A quantitative analysis of environmental associations in ... · sauropods preferred coastal...

A quantitative analysis of environmental associations insauropod dinosaurs

Philip D. Mannion and Paul Upchurch

Abstract.—Both the body fossils and trackways of sauropod dinosaurs indicate that they inhabited arange of inland and coastal environments during their 160-Myr evolutionary history. Quantitativepaleoecological analyses of a large data set of sauropod occurrences reveal a statistically significantpositive association between non-titanosaurs and coastal environments, and between titanosaurs andinland environments. Similarly, ‘‘narrow-gauge’’ trackways are positively associated with coastalenvironments and ‘‘wide-gauge’’ trackways are associated with inland environments. The statisticalsupport for these associations suggests that this is a genuine ecological signal: non-titanosaursauropods preferred coastal environments such as carbonate platforms, whereas titanosaurs preferredinland environments such as fluvio-lacustrine systems. These results remain robust when the data setis time sliced and jackknifed in various ways. When the analyses are repeated using the moreinclusive groupings of titanosauriforms and Macronaria, the signal is weakened or lost. These resultsreinforce the hypothesis that ‘‘wide-gauge’’ trackways were produced by titanosaurs. It is commonlyassumed that the trackway and body fossil records will give different results, with the formerproviding a more reliable guide to the habitats occupied by extinct organisms because footprints areproduced during life, whereas carcasses can be transported to different environments prior to burial.However, this view is challenged by our observation that separate body fossil and trackway data setsindependently support the same conclusions regarding environmental preferences in sauropoddinosaurs. Similarly, analyzing localities and individuals independently results in the sameenvironmental associations. We demonstrate that conclusions about environmental patterns amongfossil taxa can be highly sensitive to an investigator’s choices regarding analytical protocols. Inparticular, decisions regarding the taxonomic groupings used for comparison, the time rangerepresented by the data set, and the criteria used to identify the number of localities can all have amarked effect on conclusions regarding the existence and nature of putative environmentalassociations. We recommend that large data sets be explored for such associations at a variety ofdifferent taxonomic and temporal scales.

Philip D. Mannion and Paul Upchurch. Department of Earth Sciences, University College London, GowerStreet, London WC1E 6BT, United Kingdom. E-mail: [email protected]; [email protected]

Accepted: 9 September 2009

Introduction

Sauropods were a major Mesozoic radia-tion of gigantic herbivorous dinosaurs andincluded the largest known terrestrial animals(Wilson 2002; Upchurch et al. 2004) (Fig. 1).They make their first appearance in the fossilrecord in the Late Triassic (Buffetaut 2000;Yates and Kitching 2003), as also inferredfrom ghost ranges (Upchurch 1995; Sereno1999), and had achieved a global distribution(excluding Antarctica) by the Middle Jurassic(McIntosh 1990; Upchurch et al. 2002;Weishampel et al. 2004a). A major extinctionevent occurred at the Jurassic/Cretaceousboundary, with 60% of lineages and 80% ofgenera disappearing at this time (Upchurchand Barrett 2005; Barrett et al. 2009). Duringthe Early Cretaceous, the remaining sauropod

lineages gradually dwindled to extinction,except for the rebbachisaurids and titanosaurs(Fig. 2). Members of the latter group wereextremely rare in the Jurassic, but this cladediversified during the Cretaceous, producingover 50 genera (Salgado et al. 1997; Wilsonand Upchurch 2003; Upchurch et al. 2004;Curry Rogers 2005). Titanosaurs were asignificant part of many terrestrial ecosystemsbefore becoming extinct at the end of theCretaceous along with the other non-aviandinosaurs (Salgado 2001).

Over the past two decades, our knowledgeof sauropod paleoecology has been improvedby the discovery of new material, the devel-opment of taxonomic schemes based ondetailed phylogenetic analyses, and investiga-tions of the depositional environments that

Paleobiology, 36(2), 2010, pp. 253–282

’ 2010 The Paleontological Society. All rights reserved. 0094-8373/10/3602–0005/$1.00

FIGURE 1. A simplified cladogram showing the relationships between the major sauropod lineages and the stem- andnode-based names currently in use (based on Wilson and Upchurch 2003, 2009, and Upchurch et al. 2004).

FIGURE 2. Taxic diversity curves for non-titanosaur (gray line) and titanosaur (black line) genera during the Jurassicand Cretaceous (based on the information on generic occurrences in the data set utilized in the current study—seeSupplementary data). Data are plotted against the absolute timescale of Gradstein et al. (2004). Abbreviations: Hett,Hettangian; Sine, Sinemurian; Plie, Pliensbachian; Toa, Toarcian; Aal, Aalenian; Baj, Bajocian; Bath, Bathonian; Call,Callovian; Oxf, Oxfordian; Kimm, Kimmeridgian; Tith, Tithonian; Ber, Berriasian; Val, Valanginian; Haut, Hauterivian;Bar, Barremian; Apt, Aptian; Alb, Albian; Cen, Cenomanian; Tur, Turonian; Con, Coniacian; San, Santonian; Cam,Campanian; Maa, Maastrichtian.

254 PHILIP D. MANNION AND PAUL UPCHURCH

yield sauropod body fossils and trackways(see reviews in Upchurch et al. 2004; Carpen-ter and Tidwell 2005; Curry Rogers andWilson 2005). Paleoenvironmental analysesdemonstrate that sauropods inhabited a widevariety of environments ranging from flood-plain and fluvio-lacustrine settings in theMorrison Formation of North America, tonearshore estuarine conditions in the Tenda-guru Formation of Tanzania (Dodson et al.1980; Russell et al. 1980). However, it isdifficult to generalize about sauropods as awhole, or identify the habitat preferences ofgroups within Sauropoda, because mostpaleoecological studies have focused on asingle geological formation (e.g., Dodson et al.1980) or geographical area (e.g., Lehman 1987;Lucas and Hunt 1989) (see also Butler andBarrett 2008 for a discussion of this issue).Some of the key questions that remain to beanswered include:

N What, if any, habitat preferences existedamong different sauropod groups?

N Can we detect evidence for true habitatpreferences against the background ‘‘noise’’generated by our patchy and biased sam-pling of the fossil record?

N How did habitat preferences relate to themorphological adaptations displayed byeach type of sauropod?

N Did habitat preferences change duringsauropod evolution?

N Did habitat preferences constrain and/orpromote changes in sauropod diversity andgeographic distribution?

Only a few studies have attempted toaddress these questions through quantitativeanalyses using extensive data sets. For exam-ple, Lockley et al. (1994) used the trackwayrecord to demonstrate that (1) sauropodsoften walked on submerged substrates incoastal and deltaic settings; (2) tracksites aremainly associated with tropical and subtropi-cal latitudes (mean average Northern Hemi-sphere paleolatitude 5 25u); and (3) themajority of trackways occur in semiarid orseasonal climatic environments where carbon-ate deposition was taking place (i.e., inlacustrine settings or in marine carbonateplatform environments). Butler et al. (2007)

and Butler and Barrett (2008), using anextensive data set of Cretaceous dinosaurs,confirmed the strength of the sauropodichnofossil record in coastal environments.However, they disagreed that sauropod track-ways were overrepresented in coastal envi-ronments relative to trackways of other her-bivorous groups and suggested that thepreservation potential for footprints was en-hanced in these habitats. The body fossilevidence, they maintained, suggested ‘‘moredistal, or inland (away from channels), pa-leoenvironmental preferences for sauropods,at least during the Cretaceous, when com-pared to contemporaneous clades such asNodosauridae and Hadrosauridae’’ (Butlerand Barrett 2008: pp. 1030–1031). Thus, bothLockley et al. (1994) and Butler and Barrett(2008) focused on the environments occupiedby sauropods as a whole, and discussion of‘‘within sauropod’’ patterns was limited tothe distributions of Macronaria versus non-Macronaria (see Fig. 1).

Our study uses a large data set comprisingthe depositional environments for virtually allidentifiable sauropod trackways and bodyfossils. We compare the distributions of‘‘narrow-gauge’’ trackways and non-titano-saur body fossils with those of ‘‘wide-gauge’’trackways and titanosaur body fossils, usingPearson’s chi-square tests to evaluate thestatistical support for putative nonrandomassociations. Finally, we evaluate the implica-tions of our results for understanding thehabitat preferences of sauropods, and discusstheir relevance to existing hypotheses ofsauropod paleoecology and evolutionary his-tory.

Materials and Methods

Data

Exploring the possible environmental pref-erences in extinct terrestrial vertebrates re-quires information on the spatiotemporaldistributions of fossils and their depositionalenvironments. We have constructed a globaldatabase of all sauropod occurrences, includ-ing detailed information on geographic loca-tion, geological setting, paleoenvironmentalindicators, stratigraphic age, and taxonomic

ENVIRONMENTAL ASSOCIATIONS IN SAUROPODS 255

composition. These data were gathered fromthe primary literature, the Paleobiology Data-base (www.paleodb.org; Carrano 2008), andpersonal observations of specimens in museumcollections. This global data set was thenfiltered to remove tracksites and body fossilsof dubious sauropod affinity and thoseoccurrences identified only as ‘‘Sauropodaindet.’’ The resulting database (see Supple-mentary Materials online at http://dx.doi.org/10.1666/08085.s1) contains informationfrom 896 localities (706 body fossil localitiesand 190 tracksites; Fig. 3) that is estimated torepresent approximately 1988 sauropodindividuals (1355 based on body fossils and633 trackmakers). The criteria for recognizingseparate localities and individuals arediscussed in ‘‘Units of Analysis’’ below.

Methodological Approach and Analyses

Analytical Protocol and Statistical Evalua-tion.—The rationale for investigating andtesting environmental associations is as fol-lows. Suppose we have two environments (Aand B) and two types of organism (X and Y).The null hypothesis is that there is nodetectable environmental preference or asso-ciation: that is, X and Y are distributed acrossA and B with no significant skew or bias. Inecological studies, this null hypothesis can betested using a variety of statistical tests, thetwo most commonly used being the G-testand Pearson’s chi-square test (Sokal andRohlf 1987; Waite 2000; Hammer and Harper2006). Although the two tests, which are bothused to determine the ‘‘goodness of fit’’between observed and expected values,

FIGURE 3. Distribution of all sauropod occurrences used in this study, plotted on modern-day world maps. A,Distribution of sauropod body fossil localities. B, Distribution of sauropod tracksite localities. Images producedin ArcGIS.


generally give the same results (Sokal andRohlf 1987), the G-test has the disadvantage ofbeing impossible to calculate if some ob-served values are zero, because logarithmsare used in its calculation (Waite 2000).Because many of our analyses include ob-served values of zero, Pearson’s chi-squaretest is the more appropriate method for thisstudy. The use of chi-square tests in paleoen-vironmental analyses also has precedents infossil invertebrate studies (e.g., Peters andBork 1999; Scholz and Hartman 2007; DeFrancesco and Hassan 2008) as well as thedinosaur analysis of Butler and Barrett (2008).In the context of paleoecology, uneven sam-pling of fossils from different environmentsintroduces a complicating factor. If we havecollected three times as many fossils of groupX from environment A relative to environ-ment B, then a non-skewed distributionwould be one in which three times as manymembers of group X are found in A com-pared to B. We must therefore take intoaccount the relative sampling of our taxontypes and environmental categories whencalculating the expected number of occur-rences. Thus, the expected number of occur-rences of taxon X in environment A (EXA) isgiven by

EXA~NT| NX=NTð Þ| NA=NTð Þ½ �, ð1Þwhere NX is the number of occurrences ofmembers of group X, NA is the number ofoccurrences of environment A, and NT is thetotal number of occurrences (modified fromWaite 2000: see also Butler and Barrett 2008).Equation (1) simplifies to

EXA~ NX|NAð Þ=NT: ð2Þ

The other three expected values requiredfor the chi-square test are given by

EXB~ NX|NBð Þ=NT ð3Þ

EYA~ NY|NAð Þ=NT ð4Þ

EYB~ NY|NBð Þ=NT: ð5Þ

The formulae above were used to calculatethe expected values shown in Tables 2–10.For example, if we designate X and Y torepresent narrow- and wide-gauge trackways

respectively, and A and B to represent inlandand coastal localities where such tracks arefound, then EXA is the expected number oftimes that narrow-gauge trackways shouldoccur in inland environments if there is noskew in the distribution. Thus, using theobserved values listed for Analysis 5 (Ta-ble 2), EXA 5 (41 3 116)/190 5 25.032(compared to the observed value of 14).

In this study, we have applied the chi-square test 134 times in order to exploredifferent aspects of our data set (see below).Therefore, if we used the standard p-value forstatistical significance of 0.05 we risk incor-rectly attributing significance to some of ourpairwise comparisons. Two tests are com-monly used for identifying which of thesepairs of samples are significantly different.The first of these, Tukey’s HSD, has thedisadvantage of reporting too high p-valueswhen sample sizes are unequal (Hammer andHarper 2006). The Bonferroni test also sub-jects the samples to pairwise comparisons,akin to the chi-square analyses, but usesmuch lower significance levels (Rice 1989)and is thus the more appropriate methodhere. This correction states that the p-value fordetermining statistical significance is given bya/n, where a is the original desired p-value(in this case 0.05) and n is the number ofanalyses (5134). Thus, the p-value we haveused to determine statistical significance is3.73 3 1024. By lowering the significancelevel, the Bonferroni correction greatly re-duces the likelihood of incorrectly attributingsignificance to our pairwise comparisons(Waite 2000). A second advantage of theBonferroni correction is that it can be usedwhen the multiple analyses are independentor non-independent from each other, as is thecase in this study (see ‘‘Analyses and Results’’below).

Units of Analysis.—In order to study envi-ronmental associations using chi-square anal-yses we require counts of the number of timesa taxon occurs in a particular environment,which can be estimates of either (1) thenumber of individual organisms belongingto a given taxon present in each habitat type,or (2) the number of localities where a giventaxon occurs in each habitat type. Each type of


estimate has its advantages and disadvan-tages.

Accurate estimation of the number ofindividual fossil organisms is often problem-atic, especially when dealing with fragmen-tary vertebrate and plant remains. In thisstudy we estimated the minimum number ofindividuals (MNIs) for each locality. The useof MNIs is a standard tool in archaeological,paleoecological and taphonomic studies (e.g.,Grayson 1973; Badgley 1986; Gilinsky andBennington 1994; Davis and Pyenson 2007).For example, if a quarry consisted of closelyassociated material belonging to the sametaxon, with no duplication or size variation ofelements, then we would consider this asingle individual. However, this method ofestimating the MNIs does have the disadvan-tage of unavoidably undercounting the num-ber of individuals. Estimating the MNIs fromtrackways can be even more difficult becausea single trackmaker could have made manydifferent tracks, and the relative size of footimpressions can be affected by under-printingand over-printing (Day et al. 2002, 2004;Milan and Bromley 2006). Without informa-tion on the tracksite that indicated otherwise,we assumed that a tracksite represented nomore than one trackmaker per recognizabletaxon. This assumption was made even whenan author stated that a locality had produced‘‘trackways’’ (e.g., Lockley et al. 2006). Futureanalyses might wish to assume the presenceof two individual trackmakers in such situa-tions, though we doubt that this refinementwould have a major effect on our conclusionsbecause it would add roughly equal numbersof narrow- and wide-gauge individual track-makers to both coastal and inland environ-ments.

Although locality-based estimates avoidsuch problems of estimating number ofindividuals (particularly for trackway data),boundaries between ‘‘separate’’ localities canbe somewhat arbitrary. For example, areassuch as Como Bluff in the Late JurassicMorrison Formation of North America haveproduced sauropod material from approxi-mately the same horizons in several closelysituated quarries (Ostrom and McIntosh1966). In our study, localities are based on

the divisions used in recent data sets (e.g.,Lockley et al. 1994; Weishampel et al. 2004a;Carrano 2008; www.paleodb.org) and aredefined as separate geographic locations anddiscrete stratigraphic levels that yield bodyfossils and/or trackways (modified fromLockley et al. 1994: p. 234). Each locality iscounted only once, irrespective of the numberof individuals present.

Environmental Categories.—The depositionalenvironments that have yielded body fossilsand trackways have been allocated to one oftwo broad categories: ‘‘inland’’ and ‘‘coastal.’’Sauropod fossils are occasionally recoveredfrom marine deposits (0.01% of occurrences),but we excluded these because it seemshighly unlikely that sauropods could havelived in marine environments. Inland envi-ronments include fluvial, lacustrine, flood-plain and eolian settings; coastal environ-ments comprise estuarine, deltaic, lagoonal,and carbonate platform settings. The divisioninto inland versus coastal is only one of themany possible ways of combining the differ-ent environments into categories for thepurposes of analysis. We focus on the in-land/coastal division because these cate-gories have yielded environmental associa-tions among other dinosaurs (Butler et al.2007; Butler and Barrett 2008) and becauseprevious studies of sauropod paleoecologyhave made claims concerning preferences forone of these broad habitat types (e.g., Lehman1987; Lucas and Hunt 1989).

Taxonomic Categories.—The search for en-vironmental associations within Sauropodarequires that the taxa and trackways bedivided into at least two distinct types orcategories. At present, most sauropod track-ways can only be placed in one of two (orperhaps three) broad categories: i.e., narrow-gauge and wide-gauge with, or without,manus-claw and phalangeal impressions(Fig. 4) (Farlow et al. 1989; Lockley et al.1994; Wilson and Carrano 1999; Day et al.2004). Narrow-gauge trackways have foot-prints that approach (or intersect) the midline,with pollex claw impressions preserved (e.g.,Parabrontopodus: Fig. 4A), whereas wide-gauge trackways have footprints placed wellaway from the midline and pollex claw


impressions are frequently absent (e.g., Bron-topodus: Fig. 4B). On the basis of severalfemoral morphological features (such as thelateral deflection of the proximal one-third ofthe femur) which suggest that titanosaursheld their limbs farther from the midline thanother sauropods (Fig. 4C,D), Wilson andCarrano (1999) argued that titanosaurs wereresponsible for producing these wide-gaugetracks. This interpretation is also supportedby the derived loss of pollex claws andmanual phalanges in both titanosaurian skel-etal remains and many wide-gauge trackways(Salgado et al. 1997; Day et al. 2002), thoughthis may not be the case in basal titanosaursor the more inclusive Titanosauriformes

(Fig. 1) (Day et al. 2004). It is difficult toassign sauropod trackways to their track-maker more precisely than either narrow-gauge or wide-gauge types because of thevery limited number of derived features inthe fore and hind feet of distinct sauropodlineages that can be unambiguously recog-nized in trackways (though see Wilson 2005a;Wright 2005). However, when trace fossilscan be accurately assigned to particular taxa,analyses of separate body fossil and tracefossil data sets have the advantage that theyyield two independent assessments of theenvironmental associations for the organismsconcerned. Such an approach might reinforcethe support for the hypothesis that a given

FIGURE 4. The two main sauropod track types: ‘‘narrow-gauge’’ trackway (A) and ‘‘wide-gauge’’ trackway (B).Reconstructed pelvic girdles and hindlimbs, in anterior view, of the non-titanosaur Camarasaurus (C) and the titanosaurOpisthocoelicaudia (D). The vertical lines in A and B illustrate the distance separating left and right prints in thetrackways. Images modified from Lockley et al. (1994) and Wilson and Carrano (1999).


clade made the trace fossil, or might revealincongruence that would require reassess-ment of the initial identification of theputative trace fossil makers. If congruencebetween the body fossil and trace fossilsignals exists, it is then justifiable to unitethe data sets to produce a ‘‘combined evi-dence’’ analysis of the environmental associa-tions. Thus, trackways and body fossils havetheir own particular advantages and disad-vantages for paleoecological research and wetherefore analyze both types of data sepa-rately and in combination.

With regard to body fossils, there is a muchwider choice of possible comparisons, such asneosauropods versus non-neosauropods, ti-tanosaurians versus non-titanosaurians, ordiplodocoids versus macronarians. In thisstudy, the majority of the analyses examinethe environmental associations of titanosaursversus non-titanosaurs. This enables us tocarry out ‘‘combined evidence’’ analyses inwhich narrow-gauge trackways and non-titanosaur body fossil data are combinedand compared with wide-gauge trackwaysplus titanosaur body fossil data, providingtwo independent lines of evidence to test forenvironmental associations. In addition, theutilization of narrow- and wide-gauge loco-motor styles might be related to the biome-chanical demands of different habitats, makingit a suitable starting point for investigatingenvironmental associations among sauropods.

Wilson and Carrano (1999) noted that basaltitanosauriforms such as Brachiosaurus andEuhelopus have femoral morphologies that areintermediate between those in narrow-gaugenon-titanosaurs and wide-gauge titanosaurs.We therefore also analyze the body fossil datausing a division into non-titanosauriformsand titanosauriforms, as well as a partitioninto non-titanosaurian titanosauriforms andtitanosaurs to test for differences withinTitanosauriformes.

Macronaria is a major sauropod clade thatis largely composed of the Titanosauriformesand a few basal genera such as Camarasaurus(Fig. 1). A recent study of Cretaceous dino-saurian distributions suggested that ‘‘Sauro-pods show little evidence for broad environ-mental associations: a significant negative

association between Macronaria and coastalenvironments may be a result of taphonomicprocesses’’ (Butler et al. 2007: pp. 54–55).Furthermore, Butler and Barrett (2008: Ta-ble 1) also reported a statistically significant(p , 1 3 1024) positive association betweenMacronaria and inland environments. We testthis possibility by comparing macronarianswith non-macronarians and by restricting ourdata set to Cretaceous occurrences. In addi-tion, we reanalyze the Butler and Barrett(2008) data set to look for patterns at thetaxonomic levels of titanosauriforms andtitanosaurs.

A Note on Paraphyly.—Division of thesauropod data into non-titanosaurs versustitanosaurs, non-titanosauriforms versus titano-sauriforms and macronarians versus non-macronarians means that we are comparingthe distributions of a paraphyletic assemblagewith those of a monophyletic group. The useof paraphyletic groups in paleobiologicalanalyses is potentially problematic becausesuch groups have an arbitrary taxonomiccontent. If we choose to use a differenttaxonomic definition, the boundaries aroundthe paraphyletic assemblage change andtherefore its contents also change. However,the use of paraphyletic groups is justifiedwhen such assemblages represent evolution-ary grades or ecological communities (seePeters [2008] for an example concerningSepkoski’s [1984] marine invertebrate evolu-tionary faunas, and Wilson and Carrano [1999]who carried out statistical analyses of femoralmeasurements based on a division into ‘‘ti-tanosaurs’’ and ‘‘other sauropods’’). For ex-ample, when analyzing the ecological conse-quences of limblessness, it would be legitimateto compare one or more of the monophyleticlimbless squamate groups (such as snakes,amphisbaenians) with the paraphyletic gradeof ‘‘lizards’’ in which limbs are retained. Insuch analyses, the boundaries of a paraphy-letic group are not arbitrary because they areset by the retention of one or more symplesio-morphies that are related to the biomechanicsand/or ecology of the group. The synapomor-phies which mark the boundary between aparaphyletic grade and one of its monophy-letic descendant clades might relate to an


evolutionary shift in physiology, behavior,habitat, etc., so that the only meaningful wayto explore the resulting differences in environ-mental association would be to compare thedistributions of the paraphyletic assemblageand the monophyletic group. It is conceivablethat the acquisition of the wide-gauge stancein titanosaurs enabled them to invade orinhabit a different set of environments, andwe therefore believe that our comparisonsamong sauropods are potentially ecologicallymeaningful.

Uneven Sampling of Environments throughTime.—Trends in the diversity and abun-dance of taxa through time, combined withuneven sampling of depositional environ-ments, could create an additional problemfor paleobiologists searching for environ-mental associations. Suppose, for example,that the ratio of inland to coastal localitiessampled from the Jurassic is 1:4 and from theCretaceous is 4:1. Furthermore, suppose thattitanosaurs were rare during the Jurassic andcommon during the Cretaceous relative tonon-titanosaurs, so that the ratio of titano-saurs to non-titanosaur individuals is 1:10 inthe Jurassic and 10:1 in the Cretaceous.Finally, suppose that titanosaurs and non-titanosaurs exhibited no environmental pref-erences so that they were evenly distributedacross the two environmental categories.Under this scenario, if we collected 110sauropod individuals from the Jurassic,approximately ten of these would be titano-saurs and 100 would be non-titanosaurs. Ifthese taxa display no environmental prefer-ences, then we would expect eight titano-saurs and 80 non-titanosaurs from thecoastal localities and two titanosaurs and20 non-titanosaurs from the inland localities.If we then collected 110 sauropod individ-uals from the Cretaceous, we would expect80 titanosaurs and eight non-titanosaursfrom the inland localities, and 20 titanosaursand two non-titanosaurs from the coastallocalities. The total (Jurassic + Cretaceous)data set would comprise 82 titanosaurs and28 non-titanosaurs from inland localities,and 28 titanosaurs and 82 non-titanosaursfrom coastal localities. This skewed distribu-tion passes the chi-square test (p , 1 3 1025),

but the uneven distribution does not resultfrom environmental preferences: such askew is a by-product of the combination oflong-term trends in the relative abundancesof the two taxon categories and two envi-ronmental categories. Thus, given admittedlysomewhat contrived conditions, a statisticallyrobust, but nonetheless artifactual, environ-mental association can be generated.

Evidence suggests that investigations ofenvironmental associations in sauropodsmust deal with precisely the scenario outlinedabove. Our current understanding of sau-ropod diversity patterns is that non-titano-saur lineages were diverse during the Ju-rassic, declined in the Early Cretaceous, andbecame extinct in the early Late Cretaceous,whereas titanosaurs were rare in the Jurassicbut increased in diversity and abundance inthe Cretaceous (Upchurch and Barrett 2005)(Fig. 2). To investigate the possible effects offluctuations of inland and coastal environ-ments, we calculated the relative samplingrates of these two environments for sauropod-bearing localities for each stage of the Jurassicand Cretaceous. However, to avoid thepotential circular reasoning that arises fromonly considering sauropod-bearing localities(i.e., it is possible that there were as manycoastal localities capable of preserving large-bodied terrestrial vertebrates during theCretaceous as there were during the Jurassic,but we might recognize fewer of them duringthe Cretaceous because we have collecteddata only on sauropod-bearing localities andbecause titanosaurs preferred inland habi-tats), we have also compiled information onthe numbers of ornithischian-bearing local-ities (www.pbdb.org; Carrano 2008).

There is a marked difference in the relativesampling rates for coastal and inland sau-ropod-bearing localities during the Jurassicand Cretaceous. Table 1 shows that thepercentage of coastal sauropod-bearing local-ities during the Jurassic is 19.4%, whereasduring the Cretaceous this falls to 8.4%. Fornumbers of ornithischian-bearing localities,Table 1 shows that coastal environmentscontribute 23.2% of Jurassic and only 7.5%of Cretaceous localities. Given that severalornithischian clades apparently display a


positive association with marine/coastal lo-calities during the Cretaceous (Butler andBarrett 2008), the sauropod and ornithischiandata taken together and separately (Table 1)suggest that coastal localities are indeedunderrepresented during the Cretaceous rel-ative to the Jurassic. Thus, we need toeliminate the possibility that statistically sig-nificant environmental associations relatingto titanosaurs versus non-titanosaurs are anartifact of uneven sampling of inland andcoastal habitats during the Jurassic andCretaceous.

The deleterious effects of trends in therelative abundance and/or diversity of taxaand environments can be ameliorated bysearching for environmental associationsusing narrower time bins. If titanosaurs andnon-titanosaurs displayed genuine habitatpreferences, then statistically robust environ-

mental associations should persist even whenthe data are drawn from a narrower timewindow in which the relative abundance ofinland and coastal environment types doesnot change markedly. We therefore repeat thechi-square analyses using Cretaceous, Early-early Late Cretaceous (Berriasian–Coniacian)and stage-level time slices.

‘‘Jackknifing’’ the Data.—One danger withour approach is that the results may bedominated by short-lived windows of excep-tional preservation that overwhelm signals inthe rest of the data set. This is a particularconcern with our trackway data because 73out of 190 localities (38.4%) are based on amega-tracksite from the Early Cretaceous ofthe Republic of Korea (Lim et al. 1989; Lee etal. 2000; Lockley et al. 2006). Thus, thedecision to treat this mega-tracksite as 73separate localities, rather than one large

TABLE 1. The relative abundances of inland and coastal environments during the Late Triassic to Late Cretaceous,based on the numbers of localities producing sauropod body fossils, ornithischian body fossils, and combinedsauropod and ornithischian body fossils.

Ma Stage

Sauropod body fossils + tracksites Ornithischian body fossils + tracksites

Inland localities Coastal localities Inland localities Coastal localities

228 Carnian 3 0 2 2216.5 Norian 5 0 2 0203.6 Rhaetian 2 1 1 0199.6 Hettangian 7 7 30 5196.5 Sinemurian 10 4 28 1189.6 Pliensbachian 7 3 5 1183 Toarcian 8 3 5 0175.6 Aalenian 8 1 1 1171.6 Bajocian 15 2 2 1167.7 Bathonian 21 10 1 1164.7 Callovian 20 1 4 0161.2 Oxfordian 24 3 11 1155.7 Kimmeridgian 241 38 33 11150.8 Tithonian 234 71 35 24145.5 Berriasian 30 3 22 3140.2 Valanginian 31 2 24 2136.4 Hauterivian 44 4 35 3130 Barremian 59 13 55 2125 Aptian 101 8 97 7112 Albian 140 28 196 3199.6 Cenomanian 41 3 39 593.5 Turonian 53 1 16 189.3 Coniacian 46 1 12 185.8 Santonian 38 0 19 183.5 Campanian 84 4 268 970.6 Maastrichtian 118 5 117 3

PeriodSauropods: % Coastal

relative to totalOrnithischians: % Coastal

relative to totalSauropods + Ornithischians: %

Coastal relative to total

Triassic 9.1% 28.6% 16.7%Jurassic 19.4% 23.2% 20.2%Cretaceous 8.4% 7.5% 7.6%


locality, could make an important differenceto our results (see ‘‘Units of Analysis’’ above).We have therefore restricted analysis oftrackway data to the 118 localities that remainafter removal of all but one of the Koreantracksites. However, subsuming 73 separateKorean tracksites into just one locality couldbe regarded as overly severe, given that atleast six main tracksites can be distinguishedfrom each other on the basis of spatial andstratigraphic separation (Lim et al. 1989; Leeet al. 2000; Lockley et al. 2006). Therefore, wehave also run the analyses with Koreantracksites represented by these six regions(see also Wright 2005). Finally, the U.S. trackrecord also constitutes a large proportion(17%) of the tracksite data set, including alarge number of Early Cretaceous coastallocalities; thus, our analyses have been runwith and without these tracksites.

Analyses and Results

The 134 analyses outlined below have beendivided into three groups. Those in the firstgroup (nos. 1–14, Table 2) use the wholeMesozoic data set to determine whethertitanosaurs (or one of the slightly moreinclusive groups represented by the Titano-sauriformes and Macronaria) and wide-gaugetrackways occur more often than expected ineither inland or coastal environments. Be-cause these analyses are based on the mostdata, we give them the greatest weight in ourdiscussion of the implications of our results.Analyses in the second group (nos. 15–115,Tables 3–9) examine the effects of time-slicingthe data sets. Those in the final group(Analyses 116–134, Table 10) determine theeffects of jackknifing the trackway data byremoving the U.S. and South Korean local-ities. It should be noted that the 134 separateanalyses have complex relationships in termsof their independence or non-independence.Many of the time slice analyses are effectivelyindependent of each other; for example,Bathonian and Callovian stage data sets(e.g., nos. 45 and 46) share no data points incommon. Similarly, analyses based solely onbody fossil data (e.g., nos. 1 and 4) areindependent of those based purely on track-

way data (e.g., nos. 2 and 5). However, manyother analyses are wholly or partly non-independent; for example, the data in Analy-sis 4 (all body fossil localities; Table 2) over-lap substantially with those in Analysis 18(Cretaceous body fossil localities; Table 3).The number of data supporting each statisti-cally significant result, and the independenceof separate analyses, will be considered whenwe discuss the implications of the results.

Analyses 1–14: Mesozoic Data Sets

Titanosaurs versus Non-Titanosaurs.—Anal-yses 1–6 (Table 2) compare the distributionsof titanosaur and non-titanosaur body fossils,narrow-gauge and wide-gauge trackways,and body fossils plus trackways, based onestimated numbers of individuals (Analyses1–3) and localities (Analyses 4–6). Analyses 1–5 produce very low p-values (p 5 1 3 1025 to 73 1024), whereas Analysis 6 is non-significant(p , 0.0677). These results suggest thattitanosaurs and wide-gauge tracks occurmore often than expected in inland environ-ments, whereas non-titanosaurs and narrow-gauge trackways occur more often thanexpected in coastal environments.

Titanosauriforms versus Non-Titanosauri-forms.—Analyses 7–10 (Table 2) examine pos-sible environmental associations among titano-sauriform and non-titanosauriform body fossildata by themselves and with the trackway dataadded, based on estimated numbers of individ-uals and localities. Three of these analyses (7, 9and 10) fail to produce statistically significantresults (p 5 0.1296 to 0.2537). However, resultsof Analysis 8, which combines body fossil andtrackway data for individuals, are significant (p5 2 3 1024), suggesting that titanosauriformsoccur more often than expected in coastalenvironments, whereas non-titanosauriformsoccur more often than expected in inlandenvironments. This is the reverse of theenvironmental association found among titano-saurs and non-titanosaurs, which we hypothe-size occurs because the data from non-titano-saurian (‘‘basal’’) titanosauriforms is somehow‘‘swamping’’ the inland signal recovered inAnalyses 1–5 (see below).

To test for this swamping by non-titano-saurian titanosauriforms, we restricted the


TABLE 2. Summary of analyses (1–14) of potential environmental associations among sauropods, based on thecomplete Mesozoic data sets (see ‘‘Analyses and Results’’ for details). Abbreviations: E, expected number; O, observednumber; T, total number. The following symbols next to an analysis number denote: (1) an asterisk (*) denotes asignificant result which supports titanosaurs preferring inland environments; (2) two asterisks (**) denote a significantresult which supports titanosaurs preferring coastal environments; (3) a hash sign (#) denotes a significant resultwhich supports titanosauriforms preferring inland environments; (4) two hash signs (##) denote a significant resultwhich supports titanosauriforms preferring coastal environments. Significant results are those with p-levels that areless than 3.78 3 1024.

Analysis No. Comparison Inland Coastal p-value

1* Body fossil individuals: T 5 1360 T 5 1226 T 5 134 , 1 3 1025

Non-titanosaurs: T 5 987 O 5 864 O 5 123E 5 889.75 E 5 97.25

Titanosaurs: T 5 373 O 5 362 O 5 11E 5 336.25 E 5 36.75

2* Trackway individuals: T 5 633 T 5 236 T 5 397 , 1 3 1025

Narrow-gauge: T 5 187 O 5 30 O 5 157E 5 69.72 E 5 117.28

Wide-gauge: T 5 446 O 5 206 O 5 240E 5 166.28 E 5 279.72

3 Body fossil and trackway individuals: T 5 1993 T 5 1462 T 5 531 7 3 1024

Non-titanosaurs and narrow-gauge: T 5 1174 O 5 894 O 5 280E 5 861.21 E 5 312.79

Titanosaurs and wide-gauge: T 5 819 O 5 568 O 5 251E 5 600.79 E 5 218.21

4* Body fossil localities: T 5 706 T 5 624 T 5 82 , 1 3 1025



5* Trackway localities: T 5 190 T 5 116 T 5 74 8 3 1025


Wide-gauge: T 5 150 O 5 102 O 5 48E 5 91.58 E 5 58.42

6 Body fossil and trackway localities: T 5 896 T 5 740 T 5 156 0.0677Non-titanosaurs and narrow-gauge: T 5 520 O 5 418 O 5 102

E 5 429.46 E 5 90.54Titanosaurs and wide-gauge: T 5 387 O 5 329 O 5 58

E 5 319.62 E 5 67.387 Body fossil individuals: T 5 1360 T 5 1226 T 5 134 0.2145

Non-titanosauriforms: T 5 819 O 5 745 O 5 74E 5 738.30 E 5 80.70

Titanosauriforms: T 5 541 O 5 481 O 5 60E 5 487.70 E 5 53.30

8## Body fossil and trackway individuals: T 5 1993 T 5 1462 T 5 531 2 3 1024

Non-titanosauriforms and narrow-gauge: T 5 1006 O 5 775 O 5 231E 5 737.97 E 5 268.03

Titanosauriforms and wide-gauge: T 5 987 O 5 687 O 5 300E 5 724.03 E 5 262.97

9 Body fossil localities: T 5 706 T 5 624 T 5 82 0.1296Non-titanosauriforms: T 5 370 O 5 323 O 5 47

E 5 327.03 E 5 42.97Titanosauriforms: T 5 359 O 5 309 O 5 50

E 5 317.30 E 5 41.7010 Body fossil and trackway localities: T 5 896 T 5 740 T 5 156 0.2537



11* Body fossil individuals: T 5 541 T 5 481 T 5 60 , 1 3 1025

Non-titanosaurian titanosauriforms: T 5 168 O 5 119 O 5 49E 5 149.37 E 5 18.63



body fossil data set to Titanosauriformes,partitioning the data into titanosaurs andnon-titanosaurs (Analyses 11 and 12, Table 2).Both analyses display strong support (p , 1 3

1025) for titanosaurs occurring more oftenthan expected in inland environments, andbasal titanosauriforms occurring more oftenthan expected in coastal ones.

Macronarians versus Non-Macronarians.—Analyses 13 and 14 (Table 2) compare thedistributions of Macronaria and non-Macro-naria for all sauropods. Both analyses fail thechi-square test (Table 2), indicating that thereis no significant support for environmentalassociations when sauropods are partitionedinto Macronaria versus non-Macronaria (con-tra Butler and Barrett 2008; see below).

Analyses 15–115: Time-Slicing the Data

Cretaceous Time Slices.—Analyses 15–17(Table 3) repeat Analyses 1, 5, and 7 respec-tively, but are restricted to Cretaceous sau-ropods alone. Analysis 15 passes the chi-square tests (p , 1 3 1025), demonstratingthat the restriction of the data set does notaffect conclusions based on Analysis 1. Theprevious positive association between wide-gauge trackways and inland environmentsand narrow-gauge trackways and coastalenvironments (Analysis 5) disappears whenonly Cretaceous data are examined (Analysis16), probably because there are very fewnarrow-gauge trackways known from theCretaceous (six in our data set; see Supple-mentary Materials). Analysis 7 failed the chi-

square test, but Analysis 17 yields a positiveassociation between titanosauriforms and in-land environments and non-titanosauriformsand coastal environments. This result prob-ably occurs because restricting ‘‘Titanosaur-iformes’’ to Cretaceous taxa alone raises theproportion of titanosaurs.

Analyses 18 and 19 examine the distribu-tions of Macronaria and non-Macronariabased on just the Cretaceous part of our dataset. As before there is no significant supportfor environmental associations when sauro-pods are partitioned into these two groups,although the p-values are substantially lower(p , 0.14) than for when all sauropods areconsidered (Analyses 13 and 14, p . 0.5). Thisphenomenon seems to occur because thetaxonomic content of ‘‘Cretaceous Macro-naria’’ is very similar to that of Titanosaur-iformes and Titanosauria.

Analysis of Butler and Barrett’s (2008) dataset of Cretaceous sauropods (Analyses 20–23)uses another time slice of the total availabledata to examine environmental associationsbetween titanosaurs and non-titanosaurs, andbetween titanosauriforms and non-titano-sauriforms. Although results are consistentwith positive associations between titano-saurs/titanosauriforms and inland environ-ments, none pass the chi-square test (p .

0.0017, Table 3).Within-Cretaceous Time Slices.—Analyses

24–29 (Table 3) repeat Analyses 1–6, but withthe data restricted to the Early and early LateCretaceous (Berriasian–Coniacian). Analyses

TABLE 2. Continued.

Analysis No. Comparison Inland Coastal p-value

12* Body fossil localities: T 5 362 T 5 310 T 5 52 , 1 3 1025

Non-titanosaurian titanosauriforms: T 5 122 O 5 82 O 5 40E 5 104.48 E 5 17.52


13 Body fossil individuals: T 5 1360 T 5 1225 T 5 135 0.9383Non-macronarians: T 5 565 O 5 509 O 5 56

E 5 508.92 E 5 56.08Macronarians: T 5 795 O 5 716 O 5 79

E 5 716.08 E 5 78.9214 Body fossil localities: T 5 706 T 5 624 T 5 82 0.5435

Non-macronarians: T 5 307 O 5 268 O 5 39E 5 271.34 E 5 35.66

Macronarians: T 5 467 O 5 412 O 5 55E 5 412.76 E 5 54.24


TABLE 3. Summary of analyses (15–29) of potential environmental associations among sauropods, based on larger timeslices such as ‘‘Cretaceous,’’ ‘‘Early Cretaceous,’’ etc. (see ‘‘Analyses and Results’’ for details). For abbreviations andsymbols, see legend to Table 2.

Analysis no. Comparison Inland Coastal p-value

15* Body fossil individuals (Cretaceous only): T 5 526 T 5 499 T 5 27 , 1 3 1025



16 Trackway localities (Cretaceous only): T 5 131 T 5 97 T 5 34 0.6748Narrow-gauge: T 5 6 O 5 4 O 5 2

E 5 4.44 E 5 1.56Wide-gauge: T 5 125 O 5 93 O 5 32

E 5 92.56 E 5 32.4417# Body fossil individuals (Cretaceous only): T 5 526 T 5 499 T 5 27 , 1 3 1025



18 Body fossil localities (Cretaceous only): T 5 525 T 5 496 T 5 29 0.0574Non-macronarians: T 5 54 O 5 48 O 5 6

E 5 51.02 E 5 2.98Macronarians: T 5 471 O 5 448 O 5 23

E 5 444.98 E 5 26.0219 Body fossil localities (Cretaceous only): T 5 334 T 5 312 T 5 22 0.1351

Non-macronarians: T 5 41 O 5 36 O 5 5E 5 38.30 E 5 2.70

Macronarians: T 5 297 O 5 279 O 5 18E 5 277.44 E 5 19.56

20 Body fossil localities (Butler and Barrett Cretaceousdata): T 5 175

T 5 166 T 5 9 0.0019



21 Body fossil and trackway localities (Butler andBarrett Cretaceous data): T 5 177

T 5 168 T 5 9 0.0017



22 Body fossil localities (Butler and Barrett Cretaceousdata): T 5 175

T 5 166 T 5 9 0.0178



23 Body fossil and trackway localities (Butler andBarrett Cretaceous data): T 5 177

T 5 168 T 5 9 0.0163



24* Body fossil individuals (Early-mid Cretaceous only):T 5 259

T 5 237 T 5 22 1 3 1024



25 Trackway individuals (Early–mid Cretaceous): T 5284

T 5 180 T 5 104 0.2769


Wide-gauge: T 5 281 O 5 179 O 5 102E 5 178.1 E 5 102.9


25–29 produce non-significant p-values. How-ever, the remaining analysis (no. 24, body-fossil individuals) still supports the positiveassociation between titanosaurs and inlandenvironments (p 5 1 3 1024).

Stage-Level Time-Slicing.—Analyses 30–115(Tables 4–9) replicate Analyses 1–6, but withdata restricted to a single European stage ofthe Jurassic or Cretaceous. Of these 86analyses, only 22 produced statistically sig-nificant results: 13 in which the associationbetween titanosaurs and/or wide gaugetrackways and inland environments wasnegative, and nine in which it was positive.Although these results appear to conflict, itshould be noted that 12 of the 13 analyseswith ‘‘negative’’ results were of Middle andLate Jurassic stages (the exception is Analysis66—Albian body fossil and trackway individ-uals), whereas all nine analyses supportingthe positive association with inland environ-ments were of Cretaceous stages. Theseresults reflect the fact that the earliest wide-gauge trackways (e.g., the Middle JurassicArdley site) and the earliest titanosaurianbody fossils (Janenschia, from the Late Jurassicof Tendaguru, Tanzania) are preserved incoastal environments (Aberhan et al. 2002;

Day et al. 2002, 2004). This issue will beaddressed further in the ‘‘Discussion’’. Fornow, it is sufficient to note that even when thedata set is time-sliced to stage level, manystatistically significant environmental associa-tions occur, suggesting that the results cannotbe explained merely as an artifact created byuneven sampling of the two environmenttypes from deposits of Jurassic and Creta-ceous age.

Analyses 116–134: ‘‘Jack-knifing’’ the Data

Following exclusion of the Korean track-sites, environmental associations are notsupported by analyses 119–121 (Table 10, p. 0.2432), which are based on the number oflocalities. Analyses 116–118, based on numberof individuals, pass the chi-square test (p ,

13 1025). Analysis 116 (based on trackwaydata alone) continues to support the positiveassociations between wide-gauge trackwaysand inland environments and narrow-gaugetrackways and coastal environments. How-ever, the addition of the body fossil data inAnalyses 117 and 118 reverses the polarity ofthese associations. Analysis 122 explores whathappens when the 73 Korean tracksites aretreated as six separate localities, but this

TABLE 3. Continued.


26 Body fossil and trackway individuals (Early–midCretaceous): T 5 543

T 5 417 T 5 126 0.0193



27 Body fossil localities (Early–mid Cretaceous): T 5 162 T 5 144 T 5 18 0.0039Non-titanosaurs: T 5 95 O 5 79 O 5 16

E 5 84.44 E 5 10.56Titanosaurs: T 5 73 O 5 71 O 5 2

E 5 64.89 E 5 8.1128 Trackway localities (Early–mid Cretaceous): T 5 120 T 5 88 T 5 32 0.1126


Wide-gauge: T 5 117 O 5 87 O 5 30E 5 85.80 E 5 31.20

29 Body fossil and trackway localities (Early–midCretaceous): T 5 282

T 5 232 T 5 50 0.7195




TABLE 4. Results of Analyses 30–43 (sauropod body fossil individuals) per European stage. Only stages where bothtitanosaurs and non-titanosaur body fossils are known are shown. See Table 2 for explanation of abbreviationsand symbols.

Analysis no. Stage

Non-titanosaurs Titanosaurs

p-valueInland Coastal Inland Coastal

30** Kimmeridgian O 5 582 O 5 38 O 5 0 O 5 6 , 1 3 1025

E 5 576.42 E 5 43.58 E 5 5.58 E 5 0.4231** Tithonian O 5 489 O 5 70 O 5 0 O 5 8 , 1 3 1025

E 5 482.10 E 5 76.90 E 5 6.90 E 5 1.1032 Berriasian O 5 20 O 5 0 O 5 6 O 5 0 1

E 5 20.00 E 5 0 E 5 6.00 E 5 033 Valanginian O 5 13 O 5 0 O 5 19 O 5 0 1

E 5 13.00 E 5 0 E 5 19.00 E 5 034 Hauterivian O 5 35 O 5 4 O 5 20 O 5 0 0.1371

E 5 36.36 E 5 2.64 E 5 18.64 E 5 1.3635 Barremian O 5 52 O 5 13 O 5 33 O 5 0 0.0058

E 5 56.38 E 5 8.62 E 5 28.62 E 5 4.3836 Aptian O 5 50 O 5 6 O 5 28 O 5 0 0.0723

E 5 52.00 E 5 4.00 E 5 26.00 E 5 2.0037 Albian O 5 61 O 5 4 O 5 29 O 5 0 0.1733

E 5 62.23 E 5 2.77 E 5 27.77 E 5 1.2338 Cenomanian O 5 23 O 5 1 O 5 36 O 5 1 0.7567

E 5 23.21 E 5 0.79 E 5 35.79 E 5 1.2139 Turonian O 5 11 O 5 0 O 5 52 O 5 0 1

E 5 11.00 E 5 0 E 5 52.00 E 5 040 Coniacian O 5 5 O 5 0 O 5 51 O 5 0 1

E 5 5.00 E 5 0 E 5 51.00 E 5 041 Santonian O 5 3 O 5 0 O 5 46 O 5 0 1

E 5 3.00 E 5 0 E 5 46.00 E 5 042 Campanian O 5 4 O 5 0 O 5 126 O 5 2 0.8018

E 5 3.94 E 5 0.06 E 5 126.06 E 5 1.9443* Maastrichtian O 5 5 O 5 5 O 5 189 O 5 0 , 1 3 1025

E 5 9.75 E 5 0.25 E 5 184.25 E 5 4.75

TABLE 5. Results of Analyses 44–55 (sauropod tracksite individuals) per European stage. Only stages where bothtitanosaurs and non-titanosaurs are known are shown. See Table 2 for explanation of abbreviations and symbols.

Analysis no. Stage

Narrow-gauge Wide-gauge


44** Bajocian O 5 8 O 5 4 O 5 0 O 5 15 2 3 1024

E 5 3.56 E 5 8.44 E 5 4.44 E 5 10.5645** Bathonian O 5 17 O 5 31 O 5 0 O 5 35 8 3 1025

E 5 9.83 E 5 38.17 E 5 7.17 E 5 27.8346 Callovian O 5 8 O 5 4 O 5 1 O 5 0 0.4857

E 5 8.31 E 5 3.69 E 5 0.69 E 5 0.3147 Kimmeridgian O 5 8 O 5 70 O 5 8 O 5 15 0.0046

E 5 12.36 E 5 65.64 E 5 3.64 E 5 19.3648 Tithonian O 5 3 O 5 91 O 5 8 O 5 39 0.0039

E 5 7.33 E 5 86.67 E 5 3.67 E 5 43.3349 Berriasian O 5 0 O 5 1 O 5 34 O 5 8 0.0497

E 5 0.79 E 5 0.21 E 5 33.21 E 5 8.7950* Valanginian O 5 0 O 5 1 O 5 32 O 5 1 6 3 1025

E 5 0.94 E 5 0.06 E 5 31.06 E 5 1.9451* Hauterivian O 5 0 O 5 1 O 5 32 O 5 1 6 3 1025

E 5 0.94 E 5 0.06 E 5 31.06 E 5 1.9452* Barremian O 5 0 O 5 1 O 5 41 O 5 0 , 1 3 1025

E 5 0.98 E 5 0.02 E 5 40.02 E 5 0.9853* Aptian O 5 0 O 5 1 O 5 100 O 5 1 , 1 3 1025

E 5 0.98 E 5 0.02 E 5 99.02 E 5 0.9854 Albian O 5 1 O 5 0 O 5 137 O 5 86 0.4337

E 5 0.62 E 5 0.38 E 5 137.38 E 5 85.6255 Campanian O 5 3 O 5 0 O 5 11 O 5 9 0.1376

E 5 1.83 E 5 1.17 E 5 12.17 E 5 7.83


version of the data set still fails the chi-squaretest (p 5 0.23, Table 10).

Analyses 123–128 (Table 10) exclude theU.S. trackway data. Four of these analyses(124, 125, 127, and 128) fail the chi-square test.The trackway data alone (Analyses 123 and126) pass the chi-square tests (p 5 1 31025)and still support the associations betweenwide-gauge trackways and inland environ-ments and narrow-gauge trackways andcoastal environments.

Finally, Analyses 129–134 (Table 10) ex-plore the consequences of excluding boththe Korean and U.S. trackway data (56% ofthe tracksite data set). Analyses 129 and 131(based on number of individuals) both passthe chi-square test (p , 6 3 1024, Table 10),with the former supporting the positiveassociation between titanosaurs/wide-gauge

trackways and inland environments and non-titanosaurs/narrow-gauge trackways andcoastal environments, but the latter reversingthe polarity of this association. Analysis 131 isnot surprising, however, because severalother analyses based on the titanosauri-form/non-titanosauriform categories producesimilar results (e.g., Analysis 8, Table 2). Thethree remaining locality-based analyses(nos. 132-134) all fail the chi-square tests(Table 10).

Discussion

Environmental Associationsamong Sauropods

In this section we discuss the taxonomiclevel at which we believe the environmentalassociations hold true, and the polarity of

TABLE 6. Results of Analyses 56–72 (sauropod body fossil and tracksite individuals) per European stage. Only stageswhere both titanosaurs and non-titanosaurs are known are shown. See Table 2 for explanation of abbreviationsand symbols.

Analysis no. Stage

Non-titanosaurs and narrow-gauge Titanosaurs and wide-gauge


56** Bajocian O 5 52 O 5 4 O 5 0 O 5 15 , 1 3 1025

E 5 41.01 E 5 14.99 E 5 10.99 E 5 4.0157** Bathonian O 5 90 O 5 33 O 5 0 O 5 35 , 1 3 1025

E 5 70.06 E 5 52.94 E 5 19.94 E 5 15.0658 Callovian O 5 89 O 5 4 O 5 1 O 5 0 0.8376

E 5 89.04 E 5 3.96 E 5 0.96 E 5 0.0459** Kimmeridgian O 5 590 O 5 108 O 5 8 O 5 21 , 1 3 1025



E 5 18.26 E 5 2.74 E 5 41.74 E 5 6.2662 Valanginian O 5 13 O 5 1 O 5 51 O 5 1 0.3078

E 5 13.58 E 5 0.42 E 5 50.42 E 5 1.5863 Hauterivian O 5 35 O 5 5 O 5 52 O 5 1 0.0391

E 5 37.42 E 5 2.58 E 5 49.58 E 5 3.4264* Barremian O 5 52 O 5 14 O 5 74 O 5 0 3 3 1025

E 5 59.40 E 5 6.60 E 5 66.60 E 5 7.4065* Aptian O 5 50 O 5 7 O 5 128 O 5 1 3 3 1024

E 5 54.55 E 5 2.45 E 5 123.45 E 5 5.5566** Albian O 5 62 O 5 4 O 5 166 O 5 86 , 1 3 1025


E 5 21.82 E 5 2.18 E 5 38.18 E 5 3.8268 Turonian O 5 11 O 5 0 O 5 65 O 5 2 0.5633

E 5 10.72 E 5 0.28 E 5 65.28 E 5 1.7269 Coniacian O 5 5 O 5 0 O 5 64 O 5 2 0.6938

E 5 4.86 E 5 0.14 E 5 64.14 E 5 1.8670 Santonian O 5 3 O 5 0 O 5 46 O 5 0 1


E 5 6.50 E 5 0.50 E 5 137.50 E 5 10.5072 Maastrichtian O 5 5 O 5 5 O 5 202 O 5 49 0.0198

E 5 7.93 E 5 2.07 E 5 199.07 E 5 51.93


TABLE 7. Results of Analyses 73–86 (sauropod body fossil localities) per European stage. Only stages where bothtitanosaurs and non-titanosaurs are known are shown. See Table 2 for explanation of abbreviations and symbols.

Analysis no. Stage

Non-titanosaurs Titanosaurs


73** Kimmeridgian O 5 235 O 5 23 O 5 0 O 5 5 , 1 3 1025


E 5 222.57 E 5 52.43 E 5 5.67 E 5 1.3375 Berriasian O 5 17 O 5 0 O 5 6 O 5 0 1

E 5 17.00 E 5 0 E 5 6.00 E 5 076 Valanginian O 5 10 O 5 0 O 5 15 O 5 0 1

E 5 10.00 E 5 0 E 5 15.00 E 5 077 Hauterivian O 5 22 O 5 4 O 5 16 O 5 0 0.0094


E 5 36.8 E 5 9.20 E 5 12.00 E 5 3.0079 Aptian O 5 32 O 5 6 O 5 22 O 5 0 0.0525

E 5 34.07 E 5 3.93 E 5 19.72 E 5 2.2880 Albian O 5 42 O 5 4 O 5 23 O 5 0 0.1552


E 5 19.05 E 5 0.95 E 5 25.71 E 5 1.2982 Turonian O 5 10 O 5 0 O 5 41 O 5 0 1

E 5 10.00 E 5 0 E 5 41.00 E 5 083 Coniacian O 5 4 O 5 0 O 5 40 O 5 0 1

E 5 4.00 E 5 0 E 5 40.00 E 5 084 Santonian O 5 2 O 5 0 O 5 40 O 5 0 1


E 5 2.93 E 5 0.07 E 5 80.98 E 5 2.0286* Maastrichtian O 5 4 O 5 2 O 5 113 O 5 0 , 1 3 1025

E 5 5.90 E 5 0.10 E 5 111.08 E 5 1.92

TABLE 8. Results of Analyses 87–98 (sauropod tracksite localities) per European stage. Only stages where bothtitanosaurs and non-titanosaurs are known are shown. See Table 2 for explanation of abbreviations and symbols.

Analysis no. Stage

Narrow-gauge Wide-gauge


87 Bajocian O 5 1 O 5 1 O 5 0 O 5 1 0.3913E 5 0.67 E 5 1.33 E 5 0.33 E 5 0.67

88 Bathonian O 5 6 O 5 7 O 5 0 O 5 2 0.2106E 5 5.57 E 5 7.43 E 5 0.86 E 5 1.14

89 Callovian O 5 1 O 5 1 O 5 1 O 5 0 0.3871E 5 1.33 E 5 0.67 E 5 0.67 E 5 0.33

90 Kimmeridgian O 5 4 O 5 8 O 5 3 O 5 5 0.8495E 5 4.20 E 5 7.80 E 5 2.80 E 5 5.20

91 Tithonian O 5 3 O 5 9 O 5 6 O 5 9 0.4113E 5 4.00 E 5 8.00 E 5 5.00 E 5 10.00

92 Berriasian O 5 0 O 5 1 O 5 7 O 5 2 0.1073E 5 0.70 E 5 0.30 E 5 6.30 E 5 2.70

93 Valanginian O 5 0 O 5 1 O 5 6 O 5 1 0.0641E 5 0.75 E 5 0.25 E 5 5.25 E 5 1.75

94 Hauterivian O 5 0 O 5 1 O 5 6 O 5 1 0.0641E 5 0.75 E 5 0.25 E 5 5.25 E 5 1.75

95* Barremian O 5 0 O 5 1 O 5 11 O 5 0 3 3 1024

E 5 0.92 E 5 0.08 E 5 10.08 E 5 0.9296* Aptian O 5 0 O 5 1 O 5 49 O 5 1 ,1 3 1025

E 5 0.96 E 5 0.04 E 5 48.04 E 5 1.9697 Albian O 5 1 O 5 0 O 5 78 O 5 24 0.5826

E 5 0.77 E 5 0.23 E 5 78.23 E 5 23.7798 Campanian O 5 3 O 5 0 O 5 2 O 5 2 0.1455

E 5 2.14 E 5 0.86 E 5 2.86 E 5 1.14


these associations. Whether these associationsrepresent genuine habitat preferences or aremerely sampling artifacts is also addressed,with possible explanations provided for eachhypothesis.

Taxonomic Level and Association Polarity.—All of the statistically significant resultswere obtained in analyses involving titano-saurs/non-titanosaurs, titanosauriforms/non-titanosauriforms, and/or wide-gauge/narrow-gauge trackways, whereas none of the resultsinvolving Macronaria/non-Macronaria werestatistically significant. These results indicatethat Butler et al.’s (2007) and Butler andBarrett’s (2008) suggestion that macronariansare positively associated with inland habitats isprobably incorrect (N.B. these authors were

also skeptical about the significance of theirmacronarian result, suggesting that it mighthave been generated by a taphonomic artifact).By limiting their data set to Cretaceous forms,Butler and colleagues effectively restrictedcomposition of this macronarian group largelyto titanosauriform taxa because virtually allnon-titanosauriform macronarians are Jurassicin age (e.g., Camarasaurus). Had they insteadconsidered all sauropods, the positive associa-tion between macronarians and inland envi-ronments would have been much weaker orperhaps nonexistent. The reason our analysesat the Macronaria/non-Macronaria level, evenwhen restricted to Cretaceous forms, failed tosupport an environmental association is thatwe used a larger data set: when marine

TABLE 9. Results of Analyses 99–115 (sauropod tracksite and body fossil localities) per European stage. Only stageswhere both titanosaurs and non-titanosaurs are known are shown. See Table 2 for explanation of abbreviationsand symbols.

Analysis no. Stage

Non-titanosaurs andnarrow-gauge

Titanosaurs andwide-gauge


99 Bajocian O 5 16 O 5 1 O 5 0 O 5 1 0.0049E 5 15.00 E 5 2.00 E 5 0.88 E 5 0.12

100 Bathonian O 5 22 O 5 9 O 5 0 O 5 2 0.0381E 5 21.00 E 5 10.00 E 5 1.35 E 5 0.65

101 Callovian O 5 21 O 5 1 O 5 1 O 5 0 0.8144E 5 20.95 E 5 1.05 E 5 0.95 E 5 0.05

102** Kimmeridgian O 5 239 O 5 31 O 5 3 O 5 10 ,1 3 1025

E 5 241.86 E 5 38.14 E 5 11.23 E 5 1.77103** Tithonian O 5 228 O 5 59 O 5 6 O 5 16 ,1 3 1025


E 5 16.36 E 5 1.64 E 5 13.64 E 5 1.36105 Valanginian O 5 10 O 5 1 O 5 21 O 5 1 0.6122

E 5 10.33 E 5 0.67 E 5 20.67 E 5 1.33106 Hauterivian O 5 22 O 5 5 0 5 22 O 5 1 0.0416


E 5 38.51 E 5 8.49 E 5 21.31 E 5 4.69108 Aptian O 5 32 O 5 7 O 5 71 O 5 1 0.0014

E 5 36.14 E 5 2.86 E 5 66.72 E 5 5.28109 Albian O 5 43 O 5 4 O 5 101 O 5 24 0.0932


E 5 18.64 E 5 1.36 E 5 27.02 E 5 1.98111 Turonian O 5 10 O 5 0 O 5 45 O 5 1 0.6383

E 5 9.81 E 5 0.19 E 5 45.15 E 5 0.85112 Coniacian O 5 4 O 5 0 O 5 44 O 5 1 0.7655

E 5 3.91 E 5 0.09 E 5 44.04 E 5 0.96113 Santonian O 5 2 O 5 0 O 5 40 O 5 0 1


E 5 5.73 E 5 0.27 E 5 83.05 E 5 3.95115* Maastrichtian O 5 4 O 5 2 O 5 115 O 5 4 2 3 1024

E 5 5.76 E 5 0.24 E 5 114.16 E 5 4.84


TABLE 10. Results of Analyses 116–134 (‘Sensitivity analyses). See Table 2 for explanation of abbreviationsand symbols.


116* Trackway individuals minus Korean data: T 5 519 T 5 122 T 5 397 , 1 3 1025

Narrow-gauge: T 5 183 O 5 26 O 5157E 5 43.02 E 5 139.98

Wide-gauge T 5 336 O 5 96 O 5 240E 5 78.98 E 5 257.02

117** Body fossil and trackway individuals minus Koreandata: T 5 1879

T 5 1348 T 5 531 , 1 3 1025



118## Body fossil and trackway individuals minus Koreandata: T 5 1879

T 5 1348 T 5 531 , 1 3 1025



119 Trackway localities minus Korean data: T 5 117 T 5 43 T 5 74 0.2432Narrow-gauge: T 5 38 O 5 11 O 5 27

E 5 13.97 E 5 24.03Wide-gauge: T 5 80 O 5 32 O 5 48

E 5 29.40 E 5 50.60120 Body fossil and trackway localities minus Korean

data: T 5 823T 5 667 T 5 156 0.5896



121 Body fossil and trackway localities minus Koreandata: T 5 823

T 5 667 T 5 156 0.0513



122 Trackway localities with Korea reduced to the 6 mainlocalities: T 5 123

T 5 49 T 5 74 0.23


Wide-gauge: T 5 85 O 5 37 O 5 48E 5 33.86 E 5 51.14

123* Trackway individuals minus USA data: T 5 502 T 5 222 T 5 280 ,1 3 1025


Wide-gauge: T 5 361 O 5 195 O 5 166E 5 159.65 E 5 201.35

124 Body fossil and trackway individuals minus USAdata: T 5 1862

T 5 1448 T 5 414 0.1143



125 Body fossil and trackway individuals minus USAdata: T 5 1862

T 5 1448 T 5 414 0.0044



126* Trackway localities minus USA data: T 5 157 T 5 107 T 5 50 ,1 3 1025


Wide-gauge: T 5 124 O 5 96 O 5 28E 5 84.51 E 5 39.49


deposits and indeterminate sauropod speci-mens are excluded, the Butler and Barrett(2008) data set comprises 177 Cretaceoussauropod-bearing localities, whereas theequivalent figure for our data set is 475. Thisincrease reflects additions since Butler col-lected his data in 2006 as well as personalobservations of museum collections that addedsubstantially to our data set.

Determination of the environmental asso-ciations of Titanosauriformes and Titano-

sauria is bound up with the issue of polarity(i.e., whether the positive association is withinland or coastal habitats). Of the 41 analysesthat produced statistically significant results,24 support a positive association betweentitanosaurs, titanosauriforms, and/or wide-gauge trackways and inland environments,and 17 support the opposite (negative) asso-ciation. For ease of discussion, we will termthese conflicting patterns ‘‘titanosaurs preferinland’’ and ‘‘titanosaurs prefer coastal.’’

TABLE 10. Continued.


127 Body fossil and trackway localities minus USA data:T 5 863

T 5 731 T 5 132 5 3 1024



128 Body fossil and trackway localities minus USA data:T 5 863

T 5 731 T 5 132 0.2121

Non-titanosauriforms and narrow-gauge: T 5 404 334 (342.21) 70 (61.79)Titanosauriforms and wide-gauge: T 5 483 O 5 405 O 5 78

E 5 409.12 E 5 73.88129* Trackway individuals minus Korean and USA data:

T 5 388T 5 108 T 5 280 3 3 1024


Wide-gauge: T 5 251 O 5 85 O 5 166E 5 69.87 E 5 181.13

130 Body fossil and trackway individuals minus Koreanand USA data: T 5 1748

T 5 1334 T 5 414 6 3 1024



131## Body fossil and trackway individuals minus Koreanand USA data: T 5 1748

T 5 1334 T 5 414 1 3 1025



132 Trackway localities minus USA and Korean data: T 5

84T 5 34 T 5 50 0.0788


Wide-gauge: T 5 54 O 5 26 O 5 28E 5 21.86 E 5 32.14

133 Body fossil and trackway localities minus Korean andUSA data: T 5 790

T 5 658 T 5 132 0.0241



134 Body fossil and trackway localities minus Korean andUSA data: T 5 790

T 5 658 T 5 132 0.2102




Detailed examination of these results revealssome important and interesting patterns thathelp to resolve this conflict.

The results of Analyses 1–10 should carrythe greatest weight because they are based onthe most data. Analyses 1–5 support the‘‘titanosaurs prefer inland’’ pattern, and allbut Analysis 3 are independent of each otherin that they do not share any data in common.Only one (no. 8) of the four analyses based onthe total data set supports the ‘‘titanosaursprefer coastal’’ pattern. One way to reconcilethis contradiction is to postulate that basaltitanosauriforms were indeed positively asso-ciated with coastal environments, and that aless inclusive clade such as Titanosauriasubsequently switched its preference to onefor inland habitats. This proposal receivessupport from three separate lines of evidence:

1. Analyses 11 and 12 partition Titanosaur-iformes into basal forms (non-titanosaurs)and titanosaurs. The results support posi-tive associations between the basal formsand coastal environments and between themore derived titanosaurs and inland en-vironments.

2. Of the 101 time-sliced analyses (Tables 3–9), 14 support the ‘‘titanosaurs preferinland’’ pattern and 12 support the oppo-site pattern. However, 11 of the 12 contra-dictory analyses were generated by Middleand Late Jurassic time slices, whereas all 14of the results supporting the inland pref-erence were generated by Cretaceous timeslices. Many of these time-sliced analysesare non-independent because they overlaptemporally or they are based on commondata (e.g., data sets comprising body fossilindividuals and data sets comprising thesedata plus trackway individuals). The 52stage-level time-sliced analyses in Ta-bles 4, 5, 7, and 8, however, effectivelyuse independent data sets (if we assume,for example, that the number of individ-uals is independent from the number oflocalities). Six of these (all Jurassic) supportthe coastal preference and eight (all Cre-taceous) support the preference for inlandhabitats. This marked temporal division isconsistent with the view that the earliest

and most basal titanosauriforms and/ortitanosaurs occurred more often than ex-pected in coastal habitats, but the morederived Cretaceous forms displayed apreference for inland habitats.

3. The positive association between earlyand/or basal titanosaurs and coastal habi-tats could partly be an artifact of taxon-omy. There are very few confirmed titano-saur body fossils from the Jurassic: themain evidence is Janenschia robusta fromthe Kimmeridgian–Tithonian of Tenda-guru, Tanzania. Although considered atitanosaur by many workers (Janensch1929; McIntosh 1990; Jacobs et al. 1993;Upchurch 1995; Wilson and Sereno 1998),both the titanosaurian affinities and con-generic status of the eight individualscurrently assigned to this taxon haverecently been doubted (Bonaparte et al.2000). Given that the dinosaur-bearingbeds of Tendaguru represent coastal habi-tats (Aberhan et al. 2002), the incorrectassignment of all or some of the Janenschiaindividuals to the Titanosauria could haveobscured the ‘‘titanosaurs prefer inlandhabitats’’ pattern, especially for Jurassictime slices.

Habitat Preference or Sampling Artifact?—Asdiscussed earlier, a statistically significantassociation between an assemblage of taxaand a given environment might not, by itself,provide evidence of a genuine habitat pref-erence. Skewed distributions can arise as aresult of other factors, such as long-termtrends in taxon diversity and/or abundancecombined with long-term trends in therelative sampling rates of the different envi-ronmental categories. The observations thattitanosaurs appear to be more diverse andabundant during the Cretaceous than duringthe Jurassic, and that there are approximatelythree times as many coastal localities (relativeto the total number of localities) in theJurassic compared to the Cretaceous, raiseconcerns that the skewed distributions ofsauropods are artifacts rather than genuineecological signals. However, when analysesare run at the stage level (i.e., Analyses 30–115: Tables 4–9), we find that statistically


significant environmental associations persist,even though such analyses disrupt the effectsof the long-term trend in the relative sam-pling of inland and coastal environments.

Further support for genuine ecologicalpreferences comes from the observation thatonly eight localities in our data set haveyielded both titanosaurs and non-titanosaurs.Three of these eight localities are Jurassic andcoastal whereas the other five are Cretaceousand inland localities (see SupplementaryMaterials), which is consistent with thehypothesis that the earlier and more basaltitanosaurs were more likely to occur incoastal environments than the Cretaceousforms. If titanosaurs and non-titanosaurshad occupied all environments with nodiscernible preference, many more localitiesshould yield evidence for both types ofsauropod coexisting. Although our simpledivision into inland and coastal categoriesmay be so crude as to partially obscure theprecise nature of ecological preferences (seebelow), the relative rarity of ‘‘shared local-ities’’ points to genuine ecological partition-ing or separation between titanosaurs andnon-titanosaurs rather than mere samplingartifacts. We tentatively suggest, therefore,that our results indicate evidence for habitatpreferences among sauropod groups, and arenot merely artifacts created by trends in therelative sampling of different environments.

Strength of the Habitat Preference.—The rawdata (see Supplementary Materials) and the‘‘observed’’ and ‘‘expected’’ values in Ta-bles 2–10 demonstrate that titanosaur bodyfossils and/or wide-gauge trackways oftenoccur in coastal habitats, and non-titanosaurbody fossils and narrow gauge trackwaysoften occur in inland habitats. For example,consider Analysis 1 (Table 2). In this case(body fossil individuals), there are 864 non-titanosaurs (expected value 5 889.75) and 362titanosaurs (expected value 5 336.25) ininland environments. There are fewer non-titanosaurs and more titanosaurs than ex-pected, and this skew is large enough to resultin a statistically significant p-value. Never-theless, the observed values indicate thatmany non-titanosaurs were present in inlandenvironments even though they apparently

displayed a preference for coastal environ-ments. Similar observations were made forCretaceous herbivorous dinosaurian cladesby Butler and Barrett (2008). Such resultsmight be interpreted in two ways, which weterm the ‘‘weak preference hypothesis’’ andthe ‘‘strong preference plus noise hypoth-esis.’’

The ‘‘weak preference hypothesis’’ sug-gests that the occurrences of non-titanosaursand titanosaurs across the inland and coastalenvironmental categories represent a largelyaccurate picture of sauropod distributions.Thus, the skewed environmental distributionswould reflect a subtle difference between thetwo groups, such as the relative amounts oftime that members of each group spent ineach environment, or the relative abundancesof each group. If this interpretation is correct,then the difference between titanosaurs andnon-titanosaurs should be relatively minor,just large enough to produce a statisticallydetectable skew in a large data set.

Alternatively, the ‘‘strong preference plusnoise hypothesis’’ suggests that the habitatpreferences of titanosaurs and non-titano-saurs were significantly different, but thatthe strength of this signal has been reducedby ‘‘noise’’ in the data set. There are severalpossible sources of such noise:

N Estimating the number of individuals basedon body fossils or trackways is imprecise,and determining the number of localitiescan also be problematic.

N Evidence of habitat ‘‘occupation’’ mayactually be the result of postmortem trans-port of body fossils. This could be tested byrepeating our analyses using a more strin-gently filtered version of our data set (i.e.,by excluding disarticulated or very incom-plete specimens), an endeavor that liesoutside of the scope of the current study.

N The partitioning of taxa into two categoriescould introduce noise into an analysis.Suppose, for example, that a genuinedifference in habitat preferences existedbetween titanosaurs and non-titanosaurs.If our analyses are limited to titanosauri-forms versus non-titanosauriforms, the re-sults might still be statistically robust


because the taxonomic contents of Titano-sauriformes and Titanosauria are verysimilar. Nonetheless, Titanosauriformescontains some non-titanosaurs that exhib-ited a preference for coastal environments,and by including these taxa within the samecategory as members of Titanosauria, wewould weaken the environmental associa-tions signal.

N Taxonomic and phylogenetic errors couldobscure or weaken an environmental asso-ciation signal. Because the phylogeneticrelationships of basal titanosaurs and basaltitanosauriforms are poorly understood(Upchurch et al. 2004; Curry Rogers2005), the contents of our titanosaur/non-titanosaur and titanosauriform/non-titano-sauriform categories could be inaccurate.Future developments in sauropod phyloge-netics might result in some of the taxa wehave classified as titanosaurs or titanosauri-forms shifting in relative position, producinga strengthening or weakening of the putativeenvironmental associations signal (e.g., seediscussion of Janenschia, above).

N Henderson (2006) modeled the position ofthe center of mass in several sauropods andargued that all large sauropods (over,12 tons) would have been constrained toadopt a wide-gauge stance in order tomaintain stability during locomotion. Thiswould mean that many large non-titano-saurian taxa, such as Apatosaurus, Camara-saurus, and Turiasaurus, would have pro-duced wide-gauge trackways, potentiallyintroducing errors into all of our ‘‘combinedevidence’’ analyses where we have groupedtitanosaur body fossils with wide-gaugetrackways and non-titanosaur body fossilswith narrow-gauge trackways. However,many large-bodied non-titanosaurian sau-ropods lack most or all of the modificationsto the hindlimb which Wilson and Carrano(1999) identified as adaptations for a wide-gauge stance.

N Our simple division of habitats into inlandversus coastal types might partially obscurethe true habitat preferences of titanosaursand non-titanosaurs. If, for example, titano-saurs actually preferred relatively aridconditions, and tended to occupy semi-arid

inland environments, then including mesichabitats (e.g., fluvio-lacustrine facies) in theinland category could hide their truehabitat preferences. This may explain why,as discussed above, titanosaurs and non-titanosaurs are rarely found at the samelocalities despite occurring in both habitatcategories. Additionally, reports of sauro-pod skeletons from ‘‘fluvial’’ (inland) en-vironments may obscure the fact that asetting is actually much closer to the coast-line (J. A. Wilson personal communication2009).

We suggest that elements of both the ‘‘weakpreference hypothesis’’ and ‘‘strong prefer-ence plus noise hypothesis’’ are supported byour data set and analyses. Despite thestatistically significant environmental associa-tions of titanosaurs and non-titanosaurs, bothtypes of sauropods probably spent consider-able time in both types of habitat. However,one or more sources of noise probably haveblurred the habitat preference signal, perhapsmaking it appear much more subtle than itwas.

Nature of the Habitat Preference.—If sauro-pods did have habitat preferences, then (1)which aspects of the environments wererelevant to each group’s preference? and (2)Are any of the morphological differencesbetween titanosaurs and non-titanosaurslinked to these habitat preferences? Belowwe outline two broad hypotheses that mightaccount for the proposed habitat preferences.

The resource exploitation hypothesis sug-gests that the habitat preference is linked toparticular resources in each habitat (such asdifferent types of plant fodder). If thishypothesis is correct, particular plant typesshould display nonrandom associations withinland and coastal habitats, and titanosaursand non-titanosaurs should have possesseddifferent feeding mechanisms adapted toexploit these particular resources. Althoughthe patchiness of our sampling of both thesauropod and plant fossil records hamperstesting of this idea, several distinctive fea-tures of titanosaur skulls and postcraniaplausibly can be linked to novel feedingmechanisms (Calvo 1994; Upchurch and


Barrett 2000; Curry Rogers and Forster 2001;Wilson 2002, 2005b; Upchurch et al. 2004;Barrett and Upchurch 2005). For example, ifthe wide-gauge stance, increased flexibilityof the dorsal vertebral column, short pro-coelous tails, and anteriorly flaring ilia oftitanosaurs are related to a tripod stance(Borsuk-Bialynicka 1977; Powell 1992; Wil-son and Carrano 1999), then such a posturemay have been used during high-browsingon particular types of plant (though seeHenderson 2006).

The locomotion/stance hypothesis proposes alinkage between the wide-gauge stance oftitanosaurs (and other anatomical specializa-tions) and some physical (perhaps topograph-ical or substrate-related) aspect of inlandenvironments. The adaptive significance ofthe wide-gauge stance is still poorly under-stood, though there can be little doubt thatthis titanosaurian feature would have had amajor effect on many aspects of locomotionand behavior. If, for example, it increased theanimal’s stability, it might thus have facili-tated crossing of uneven or sloping terrain.Wilson and Carrano (1999) noted that thewide-gauge stance is associated with severalother anatomical modifications, all of whichsuggest that titanosaurs had a wider range ofmotion in the trunk and tail regions and in thefore and hind limbs, which collectively mighthave enhanced their ability to rear into atripod stance (Wilson and Carrano 1999) and/or move more quickly (Apesteguıa 2005).Thus, even if inland and coastal habitatspossessed approximately the same resources,titanosaurs may have found it easier toexploit these resources in the inland habitatsthan did non-titanosaurs. This hypothesiscould be tested by using the biomechanicalapproaches proposed by Henderson (2006)and Hutchinson et al. (2007) to modeltitanosaurs and non-titanosaurs walking andturning at different speeds on a variety ofterrains and substrates.

Sauropod Evolutionary History

Figure 2 illustrates how the diversity of non-titanosaur lineages declined through the Cre-taceous, while, at the same time, titanosaursradiated strongly (see also Barrett and Up-

church 2005; Upchurch and Barrett 2005). Whytitanosaurs should have been so scarce duringthe Jurassic and so dominant in the Cretaceous(especially the Late Cretaceous) is not under-stood, though this pattern probably reflectssampling biases in the fossil record. It isinteresting to note, for example, that titano-saurian body fossils are extremely scarceduring the Jurassic (0.01% of the Jurassic partof our body fossil individual data set), whereaswide-gauge trackways and tracksite localitiesmake up 58% and 42% of the Jurassic track dataset, respectively. This disparity seems anom-alous and may indicate that early titanosaursoccupied environments with low preservationpotentials for body fossils, and/or that some ofthe sauropod taxa known from the Middle andLate Jurassic might be currently unrecognizedmembers of the basal titanosaurian radiation.The decline of non-titanosaurs throughout theEarly and early Late Cretaceous is less easilyexplained as a sampling artifact because weobserve a decrease in the abundance anddiversity of non-titanosaurian body fossilsand narrow-gauge trackways (both are absentfrom the Coniacian onwards). Table 1 showsthat the number of coastal localities producingherbivorous dinosaur material decreases mark-edly from the Jurassic to the Cretaceous. Itseems very improbable that there was agenuine decrease in the number or areal extentof coastal habitats during the Cretaceous; ifanything, continental fragmentation during theCretaceous should have increased the amountof available coastline. If the relative extent ofcoastal to inland habitats remained the same(or even increased) during the Cretaceous, thenthe observation that fewer herbivorous dino-saurs were living in coastal environmentsrequires explanation. Sauropods, and perhapscertain ornithischian groups, might have beenforced to occupy inland habitats more fre-quently during the Cretaceous because coastalenvironments became less hospitable. If non-titanosaurs were less well equipped thantitanosaurs to deal with conditions and/orresources in the inland habitats, this mighthave contributed to their decline in the EarlyCretaceous.

In recent years, several authors have re-marked on the convergence in feeding and/or


locomotor systems between diplodocoids(particularly rebbachisaurids) and titanosaurs(Upchurch 1999: pp. 119–120; Curry Rogersand Forster 2004; Apesteguıa 2005; Barrettand Upchurch 2005). Rebbachisaurids seemto have diversified during the Early Creta-ceous and were the last non-titanosauriangroup of sauropods to go extinct. Althoughthese observations suggest that the conver-gence was driven by environmental changesduring the Early and early Late Cretaceous,data on the environmental associations ofthese groups, particularly rebbachisaurids,are too scarce to confirm this. However, it isworth noting that of the 25 rebbachisauridsincorporated into this analysis, only one (theputative form, Amazonsaurus [Carvalho et al.2003]) was recovered from a coastal environ-ment, suggesting a possible ‘‘environmentalconvergence’’ between titanosaurs and rebba-chisaurids.

Our data on the numbers of inland andcoastal localities are based on localities whereherbivorous dinosaurs are found. A morerigorous quantitative approach to analyzingenvironmental changes through the Jurassicand Cretaceous would include data on theareal extent of inland and coastal facies,including sediments that do not containdinosaur fossils. A second line of inquirywould be to examine how potential forageplants are associated with habitat type andwhether these plants declined in diversityand/or abundance during the Cretaceous.

Wider Implications—Methodological Issues

Although this study focuses on sauropodpaleoecology, our analytical protocols andresults raise issues that are of much widersignificance, especially with regard to themethodology of establishing environmentalassociations. Some key points are discussedbriefly below.

Body Fossils Compared with Trackways.—Body fossils provide the bulk of informationon sauropods. For example, in our data set706 and 190 localities yielded sauropod bodyfossils and trackways respectively (Supple-mentary Materials; Fig. 3). Likewise, the con-tributions of body fossils and trackways to

Butler and Barrett’s (2008) data set on Creta-ceous herbivorous dinosaurs were 92.5% and7.5% respectively. Body fossils also have theadvantage of being assignable (often) todistinct clades (e.g., Brachiosauridae, Salt-asauridae, Dicraeosauridae) or particular gen-era and species. However, body fossils can betransported to different environments afterthe animal dies, whereas trackways provide adirect record of where the animal actuallystood while alive (Thulborn 1982; Lockley1991; Wilson and Carrano 1999; Carrano andWilson 2001). Our analyses provide an op-portunity to compare the relative perfor-mances of trackway-based and body fossil-based data sets. A survey of Tables 2–10reveals that both body fossil data (13 anal-yses) and trackway data (14 analyses) yieldstatistically significant results, and trackwaydata do not in general produce lower p-valuesthan the body fossil data. If postmortemtransport of body fossils has had a strongmasking effect on habitat preferences, thenthe analyses based solely on trackwaysshould have provided stronger support forenvironmental associations than those basedjust on body fossils.

Localities versus Individuals.—The use ofnumbers of individuals in this type ofpaleoecological study appears to be novel, atleast with regard to vertebrates. Our defini-tion of a ‘‘locality’’ is perhaps more arbitrarythan the definition of an ‘‘individual,’’ butbecause of the error that could be associatedwith estimating numbers of individuals fromtrackways or from fragmentary skeletons, wehad expected that numbers of individualswould provide a less reliable guide toenvironmental associations. However, ouranalyses suggest that, if anything, individ-ual-based analyses are more likely to findevidence for environmental associations thanare locality-based ones. This phenomenonmay be related to the fact that the numberof individuals cannot be less than, and willoften exceed, the number of localities. If theskew in the spatial distributions of the twotaxon categories is subtle, then individual-based analyses will reveal the skew betterbecause of the larger number of data points.Alternatively, even if there are substantial


errors in the estimation of the numbers ofindividuals, these errors might be randomwith respect to environmental and taxoncategories. If, for example, a habitat prefer-ence is expressed in terms of how much timeeach taxon spends in a given habitat, or therelative abundances of these taxa in eachhabitat, locality-based estimates of occur-rences cannot capture this information be-cause they record only presence or absence ofa taxon. Thus, individual counts may beuseful in capturing aspects of paleoecologythat are ignored by locality-based counts,even when estimating the numbers of individ-uals is prone to significant error. This hasparallels with modern ecological studies at-tempting to assess population size, where totalcounts are often impractical as a consequenceof time, costs, and size of area (Waite 2000).Consequently, population size must be esti-mated with alternative techniques such assample counts or capture-mark-recapturemethods (Burnham and Overton 1979; Bloweret al. 1981; Chao 1987; Waite 2000), using avariety of statistical approaches (see Colwelland Coddington 1994; Krebs 1999; Waite 2000;Sutherland 2006). Although some of theseanalytical methods have been implementedin paleoecological analyses (e.g., Harringtonand Jaramillo 2007), they have yet to beapplied to fossil vertebrates.

The Costs and Benefits of Time-Slicing andSensitivity Analyses.—We have outlined abovea hypothetical scenario in which paralleltrends in the diversity of two taxon assem-blages and the preservation rates of twoenvironmental categories could create artifac-tual support for environmental associations.Exploring various subsets of the data canameliorate these difficulties. One of the mostuseful approaches is to time slice the data inorder to disrupt long-term trends in diversityand environmental preservation. Althoughsuch studies can reveal that putative environ-mental associations have changed throughtime, the key disadvantage of time-slicing isthat as time slices become narrower, theyinclude fewer data points. Thus, even thoughthe data set as a whole contains such signals,numerous analyses may fail to find anystatistically significant results (see Tables 3–9

for examples). Time-slicing is a key tool in thesearch for environmental associations becauseit allows the researcher to ‘‘fine-tune’’ thetemporal range and taxonomic level of theproposed signals, but it should be appliedwith caution because it can be misleadingabout the taxonomic level at which a pro-posed environmental association occurs (e.g.,Cretaceous ‘‘Macronaria’’ has virtually thesame taxonomic content as ‘‘Titanosauri-formes’).

In this study, we re-analyzed our data afterremoving the Korean and U.S. tracksite dataand found that our conclusions regardingenvironmental associations are affected by thepresence or absence of these data blocks,although several analyses continue to supporta positive association between titanosaursand inland habitats (Table 10). Uneven sam-pling of the fossil record is a major topic ofconcern, particularly with regard to thetemporal distributions of fossil taxa and thereconstruction of diversity curves (Raup 1972;Smith 2001; Peters and Foote 2001, 2002;Peters 2005, 2008; Upchurch and Barrett2005; Smith and McGowan 2007; McGowanand Smith 2008; Barrett et al. 2009; Butler et al.2009). The effect of uneven sampling onanalyses of the spatial distribution of taxahas received considerably less attention, eventhough the ambiguity of ‘‘absence’’ likelyaffects both paleobiogeographic and paleoeco-logical analyses (Ronquist 1997; Lieberman2000; Hunn and Upchurch 2001; Upchurchand Hunn 2002). Ultimately, the analysis ofthe spatial distributions of fossil taxa mightbenefit from some form of rarefaction ap-proach in which repeated subsamples of thedata are selected at random and analyzed forassociations or nonrandom area relationships.However, integration of such methods intothe protocol for searching for environmentalassociations would require the creation ofspecialized software and lies outside of thescope of our study. Pending emergence ofsuch software, we urge other researchers toexplore their data sets by removing majorblocks of data, changing the boundaries oftime slices, and altering the criteria used todefine localities and estimate numbers ofindividuals.


Conclusion

Our results suggest that titanosaurs andnon-titanosaurs display statistically signifi-cant associations with inland and coastalenvironments respectively. These signals oc-cur when body fossils and trackway data aretreated separately and together, for bothlocality-based and individual-based countsof occurrences. We interpret this pattern tomean that sauropod groups displayed habitatpreferences, although the precise nature andstrength of this preference are not clear atpresent. Wilson and Carrano’s (1999) hypoth-esis that wide-gauge trackways were made bytitanosaurs is reinforced by the observationthat separate analyses of titanosaur bodyfossils and wide-gauge trackways displaythe same positive association with inlandhabitats. Finally, although the decline ofnon-titanosaurs and diversification of titano-saurs during the Early and early Late Creta-ceous cannot be linked directly to habitatpreferences, a better understanding of suchpreferences may help explain these events inthe future.

Ecologists and invertebrate paleontologistshave been investigating environmental asso-ciations for several decades, but the search forstatistically robust associations among fossilvertebrate taxa is in its infancy. Although thislag is partly the result of suitable databasesonly recently becoming available (e.g., thePaleobiology Database), this is a long overduefocus for the field of vertebrate paleontology.It is crucial that paleobiologists test theirecological and evolutionary hypotheses usinganalytical methods and statistical tests thatcan distinguish genuine signals from thebackground noise generated by missing dataand sampling biases. At the same time, thesetechniques must be applied, and their resultsinterpreted, with subtlety and caution. Thecurrent study has demonstrated that both‘‘total evidence’’ and time-slicing approacheshave their costs and benefits, and that paralleltrends in diversity and the representation ofenvironments can create skews in spatialdistributions that result in statistically sig-nificant but nonetheless artifactual supportfor environmental associations. We hope,

therefore, that this study not only sheds somelight on the evolution of sauropod dinosaurs,but also will stimulate more detailed quanti-tative analyses of ecological relationships inother extinct organisms.

Acknowledgments

We are grateful to the numerous institu-tions that allowed the study of material usedin compiling the data set. The authors wouldlike to express their gratitude to R. J. Butlerfor help with the statistical element of thiswork, as well as for providing data that wasunpublished at the time, and to P. M. Barrettfor helpful comments and discussion. Re-views by J. A. Wilson, M. Uhen, and C.Redman greatly improved the quality of thiswork, as did editorial comments by M. T.Carrano. Lastly, suggestions by N. Atkinsgreatly improved the clarity of this work. P.D. Mannion’s research was supported by aUniversity College London Natural Environ-ment Research Council (NERC) studentship(NER/S/A/2006/14347).

Literature Cited

Aberhan, M., R. Bussert, W.-D. Heinrich, E. Schrank, S. Schultka,

B. Sames, J. Kriwet, and S. Kapilima. 2002. Paleoecology and

depositional environments of the Tendaguru Beds (Late

Jurassic to Early Cretaceous, Tanzania). Mitteilungen aus dem

Museum fur Naturkunde in Berlin, Geowissenschaftliche Reihe

5:17–42.

Apesteguıa, S. 2005. Evolution of the hyposphene-hypantrum

complex within Sauropoda. Pp. 248–267 in Carpenter and

Tidwell 2005.

Badgley, C. E. 1986. Counting individuals in mammalian fossil

assemblages from fluvial environments. Palaios 1:328–338.

Barrett, P. M., and P. Upchurch. 2005. Sauropod diversity through

time: possible macroevolutionary and paleoecological implica-

tions. Pp. 125–156 in Curry Rogers and Wilson 2005.

Barrett, P. M., A. J. McGowan, and V. Page. 2009. Dinosaur

diversity and the rock record. Proceedings of the Royal Society

B 276:2667–2674.

Blower, J. G., L. M. Cook, and J. A. Bishop. 1981. Estimating the

size of animal populations. Allen and Unwin, London.

Bonaparte, J. F., W.-D. Heinrich, and R. Wild. 2000. Review of

Janenschia Wild, with the description of a new sauropod from

the Tendaguru beds of Tanzania and a discussion on the

systematic value of procoelous caudal vertebrae in the

Sauropoda. Palaeontographica, Abteilung A 256:25–76.

Borsuk-Bialynicka, M. 1977. A new camarasaurid sauropod

Opisthocoelicaudia skarzynskii, gen. n., sp. n. from the Upper

Cretaceous of Mongolia. Paleontologica Polonica 37:1–64.

Buffetaut, E., V. Suteethorn, G. Cuny, H. Tong, J. Le Loeuff, S.

Khansubha, and S. Jongautchariyakul. 2000. The earliest known

sauropod dinosaur. Nature 407:72–74.

Burnham, K. P., and W. S. Overton. 1979. Robust estimation of

population size when capture probabilities vary among

animals. Ecology 60:927–936.


Butler, R. J., and P. M. Barrett. 2008. Paleoenvironmental controls

on the distribution of Cretaceous herbivorous dinosaurs.

Naturwissenschaften 95:1027–1032.

Butler, R. J., P. M. Barrett, P. Kenrick, and M. G. Penn. 2007.

Paleoenvironmental controls on the distribution of Cretaceous

herbivorous dinosaurs. Journal of Vertebrate Paleontology

27(Suppl. to No. 3):54A–55A.

Butler, R. J., P. M. Barrett, S. Nowbath, and P. Upchurch. 2009.

Estimating the effects of the rock record on pterosaur diversity

patterns: implications for hypotheses of bird/pterosaur com-

petitive replacement. Paleobiology 35:432–446.

Calvo, J. O. 1994. Jaw mechanics in sauropod dinosaurs. In M. G.

Lockley, V. F. dos Santos, C. A. Meyer, and A. P. Hunt, eds.

Aspects of sauropod paleobiology. GAIA 10:183–193.

Carpenter, K., and V. Tidwell. 2005. Thunder lizards: the

sauropodomorph dinosaurs. Indiana University Press, Bloo-

mington.

Carrano, M. T. 2008. Taxonomy and classification of non-avian

Dinosauria. Paleobiology Database Online Systematics Archive

4 (www.paleodb.org).

Carrano, M. T., and J. A. Wilson. 2001. Taxon distributions and

the vertebrate track record. Paleobiology 27:563–581.

Carvalho, I. S., L. S. Avilla, and L. Salgado. 2003. Amazonsaurus

maranhensis gen. et sp. nov. (Sauropoda, Diplodocoidea) from

the Lower Cretaceous (Aptian–Albian) of Brazil. Cretaceous

Research 24:697–713.

Chao, A. 1987. Estimating the population size for capture-

recapture data with unequal capturability. Biometrics 43:783–

791.

Colwell, R. K., and J. A. Coddington. 1994. Estimating terrestrial

biodiversity through extrapolation. Philosophical Transactions

of the Royal Society of London B 345:101–118.

Curry Rogers, K. A. 2005. Titanosauria: a phylogenetic overview.

Pp. 50–103 in Curry Rogers and Wilson 2005.

Curry Rogers, K. A., and C. A. Forster. 2001. The last of the

dinosaur titans: a new sauropod from Madagascar. Nature

412:530–534.

——— . 2004. The skull of Rapetosaurus krausei (Sauropoda:

Titanosauria) from the Late Cretaceous of Madagascar. Journal

of Vertebrate Paleontology 24:121–144.

Curry Rogers, K. A., and J. A. Wilson. 2005. The sauropods:

evolution and paleobiology. University of California Press,

Berkeley.

Davis, E. B., and N. D. Pyenson. 2007. Diversity biases in

terrestrial mammalian assemblages and quantifying the differ-

ences between museum collections and published accounts: a

case study from the Miocene of Nevada. Palaeogeography,

Palaeoclimatology, Palaeoecology 250:139–149.

Day, J. J., P. Upchurch, D. B. Norman, A. S. Gale, and H. P.

Powell. 2002. Sauropod trackways, evolution, and behavior.

Science 296:1659.

Day, J. J., D. B. Norman, A. S. Gale, P. Upchurch, and H. P.

Powell. 2004. A Middle Jurassic dinosaur trackway site from

Oxfordshire, UK. Palaeontology 47:319–348.

De Francesco, C. G., and G. S. Hassan. 2008. Dominance of

reworked fossil shells in modern estuarine environments:

implications for paleoenvironmental reconstructions based on

biological remains. Palaios 23:14–23.

Dodson, P., A. K. Behrensmeyer, R. T. Bakker, and J. S. McIntosh.

1980. Taphonomy and paleoecology of the dinosaur beds of the

Jurassic Morrison Formation. Paleobiology 6:208–232.

Farlow, J. O., J. G. Pittman, and J. M. Hawthorne. 1989.

Brontopodus birdi, Lower Cretaceous sauropod footprints from

the U.S. Gulf Coastal Plain. Pp. 371–394 in D. D. Gillette and M.

G. Lockley, eds. Dinosaur tracks and traces. Cambridge

University Press, Cambridge.

Gilinsky, N. L., and J. B. Bennington. 1994. Estimating numbers of

whole individuals from collections of body parts: a taphonomic

limitation of the paleontological record. Paleobiology 20:245–

258.

Gradstein, F. M., J. G. Ogg, and A. G. Smith. 2004. A geological

timescale 2004. Cambridge University Press, Cambridge.

Grayson, D. K. 1973. On the methodology of faunal analysis.

American Antiquity 39:432–439.

Hammer, Ø., and D. A. T. Harper. 2006. Paleontological data

analysis. Blackwell, Oxford.

Harrington, G. J., and C. A. Jaramillo. 2007. Paratropical floral

extinction in the Late Paleocene–Early Eocene. Journal of the

Geological Society, London 164:323–332.

Henderson, D. M. 2006. Burly gaits: centers of mass, stability, and

the trackways of sauropod dinosaurs. Journal of Vertebrate

Paleontology 26:907–921.

Hunn, C. A., and P. Upchurch. 2001. The importance of time/

space in diagnosing the causality of phylogenetic events:

towards a ‘‘chronobiogeographical’’ paradigm? Systematic

Biology 50:1–17.

Hutchinson, J. R., V. Ng-Thow-Hing, and F. C. Anderson. 2007. A

3D interactive method for estimating body segmental param-

eters in animals: application to the turning and running

performance of Tyrannosaurus rex. Journal of Theoretical

Biology 246:660–680.

Jacobs, L. L., D. A. Winkler, W. R. Downs, and E. M. Gomani.

1993. New material of an Early Cretaceous titanosaurid

sauropod dinosaur from Malawi. Paleontology 36:523–534.

Janensch, W. 1929. Material und Formegehalt der Sauropoden in

der Ausbeute der Tendaguru-Expedition, 1909–1912. Palaeon-

tographica 2(Suppl. VII):3–34.

Krebs, C. J. 1999. Ecological methodology, 2d ed. Addison-Wesley

Longman, Menlo Park, Calif.

Lee, Y.-N., S.-Y. Yang, S.-J. Seo, K.-S. Baek, D.-J. Lee, E.-J. Park,

and S.-W. Han. 2000. Distribution and paleobiological signifi-

cance of dinosaur tracks in the Jindong Formation (Albian) in

Kosong County, Korea. Paleontological Society of Korea Special

Publication 4:1–12.

Lehman, T. M. 1987. Late Maastrichtian paleoenvironments and

dinosaur biogeography in the western interior of North

America. Palaeogeography, Palaeoclimatology, Palaeoecology

60:189–217.

Lieberman, B. S. 2000. Paleobiogeography: using fossils to study

global change, plate tectonics, and evolution. Kluwer Academic

/Plenum, New York.

Lim, S.-K., S.-Y. Yang, and M. G. Lockley. 1989. Large dinosaur

footprint assemblages from the Cretaceous Jindong Formation

of southern Korea. Pp. 333–336 in D. D. Gillette and M. G.

Lockley, eds. Dinosaur tracks and traces. Cambridge University

Press, Cambridge.

Lockley, M. G. 1991. Tracking dinosaurs: a new look at an ancient

world. Cambridge University Press, Cambridge.

Lockley, M. G., C. A. Meyer, A. P. Hunt, and S. Lucas. 1994. The

distribution of sauropod tracks and trackmakers. In M. G.

Lockley, V. F. dos Santos, C. A. Meyer, and A. P. Hunt, eds.

Aspects of sauropod paleobiology. GAIA 10:233–248.

Lockley, M. G., K. Houck, S.-Y. Yang, M. Matsukawa, and S.-K.

Lim. 2006. Dinosaur-dominated footprint assemblages from the

Cretaceous Jindong Formation, Hallayo Haesang National

Park, Goseong County, South Korea: evidence and implica-

tions. Cretaceous Research 27:20–101.

Lucas, S. G., and A. P. Hunt. 1989. Alamosaurus and the sauropod

hiatus in the Cretaceous of the North American western

interior. In J. O. Farlow, ed. Paleobiology of the dinosaurs.

Geological Society of America Special Paper 238:75–85.

McGowan, A. J., and A. B. Smith. 2008. Are global Phanerozoic

marine diversity curves truly global? A study of the relation-

ship between regional rock records and global Phanerozoic

marine diversity. Paleobiology 34:80–103.


McIntosh, J. S. 1990. Sauropoda. Pp. 345–401 in D. B. Weishampel,

P. Dodson, and H. Osmolska, eds. The Dinosauria, 1st ed.

University of California Press, Berkeley.

Milan, J., and R. G. Bromley. 2006. True tracks, undertracks and

eroded tracks, experimental work with tetrapod tracks in

laboratory and field. Palaeogeography, Palaeoclimatology,

Palaeoecology 231:253–264.

Ostrom, J. H., and J. S. McIntosh. 1966. Marsh’s dinosaurs. Yale

University Press, New Haven, Conn.

Peters, S. E. 2005. Geological constraints on the macroevolu-

tionary history of marine animals. Proceedings of the National

Academy of Sciences USA 102:12,326–12,331.

———. 2008. Environmental determinants of extinction selectiv-

ity in the fossil record. Nature 454:626–629.

Peters, S. E., and K. B. Bork. 1999. Species-abundance models: an

ecological approach to inferring paleoenvironment and resolv-

ing paleoecological change in the Waldron Shale (Silurian).

Palaios 14:234–245.

Peters, S. E., and M. Foote. 2001. Biodiversity in the Phanerozoic:

a reinterpretation. Paleobiology 27:583–601.

———. 2002. Determinants of extinction in the fossil record.

Nature 416:420–424.

Powell, J. E. 1992. Osteologıa de Saltasaurus loricatus (Sauropoda-

Titanosauridae) del Cretacico Superior del Noroeste Argentino.

Pp. 165–230 in J. L. Sanz and A. D. Buscalioni, eds. Los

dinosaurios y su entorno biotico. Instituto ‘‘Juan de Valdes,’’

Cuenca, Spain.

Raup, D. M. 1972. Taxonomic diversity during the Phanerozoic.

Science 177:1065–1071.

Rice, W. R. 1989. Analyzing tables of statistical tests. Evolution

43:223–225.

Ronquist, F. 1997. Dispersal–vicariance analysis: a new biogeo-

graphic approach to the quantification of historical biogeogra-

phy. Systematic Biology 46:195–203.

Russell, D., P. Beland, and J. S. McIntosh. 1980. Paleoecology of

the dinosaurs of Tendaguru (Tanzania). Memoires de la Societe

Geologique de France 139:169–175.

Salgado, L. 2001. Los sauropodos de Patagonia: sistematuca,

evolucıon y paleobiologıa. Pp. 139–168 in Actas de Las II

Journadas Internacionales sobre Paleontologıa de Dinosaurios

y su Entorno. Salas de los Infantes, Burgos, Spain.

Salgado, L., R. A. Coria, and J. O. Calvo. 1997. Evolution of

titanosaurid sauropods. I. Phylogenetic analysis based on the

postcranial evidence. Ameghiniana 34:3–32.

Scholz, H., and J. H. Hartman. 2007. Paleoenvironmental

reconstruction of the Upper Cretaceous Hell Creek Formation

of the Williston Basin, Montana, USA: implications from the

quantitative analysis of unionoid bivalve taxonomic diversity

and morphologic disparity. Palaios 22:24–34.

Sepkoski, J. J., Jr. 1984. A kinematic model of Phanerozoic

taxonomic diversity. III. Post-Paleozoic families and mass

extinctions. Paleobiology 10:246–267.

Sereno, P. C. 1999. The evolution of dinosaurs. Science 284:2137–

2147.

Smith, A. B. 2001. Large-scale heterogeneity of the fossil record:

implications for Phanerozoic biodiversity studies. Philosophi-

cal Transactions of the Royal Society of London B 356:351–367.

Smith, A. B., and A. J. McGowan. 2007. The shape of the

Phanerozoic marine paleodiversity curve: how much can be

predicted from the sedimentary rock record of Western

Europe? Paleontology 50:765–774.

Sokal, R. R., and F. J. Rohlf. 1987. Introduction to biostatistics. W.

H. Freeman, New York.

Sutherland, W. J. 2006. Ecological census techniques: a handbook,

2d ed. Cambridge University Press, Cambridge.

Thulborn, R. A. 1982. Speeds and gaits of dinosaurs. Palaeogeog-

raphy, Palaeoclimatology, Palaeoecology 38:227–256.

Upchurch, P. 1995. Evolutionary history of sauropod dinosaurs.

Philosophical Transactions of the Royal Society of London B

349:365–390.

———. 1999. The phylogenetic relationships of the Nemegtosaur-

idae (Saurischia, Sauropoda). Journal of Vertebrate Paleontol-

ogy 19:106–125.

Upchurch, P., and P. M. Barrett. 2000. The evolution of sauropod

feeding mechanisms. Pp. 79–122 in H.-D. Sues, ed. The

evolution of herbivory in terrestrial vertebrates: perspectives

from the fossil record. Cambridge University Press, Cambridge.

———. 2005. A phylogenetic perspective on sauropod diversity.

Pp. 104–124 in Curry Rogers and Wilson 2005.

Upchurch, P., C. A. Hunn, and D. B. Norman. 2002. An analysis of

dinosaurian biogeography: evidence for the existence of

vicariance and dispersal patterns caused by geological events.

Proceedings of the Royal Society of London B 269:613–622.

Upchurch, P., P. M. Barrett, and P. Dodson. 2004. Sauropoda. Pp.

259–322 in Weishampel et al. 2004b.

Waite, S. 2000. Statistical ecology in practice: a guide to analysing

environmental and ecological field data. Pearson Education

Limited, Harlow, U.K.

Weishampel, D. B., P. M. Barrett, R. E. Coria, J. Le Loeuff, E. S.

Gomani, Z. Zhao, X. Xu, A. Sahni, and C. Noto. 2004a. Dinosaur

distribution. Pp. 517–606 in Weishampel et al. 2004b.

Weishampel, D. B., P. Dodson, and H. Osmolska, eds. 2004b. The

Dinosauria, 2d ed. University of California Press, Berkeley.

Wilson, J. A. 2002. Sauropod dinosaur phylogeny: critique and

cladistic analysis. Zoological Journal of the Linnean Society

136:217–276.

———. 2005a. Integrating ichnofossils and body fossil records to

estimate locomotor posture and spatiotemporal distribution of

early sauropod dinosaurs: a stratocladistic approach. Paleo-

biology 31:400–423.

———. 2005b. A redescription of the skull of Nemegtosaurus

mongoliensis (Dinosauria—Sauropoda) and its relevance to

Cretaceous titanosaur diversity. Journal of Systematic Paleon-

tology 3:283–318.

Wilson, J. A., and M. T. Carrano. 1999. Titanosaurs and the origin

of ‘‘wide-gauge’’ trackways: a biomechanical and systematic

perspective on sauropod locomotion. Paleobiology 25:

252–267.

Wilson, J. A., and P. C. Sereno. 1998. Early evolution and higher-

level phylogeny of sauropod dinosaurs. Society of Vertebrate

Paleontology Memoir 5:1–68.

Wilson, J. A., and P. Upchurch. 2003. A revision of Titanosaurus

(Dinosauria Sauropoda), the first ‘‘Gondwanan’’ dinosaur

genus. Journal of Systematic Paleontology 1:125–60.

———. 2009. Redescription and reassessment of the phylogenetic

affinities of Euhelopus zdanskyi (Dinosauria: Sauropoda) from

the Late Jurassic or Early Cretaceous of China. Journal of

Systematic Paleontology 7:199–239.

Wright, J. L. 2005. Steps in understanding sauropod biology: the

importance of sauropod tracks. Pp. 252–284 in Curry Rogers

and Wilson 2005.

Yates, A. M., and J. W. Kitching. 2003. The earliest known

sauropod dinosaur and the first steps towards sauropod

locomotion. Proceedings of the Royal Society of London B

270:1753–1758.


Date post:	09-Jun-2020
Category:	Documents
Upload:	others
View:	3 times
Download:	0 times

A quantitative analysis of environmental associations in ... · sauropods preferred coastal...

Documents