Small populations of Palaeolithic humans in Cyprus hunted endemic megafauna to extinction

Abstract

The hypothesized main drivers of megafauna extinctions in the late Quaternary have wavered between over-exploitation by humans and environmental change, with recent investigations demonstrating more nuanced synergies between these drivers depending on taxon, spatial scale, and region. However, most studies still rely on comparing archaeologically based chronologies of timing of initial human arrival into naïve ecosystems and palaeontologically inferred dates of megafauna extinctions. Conclusions arising from comparing chronologies also depend on the reliability of dated evidence, dating uncertainties, and correcting for the low probability of preservation (Signor–Lipps effect). While some models have been developed to test the susceptibility of megafauna to theoretical offtake rates, none has explicitly linked human energetic needs, prey choice, and hunting efficiency to examine the plausibility of human-driven extinctions. Using the island of Cyprus in the terminal Pleistocene as an ideal test case because of its late human settlement (~14.2–13.2 ka), small area (~11 000 km²), and low megafauna diversity (2 species), we developed stochastic models of megafauna population dynamics, with offtake dictated by human energetic requirements, prey choice, and hunting-efficiency functions to test whether the human population at the end of the Pleistocene could have caused the extinction of dwarf hippopotamus (Phanourios minor) and dwarf elephants (Palaeoloxodon cypriotes). Our models reveal not only that the estimated human population sizes (n = 3000–7000) in Late Pleistocene Cyprus could have easily driven both species to extinction within < 1000 years, the model predictions match the observed, Signor–Lipps-corrected chronological sequence of megafauna extinctions inferred from the palaeontological record (P. minor at ~12–11.1 ka, followed by P. cypriotes at ~10.3–9.1 ka).

1. Introduction

Explanations for the global extinction of hundreds of large terrestrial species during the late Quaternary [1] have matured from relying on simple binary drivers to a more nuanced demonstration of synergistic mechanisms varying across taxa and regions [2–9]. However, temporal variation in species composition inferred from the zooarchaeological record is still often attributed either to (i) changing environmental conditions altering natural abundances, (ii) humans depleting populations through subsistence offtake, or (iii) a combination of the two [10–13]. Yet the relative contribution of these two mechanisms and/or their combination to the loss of megafauna during the Late Pleistocene and early Holocene are still largely examined based on inferred chronologies of relative human appearance and megafauna extinctions [5,8,14]. When the estimated window of human appearance to naïve ecosystems estimated from archaeological evidence precedes (but not by too much) palaeontologically inferred extinction dates, the conclusion tends to invoke human endeavour as the primary cause of the extinction [15,16]. On the other hand, when the palaeontological record suggests an extinction event occurred well before inferred human arrival, the assumed mechanism underlying the extinction tends to be environmental change. Here, proxy data indicating large climatological fluctuations [17] or via species distributions derived from climate niche models [13] in the period immediately before inferred extinction tend to be the basis for conclusions that environmental change drove regional extinctions of large terrestrial species.

Despite recent analytical advances in such (spatio-) temporal analyses [8,18], the quality and robustness of the underlying date estimates are still central to conclusions arising from the comparison of appearance-extinction chronologies. Acknowledging too that robust time series of a species’s decline to extinction and the clear, unambiguous dates of initial human arrival are extremely rare, even high-quality data can still only hypothesize the mechanisms underlying the overarching causes [19]. Regardless, analyses comparing chronologies provide only weak inference because they do not explicitly incorporate the mechanisms of extinction (or persistence). In other words, mechanistic approaches are still required to determine how particular populations of humans were able to drive specific species to extinction, and how environment changes could lead to loss of fitness and the eventual demise of an entire species.

There are in fact few quantitative or modelled examples of plausible ecological mechanisms driving extinction, whether the main determinants were human over-exploitation, environmental change, or a combination of both. Exceptions include mechanistic models of varying complexity that have been developed to discern the dynamics of megafauna extinctions in response to humans [15,19–26]; however, none of these models has explicitly included the energetic needs of Palaeolithic hunter–gatherers, hunting efficiency, and prey selection and converted these parameters into equivalent animal offtake rates by humans (although Alroy [15] and Lima-Ribeiro & Diniz-Fihlo [26] do include some of these elements). The main reasons for this gap likely arise from the complexity of hunter–gatherer foraging systems [27], a lack of relevant data, and uncertainties regarding human patterns of expansion and settlement [28].

These three characteristics—human energetic needs, hunting efficiency, and prey selection—are essential to convert inferred human population sizes into ecologically plausible prey-offtake rates. Energetic demand is the required nutrient intake derived from different sources (plants and animals) to maintain survival and reproduction [29,30]. In age-structured populations, energetic demand also varies according to individual age and sex [31]. In terms of animal-derived nutrients, hunter–gatherers also must balance the costs of the hunting activity (e.g. travel, landscape accessibility, risk, carcass transport, and processing) with the potential energetic gains from successful procurement of prey [13,31–35]. Prey selection is an inherent component of these decisions, with larger, swifter (and potentially more dangerous) prey providing more potential energy, but simultaneously presenting a higher risk of failed acquisition and injury/death [13,32]. In other words, the numbers of available prey in any landscape do not necessarily translate into realized energy acquisition in predators, including humans, so accounting for these characteristics is necessary to test hypotheses regarding the plausibility of driving prey populations to extinction via hunting.

The island of Cyprus in the eastern Mediterranean offers an ideal set of conditions to test whether recently arrived populations of pre-agropastoralist humans had the capacity to drive megafauna species to extinction. Cyprus is an insular environment with a maximum area at the approximate period of human arrival (~14.2–13.2 ka) [36] of only ~11 000 km², making spatial heterogeneity in archaeological and palaeontological evidence less important for inferring regional trends compared to large regions such as Eurasia with considerably larger gaps in spatial coverage of the available evidence [21], South America [5], or Sahul [8,16]. Most importantly, there were only two megafauna species on the island when people first arrived (although there were other, smaller terrestrial species recovered from zooarchaeological records—a genet Genetta plesictoides, a murid Mus sp., a shrew Crocidura suaveolens praecypria, and a megachiropteran [37]), making models of prey choice more tractable to construct compared to those situated in more biodiverse environments. Neither is there any evidence that Cyprus had predators large enough to kill either species [38,39] prior to humans arriving. In addition, the patterns of initial human arrival and spread in Cyprus have recently been established with considerable certainty [36].

The two ‘large’ (>100 kg body weight) species present in Cyprus when people first arrived were the dwarf hippopotamus Phanourios minor and the dwarf elephant Palaeoloxodon cypriotes [40]. Phanourios minor was the smallest dwarf hippopotamus in the Mediterranean region [39] and weighed ~130 kg at adulthood [41]. It was adapted to a largely terrestrial, browsing lifestyle given its lower orbits and nostrils [42,43], loss of the 4th molar, brachydont molars, and a shortened and narrow muzzle [43–45] when compared to semi-aquatic forms. Ancient mitogenomic analyses suggest the P. minor diverged from the common hippopotamus Hippopotamus amphibius approximately 1.36−1.58 Mya [46]; the Cypriot species is therefore potentially better placed in the Hippopotamus rather than the Phanourios genus [46], although this recommendation is not yet widely accepted. Palaeoloxodon cypriotes weighed only about 530 kg and was therefore <10% of the size of its mainland ancestor [41]. The species probably derived from the straight-tusked elephant P. antiquus that inhabited Europe and Western Asia during the Middle and Late Pleistocene [47]. Analyses of other species of dwarf elephant from the Mediterranean (e.g. Palaeoloxodon ex gr. P. falconeri from Sicily) suggest their small size implied a shorter life span than its mainland ancestor, more rapid growth, shorter gestation, larger relative brain size, and different thermoregulatory mechanisms [48].

The arrival of an efficient, novel predator (humans) was therefore potentially catastrophic to these predator-naïve populations. Median human population sizes predicted for Cyprus during the Late Pleistocene have been previously estimated in the several thousands, from initial arrival at 14.3−13.2 ka to settlement of the entire island in as little as 200 years [36]. Rapid growth during that time also estimates that the human population could have numbered >10 000 within <400 years from initial arrival (median = 4300 after 400 years) [36]. Yet despite strong evidence that large accumulations of P. minor and P. cypriotes bones are anthropogenic in origin [40,42,49], and global evidence that the likelihood of extinction is highest in the most extreme island dwarfs and giants [50], many contend that humans played no part in their extinction [38,51–53].

In this article, we hypothesize that pre-agropastoralist human populations in Cyprus were capable of driving these megafauna species to extinction. To test this hypothesis, we first (i) re-examined the extinction chronology for P. minor and P. cypriotes, accounting for both dating uncertainty and the Signor–Lipps effect [54,55]—the low probability of archaeological or palaeontological evidence being preserved or discovered, such that first and last dates in a time series almost never indicate the true dates of initial appearance or extinction, respectively [reviewed in [56–58]). Our new Signor–Lipps-corrected windows of extinction for both species now also account for dating uncertainty. (ii) Next, we developed stochastic, cohort-based models of the population dynamics for both P. minor and P. cypriotes to estimate the offtake rates necessary to drive equilibrium populations of these two species to extinction in a Cyprus-equivalent area. Finally, (iii) we expanded the demographic models to include both hunting functions and the energetic requirements of pre-agropastoralist human populations to express offtake in terms of ‘meat equivalents’ for human consumption. This approach allowed us to estimate the size of the palaeo-Cypriot human population required to drive both species to extinction, as well as the most ecologically realistic chronologies of any ensuing extinction events. We show not only that the estimated human population sizes in Late Pleistocene Cyprus could have easily driven both species to extinction, but the predictions also match the observed chronological sequence of extinctions inferred from the palaeontological record.

2. Methods

We designed the approach to estimate the probability of extinction of P. minor and P. cypriotes as a function of varying human population sizes. In brief, §2a describes the application of a Signor–Lipps correction to the time series of available age estimates for each species to determine the likely sequence of extinction derived from the palaeontological record. Section 2b describes how we estimated the vital rates (e.g. survival, fertility, age at primiparity, longevity) for each species to build the Leslie matrix projection models described in §2c. Section 2c also describes the density-feedback catastrophe functions that modify the survival probabilities in the Leslie matrices. Section 2d describes how we used the Leslie matrices to calculate the probabilities of extinction with incrementing rates of offtake, and §2e describes how we translated human energetic needs, hunting efficiency and prey selection to estimate the probability of extinction of both megafauna species as a function of incrementing human population size.

(a). Inferring windows of extinction from the palaeontological record

To estimate a Signor–Lipps-corrected window of extinction for both P. minor and P. cypriotes, we sourced available radiocarbon-dated time series [40,52]. We applied a quality-rating protocol [59] to the radiocarbon dates using a customized R [60] function FosSahul_Rating (available at figshare.com/articles/dataset/FosSahul_2_0_database_and_R_code/8796944) [61] that automates the quality-rating protocol for radiocarbon age estimates described in Rodríguez-Rey et al. [59]. The quality-rating algorithm uses information such as the type of material dated, quality pre-assessment, pre-treatment and association to calculate a quality rating from A* (highest quality), A (high-quality), B (possibly reliable) to C (unreliable) [59]. We obtained as much of this information as possible from the source papers [40,52], and then ran the quality-rating algorithm in R. None of the dates for either species achieved a quality rating >B, so we removed all C-rated dates from the series and applied the Signor–Lipps correction to estimate a possible (although only possibly reliable) window of extinction, as we explain below.

Using the uncalibrated, quality-rated radiocarbon dates described above, we applied the calibration-resampled inverse-weighted McInerny method (CRIWM) using the Rextinct package [57] (the most up-to-date and advanced method available that incorporates dating uncertainty in its calculation of probable extinction windows) in R [60]. Rextinct first calibrates the radiocarbon dates to calendar years before present (BP) based on a user-defined curve (we used the IntCal20 calibration curve) [62], and then resamples the intervals in the time series to provide a 95% confidence interval for the estimated extinction date (script ‘dwarf hippo & elephant extinction dates.R’; doi:10.5281/zenodo.10561469 [63]).

(b). Demographic parameter estimation

To build age-structured population models for the two extinct species, we applied allometric, phylogenetic, and measured relationships to predict plausible component demographic rates. We used the estimated adult body mass of 132 kg for P. minor and 531 kg for P. cypriotes [41]. For each species, we calculated the maximum rate of instantaneous population growth (r _m), theoretical equilibrium population densities (D, km⁻²), maximum age (ω), fecundity (F), age at first breeding (α), and age-specific fertilities (m_x ) and survival probabilities (S_x ) following the equations provided in electronic supplementary material, appendix I.

(c). Leslie matrix projections for the megafauna species

From the estimated demographic rates for each species, we constructed a pre-breeding, ω+1 (i) × ω+1 (j) element (representing ages from 0 to ω years old), Leslie transition matrix (M) for females only (males are demographically irrelevant assuming equal sex ratios). Fertilities (m_x ) occupied the first row of the matrix, survival probabilities (S_x ) occupied the sub-diagonal, and we set the final diagonal transition probability (M _i,j ) to S_ω (script ‘base hippo & elephant models.R’, lines 167–183, 437–453; [63]). Multiplying M by a population vector n estimates total population size at each forecasted time step [64]. Here, we used n ₀ = AD Mw, where w = the right eigenvector of M (stable stage distribution) and A = the surface area of Cyprus at 14 ka (approximate period of arrival of humans) applied in the stochastic extinction scenario (A = 11 194 km²) [36] (script ‘base hippo & elephant models.R’, lines 186–189, 456–458; [63]).

To avoid an exponentially increasing population of each megafauna species without limit generated by a transition matrix and optimized to produce values as close to r _m as possible, we applied a theoretical compensatory density-feedback function. This procedure ensures that the long-term population dynamics were approximately stable by creating a second logistic function of the same form as m_x to calculate a modifier (S _mod) of the S_x vector according to total population size (Σn):

$${\displaystyle S_{\text{mod}}=\frac{a}{1+{\left(\frac{\mathrm{\Sigma }n}{b}\right)}^c}.}$$ (2.1)

We adjusted the a, b and c constants for each megafauna species in turn so that a stochastic projection of the population remained stable on average for 40 generations (40G), where

$${\displaystyle G=\frac{\log \left({\left({\mathsf{v}}^{\mathsf{T}}\mathsf{M}\right)}_{\text{1}}\right)}{{\lambda }_{\text{1}}},}$$ (2.2)

and (v ^T M)₁ = the dominant eigenvalue of the reproductive matrix R derived from M, and v = the left eigenvector [64] of M (script ‘base hippo & elephant models.R’, lines 217–240, 486–509; [63]). Although arbitrary, we chose a 40G projection time as a convention of population viability analysis to standardize across different life histories [65,66].

The projections were stochastic in that we β-resampled the S_x vector assuming a 5% standard deviation of each S_x and Gaussian-resampled the m_x vector at each yearly time step to 40G. We also added a catastrophic die-off function to account for the probability of catastrophic mortality events (C) scaling to generation length among vertebrates [67]:

$$C=\frac{p_c}{G}$$ (2.3)

where p_C = probability of catastrophe (set at 0.14) [67]. Once invoked at probability C, we applied a β-resampled proportion centred on 0.5 to the β-resampled S_x vector to induce a ~50% mortality event for that year [19], as we assumed that a catastrophic event is defined as ‘… any 1 yr peak-to-trough decline in estimated numbers of 50% or greater’ [67]. Finally, for each species we rejected the first G years of the projection as a burn-in to allow the initial (deterministic) stable stage distribution to stabilize to the stochastic expression of stability under compensatory density feedback [19]. We ran 10 000 stochastic iterations of each model starting with allometrically predicted stable population size divided into age classes according to the stable stage distribution. We projected all runs to 40G for both species (removing the first G values as burn-in) (script ‘base hippo & elephant models.R’; [63]).

(d). Offtake simulation

To determine relative susceptibility to offtake, first, we progressively removed individuals from the n population vector, with age-relative offtake following the stable stage distribution of the target species. We then progressively increased the offtake and calculated the proportion of 10 000 stochastic model runs where the final population size fell below a quasi-extinction (E _Q ) of 50 female individuals (100 total individuals assuming 1 : 1 sex ratios). This threshold is based on the updated minimum size below which a population cannot avoid inbreeding depression [68]. This basic scenario does not link offtake to human dietary requirements or hunting capacity, nor does it translate offtake to resident human population sizes—it only establishes a relationship between gross offtake rates (individuals removed per projection interval) and the probability of quasi-extinction (script ‘offtake hippo & elephant models.R’; [63]).

(e). Animal growth rates

For P. minor, we sourced several parameters to estimate female and male Von Bertalanffy growth functions of the form

$$M_x=M_{\text{max}}-\left(M_{\text{max}}-M_{\text{0}}\right)e^{-kx}$$ (2.4)

where M ₀ = mass at birth (kg), M _max = maximum adult body mass (kg), k = growth rate constant, and x = age in years. We used 3 ages per sex (birth, age at sexual maturity, longevity) and corresponding mass estimates to fit the Von Bertalanffy equations using the nls function in R. We estimated sex-specific age at sexual maturity for females following equation S5 (electronic supplementary material, appendix I), and adjusted for males by multiplying α _female by the ratio of mean age at sexual maturity for male (6–13 years) and female (7–15 years) extant pygmy hippopotamus (Choerpsis liberiensis) [69]. For the corresponding masses, we calculated a ratio of maximum adult weight for P. minor (132 kg) to the extant pygmy hippopotamus (C. liberiensis; 179–273 kg) [69], and then used this ratio to correct size at birth for female (4.5 kg) and male (6.2 kg) C. liberiensis [70], and estimated size at sexual maturity based on the observation that female H. amphibius reach sexual maturity at 0.78 of maximum weight [71], and males at 0.65, as well as the maximum female and male weights estimated for C. liberiensis (179 and 273 kg, respectively [69]). The fitted P. minor female and male Von Bertalanffy growth equations (electronic supplementary material, appendix II, figure S1) estimated median k = 0.3722 and 0.2972, respectively (script ‘meat equivalents hippo & elephant models.R’, lines 62−108; [63]).

For P. cypriotes, we first calculated the mean adult male and female masses for African savanna elephants (Loxodonta africana) [72], and proportioned the ratio across the mean mass for both sexes to estimate equivalent female and male maximum masses for P. cypriotes. We then substituted these masses into the sex-specific growth equations estimated for Asian elephants [73] of the form

$${\displaystyle M_x=M_{max}{\left(\text{1}-e^{-k\left(x+a\right)}\right)}^{\text{3}},}$$ (2.5)

where k = 0.092 (females) or 0.149 (males), a = 6.15 (females) or 3.16 (males), and x = age in years (electronic supplementary material appendix II, figure S1) (script ‘meat equivalents hippo & elephant models.R’, lines 124−142; [63]).

(f). Edible meat

To estimate the amount (mass) of edible meat (‘meat weight’, including bone marrow) that can be obtained from a carcass of a large herbivore, we obtained data on the edible proportions (η) of several species of large ungulate [74]. There were multiple total weights of the edible portion available for the following species: barren-ground caribou (Rangifer tarandus groenlandicus), woodland caribou (R. tarandus caribou), moose (Alces alces), and muskox (Ovibos moschatus). We then divided these weights by the mean total mass (both sexes) of each species obtained from the following sources: R. tarandus groenlandicus [75,76], R. tarandus caribou [76], A. alces [77] and O. moschatus [78]. While no edible-meat data on similar-sized elephants or hippopotamus exist, the values we obtained for other species of similar size are indicative of the approximate edible meat proportions of Phaniouros and P. cypriotes (we also test the relative importance of variation in this parameter in the global sensitivity analysis provided in electronic supplementary material, appendix III). We then bootstrapped (10 000 iterations) the mean and standard deviation of these proportions combining all species to provide a global mean proportion η = 0.314 ± 0.095 (script ‘meat equivalents hippo & elephant models.R’, lines 144–197; [63]) that we used in the stochastic hunting scenarios described below.

(g). hunter–gatherer diet

We first obtained estimates of the daily energy intake for hunter–gatherers [34] for both adult females (ξ _f = 1877 ± 364 kCal day⁻¹ = 7853 ± 1523 kJ day⁻¹) and adult males (ξ _m = 2649 ± 395 kCal day⁻¹ = 11083 ± 1653 kJ day⁻¹). Assuming the proportion of meat in the diet of hunter–gatherers [29] (ζ) was 0.65, we translated meat into energy equivalents based on the mean value for African elephants (L. africana) of 130 kCal (μ) 100 g⁻¹ meat [32,33]. With these values, we can estimate the total amount of meat consumed by an average adult female (342.6 kg) and male (483.4 kg) per year (script ‘meat equivalents hippo & elephant models.R’, lines 199−230; [63]). Compared to the highest per-capita meat-consuming countries today (e.g. USA: 101.9 kg person⁻¹ yr⁻¹; Israel: 90 kg person⁻¹ yr⁻¹; Australia: 89.6 kg person⁻¹ yr⁻¹; data for 2019) [79], the estimated values for palaeo-hunter–gatherers are just over 4 times higher.

To create a function of annual meat requirements for each human age class from 0 to longevity, we obtained data on age-specific protein requirements for people [31] (50 kg adult: 40 g protein day⁻¹, 14–18-year-olds: 37 g day⁻¹, 9–13-year-olds: 24 g day⁻¹, 4–8- year-olds: 13.5 g day⁻¹, 1–3-year-olds: 9.2 g day⁻¹) that we first transformed to proportions of the adult requirement (I_p ) to which we fitted a logistic power function of the form

$${\displaystyle I_p=a\frac{b}{\left(a-b\right)e^{-cx}+b},}$$ (2.6)

to estimate age- (year-) specific proportions and correct for the age class consuming meat (a = 1.1381, b = 0.1393, c = 0.1983, and x = age in years) (electronic supplementary material, appendix II, figure S2) (script ‘meat equivalents hippo & elephant models.R’, lines 233−270; [63]).

(i). Prey choice

Different species provide different returns to human hunters based on components such as capture probability, animal body size, edible portion, and handling costs (e.g. pursuit time, butchering time, and preparation time) [32]. We therefore applied the equations of Yaworsky et al. [13] using the adult mass estimates of both species to estimate the mean and standard deviation of post-encounter return rate (π; cal h⁻¹) (defined as the energy provided divided by handling costs [13]):

$${\pi }_s=\text{60}\frac{e_s\left(1-p_s\right)}{c_s+\left(1-p_s\right)h_s}$$ (2.7)

where e = energetic payoff (cal), p = probability of acquisition failure, c = pre-acquisition handling time (min), h = post-acquisition handling time (min), s subscript indicates value for species s, and the multiplier 60 converts to energy h⁻¹. Using the coefficients and their standard errors that estimate the parameters in equation (2.7), we developed a resampling approach where we produced 100 000 samples of π for P. minor (π _Pm) and P. cypriotes (π _Pc), and then calculated the number of times where π _Pm > π _Pc. The higher average relative return rate of P. minor compared to P. cypriotes results from the increased handling costs of larger species due to higher probabilities of failed pursuit [13]. This sum divided by the total number of iterations (100 000) gives the relative probability of targeting P. minor over P. cypriotes (ψ = 0.773). Therefore, we assumed that ψ represented the relative likelihood of selecting P. minor versus P. cypriotes that we resampled stochastically (see below) following a β distribution with mean = ψ and an arbitrary standard deviation = σ _ψ =pψ(ψ) = 0.05ψ (but see global sensitivity analysis in electronic supplementary material, appendix III regarding the choice of pψ = 0.05) (script ‘meat equivalents hippo & elephant models.R’, lines 283–391; [63]).

(h). Hunting simulation

We developed a stochastic simulation similar to the offtake simulation described above, but instead of sequentially reducing the n vectors for each species separately, we incremented the number of humans on the island of Cyprus and converted this number into megafauna meat equivalents to sustain the human population. This approach not only required translating animals culled into protein energy required by humans of different ages and sexes, we also incorporated a prey-selection function (described in the previous section) as well as a density feedback on the meat portion of the human diet fulfilled by megafauna sources.

Assuming that 0.65 of the age-specific human energy requirements were provided by meat (ζ) on average [29] (but see global sensitivity analysis in electronic supplementary material, appendix III, and discussion of non-megafauna food resources in electronic supplementary material, appendix IV regarding this value), we developed an arbitrary feedback function whereby the proportion of ‘other’ meat sources (e.g. marine fish and shellfish, marine chelonians, birds, and small terrestrial animals; see also electronic supplementary material, appendix IV) [31] increased from 0.33 at the time of initial human arrival (assuming the populations of P. minor and P. cypriotes were at their maximum equilibrium sizes), approaching 1.0 sigmoidally as the megafauna populations approached extinction. This function accounts for decreasing prey encounter rates by human hunters as the prey populations dwindle, such that 0.67 of the meat requirements are provided by megafauna sources at maximum megafauna densities, and approach 0 as those populations approach extinction. In other words, when humans first arrive in Cyprus, the total proportion of the diet attributed to megafauna is 0.67ζ = 0.4355, which falls towards 0 as the megafauna populations approach extinction.

Starting with an initial human population size of 1000 (i.e. 500 females)—a large-enough predator population to elicit some extinctions during the iterative process—we applied the age-specific energy requirements to the stochastic age structure generated by the model at each time step. Next, we β-resampled the probability of successfully acquiring P. minor relative to P. cypriotes, and then calculated the number of individuals across their age distribution required to fulfil this human meat requirement (i.e. using the female- and male-specific growth curves for both species). We assumed that humans did not select male or female prey preferentially (relative proportion female versus male prey taken φ = 0.5, but see global sensitivity analysis in electronic supplementary material, appendix III). We then removed these meat equivalents in terms of individuals culled from the megafauna n vectors at each time step, projecting those populations through to 80 generations (80G: to allow a sufficiently large human population enough time to drive a megafauna population to extinction) in each iteration. In cases when the P. minor population declined enough to where it could no longer supply sufficient meat as the prey with the highest energy return rate (even after accounting for the change in ‘other’ meat categories described above), we transferred that meat requirement to P. cypriotes by removing the equivalent number of P. cypriotes individuals to account for the missing meat requirement.

We then incremented the number of humans on the island and recorded the probability of quasi-extinction for each megafauna species, as well as the median time (years) required to drive each population to extinction. For each human population size increment we tested, we assumed that human population remained stable during a 80 prey-generation projection interval, so our extinction predictions are necessarily conservative (script ‘meat equivalents hippo & elephant models.R’, lines 393−615; [63]). We ran all code on the Flinders University High-Performance Computing facility DeepThought [80], and all code and data required to repeat the analyses are available at [63].

3. Results

(a). Estimated extinction windows

After quality rating, there were 5 dates for P. minor from the original 13 provided by Zazzo et al. [52] that had a B rating, and 14 B-rated dates for P. cypriotes from the original 30 provided by Wigand & Simmons [81]. The resultant windows of extinction estimated using the CRIWM unbiased algorithm on these B-rated dates were 11 995–11 092 BP for P. minor, and 10 347–9073 BP for P. cypriotes. Compared to the arrival window of 14 257−13 182 estimated for pre-agropastoralist humans in Cyprus [36], the Signor–Lipps-corrected megafauna extinction windows suggest that P. minor went extinct 1187−3165 years after human arrival, and P. cypriotes went extinct 2835−5184 after human arrival (figure 1), although with the caveat that the extinction windows are not based on the highest-quality radiocarbon age estimates.

Figure 1.

Human arrival window (grey vertical bar; estimated [36] using the calibration-resampled inverse-weighted McInerny method, CRIWM algorithm [57]) and the CRIWM-estimated windows of extinction (red vertical bars) for Phaniouros minor and Palaeoloxodon cypriotes. Also shown are the hindcasted temperature anomalies (°C, relative to the present) for Cyprus derived from the HadCM3 [82] and TraCE21ka [83,84] global circulation models. Also shown are major climatic periods: Last Glacial Maximum (LGM), Bølling–Allerød interstadial (B-O), Younger Dryas (YD), and the early to mid-Holocene.

(b). Demographic estimates

For P. minor and P. cypriotes, respectively, the allometric equations predicted age at first breeding (α) = 4 and 5 years, maximum longevity (ω) = 36 and 43 years, equilibrium density = 1.28 and 0.46 individuals km⁻² (corresponding to populations of 14 280 and 5098 individuals assuming a land area [36] of Cyprus at 14 ka = 11 194 km²), and maximum instantaneous rate of exponential increase (r _m) = 0.22 and 0.15. Parameter values and stable age distributions for both deterministic Leslie matrices are provided in electronic supplementary material, appendix II and figure S3.

(c). Offtake simulation

The offtake simulations demonstrated that P. cypriotes was more susceptible to extinction than P. minor (figure 2), which is expected given the slower life history of the former. Once the annual offtake of P. cypriotes began to exceed 200 individuals, the probability of quasi-extinction climbed precipitously, becoming close to 1.0 at an average annual offtake of ~350 individuals (figure 2). The extinction probability of the smaller P. minor only began to increase after an average annual offtake of ~650 individuals, reaching near certainty at ~1000 animals removed annually (figure 2).

Figure 2.

Probability of quasi-extinction (E _Q ) of dwarf hippopotamus (Phanourios minor) and dwarf elephant (Palaeoloxodon cypriotes) as a function of the number of individuals removed per year (following the stable stage distribution).

(d). Hunting simulation

However, these relative susceptibilities reverse when we consider the second set of simulations estimating offtake as a function of human dietary requirements and prey choice. Because of the higher relative return rate of P. cypriotes, as well as their ~4-fold greater mass compared to P. minor, the elephant’s extinction probability was lower than for the hippopotamus across the range of human population sizes eliciting some extinction risk (figure 3). Here, P. minor extinction risk began to increase once the human population on the island exceeded 3000 and was near certain at human population sizes of ~4500 (figure 3). The extinction risk of P. cypriotes similarly began to rise at human population sizes >3000, but increased more slowly than for P. minor, eventually achieving near-certain extinction risk at an island-wide human population of around 7500 (figure 3). The time to drive the population of P. minor to extinction correspondingly declined from around 800 years at a human population of ~3700, to <100 years at a human population of 4500 (figure 3). Meanwhile, the time to drive the population of P. cypriotes to extinction declined from around 800 years at a human population of ~6300, to <100 years at a human population of just over 7000 (figure 3).

Figure 3.

Probability of quasi-extinction (E _Q ) of dwarf hippopotamus (Phanourios minor) and dwarf elephant (Palaeoloxodon cypriotes) as a function of the number of people living in Cyprus (left axis; red), and median number of years to extinction for each species (right axis; blue).

4. Discussion

Conclusions drawn about the role of palaeo-human exploitation on the extinction of megafauna species are too often predicated on an uncritical comparison of uncertain chronologies and do not typically examine the ecological plausibility of extinctions based on predator-prey dynamics or human energetic requirements [53]. Such weak inference limits our ability to determine the extent to which Palaeolithic peoples were able to drive megafauna extinctions. Using detailed reconstructions of energetic demand, diet composition, prey selection, and hunting efficiency, our conservative stochastic model not only demonstrates that 3000–7000 pre-agropastoralist humans in Cyprus (human population sizes well within predicted estimates for the island from 13 ka onward) [36] could have driven both dwarf hippopotamus and dwarf elephants to extinction within < 1000 years, the predicted chronology of extinctions (first P. minor, then P. cypriotes) matches the Signor–Lipps-corrected extinction sequence derived from independent palaeontological evidence. Given the island’s small size and relatively simple megafauna assemblage (two species), we have provided some of the first evidence for the mechanisms allowing Palaeolithic societies to drive large megafauna species to extinction soon after first contact. Our results therefore provide strong support for the hypothesis that Palaeolithic peoples were at least partially responsible for megafauna extinctions globally during the Late Pleistocene and early Holocene [1,4,5,8,10–26,85].

Elsewhere, the zooarchaeological record of large-mammal remains during the Late Upper Palaeolithic (14.0–12.6 ka) is influenced strongly by cost-benefit regimes arising from human decision-making, meaning that the abundance of zooarchaeological remains does not necessarily reflect animal densities in the landscape [13]. As such, the higher relative return rate of hippopotamus compared to elephants (expressed as ψ in the model) was an important determinant of the relative extinction chronology of the two species. However, ψ was a weak driver of variation in extinction risk predicted by our model (electronic supplementary material, figure S4). Instead, the most important determinant of extinction risk for both species was the proportion of edible meat that could be derived from a single carcass (η). While we determined η from measured edible proportions of several Arctic species, it is possible that pre-agropastoralists in Cyprus were able to obtain higher portions, thereby reducing the number of individual hippopotamus or elephants killed to supply their human energetic requirements.

We also incorporated a function that increased the proportion of ‘other’ meat sources in the diet as megafauna were depleted, based on empirical data that hunter–gatherers of the Late Pleistocene pursued and exploited a broad range of prey [33]. Here, the function modifies ζ (proportion of meat in the human diet) by shifting the source of prey items away from the two large megafauna species as they become rarer in the environment. In other words, 0.67 of ζ = 0.65 (i.e. 0.67 × 0.65 = 0.4355) is attributable to megafauna consumption at maximum megafauna densities (i.e. when people first arrive), henceforth declining to 0 as megafauna populations are driven to extinction (i.e. all meat is supplied by the ‘other’ category). But this function was partially arbitrary because we do not know the shape of the relationship between megafauna abundance and reliance on other meat sources. However, the assumption that Palaeolithic peoples would have preferentially selected megafauna over other meat sources until rarity of the former forced them to rely more on the latter is supported by other studies demonstrating the relative benefits of selecting large game in many hunter–gatherer societies [13,32,33]. Nonetheless, we also tested the effect of halving ζ on the probability of quasi-extinction of both megafauna species (i.e. ζ ′ = 0.325). With all other parameters maintained as in the original model, reducing ζ to 0.325 effectively doubles the size of the human population required to cause a high probability of extinction (E_Q>0.5) (e.g. from ~4000 to 7800 people for P. minor, and from~6500 to 13 400 people for P. cypriotes; electronic supplementary material, figure S6). Given that the human population in Cyprus could have grown to as many as 10 000 within 320 years following initial entry between 14.257 and 13.182 ka, and to >15 000 people by around 11 ka [36], the plausibility of human-caused extinctions of both species is still maintained even using the extreme ζ ′ = 0.325 value (see also electronic supplementary material, appendix IV).

While the initial (pre-human) population sizes of both species of course influence extinction risk, there was only a modest influence of initial population size of hippopotamus on that species’s extinction rate. The lack of a strong influence of the initial population size of elephants on that species’s extinction rate is partially a function of the two-prey model favouring hippopotamus as human prey over elephants. Furthermore, while our estimates of initial population size were derived from ecological theory, population densities would have varied spatially according to habitat diversity and island topography, including the presence of a large mountain range (Troodos Mountains) in the western region of Cyprus. While fossil sites for P. minor and P. cypriotes remains span most of the island (electronic supplementary material, appendix V), there are still large spatial gaps within the Troodos Mountains region and the far northeast (electronic supplementary material, appendix V).

We did not incorporate any theoretical prediction of how environmental change might have exacerbated the extinction risk our model predicted. Climate hindcasts from the HadCM3 atmosphere-ocean general circulation model [82] for Cyprus during the period from 14 ka to 10 ka predicted a mean temperature rise of ~1.5℃ (figure 1) and a 44 mm increase in annual precipitation. However, the temporally finer-resolution (seamless) TraCE21ka climate model [83,84] predicts a more dynamic climate during this same period, with a ~1 ℃ rise during the latter half of the Bølling–Allerød Interstadial (14.6–12.9 ka), followed by a ~1.4℃ decline during the Younger Dryas, and then a ~2.0℃ rise by 10 ka (figure 1). With increasing evidence for extinction synergies [86] between human over-exploitation and environmental change in the demise of late Quaternary megafauna extinctions [2–8], such simultaneous temperature and precipitation fluctuations could have exacerbated the extinction risk of both dwarf hippopotamus and elephants in Cyprus. Indeed, there is evidence for human- and climate-mediated collapse of ecological networks in ancient Egypt [87], and Saltré et al. [8] concluded that combinations of aridification and human presence contributed to the local extinction of many megafauna species in Sahul. Our predictions of extinction risk arising solely from human over-exploitation should therefore be considered conservative.

In addition to the feasibility of pre-agropastoralist humans driving both megafauna species in Cyprus to extinction demonstrated by our conservative models, we also argue that it was an appealing destination for early Palaeolithic explorers. The notion that Cyprus was an ‘impoverished’ landscape [38–40,88–90] is not supported either by climate models hindcasting net primary production [36] or from archaeobotanical records [91,92]. Indeed, evidence from pollen analysis of the early Holocene suggests that Cyprus was covered by dense forests of typical Mediterranean trees and shrubs (e.g. carob Ceratonia siliqua, cypress Cupressus spp., juniper Juniperus spp., kermes oak Quercus coccifera, Aleppo oak Q. infectoria, bay laurel Laurus nobilis, olive Olea europaea, and oriental plane Platanus orientalis) [91,92]. Eratosthenes reported in the third century BCE that the island was ‘thickly overgrown with forests’ [93], even in the arid central plain of Mesaoria [94]. In the Classical period, Cyprus was referred to as a ‘green island’, exporting timber and specializing in ship building [95]. Such a diverse, prey-filled landscape would therefore have been a highly sought destination once discovered by Palaeolithic peoples [36].

Ethics

This is a modelling study; there are no special ethical requirements.

Data accessibility

All R [61] code and data required to repeat the analyses are available at [63].

Supplementary material is available online [96].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

C.J.A.B.: conceptualization, data curation, formal analysis, investigation, methodology, project administration, resources, software, validation, visualization, writing—original draft, writing—review and editing; F.S.: data curation, formal analysis, investigation, methodology, resources, writing—original draft, writing—review and editing; S.A.C.: resources, writing—original draft, writing—review and editing; C.R.: conceptualization, data curation, formal analysis, investigation, methodology, resources, writing—original draft, writing—review and editing; T.M.: conceptualization, data curation, funding acquisition, investigation, project administration, resources, supervision, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This work was co-financed by the European Regional Development Fund and the Republic of Cyprus through the Research and Innovation Foundation (EXCELLENCE/0421/0050) for the project Modelling Demography and Adaptation in the Initial Peopling of the Eastern Mediterranean Islandscape (MIGRATE, 2022–2024). S.A.C. supported by the Centre for Ecological Dynamics in a Novel Biosphere (ECONOVO) funded by a Danish National Research Foundation grant (DNRF173). Funding provided to C.J.A.B. and F.S. by the Australian Research Council Centre of Excellence for Australian Biodiversity and Heritage (CE170100015).