The uncertain blitzkrieg of Pleistocene megafauna


  • Barry W. Brook,

    Corresponding author
    1. Key Centre for Tropical Wildlife Management, Charles Darwin University, Darwin 0909, Northern Territory, Australia
    2. Center for Ecological Research, Kyoto University, Otsu 520-2113, Japan
    Search for more papers by this author
  • David M. J. S. Bowman

    1. Key Centre for Tropical Wildlife Management, Charles Darwin University, Darwin 0909, Northern Territory, Australia
    Search for more papers by this author

Barry W. Brook,


We investigated, using meta-analysis of empirical data and population modelling, plausible scenarios for the cause of late Pleistocene global mammal extinctions. We also considered the rate at which these extinctions may have occurred, providing a test of the so-called ‘blitzkrieg’ hypothesis, which postulates a rapid, anthropogenically driven, extinction event. The empirical foundation for this work was a comprehensive data base of estimated body masses of mammals, comprising 198 extinct and 433 surviving species > 5 kg, which we compiled through an extensive literature search. We used mechanistic population modelling to simulate the role of human hunting efficiency, meat off-take, relative naivety of prey to invading humans, variation in reproductive fitness of prey and deterioration of habitat quality (due to either anthropogenic landscape burning or climate change), and explored the capacity of different modelling scenarios to recover the observed empirical relationship between body mass and extinction proneness. For the best-fitting scenarios, we calculated the rate at which the extinction event would have occurred. All of the modelling was based on sampling randomly from a plausible range of parameters (and their interactions), which affect human and animal population demographics. Our analyses of the empirical data base revealed that the relationship between body mass and extinction risk relationship increases continuously from small- to large-sized animals, with no clear ‘megafaunal’ threshold. A logistic ancova model incorporating body mass and geography (continent) explains 92% of the variation in the observed extinctions. Population modelling demonstrates that there were many plausible mechanistic scenarios capable of reproducing the empirical body mass–extinction risk relationship, such as specific targeting of large animals by humans, or various combinations of habitat change and opportunistic hunting. Yet, given the current imperfect knowledge base, it is equally impossible to use modelling to isolate definitively any single scenario to explain the observed extinctions. However, one universal prediction, which applied in all scenarios in which the empirical distribution was correctly predicted, was for the extinctions to be rapid following human arrival and for surviving fauna to be suppressed below their pre-‘blitzkrieg’ densities. In sum, human colonization in the late Pleistocene almost certainly triggered a ‘blitzkrieg’ of the ‘megafauna’, but the operational details remain elusive.


The Late Pleistocene saw a global loss of vertebrate species diversity of perhaps unrivalled magnitude in the Cainozoic (Alroy, 1999). Surviving mammals, birds and reptiles are generally smaller than the species that died out, collectively known as the extinct ‘megafauna’. The bulk of evidence points to an anthropogenic role in the extinctions (Kay, 1998), although the effect of humans is not clear-cut (Beck, 1996; Grayson, 2001; Brook & Bowman, 2002), as evidenced by the world-wide survival of some large vertebrates, and in particular the diverse African large mammal assemblage. The survival of the African ‘megafauna’ has usually been explained as being a consequence of their coevolution with human predators (Martin, 1973, 1984; Flannery, 1999), or alternatively, the apparently limited modification of that continent's vegetation types by late Quaternary climate change (Owen-Smith, 1988). Another discrepancy concerns the persistence of species that are of a similar size (and often related) to extinct species, and has been attributed to contrasting habitat preferences (Alroy, 2001), the uneven effect of changed habitat conditions caused by anthropogenic landscape burning (Miller et al., 1999) or hunting of ‘keystone herbivores’ (Owen-Smith, 1988), and variation in the capacity of prey species to find refugia (Johnson, 2002). Evaluation of these hypotheses presents an enormous intellectual challenge, given the limited ability to undertake field observations or experimental studies, and is therefore increasingly tackled by theoretical modelling (Brook & Bowman, 2002).

The definition of ‘megafauna’ is of critical importance in testing various hypotheses for these extinctions, because allometric body mass relationships (e.g. Damuth, 1981; Hennemann, 1983) are used to underpin the models (see Alroy, 2001; Brook & Bowman, 2002). Yet despite this reliance, there has been surprisingly limited consideration of the comparative distribution of body masses of extinct and surviving species. Indeed, most authors have restricted their discussion of extinct ‘megafauna’ by reference to some arbitrary body mass threshold, often set at around 45 kg (e.g. Caughley & Krebs, 1983; Martin & Klein, 1984; Beck, 1996; Lambert & Holling, 1998; Flannery, 1999; Miller et al., 1999; Gittleman & Gompper, 2001; Roberts et al., 2001), despite there being no obvious functional basis for this threshold (Owen-Smith, 1988). A point often overlooked is that many medium- and some small-sized mammals also went extinct (Johnson, 2002).

To date, this extinction debate has been stymied by a lack of direct tests of model hypotheses using relevant empirical data (Grayson & Meltzer, 2003). One significant exception is Johnson's (2002) innovative use of known and inferred within taxonomic family body mass relationships to demonstrate a strong effect of reproductive rate on the likelihood of extinction of Late Quaternary mammals. Whilst Johnson (2002) successfully falsified the proposition that hunting was body-size selective (i.e. he showed large prey were not targeted preferentially), he did not determine directly the relationship between body mass and extinction risk. Furthermore, implicit in his analysis was the assumption that only reproductive rate correlates with body mass, when it is in fact often a robust proxy for many other key demographic factors (i.e. relative abundance/density, geographical and home range size, resource use, mobility, population variability, generation length and complexity of social structure (Blackburn & Gaston, 1994; McKinney, 1997; Gardezi & da Silva, 1999), all of which can strongly influence the intrinsic vulnerability of taxa (see Raup, 1994; Purvis et al., 2000).

Building on Johnson's approach, we provide the most comprehensive comparison of the body masses of extinct and surviving ‘megafauna’ yet published, and use these data to evaluate directly both phenomenological and mechanistic explanations for these extinctions using statistical and population modelling.


Body mass data

We compiled an exhaustive data set of estimated body masses for a diverse range of 198 extinct and 433 surviving terrestrial Late Quaternary eutherian and marsupial mammal species > 5 kg, from Africa, Australia, Eurasia, North and South America and Madagascar. For each continent, the relationship between body mass and extinction probability was described using logistic regression (following Lessa et al., 1997). However, in order to develop a ‘universal’ extinction–body mass model, we subsequently used the empirical relationship developed by Burness et al. (2001) to correct for the strong relationship between body mass and continental area by dividing each species body mass by the maximal predicted value for their respective land mass, and regressed extinction risk against this single predictor variable.

Extinction mechanisms

To evaluate mechanistic explanations, we developed a relatively simple model of human and megafauna population dynamics, based on five key parameters (Brook & Bowman, 2002): maximal replacement rate (rm) and equilibrium density (D) of megafaunal prey populations, density of human populations (H), maximal rate of off-take by human hunters (O), and relative naivety of prey (z). We used this model to simulate 150 alterative scenarios of human impact on megafaunal prey populations (i.e. a small but strategically stratified subset of possibilities), with model parameters assigned a range of plausible random combinations (from the ranges given in Table 1), and in each scenario it was iterated 1000 times for each of seven size classes of area-corrected log10 body mass shown in Fig. 1. Simulation results (proportion of iterations extinct) were compared with the empirical extinction–body mass relationship based using the mean squared deviation.

Table 1.  Definition and distribution of parameters in the megafauna simulation model, which was used to generate scenarios to fit against the empirical body mass–extinction risk relationship
ParameterDefinitionSampling distribution
Prey replacement rate (rm)log10(rm) = 0.72 − 0.27 log10 mass (g)Normal (SD = 0.295)
Prey density (D)log10(D) = 4.23 − 0.75 log10 mass (g)Normal (SD = 0.456)
Initial human population (N)Population size of human foundersUniform (50,150)
Human growth rate (gm)Maximum percentage increase per yearUniform (1%, 2.5%)
Human density (H)Number of persons km−2Uniform (0.02, 0.21)
Hunting off-take (O)Prey items per person per yearDiscrete (0, 5, 10, 15, 30)
Prey ‘naivety’ (z)Scaling parameter (dimensionless)Discrete (1.1, 1.2, 1.3, 1.5, 1.7)
Hunting success (HS)HS = ODz/(0.4 + Dz)n/a
Habitat quality (Mrm,MD)[rm × Mrm], [D × MD]Discrete (1, 0.9, 0.8, 0.7, 0.5)
Figure 1.

Frequency distribution of continental area-corrected log10 body mass (g) of 198 Late Quaternary extinct mammals (black bars) and 433 modern surviving mammals (white bars) with a body mass > 5 kg: (a) The disparately shaped distributions of extinct vs. surviving mammals; (b) a reconstructed (i.e. pre-extinction) mammalian body mass distribution.

Rate of extinctions

To examine the rate of extinction, we introduced an initial founder human population of 50–150 individuals (based on Martin, 1973; Holdaway & Jacomb, 2000; Alroy, 2001), with prey populations initially set at maximum densities and their carrying capacities determined according to small (500,000 km2), intermediate (8 million km2) and large (25 million km2) land-mass areas. Although our model ignores the potential complexities of spatial heterogeneity, Alroy's (2001) model generated very similar results under either spatially implicit or spatially explicit assumptions.

Online appendices

A complete listing of the species-by-species body mass data (with a catalogue of source literature) and a detailed description of the statistical analyses, potential biases, and structure and assumptions of the population model, is given in two electronic appendices. (see Supplementary Material)


Statistical modelling

Our analysis of 198 extinct and 433 surviving terrestrial Late Quaternary mammal species, weighing 5–6000 kg, reveals that whilst extinct species were certainly significantly larger than survivors [median log10 body mass (g) (extinct) = 5.18; (surviving) = 4.25, Mann–Whitney W = 111,099, P < 0.0001], there is no clear body mass threshold at which to logically separate the two groups (Fig. 1). The body size distribution of surviving mammals in these size classes follows an approximately negative exponential distribution (see also Blackburn & Gaston, 1994; Gardezi & da Silva, 1999) whereas the distribution of extinct species is more symmetrically distributed, with the majority of species clustered around an intermediate body size of 50–500 kg.

This continuum of vulnerability is almost perfectly described by a binary logistic ancova model relating the probability of extinction to log10 body mass (g) and ‘continent’ (categorical variable), which correctly predicts the actual fate of species in 91.7% of cases (Somer's D = 0.83, G = 367.3, d.f. = 7, P < 0.0001). Although the universal outcome was that the larger a mammal species, the greater its risk of extinction, absolute risk clearly differed according to ‘continent’, with larger areas able to support larger persisting ‘megafauna’. A global model, which uses the area-corrected log10 body mass (g) as a single predictor variable in a logistic regression, also results in a highly significant relationship (G = 242.5, d.f. = 1, P < 0.0001, concordance = 86.2% or 80.3% without area correction).

Mechanistic population modelling

We found it possible to closely match the empirical relationship between extinction risk and body mass using a relatively simple mechanistic simulation model (Fig. 2b, continuous line). Of the 150 different combinations of off-take rate (O), prey naivety (z), and suppression of prey reproductive rates (rm) and habitat quality (D) (see Table 1 and online Appendices), the scenario that most closely matched the empirical relationship was as follows: moderate levels of hunting off-take (maximum harvest of five prey individuals per person per year, but unrelated to body mass of prey), only minor deviation from maximal prey naivety (z = 1.1), and a moderate (20%) suppression of prey fitness and habitat quality [mean squared error (MSE) = 0.0025, correlation (r) = 0.993]. Yet almost equally good fits were obtained with by holding O = 5 and z = 1.1 and substituting different levels of suppression of rm and D (e.g. MSE = 0.0050, r = 0.991, when O = 5, z = 1.1, rm = 0.9 and D = 0.7). If a higher level of O = 10 is assumed, the optimal z increases to 1.2, and the fit deteriorates only slightly (MSE = 0.0076, r = 0.984).

Figure 2.

Relationship of proportion of extinct to surviving mammals vs. continental area-corrected log10 body mass (ACBM), where logit[P] = 2.833 + 2.195 ACBM. (a) Phenomenological model: a logistic regression equation fitted through these data, which provides an 86.2% predictive concordance with the empirical data. (b) Mechanistic model: the predictions of a simulation model in which either (1) the human off-take term (O) was tuned to match the consumption rate required for subsistence hunting (dashed line), and (2) the scenario that most closely matched the empirical data (solid line), in which hunting off-take (O) varied randomly between U[0, 5] (irrespective of body mass), the predation efficiency [prey naivety] term (z) varied randomly between U[1, 1.1], and the maximal prey growth rate (rm) and equilibrium density (D) were diminished by 20%, due, for instance, to habitat alteration by humans and/or climate change.

In sum, a large number of plausible parameter combinations can result in a robust fit to the empirical data. However, a poor and contradictory fit (MSE = 0.265, r = −0.931) was obtained for those scenarios in which hunting off-take was assumed to be geared solely towards acquiring sufficient resources to meet subsistence requirements – such that more smaller than larger prey were harvested in order to provide the same volume of meat (Fig. 2b, dashed line). In aggregate, these results suggest strongly that if hunting was the major cause of the extinctions, it was conducted for purposes other than just subsistence (e.g. opportunism, social prestige).

Rapidity of extinction wave

The rate of ‘megafaunal’ extinctions predicted by our model following the establishment of human populations was surprisingly rapid under all scenarios (Fig. 3a) and, moreover, insensitive to assumptions about either body mass or continental/island area. For example, using the empirically derived body mass–extinction risk relationship, an intermediate sized landmass (e.g. 8 million km2), mammalian species with a predicted extinction risk during the Late Pleistocene of 10%, 25% and 50% had median times to extinction of 778, 741 and 711 years. For the corresponding 90%, 75% and 50% of surviving species, the median degree of population reduction was 3.1%, 6.3% and 14.1%, although some species populations were reduced to densities far below these median values (Fig. 3b), and were therefore functionally extinct due to the stochastic hazards associated with small population size. For a larger landmass (e.g. 25 million km2), the median time to extinction for a mammal with a 50% extinction risk was 782 years (median density reduction in survivors = 19%), whilst for a large island (e.g. 500,000 km2), it is 547 years (5.9%). The common prediction for all body sizes and landmass areas was for a rapid extinction of a proportion of the fauna following human arrival, and for swift and sustained density suppression of the survivors.

Figure 3.

Case study of a medium-sized land mass (8 million km2). (a) Distribution of simulated extinctions times following human arrival, and (b) percentage reduction in average population size for persisting populations, of a hypothetical megafaunal prey species with a 50% extinction risk. Results are derived from the scenario described in Fig. 2b, but are effectively identical in all scenarios and for larger and smaller landmass areas.


In contrast to recent assertions (Johnson, 2002), we contend that body mass does provide a powerful predictor of the selectivity of extinctions in the Late Pleistocene mammal fauna, for both continent-specific and global models. Moreover, our comprehensive compilation of the body masses of 631 species demonstrates that no single threshold size obviously delineates the victims of extinction from the survivors. These findings are explicable when one considers the characteristically continuous and pervasive way in which a wide range of life history attributes commonly associated with extinction vulnerability, such as reproductive rate, generation length, relative abundance, home range size and population variability (Raup, 1994; McKinney, 1997; Purvis et al., 2000), scale to body mass (Damuth, 1981; Caughley & Krebs, 1983; Hennemann, 1983; Owen-Smith, 1988). Thus as species become larger, they will not only tend towards lower reproductive rates [the crux of Johnson's (2002) argument], but will also be rarer, less able to respond rapidly to changing circumstances, more prone to Allee effects (e.g. social disruption), and exist closer to the minimum viable population size threshold below which stochastic hazards predominate.

A feature of the Late Pleistocene extinctions that has perplexed some authors concerns the differential survival of geographically overlapping, often phylogentically proximate, animals of equivalent body sizes; a classic example being the extinct Australian Sthenurine kangaroos, many of which were smaller than surviving macropods (Johnson & Prideaux, 2003). Alroy (2001) speculated that the differential extinction of bison species was due to ecological factors not included in his overkill model, whilst Brook & Bowman (2002) suggested the explanation may lie with some combination of differential naivety of prey, selective hunting of some species, and contrasting ability to tolerate hunting pressure (due to fortuitous ‘preadaption’) following changes in habitat conditions induced by landscape burning, climatic change or both (see also Kay, 1998). Johnson (2002) favoured the role of behavioural traits (e.g. arboreal or nocturnal species) or habitat preferences (e.g. closed canopy or cold tolerant vegetation), which likely conferred protection against human predation. However, our study points to the importance of probabilistic effects for taxa of the same size, which blurs the boundary between extinct ‘megafauna’ and extant ‘mesofauna’, because no rigid body mass threshold exists. Thus the idea that this particular ‘problem’ warrants a mechanistic explanation, is unfounded.

Our modelling demonstrates that there are many divergent explanatory paths capable of satisfactorily reproducing the empirical body mass–extinction risk relationship, including multi-causal scenarios in which hunting impacts may have been amplified by anthropogenic or climate-driven deterioration of habitat quality (e.g. Stuart, 1991). In previous modelling exercises, the principle of parsimony has been used as a basis for ignoring explicitly multi-causal scenarios in favour of single factor explanations. However, this philosophy is flawed, because all other factors are nevertheless implicitly and arbitrarily (given the current knowledge base) assigned fixed values such that it is difficult for the model to deviate from a predetermined path (Brook & Bowman, 2002). Nonetheless, modelling can isolate some contingencies. For instance, if overhunting is assumed to have been a major driver of the extinctions, then our results show (see Fig. 3) that it must have been conducted for purposes beyond simple subsistence (see also Choquenot & Bowman, 1998), or that increases in dietary breadth by prey-switching was a fundamental moderating mechanism (Winterhalder & Lu, 1997; Grayson, 2001).

An important general prediction is manifest under all scenarios – those species driven to extinction by invading human hunters (with or without concomitant habitat alteration), succumb rapidly (i.e. within a few thousand years). Indeed, even for scenarios we have not explored in detail, such as the preferential or opportunistic harvesting of juveniles, ambush hunting of waterholes during droughts, or the cascading affect of loss of keystone species, would all tend to accelerate the rate at which the extinctions unfolded. This finding agrees strongly with other, independently derived megafaunal simulation models (e.g. Mosimann & Martin, 1975; Choquenot & Bowman, 1998; Holdaway & Jacomb, 2000; Alroy, 2001). Gradual die-offs through steady hunting attrition over tens of millennia, as envisaged by Owen-Smith (1988) and Johnson (2002), are unlikely to occur. In other words, overkill, if it happens, will inevitably result in a ‘blitzkrieg’ (sensuMartin, 1984), albeit a selective one. The survivors are suppressed rapidly to some lower ‘equilibrium’ population density, but as a result are left more vulnerable to later extinction by future stresses (Balmford, 1996; Burney et al., 2003; Kerr, 2003). From a geological perspective, the later extinction of these taxa would, in many cases, nevertheless appear directly coincident with human activities. These model predictions support contemporary research findings that (1) extinction due to naivety or low natural density will most likely occur if prey are unable to sustain the initial wave of new predation (Gittleman & Gompper, 2001), and (2) new predators can change the tolerance of surviving species to environmental change (e.g. Schoener et al., 2001).


The realization that the observed body mass–extinction risk relationship can be derived from a galaxy of permutations of the parameters described in Table 1 relieves the need to argue that precisely the same mechanism was responsible for the megafaunal extinctions following the arrival of humans in ‘naive’ lands. Winnowing the ensemble of plausible explanations for the extinctions requires the further development of a robust geochronological framework – permitting a more precise determination of the overlap between megafauna and humans (Beck, 1996; Roberts et al., 2001; Brook & Bowman, 2002). For example, once direct dating of bone becomes routine, a direct test can be made of the model prediction of an initial rapid defaunation, followed by gradual extinction of species unable to tolerate new stresses. Palaeoenvironmental data and stable isotope analyses of fossil remains are also required, to identify habitat preferences and feeding ecologies of extinct ‘megafauna’ from each continent (Owen-Smith, 1988; Miller et al., 1999; Johnson & Prideaux, 2003). Geochronological control of climate change is also critical, as such events are the driving force behind palaeoecological changes and isotopic variations (Grayson et al., 2001). We conclude that human colonization in the late Pleistocene triggered a ‘blitzkrieg’ of the ‘megafauna’, but the operational details remain uncertain.


We thank Chris Johnson, Richard Roberts, Navjot Sodhi, Tim Flannery, Peter Whitehead and Norio Yamamura and an anonymous referee for comments and suggestions, and Guy Pardon for helping to compile the data base. This work was undertaken by the Key Centre for Tropical Wildlife Management, Charles Darwin University, under funding from the Australian Research Council and the Center for Ecological Research through the Kyoto University Visiting Scholars Programme.

Supplementary materials

The following material is available from

Appendix S1 Detailed description of analyses and models.

Appendix S2 Source data on body masses.


Dr Barry Brook is a population ecologist based in Darwin, Northern Territory. His research focuses on extinction theory, conservation of biodiversity, management of wildlife populations, and how past extinction crises inform on present-day environmental issues.

Professor David Bowman has worked throughout northern Australia and has a particular interest in landscape fire, indigenous ecological knowledge, and in the application of historical and ecological biogeography to land management. His recently published book entitled Australian rain forests: Islands of green in a land of fire (Cambridge University Press) presented a challenging new theory concerning the flammability of Australian vegetation.