sign in sign up
<<
Home , Local extinction, Quaternary extinction

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/295919520

A quantitative model for distinguishing between climate change, impact, and their synergistic interactions as drivers of the late megafaunal

Article · October 2015

READS 143

4 authors, including:

Emily Lindsey Natalia A. Villavicencio University of California, Berkeley University of California, Berkeley

15 PUBLICATIONS 735 CITATIONS 5 PUBLICATIONS 13 CITATIONS

SEE PROFILE SEE PROFILE

Anthony D Barnosky University of California, Berkeley

64 PUBLICATIONS 3,182 CITATIONS

SEE PROFILE

Available from: Emily Lindsey Retrieved on: 02 August 2016 A QUANTITATIVE MODEL FOR DISTINGUISHING BETWEEN CLIMATE CHANGE, HUMAN IMPACT, AND THEIR SYNERGISTIC INTERACTION AS DRIVERS OF THE LATE QUATERNARY MEGAFAUNAL EXTINCTIONS

CHARLES R. MARSHALL,1 EMILY L. LINDSEY,1 NATALIA A. VILLAVICENCIO,1 AND ANTHONY D. BARNOSKY2

1Department of Integrative Biology and Museum of , University of California, Berkeley, Berkeley, CA 94720 USA

2Department of Integrative Biology, Museum of Paleontology, and Museum of Vertebrate Zoology, University of California, Berkeley, Berkeley, CA 94720 USA

ABSTRACT.—A simple quantitative approach is presented for determining the relative importance of climate change and human impact in driving late Quaternary megafaunal extinctions. This method is designed to determine whether climate change or human impact alone can account for these extinctions, or whether both were important, acting independently (additively) and/or synergistically (multiplicatively). This approach is applied to the megafaunal in the Última Esperanza region of southern Chile. In this region, there is a complex pattern of extinction. Records of environmental change include temperature proxies and records that capture the transition from cold grasslands to warmer, moister , as well as evidence of initial human arrival. Uncertainty in extinction times and time of human arrival complicates the analysis, as does uncertainty about the size of local human populations, and the nature, strength, and persistence of their impacts through the late and early . Results of the Última Esperanza analysis were equivocal, with evidence for climate- and human-driven extinction, with each operating alone or additively. The results depend on the exact timing of extinctions and human arrival, and assumptions about the kinds of pressures put on the . There was little evidence for positive synergistic effects, while the unexpected possibility of negative synergistic interactions arose in some scenarios. Application of this quantitative approach highlights the need for higher precision dating of the extinctions and human arrival, and provides a platform for sharpening our understanding of these megafaunal extinctions.

INTRODUCTION Alroy, 2001; Koch and Barnosky, 2006; Turvey, 2009), while in other areas, it appears that climate The cause(s) of the late Quaternary megafaunal change likely triggered extinction of some taxa extinctions continue to generate debate about the (Barnosky, 1985, 1986; Stuart et al., 2002, 2004; relative roles of humans (Martin, 1966, 1967, Guthrie, 2003, 2006). The case has also been 1973, 1984, 1990; Alroy, 1999, 2001; Flannery made for a “one-two punch” of human impact and and Roberts, 1999; Roberts et al., 2001), climate climate change exacerbating extinction intensity (Graham and Lundelius, 1984; Grayson, 1984, (Barnosky et al., 2004; Koch and Barnosky, 2006; 2001; Stuart, 1999; Stuart et al., 2002, 2004; Barnosky and Lindsey, 2010; Brook and Grayson and Meltzer, 2003; Trueman et al., 2005; Barnosky, 2012; Villavicencio et al., 2015). Wroe et al., 2006, 2013), and possible synergy However, few studies have attempted to between the two (Guilday, 1967, 1984; Barnosky quantitatively assess the relative importance of et al., 2004; Guthrie, 2006; Koch and Barnosky, humans, climate change, and/or interactions 2006; Barnosky and Lindsey, 2010; Brook and between the two in determining extinction Barnosky, 2012) in triggering extinctions. There is intensity. Moreover, when interaction effects are strong evidence that humans played a key role in postulated, there is seldom an effort to distinguish megafaunal extinctions in many geographic areas between additive effects—that is, climate change (MacPhee, 1999; Martin and Steadman, 1999; accounting for the extinction of some taxa and In: Earth-Life Transitions: Paleobiology in the Context of Earth System Evolution. The Paleontological Society Papers, Volume 21, P. David Polly, Jason J. Head, and David L. Fox (eds.). The Paleontological Society Short Course, October 31, 2015. Copyright © 2015 The Paleontological Society. MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION humans accounting for others—versus true flourishes under warmer conditions)? How did synergy, where interaction between human impact vegetation change affect abundances, and climate change multiplies the extinction species interactions, and the ability of humans to intensity beyond the simple additive effects of the interact with them? To what extent did two factors acting independently. megafaunal change feed back onto vegetation Here, we present a new quantitative approach change, via the process of (Gill et al., designed to determine additive versus synergistic 2009, 2012; Galetti and Dirzo, 2013; Young et al., roles of human impact and climate change in 2013, 2014; Dirzo et al., 2014; Gill, 2014; driving extinction. The method ascribes Barnosky et al., 2015)? How did climate change, probabilities to the independent roles of human human activity, and megafaunal extinctions affect impact, non-human environmental change, and local fire regimes, and vice versa (Bond, 2005; interactions between the two in explaining the Bond and Keeley, 2005)? How did human- and observed temporal pattern of in a climate-driven changes elsewhere in South specified geographic region. The method is America impact the Última Esperanza region? described and applied to the Última Esperanza Beyond these fundamental considerations, region (Villavicencio et al., 2015), located in the shortcomings in the fossil record need to be taken Patagonia region of southern Chile. into account. For example, given the incompleteness of the record, when did the WHAT SIMPLE QUANTITATIVE MODELS various taxa actually become locally extinct? CANNOT CAPTURE When did humans first arrive? Moreover, to what extent might changes in the behaviors of taxa give While the fossil and paleoenvironmental records the false appearance of extinction? For example, of Última Esperanza are relatively rich (see in the study area, does the early disappearance of below), developing a comprehensive model of the fossil big cats indicate their extinction from the factors responsible for the extinction of the area, or simply that they moved elsewhere when megafauna in this, or any other region, is almost humans began to utilize caves that big cats had impossible. Such a model would require detailed formerly found hospitable, and that were a understanding of how the number of humans favorable depositional environment for present changed over time, whether they were fossilization? permanent residents in the region, what their preferences and practices were, how the A SIMPLE QUANTITATIVE METHOD FOR various taxa responded to their presence (e.g., ASSESSING CAUSAL AGENTS OF whether after first contact some species learned to EXTINCTION AND THEIR POTENTIAL be wary of humans), what the interactions INTERACTION between the various taxa were and how these might have changed after human contact, among Given the lack of information needed to develop a many other questions. full mechanistic model of the megafaunal Quantifying non-human effects is equally extinctions, we present a simple correlative difficult. For example, in the focal region of approach that can distinguish between human southwestern Patagonia, while climate warming is impact, environmental change, and any thought to be the primary driver of environmental interactions between the two, in accounting for change as the region came out of the last , the late Quaternary extinctions. This new method how did changes in climate actually translate into is designed to make use of available fossil and biotic transformations that might lead to environmental data, and accommodates the extinction? To what extent was the observed incompleteness of the fossil record. It requires temperature increase significant? How did rising input of three basic kinds of data: 1) the extinction sea level from the melting ice impact the region times of the megafauna, derived ideally from (particularly in the study area of southern multiple radiocarbon dates on multiple specimens; Patagonia)? Did rainfall change in any significant 2) a quantitative time series that records way, either seasonally or annually? Are indirect environmental change, such as pollen percentages effects of climate change more important than of environmentally sensitive plants, or direct changes in temperature or precipitation temperature proxies, such as those that come from (e.g., by triggering the replacement of cold- stable isotopes; and 3) some measure of probable adapted grasslands by Nothofagus , which human impact; for example, numbers of

2 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21 archaeological sites distributed through time, or while many people think of quantitative methods changes in cultural toolkits or numbers and kinds in terms of hypothesis testing, they can also lead of artifacts that may indicate changes in to hypothesis generation. In the context of the population density or hunting efficiency. In modern crisis, negative synergy is the practice, the human data is the most difficult to equivalent of asking if human impacts (e.g., assemble and interpret. Thus, a range of human- conscious conservation practices, such as moving impact values is used to evaluate the sensitivity of species out of a region where they are threatened the method to differing intensities of human by development) combined with climate change pressures. (e.g., making a formerly inhospitable region able The model itself simply tests for correlations to support the ) could help in time between megafaunal extinction (E), non- inhibit extinction. human environmental change, which for Because this method is based on the search convenience is referred to as climate change (ΔC), for temporal correlations between the three key and human impact (ΔH). Extinction = Climate variables (human impact, climate change and change + Human impact + Synergy between extinction), it requires data from multiple time climate change and human impacts, or: intervals. However, the method will not work if everything changed simultaneously, in the same E = a ΔC + b ΔH + c ΔC ΔH Eq. (1) way that simple linear regression is unable to find the slope of a line if there is only one data point. where the parameter a is a measure of the strength Thus, one of the reasons the Última Esperanza of the contribution of climate change to region of is used as a case study is extinction, b the strength of the human impacts, because the temporal pattern of climate change, and c the strength of interaction between climate human arrival, and megafaunal extinction is change and human impacts. If a parameter value complex, which provides the necessary data. The is indistinguishable from zero, there is no heterogeneous patterns of extinction and human evidence that the variable (climate change, human and environmental change in southern South impact, or synergy between the two) played a America (Villavicencio et al., 2015) offer the significant role in the observed extinctions. possibility of teasing out the relative importance of these drivers and their interaction in explaining Graphical interpretation of Eq. (1) the megafaunal extinctions. Figure 1 shows graphical depictions of the ideas captured by Eq. (1): A, B show the assumed linear Learning to ‘converse’ with quantitative relationship between extinction intensity and methods human impact or climate change acting alone; C In the context of this Short Course, we emphasize shows the effect if both human impact and climate the importance of the appropriate use of change play a role in extinction, but without quantitative tools. These tools should be used in synergy (that is, there is simply an additive the same way one might use a geological effect); D, E show the combinations of synergistic hammer: understand your goals (e.g., extracting effects and one or the other of human impact or fossils in the case of a hammer), and remain in climate change; F shows purely synergistic control of the tool at all times. Furthermore, to interaction between human and climate impacts, gain maximum insight, one must be aware of i.e., climate or humans alone are insufficient to what a tool can and cannot do. In the case of cause extinction, but together, they initiate the geological hammers, they are good for some crossing of an ; G depicts the tasks, such as breaking off rock samples, poor at relationship if there is a negative synergistic others (e.g., extracting fragile bones from effect, i.e., if the presence of humans and climate ), and can be used in different ways change somehow combine to inhibit extinction depending on the nature of fossils and the rock/ intensity; and H shows the predicted extinction sediment they are in. In the case of the tool intensity if there are both additive and synergistic developed here, the need to establish a effects of humans and climate change in driving ‘conversation’ between the tool and the data is extinction. While we have never seen this important, and is developed through sensitivity proposed before, Eq. (1) is general enough that it analyses. In our analysis of the southern South allows for the possibility of negative synergy, and American megafaunal extinction, this leads to illustrates an interesting aspect of model building: some unexpected new ideas, and helps establish

3 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION

FIGURE 1.—Graphical expression of Eq. (1), used to establish the relative importance of human impact and climate change on extinction intensity, including the nature of the interaction, if any, between them. The parameters a and b are given values of 1, while c is given a value of 2. Both human impacts and climate change have been normalized so that the maximum impact or change has a value of 1. Thus, the maximum possible extinction intensity is 4 (i.e., E = a ΔC + b ΔH + c ΔC ΔH = 4 when a = 1, b = 1, c = 2, ΔCmax = 1, and ΔHmax = 1: see panel H). See text for discussion. what additional data are needed to more fully surfaces depicted in Fig. 1 best fits the data. The understand the extinctions. program is written in R and provided in the Appendix. Values for the three different kinds of Using the method data required (the temporal records of extinction, To illustrate the use of Eq. (1), three hypothetical climate change and human impacts) were input, data sets were analyzed using non-linear least then non-linear least regression used to calculate squares to determine whether the values of the the correlation between the temporal changes in parameters a, b, and c in Eq. (1) were the three variables, and finding the best-fit values significantly different from zero. In effect, non- for a, b, and c, giving the probability that the linear least squares determines which of the values are significantly different from zero.

4 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21

FIGURE 2.—Hypothetical scenario where climate and human impacts are both important, but where there is no evidence of synergistic effects. Stratigraphic ranges and extinction times (E) for six taxa are shown for 10 time bins at a temporal resolution of 1,000 , roughly that seen in the Última Esperanza data. Non-human environmental change (ΔC) is captured by change in the proportion of forest cover (see text for further discussion). Human impact (ΔH) is assumed to be in proportion to the growth in population size. It is assumed here that once human population reached its stable size (in time bin 9), their continued presence no longer constituted an extinction threat for the remaining taxa (the ‘initial impact only’ scenario described in the text) – thus in time bin 10, ΔH is scored “0”. Positive parameter values indicate a positive Hypothetical Case 2: Synergistic effects only. relationship between extinction and the driver of —Figure 3 shows a second hypothetical example extinction, while negative parameter values where human impact and forest change work in indicate a negative relationship (see Fig. 1G). concert to cause extinction. When there is change Hypothetical Case 1: Climate and human in forest cover alone (time bin 5), or increase in impacts both important with no synergistic effects. human impacts alone (time bin 8), there is no —Figure 2 shows how data can be extracted from extinction. However, when both change at the stratigraphic ranges and environmental proxy same time, there are several extinctions (time bins information, and how these data are converted 6, 7). Also, evidence of synergistic interactions is into a format that can be used by Eq. (1). In this supported by the fact that there are an equal case, extinctions (E) are concentrated in two number of extinctions in time bins 6 and 7, but in pulses—one when there is a transition from one bin, climate change is bigger, while in the grassland to forest (Fig. 2; ΔC column, mainly in other, human change is bigger. time bins 4 and 5), and one with human arrival Analysis of the data using Eq. (1) confirms and population growth (Fig. 2, ΔH column, time that there is no significant support for either bins 8, 9). climate change (p = 0.610, a = –1.01 + 1.89) or Application of Eq. (1) yields the result human arrival and growth (p = 0.952, b = 0.06 + expected through visual inspection of the data: 0.93) playing a role in the extinctions on their significant support for climate change (p = 0.021, own, but there is significant support for a positive a = 1.97 + a standard error of 0.66) and human synergistic effect between climate change and impact (p = 0.003, b = 2.99 + 0.65) on extinction, human arrival and growth (p = 0.048, c = 21.9 + while there is no support for a synergistic effect 9.0). between climate change and human impact (p = Hypothetical Case 3: Sensitivity analysis and 0.869, c = –5.1 + 29.5), i.e., the parameter c is not an unexpected discovery.—As part of significantly different from zero. understanding how Eq. (1) ‘converses’ with the

5 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION

FIGURE 3.—Hypothetical scenario where climate and human impacts are both individually unimportant, but where there is evidence of synergistic effects (see Figs. 1, 2 for further explanation). data, three experiments were run, each based on effect is as follows: First, in time bin 7, the human minor changes to Case 2 above (Table 1). impact is twice that indicated for time bin 6, yet In the first experiment (column E[1]; Table 1), there is the same number of extinctions. Second, most extinction occurs when there is both there is also climate change in bins 6 and 7, and it substantial climate change and human impact, is also higher in bin 7. Thus, the analysis with two extinctions in time bins 6 and 7. Thus, identifies negative synergy between the higher we expected to see a human effect, a climate human impact and higher climate change in bin 7 effect, and possibly a synergistic effect. However, to explain why the number of extinctions is the analysis with Eq. (1) showed that while there is a same in bins 6 and 7 despite the higher human strong human effect (p = 0.005, b = 13.57 + 3.31), impacts in bin 7. there is no evidence of a climate change effect (p To test this conclusion, the number of = 1.000, a = 0.000 + 3.78). More surprisingly, extinctions in time bin 7 was increased from two while there is evidence of synergistic effect (p = to five (Table 1, column E[2]), so the extinction 0.05, c = –14.29 + 6.04), the effect is negative. severity is roughly proportional to the human Careful scrutiny of the data reveals why Eq. impact. As expected, Eq. (1) still indicates that (1) leads to these counterintuitive results. To human impact is important (p = 0.042, b = 8.214 understand why there is no support for the impact + 3.31), but this alone is now sufficient to explain of climate change, the two key observations are: the patterns of change (Table 1). There is still no 1) when there is human impact, there is always support for a climate change effect (p = 1.000, a = extinction (time bins 6 and 7); and, 2) while there 0.000 + 3.78,), and there is now no need to invoke is also climate change in those two time bins, a negative synergistic human–climate effect (p = there is also climate change in time bin 5, but with 0.275, c = 7.14 + 6.04). no extinction. Thus, the model recognizes that If this logic is correct, increasing the number human change can explain all the extinctions, and, of extinctions in time bin 7 should support given this, dismisses climate because there is no positive synergistic effects between human and extinction in time bin 5 when there is climate climate change. As a test, increasing the number change. of extinctions in time bin 7 to eight (extinction Given the identification of human impacts as scenario E[3], Table 1) leads to a significantly the primary driver of extinction, the reason positive synergistic effect (p = 0.002, c = 28.57 + analysis of E(1) supports a negative synergistic 6.04). Synergistic interaction alone is enough to

6 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21

TABLE 1.—Hypothetical data used to explore properties of Equation (1). Time Years BP ΔC ΔH E(1) E(2) E(3) Bin 10 8,000–7,000 0 0 0 0 0 9 9,000–8,000 0 0 0 0 0 8 10,000–9,000 0 0 0 0 0 7 11,000–10,000 0.6 0.4 2 5 8 6 12,000–11,000 0.25 0.2 2 2 2 5 13,000–12,000 0.1 0 0 0 0 4 14,000–13,000 0 0 0 0 0 3 15,000–14,000 0 0 1 1 1 2 16,000–15,000 0 0 0 0 0 1 17,000–16,000 0 0 0 0 0 Significant climate effect? –– –– –– Significant human effect? Positive Positive –– Significant synergistic effect? Negative –– Positive explain the extinction pattern, so there is no few percent, with only limited data (only a few support for human (p = 0.417, b = 2.86 + 3.31) or tosses), the statistical power to reject the null climate change being important (p = 1.000, c = hypothesis that the coin was fair cannot be 0.000 + 3.78). generated because it takes many coin tosses to Take-home lessons from the hypothetical case reveal subtle bias. studies.—1) The method is sensitive to small The method presented here tries to determine changes to the data, especially when the total which of the basic relationships depicted in Fig. 1 amount of data is small; 2) Marked changes in best explain the available data. With limited data, outcome resulting from slight changes to the data as in the southern Chilean dataset analyzed below, emphasize the important of performing sensitivity there may be insufficient data to correctly identify analyses; 3) The third hypothetical data set the factors that might correlate with the developed to further understand how Eq. (1) extinctions. Thus, non-significant results might works showed significant support for negative simply indicate a lack of power. synergistic interaction between humans and climate change. Initially, the result of negative ÚLTIMA ESPERANZA MEGAFAUNAL synergy had us worried that our approach was EXTINCTION DATA fundamentally flawed; however, via sensitivity analyses E(2) and E(3) the surprising result was The relevant empirical data are shown in Figure 4. explained. While the full data are discussed in Villavicencio When quantitative analysis yields results that et al. (2015), the demands of Eq. (1) require some are counterintuitive, it might be because the explanation. quantitative tool is, in fact, inappropriate for the problem at hand, or, there may be something Measuring extinction about the tool that was not previously appreciated. Extinction was measured by counting the absolute The only way to find out is to ‘converse’ with the number of taxa inferred to have disappeared in selected method. each time interval. There are other ways of A question of power.—In statistical analysis, measuring extinction: for example, using ‘power’ refers to the ability to distinguish between proportional extinction rates in each time interval, alternative hypotheses, specifically, the ability to i.e., expressing extinction as proportion of reject the null hypothesis when an alternative is megafauna species extinct relative to those that true, and there is a close relationship between were present at the time. This would effectively power and the amount of data used. For example, up-weight later-occurring extinctions, given that if a tossed coin were biased towards heads by a each successive extinction represents a loss of an

7 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION

FIGURE 4.—Fossil and environmental data for the Última Esperanza region of southern Chile (Villavicencio et al., 2015), and the way these data have been coded for analysis with equation (1). The data have been discretized into 1,000- time bins. For the megafauna and human fossil records each point represents a robust (following Barnosky and Lindsey, 2010) radiocarbon date on a single fossil, converted to calendar years BP. The 95% confidence intervals on those ages are indicated, and are asymmetrical due to the non-linear conversion from radiocarbon years to calendar years. Estimated times of extinction for the megafauna and time of arrival of humans, using the GRIWM method, are shown by black bars within the adjacent dashed boxes with gray normal distributions representing the 95% confidence bands on those estimates. The forest-cover values were estimated from pollen data, and changes in temperature are based on an EPICA Dome C . increasingly larger proportion of the surviving beyond the last observed extinction, and because megafauna. However, this approach requires in most of the analyses, there is no human or knowing the standing diversity of megafauna in climate change in the later time points. all time bins, and sufficiently well-dated specimens are not available to make these Determining extinction and arrival times estimates. It is also unclear whether the extinction There are two sources of uncertainty in the times of the last megafaunal taxon to go extinct should of local extinction of the megafauna and the be any more significant than the extinction of any arrival time of humans: 1) uncertainty in the other single taxon. radiocarbon dates on the fossils; and 2) the Note that Eq. (1) does not account for the fact incompleteness of the fossil record. Even with that at some point, there may no longer be any perfect age control for the fossils, there typically taxa left as extinction candidates, which can lead will be a gap between the youngest fossil and the to spurious results. Thus, there might be climate actual time of extinction (Strauss and Sadler, and/or human changes that would have led to 1989; Marshall, 1990, 2010). To accommodate extinction if more taxa were available. If this were these complexities, Villavicencio et al. (2015) the case, Eq. (1) would down-weight its used a Gaussian resampled inverse-weighted assessment of the importance of these drivers of method (GRIWM) developed by Bradshaw et al. extinction if human or climate change correlated (2012), which is a modification of a statistical with previous extinction did not cause more approach that was originally constructed for extinction at a later time simply because there inferring recent extinctions based on sighting were no more taxa to go extinct. This is not a records (McInerny et al., 2006). GRIWM significant problem in the Última Esperanza data, estimates true times of local extinction and largely because the time series barely extends arrival. The method adds a representative gap

8 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21 drawn from the observed stratigraphic range onto change were used: 1) estimates of 100-year mean the end-point of stratigraphic ranges, which temperatures from the EPICA Dome C (EDC) represents an unbiased estimate of the true time of from east Antarctica derived from deuterium disappearances or appearances. To accommodate measurements from an ice core (δDice; Fig. 4); nonrandom fossilization, it up-weights younger and 2) proportion of Nothofagus forest cover in gaps in the stratigraphic range to account for the the Última Esperanza region (Fig. 4). We suggest fact that local abundances, and thus preservation that the change in forest cover captures climate rates, are likely to drop approaching the extinction change more completely than simple temperature time (or, in the case of arrival times, it up-weights change. The reason for this is twofold. First, the older gaps). Uncertainties associated with the change in average temperature does not radiometric dates leads to a 95% confidence band necessarily provide the best proxy for biologically around the estimated time of extinction or arrival relevant climate changes, which include such (Fig. 4). Villavicencio et al. (2015) used R code features as the frequency and magnitude of provided by Saltré et al. (2015; Appendix A). This temperature extremes, change in precipitation technique was also applied to estimate the timing both annually and seasonally, etc. In contrast, of first human arrival. change in vegetation integrates a spectrum of physical climate-change parameters. Second, Temporal resolution vegetation change is directly biologically relevant Given the uncertainties in the times of to extinction given the reliance of mammalian disappearance of the megafaunal taxa and the communities on the local vegetation—in fact, the arrival time of humans (Fig. 4), the time series nature of local vegetation is a powerful predictor shown in Figure 4 was divided into 1,000 year of animal distributions. In the case of the Última intervals, recording the number of extinctions, the Esperanza megafauna, vegetation change is time of human arrival and hypothesized particularly relevant given the presence of population increase, and the change in the taxa and the transition from cold grasslands to Nothofagus forest cover and temperature. The warmer, moister forest during the extinction placement of the boundaries is arbitrary, and so interval. two sets of analyses were run: the first with The proportion of forest cover from two temporal boundaries placed at the beginning of pollen cores close to the Última Esperanza the millennia (e.g., 12,000 years BP, 11,000 years megafauna sites was estimated (Villavicencio et BP, etc.), and the second with the boundaries al., 2015, fig. 1). The first core was taken from the placed at the half-millennial boundaries (e.g., center of a small surrounded by a , Lago 12,500 years BP, 11,500 years BP, etc.). Eberhard, ~4 km south west of the megafauna While the uncertainty in the estimated times sites. The second was taken from Pantano of extinction and human arrival led to the choice Dumestre, ~26 km to the south, which began as of a 1,000 year temporal resolution for our lake but filled in to become a bog ~14,600–14,900 analyses (the 95% confidence bands on the years BP. A composite forest cover curve was extinction and human arrival span 1,090 years on constructed using the available data from Lago average), including too many time bins may Eberhard from 10,400–12,600 years BP, which compromise the ability of the method to correctly was then correlated to the Pantano Dumestre core identify the correlates of extinction. For example, to extend the record further back in time to just if there are many more time bins than extinctions, over 14,000 years BP. The pollen percentages at then there may be many situations where changes Pantano Dumestre were rescaled to match the in climate or human impacts do not correlate with Lago Eberhard core where their records extinction simply because there are insufficient overlapped to provide quantitative continuity extinctions in the data set. In these cases, the between the two cores. The Lago Eberhard record method will inappropriately down-weight the was used as the primary record despite its shorter significance of these factors. In general, the temporal coverage because it is closer to the number of temporal bins ideally should paleontological and archaeological sites approximate the number of extinctions in the data examined, and because, as a bog, Pantano set. Dumestre was assumed to have a potentially more localized pollen representation. The time interval Measuring climate change studied captures the rise from essentially no tree Two measures of non-human environmental cover to full forest cover, which corresponds to

9 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION about 70% Nothofagus pollen in the Lago measure of non-human environmental change, Eberhard core. there is a lack of good quantitative measures of In the absence of additional data, it was human population change with time. Second, assumed that forest cover was essentially absent following directly from the discussion of time when Nothofagus pollen accounted for only a few lags above, and assuming there was a causal percent of total pollen grains prior to 14,000 year relationship between humans and extinction, it is BP, and that there was effectively dense forest not known if human impact was constant since cover from 10,400 years BP to the top of the their first arrival or increased through time, nor is analysis window. the nature of that impact known. If there were impacts, they could have been direct, such as Dealing with time lags between extinction hunting, or indirect, such as energetic constraints times and time of onset of drivers of extinction on how many megafauna bodies (including Equation (1) assumes that the response of the human bodies) a given patch of real estate could system (extinctions) occurred in the same support (Barnosky, 2008). temporal bin as the change in the driver of For Última Esperanza, there are no extinction. Thus, at a temporal resolution of 1,000 quantitative data on how human population sizes years, this assumption means that the equation in the area, or the nature of their impacts, may can only capture the dynamics of extinction if have fluctuated or changed with time. Thus, here extinction occurred no more than 1,000 years we used the estimated arrival time of humans, and after the onset of driver of extinction. Given that the number of caves where archaeological the time of onset of the driver is unlikely to have evidence has been dated (there are just two), as a occurred exactly at the beginning of the temporal measure of relative population size in this bin, nor the extinction exactly at the end of the analysis. temporal bin, the equation will more typically accommodate a time lag of a few hundred years. Initial analyses For the megafaunal response to changes in The initial analyses were based on the following forest cover, this time lag seems reasonable: if a assumptions and data treatments: 1) Data were taxon is unable to live in forested , then parsed into 1,000-year time intervals. Two that taxon would be expected to disappear shortly analyses were performed, one where temporal after the amount of forested landscape reached boundaries were placed at the beginning of the critical thresholds—certainly within 1,000 years, millennia (the ‘millennial boundary’ analyses), and probably within a few hundred years. and the other at the half-millennial boundaries However, the validity of this assumption is (the ‘half-millennial boundary’ analyses). 2) less certain for megafaunal response to human Times of extinction were assumed to correspond arrival and presence. It is possible that to the median estimate of the extinction times megafaunal extinctions may have occurred within based on the GRIWM confidence intervals, and 1,000 years, but conceivably could have been that the time of arrival of humans similarly more protracted if attrition of megafauna corresponded to the median position of the populations was sufficiently slow (Mosimann and confidence interval on their arrival time (see Fig. Martin, 1975; Alroy, 2001; Brook and Bowman, 4). 3) Humans were assumed to only have had an 2004; Koch and Barnosky, 2006). Adding a impact in the time bins in which we surmise their parameter to the model that would allow it to find numbers might have been increasing as measured the best-fit time lag between human arrival and by the number of caves where human remains extinction was considered, but doing so assumes have been found – this is called the ‘initial impact that humans were a causal agent, which is a only’ hypothesis. hypothesis we want to evaluate. An alternative In all analyses, forest cover was normalized approach is to code human presence as a potential so that maximum forest cover was scored as 1.0 cause of extinction regardless of whether the (with Nothofagus pollen percentages of ~70% population was changing, which is explored standardized to 1.0). Similarly, human impact was below. normalized so that maximum human presence (measured crudely by the number of caves Measuring human impacts yielding human fossils) was scored as 1.0. There are two issues that make the quantification of human impacts difficult. First, unlike the Results for millennial boundary analysis

10 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21

Application of Eq. (1) to the data (Fig. 4) yielded possibility is that the presence of Nothofagus highly significant climate (p = 0.005, a = 3.28 + forest made hunting by humans more challenging, 0.70) and human effects (p = 0.008, b = 2.00 + thus reducing the impact of humans on the 0.47), as well as an essentially significant, but megafauna populations. completely unexpected negative synergistic effect To assess the assumption that humans had (p = 0.051, c = –24.75 + 9.7). little effect on the megafauna when they first The reasons for significant human and climate arrived due to low initial population size, their effects are easy to determine: there are extinctions first impact 14–13 ka ago was down-weighted when there is human impact alone (13–12 ka from 0.5 to 0.1. With the climate impact set to ago), and extinctions when there is climate 0.011 (reflecting the fact that the habitat was still change alone (12–11 ka ago) (Fig. 4). Given this, largely grassland despite the increase in the signal for negative synergy can be understood. Nothofagus pollen), the p-value for a negative In the 14–13 ka time interval, there is both human synergistic effect became non-significant (0.639), impact and forest cover change, but no extinction. as anticipated. However, given the data from 13–11 ka ago, there Without more data, it is difficult to understand should have been about 1.5 extinctions 14–13 ka whether the negative synergy is real, and what it ago if the human and climate factors responsible means if it is. However, the framework for the extinctions 13–11 ka operated additively: a established by Eq. (1) enables a much richer way human impact of 1.0 corresponds to 2 extinctions of thinking about the megafaunal extinctions in 13–12 ka ago, and a forest cover change of 0.64 any case. corresponds to 2 extinctions 12–11 ka ago, Experimenting with the nature of human followed by another extinction with a further 0.21 impacts.—In the above analyses, it was assumed vegetation change in the millennium after that. that humans only posed a threat to the megafauna Thus, the 0.5 human impact 14–13 ka ago should in the millennium of their arrival and in the correspond to about 1 extinction, and the 0.11 following millennium where there is evidence of forest change in that time bin should correspond population increase. However, it is possible that to about half an extinction. However, given that humans remained a threat after their initial impact this is the only interval where both human and —the ‘persistent impact’ hypothesis. This was climate impacts are registered together, the non- tested by changing the last three values in the ∆H linear regression yields a result of negative time series (Fig. 4) from “0”s to “1”s. With synergy to explain the lack of extinction in the human impact coded this way, none of the interval. To test this reasoning, the vegetation parameters were significantly different from zero. change was down-weighted from 0.11 to 0.011, However, as noted above, recording human and, as expected, the p-value for negative energy impact when there were no megafaunal taxa left became less significant from 0.051 to 0.106. to be driven to extinction can confound the (Note that down-weighting the vegetation cover analysis, so the time interval (9–10 ka ago) after during this time interval seems reasonable, given the last megafaunal extinction was eliminated and that an increase of forest cover from ~3% to 14% the analysis was run again. This yielded only many have had little, if any, effect on most taxa). human impacts as being important (p = 0.044), Interpreting negative synergy.—When but not vegetation change (p = 0.30) or synergy (p establishing the quantitative framework employed = 0.29). With these data, there is always here, it did not occur to us that there could be extinction with human impacts, but not always negative synergy between vegetation change and extinction with vegetation change, so the non- human impacts. So, a question arises: if the linear regression puts all the weight on the human negative synergy is real, what could it mean? One component. Note that it is only in hindsight that possibility is that whatever was responsible for there were no more megafauna left to be driven to the increase in vegetation (perhaps increased extinction in the last time bin: three surviving temperature, or less severe or frequent extreme megafauna species are known from Ultima weather) may also have promoted increases in Esperanza, including the (Lama megafaunal numbers that more than compensated guanicoe), puma (Puma concolor), and huemul for the onset of human predation. In fact, it is (Hippocamelus bisulcus), and at least the first two possible humans began to frequent the Última were present in the region in the Pleistocene. Esperanza region precisely because there was an Because these taxa might have become extinct, a increase in the abundance of megafauna. Another case could also be made for not eliminating the

11 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION last time bin. human impact is re-coded under the ‘persistent Finally, an experiment was run with an impact’ scenario (ΔH = 1 in time bins 6–9), intermediate between the ‘initial impact only’ and human impact emerges as the most important ‘persistent impact’ scenarios: the ‘high initial factor, with a significant p value (0.051), with impact, weak persistence’ scenario. This was climate effects (p = 0.887) and human-climate achieved by retaining the initial impact coding of synergy (p = 0.751) unimportant. If the youngest a 0.5 and 1.0 in bins 15–13 ka ago, followed by time bin (bin 9) is eliminated (where there is no 0.1, 0.2, and 0.3 in the next three time bins. These extinction or climate change), the p value for latter values, which are meant to represent human impacts, as expected, is even stronger ongoing attritional pressure on the taxa which (0.021). survived any initial human impact, are relatively For the ‘high initial impact, weak persistence’ small and, as might be expected, the results were scenario, there was no significant support for any similar to the ‘initial impact analyses’, except parameter, with the strongest signal being for with weakened p values (human impact, p = human impact (p = 0.11). 0.029, climate effects, p = 0.021, negative human–climate synergy, p = 0.151). Summary of initial analyses The analyses above show that the results are Results for half millennial boundary analysis sensitive to two main features of the data. The For the half-millennial-boundary analysis (Table first is where the temporal boundaries are placed, 2) with the human ‘initial impact only’ scenario, and the second is how human impact is assessed. the only significantly supported parameter was for In the millennial-boundary analyses, humans first climate effects (p = 0.015, a = 2.78 + 0.82), with occupy both cave sites the same millennium that no significant human effect (p = 0.61, b = 3.50 + the big cats go extinct, with no accompanying 6.49), or synergistic effect (p = 0.701, c = –86.07 forest cover change; thus, there is a strong human + 213.35). The reason is fairly straightforward— impact signal. There are also extinctions when the the bulk of the extinctions are spread out over vegetation cover dramatically increases, so there four millennia, while there is only postulated is a climate signal too. significant human impact in one of those However, in the half-millennial analysis, the millennia, but vegetation change occurred during big cat extinctions are staggered over two time three of the four. intervals, one of which is younger than the arrival As in the millennial-boundary analysis, when of humans, and there is vegetation change in both of these intervals. Given the subsequent TABLE 2.—Data for Última Esperanza for half- extinctions that correlate with further forest millennial time bins (see Fig. 4 for raw data and the increase, the human impact signal disappears but millennial time bins). For humans their impact was the climate signal remains. The human signal scored as 0.1 for bin 4, given that the GRIWM reappears (at the expense of the climate signal) if confidence interval falls in that bin, but for which there is no direct evidence of human presence. In the one assumes that human presence remains a following bin (bin 5), the time interval encompasses persistent threat through time. both caves where human remains have been found, so In the analyses above, it is assumed that it is the human impact was scored as 1.0. known when, to the nearest 1,000 years, each taxon became extinct, and when humans arrived. Time Bin Years BP E ΔH ΔC However, as shown in Fig. 4, extinction/arrival 9 9,500–8,500 0 0 0 times are not known with that degree of precision. 8 10,500–9,500 1 0 0 Below, we assess the impact of uncertainty in the extinction and arrival times on the results of the 7 11,500–10,500 2 0 0.76 analysis, as well as the impact of using 6 12,500–11,500 1 0 0.13 temperature as a measure of climate change. 5 13,500–12,500 1 1 0.03 SENSITIVITY ANALYSES 4 14,500–13,500 0 0.1 0.06 3 15,500–14,500 1 0 0 Uncertain extinction and arrival times In the analyses above, it is assumed that 2 16,500–15,500 0 0 0 extinction/arrival times correspond to the most 1 17,500–16,500 0 0 0 likely time interval of extinction/arrival (‘best

12 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21 bin’) as judged by the GRIWM confidence arrival. Analysis of these extinction scenarios with intervals (Fig. 4). However, the bands on the Eq. (1) (Table 3B) shows that in half of the confidence interval are sufficiently broad that the bootstrapped extinction scenarios, a significant assurance that all the extinction/arrival times human impact is found, and in eight of ten actually occurred in the ‘best bins’ (the ‘best bin scenarios a significant climate effect is seen. pattern’) is quite small. This can be calculated by Interestingly, only one scenario shows significant multiplying the cumulative probabilities that each (p < 0.05), negative synergy, with two other species became extinct (or arrived) in its ‘best scenarios showing p values for negative synergy bin,’ so for the millennial-boundary analysis, this between 0.1 and 0.05. probability is only 2.9%, and for the half- Table 4 shows the results for the bootstrapped millennial boundary analysis, it is 2.7%. extinction patterns (Table 4A) for the half Thus, the fossil data are consistent with a millennial-boundary time bins. In none of the range of different extinction patterns. For bootstrapped scenarios for the human ‘initial example, with the millennial-boundary time bins, impact only’ scenario (Table 4B) did any of the may have become extinct in one of parameters find significant support, even at the p three bins, 14–13 ka ago (5.3%), 13–12 ka ago = 0.10 level, despite the strong support for climate (89.4%), and 12–11 ka ago (5.3%), while effects in the ‘best bin’ scenario. The reason the may have become in extinct in one of ‘best bin’ analysis, but not the bootstrapped two intervals, etc. (Fig. 4). Multiplying the analyses, yielded a significant result appears to be number of intervals each taxon might have gone because in the ‘best bin’ treatment, support for extinct/arrived across all taxa yields 824 possible climate effects comes from the two extinctions extinction/arrival patterns with the millennial- (an equid and camelid) from 11.5–10.5 ka ago, the boundary time boundaries, and 2,880 possible time of maximum increase in the forest cover extinction patterns with the half-millennial time from about 10–15% to ~90% (see time bin 7 in boundaries, ignoring possible extinction/arrival Table 2). But, in none of the bootstrapped times that are more than two standard deviations extinction scenarios is there more than one from their best estimates. The most likely extinction in that time bin (data not shown). extinction/arrival scenario corresponds to the Clearly, to be certain of a climate effect (as ‘best bin’ pattern, while some other patterns are measured by forest cover), tighter control is highly unlikely (e.g., it is highly unlikely that all needed on extinction times of the equid and the the species extinctions occurred in the youngest- camelid. most tails of the range of their possible extinction For the ‘persistent impact’ human scenario times). (Table 4C), while there is no support for any of Testing all possible patterns of extinction/ the parameters in the ‘best bin’ analysis, there is arrival time is impractical, in part because of the usually strong support for human effects among difficulty in visualizing the results, especially the bootstrapped extinction patterns, which occurs given that the p-values cannot be averaged across for a variety of reasons. extinction scenarios because this could mask less- common scenarios with significant results, and Other measures of non-human environmental make it difficult to detect dependencies among the change drivers of extinction (e.g., some drivers may In the analyses above, the increase in forest cover never act together, or always act together). was used as the proxy for non-human A bootstrapping method for determining the environmental change. However, an estimate of range of most plausible extinction scenarios was the change in mean annual temperatures (Fig. 4) developed for each scenario. where a random is also available from the Antarctic EPICA Dome extinction time was drawn from the normally C ice core. Here, we explore how using distributed range of possible extinction times for temperature change instead of forest cover affects each species. the analysis. Table 3A shows the results of 10 bootstrapped Simple temperature change.—For the extinction/arrival scenarios for the millennial- ‘millennial-boundary’ time bins and the human boundary time bins, along with the ‘best bin ‘initial impact only’ scenario, none of the pattern,’ and for a literal reading of the fossil parameters were significantly correlated with the record, where the last and first appearance dates extinction times for any of the extinction were treated as the actual times of extinction/ scenarios (even at p = 0.10), while for the

13 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION

TABLE 3.—A. Bootstrapped extinction/arrival patterns for the ‘millennial boundary’ time bins, recorded as the number of bins below (negative numbers), within (0), or above (positive numbers) the most likely time bin of extinction or arrival. Also given is the most likely (‘Best Bin’) extinction/ arrival pattern, and the pattern observed if one takes the fossil record at face value, the ‘literal’ pattern seen before the application of confidence intervals on the stratigraphic end-points. B and C. Results of analysis of the bootstrapped extinction/arrival patterns for the two primary human impact scenarios employed. Grey boxes = p values < 0.10, bold values are p values < 0.05. A. Extinction/arrival scenarios Best Bootstrapped Extinction Scenario Extinction Bin Literal 1 2 3 4 5 6 7 8 9 10 Smilodon 0 -1 1 0 0 0 0 0 0 0 0 0 Panthera 0 0 0 -1 0 0 0 0 -1 0 0 0 0 1 -2 0 0 0 0 0 -1 -1 0 1 Lama owenii 0 0 0 1 0 1 0 0 1 -1 -1 1 Vicugna 0 0 0 -2 -1 1 0 -2 0 -1 0 0 0 0 -1 0 0 -1 -1 0 -1 -1 -1 0 Human arrival 0 1 0 1 1 0 0 1 0 0 0 0

B. Values of p derived from equation (1): Human scenario = initial impact only ‘Climate’ (a) 0.01 0.01 0.39 0.00 0.00 0.20 0.04 0.00 0.13 0.58 0.13 0.63 Human (b) 0.01 0.01 0.37 0.00 0.00 0.01 0.03 0.00 0.03 0.12 0.03 0.00 Synergy (c) -0.05 -0.05 0.78 -0.13 -0.02 -0.07 -0.15 -0.13 -0.18 -0.40 -0.18 -0.03

C. Values of p derived from equation (1): Human scenario = persistent impact ‘Climate’ (a) 0.53 0.53 0.58 0.53 -0.51 -0.53 -0.53 0.53 -0.31 -0.33 -0.31 0.60 Human (b) 0.13 0.13 0.06 0.08 0.11 0.06 0.10 0.08 0.01 0.01 0.01 0.08 Synergy (c) 0.47 0.47 -0.55 -0.60 0.45 0.58 0.50 -0.60 0.33 0.41 0.33 0.68

‘persistent impact’ human scenario, there was change in temperature that is important, but the significant correlation between human impacts absolute temperature, analyses were rerun using and extinction for the ‘best bin’ data and nine of the cumulative temperature change that had the ten bootstrap extinction/scenarios (Table 5). occurred since the region became free of ice. Presumably for the ‘persistent impact’ scenario, Support was found only for a correlation between the fact that human change and the extinctions are cumulative temperature and extinction for the concentrated in the later part of the time series, ‘initial impact only’ human scenarios, supporting while temperature change is seen throughout, temperature change as a significant correlate with favors human impacts over climate change as an the extinctions for both the millennial and half- explanation for the extinctions. millennial time boundary time intervals (including For the ‘half millennial boundary’ time bins, all, and eight of the ten bootstrap extinction the results are very similar (Table 5), except for scenarios, respectively). Interestingly in the the ‘persistent impact’ human scenario, where millennial boundary analyses, three of the beyond the strong support for human impacts, bootstraps supported positive human-climate there was also one bootstrap extinction/arrival synergy. There was no support for a correlation scenario that gave support for a positive between human impacts and extinction. The synergistic effect between human impact and reason cumulative temperature shows a climate (p = 0.01). correlation with extinction but not simple Cumulative temperature change.––To try and temperature change is that the extinctions are capture the sense that it may not be only the concentrated in the latter part of the time series,

14 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21

TABLE 4.—A. Bootstrapped extinction/arrival patterns and results for the ‘half millennial boundary’ time bins (see Table 3 for further explanation). B, C. Results of analysis of the bootstrapped extinction/arrival patterns for the two primary human impact scenarios employed. Grey boxes = p values < 0.10, bold values are p values < 0.05. A. Extinction/arrival scenarios Best Bootstrapped Extinction Scenario Extinction Literal Bin 1 2 3 4 5 6 7 8 9 10 Smilodon 0 0 -1 0 0 0 -1 0 0 0 0 0 Panthera 0 1 0 0 0 0 0 0 0 0 0 1 Hippidion 0 0 -2 1 -2 0 -1 -2 1 1 -1 1 Lama owenii 0 1 0 1 1 1 -1 0 -1 1 1 1 Vicugna 0 0 -1 -1 -2 0 0 -1 0 -1 -1 1 Mylodon 0 -1 -1 0 0 -1 1 0 1 -1 -1 1 Human arrival 0 0 0 0 0 0 0 0 -1 0 1 0

B. Values of p derived from equation (1): Human scenario = initial impact only ‘Climate’ (a) 0.02 0.13 0.27 0.71 0.75 0.21 0.30 0.13 0.13 0.71 0.73 0.17 Human (b) 0.61 0.51 -0.45 -0.58 0.69 0.77 0.94 -0.33 -0.33 -0.58 -0.62 0.59 Synergy (c) -0.70 -0.69 0.45 0.53 -0.83 -0.84 -0.94 0.27 0.27 0.53 0.52 -0.74

C. Values of p derived from equation (1): Human scenario = persistent impact ‘Climate’ (a) 0.53 0.53 0.58 0.53 -0.51 -0.53 -0.53 0.53 -0.31 -0.33 -0.31 0.60 Human (b) 0.13 0.13 0.06 0.08 0.11 0.06 0.10 0.08 0.01 0.01 0.01 0.08 Synergy (c) 0.47 0.47 -0.55 -0.60 0.45 0.58 0.50 -0.60 0.33 0.41 0.33 0.68 which is when the cumulative temperature is also extinction scenarios. The only major anomaly was greatest. in the ‘millennial-boundary,’ ‘initial human For the ‘persistent impact’ human scenario, impact only’ scenario, where in the forest-cover there was no support for climate change analysis, a literal reading of the fossil record gave correlating with extinction, while in just three of the worst support for change in forest cover the bootstrap extinction scenarios for the ‘half playing a role in the extinctions (p = 0.68), largely millennial’ analyses, there was significant support because a literal reading of the fossil record pulls for human impact (one of which was negative), some of the extinctions earlier than the major with one bootstrap showing positive synergy, and increase in forest cover. one showing negative synergy. With the persistent human scenario, there is always human impact SUMMARY OF ÚLTIMA ESPERANZA and climate change, and we suspect that there are ANALYSES simply not enough extinctions to identify climate or humans as being strongly correlated with the Table 5 summarizes the results of the analyses of extinctions. the most likely extinction/arrival pattern, and for Ignoring the incompleteness of the fossil ten bootstrapped alternative extinction/arrival record.—This analysis was run assuming that the patterns. The primary findings are: observed stratigraphic end points, i.e., the median radiocarbon date of the latest dated specimen for Climate change each taxon, represented the true time of extinction All analyses with the most likely extinction for that taxon. Typically, the literal reading of the pattern and with human impact coded under the fossil record is different from the ‘best bin’ ‘initial impact only’ scenario found significant analyses (e.g., see Tables 3, 4), but not too support for climate change correlating with different from at least some of the bootstrapped extinction, except when using simple temperature

15 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION

TABLE 5.—Summary of analyses of the drivers of the Última Esperanza megafaunal extinction data (Fig. 4) using Equation (1). *** = p < 0.05; * = 0.05 < p < 0.10; (–) means that the parameter had a significantly negative value. A cutoff of p < 0.10 was used when counting the number of bootstrapped extinction scenarios. Extinction/arrival pattern: Most likely (‘Best bin’) Bootstrapped (of 10) Driver of Extinction: Climate Human Synergy Climate Human Synergy (a) (b) (c) (a) (b) (c) ΔC (Forest cover change) ΔH (Initial impact only) Millennial bounds *** *** *(–) 5 8 3(–) Half-millennial bounds *** –– –– –– –– –– ΔH (Persistent impact) Millennial bounds –– –– –– –– 7 –– Half-millennial bounds –– * –– 1 10 1(–) ΔC (Temperature change) ΔH (Initial impact only) Millennial bounds –– –– –– –– –– –– Half-millennial bounds –– –– –– –– –– –– ΔH (Persistent impact) Millennial bounds –– * –– –– 9 –– Half-millennial bounds –– *** –– –– 9 1 ΔC (Cumulative temp. change) ΔH (Initial impact only) Millennial bounds *** –– –– 10 –– 3 Half-millennial bounds *** –– –– 8 –– –– ΔH (Persistent impact) Millennial bounds –– –– –– –– –– –– Half-millennial bounds –– –– –– –– 2, 1(–) 1, 1(–) change. Typically, there was modest to strong boundaries’ when using change in forest cover as support for this conclusion among the a proxy for climate change and the ‘initial impact bootstrapped extinction/arrival scenarios. The only’ scenario. Human impact was also supported support for climate as a driver disappears when when the half-millennial boundaries were used coding human impact under the ‘persistent with the ‘persistent impact’ scenario with change impact’ scenario, given that human impact is in forest cover and change in temperature. The always present. Understanding how humans sensitivity of the analysis to the placement of the interacted with and affected the megafaunal temporal boundaries indicates that if humans were species is key to understanding the significance of important, then either there was a time lag in the the climate change signal seen in the data. response of the megafauna to human arrival, and/ Cumulative temperature change also strongly or humans remained a threat after the initial correlated with extinction for the human ‘initial arrival and growth. Typically, most bootstrapped impact only’ scenario. extinction/arrival scenarios supported the most likely extinction/arrival scenario analyses. Human impact Human impact was supported in the most likely Synergistic interaction between climate change extinction/arrival scenario with ‘millennial and human impacts

16 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21

Usually there was no support for positive are several ways of capturing that threat, and the synergistic effects between climate change, results are somewhat sensitive to the way the however measured, and human impacts, except impact is quantified. There is a close relationship for rare bootstrap extinction patterns (Table 5). between the time lag between human arrival and However, support for negative synergy was seen impact, the binning of time into discrete intervals, for the most likely extinction/arrival scenario (and and the nature of persistent human impacts. three of ten bootstraps) with the millennial boundaries for the ‘initial impact only’ human 4) Measuring and change in forest cover. As discussed The results also depend on the proxies used to above, this effect was diminished if the impact of represent climate or environmental change, and humans in the first millennium of contact was how those proxies are expressed quantitatively reduced, and if the magnitude of ‘climate’ change (e.g., as relative or cumulative change). was also reduced when the forest cover was still small. Among all the analyses, there were only FUTURE DIRECTIONS two other cases of negative synergy (see Table 5). Thus, the idea of negative synergy, while an Need for better data intriguing possibility, requires more data to The sensitivity analyses make it clear that more evaluate. data are needed to have a clear picture of the causes and nature of the late Quaternary CONCLUSIONS megafaunal extinctions in the Última Esperanza region. Currently, there is an urgent need for more The analytic approach developed here provides a radiocarbon dates for most megafauna taxa that platform for explicit data analysis and went extinct in order to narrow the time span over exploration, offering the opportunity for deeply which extinction is most probable. With the sharpening the analyses of the causal relationships available data, a literal reading of the fossil between extinction and their putative drivers. record, where the youngest radiocarbon (or other) The analysis of the megafaunal extinction date is considered to represent the time of pattern in the Última Esperanza region of extinction, is likely to give erroneous conclusions southern South America with this approach does about how extinction correlates with potential not provide unambiguous results. There are three causal mechanisms because the actual extinction primary sources of uncertainty, with a fourth date is virtually certain to be a few hundred to a factor also playing a role. few thousand years younger than the last radiocarbon date. Scenarios that rely on 1) Precision of extinction and arrival times correlative analyses seldom take this into account The results depend critically upon when humans (but see Bradshaw et al., 2012), but need to do so. arrived and when the megafauna went locally There is also a critical lack of information extinct. A higher precision temporal control is about how human population sizes grew in needed, and, in fact, with this framework, it is particular regions after first contact with possible to estimate how many more high- megafauna. Such data is becoming available precision radiocarbon dates would be needed to through compilations of archaeological evidence remove this source of ambiguity. in ways that should allow quantitative representation of human impact from the late 2) Discretizing the temporal data Pleistocene through the Holocene (Goldberg et The results depend on how time is binned into al., 2015), but these data have generally been discrete intervals. This indicates that time lags are widely scattered through a diverse literature. a crucial component of the megafauna–human Of the kinds of data needed for this model, interactions. the data that perhaps is most abundant at this point is paleoenvironmental proxy data, in such 3) The nature of human impacts forms as pollen and charcoal records and stable- The results depend on the nature of human impact isotope analyses, that can often provide on the megafauna, and whether their impact was continuous time-series for selected areas. only in the first one-to-two millennia after first However, an important issue in bringing together contact, or whether they represented a persistent all these data sets is that they must all come from threat. If humans were a persistent threat, there the same geographic area—humans, climate,

17 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION other animals, and plants all interact in a specific REFERENCES place, and the driving forces of change can be quite different geographically from region to Alroy, J. 1999. Putting 's end- region. Pleistocene megafaunal extinction in context: Large-scale analyses of spatial patterns, extinction Need to develop ways of rigorously dealing rates, and size distributions, p. 105–143. In R. D. with ambiguity E. MacPhee (ed.), Extinctions in Near Time. Kluwer Academic/ Plenum Publishers, New York. Fossil and geologic data, even in the best cases, Alroy, J. 2001. A multispecies overkill simulation of are incomplete and therefore will often fall short the end-Pleistocene megafaunal mass extinction. of giving the one ‘right’ answer. Therefore, to Science, 292:1893–1896. understand which of the range of potential Barnosky, A. D. 1985. Taphonomy and herd structure interpretations is most probable, quantitative of the extinct Irish , Megaloceros giganteus. exploratory models such as the one discussed here Science, 228:340–344. are essential. Such models, when informed by Barnosky, A. D. 1986. "Big game" extinction caused sound data, offer a new way forward in the long- by climatic change: standing debate about whether humans, climate (Megaloceros giganteus) in Ireland. Quaternary change, or interactions between the two ultimately Research, 25:128–135. drove half the world’s large-bodied species to die Barnosky, A. D. 2008. Megafauna biomass tradeoff as a driver of Quaternary and future extinctions. out as the Pleistocene gave way to the Holocene. Proceedings of the National Academy of Sciences They may also prove useful in understanding how USA, 105(suppl. 1):11543–11548. the same drivers—humans and climate change, Barnosky, A. D., P. L. Koch, R. S. Feranec, S. L. Wing, this time human-caused—are likely to manifest as and A. B. Shabel. 2004. Assessing the causes of the Holocene transforms into a new age, the late Pleistocene extinctions on the continents. . Science, 306:70–75. Barnosky, A. D., and E. L. Lindsey. 2010. Timing of Generality of the method Quaternary megafaunal extinction in South Finally, this approach is generalizable to any America in relation to human arrival and climate situation where there are multiple drivers of change. Quaternary International, 217:10–29. extinction (or origination for that matter) where Barnosky, A. D., E. L. Lindsey, N. A. Villavicencio, E. Bostelmann, E. A. Hadly, J. Wanket, and C. R. there are quantitative data available on the Marshall. 2015. The variable impact of Late temporal correlations between the extinction and Quaternary megafaunal extinction in causing the potential drivers of those extinctions. ecological state shifts in North and South America. Proceedings of the USA National Academy of ACKNOWLEDGMENTS Sciences, in press. Bond, W. J. 2005. Large parts of the world are brown We acknowledge support from the National or black: A different view on the 'Green World' Science Foundation EAR 1148181. Conversations hypothesis. Journal of Vegetation Science, with the following colleagues enriched our 16:261–266. understanding of late and Bond, W. J., and J. E. Keeley. 2005. Fire as a global human in South America: '': the and evolution of flammable . Trends in Ecology and Luis Borrero, Enrique Bostelmann, Amy Evolution, 20(7):337–394. Goldberg, Elizabeth Hadly, Fabiana Martín, Bradshaw, C. J. A., A. Cooper, C. S. M. Turney, and B. Patricio Moreno, Alexis Mychajliw, Gustavo W. Brook. 2012. Robust estimates of extinction Politis, and José Luis Prado. We thank P. David time in the geological record. Quaternary Science Polly, Jason Head, and David Fox, the organizers Reviews, 33:14–19. of this Paleontological Society Short Course, for Brook, B. W., and A. D. Barnosky. 2012. Quaternary inviting us to participate. Thanks to Patricia extinctions and their link to climate change, p. Holroyd for feedback on the manuscript, and 179–198. In L. Hannah (ed.), Saving a Million special thanks to P. David Polly for a critical Species. Island Press, Washington, D. C. review of the manuscript, and for confirming our Brook, B. W., and D. M. J. S. Bowman. 2004. The results with Mathematica. This is UCMP uncertain blitzkrieg of . Journal of , 31:517–523. Contribution 2067. Dirzo, R., H. S. Young, M. Galetti, G. Ceballos, N. J. B. Isaac, and B. Collen. 2014. Defaunation in the

18 THE PALEONTOLOGICAL SOCIETY PAPERS, V. 21

Anthropocene. Science, 345:401–406. Nature, 426:169–171. Flannery, T. F., and R. G. Roberts. 1999. Late Guthrie, R. D. 2006. New carbon dates link climatic Quaternary extinctions in : an change with human colonization and Pleistocene overview, p. 239–255. In R. D. E. MacPhee (ed.), extinctions. Nature, 441:207–209. Extinctions in Near Time: Causes, Contexts, and Koch, P. L., and A. D. Barnosky. 2006. Late Consequences. Kluwer Academic/Plenum Quaternary extinctions: State of the debate. Publishers, New York. Annual Review of Ecology Evolution and Galetti, M., and R. Dirzo. 2013. Ecological and Systematics, 37:215–250. evolutionary consequences of living in a MacPhee, R. D. E. 1999. Extinctions in Near Time: defaunated world. Biological Conservation, Causes, Contexts, and Consequences. Kluwer 163:1–6. Academic/Plenum, New York. Gill, J. L. 2014. Ecological impacts of the late Marshall, C. R. 1990. Confidence intervals on Quaternary megaherbivore extinctions. New stratigraphic ranges. Paleobiology, 16:1–10. Phytologist, 201:1163–1169. Marshall, C. R. 2010. Using confidence intervals to Gill, J. L., J. W. Williams, S. T. Jackson, J. P. Donnelly, quantify the uncertainty in the end-points of and G. C. Schellinger. 2012. Climatic and stratigraphic ranges, p. 291–316. In J. Alroy and megaherbivory controls on late-glacial vegetation G. Hunt (eds.), Quantitative Methods in dynamics: a new, high-resolution, multi-proxy Paleobiology, The Paleontological Society Papers record from Silver Lake, . Quaternary 16, Yale Press, New Haven. Science Reviews, 34:68–80. Martin, P. S. 1966. African and Pleistocene overkill. Gill, J. L., J. W. Williams, S. T. Jackson, K. B. Nature, 212:339–342. Lininger, and G. S. Robinson. 2009. Pleistocene Martin, P. S. 1967. Prehistoric overkill, p. 75–120. In P. megafaunal collapse, novel plant communities, S. Martin and H. E. Wright, Jr (eds.), Pleistocene and enhanced fire regimes in North America. Extinctions: The Search for a Cause. Yale Science, 326:1100–1103. University Press, New Haven. Goldberg, A., A. M. Mychajliw, D. Sztraicher, and E. Martin, P. S. 1973. The discovery of America. Science, A. Hadly. 2015. Evidence for the peopling of 179:969–974. South America: archeological and genetic Martin, P. S. 1984. Prehistoric overkill: the global perspectives, 84th Annual Meeting of the model, p. 354–403. In P. S. Martin and R. G. Klein American Association of Physical Anthropology. (eds.), Quaternary Extinctions: A Prehistoric American Association of Physical Anthropology, Revolution. Press, Tucson. St. Louis, Missouri. Martin, P. S. 1990. 40,000 years of extinction on the Graham, R. W., and E. L. Lundelius, Jr. 1984. "planet of doom". Palaeogeography, Coevolutionary disequilibrium and Pleistocene Palaeoclimatology, Palaeoecology, 82:187–201. extinction, p. 223–249. In P. S. Martin and R. G. Martin, P. S., and D. W. Steadman. 1999. Prehistoric Klein (eds.), Quaternary Extinctions: A Prehistoric extinctions on islands and continents, p. 17–55. In Revolution. University of Arizona Press, Tucson. R. D. E. MacPhee (ed.), Extinctions in Near Time: Grayson, D. K. 1984. Archaeological associations with Causes, Contexts, and Consequences. Kluwer extinct Pleistocene in North America. Academic/Plenum Publishers, New York. Journal of Archaeological Science, 11:213–221. McInerny, G. J., D. L. Roberts, A. J. Davy, and P. J. Grayson, D. K. 2001. Did human hunting cause mass Cribb. 2006. Significance of sighting rate in extinction? Science, 294:1459. inferring extinction and threat. Conservation Grayson, D. K., and D. J. Meltzer. 2003. A requiem for Biology, 20:562e567. North American overkill. Journal of Mosimann, J. E., and P. S. Martin. 1975. Simulating Archaeological Science, 30:585–593. overkill by Paleoindians. American Scientist, Guilday, J. E. 1967. Differential extinction during late- 63:304–313. Pleistocene and Recent times, p. 121–140. In P. S. Roberts, R. G., T. F. Flannery, L. K. Ayliffe, H. Martin and H. E. Wright, Jr (eds.), Pleistocene Yoshida, J. M. Olley, G. J. Prideaux, G. M. Laslett, Extinctions: The Search for a Cause. Yale A. Baynes, M. A. Smith, R. Jones, and B. L. University Press, New Haven. Smith. 2001. New ages for the last Australian Guilday, J. E. 1984. Pleistocene extinction and megafauna: Continent-wide extinction about environmental change: case study of the 46,000 years ago. Science, 292:1888–1892. Appalachians, p. 250–258. In P. S. Martin and R. Saltré, F. É., B. W. Brook, M. Rodríguez-Rey, A. G. Klein (eds.), Quaternary Extinctions: A Cooper, C. N. Johnson, C. S. M. Turney, and C. J. Prehistoric Revolution. University of Arizona A. Bradshaw. 2015. Uncertainties in specimen Press, Tucson. dates constrain the choice of statistical method to Guthrie, R. D. 2003. Rapid body size decline in infer extinction time. Quaternary Science Alaskan Pleistocene horses before extinction. Reviews, 112:128–137.

19 MARSHALL ET AL.: QUANTIFYING CAUSES OF MASS EXTINCTION

Strauss, D., and P. M. Sadler. 1989. Classical APPENDIX confidence intervals and Bayesian probability estimates for ends of local taxon ranges. R CODE FOR RUNNING NON-LINEAR LEAST Mathematical Geology, 21:411–427. SQUARES ANALYSES Stuart, A. J. 1999. Late Pleistocene megafaunal extinctions: a European perspective, p. 257–270. This is the simplest possible code for applying In R. D. E. MacPhee (ed.), Extinctions in Near equation (1) to the extinction, human impacts, and Time: Causes, Contexts, and Consequences. climate change data. More sophisticated code might Kluwer Academic/Plenum Publishers, New York. read in multiple data sets from an excel spreadsheet, Stuart, A. J., P. A. Kosintsev, T. F. G. Higham, and A. for example, and analyze and organize the output into a M. Lister. 2004. Pleistocene to Holocene single file all in one step. extinction dynamics in giant and woolly . Nature, 431:684–690. Step 1: Enter data directly into R. The data below are Stuart, A. J., L. D. Sulerzhitsky, L. A. Orlova, Y. V. those shown in Table 1 for extinction scenario E(1). Kuzmin, and A. M. Lister. 2002. The latest woolly For convenience, ΔC is designated C and ΔH is (Mammuthus primigenius designated H. Blumenbach) in and : a review of the current evidence. Quaternary Science Reviews, > E <- c(0,0,1,0,0,2,2,0,0,0) 21(14–15):1559–1569. > C <- c(0,0,0,0,0.1,0.25,0.6,0,0,0) Trueman, C. N. G., J. H. Field, J. Dortch, B. Charles, > H <- c(0,0,0,0,0,0.2,0.4,0,0,0) and S. Wroe. 2005. Prolonged coexistence of Step 2: Combine the three vectors into a data frame humans and megafauna in Pleistocene . called ‘d’. Proceedings of the USA National Academy of Sciences, 102:8381–8385. > d <- data.frame(E, C, H) Turvey, S. T. 2009. Holocene Extinctions. Oxford University Press, Oxford, 364 p. Step 3: Print the data frame to the screen to make sure Villavicencio, N. A., E. L. Lindsey, F. M. Martin, L. A. that it has been entered properly by typing ‘d’. Borrero, P. I. Moreno, C. R. Marshall, and A. D. Barnosky. 2015. Combination of humans, climate, > d and vegetation change triggered Late Quaternary megafauna extinction in the Última Esperanza Step 4: Run the non-linear least squares. Values of the region, southern Patagonia, Chile. Ecography, initial estimates of a, b, and c are here set to 0.5, but 38:1–16. the values chosen typically make no difference. Wroe, S., J. Field, and D. K. Grayson. 2006. Megafaunal extinction: climate, humans and > fit <- nls(E~a*C+b*H+c*C*H, assumptions. Trends in Ecology and Evolution, start=list(a=0.5, b=0.5, c=0.5), data=d) 21:61–62. Wroe, S., J. H. Field, M. Archera, D. K. Grayson, G. J. If you get the error “‘Error in nlsModel(formula, mf, start, wts): singular gradient matrix at initial parameter Price, Julien Louys, J. T. Faith, G. E. Webb, I. estimates”’ it most likely means that your data are not Davidson, and S. D. Mooney. 2013. Climate sufficient to estimate the parameter values. change frames debate over the extinction of megafauna in Sahul (Pleistocene Australia–New Step 5: View the results. Guinea). Proceedings of the USA National Academy of Sciences, 110(22):8777–8781. > summary(fit) Young, H. S., R. Dirzo, K. M. Helgen, D. J. Mccauley, S. A. Billeter, M. Y. Kosoy, L. M. Osikowicz, D. J. You should get a table that provides the values, Salkelde, T. P. Young, and K. Dittmarh. 2014. standard errors, t values and significance levels, Pr(>| Declines in large wildlife increase landscape-level t|), for a, b, and c. For the data above the values are: a prevalence of rodent-borne disease in . = –1.863e-09 + 3.78, t = 0.000, p = 1.00000; b = 13.57 Proceedings of the USA National Academy of + 3.309, t = 4.101, p = 0.00457 **; c = –14.29 + 6.037, Sciences, 111(19):7036–7041. t = –2.366, p = 0.04987*. Young, H. S., D. J. Mccauley, K. M. Helgen, J. R. Goheen, E. Otárola-Castillo, T. M. Palmer, R. M. When you change the data (by repeating Step 1 for any Pringle, T. P. Young, and R. Dirzo. 2013. Effects or all of the vectors) it is easy to forget to update the of mammalian herbivore declines on plant data frame (Step 2) – Step 3 is used to make sure the communities: observations and experiments in an data frame has been updated. African savanna. Journal of Ecology, 101:1030– 1041.

20