Frequent disturbances enhanced the resilience of past human populations

Understanding the range of past responses of human societies to disturbances is a global priority and can inform solutions to future crises. Despite many case studies of cultural collapse, transformation and persistence, comparability between cases of resilience remains limited and debated. Few studies explicitly model impacts, recovery and adaptation over long-term histories, and a frequent focus on extreme events risks overemphasizing short-term success while obscuring broader vulnerabilities. For example, the Greenland Norse’s shift to a narrow marine diet mitigated short-term crop failure risks but increased vulnerability to long-term cooling. This study establishes a global comparative approach to long-term resilience, measuring populations’ capacity to withstand change and their rate of recovery following disturbances using summed probability distributions (SPDs) of calibrated radiocarbon dates as a proxy for relative human activity or population levels. The research addresses two key questions: (1) how quickly do past populations recover after downturns; and (2) what factors mediate past resistance and resilience to downturns?

Prior work has emphasized historical case studies of collapse and resilience, but there is no standardized framework for cross-regional, long-term comparisons of population resilience. Debates persist on how best to characterize collapse versus transformation, and many studies lack formal modeling of impacts, recovery and adaptation or do not account for long-term baseline variability across cultural and environmental settings. A focus on extreme events may bias understanding toward local successes while neglecting systemic vulnerabilities. Archaeological demography has advanced methods to infer population trends from radiocarbon dates, and macroecology suggests that resilience patterns often resolve at centennial or longer timescales. Nevertheless, systematic comparative evidence across disturbance categories and land-use systems has been limited. This study builds on radiocarbon SPD approaches and recent formal tests of resistance-resilience in archaeological time series to provide a comparative, global-scale analysis.

The authors conducted a meta-analysis of 16 published regional studies that used archaeological radiocarbon data to reconstruct palaeodemographic trends. Inclusion criteria required: (1) evidence for statistically significant population downturns; (2) regional scope (not single-activity subsets); and (3) availability of radiocarbon datasets. Studies lacking any criterion were excluded, as were cases without reported downturns. Where necessary, datasets were augmented with entries from the People3k database to match original study areas. The sample spans the Arctic to the tropics and covers 30,000 years. Downturns were reproduced using a standardized protocol based on SPDs of calibrated radiocarbon dates and null-model expected growth trajectories. Population downturns were defined as intervals where observed SPDs fell significantly below expectations. For each of 154 downturns (median 8.5 per region), the authors extracted numerical metrics capturing severity, chronology and frequency, including: overall duration; time to SPD minimum; start and end dates of the downturn (Ta, Te); baseline SPD at start (b); minimum SPD during downturn (x); and SPD at the end (e). They computed resistance (depth of downturn relative to baseline, range 0–1) and resilience (rate of recovery relative to impact, range −1 to 1) using functions of b, x and e. They also recorded disturbance categories (environmental, cultural, mixed, unclear) and specific types (e.g., aridity, mobility, variability, colonialism, cooling), dominant land-use pattern (hunter-gatherer, maritime forager, low-level food production, agriculture, agropastoralism, mixed), and evidence of cultural change during or after downturns (boolean). To compare speeds of downturns across variable durations, time to minimum was normalized by duration to create a relative pace index (higher values indicate slower downturns). The cumulative number of downturns per region was normalized by duration (events per millennium) and log-transformed to address skewness, yielding the key predictor “frequency of downturns.” A suite of hierarchical linear mixed models tested associations between resistance, resilience, and candidate fixed effects (including frequency of downturns, relative pace, disturbance category, land use), with study region included as a random effect to account for geographical variability. Additional models treated the frequency of downturns itself as a response variable to test its associations with land use and disturbance type. Model selection used information criteria, and significance was assessed with two-sided tests.

• Across 154 recorded downturns, frequent exposure to disturbances is strongly associated with greater resistance (shallower downturns) and higher resilience (faster/closer recovery). Frequency of downturns was the only fixed variable consistently retained by model selection and showed the strongest relationship to both resistance and resilience (P < 0.001 for both).
• Effect sizes indicate a stronger influence on resistance (η² = 0.46, P < 0.001) than on resilience (η² = 0.29, P < 0.001), suggesting partially distinct processes governing withstanding versus recovering.
• Disturbance categories: environmental (n = 48; 31%), mixed cultural–environmental (n = 33; 21%), and unclear (n = 65) were common; cultural types were also represented. The most common specific disturbance types were aridity (n = 31, environmental) and mobility (n = 20, cultural).
• Resistance and resilience distributions: only about one-third of downturns had resistance values indicating a drop of more than 50% from pre-downturn activity (n = 53; 34.4%). Most downturns (n = 133; 86%) ended before fully returning to baseline SPD values, yet resilience was relatively high overall (median = 0.64; n = 154), with 40% (n = 63) reaching at least 90% of pre-downturn conditions by the end.
• No significant differences in resistance or resilience were detected across high-level disturbance categories (ANOVA: resistance F = 0.541, P = 0.65; resilience F = 0.04, P = 0.98; d.f. = 3; n = 154).
• Geography: resistance showed no systematic geographical effect except higher values in the Caribbean archipelago and the Italian peninsula. Resilience was significantly lower in the Central China Plains, the Caribbean archipelago and the Korean Peninsula compared with other regions; several downturns in these areas had negative resilience values (n = 11; 39% of their downturns), indicating SPD values exceeded start values by the end despite remaining below null expectations.
• Temporal scales: downturn durations were mostly decadal (≤50 years; n = 47; 30.5%) to centennial (100–500 years; n = 48; 31%), with a median duration of 98 years. Time to SPD minimum skewed toward decadal timescales; only one downturn (starting with the 8.2 ka event in the Near East) exceeded a millennium (2,070 years) to minimum.
• Land use mediates exposure: agricultural and agropastoral land-use systems were associated with significantly higher rates of downturns per millennium than low-level food production, marine foraging or mixed subsistence. Over the Holocene, the proportion of food-producing land use increased through time during downturns.
• Disturbance frequencies by type: mobility changes (median 2.26 events per millennium; n = 20) and high environmental variability (median 2.19; n = 17) had the highest rates; cooling had the lowest (median 0.761; n = 4). Regional extremes included the South African Greater Cape Floristic Region (median 2.41 per millennium; n = 17 over 9,950 years) and the Korean Peninsula (median 0.58; n = 3 over 4,000 years).
• Interpretation: The global shift toward food production likely increased populations’ exposure and vulnerability to downturns but also enhanced adaptive capacity via repeated exposure, producing higher resistance and resilience over time.

The analysis indicates a robust, positive association between the frequency of population downturns and both resistance and resilience across diverse prehistoric societies. This suggests a common, emergent mechanism whereby repeated exposure to disturbance fosters adaptive capacities, though the processes that underlie withstanding versus recovery may differ. The results align with disturbance ecology, where frequent disturbances can enhance long-term resilience of ecosystem services via localized declines. Land use importantly mediates exposure: food-producing systems (farming and herding) are associated with more frequent downturns. As food production became more prevalent during the Holocene, populations appear to have faced more frequent demographic stress, yet also developed higher capacity to resist and recover—illustrating trade-offs inherent in these socio-ecological systems. Historical examples (e.g., gradual depopulation in the Americas post-1492; protracted rural depopulation preceding Rome’s collapse) support multidecadal-to-centennial scales for such dynamics. Potential mechanisms include biased cultural transmission during downturns, opportunities for landscape learning, strengthening of knowledge networks, investments in infrastructure, enhanced social cohesion and technological innovation. However, increasing population thresholds via intensified land use can raise the risk of more severe collapses with diminishing returns. Outcomes likely depend on contingent factors such as the ecology and diversity of domesticates, political complexity, and settlement–environment configurations. Overall, the findings underscore that population downturns are not merely failures but may play a constructive role in long-term demographic resilience, helping sustain humanity’s long-run population growth via feedbacks among vulnerability, resistance and recovery.

This study provides a global, comparative framework for quantifying resistance and resilience of past human populations using radiocarbon SPD proxies across 30,000 years. The central finding is that more frequent downturns are associated with higher resistance and resilience, with land-use practices (especially agriculture and agropastoralism) linked to increased downturn frequency. Although environmental, cultural and mixed drivers vary widely, differences in resistance and resilience across these categories were not statistically significant. Downturns typically span multidecadal to centennial timescales, and full returns to pre-downturn baselines are often interrupted by subsequent events. The main contributions are: (1) establishing resistance and resilience metrics applicable across regions; (2) demonstrating frequency of downturns as the strongest predictor of both outcomes; and (3) identifying land use as a mediator of exposure. Future research should expand the dataset to increase explanatory power, develop multiscalar models that distinguish exogenous and endogenous processes, and assemble comparable evidence across disturbance categories to investigate causal pathways linking exposure frequency, land use and demographic resilience under anticipated increases in environmental variability.

• The meta-analytic approach trades fine-grained specificity for broad comparability; mechanisms are inferred from aggregated patterns rather than directly observed.
• SPDs are a proxy for relative population or activity and may be influenced by sampling intensity, preservation and research biases.
• The study includes only regions with reported significant downturns; cases without downturns were excluded, potentially biasing exposure frequency estimates.
• Disturbance drivers were sometimes classified as unclear, and causal attributions rely on synthesis of prior regional studies and expert interpretation.
• Mixed-effects models retain study region as a random effect due to geographical variability; unobserved regional factors may remain confounded.
• Sample size (n = 154 downturns across 16 regions) limits power for detecting differences across disturbance categories and complex interactions.
• The framework does not disentangle exogenous impacts from endogenous socio-cultural responses, nor does it model cascading effects explicitly.