A survey of dinosaur diversity by clade, age, place of discovery and year of description

Dinosaur diversity is analyzed in terms of the number of valid genera within each major clade, Mesozoic age, place of discovery and year of description. Av e s( Archaeopteryx +N eornithes) is excluded. Nomina nuda and nomina dubia are not counted. The results sho w4 51 valid dinosaurian genera at the end of 2001, of which 282 are saurischian (112 sauropodomorphs and 170 theropods, including 93 coelurosaurs) and 169 ornithischian, including 11 pachycephalosaurs, 26 ceratopsians, 60 ornithopods, 12 stegosaurs, and 38 ankylosaurs. Thirty-eight genera arose in the Triassic, 124 in the Jurassic, and 289 in the Cretaceous, of which a disproportionately high number — 85 and 47 — are from the Campanian and Maastrichtian. The Kimmeridgian wa st he most productive age, with an average of 11.18 ne wg enera per million years. The Kimmeridgian sa wa nu nparalleled boom in sauropod diversity ,w ith 20 ne ws auropod genera arising in its 3.4 million years, an av erage of one ne ws auropod every 170,000 years. Asia was the most productive continent with 149 genera, followed by North America (135), Europe (66), South America (52), Africa (39), Australasia (9), and finally Antarctica (1). Just three countries account for more than half of all dinosaur diversity ,w ith 231 genera between them: the U.S.A (105), China (73), and Mongolia (53). The top six countries also include Argentina (44), England (30), and Canada (30), and together provide 335 dinosaur genera, nearly three quarters of the total. The rate of naming ne wd inosaurs has increased hugely in recent years, with more genera named in the last 19 years than in all the preceding 159 years. The results of these analyses must be interpreted with care, as diversity in ancient ecosystems is percei ve dt hrough a series of preservational and human filters yielding observed diversity patterns that may be very different from the actual di versity.


Introduction
Although it is fundamental to matters of Mesozoic palaeoecology,the subject of dinosaur diversity has receivedsurprisingly little attention in the literature. The principal contributions have been those of Dodson and his collaborators (Dodson 1990, Dodson and Dawson 1991, Dodson 1994, Holmes and Dodson 1997. Dodson (1990) surveyed the non-avian dinosaur genera described at that time and concluded that only slightly more than half were valid (285 of 540). He briefly discussed the geographical distribution of the genera, and in more detail the changing levels of observed diversity through the Mesozoic. He estimated the total number of dinosaurian genera at 900-1200, based on estimated genus longevity of 7.7 million years, concluding that the record was at that time about 25% complete. Dodson and Dawson (1991) discussed the process by which the fossil record of dinosaurs has been assembled, analysing the rate of description of newgenera and considering this rate separately for the six countries most fertile in dinosaur genera. Theyalso considered the differing levels of interest in different dinosaur groups and howthis may have biased the publication record. Dodson (1994) covered similar ground, but with more emphasis on the effect of geographical and taxonomic biases on the fossil record. He considered what the record in the last fewmillion years of the Mesozoic implies about dinosaur extinction, concluding that a decline in diversity set in before the end of the Maastrichtian.
Finally, Holmes and Dodson (1997) updated Dodson's1990 analysis with the 51 new genera named between 1989 and 1995, briefly discussing the age, country and infraorder of the newgenera.
No published analysis considers the explosion in newgenera since 1997, or analyzes diversity by clade or at all comprehensively by place of origin. The present study attempts to address these deficiencies by offering four different analyses of a single dataset describing the dinosaurian genera considered valid as at the end of 2001. The data-set itself and the analysis program are both freely available (Taylor 2004a(Taylor , 2004b.
Adistinction must be made between observed diversity and actual diversity.A ll these analyses necessarily work with information about the former.W ecan neverknowthe actual levels of diversity in anyancient ecosystem: the set of valid genera that we have today is the result of a series of chances including which animals were fossilised, which fossils surviveduntil the present, which surviving fossils are in exposed outcrops, which exposed fossils have been collected, which collected fossils have been prepared and which prepared fossils have been described. However, increased understanding of the preservational and human factors that bias the record of observed diversity may in the future allowincreasingly accurate estimates of actual diversity to be made.

Source of Data
The analyses in this paper are all derivedfrom a single data-set (Taylor 2004a) representing dinosaur phylogeny, geology,geographyand history.T he initial version of the data was obtained with permission from the Dinosauricon web-site (Keesey2001). This data-set was assembled oversev eral years by the web-site author,with reference to the scientific literature, to reflect a consensus of then-current ideas about dinosaurian phylogenyand taxonomy.T he data-set has been progressively reviewed on an informal basis since its inception, so it is perhaps closer to being a peer-reviewed database than anyother.
The initial data-set taken from this web-site has been modified by the author to include all valid genera named to the end of the year 2001, and to reflect a more up to date understanding of the classification and age of some taxa. This updating was done with reference to Glut 2003 and also to numerous papers referenced by the very helpful DinoData web site (Bervoets 2004). Genera described since 2001 are not included.

Analysis Program
The analyses are all produced by a single program (Taylor 2004b) which reads the entire data-set into memory and arranges the taxa into a prescribed phylogenetic tree. The clade Av es(birds) is then excised from the tree, and the remaining structure is processed in a number of ways to produce the different analyses. Forthe purposes of these analyses, the definition of Avesisthat of Chiappe 1992, being the most recent common ancestor of Archaeopteryx and modern birds together with all its descendents.
Note that this program does not perform a phylogenetic analysis, but analyses genus data within the framework of a prescribed phylogeny. Phylogenies are always contentious, but for the purpose of the current study it seemed best to re-use rather than to replicate prior work, using an available phylogenyuncritically.T he purpose of this paper is not to advocate a particular phylogeny: the phylogenypresented here is the hypothesis, not the conclusion. In anycase, the more interesting results of this study mostly pertain to highlevelclades and are therefore not much affected by uncertainties about the details of lowlevelphylogenywithin those clades.
The first analysis is of the phylogenyitself, and simply lists all the taxa described in the data-set in a form that illustrates the hypothesised relationships, and counts the number of genera included in each taxon. The second analysis notes the first geological age in which each genus occurred and counts howmanygenera arose in each age, epoch and period of the Mesozoic. The third analysis counts the number of genera described from each country and state, aggregating up to continent and supercontinent. The fourth counts the number of genera described each year since 1824.

Genera and Species
This study only counts genera and makes no attempt to consider species. Forextant animals, it can be argued that species are objectively real while genera are merely a convenient abstraction (Cantino et al. 1999, Lee 2003. Forextinct animals, however, the converse appears to be the case. The biological concept of species is not testable with fossils, and therefore inapplicable, so a morphological concept must be used; and while there is broad consensus on the degree of variation that constitutes a generic difference between fossils, there is little agreement overhow toseparate fossil species in the Dinosauria. For example, the number of valid species in Triceratops has variously been placed at 10 (Hatcher et al. 1907), six (Lull 1933), one (Ostrom and Wellnhofer 1986;Lehman 1990) and most recently two (Forster 1990(Forster , 1996. Although separation at the genus levelisless contentious than at the species levelfor extinct animals, it is still by no means universally agreed upon. There are manydinosaur genera that some authorities consider distinct while others do not -for example, the allosaurid Saurophaganax (Chure 1995) is considered by some workers merely to represent large Allosaurus specimens (for example, Lucas 1987, Paul 1988 pp. 312-313).
Ultimately,classification of dinosaur specimens into genera and species is as much an art as a science; so while there is some consensus at the genus level, there is no single, definitive list of valid genera. The genera considered valid in this study therefore represent one perspective among many.
In the current data set, 381 of the 451 valid genera are monospecific. Forty-six contain twospecies, 17 contain three species and only sevengenera contain more than three recognised species. Of these, Camarasaurus, Cetiosaurus, Chasmosaurus and Edmontonia each have four species in this data-set, though Upchurch andMartin's(2002, 2003) recent work on Cetiosaurus has since reduced it to a single valid species, C. oxoniensis. Iguanodon and Mamenchisaurus each have sev enspecies, and Psittacosaurus has eight. This givesatotal of 562 dinosaur species, for an average of just 1.25 species per genus.

Number of Genera by Clade
[ Fig. 1] Fig. 1 givesahigh-leveloverviewofthe phylogenyused in this study.S ee the Appendix for full details. Clade sizes are of course subject to the accuracyofthe phylogenyused: however, high-leveldinosaur phylogenyappears to have been relatively stable overthe last fewyears, with instability largely at ''family level''and lower within these higher clades, especially the Coelurosauria.
The 93 coelurosaurs represent 56% of the 170 theropods. Wilson and Upchurch's(2003) observation that titanosaurs represent approximately one-third of sauropod diversity is corroborated by this study: theysupply 34 of the 93 sauropod genera. This contrasts with Curry Rogers and Forster's(2004) assertion that titanosaurs comprise nearly half of all known sauropod genera.
[ Fig. 2] Fig. 2 shows the relative sizes of the major dinosaur groups. It is surprising that saurischians outnumber ornithischians so heavily -theyare about 66% more diverse. More surprising still is the predominance of theropods: the total number described (170) is greater than the number of sauropodomorphs (112) or ornithischians (169). They account for 37% of all known dinosaur genera. This is in spite of the theropods' having a much more conservative body-plan than the ornithischians, which display remarkable morphological diversity encompassing ceratopsians, ornithopods and stegosaurs.
Of the total of 16 diplodocoidean genera, 12 fall within Flagellicaudata = Clade(Dicraeosaurus + Diplodocus), Harris and Dodson 2004. All but one of these 12 arose during the Kimmeridgian: the sole exception is Amargasaurus from the Hauterivian, about 20 million years later.
If the clades Ornithomimosauria and Therizinosauria are considered herbivorous or omnivorous (Kobayashi et al. 1999, Barsbold andMaryanska 1990) then the remaining, carnivorous, theropods number 151 -fully one third of all dinosaur genera. This is an unusually high proportion of total diversity for carnivores to attain within an ecosystem. (Holtz et al. (1998) have also suggested based on tooth-serration density that troodontids may have been omnivorous, but this idea is not widely accepted.)

Number of Genera by Geological Age
[ Table 1] Table 1 shows that observed dinosaur diversity generally increases through time, with 38 genera having arisen in the Triassic, 124 in the Jurassic, and 289 in the Cretaceous. This imbalance is partly due to the origin of the dinosaurs only in the Carnian (Late Triassic), butevenwhen diversity across the three periods is normalised by duration, the trend towards greater diversity is evident. The 38 Triassic genera occurred in the 21.7 million years from the beginning of the Carnian to the end of Rhaetian, giving a genus density,or GD, of 1.75 genera per million years. The 124 Jurassic genera arose in 61.5 million years for a GD of 2.02 and the 289 Cretaceous genera arose in 79.2 million years for a GD of 3.65. One reason for this bias towards greater observed diversity in more recent times may simply be that older fossils have had more time in which to be destroyed by processes such as erosion (Molnar 1997). [ Table 2] Table 2 shows the GD of individual ages. This givesamore realistic indication of the levels of dinosaurian diversity in each age than the non-normalised figures: for example, the Kimmeridgian is nowseen to have been more fertile in its rate of producing new genera than the Campanian, eventhough the latter period gav e rise to more than twice as manygenera as the former.S imilarly,the Santonian was more fertile than the Albian despite having originated fewer than one third as manynew genera.
Three ages stand out as much more diverse than others. The Kimmeridgian has a very high GD of 11.18; and the last twoages of the Mesozoic, the Maastrichtian and Campanian, have GDs of 7.83 and 6.80 respectively.N oother age has a GD greater than 4.00. While no doubt sampling biases account for some of the GD irregularity,there does appear to have been a substantial and sustained flurry of diversity in the last twenty million years or so of the Mesozoic.
The highest apparent spike, in the Kimmeridgian, is exaggerated by the rule used in this analysis that a genus is attributed only to the age in which it first arose and not also to subsequent ages in which it survived. Of the 38 genera designated as Kimmeridgian in this analysis, fully 22 may have persisted into the Tithonian, so the apparent fall-offof diversity between these ages is not entirely real.
[ Fig. 4] Fig. 4 shows that diversity levels correlate only weakly with eustatic level(that is, global sea level). Thus this study does not strongly corroborate the claims of Haubold 1990 andHunt et al. 1994 that taphonomic biases cause observed dinosaur diversity to be highest at times of highest eustatic level.
The large genus-count of 24 for the Carnian, the earliest age in which dinosaurs appeared, implies that initial dinosaur diversification was rapid. Of those genera, one quarter are ornithischian, but all six are too basal to assign to more specific clades. Of the saurischians, four are sauropodomorphs (of which none are sauropods) and the remaining 14 are theropods, of which six fall within Neotheropoda and eight are more basal. Twelvemore newgenera arose in the Norian, including the earliest known sauropod, Isanosaurus.
[ Table 3] Table 3 shows long intervals between the earliest and subsequent genera within manyof the major clades. Forexample, the next recorded coelurosaur after the therizinosauroid maniraptor Eshanosaurus,inthe Hettangian, is Ozraptor from the Bajocian, 29 million years later.I thas been suggested that Eshanosaurus may in fact be a prosauropod (MatthewC.Lamanna, pers. comm. to Xu, Zhao and Clark), though this alternative identification has not yet been published. Similarly Yaverlandia's status as the oldest pachycephalosaur is not firmly established, as its pachycephalosaurian affinities have been questioned (Sullivan2000). Adifferent situation pertains for ceratopsians. After Chaoyangsaurus,the next recorded genus seems to be Archaeoceratops.T he exact ages of these genera are not firmly established; but the former may be Bathonian, and the Xinminbao Group in which latter wasfound seems to be Barremian (Tang et al. 2001), indicating a gap of about 42 million years. However, the ceratopsian identity of Chaoyangsaurus appears to be secure as the type specimen has a rostral bone (Zhao, Cheng and Xu 1999), implying a long ghost lineage.

Number of Genera by Place of Discovery
Dinosaur palaeontology begain in England, and shortly thereafter developed in mainland Europe. Accordingly,European genera dominated counts for the first 66 years (1824-1889) before the gathering pace of research in North America established it as the most productive continent for more than a hundred years. As late as 1883, Europe, with 17 genera, still had nearly twice as manydinosaurs as North America, with only nine. But in the late 1800s, the American railways movedwest opening up newareas for fossil prospecting, so that by 1890 North America had overtaken Europe, with 19 genera to Europe's17. In the sevenyears since 1883, 10 newgenera had been named from North America, but none were named in Europe that are still considered valid today.I n1890, the twoestablished continents between them accounted for all but three of the dinosaur genera then known, with the others made up of twoAfrican dinosaurs (Massospondylus and Euskelosaurus,from Lesotho) and just one from Asia (Titanosaurus,from India).
After the description of Titanosaurus in 1877, there was a 45-year gap before the next Asian dinosaurs were named (Indosuchus, Protoceratops and Psittacosaurus in 1923). But since the early 1970s, the rate of newdiscoveries in Asia has been more rapid than in the West. By the end of 1993, Asia had finally overtaken North America as the most productive continent, with 104 genera compared to North America's99. By this point, the number of Europen genera had climbed only slowly to 51, less than half as manyas Asia.
[ Table 4] Table 4 shows the breakdown of dinosaur genera by place of description as at the end of 2001.  6 shows that just three countries account for more than half of all dinosaur diversity, with 231 genera between them: the U.S.A (105), China (73), and Mongolia (53). The top six countries also include Argentina (44), England (30), and Canada (30), and together provide 335 dinosaur genera, nearly three quarters of the total. The last year in which no newdinosaurs were described was 1961; the last before that was1949. This means that newdinosaurs have been described in every year but one of the last 52.
Apart from the general upward trend, there is little pattern to the year-by-year frequency of naming: for example, 1997, with just fivenew genera, was a relatively barren year sandwiched between twobumper crops: 14 in 1996 and 25 in 1998.
[ Fig. 9] Fig. 9 shows hownaming frequencyhas varied decade by decade. Atrend is evident: apart from anomalously lowfigures for the four decades from the 1930s to the 1960s, the tendencyisfor the naming rate to growexponentially.T his four-decade fall-off corresponds with a period in which mammal palaeontology dominated the field (Bakker 1975), brought to an abrupt end in the 1970s by the ''dinosaur renaissance'', widely considered to have been catalyzed by Ostrom'sdescription and osteology of Deinonychus antirropus (Ostrom 1969a(Ostrom , 1969b. After this period, the exponential naming rate seems to pick up as though the 30s-60s had neverhappened, with more than twice as many genera named in the 1970s (56) as in the last decade before the gap, the 1920s (25 genera). The 56 genera described in the 1970s outnumber the total of 48 from the previous four decades combined: 17 in the 1930s, four in the 1940s, 14 in the 1950s and 13 in the 1960s.
[ Fig. 10] The origin of dinosaur palaeontology in the northern hemisphere, and the more recent increase of work in the southern hemisphere, is reflected in the history of newgenera from each of the twoMesozoic supercontinents (Fig. 10). As late as 1913, only four genera were known from Gondwana: Massospondylus (1854), Euskelosaurus (1866), Argyrosaurus (1893) and Genyodectes (1901). By that same year,63genera were known from Laurasia -nearly 16 times as many. By1932, the situation had started to evenup, with 16 Gondawanan genera to 98 Laurasian, for a factor of 6.13. At the end of 2001, the 230 Laurasian genera still significantly outnumber the 101 Gondwanan genera, but the factor of 2.28 indicates that the gap is closing.
These figures should not be taken at face value, however, asGondwana and Laurasia did not exist as complete, distinct landmasses throughout the whole of the Mesozoic. Their history is rather complex, with the various plates repeatedly joining and dividing in various combinations, and with epicontinental seaways dividing individual plates into multiple palaeobioprovinces (Le Loeuff1997). Note also that ''Gondwana''inthe sense used here includes only the modern southern continents Africa, Antarctica, Australasia and South America, omitting parts of Europe such as Italy and Austria that were part of the southern landmass during the Mesozoic.

Reasons for Variations in Diversity
The greater observed diversity of certain groups and ages is due to manyfactors, some of which are discussed below.
Geological Preservational Bias. -R aup (1972), working with the record of marine invertebrates throughout the Phanerozoic era, demonstrated a strong correlation between observed diversity levels and the volume of sedimentary rock available from each age. This may be the single most significant factor affecting observed diversity through time.
Anatomical Preservational Bias. -P hysical properties of the skeletons of different taxa affect the likelihood of preservation. The pneumatised and relatively fragile bones of theropods would generally be more susceptible to damage than the relatively robust bones of sauropods and ornithischians. However, observed diversity figures do not reflect this expectation, presumably because other factors outweigh this one.
Ecological Preservational Bias. -M anytheropods, being opportunistic scavengers, would have favoured carrion-rich environments such as sea margins, which confer a greater likelihood of preservation than the open plains that might have been favoured by most herbivores. This factor goes some way towards explaining whywe observeadisproportionately high number of theropod taxa. Forexample, the Santana Formation preserves four theropods and no other dinosaurs; the Solnhofen limestone preserves three theropods and no other dinosaurs. It is unlikely that there were no herbivorous dinosaurs in these ecosystems, but theyprobably livedand died in drier nearby environments, and so have not been preserved.
Differential Splitting and Lumping. -I tispossible that some clades have been over-split by workers keen to establish newgenera in ''glamourous''families, when working with specimens for which the degree of morphological difference from existing genera is not as great as would otherwise be expected. For example, the eight genera in the morphologically conservative group Tyrannosauridae are perhaps more than would have been established for specimens varying to a similar degree in another family (Currie 2003).
In the same vein, there may be a tendencyfor large sauropods to be assigned newgeneric names when theyare not really merited. Forexample, the dorsal vertebra that was the type specimen of ''Ultrasauros'' (Jensen 1985) is nowreferred to Supersaurus. ( Curtice et al. 1996); Dystylosaurus (Jensen 1985) also appears to be synonymous with Supersaurus (Curtice and Stadtman 2002); and Seismosaurus (Gillette 1991) may be merely a large Diplodocus (Lucas and Heckert 2000).
Focus of Current Work. -T he frequencywith which newgenera are described in different groups may simply reflect the number of workers in those groups. At present, there seem to be manymore theropod workers than sauropod workers, and yet fewer who specialise in ornithischians. Manyornithopod specimens collected on expeditions remain in their plaster jackets while the theropods are prepared and described first. This current focus is reflected in the naming frequencyinrecent years, which is progressively skewing the record towards theropods. Forexample, in 1996, 10 newtheropods were described butonly twonew ornithischians. The 111 newgenera described in the years from 1996 to 2001 are made up of 47 theropods (42%), 32 sauropodomorphs (29%) and 32 ornithischians (29%).
Actual Diversity. -W ith these other factors taken into account, the observed diversity numbers are indicative ofthe actual diversity of the living animals. But caution must be exercised when interpreting observed diversity numbers. Forexample, the last fourteen years of history strongly indicate that Dodson's(1990) estimates of total dinosaur diversity at 900-1200 genera to be well short of the true number.

The Kimmeridgian Sauropod Boom
Among the diversity anomalies shown by this study,perhaps the most puzzling is the large number of newsauropod genera that arose during the Kimmeridgian. The total of 20 genera comprises 12 from the The twoChinese genera lie outside Neosauropoda, but the other 18 are all neosauropods, comprising 11 diplodocoids (all falling within Flagellicaudata) and six macronarians, with the position of Haplocanthosaurus uncertain: it resolves as a diplodocoid or macronarian, or just outside Neosauropoda, depending on which other taxa are included in the analysis (Upchurch 1998).
Apart from the Kimmeridgian, The next most diverse ages for sauropods are the Campanian (nine genera), Albian (eight), Maastrichtian (seven) and Bathonian (five). The average number of newsauropods that arose in each million years of the Kimmeridgian (sauropod Genus Density,orsauropod GD) is 5.88 -a newsauropod genus every 170,000 years. The ages with the next highest sauropod GDs are the Maastrichtian (1.17), Bathonian (1.04) and Hauterivian (0.80). This high value for the Maastrichtian is contrary to widespread orthodoxy that sauropods were in decline at the end of the Mesozoic.
There are several possible causes for the sudden (in geological terms) Kimmeridgian boom in observed sauropod diversity.
Av a ilability of Strata. -A lthough high observed diversity in the Kimmeridgian is most pronounced for sauropods, diversity is also high for theropods (five) and ornithischians (13). This is due in part to the exposure of Morrison-Formation strata across a wide area of more than a million square kilometers in 12 states (Dodson et al. 1980). Accordingly,the Morrison Formation is particularly well studied.
Similarly,the observed diversity spikeinthe Campanian is partially attributable to the exposure of the Two Medicine Formation across a wide area of Montana and Alberta (and also partly just to the length of this age -12.5 million years, more than three times the length of the Kimmeridgian). Horner and Dobb (1997, pp. 192-196) observed that the high diversity of ostensibly contemporary Campanian centrosaurines actually represents a stratigraphic sequence, in which Styracosaurus,' 'centrosaurine 1''(not yet described), Einiosaurus, Achelousaurus and Pachyrhinosaurus occur successively within and immediately above the upper Two Medicine Formation. In other words, the seeming high diversity is really an artifact of over-coarse granularity in our time divisions, and the large number of genera actually reflects an unusually rapid turnoverrather than many contemporary centrosaurines. However, this situation does not pertain in the case of Morrison sauropod diversity,asthe Morrison sauropods all overlap in time (Turner and Peterson 1999).
Preservational Environment. -M orrison sediments represent an enormous alluvial plain rich in lucustrine and floodplain environments that were conducive to fossilisation (Dodson et al. 1890). So an unusually high proportion of the Morrison fauna has probably been preserved well enough to be identified reliably.
Taxonomic Over-Splitting. -A sdiscussed above,sev eral of the Kimmeridgian sauropods currently considered valid may in fact belong to the same genus. Following on from the Dystylosaurus and Seismosaurus referrals previously mentioned, more synonymisations are likely.F or example, Supersaurus may be congeneric with Barosaurus (Curtice 2003), and Giraffatitan may not be distinct from Brachiosaurus (Wilson and Sereno 1988).
In general, larger animals seem to be more susceptible to over-splitting than small ones. In part, this is because theytend to live longer,sotheyhav e more time in which to accumulate individual variations that can be mistaken for generic differences. Over time, individual muscles may hypertrophyoratrophy, with consequent changes in the skeleton. Furthermore, variation is easier to see in larger specimens. These effects may go some waytow ards explaining whymore genera are erected for large animals than for smaller ones.
Niche Partitioning. -E venallowing for biases arising from availability of strata, preservational environment and taxonomic over-splitting, the Kimmeridgian in general, and the Morrison ecosystem in particular,still appears remarkably diverse. No recent ecosystem evenapproaches such richness of large animals. Some areas of Africa support four herbivores massing 1000 kg or more (elephants, rhinos, hippos and giraffes), butnomore; and the last three of these, while ''large''bycontemporary standards, are small compared with eventhe smallest of the 12 Morrison sauropods. One candidate explanation for this diversity is niche partitioning: an ecology in which different sauropod genera favoured different foods, or livedindifferent environments, thereby avoiding intergeneric competition. Stevens and Parrish (1999) investigated neck mobility in Apatosaurus and Diplodocus by computer modelling of the cervical zygapophyseal articulations. Their results indicate that the former,despite its shorter neck, could feed rather higher than the latter (6m vs. 4m above ground level). Brachiosaurus could reach much higher still: a feeding height of 15m has been suggested (Paul 1998). These differing feeding heights suggest different dietary specialisations. In general, of course, the sauropods with higher reach would have been equally capable of browsing at lower levels; but Christian and Heinrich (1998) suggest, rather improbably,that Brachiosaurus brancai may have had a very limited vertical feeding range due to the difficulty of moving its neck far from the vertical pose that their study favours.
Studies of dental wear (Fiorillo 1998) indicate that different sauropods may have fed on different plants: coarse scratches on Camarasaurus teeth suggest that its food was gritty, whereas the finer scratches on Diplodocus teeth indicate a grit-free diet. As the concentration of grit tends to be higher at lower levels, this implies a low-levelfeeding strategy for Camarasaurus while Diplodocus probably browsed at a higher level-a conclusion that contradicts the horizontal neck posture suggested for Diplodocus by Stevens and Parrish (1999). Barrett and Upchurch (1994) argued that specialisations in the skull of Diplodocus indicate an unusual mechanism for cropping vegetation, with the characteristic labial wear-facets on both upper and lower teeth explained by its use of different jawactions for high and lowbrowsing. Theyspeculated that the differences between this feeding method and the less specialised method used by Brachiosaurus and Camarasaurus indicates some ecological separation.
In modern ecosystems, hippos and rhinos do not compete with each other because the former are largely aquatic. (Although theyfeed mostly on land, hippos remain close to water,whereas rhinos favour open grassland.) It is tempting to imagine that the Morrison sauropods might have niche-partitioned similarly,with some but not all being semiaquatic. However, multiple lines of evidence showthat sauropods were very poorly adapted for such a lifestyle: their feet are proportionally smaller than those of almost all terrestrial vertebrates, generating pressures about twice those of domestic cattle (Alexander 1989); and Coombs (1975) makes a strong biomechanical argument that the deep, relatively narrowtorsos of sauropods are an adaptation for carrying weight in terrestrial locomotion.
While all Morrison sauropods were primarily terrestrial, some difference in partiality to wetter and dryer environments may nevertheless be indicated. Forexample, differences in limb and foot bones suggest that Apatosaurus and Diplodocus were better suited to traversing wet sediments than Camarasaurus was(Bonnan in press). However, Dodson et al. (1980) analysed the occurrence of several sauropod taxa within the four major lithofacies of the Morrison and concluded that large herbivorous dinosaurs were not aquatic, nor evensemi-aquatic in the style of the hippo. '' Diplodocus and Camarasaurus resemble elephants in their patterns of distribution.'' Howev er, the same authors also ''believe that the distribution of large dinosaurs in the Morrison reflects ecological factors, not patterns of rapid evolution or extinction at the generic level.''

FutureW ork
Follow-up studies might usefully relate the results of the individual analyses in this report to each other.F or example, more work could be done on the age-distribution of particular clades, and on the tendencyofcertain clades to occur more commonly within particular continents. Similarly,the changing ''fashionability''ofdifferent clades through history could be determined by observing the varying rates at which newgenera have been named within those clades at different times.
It would be interesting to investigate the correlation between observed diversity levels and variables such as the levelofatmospheric oxygen (Berner and Canfield 1989), the levelof atmospheric carbon dioxide (Berner 1990(Berner , 1994 and average surface temperature (Frakes et al. 1992).
The Kimmeridgian sauropod boom is worthyoftreatment in much greater depth than it has receivedinthis study.T he Morrison ecosystem that simultaneously supported so manyvery large animals is without parallel. It is a mystery not only howsomany sauropod genera survivedascontemporaries, but also howtheyarose within so short a space of time from one another.I norder to fully explain the Kimmeridgian boom, it will be necessary to understand whyrelatively fewsauropods arose during the immediately preceding Oxfordian (four sauropod genera) and Callovian (three); and whysofew new sauropods arose in the immediately subsequent Tithonian (one genus) and Valanginian (two).
Finally,much effort is wasted at present by numerous workers each maintaining their owndatabases of valid dinosaur genera, their ages, countries, etc., similar in spirit to the one used in this study.I twould be useful to establish a single canonical list, maintained by a committee of experts, and made freely available on the Internet in a well-defined format for all who wish to work with it.

Conclusions
Analysing dinosaur diversity data yields a number of surprising results, chief among which are the high diversity of theropods compared with ornithischians, the diversity spikes in the Kimmeridgian, Maastrichtian and Campanian, the long intervals between the first and subsequent recognised genera of some clades, and the increasing rate in the naming of newgenera.
It is apparent that the diversity patterns observed from current data are extremely uneven. It is difficult to interpret some of the findings of this study,particularly those pertaining to geographical distribution, because observed diversity is affected by so manyfactors, both preservational and human. Dinosaur diversity is a system of manyvariables (taxonomic, geographic, geological, historical and others), correlated in complexways and to varying degrees. It is not always possible to study variations in anyone of these variables in isolation.
Much work remains to be done in analysing dinosaur diversity,particularly in correlating phylogenetic, geographic and stratigraphic information. That work could best be facilitated by collaboration on the creation and maintenance of a publicly owned database of dinosaur genera, so that different workers could more easily devise and perform different analyses without first having to replicate each other'sspadework in assembling data.

Acknowledgements
The rawdata used in this study were derivedfrom the work of T.Michael Keeseyfor his web-site The Dinosauricon (Keesey2001). The data used to generate the Dinosauricon site is freely available at http://dinosauricon.com/data/ and the modified versions used in this study are also available (Taylor 2004a). This paper would neverhav e been written without the initial encouragement and subsequent criticism provided by MathewJ.W edel (University of California Museum of Paleontology). This work has been greatly improvedbythe comments made on an earlier draft by Dr. David M. Martill (University of Portsmouth).
Finally,thanks are due to the members of the Internet'sDinosaur Mailing List (http://dinosaurmailinglist.org) for manyuseful and informative discussions, and also for manyuseless but entertaining ones.

Table 1
Number of dinosaur genera by geological age, from oldest to youngest. The number of genera shown for each epoch includes those for all the ages it contains as well as those of uncertain position within the epoch. The number of genera shown for each period includes those for all the epochs it contains. Numbers in square brackets after epoch names indicate genera whose first occurrence is within the epoch but cannot be more precisely stated. Each genus is counted only in the earliest age in which it occurs, so that the total number of genera counted in this analysis is equal to the total number of valid genera. The bar-graph clearly shows the bias in the fossil record towards the Carnian, Kimmeridgian and mid-and late Cretaceous.  Table 2 Genus Density (GD) by geological age, found by dividing the number of genera arising in each age by the length of the age in millions of years. Each genus is counted only in the earliest age from which it is known.  Table 3 First and second genera occurring within some important clades, with the difference between their ages, measured from the beginning of the age in each case. The difference between the ages of the earliest and subsequent genera within each clade is the length of the implied ghost lineage at the base of that clade. Genera whose age is not known to the resolution of a single age are discounted: for example, Losillasaurus is only known to be Late Jurassic, so it is ineligible to be the oldest Diplodocoid.  Table 4 Number of genera by place of discovery.T he number of genera shown for each country includes those for all the states it contains; numbers for continents include those for their countries; and numbers for the supercontinents include those for the continents that make them up. Each genus is counted only in the country in which it was first discovered, even if specimens have subsequently been discovered in other countries.  Fig. 1. Breakdown of dinosaur diversity by phylogeny. The number of genera included in each clade is indicated in parentheses. Non-terminal clades additionally have,in square brackets, the number of included genera that are not also included in one of the figured subclades. Forexample, there are 63 theropods that are neither carnosaurs nor coelurosaurs. The thickness of the lines is proportional to the number of genera in the clades theyrepresent.