Climate change, not human population growth, correlates with Late Quaternary megafauna declines in North America

The study examines the debated causes of Late Quaternary megafaunal extinctions in North America, contrasting anthropogenic overkill hypotheses (overhunting, blitzkrieg/sitzkrieg variants) with climate-driven explanations. The research question is whether through-time changes in megafauna population levels are better explained by human population dynamics, climate change, or a combination of both. The authors highlight limitations of using first/last appearance dates and commonly used summed probability density functions (SPDFs), which can conflate demographic signals with radiocarbon chronological uncertainty. They aim to apply a method that explicitly accounts for this uncertainty to test the correlation of megafauna declines with human populations and climate fluctuations, particularly across key events like the Bølling-Allerød and Younger Dryas.

Prior work includes overkill models proposed by Paul Martin and variants suggesting rapid or gradual human-driven extinctions, as well as arguments emphasizing climate and environmental changes at the end of the Pleistocene. SPDF-based studies have inferred trends such as pre-human declines, peaks during interstadials followed by declines with peatland spread, and mixed human–climate influences. However, substantial critiques note SPDF biases: taphonomic degradation, sampling adequacy, calibration artifacts, and the conflation of process variation with dating uncertainty. Simulation studies indicate SPDF regressions can lead to spurious correlations. Other lines of evidence (aDNA, plant and non-mammalian extinctions, body-size reductions, range shifts) suggest complex, climate-linked ecological restructuring with potentially limited human roles.

Data: The megafauna database (Broughton and Weitzel) includes 521 directly radiocarbon-dated individuals (contiguous U.S. and adjacent Canada). A chronologically cleaned subset (to reduce multiple dates per individual/event) yielded 432 dates for supplementary checks. The human proxy (15–10 ka) comprised 938 vetted anthropogenic radiocarbon dates (from CARD; cleaned to remove duplicates, anomalous results, non-anthropogenic contexts, and ambiguous megafauna-associated dates). Climate proxy used the NGRIP δ18O record: a 50-year smoothed series for primary analyses and an annually resolved series with explicit measurement and chronological uncertainty for extended analyses. A tephra-based Northern Hemisphere taphonomic proxy (Surovell et al.) was included as a covariate to control for time-dependent sample loss. Radiocarbon-dated Event Count (REC) modelling: The authors applied a Bayesian hierarchical REC regression framework that explicitly accounts for chronological uncertainty by generating Radiocarbon-dated Event Count Ensembles (RECEs). For each dated sample, a probable date was drawn from the 98% confidence region of its calibrated probability distribution; counts per year were tallied to form one plausible event-count sequence. Repeating this process generated an ensemble of alternate plausible sequences reflecting dating uncertainty. Each RECE member served as the dependent series in a Negative-Binomial REC regression, appropriate for count data with temporal spread induced by uncertainty. Chronological uncertainty in covariates (human REC, taphonomic proxy, and NGRIP climate series) was likewise incorporated by sampling probable sequences for each covariate. Model design: Parallel analyses included (1) megafauna vs. human proxy (humans-only), (2) megafauna vs. climate proxy (climate-only), and (3) megafauna vs. both human and climate proxies (humans+climate), always controlling for taphonomy. Response datasets were analyzed at three levels: all megafauna combined, by taxa (Equus sp., Mammuthus sp., Mammut americanum, Nothrotheriops shastensis, Smilodon fatalis), and regionally (Great Lakes: mammoth, mastodon; Southwest: mammoth, sloth). Primary analyses covered 15.0–11.7 ka; an extended climate analysis used annually resolved NGRIP and expanded the period to 20.0–10.0 ka. Significance was assessed by whether the posterior 95% credible interval for a regression coefficient excluded zero. Computational constraints led to subsampling: RECEs of 50 member sequences, and temporal decimation to every 10th year, shown previously to preserve millennial-scale patterns. Software: Analyses were conducted in R with Nimble for MCMC, and ggplot2/ggpubr for visualization; code and supplementary materials are provided by the authors.

- Across all models and scales (all megafauna, by taxon, and by region), the human population proxy coefficient (β_Humans) was not significantly different from zero after controlling for taphonomy, indicating no persistent through-time relationship between human and megafauna population levels from 15.0 to 11.7 ka. - The taphonomic proxy (β_TA) showed no consistent significant effect, suggesting no strong temporal taphonomic trend in megafauna sample frequency over the examined interval. - The climate proxy (NGRIP δ18O) consistently correlated with megafauna population size: climate-only models yielded positive regression coefficients with posterior means typically around ≥0.05 and often ~0.1 or higher; combined models (humans+climate) produced similar positive, non-zero estimates for β_Climate. - Extended analyses using the annually resolved NGRIP record with explicit uncertainty and a broader 20–10 ka span reproduced the positive, significant climate–megafauna correlation with similar effect sizes, affirming robustness. - Taxon timing nuances: declines in horse (Equus) and saber-tooth cat (Smilodon) populations predated final declines in mammoths and mastodons, with some declines aligning with the terminal Bølling-Allerød and onset of the Younger Dryas. - Overall, decreases in global temperature (lower δ18O) correlate with declines in megafauna population proxies; human population changes do not show a persistent effect.

The findings directly address the overkill vs. climate debate by showing no significant through-time correlation between human population proxies and megafauna population declines, contradicting simple overkill models centered on widespread overhunting by rapidly expanding human populations. In contrast, a consistent positive association between megafauna proxies and the NGRIP δ18O climate proxy indicates that cooler conditions correlate with lower megafauna counts, supporting climate-driven demographic fluctuations. The extended analysis confirms this relationship across a longer interval and with an unsmoothed, uncertainty-aware climate record, strengthening the inference. The results suggest that earlier mixed-cause conclusions derived from SPDF-based regressions may be artifacts of methodological limitations, as SPDFs confound process variation with chronological uncertainty. While the data do not rule out more complex human influences—such as disruption of metapopulation connectivity, trophic cascades via selective impacts on keystone herbivores, or altered predator–prey dynamics—the primary driver detectable in the aggregated temporal record is climate. The temporal coincidence of final megafauna declines with the onset of the Younger Dryas, characterized by abrupt cooling, increased seasonality, rising atmospheric CO2, and rapid vegetation shifts, is consistent with climate as a dominant factor. Regional analyses (Great Lakes and Southwest) indicate variability in local climate and vegetation responses, yet broadly align megafauna declines with the Bølling-Allerød/Younger Dryas transitions.

Using REC modelling that explicitly accounts for radiocarbon chronological uncertainty, the study finds no persistent through-time relationship between human population levels and North American megafauna population declines, while demonstrating a robust, positive correlation between megafauna counts and the NGRIP δ18O climate proxy. The results undermine simple overkill explanations and instead support climate-driven megafauna declines, with the Younger Dryas likely playing a key role. Methodologically, the paper advances the field by illustrating the pitfalls of SPDF-based analyses for extinction dynamics and by providing a more appropriate Bayesian framework for time-uncertain count data. Future research should: (1) expand high-quality, directly dated faunal and archaeological datasets across broader spatial and temporal scales; (2) integrate additional, regionally resolved climate and ecological proxies with uncertainty quantification; (3) refine event identification to minimize duplicate-dating of single individuals/events; (4) develop models incorporating species-specific ecology, metapopulation structure, and habitat connectivity; and (5) further test alternative human-impact hypotheses (e.g., fragmentation, keystone effects) within uncertainty-aware, mechanistic modelling frameworks.

- Data limitations: Despite large databases for North America, sample sizes remain modest relative to spatial and temporal scales, and biases persist (taphonomic loss, uneven sampling effort, calibration artifacts, spatial/temporal adequacy). - Chronological uncertainty: Although REC models account for uncertainty, computational constraints required using RECEs with 50 members and decadal subsampling, potentially underrepresenting tail uncertainty or high-frequency dynamics. - Proxy constraints: The primary climate proxy (NGRIP δ18O) is a broad Northern Hemisphere signal; regional climatic heterogeneity may not be fully captured. Even with the annually resolved record, uncertainties in alignment with regional ecological changes remain. - Human proxy quality: Anthropogenic radiocarbon data (CARD-derived) are vetted but may still include residual biases; the temporal and spatial resolution may not capture localized human-megafauna interactions. - Event identification: Potential multiple dates from single individuals/events necessitated a cleaned subset; although results were consistent, some undetected duplications or exclusions could persist.