Content uploaded by Donald Esker
Author content
All content in this area was uploaded by Donald Esker on Feb 25, 2017
Content may be subject to copyright.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
A multivariate analysis of the ecology of North American Pleistocene bears, with a focus on
Arctodus simus.
DONALD A. ESKER,1 W. JUSTIN WILKINS,1 and LARRY D. AGENBROAD1
1The Mammoth Site of Hot Springs South Dakota Incorporated, Hot Springs, South Dakota
57747 U.S.A., done@mammothsite.org, justinw@goldenwest.net, larrya@mammothsite.org
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
ABSTRACT—Arctodus has been the subject of intense research over the last century. Much of
this research focused on the intractable question of why it became extinct. The prevailing
hypotheses so far are either displacement by invading Eurasian ursine bears, or loss of prey
species. Past studies have focused on the anatomy of the animal but little has been done to place
the animal in its ecological context. By using multivariate analysis, we have revealed the
existence of 10 distinct Pleistocene ursid communities. The distribution of bear taxa in these
assemblages reveal whether tremarctine bears like Arctodus genuinely competed with ursines. If
bear taxa were segregated by cluster, the potential for competition would have been minimized.
Our results show the contrary – that ursine and tremarctine bears lived in the same sorts of areas
and with the same potential prey species – making competition from ursines a viable hypothesis.
ALTERNATE-LANGUAGE SUMMARY— Durante el último siglo Arctodus ha sido el objeto
de una intensa investigación. Gran parte de esta investigación se centra en la difícil pregunta de
por qué se extinguieron. Las hipótesis vigentes hasta ahora son, el desplazamiento debido a la
invasión de úrsidos de la subfamilia ursinae, procedentes de Eurasia, en particular U. arctos, y la
extinción de las especies que le servían de alimento. Estudios anteriores se centran en la
anatomía del animal, pero poco se ha hecho para situarlo en su ámbito ecológico. Mediante el
uso del análisis multivariado se ha puesto de manifiesto la existencia de 10 comunidades de
úrsidos del Pleistoceno diferentes. La distribución de los taxones en estas comunidades nos
indica si úrsidos de la subfamilia tremarctinae, como Arctodus, realmente compitieron con
úrsidos de la subfamilia ursinae. Si separamos los taxones por grupo la posibilidad de
competición se reduce al mínimo. Nuestros resultados demuestran lo contrario, ursinae y
tremarctinae convivían en el mismo tipo de áreas y se alimentaban del mismo tipo de presas. Se
conlcuye que la competencia con ursinae es una hipótesis viable.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
INTRODUCTION
The realization that the ursine bears were not the only ursids in the Pleistocene of North
America came with Leidy‘s 1854 description of the tremarctine Arctodus pristinus. Since then,
the record for North American Pleistocene bears has grown to respectable proportions. Data
gleaned from the Paleobiology Database and FaunMap in 2009 produced a list of no fewer than
193 Pleistocene ursid sites. Despite such a solid record, understanding how the different species
of Ice-Age bruins lived and why some survived to modern times while others have not has
proved difficult.
There are currently three hypotheses regarding the extinction of North American
tremarctines. First, that extinction of prey species (whatever the cause) deprived them of their
food source. Second, that climate change made North America an inhospitable environment.
Finally, that ursine bears, particularly Ursus arctos, out-competed and displaced them. This
paper will examine the last of these hypotheses.
Up to this point most authors have attempted to examine the question by examining the
anatomy of the extinct tremarctine bears, with an eye towards seeing how they differed
ecologically and behaviorally from extant ursine bears. Few papers have tried to look at the
problem from a purely statistical, biogeographic, and ecological perspective. In its attempt to
exclude Ursus as a factor in the tremarctines‘ extinction, this paper represents such a study. If it
can be shown that Ursus and the tremarctines inhabited different faunal assemblages, it would
remove the possibility of competition, and thus falsify the hypothesis. If no such separation can
be shown, the possibility that competition for niche space was a factor in the extinction remains.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
HISTORY OF STUDY
Initial Description
Early work on North American tremarctinae was primarily descriptive. When Leidy
named the genus Arctodus, he provided a qualitative description of the single molar he found
(Leidy, 1854). Cope‘s publication naming A. simus (Arctotherium simum) was based on a nearly
complete skull, but he still refrained from speculating about tremarctines‘ place in the
Pleistocene landscape (Cope, 1879).
Bjorn Kurten was the first person to so speculate in his examination of Tremarctos floridanus.
Based on similarities with Ursus spelaeus, he concluded that T. floridanus was likewise an
herbivore (Kurten, 1966). This agrees fairly well with what is known about the extant
tremarctine, T. ornatus‘ diet (Troya, Cuesta et al., 2004). In his book, Pleistocene Mammals of
North America, Kurten reiterates this idea adding that since T. floridanus is frequently found at
the same sites as Ursus americanus, it may have inhabited a different – presumably more
herbivorous – niche than its ursine cousin (Kurten and Anderson, 1980). Kurten never posited
any hypothesis for the animal‘s extinction.
Arctodus as an Active Carnivore
Kurten was also the first to hypothesize regarding Arctodus’ habits. He cast A. simus as
North America‘s largest active carnivore – though he doesn‘t discount the possibility that plants
made up a small part of its diet – with the less derived A. pristinus as a more typical omnivorous
bear. He sees similarities with Ursus arctos, to such an extent that he dubs the ancient bruin the
‗vicar‘ of U. arctos. His conclusions were based on the osteology of the animal – particularly the
disproportionately long legs and broad, stout skull (Kurten, 1967). In his book, he posits the
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
hypothesis that competition from U. arctos was a factor in A. simus‘ demise (Kurten and
Anderson, 1980). Valerius Geist reiterated Kurten‘s interpretation of A. simus as a vicious
predator, even intimating that the presence of the beast for a time prevented the migration
humans from Siberia to Alaska (Geist, 1986).
Arctodus as an Herbivore
Evidence in favor—The first major challenge to Kurten‘s interpretation came in 1985.
Steven Emslie and Nicholas Czplewski conducted a morphometric study comparing A. simus and
Tremarctos ornatus. They concluded that the features Kurten noted indicating cursoriality and
feline-like skull were not reliable indicators of carnivory. They also noted that close
morphological similarity with the largely herbivorous T. ornatus seemed to indicate that A. simus
had a similarly herbivorous diet. Furthermore, they felt that the co-occurrence of Arctodus and
Ursus species at some sites provided strong evidence that the two genera had different dietary
habits (Emslie and Czaplewski, 1985). Sorkin came to the same conclusion twenty-one years
later, after comparing the morphology of Arctodus and its close relative Agriotherium to that of
the – presumed – carnivore Hemicyon. He came to the conclusion that gross dental and
mandibular morphology of the tremarctine bears simply do not support carnivory, although he
left open the possibility of scavenging (Sorkin, 2006).
Mitigating factors—If their comparison with Tremarctos is valid, Emslie and
Czplewski‘s hypothesis on Arctodus‘ herbivory is still predicated on the idea that T. ornatus is
more-or-less herbivorous. While accepted as fact by many authors in the paleontological
literature, modern mammalogists are less certain on the matter. Scat studies have certainly shown
that T. ornatus may get a substantial percentage of its diet from plants, but authors have
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
wondered whether the more easily digestible meat is simply not preserved in scat (Troya, Cuesta
et al., 2004). Goldstein et al. wrote an entire paper on livestock predation in T. ornatus in 2006.
This proves they certainly can and do consume flesh, although it does little to indicate how
important meat is to Tremarctos‘ diet (Goldstein, Paisley et al., 2006). With scat studies
potentially less reliable than supposed, and direct observation of behavior in the wild limited, the
case for a strictly herbivorous T. ornatus is not as strong as previously thought. Even if the genus
is currently a strict herbivore, it has even been proposed this a geologically recent development.
Tremarctos, while initially omnivorous became more herbivorous after the invading a more
predator-rich South America during the Great American Biotic Interchange (Figueirido and
Soibelzon, 2009).
Arctodus as a Scavenger
Isotope analysis—The next area of study to shed light on the problem of Arctodus‘ diet
is stable isotope analysis. Among the many isotopes available for study, δ15N and δ13C reveal the
most about dietary habits. Δ15N tends to increase with trophic level, while δ13C tends to decrease
with trophic level. When plotted against one another, terrestrial herbivores, terrestrial carnivores,
and marine carnivores fall into distinct clusters. Matheus used this fact to hypothesize about the
diets of Beringian A. simus and U. arctos. In this study A. simus comes out as significantly more
carnivorous than contemporaneous and sympatric U. arctos (Matheus, 1995).
Anatomy and enamel structure—Matheus reconciles these results with Emslie and
Czplewski‘s findings showing a lack of predatory adaptations by proposing that A. simus was a
dedicated scavenger (Matheus, 1995). This view was echoed by researchers at the Hot Springs
Mammoth Site (HSMS) who, like Matheus, see Arctodus’ appendicular morphology as an
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
adaptation to seeing over tall grass and traveling long distances, and see the beast‘s cranial
morphology as an adaptation to bone-cracking (Baryshnikov, Agenbroad et al., 1994). While
generally siding with Emslie and Czplewski, Sorkin concedes that meat obtained from
scavenging might make up a significant part of Arctodus‘ diet at times (2006). The enamel
structure present in A. simus seems to support the idea as well. The zig-zag Hunter-Schreger
bands (HSB) present in Arctodus most closely resemble those in bone-cracking carnivores like
hyena and wolverine (Stefen, 2001). In fairness, zig-zag HSB is of use to any mammal that eats
durable food, vegetable or animal. Still, the only herbivore Stefen studied that possessed zig-zag
HSB was the bamboo-cracking Ailuropoda (2001). The osteophagus hypothesis is further
corroborated by the HSMS specimen 93HS031, a rib with clear canine impressions that may
have been produced by an A. simus.
Arctodus as an Omnivore
Morphometric evidence—The final hypothesis assumes that Arctodus‘ diet was the
same as has been assumed for the ancestor of all modern bears–omnivorous. All of the authors
previously reviewed conceded that Arctodus probably wasn‘t strictly limited to any particular
food source, but they all interpreted the animal as having been more specialized towards active
predation, herbivory, or scavenging than modern bears are. Figueirido et al. beg to differ. They
performed landmark morphometric analysis on the skulls of extant and extinct bears, and plotted
them in morphospace to see whether A. simus falls more in line with herbivorous bears, like
Ailuropoda and T. ornatus, or more with carnivores, like Ursus maritimus. They found neither.
Rather, the proportions of A. simus‘ skull fall in with the omnivorous U. arctos, U. americanus,
U. thibetanus, and Helarctos malayanus (Figuerido, Palmqvist et al., 2009).
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Diet of extant bears—Complicating matters is the fact that extant bears have very
flexible dietary preferences over all scales of time and space. Already mentioned is the fact that
Tremarctos, while strongly associated with herbivory is far from strictly herbivorous now, and
may have been more omnivorous in the past (Goldstein, Paisley et al., 2006) (Figueirido and
Soibelzon, 2009). Ursid bears seem to be similarly flexible adopting whatever food source is
most plentiful and least competed for. For example U. arctos in Alaska make heavy use of
salmon – up to 66% of their diet (Fortin, Farley et al., 2007). In Kluane National Park in the
Yukon, however, the largest fraction of U. arctos’ diet is plants, making up over 80% of the
animals‘ food (Desrochers, Provencher et al., 2002). The diet of U. americanus is similarly
variable. In Utah meat makes up around 30% of the yearly diet, while Floridian bears
concentrate on vegetable matter and insects, with vertebrate matter making up only 7% of the
yearly diet (Black, 2004) (Stratman and Pelton, 1999).
Causes of Extinction
With the lack of consensus on Arctodus‘ habits, is there any consensus on the reasons for
its extinction? No, but some dietary hypotheses are more conducive to some mechanisms of
extinction than others.
Extinction of prey or territorial competition—If Arctodus was a specialized predator
or scavenger, it would have been vulnerable to extinction of its favorite prey species. It thus may
have simply been along for the ride when much of the herbivorous charismatic megafauna went
extinct at the end of the Pleistocene (Matheus, 1995) (Baryshnikov, Agenbroad et al., 1994). If
we assume that Arctodus was a scavenger, there is also room for an alternative hypothesis:
territorial competition from invading U. arctos. If U. arctos was significantly more aggressive
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
than Arctodus, perhaps it displaced the larger bear despite not competing for the same resources
(Agenbroad, 1999).
Loss of browse or the end Pleistocene killer—If Arctodus was largely herbivorous, it
may have gone extinct when climate change at the Pleistocene / Holocene boundary eliminated
its food source. That is, they may have been dependant on non-analog communities of plants and
animals that simply didn‘t exist anymore. Alternatively, it may have suffered whatever fate the
other large herbivores succumbed to.
Competition for food—Finally, if Arctodus was a typically omnivorous bear, as
hypothesized by Figueirido et al., competition from U. arctos for food becomes a viable
hypothesis. In his recent paper in the Journal of Vertebrate Paleontology, he said that, ―. . . A.
simus did so [opportunistically scavenge] in a similar manner as some North American
populations of brown bears (e.g. Alaska and the Yukon) currently do so‖ (Figueirido, Perez-
Claros et al., 2010). Inhabiting the same niche-space would certainly put Arctodus (both A. simus
and A. pristinus) in competition with the invading Eurasian U. arctos.
PROCEDURES
Falsifying the competition hypothesis—Is there any way to put this last hypothesis to
the test? That is, is there any anyway to provide U. arctos with an ‗alibi‘ for the extinction of
Arctodus? They co-occur in several sites, and some authors have put this forward as evidence of
niche partitioning that allowed a sympatric relationship, without significant competition (Emslie
and Czaplewski, 1985). Without good dates for every level of these sites, however, we can‘t
exclude the possibility that there is significant time-averaging, and that A. simus was gone by the
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
time U. arctos took over. The only way to conclusively reject competition between these taxa is
to show that they did not share a niche, and at present the best way to do that is to show that they
weren‘t part of the same communities. Do we have the tools and data required to do that?
Use of multivariate analysis—Indeed we do. Judicious use of multivariate analysis
could reconstruct Pleistocene communities by looking at which taxa tend to co-occur. If A. simus
and U. arctos (and the other Pleistocene bears) tend to be found in separate communities, there
simply wouldn‘t be any opportunity for competition. Multivariate analysis has long been a part
of modern ecological studies and invertebrate paleontology but their popularity has lagged in
vertebrate paleontology because meaningful statistical analysis requires large data sets
containing at least hundreds of specimens – a rare occurrence when working with macrofauna.
Occurrence databases—To solve this problem, we used two pre-existing databases that
provide ample information on occurrences of Pleistocene mammal taxa: the Paleobiology
Database (PBDB) and FaunMap. At this writing, the PBDB provides occurrence data for
169,924 species found in 97,367 collections. Data was collected and modified from the site using
the parameter ―mammalia‖ as well as the 61 collection numbers for North American Pleistocene
ursid sites found in a previous search (Alroy, Sommers et al., 2009). The FaunMap database
focuses on Pleistocene mammals, featuring 2,919 North American sites up to 40,000 years old
(Graham and Lundelius, 1996). Please note that FaunMap has been substantially updated since
the data for this study was collected (Graham and Lundelius, 2010). The combined information
from these two sources produced a presence-absence matrix of 193 bear-containing sites (61
from the PBDB and 162 from FaunMap) and 427 mammalian species. To improve the statistical
rigor of the study, sites with fewer than five taxa and species with fewer than five occurrences
were eliminated, leaving a database with 142 sites and 181 species (APPENDIX 1).
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Producing a similarity matrix—This data was imported into the Paleontological
Statistics program PAST, using a Raup-Crick similarity matrix to compare the number of taxa
that two sites have in common to the number of sites expected if the taxa were distributed
randomly (Hammer and Harper, 2010). In this comparison, two sites with no taxa in common
will have a coefficient of 0.0, and two sites with all taxa in common will have a coefficient of
1.0. The Raup-Crick similarity coefficient was chosen as it works well with a presence-absence
matrix and does not require the relative abundance of data required by other similarity
coefficients (Raup and Crick, 1979). This was important because some of the collections in the
PBDB and sites in FaunMap do not provide abundance data.
Producing a dendrogram and map—PAST was then used to generate a dendrogram
from the similarity matrix with ten clusters of Raup-Crick similarity coefficients greater than 0.8.
The clusters of the dendrogram show which sites tend to have the same taxa (McCune, Grace et
al., 2002). The location of the sites in each cluster was plotted on Google Earth to visualize
possible geographic bases for the similarity between clusters (Google, 2010).
Finding a proxy for environment—Faunal lists were then compiled for each of the
clusters, and the abundance of ursid taxa in each cluster was plotted as a pie-chart. The extant
species in each cluster were checked in the online database, the Animal Diversity Web or ADW
to determine their biome preferences (Myers, Espinosa et al., 2006). The ADW applies several
categories as biome preferences for extant animals. Of these, 10 apply to the mammals that
commonly co-occur with the ursids under study: tundra, taiga, desert, grassland, chaparral,
forest, rainforest, scrubforest, mountain, and wetland/water. The preferences of the extant
mammal species in each cluster were treated as rough proxies for the biome(s) represented by the
sites in the cluster. In establishing this, the amount of statistical weight appropriate for each
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
species was determined by how specialized a species was for a particular biome. For example,
the occurrence of Ochotona princeps in a cluster awarded 1/2 point each to the mountain and
taiga biomes, as these are the only places O. princeps occurs. Puma concolor, however, only
contributes 1/7th of a point to the scores for each of the seven biomes it inhabits. This is because
a habitat specialist like Ochotona tells you more about the biome of a site than a generalist like
Puma.
Displaying environmental differences between clusters—The number of points
accrued for each biome was divided by the total number of points for the cluster, and each was
plotted as a pie-chart. This was also done for Pleistocene bear sites as a whole, to develop a
rough picture of ‗average‘ Pleistocene bear habitat. These measures are all strictly relative, so the
average computed here serves as a benchmark. The biome preferences for each cluster were
subtracted from the average preferences from all the ursid sites and plotted as bar graphs, to
more clearly illustrate environmental differences between the clusters. Information about the
biomes was combined to make more general inferences about the climate in each cluster. Higher
than average scores in tundra, desert, and grassland were considered evidence of an open habitat.
Forested habitats could be identified by high scores in taiga, forest, rainforest, and scrubforest.
Dry climates were evidenced by higher than average numbers of taxa with preferences for desert,
chaparral, or scrubforest biomes, while rainforest and wetland/water based biome preferences
point to a moister environment. These measures are by necessity qualitative, but useful. All of
this information was then compared to the distribution of ursine and tremarctine taxa between the
clusters to provide a better understanding of whether the two groups interacted.
Non-metric multi-dimensional scaling—To better visualize how the clusters differed
from one another, the same data used for the cluster analysis was fed into PAST‘s non-metric
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
multi-dimensional scaling (NMDS) feature, again using Raup-Crick similarity. NMDS is a
variety of multivariate analysis that aims to project multidimensional space down to two or three
dimensions. For the purposes of this study it is superior to similar methods, like principal
components / coordinates analysis because instead of using sites as endpoints for axes, it
calculates its own axes in such a way as to explain the largest amount of variation in the data. In
NMDS, the output is a set of two or three dimensional coordinates for each site, with proximity
between sites indicating similarity in faunal composition (McCune, Grace et al., 2002).
Interpreting NMDS axes—By viewing this information in three dimensions, it is
possible to extract information from the data that isn‘t revealed in cluster analysis. The three axes
that define the space in which the points sit represent the three most important conditions
controlling the distribution and clustering of points. By looking at what differs between sites at
the extremes of each axis, it is possible to determine what these conditions are. Moreover, by
looking at how discreet the clusters recovered from the cluster analysis are, it is possible to
estimate how well these conditions explain the variation in the data. If the volume occupied by
one cluster overlaps considerably with other clusters, we see that three axes we recovered do a
poor job of explaining the variation. If each cluster occupies a discreet space, it indicates that the
clustering can be robustly explained by the conditions controlling each axis. The sites were
viewed in three dimension by loading the coordinates for each site from PAST into the three
dimensional modeling software MeshLab. Each cluster was input to the program as a separate
point cloud with both opaque and transparent convex hulls projected around them (Callieri,
Corsini et al., 2010).
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
RESULTS & DISCUSSION
The Sites as a Whole
The 10 clusters—The first result of this study was the creation of a dendrogram,
showing relative similarity between 142 Pleistocene bear sites, based on the presence or absence
of 181 mammalian species (Fig. 1). The result was the discovery of ten distinct clusters of bear
sites, each linked with similarity coefficients of 0.8 or higher. Please note: these ten faunal
assemblages and the habitats they seem to represent should by no means be taken as an
exhaustive list for the Pleistocene. There are doubtless many Pleistocene faunal assemblages in
which bears have not been found, and many Pleistocene habitats where bears simply did not live.
The map—Plotting the location of each site on Google Earth, and color-coding by
cluster membership revealed a strong geographic bias in nine of the ten clusters (Fig. 2).
Climate/habitat approximation—Using the habitat preferences of the extant members
of each community represented by the ten clusters, we were able to roughly characterize the sort
of habitat each cluster represented, as well as character of all Pleistocene bear sites (Fig.
3)(APPENDIX 2). The variation was considerable, and it appears that all but two of the bear taxa
examined -- A. pristinus and T. floridanus -- were remarkably cosmopolitan.
NMDS—The same data that produced the cluster diagram was used to perform 3D
NMDS analysis, projecting location of the sites in multivariate space down to just three
dimensions. These dimensions appear to represent latitude, longitude, and age, and account for
much of the variation in the original data. That is, sites with similar mammalian communities
tend to be close geographically and in time as is the case in nine of the ten clusters (Figs. 4a-4f).
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Individual Clusters
Below are descriptions of each of the clusters revealed in the original analysis, from I to
X. The geographic range of sites in the cluster is given, along with the approximate age, when it
was known. Also listed is how the habitat differed from the ‗average‘ Pleistocene bear site, as far
as could be determined by proxies. Finally, each summary describes the result of the NMDS
analysis for the cluster.
Cluster I—The sites in this cluster are exclusively in the American west, and consist of
sites found in the PBDB (Fig. 5). The ancillary data recorded there indicates that some of these
sites date from the early Pleistocene, and perhaps even the late Pliocene. Based on the
preferences of the extant fauna found there, the sites in Cluster I point to represent habitats that
are moderately warmer and much more wet and grassy than is typical for Pleistocene bear sites
(Fig. 6). All of the bears found in Cluster I sites were Arctodus simus specimens (Fig. 7). In
NMDS space, this cluster takes up a fairly large volume, despite having relatively few sites. It is
on the left side of axis 1, which seems to be mostly older sites, it straddles a great deal of axis 2,
indicating that membership in this cluster isn‘t strongly controlled by latitude, and exists near the
far end of axis 3, which contains mostly western sites (Fig. 4).
Cluster II—Restricted entirely to Mexico, sites in Cluster II seem to have been slightly
hotter, dryer, and more forested than typical for the average bear site (Figs. 8-9). Mexico seems
to have been completely dominated by tremarctine bears during the Pleistocene, with half of all
bears belonging to A. simus and the balance belonging to Tremarctos floridanus (Fig. 10). In
NMDS space, this small cluster occupies a space adjacent to Cluster I, but with a more distinctly
―southern‖ faunal assemblage (Fig. 4).
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Cluster III—These sites are primarily in the western US, north of Arizona and New
Mexico, with an outlier in southern Alberta (Fig. 11). These sites are much cooler, and slightly
more arid and grassy than is typical of fossil ursid digs (Fig. 12). Many of these sites represent
cave faunas. The vast majority of bears from this cluster are urine, with twelve Ursus arctos, five
U. americanus, and only two belonging to A. simus (Fig. 13). In NMDS space, Cluster III lies on
coordinates that interpreted as the Late Pleistocene northwest (Fig. 4).
Cluster IV—Cluster IV has a very similar geographic distribution to the sites in Cluster
III, but extends slightly further east (Fig. 14). The preferred habitats of the creatures at these sites
seem to be slightly cooler, dryer, and grassier than the mean Pleistocene bear site (Fig. 15).
FaunMap site 0759, The Mammoth Site is the eastern-most site in this cluster. A slim majority of
bears at these sites (six) belong to the tremarctine bear species A. simus, with two occurrences
each of U. arctos and U. americanus (Fig. 16). In NMDS space the difference between Clusters
III and IV is made clear: Cluster IV is further to the left on axis one, indicating that the sites are
probably older (Fig 4).
Cluster V—These localities are largely confined to the southwestern United States and
northern Mexico, with two peculiar outliers in northern California (Fig. 17). These sites are
slightly colder, and much more arid and open than is typical of Ice Age bear sites (Fig. 18).
Many of these sites are caves. Ursine bears (thirteen U. americanus and five U. arctos)
outnumber tremarctine bears (six A. simus) by nearly three-to-one in these sites (Fig. 19).
According to the NMDS analysis, Cluster V should primarily consist of sites in the Middle to
Late Pleistocene southwest (Fig. 4).
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Cluster VI—Limited to Florida, the animals found at these locales seem to indicate a
very hot, very wet, and very densely forested environment (Figs. 20-21). The majority of these
sites seem to date from earlier in the Pleistocene, or perhaps the late Pliocene. Arctodus pristinus
was the only bear found in these early Ice-Age Florida sites (Fig. 22). NMDS analysis indicates
that Cluster VI includes the oldest sites examined in this study. Interestingly, this cluster spans
much of axis 2, which would seem to indicate a larger latitudinal range than was actually present
(Fig. 4).
Cluster VII—This cluster is restricted to a handful of caves in central Texas (Fig. 23).
The fauna preserved here hint at an environment that was much hotter, much drier, and grassier
than was typical (Fig. 24). These Texan bears are 100% ursine, with five of the specimens being
of U. americanus, and only one from U. arctos (Fig. 25). The NMDS analysis indicates a fauna
that is distinctly southwestern, but possibly somewhat more recent than the sites in Cluster V
(Fig 4).
Cluster VIII—The majority of the sites in this cluster are found in one of two
geographic areas – the Appalachian and Ozark mountain ranges. There are two significant
outliers, one in New Mexico and one in Minnesota (Fig. 26). Nearly 76% of these sites are in
caves or fissures. The bears that lived here inhabited an environment that was slightly wetter,
much colder, and more heavily forested than was average for the bruins of the day (Fig. 27).
Every species of North American Pleistocene bear is represented in Cluster VIII sites, but the
vast majority (twenty-nine) of the remains comes from U. americanus (Fig. 28). There were two
occurrences each of A. simus and U. arctos, and only one each of A. pristinus and T. floridanus.
The NMDS results show that these sites hosted a fauna that ranged considerably in latitude and
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
age, but that was relatively restricted in longitude (Fig. 4). This is expected from a
quintessentially eastern community that was relatively stable over the course of the Pleistocene.
Cluster IX—Cluster IX fails to provide insights into the geographic distribution (Fig.
29). The preferences of the extant animals excavated from the sites seem to indicate an
environment that was average for bears during the Pleistocene – slightly warmer, wetter, and
more forested than typical, but not much (Fig. 30). The only bear missing from Cluster IX sites is
the brown bear. The distribution of bear species is hardly even, though, with U. americanus
dominating again, with fourteen occurrences. A. simus has four occurrences, and A. pristinus and
T. floridanus had only two occurrences each (Fig. 31). In NMDS space, Cluster IX overlaps with
many of the clusters on the right half of axis 1, taking up a huge volume (Fig. 4). This indicates
that whatever characteristic that unites the diverse sites in this cluster isn‘t represented on any of
the three axes shown.
Cluster X—Most of the sites in this cluster are found on the Florida Peninsula, but many
are scattered around the American southeast. Unlike Cluster VI which was also centered on
Florida, Cluster X seems to have been overwhelmingly Rancholabrean. There are two significant
outliers, in Texas far to the west of the main group (Fig. 32). These sites were a little chillier,
more forested, and much wetter than the mean. While more forested than average, these sites
probably weren't as forested as the earlier Cluster VI sites (Fig. 33). A. pristinus still turned up in
these upper Pleistocene sites (with two occurrences), but unlike the Cluster VI sites, they no
longer predominate. Instead, T. floridanus narrowly edges out U. americanus for the most
common ursid, with eighteen and fourteen occurrences, respectively (Fig. 34). In NMDS space,
they are adjacent to Cluster VI, but further to the right on axis 1, hinting at a more recent
community (Fig. 4).
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
CONCLUSIONS
Competition Hypothesis Remains Viable
While multivariate analysis can answer many questions about Pleistocene bears, the null
hypothesis tested in this study was more narrowly defined. That is, that ursine bears, particularly
U. arctos were not in competition with and were not responsible for the extinction of North
American tremarctine bears. The results presented above do little to support this hypothesis. Of
the ten clusters recovered from the multivariate analysis, three were strictly tremarctine and one
was strictly ursine. The remaining six site clusters contained a mix of both subfamilies. This
shows that for a majority of Pleistocene sites, ursine and tremarctine bears could be found in the
same mammalian communities, and presumably the same habitats – strong evidence that there
was no barrier to co-occurrence and competition. The four clusters hosting a single subfamily are
intriguing, but not enlightening. Three of the site clusters – I, II, and VII – are small, containing
five or fewer sites, suggesting that the absence of one of the subfamilies may be a result of small
sample size rather than solid evidence that they never inhabited the area. Cluster VI contains a
larger number of sites (12), but it consists of sites dating to the early Pleistocene, so the absence
of U. arctos can almost certainly be traced to the fact that these sites pre-date their immigration
to the Americas.
Arctodus simus as a Habitat Generalist
Nor is there evidence that there was a substantive difference in habitat preference
between the subfamilies. Rather, most North American Pleistocene bears seem to have been
remarkably cosmopolitan. A. simus, U. americanus, and U. arctos all seem to have been
comfortable with their environs almost anywhere. The only bears that showed geographic
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
preference were A. pristinus and T. floridanus, which both favored the southeastern U.S. and the
Florida peninsula in particular. Climate may have been important for some Pleistocene bears, but
not A. simus.
Incidental Observations
In addition to helping to grapple with the question of the extinction of North American
tremarctine bears, this study revealed a number of other interesting observations. One is the
drying and opening out of Florida (or at least Floridian bear habitat) from the beginning of the
Pleistocene to its end. The record of the early Pleistocene as seen in Cluster VI indicates a hot,
wet, densely forested land. By the time ursine bears arrived in Pleistocene Florida was still hot,
not quite as wet, and it is substantially more grassy. Also interesting is the strong domination of
cave sites by U. americanus, possibly reflecting denning preferences.
FUTURE RESEARCH
Accurate Paleoclimate Modeling
Climate modeling of each of the sites in a cluster could confirm that these sites shared
similar environmental conditions, an important step toward an explanation of the peculiar
outliers found in the data. If the Texas sites found in Cluster X had a similar climate to those in
Florida, for example, we might be able to put the question of their odd geographic placement to
rest. Moreover, precise modeling of the climate for each site could define the environmental
constraints on each species of extinct North American bear – more useful than the relative
measures used in this study.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Better Site Dating
Accurate radiocarbon dates for each and every one of these sites would also be of
inestimable value. The present study cannot exclude competition between ursines and
tremarctines as a cause of the latter‘s extinction, but it does nothing to confirm it. Dating the first
and last occurrences of ursines vs. tremarctines in a particular area (or even within the same site)
would allow us to determine whether and for how long different bear taxa actually shared space.
If tremarctine bears disappear from the record before ursines show up in a region we could
discard competition as a driver in the tremarctines‘ extinction. If, however, tremarctines are
extirpated from an area at roughly the same time as ursines appear, it would actually be a point in
favor of competition playing a significant part the extinction.
Detailed Food Web Reconstruction
More complete occurrence data – as may be provided by the newly updated FaunMap
database – combined with more rigorous statistical techniques may also provide new insights. An
accurate reconstruction of the Pleistocene food web, as a whole and for each cluster found in this
analysis, could answer once and for all whether A. simus and U. arctos were in competition for
the same resources (Roopnarine, 2010).
Hope for the Future
The most consistent and encouraging point in each of these studies is that each question
answered will no doubt suggest new ones.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
ACKNOWLEDGMENTS
We thank Dr. J. Alroy for his work creating the Paleobiology Database, Drs. R. Graham
and E. Lundelius Jr. for their work creating the FaunMap database, and Drs. P. Meyers, T. Jones,
T. Dewey, and Msrs. R. Espinosa and G. Hammond for their work creating the Animal Diversity
Web database. Without their work making critical information available to all researchers,
projects like this would be impossible. We also thank Drs. Ø. Hammer, D.A.T. Harper, and P.D.
Ryan for their working creating the PAST statistical analysis package, along with Drs. and Msrs.
M. Callieri, M. Corsini, M. Dellepiane, F. Ganovelli, N. Pietroni, and M. Tarini for their work on
creating the Meshlab 3D visualization and editing software. Without the powerful freeware tools
they created, this endeavor would have been far more expensive. We‘d like to thank Ms. Alicia
Postigo for kindly providing the alternate-language summary, as well as her excellent proof-
reading services. Most of all, we thank D. A. Esker‘s father, Don E. Esker Jr. Without his
constant encouragement, infectious enthusiasm, deep insight, and sage advice, one of our authors
would not be where he is today. He will be sorely missed.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
WORKS CITED
Agenbroad, L.D., 1999, Giant short-faced bear (Arctodus simus), Mammoth Site Research Notes,
Volume 3: Hot Springs, SD, The Mammoth Site of Hot Springs, SD, Inc., p. 4.
Alroy, J., Sommers, M., Carrano, M., Uhen, M., Stigall, A., Bell, C., and Hendy, A., 2009, The
Paleobiology Database, in Alroy, J., ed., Accessed 2009: Santa Barbara, CA, National
Center for Ecological Analysis and Synthesis.
Baryshnikov, G., Agenbroad, L.D., and Mead, J.I., 1994, Carnivores from the Mammoth Site,
Hot Springs, South Dakota, in Agenbroad, L.D., and Mead, J.I., eds., The Hot Springs
Mammoth Site: A Decade of Field and Laboratory Reasearch in Paleontology, Geology,
and Paleoecology: Rapid City, Fenske Printing, Inc., p. 52.
Black, H.L., 2004, Black bears of Utah's east Tavaputs Plateau: Salt Lake City, Utah, Brigham
Young University, 143 p.
Callieri, M., Corsini, M., Dellepiane, M., Ganovelli, F., Pietroni, N., and Tarini, M., 2010,
MeshLab: Pisa, Italy, Istituto di Scienza e Tecnologie dell‘Informazione.
Cope, E.D., 1879, The cave bear of California: The American Naturalist, v. 13, p. 1.
Desrochers, M., Provencher, L., and Dubois, J.-M., 2002, Relationship between environmental
characteristics and the distribution of grizzly bears (Ursus arctos), Kluane National Park
(Yukon): Bulletin de recherche Département de géographie et télédétection, Université de
Sherbrooke, v. 0, p. 72.
Emslie, S.D., and Czaplewski, N.J., 1985, A new record of giant short-faced bear, Arctodus
simus, from western North America with a re-evaluation of its paleobiology:
Contributions in Science, Natural History Museum of Los Angeles County, v. 371, p. 12.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Figueirido, B., Perez-Claros, J.A., Torregrosa, V., Martin-Serra, A., and Palmqvist, P., 2010,
Demythologizing Arctodus simus, the 'short-faced' long-legged and predaceous bear that
never was: Journal of Vertebrate Paleontology, v. 30, p. 13.
Figueirido, B., and Soibelzon, L.H., 2009, Inferring paleoecology in extinct tremarctine bears
(Carnivora, Ursidae) using geometric morphometrics: Lethaia, v. 43, p. 13.
Figuerido, B., Palmqvist, P., and Perez-Claros, J.A., 2009, Ecomorphological correlates of
craniodental variation in bears and paleobiological implications for extinct taxa: an
approach based on geometric morphometrics: Journal of Zoology, v. 277, p. 10.
Fortin, J.K., Farley, S.D., Rode, K.D., and Robbins, C.T., 2007, Dietary and spatial overlap
between sympatric ursids relative to salmon use: Ursus, v. 18, p. 10.
Geist, V., 1986, Did large predators keep humans out of North America?, in Unwin, and Allen,
eds., Cultural Attitudes to Animals Including Birds, Fish and Invertebrates Volume 1:
South Hampton, University of South Hampton, p. 13.
Goldstein, I., Paisley, S., Wallace, R., Jorgenson, J.P., Cuesta, F., and Castellanos, A., 2006,
Andean bear--livestock conflicts: a review: Ursus, v. 17, p. 7.
Google, 2010, Google Earth: Santa Clara County, Ca, Google Corporation.
Graham, R.W., and Lundelius, E.J., 1996, FAUNMAP: A Database Documenting Late
Quaternary Distributions of Mammal Species in the United States, Illinois State Museum
Scientific Papers, Volume 2010: Springfield, IL.
Graham, R.W., and Lundelius, E.J., 2010, FAUNMAP II: New Data for North America with a
Temporal Extension for the Blancan, Irvingtonian and Early Rancholabrean: Berkeley,
CA, University of California, Museum of Paleontology.
Hammer, O., and Harper, D.A.T., 2010, PAST: Oslo, Norway, Hammer & Harper.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Kurten, B., 1966, Pleistocene Bears of North America: Acta Zoologica Fennica, v. 1. Genus
Tremarctos, spectacled bears, p. 120.
Kurten, B., 1967, Pleistocene Bears of North America: Acta Zoologica Fennica, v. 2. Genus
Arctodus, short-faced bear, p. 60.
Kurten, B., and Anderson, E., 1980, Ice Age Mammals of North America: New York, Columbia
University Press, 442 p.
Leidy, J., 1854, On Bison latifrons, Arctodus pristinus, Hippodon speciosus and Merycodus
necatus: Proceedings of the Academy of Natural Sciences of Philadelphia, v. 7, p. 2.
Matheus, P.E., 1995, Diet and co-ecology of Pleistocene short-faced bears and brown bears in
eastern Beringia: Quaternary Research, v. 44, p. 6.
McCune, B., Grace, J.B., and Urban, D.L., 2002, Analysis of Ecological Communities: Glenden
Beach, Oregon, MjM Software Design, 304 p.
Myers, P., Espinosa, C.S., Parr, T.J., Hammond, G.S., and Dewey, T.A., 2006, The Animal
Diversity Web, Accessed 2010: Ann Arbor, MI, University of Michigan Museum of
Zoology.
Raup, D., and Crick, R., 1979, Measurment of faunal similarity in paleontology: Journal of
Paleontology, v. 53, p. 14.
Roopnarine, P.D., 2010, Networks, extinction and paleocommunity food webs, in Alroy, J., and
Hunt, G., eds., Quantitative Methods in Paleobiology, Volume 16: The Paleontological
Society Papers, The Paleontological Society, p. 143-161.
Sorkin, B., 2006, Ecomorphology of the giant short-faced bears Agriotherium and Arctodus:
Historical Biology, v. 18, p. 20.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
Stefen, C., 2001, Enamel structure of arctoid Carnivora: Amphicyonidae, Ursidae, Procyonidae,
and Mustelidae: Journal of Mammalogy, v. 82, p. 12.
Stratman, M.R., and Pelton, M.R., 1999, Feeding ecology of black bears in northwest Florida:
Florida Field Naturalist, v. 27, p. 7.
Troya, V., Cuesta, F., and Peralvo, M., 2004, Food habits of Andean bears in the Oyacachi River
Basin, Ecuador: Ursus, v. 15, p. 3.
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
FIGURE CAPTIONS
FIGURE 1. Ten clusters of Pleistocene bear sites grouped at 80% Raup-Crick similarity.
[Planned for page width]
FIGURE 2. Map of the locations of each site in the analysis, color-coded by cluster membership,
with convex hull projected around each cluster. The extensively overlapping Cluster IX (green
arrow) does have a convex hull. [Planned for page width]
FIGURE 3. Percentage break down of potential habitats represented by mammals found at the
bear sites under study. [Planned for page width]
FIGURE 4a-c. Sites in the study placed in 3D space by NMDS, color coded by cluster
membership, with 3D convex hulls projected around the clusters. Cluster IX is omitted to allow a
clearer view of the other clusters. [Planned for column width]
FIGURE 4d-f. Sites in the study placed in 3D space by NMDS, color coded by cluster
membership, with 3D convex hulls projected around the clusters. Convex hulls are rendered
translucent to allow display of Cluster IX without obscuring other clusters. [Planned for column
width]
FIGURE 4a-f. (Single figure combining all views of figure 4. Use at your own discretion, with
both descriptions.) [Planned for page width]
FIGURE 5. Geographic distribution of sites in Cluster I. [Planned for page width]
FIGURE 6. Habitats represented in Cluster I determined by preference of mammalian proxies.
[Planned for column width]
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
FIGURE 7. % Breakdown and total occurrences of bears found in Cluster I sites. [Planned for
column width]
FIGURE 8. Geographic distribution of sites in Cluster II. [Planned for page width]
FIGURE 9. Habitats represented in Cluster II determined by preference of mammalian proxies.
[Planned for column width]
FIGURE 10. % Breakdown and total occurrences of bears found in Cluster II sites. [Planned for
column width]
FIGURE 11. Geographic distribution of sites in Cluster III. [Planned for page width]
FIGURE 12. Habitats represented in Cluster III determined by preference of mammalian proxies.
[Planned for column width]
FIGURE 13. % Breakdown and total occurrences of bears found in Cluster III sites. [Planned for
column width]
FIGURE 14. Geographic distribution of sites in Cluster IV. [Planned for page width]
FIGURE 15. Habitats represented in Cluster IV determined by preference of mammalian proxies.
[Planned for column width]
FIGURE 16. % Breakdown and total occurrences of bears found in Cluster IV sites. [Planned for
column width]
FIGURE 17. Geographic distribution of sites in Cluster V. [Planned for page width]
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
FIGURE 18. Habitats represented in Cluster V determined by preference of mammalian proxies.
[Planned for column width]
FIGURE 19. % Breakdown and total occurrences of bears found in Cluster V sites. [Planned for
column width]
FIGURE 20. Geographic distribution of sites in Cluster VI. [Planned for page width]
FIGURE 21. Habitats represented in Cluster VI determined by preference of mammalian proxies.
[Planned for column width]
FIGURE 22. % Breakdown and total occurrences of bears found in Cluster VI sites. [Planned for
column width]
FIGURE 23. Geographic distribution of sites in Cluster VII. [Planned for page width]
FIGURE 24. Habitats represented in Cluster VII determined by preference of mammalian
proxies. [Planned for column width]
FIGURE 25. % Breakdown and total occurrences of bears found in Cluster VII sites. [Planned
for column width]
FIGURE 26. Geographic distribution of sites in Cluster VIII. [Planned for page width]
FIGURE 27. Habitats represented in Cluster VIII determined by preference of mammalian
proxies. [Planned for column width]
FIGURE 28. % Breakdown and total occurrences of bears found in Cluster VIII sites. [Planned
for column width]
ESKER, WILKINS, AND AGENBROAD—MULTIVARIATE ANALYSIS OF URSIDS
FIGURE 29. Geographic distribution of sites in Cluster IX. [Planned for page width]
FIGURE 30. Habitats represented in Cluster IX determined by preference of mammalian proxies.
[Planned for column width]
FIGURE 31. % Breakdown and total occurrences of bears found in Cluster IX sites. [Planned for
column width]
FIGURE 32. Geographic distribution of sites in Cluster X. [Planned for page width]
FIGURE 33. Habitats represented in Cluster X determined by preference of mammalian proxies.
[Planned for column width]
FIGURE 34. % Breakdown and total occurrences of bears found in Cluster X sites. [Planned for
column width]
