In this paper, we use radiocarbon frequency data from the Atlantic Forest of eastern Brazil to develop an empirically grounded archaeological test for resistance-resilience, focused on the period of the Medieval Climate Anomaly (MCA). The MCA is a multi-centennial period of anomalous warming between ∼1,000 and 700 cal BP (AD 950–1250), originally identified in northern hemisphere palaeotemperature proxies (Bradley et al., 2003; Mann et al., 2009). Narrowing data gaps across tropical South America have found support for its expression as a period of abrupt hydroclimate change over central and eastern Brazil too, with precipitation proxies suggesting depressed rainfall due to a northward shift of the Intertropical Convergence Zone (ITCZ) and lowered moisture delivery across the continent (Novello et al., 2018; Deininger et al., 2019; Lüning et al., 2019). Following the MCA, the Little Ice Age (LIA) manifests in the northern hemisphere as a period of cooling, which in South America shows a signal that varies from wet to dry, depending on the position within the path of the South American Monsoon System (SAMS). Of particular interest in the Atlantic Forest biome is a dry-wet dipole to the northeast and southwest of the South Atlantic Convergence Zone (SACZ), respectively (Novello et al., 2018; Utida et al., 2020). Records from the core of the SACZ, which lies astride the central Atlantic Forest, show a strong multidecadal to centennial-scale variability at around ∼900–500 cal BP, on the cusp of the transition between MCA and LIA (Novello et al., 2018). We expand the scope of recent palaeodemographic modelling work in southern Brazil (De Souza and Riris, 2021) and neighbouring regions of South America (Azevedo et al., 2019; De Souza et al., 2019) to encompass the entire Atlantic Forest biome across eastern Brazil and evaluate the resilience of pre-Columbian populations to known climate perturbations that affected this domain.

The Atlantic Forest, originally covering 1.5 million km² over a wide latitudinal and elevation range, is recognised as a global biodiversity hotspot, composed of evergreen forests, dry forests, mangroves, and other vegetation formations with high endemism. It is also one of the most endangered Neotropical biomes, with only ca. 12% of its former extent remaining at most (Ribeiro et al., 2011; Marques et al., 2021). In addition, the Atlantic Forest has a long history of human-environmental interaction. At the time of European contact, most of the coastal strip and hinterland of the La Plata basin was inhabited by Tupi-Guarani societies, who practised polyculture and management of secondary forests, as attested by ethnohistorical and archaeological evidence (Scheel-Ybert et al., 2014; Bonomo et al., 2015). This follows millennia of landscape modification by marine-adapted shell mound (sambaqui) societies, who practised a mixed economy with forest management and some cultivation (Scheel-Ybert and Boyadjian, 2020). The Tupi-Guarani presence on the coast also overlaps with the arrival of the Aratu tradition, who occupied large circular villages in central Brazil, but whose expansion towards and impact on the Atlantic Forest are still poorly understood (Robrahn-González, 1996). At the same time, the southern Brazilian highlands in particular are increasingly recognised as an anthropogenic landscape, where a peak in human occupation (by the Southern Jê/Taquara-Itararé Tradition) coincides with the expansion of Araucaria forests—one of the vegetation types in the Atlantic Forest biome (Iriarte and Behling, 2007; Robinson et al., 2018; Pereira Cruz et al., 2020)—and the emergence of a mixed agroforestry/horticultural subsistence pattern (Corteletti et al., 2015).

The growing alignment of archaeological and palaeoecological datasets in tropical South America (Mayle and Iriarte, 2014; Iriarte et al., 2020) has helped to underscore the importance of forests to Indigenous subsistence. In what has come to be termed “polyculture agroforestry systems,” a subsistence base made up of an extensive mix of cultivated, domesticated, and wild sources (Arroyo-Kalin, 2012; Clement et al., 2015; Roberts et al., 2017; Maezumi et al., 2018; De Souza et al., 2019; Iriarte et al., 2020) with a substantial input of aquatic resources (Albuquerque and Lucena, 1991; De Borges Franco, 1998; Wüst and Barreto, 1999; Hermenegildo et al., 2017; Azevedo et al., 2019; Prestes-Carneiro et al., 2019), pre-Columbian groups are inferred to have been more resilient when compared to contemporary intensive land use systems (De Souza et al., 2019). Furthermore, humid/arid cycles are a known driver of Atlantic Forest composition and extent (Scarano and Ceotto, 2015; Ledru et al., 2016; Barros Silveira et al., 2019). Over the MCA and early LIA period, archaeological data in Amazonia and the central Brazilian Cerrado evidence heightened levels of conflict and site abandonment (Azevedo et al., 2019; De Souza et al., 2019). While the heterogeneity of late prehistoric palaeoecological trends cautions against simple mechanistic models of culture change (Bush et al., 2021), it is worth noting that by 750 cal BP population growth in neighbouring Amazonia had likely, at best, stabilised (Arroyo-Kalin and Riris, 2021). We suggest that the reliability and availability of resources under polyculture agroforestry systems may also have been affected by hydroclimatic perturbations and propose this as a mechanism through which abrupt climate change impacted pre-Columbian population trajectories in the Atlantic Forest. Our working hypothesis is that the expression of the MCA across the Atlantic Forest adversely impacted the resource base of pre-Columbian Indigenous groups practicing polyculture agroforestry, which we anticipate is reflected in proxies for ancient population change over time.

To evaluate this hypothesis, we present a novel test of resistance-resilience (Nimmo et al., 2015) specifically developed for summed probability distributions of calibrated radiocarbon dates (SPDs). SPDs are one among several methods that assume the relative abundance of radiocarbon dates will reflect broad trends in “events” over time, as a proxy for relative population change in the past (Shennan et al., 2013; Timpson et al., 2014; Crema and Bevan, 2021). SPDs are, at present, one of the most common means of aggregating radiocarbon dates. Due to this, the study of population dynamics in the past—palaeodemography (French et al., 2021)—is by necessity performed within one or several statistical frameworks to account for the chronometric uncertainties, biases, and sampling problems that are inherent to such data. A major goal of these approaches has been to provide robust alternatives to visual interpretations of SPDs, either through alternative summation methods or the development of statistical models to control for diverse sources of error (Timpson et al., 2014; Crema et al., 2016; Bronk Ramsey, 2017; McLaughlin, 2019; Carleton and Groucutt, 2021). We build upon one of the most widespread such frameworks, available in the R package rcarbon (Crema and Bevan, 2021), to implement our approach towards resistance-resilience in the Atlantic Forest biome of eastern Brazil. It affords the ability to control for variation in sampling intensities through binning, mitigation of calibration curve effects on SPD shape, and direct statistical comparison of multiple SPDs concurrently through mark permutation tests (Crema and Bevan, 2021). Although other readily available approaches, such as kernel density estimates or event count modelling (Brown, 2017; Carleton, 2021) are also theoretically amenable to the treatment we propose here, they do not currently have the same range of model-testing functionality that rcarbon offers. We also scaffold our approach with Bayesian model-fitting methods (Crema and Shoda, 2021) to build a palaeodemographic context for the present study that iterates on our prior work in the region (De Souza and Riris, 2021).

Our aims are to: (1) develop and apply palaeodemographic tests for resistance-resilience, which we define below, to enable the systematisation of knowledge on this crucial topic, and (2) demonstrate their utility for statistically comparing responses to disturbances between different regions of the Atlantic Forest while evaluating their implications. Before presenting the methods, we draw on recent reviews of resilience in archaeological research (Bradtmöller et al., 2017; Fitzhugh et al., 2019; Degroot et al., 2021; Russo and Brainerd, 2021) to contextualise our method and outline the scope of palaeodemography’s potential contribution to the study of resilience in ancient socio-environmental systems. More broadly, we situate this work within the study of “extreme events” in the deep past, bearing in mind the absence of consensus definitions for extremity (Intergovernmental Panel on Climate Change [IPCC], 2012; Gregory et al., 2015), and turn our attention to a systems approach that is focused on empirical benchmarking of culture-climate links, through the lens of palaeodemography (Broska et al., 2020).

Archaeological approaches to resilience are greatly influenced by the work of Gunderson and Holling (2002) on panarchy and adaptive cycling (Redman, 2005; Fisher et al., 2009; Kintigh et al., 2014). Recent reviews indicate that many, if not most, archaeological studies of ancient resilience draw on Resilience Theory (RT) in this sense (Holling, 1986; Folke, 2006; Bradtmöller et al., 2017; Fitzhugh et al., 2019). Within this body of work, adaptive cycling describes socio-environmental systems consisting of repeating phases of growth, conservation, release, and reorganisation in a domain of interest, which can be subsumed within (or themselves contain) cycles that operate across different spatio-temporal scales (Gunderson and Holling, 2002; Middleton, 2017). Due to the challenges associated with identifying appropriate scales and resolving different phases of adaptive cycling in the material record, studies of prehistoric resilience tend to adopt alternative strategies. Methods found within the archaeological resilience literature range from making qualitative point estimates or descriptions of conditions before and after a disturbance, to visual cross-referencing different proxies or time series (“wiggle matching,” cf. Bronk Ramsey et al., 2001; Blockley et al., 2018), as well as more rigorous quantification of relationships between different elements of a socio-ecological system (Kelly et al., 2013; Kintigh and Ingram, 2018; Carleton and Collard, 2020; DiNapoli et al., 2021).

In the vast majority of cases, resilience is employed largely as an interpretative device (Bradtmöller et al., 2017, p. 4), and consequently, the recognition of resilience (or vulnerability, stability, or fragility) in the archaeological record rests largely upon qualitative criteria (Degroot et al., 2021). Although this may offer internally consistent perspectives on how a socio-environmental system changed in response to disturbances, the findings of any single case study are rarely meaningfully comparable with those of another. Similarly, palaeodemographic studies tend to employ resilience heuristically (Leppard, 2015; Marsh, 2016; Bevan et al., 2017; Blockley et al., 2018; Riris and Arroyo-Kalin, 2019; Walsh et al., 2019; Palmisano et al., 2021), as opposed to a measurable property of a socio-ecological system like diversity, richness, or stability (Cantarello et al., 2017; Van Meerbeek et al., 2021), including in the Neotropics (De Souza et al., 2019; Ebert et al., 2019; Arroyo-Kalin and Riris, 2021). As the systematisation of archaeological knowledge been suggested to be key prerequisite for building coherent large-scale models of feedbacks between culture and climate (Kintigh et al., 2014; Perreault, 2019), we suggest there is a need to redefine resilience analytically and thereby overcome some of these limitations. At present, there are clear obstacles to this goal that are inherent to interpretative approaches.

To do so, we draw on approaches to the study of resilience from quantitative ecology, which has some noteworthy differences from the use of Resilience Theory in archaeology as described above. Although RT as high-level theory (sensu Gunderson and Holling, 2002; Folke, 2006) is similarly prominent in the ecological literature (Newton, 2021; Van Meerbeek et al., 2021), one area of vigorous research that has gone largely unappreciated in archaeological circles is its quantification, measurement, and comparison between different cases and in response to different (or common) disturbances. This drive to develop transferable metrics of system resilience (Orwin and Wardle, 2004; Pimm et al., 2019) does not have an identifiable counterpart in our field and stands in contrast to its almost exclusive use as an interpretative heuristic to frame archaeological problems (Redman, 2005; Bradtmöller et al., 2017; Table 1). Robust debate surrounds the definition of resilience metrics and the different elements of a socio-ecological system they purport to measure (Ingrisch and Bahn, 2018; Pimm et al., 2019; Van Meerbeek et al., 2021), however, it is clear that building understandings of resilience from concise operational definitions has afforded key insights into long-term biodiversity trends, informed conservation agendas, aided in the communication of scientific results, and provided a platform for ecologists to bridge the gap with stakeholders (Nimmo et al., 2015). We follow others in contending that research on prehistoric societies has a contribution to make to all of these research areas or closely analogous ones (Kintigh et al., 2014). While there is clearly an appetite on the part of archaeologists to adopt terminology across disciplines, the field has to date largely avoided the formalism necessary to obtain meaningfully comparable data on resilience (Olsson et al., 2015) that would enable such impacts to be made.

We suggest that radiocarbon-based estimates of palaeodemographic dynamics are exceptionally well-positioned to provide solutions, as they provide an absolute chronological framework for assessing fluctuations that can be cross-referenced to known perturbations and couched within a hypothesis-testing framework. As noted, climate change often figures as an important, high-level driver of prehistoric population turnover and cultural transitions, whether as a source of unpredictability (high variance), extreme (pulse) events, gradual pressure, or other mechanism (Bevan et al., 2017; Blockley et al., 2018; De Souza et al., 2019; Ebert et al., 2019; Riris and Arroyo-Kalin, 2019; Walsh et al., 2019; Palmisano et al., 2021). We now investigate whether the transient, centennial-scale effects of the MCA, in the form of a weakened South American Summer Monsoon, are reflected in persistent changes to palaeodemographic growth rates (expressed as a percent rate of change) across the Atlantic Forest, as in parts of Amazonia (De Souza et al., 2019).

Our findings affirm that a delayed demographic transition occurs in the Atlantic Forest relative to the Amazon basin following the adoption and development of sedentary agroforestry (De Souza and Riris, 2021). In contrast, in the Amazon, centres of local domestication and early adoption of exotic crops date to the early to mid-Holocene (Watling et al., 2018; Lombardo et al., 2020). Due to this, it is possible that the ability to reach carrying capacity would have been attained earlier in the Amazon than in the Atlantic Forest. Although we detected minimal differences between the logistic and exponential model fit (Supplementary Figures 2, 3), an exponential model of growth, besides being more parsimonious, is compatible with the cultural trajectories of the Atlantic Forest against those of the Amazon over the period of interest (Arroyo-Kalin and Riris, 2021). Before continuing, we underscore the persistence of Indigenous populations over our period of study. In the context of the last 1,500 years before the Conquest, Indigenous people were still experiencing unabated exponential growth (Figure 4). The relatively late occupation of the Atlantic Forest by farming populations must, however, be considered within its changing environmental context. The best example is the dispersal of the Tupi-Guarani, extremely rapid to the point where determining the precise chronology and direction of spread becomes unfeasible (Riris and Silva, 2021). Events of accelerated demic expansion, perhaps due to the migrant societies finding a new or relatively unexplored niche, may have contributed to the observed exponential growth, suggesting the Atlantic Forest societies were far from reaching carrying capacity at the eve of the Conquest. Furthermore, the role of late Holocene climate-driven forest expansion in the Atlantic Forest biome has to be considered (Smith and Mayle, 2017). As in other global cases of the expansion of agriculturists (Russell et al., 2014), the increase in the areas that could be exploited by polyculture agroforestry is likely to have facilitated the Tupi-Guarani dispersal (Iriarte et al., 2017). The expansion and contraction of certain niches would have influenced the dispersal and growth not only of the Tupi-Guarani, but also of Jê societies, whose subsistence relied on Araucaria forests in southern Brazil (Iriarte and Behling, 2007; Pereira Cruz et al., 2020).

Bearing these points in mind, we note that low resistance metrics for the Northern region, coupled with comparatively higher resilience observed in the centre and south, appear at first to correlate with the latitudinal gradient in the signal of the MCA observed in eastern Brazilian speleothems, which show a stronger deviation north of the SACZ (Novello et al., 2018). The speleothems at the core region of the SACZ, corresponding to our central sites, exhibit increased climate variability at the MCA-LIA transition (Novello et al., 2018), which may explain the decline in growth rates at the end of the MCA period (Figure 5). The northeast region of Brazil is expected to be the most sensitive to hydrological changes; climate simulations have confirmed its susceptibility to biome shifts when compared to the southern and central parts of the Atlantic Forest domain (Ledru et al., 2016; Costa et al., 2018). Relative biome (in)stability may therefore be an important factor in mediating the resistance and/or resilience of pre-Columbian populations to abrupt change, to a greater degree than absolute states such as “open” or “closed” vegetation (alternatively, forest or grassland). Quantifying tropical ecosystem stability—and hence the viability of different subsistence strategies in terms of resistance-resilience—against the backdrop of palaeodemographic dynamics is likely a useful avenue for future work. The main factor behind the observed north-south trend in resistance may be hydroclimatic regime variability, as the southern region receives precipitation from southerly, extra-tropical, oceanic sources in winter (Cruz et al., 2005; Campos et al., 2019). This is in agreement with the observation that, in the early to mid-Holocene transition (∼8,200 cal BP), the impacts of precipitation variability were lessened in more climatically stable regions, such as the southern part of South America (Riris and Arroyo-Kalin, 2019). Climate therefore seems a likely candidate for the main driver of the observed differences in resistance-resilience in our study area. Consequently, we reject the null hypothesis of no change to palaeodemographic growth rates across the Atlantic Forest during the MCA.

The different responses of Atlantic Forest societies during the MCA provides additional context for the expectations posed by De Souza et al. (2019), who argued in central-eastern Amazonia that polyculture agroforestry should provide a more resilient economic basis in the face of climate change than specialised, intensive land use systems. By separating palaeodemographic responses into the components of resistance and resilience, we are able to qualify this inference; the capacity to rebound following abrupt disturbances appear to be common across the agroforestry systems under study (resilience p > 0.05 in all cases), but responses to initial impacts (notably its timing and relative magnitude) appears to depend on local circumstances to a greater degree. In a topographically and biogeographically diverse environment, such as the Atlantic Forest, this is a relatively intuitive finding. Our findings parallel ecological resilience studies, where multiple metrics are found to be more capable of revealing different facets of the systems under study than single indices (Cantarello et al., 2017). Comparative studies will likely help to add further nuance, as well as better understandings of the components (species, cultivars, land use systems) that made up different forms of polyculture agroforestry in the past, which certainly had considerable internal variation. To this end, in addition to climatic/vegetation variability within the Atlantic Forest range, socio-cultural factors could be behind some of the observed differences in resistance-resilience. The available data suggest that the societies throughout the studied area practised similar forms of polyculture agroforestry, albeit in different niches (De Souza et al., 2020; Pereira Cruz et al., 2020), yet multiple distinct forms of social organisation existed prior to Conquest. Recent studies have emphasised pre-Columbian monumentality in southern Brazil as an indicator of territoriality and inter-group competition (De Souza et al., 2016), however, the emergence and maintenance of corporate architectural traditions can also be viewed as evidence and facilitators of strengthened intra-group cohesion (Iriarte et al., 2013). A possible consequence of more explicitly manifested and formalised social structures includes an ability to coordinate and manage resources above the kin level. Although we have focused on MCA climate as a driver of palaeodemographic impacts, we suggest that future work should consider quantitative approaches towards social and environmental factors as mediators of these impacts, and thus, investigate their possible role as facilitators of demographic and societal resilience.

We have focused here on the potentially deleterious effects of abrupt hydroclimate variability on population growth, highlighted previously elsewhere in lowland South America (Azevedo et al., 2019; Riris and Arroyo-Kalin, 2019). Our approach expects that the system(s) under study display resistance and resilience if baseline conditions are maintained or are returned to, respectively. There are, however, other scenarios for which our tests may be suitable, including inverse to those presented here. Returning to the indices as illustrated in Figure 3, one can also imagine “perturbations” to cultural systems such as the invention and spread of a technological innovation that raises growth rates over an interval of time. This hypothetical scenario would also result in values of <1 being returned in our metrics. Such a result is not because a given cultural system or group was (necessarily) adversely affected by rising growth rates, but because the nature of the system has measurably changed by the adoption of an innovation, reflected in its “low” resilience. The equations, and by extension our tests, are agnostic about whether change is a net positive or a net negative. We feel it is important to underscore this point, as it is contrary to commonplace understandings of resilience in the archaeological literature, which tends to interpret signs of “improvement” (i.e., increasing growth rates) as “high” resilience (Bradtmöller et al., 2017). Another, perhaps more intuitive, way to view resilience in this sense is that the prior cultural system was vulnerable to the introduction of an innovation, and so “collapsed” into a novel regime incorporating it. As always, caution must be exercised when deploying metrics; their broader societal and, in this case, palaeodemographic contexts are important for enabling considered interpretations of their meanings. The equations and indices make no distinction about the significance of low or high resistance-resilience values—careful consideration of the cultural system as a whole must accompany interpretation. Our aim is to encourage careful use in future applications, as the benefits of systematising the archaeological study of societal resilience are potentially great: comparability, communicability of results, rigour, and facilitating interdisciplinarity and impact.

This paper has sought to develop novel methods for the archaeological study of resilience in prehistory, a topic of significant concern (Bradtmöller et al., 2017; Russo and Brainerd, 2021). We have, both previously and separately, argued for the systematisation of the study of resilience in palaeodemography (De Souza et al., 2019; Riris and Arroyo-Kalin, 2019), in parallel to similar calls elsewhere (Bird et al., 2020). We are motivated by the need to make the archaeological study of resilience more rigorous and transparent, an essential part towards building interdisciplinary dialogue (Moser et al., 2019, p. 37). Here, we have drawn on mature research in quantitative ecology to propose a common framework for measurement of resilience in archaeology. Our method, which takes advantage of probabilistic, continuous time series data, can also be viewed as an answer to collapse-centric narratives on past societal responses to abrupt change (Middleton, 2017, p. 81; Koch et al., 2019; Degroot et al., 2021). We argue that the approach encourages attention to particular circumstances in archaeological and historical perspective, while also enabling the formalism that typifies the ecological resilience literature, and which we regard as a prerequisite to developing a comparative, deep time perspective on coupled socio-environmental changes and their consequences. We have demonstrated this by testing palaeodemographic responses to disturbances across and between biogeographically relevant zones of the South American Atlantic Forest during the Medieval Warm Period to explore the implications of hydroclimatic variability for the pre-Columbian societies of this biome. The resulting analysis of longitudinal and cross-sectional trends in palaeodemography reveal notable regional differences.

It remains hard to argue that there is a single “correct” way of viewing resilience (Middleton, 2017). Nonetheless, building a perspective that by its nature is rooted in probabilistic evidence of past population changes, itself is based on the ubiquitous use of archaeological radiocarbon dates, certainly has advantages that other proxies do not. Notably, our approach provides explicitly defined, quantitative, and reproducible alternatives to interpretative approaches, which accompany this paper for future use and refinement (Supplementary Material). We acknowledge that point estimates of resistance and/or resilience are likely imperfect reflections of a more complex picture in the past and that there will always be alternatives to the equations and tests we have implemented. Even so, we suggest that directly comparable data on prehistoric resistance-resilience provide a more useful point of entry into identifying what factors (behaviours, adaptations, institutions, ecologies) mediate the severity of disturbance impacts than incomparable ones, as well as whether trade-offs exist between different factors in terms of robustness and fragility. These issues already form the focus of substantive research (Peregrine, 2020), and the present work charts a path through them by focusing on measurable features of ancient population dynamics. Furthermore, our approach enables the archaeological study of resilience to locate commonalities of language with the ecological resilience literature.

Our method therefore paves the way for the systematic and rigorous evaluation of inferred disturbances to human societies and culture-climate links in general. We suggest that other well-known phases of abrupt climate change, for example the Younger Dryas/Bølling-Allerød, the Pleistocene/Holocene transition, the 8.2 and 4.2 ka events, and data permitting, the “Columbian Exchange” (Hamilton et al., 2021) would be ideal test beds for the approaches we advocate. While attention here has mainly been on climate as a driver, abrupt social change or reorganisation may be equally important, and interesting, targets for comparative analysis (such as periods of warfare, Edinborough et al., 2017). Archaeological palaeodemography is also uniquely positioned to offer enhanced estimates of effective population size and density, which are key to understanding the deep time earth systems impacts of humanity more comprehensively (Klein Goldewijk et al., 2017; French et al., 2021; Morrison et al., 2021). Finally, we anticipate that continuing analytical refinements in Bayesian inference (Carleton, 2021; Crema and Shoda, 2021) as well as hereto unexplored methods such as symbolic transfer entropy (Camacho et al., 2021) will open up the study of causal links between population resilience and induced disturbances.

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding author.

PR and JS: conceptualisation and writing. PR: analysis and figures. Both authors contributed to the article and approved the submitted version.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.