May if over-compensatory dynamics not population as for regarded stabilizing been potentially has dynamics. dependence population density in Classically, gradients spatial local for of forcing weather climatic of direction and magnitude both the in variation spatial large-scale of consequences the al. et Bjùrnstad 2001) 2000, and 2001). Ranta Boonstra feasible and and also (Karels (Kaitala are tractable process approaches experimental of where types signatures the both capture anal- to of insuf®cient series Stenseth be time may and although alone 2001), environments ysis Post al. 2001, et northern Coulson 1998, in 1999, al. mammals et large (Forchhammer those and other populations these of of dynamics the in work at are processes -independent Certainly, and other. purpose density-dependent the over both my position one and advocate to 2003), not is Turchin al. 1991, et Mech Messier 1954, 1987, Birch and nauseum popu- Andrewartha ad animal 1927, nearly (Elton debated in been factors, have abiotic dynamics lation and biotic of extrinsic or and dynamics intrinsic forces, of the roles relative in The species. dependence this in- im- density more it is than is independence portant nor density that population reindeer, argue main and to the caribou tended is for NAO factor the limiting that argue to tended may here example. gradients reported an results same such represent The dependence the indeed 2004). density al. along et of accordingly (Agrawal strength vary the should in- of gradients, rate spa- along the tial climate by which differentially in in¯uenced is species crease growth in population that of concluded of rate has strength the the and between dependence density relation inverse an umenting o oee,ws osiuaenwtikn about thinking new stimulate to wish however, do, I in- not is analysis this that emphasize to hasten I clg,Vl 6 o 9 No. 86, Vol. Ecology, a , pi nerii Aphis b ,wiedensity while ), doc- ) September 2005 LATITUDINAL GRADIENTS 2327 analyses reported here indicate, however, a tension be- Elton, C. 1927. Animal ecology. MacMillan, New York, New tween the two processes and suggest a potential for the York, USA. Forchhammer, M. C., T. H. Clutton-Brock, J. LindstroÈm, and climatic signal in local population dynamics to S. D. Albon. 2001. Climate and population density induce strengthen with future warming, as has been reported long-term cohort variation in a northern ungulate. Journal in the dynamics of cholera outbreaks (Rodo et al. of Animal Ecology 70:721±729. 2002). Such a strengthening might lead the dynamics Forchhammer, M. C., and E. Post. 2004. Using large-scale climate indices in climate change ecology studies. Popu- of these populations to be governed more by abiotic lation Ecology 46:1±12. than by biotic factors, a condition that may lead to Forchhammer, M. C., E. Post, N. C. Stenseth, and D. Boert- population instability (Ives 1995, Mueller and Joshi mann. 2002. Long-term responses in arctic ungulate dy- 2000). Whether the tension between density depen- namics to variation in climate and trophic processes. Pop- dence and independence hypothesized to operate in the ulation Ecology 44:113±120. Forchhammer, M. C., N. C. Stenseth, E. Post, and R. Lang- dynamics of large mammals also plays a role in spatial vatn. 1998. Population dynamics of Norwegian red deer: gradients in the dynamics of other animal species, in- density-dependence and climatic variation. Proceedings of cluding small mammals, will require at a minimum the Royal Society of London, series B 265:341±350. comparably large-scale analyses (Forchhammer and Hanski, I., L. Hansson, and H. Henttonen. 1991. Specialist predators, generalist predators, and the microtine rodent Post 2004). cycle. Journal of Animal Ecology 60:353±367. Hanski, L., and E. KorpimaÈki. 1995. Microtine rodent dy- ACKNOWLEDGMENTS namics in northern Europe: parameterized models for the I thank Jim Hurrell and Adam Phillips for constructing Fig. predator±prey interaction. Ecology 76:840±850. 2a and Danielle Garneau and Eric Long for constructing Fig. Hansson, L., and H. Henttonen. 1985. Gradients in density 2b. For comments that greatly improved an earlier version of variations of small rodents: the importance of latitude and this paper, I thank Angela Anders, Steve Pennings, and two snow cover. Oecologia 67:394±402. anonymous referees. As always, special thanks to Mads Hastie, T. J., and R. J. Tibshirani. 1999. Generalized additive Forchhammer for constructive comments; any chaff remain- models. Chapman and Hall, London, UK. ing is of my making, not his. Helle, T., and I. Kojola. 1993. Reproduction and mortality

of Finnish semi-domesticated reindeer in relation to density Feature Special LITERATURE CITED and management. Arctic 46:72±77. Helle, T., and I. Kojola. 1994. Body mass variation in semi- Agrawal, A. A., N. Underwood, and J. R. Stinchcombe. 2004. domesticated reindeer. Canadian Journal of Zoology 72: Intraspeci®c variation in the strength of density dependence 681±688. in aphid populations. Ecological Entomology 29:521±526. Hurrell, J. W. 1995. Decadal trends in the North Atlantic Albon, S. D., and T. H. Clutton-Brock. 1988. Climate and Oscillation: regional temperatures and precipitation. Sci- the population dynamics of red deer in Scotland. Pages 93± ence 269:676±679. 107 in M. B. Usher and D. B. A. Thompson, editors. Eco- Hurrell, J. W., Y. Kushnir, G. Ottersen, and M. Visbeck. 2003. logical change in the uplands. Blackwell Scienti®c, Oxford, The North Atlantic Oscillation: climatic signi®cance and UK. environmental impact. Geophysical Monograph 134. Andrewartha, H. G., and L. C. Birch. 1954. The distribution Insightful Corporation. 2001. S-plus version 6.0 for Win- and abundance of animals. University of Chicago Press, dows. Guide to statistics. Volume 1. Insightful Corporation, Chicago, Illinois, USA. Seattle, Washington, USA. Birch, L. C. 1948. The intrinsic rate of natural increase of Ives, A. R. 1995. Predicting the response of populations to an insect population. Journal of Animal Ecology 17:15± environmental change. Ecology 76:926±941. 26. Kaitala, V., and E. Ranta. 2001. Is the impact of environ- Bjùrnstad, O. N., W. Falck, and N. C. Stenseth. 1995. A mental noise visible in the dynamics of age-structured pop- geographic gradient in small rodent density ¯uctuations: a ulations? Proceedings of the Royal Society of London, se- statistical modelling approach. Proceedings of the Royal ries B 268:1769±1774. Society of London, series B 262:127±133. Karels, T. J., and R. Boonstra. 2000. Concurrent density de- Bjùrnstad, O. N., S. M. Sait, N. C. Stenseth, D. J. Thompson, pendence and independence in populations of arctic ground and M. Begon. 2001. The impact of specialized enemies squirrels. Nature 408:460±463. on the dimensionality of host dynamics. Nature 409:1001± Klein, D. R. 1968. The introduction, increase, and crash of 1006. reindeer on St. Matthew Island. Journal of Wildlife Man- Bùving, P. S., and E. Post. 1997. Vigilance and foraging agement 32:350±367. behaviour of female caribou in relation to predation risk. Klein, D. R. 1999. The roles of climate and insularity in Rangifer 17:55±63. establishment and persistence of Rangifer tarandus popu- Caughley, G., and C. J. Krebs. 1983. Are big mammals sim- lations in the high Arctic. Ecological Bulletins 47:96±104. ply little mammals writ large? Oecologia 59:7±17. Lack, D. L. 1954. The natural regulation of animal numbers. Coulson, T., E. A. Catchpole, S. D. Albon, B. J. T. Morgan, Clarendon, Oxford, UK. J. M. Pemberton, T. H. Clutton-Brock, M. J. Crawley, and Linnell, J. D. C., C. Cuyler, A. Loison, P. M. Lund, K. G. B. T. Grenfell. 2001. Age, sex, density, winter weather, Motzfeldt, T. Ingerslev, and A. Landa. 2000. The scienti®c and population crashes in Soay sheep. Science 292:1528± basis for managing the sustainable harvest of caribou and 1531. muskoxen in Greenland for the 21st century: an evaluation Davidson, J., and H. G. Andrewartha. 1948. The in¯uence and agenda. Technical Report number 34. Greenland In- of rainfall, evaporation, and atmospheric temperature on stitute of Natural Resources, Nuuk, Greenland. ¯uctuations in the size of a natural population of Thrips Lotka, A. J. 1925. Elements of physical biology. Williams imaginis. Journal of Animal Ecology 17:200±222. and Wilkins, New York, New York, USA. Elliott, J. M. 1987. Population regulation in contrasting pop- May, R. M. 1973a. Stability and complexity in model eco- ulations of trout Salmo trutta in two lake district streams. systems. Princeton University Press, Princeton, New Jer- Journal of Animal Ecology 56:83±98. sey, USA. Special Feature 2328 ot .20.Tm asi ersra n aieenviron- marine and terrestrial in lags Time 2004. E. Post, bioge- Linking 2005. Silliman. R. B. and C., S. Reid, Pennings, C. P. Post, E. Belgrano, A. Planque, of B. balance G., The Ottersen, 1935. Bailey. A. V. and J., A. Nicholson, ihlo,A .13.Teblneo nmlpopulations. animal of balance The 1933. J. A. pop- Nicholson, model in Stability 2000. Joshi. A. and D., L. Mueller, Estimating 1999. Albon. D. S. and Elston, A. D. M., J. Milner, regulating and limiting of signi®cance The 1991. F. Messier, caribouÐzoogeogra- Greenland The 1986. M. E. Meldgaard, R. and Peterson, O. R. McRoberts, E. R. D., L. Mech, ot . n .C ochme.20.Sailsynchrony Spatial 2004. Forchhammer. C. M. and E., Post, of Synchronization 2002. Forchhammer. C. M. and E., Post, in¯uence Pervasive 2001. Forchhammer. C. M. and E., Post, et.Pgs165±167 Pages ments. Ecology in strength. 2319. variation interaction latitudinal ecology: plant±herbivore community and ography North the of effects Ecological Oecologia Oscillation. 2001. Atlantic Stenseth. C. N. and Zoological the London of of Proceedings Society I. Part populations. animal ora fAia Ecology Animal of Journal Jer- New Princeton, USA. Press, sey, University Princeton ulations. sheep. Ecology Soay Animal of of ¯uc- survival Journal in climatic variation and interannual density to tuations population of contributions the Ecology deer. white-tailed Animal and of moose Journal of demography the on factors om Meddelelser Grùnland dynamics. population and taxonomy, phy, Ecology Animal populations of moose Journal 615±627. and snow. winters' deer previous of to Relationship 1987. Page. 650. eemnsi niomns mrcnNaturalist American environments. deterministic h ainlAaeyo cecs USA Sciences, of Academy of the National Proceedings with trend. the association climate in Hemisphere Northern increased recent has populations local of Nature climate. 420 large-scale by dynamics population animal ver- terrestrial a Ecology of BMC dynamics community. the tebrate in climate large-scale of UK. University Oxford Oxford, Atlantic. Press, North of the effects in Ecological variations editors. climate Belgrano, A. and Hurrell, W. a,R .1973 M. R. May, :168±171. 20 :1±88. b tblt nrnol utaigversus ¯uctuating randomly in Stability . 3 :551±598. in .C tneh .Otre,J. Ottersen, G. Stenseth, C. N. 2 68 60 :131±178. :1235±1247. :377±393. 128 1 :1±14. :5. 101 :9286±9290. 86 107 :2310± :621± RCPOST ERIC 56 : Rodo ath . .C tneh n .N jrsa.19.The 1998. Bjùrnstad. N. O. and Stenseth, C. N. T., Saitoh, popu- and stochasticity Environmental 1997. B.-E. Sñther, Chapman dynamics. population Analytical 1992. T. Royama, ot . n .C tneh 99 lmtcvraiiy plant variability, Climatic 1999. Stenseth. C. N. and E., Post, te,H 95 nagigtecue fdspernefrom disappearance of causes the Untangling 1995. H. Steen, Akaike 1986. Kitagawa. G. and Ishiguro, M. Y., Sakamoto, tneh .C,O .Bùntd n .Sio.19.Sea- 1998. Saitoh. T. and Bjùrnstad, N. O. C., N. Stenseth, hne rceig fteNtoa cdm fSciences, of climate Academy USA to National related the link of Proceedings nonstationary change? a cholera: and ENSO 1339. ouaindnmc ftevole the of dynamics population mecha- Evolution for and search Ecology a in herbivores: Trends large nisms. of dynamics lation UK. London, Hall, and hnlg,adnrhr nuae.Ecology ungulates. northern and phenology, tneh .C,O .Bùntd n .Sio.19.A 1996. Saitoh. T. and Bjùrnstad, N. O. C., N. lem- and Stenseth, voles in cycles Population 1999. C. N. Stenseth, ate,G-. .Ps,P ovy .Mne,C Parmesan, C. Menzel, A. Convey, P. Post, E. G.-R., Walther, theoret- a dynamics: population Complex 2003. P. In- Turchin, Smithsonian reindeer. Wild 1995. E. E. Syroechkovskii, Boutin, S. Boonstra, R. Tong, H. Chan, K.-S. C., N. Stenseth, oa ouaino otvoles, root of population local a Tokyo, Scienti®c, K.T.K. Japan. statistic. Criterion Information Ecology Population 61±76. on Researches Japan. Hokkaido, in oa ocn ntednmc of dynamics the on forcing sonal B series London, of Society ,X,M aca,G uh,adA .G auu.2002. Faruque. G. S. A. and Fuchs, G. Pascual, M. X., Â, rufocanus rdetfo tbet ylcppltosof populations cyclic to stable from gradient sto- a Oikos in world. dependence chastic phase and dependence density mings: Oikos hypothesis. synchrony regional 72. the for test erhso ouainEcology Population on searches Re- dynamics. population in gradients geographic modeling aren 02 clgclrsosst eetclimate recent to F. and responses Guldberg, Nature Ecological H. change. O. 2002. Fromentin, J.-M. Bairlein. Beebee, C. Press, J. T. University USA. Princeton Jersey, New Princeton, synthesis. ical/empirical USA. D.C., Washington, Libraries, stitution climatic three within Science C. populations regions. lynx dynamic M. Common Yoccoz, Canada of G. 1999. structure N. Hurrell. W. O'Donaghue, J. M. and Forchhammer, Post, E. Krebs, C. 99 :12901±12906. nHkad,Jpn rceig fteRoyal the of Proceedings Japan. Hokkaido, in 416 285 :389±395. :1071±1073. 87 :427±461. 263 :1117±1126. ltrooy rufocanus Clethrionomys ltrooy rufocanus Clethrionomys 40 irtsoeconomus Microtus clg,Vl 6 o 9 No. 86, Vol. Ecology, :85±95. 12 :143±149. Clethrionomys 80 :1322± 73 40 :65± :a : :