Twenty-first century approaches to ancient problems: Climate and society

Abstract

By documenting how humans adapted to changes in their environment that are often much greater than those experienced in the instrumental record, archaeology provides our only deep-time laboratory for highlighting the circumstances under which humans managed or failed to find to adaptive solutions to changing climate, not just over a few generations but over the longue durée. Patterning between climate-mediated environmental change and change in human societies has, however, been murky because of low spatial and temporal resolution in available datasets, and because of failure to model the effects of climate change on local resources important to human societies. In this paper we review recent advances in computational modeling that, in conjunction with improving data, address these limitations. These advances include network analysis, niche and species distribution modeling, and agent-based modeling. These studies demonstrate the utility of deep-time modeling for calibrating our understanding of how climate is influencing societies today and may in the future.

The rise and collapse of North America’s largest and one of its most-studied prehispanic centers, Cahokia (Figs. 1 and 2), has recently been examined from the concurrent perspectives of dendroclimatology (1), palynology, geology, and archaeology (2). Researchers argue that the rapid increase in its population and political scale in the last half of the 11th century AD was likely facilitated by very moist conditions [reflected in positive Palmer Drought Severity Index (PDSI) values derived from tree-ring networks] conducive to high maize production. The 12th century, marking the beginning of a population decline in the central portions of Cahokia, was much more xeric, particularly in the middle and late 1100s. The hinge between these two centuries, ∼AD 1200, was marked by a major flood in the oxbow lake system adjacent to Cahokia. The construction of defensive palisades coincided with aridification pulses in the 1100s and early 1200s.

Fig. 1.

Central portions of Cahokia, southwestern Illinois, looking north across the Grand Plaza toward Monks Mound. Image courtesy of Ira Block/National Geographic Creative.

Fig. 2.

Locations and types of major studies discussed. SDM, species distribution modeling.

Satisfying as it is in many ways, this juxtaposition of Cahokian archaeological and paleoenvironmental data also illustrates the difficulty of our task. Is even the relatively high temporal precision in this case sufficient to draw convincing causal links between climatic events and settlement changes? How much did changing PDSIs affect production of maize and other comestibles in the area surrounding Cahokia, including the likely flows of tribute into the main center? What specific problems were created by the large-scale land clearance documented by pollen spectra in this area between AD 900 and 1200, and in exactly what ways would the ∼AD 1200 flood have affected the Cahokian polity? How widespread were these floods, and to what extent did they result from forest reduction? Such uncertainties cause some specialists to emphasize various social processes, such as the contribution of factionalism and discord brought on by very high levels of immigration, in Cahokia’s decline (3).

In general, attempts to accord climate variability an appropriate role in affecting the size, placement, social forms, and cultural practices of past populations are often beset by low temporal resolution and uneven spatial representation of proxies for climate variability and the available samples of the archaeological record itself. The Cahokia example demonstrates that even when these conditions are adequately met, two important problems remain: linking proxies for climate change to variability in resources most important to human populations, and constructing a dynamic causal framework linking variability in resources to ecosystem processes, on the one hand, and—on the other hand—to variability in demography, settlement behaviors, and social organization across the micro and macro levels.

The development of new, more abundant, and more precisely dated climate proxies, and much larger archaeological datasets (4, 5) can only take us so far, and this is why models are essential. Understanding natural-human systems at multiple scales requires the development of explicit process models that couple these datasets. Here we focus on a series of related computational advances to show how they are beginning to transform archaeology into a model-based science (6–8). We highlight particularly compelling model-based studies that serve as examples for future efforts. At their best these computational models are reproducible, data-driven, and data-hungry: the interpretations and expectations they generate become more precise as they are afforded more data. This is particularly important in archaeology, where both new and legacy datasets are routinely becoming available digitally but their integration into archaeological understanding is typically slow. Model-based archaeology provides an arena for integrating new data and for generating understandings on how humans may both cause—and respond to—climate change both in the past and the future.

Challenges in Linking Climate Variability to Variability in Resources Critical to Society

Placing ancient human sites into their local environmental and climatic contexts has been a core practice in archaeology for well over half a century (9, 10). The application of new tools, such as the geographic information system (GIS), during the 1980s enabled archaeologists and geographers to more easily ask questions, such as how site locations were influenced by their distance from resources and biomes, questions inherently difficult to answer with traditional methods. However, even with the development of GIS-based technology, modern climates and environments have been used to stand in for conditions at the time of occupation. In Central Asia, for example, work that uses the best-available resources for understanding biome distribution [Landsat (11), modern land use maps (12), or observations of current biota and their relationship to site placement (13)] take contemporary vegetation distributions as proxies. We know, however, that different climatic regimes substantially changed these conditions in the past (14).

On some fronts things are getting much better. In archaeology, the creation of large databases of chronological information allows for higher-resolution cross-referencing of key events. A compilation of all previously published radiocarbon dates from China, for example, suggests that warmer and wetter periods were correlated with population growth (assuming population is reflected in radiocarbon-date frequency), whereas populations declined in cooler and more arid periods (15). Similar correlations have been demonstrated for Holocene North Africa (16), where desertification following the end of the African Humid period ∼3000 BC profoundly decreased the density of human settlement outside the Nile valley. In Europe, date frequencies document booms and busts in Neolithic populations (those maintaining plant or animal domesticates) that do not clearly correlate with climatic variability (17).

Open-access, geo-referenced, and comprehensive paleoclimatic databases from well-funded global endeavors, including CLIMAP (Climate: Long-Range Investigation, Mapping, and Prediction) (18), COHMAP (Cooperative Holocine Mapping Project) (19), and most recently the PAGES 2k Consortium (20) are the envy of archaeology because of the level of coordination they demonstrate, but even the most recent of these deals primarily with climates at the continental level. This is a problem for archaeology, because all adaptation in the relatively small-scale societies we study is local. Data from pollen cores, increasingly available online through Neotoma (21) and the European Pollen Database (22), provide one pathway for making more localized climate-field reconstructions. Changes in pollen composition in cores are used most directly to reconstruct vegetation distribution, such as biomes (23), or degree of landscape openness or cover abundance of key plant taxa (24), although with additional processing (entailing additional uncertainties and assumptions) pollen-based reconstructions of variables such as growing-season warmth, winter cold, and plant-available moisture are possible (25, 26). Depending on the approach and the question, anthropogenic disturbance frequently visible in pollen spectra can be either annoying noise to be avoided or a focus of the analysis (27). Paleoclimatic indicators derived from archaeological sites will increasingly contribute to these reconstructions, realizing the potential of archaeological sites to serve as distributed observation networks of the past (28).

These developments have supplied archaeologists with increasingly accurate data to draw links between climate processes and changes in human population size, location, and subsistence. For the most part, however, such exercises have been correlative: changes in human subsistence patterns (29) or proxies for population size (17) are correlated with various climate proxies, or even just verbally juxtaposed (30). Generally, archaeologists have not elucidated how exactly changing climates affect what humans do. How, for example, did changing rainfall and temperature impact crop production at Cahokia and its supporting areas?

Archaeologists face two especially important challenges on this front. First, building accurate understandings of climatic impacts on human societies in the past requires paleoclimate reconstructions at spatial and temporal scales relevant to human experience (31). A major challenge thus lies in downscaling: translating regional or even global climatic records into local records. The second challenge archaeologists face is that it is not just (or even primarily) the climate state that archaeologists need to know, but the effects of that state on the physical resources (e.g., location of water table and shoreline) and the local vegetation and animal communities on which societies depended.

Some climate proxies do record highly local data about variations in climatic conditions, yet these are often extrapolated to explain social or cultural change far from the proxy’s origin. Speleothem records from Dongge, Jiuxian, and Sanbao caves in China, for example, are frequently used to infer changes in the East Asian Monsoon, which are then compared with human activity hundreds of kilometers away (32). Such extrapolations likely degrade the ability of archaeologists to model how changing local monsoonal intensities affected ancient humans living far from the location of the proxy record.

In addition, not every type of paleoclimate proxy can be readily translated into continuous variables, such as temperature and precipitation that can be modeled as climate fields. δ¹⁸O levels in speleothems, for example, are translated into periods when monsoonal activity was more (or less) intense, providing only a relative measure of precipitation or heat. Similar issues plague records such as lake-level proxies. Several types of proxy records can be translated more directly into temperature or precipitation series through correlation with instrumented conditions; these include tree-ring chronologies, marine and lacustrine alkenone series, ice cores, and annually banded marine corals. Tree-ring series, in particular, have the advantage of allowing estimates of annual and even seasonal records of temperature and precipitation if appropriate species are available. Annual records of PDSI derived from tree-ring records for the past two millennia are now available at half-degree spatial resolution for large portions of the world (33, 34). Sea-surface temperatures have also been a key data source for establishing global records (35). However, translating global records that rely heavily on sea-surface temperatures can be problematic for inland regions, as sea-surface temperatures react to climatic change at a lower amplitude than areas on land. With important exceptions, few high-resolution records of paleoclimate (such as tree-ring series) extend beyond the last 2,000 y, creating an additional challenge for archaeologists working in earlier periods.

A key focus for understanding the impact of climate on humans is now to create a record that reflects local climate at high temporal and spatial resolution across an entire landscape. This is challenging because each paleoclimate proxy represents just one point on a landscape that may be very heterogeneous, for example marked by high topographic relief creating temperature and precipitation variability. To be useful for archaeologists, point data need to be interpolated to create climate-field reconstructions, and coarse-resolution climate fields need to be downscaled to reflect local conditions.

Spatial interpolation is the process of estimating a climate variable at a given prediction point in space from measured or retrodicted data derived from one or more other calibration points. Generally, prediction points are within the spatial range (and statistical distribution) of the calibration points; when prediction points are located outside the calibration range or clearly beyond the statistical distribution of the calibration data, the estimate is considered an extrapolation. Often, estimates are made for many points arranged in a grid to create a climate-field reconstruction.

Downscaling refers to the process of estimating a finer-resolution climate field from a coarser-resolution source. Techniques fall into two primary categories (36): (i) statistical downscaling, in which empirical relationships are defined between a coarse climate field and local weather, or local surface variables, such as topography, coastlines, landcover, and water bodies; and (ii) dynamical downscaling, where fine-resolution climate models are run using the results from coarse-resolution models (such as general circulation models, GCMs) as boundary conditions. A number of resources for understanding downscaling exist (36–40).

General improvement in these techniques has allowed model-based archaeologists to move away from simple correlations between climatic variability and changes in human history and pay more attention to the causal pathways responsible for these correlations (41).

Modeling Climate’s Impact on Humans

Usually archaeologists are concerned with past environmental variability because of climatic effects on the distribution, abundance, and productivity of plants and animals on which past human societies depended, although in some cases the direct effects of natural disasters (earthquakes, floods, eruptions) are the focus (42, 43). We can distinguish between macroscopic approaches to this task (which may either exploit correlations between current or past regional landscapes and various species on which humans depend, or explain plant and animal distributions in terms of climatic components) and more disaggregated approaches that try to break socio-natural systems into their constituent components to model their connections and interactions through time and across space.

Macro Models.

Correlative modeling.

One of the most popular methods for understanding how different plant, animal, and human communities responded to changes in climatic conditions is species distribution modeling (44), also known as eco-cultural niche modeling (45). Franklin et al. (44) provide an excellent review of such models in archaeology. These methods take all known manifestations of a past culture or organism to derive its niche, usually using either GARP (Genetic Algorithm for Rule-Set Prediction) (46) or MaxEnt (maximum entropy) (47). Both methods use geo-referenced information about an organism’s current location that is correlated with a suite of abiotic factors, such as precipitation, temperature, soil conditions, or even the presence of other species (48) to predict the location of areas within the species’ realized niche for which observational data may be unavailable. Similar methods have been used to create maps of past biomes (49, 50). These models can then be used to predict where an organism could have been located on a past landscape.

MaxEnt and GARP have been used in archaeology to estimate past spatial extent of human populations (45, 51) and the animals they consumed (52, 53). For example, a study on the extent of the wild progenitor of maize (Teosinte sp.) over the last interglacial that used MaxEnt helped shed light on where the wild ancestors of this crop may have initially been domesticated (54). In a recent application of both GARP and MaxEnt, Banks et al. (55) found that the three main cultures spreading Neolithic lifeways across Europe between ∼6000 and 5000 BC occupied areas with distinct and essentially nonoverlapping ecological niches. Their paleoclimate reconstruction reflected only a single mid-Holocene reconstruction, which—although certainly preferable to simply using contemporary conditions—cannot address whether changing climates affected the spread of these Neolithic societies.

Correlative models cannot establish a direct causal relationship between presence (or absence) of a species and a particular variable, such as temperature or soil quality; one cannot ask which of these factors played the most important role in explaining observed distributions. The approaches also assume a static relationship between the organism whose distribution is being explained and the organisms with which it is associated. This accords no role for behaviors, such as prey switching or niche construction, which are especially prominent in human adaptation (56). Nevertheless, their abstractions can be very usefully applied to large-scale problems. Good fits have been demonstrated between a numerical reaction/diffusion human dispersal model forced by time-varying temperature, net primary production, desert fraction, and sea-level boundary conditions from a global earth system model for the last 125 ka, and some interpretations of the archaeological and genetic records for dispersals of Homo sapiens out of Africa (57).

Mechanistic modeling.

On the other hand, we can start by viewing a species as a relatively static bundle of morphological and physiological traits that are translated into fitness components, depending on the organism’s interaction with its environment (58, 59). Such “mechanistic models” require a detailed understanding of the physiological response of species to environmental factors. For this reason, they have mainly been applied to understanding where well-documented organisms, such as grain crops, could have been grown in the past, and how productive these organisms would have been.

Mechanistic Modeling and the Origins of Agriculture.

Climate variability struck farming societies particularly strongly through effects on their crops. Correlations between changes in climate and shifts in agricultural regimes/potential productivity have been noted in many areas of the world, including the British Isles (29) and the Middle East (60). For example, rice was first cultivated in one or two centers in the Yangzi river valley that provided a great deal of summer warmth and ample water resources (61). How were farmers able to adapt this water- and heat-loving plant to higher altitudes and latitudes? Niche modeling has revealed that it was only following the development of a temperate variety of rice that farmers were able to move it into the highlands of southwestern China and northern China (62, 63).

Climate and the Spread of Farming in Asia.

A number of studies have highlighted the 4.2 ka (∼2200–2100 BC) event as a potential driver for culture change in East Asia (64), as well as elsewhere in Eurasia (65–67). A 5,500-y model of changing crop niches on the Tibetan Plateau (68) created by downscaling a hemispheric record of climate revealed that this event had a substantial effect on early agricultural systems in this area. Before the use of niche modeling, archaeologists working in eastern Tibet were puzzled by what appeared to be a fundamental reorganization in both settlement patterns and subsistence regimes around the second millennium cal BC. Niche modeling demonstrated that declines in temperature following 2100 BC made it impossible for early farmers in the eastern Himalayas to grow traditional staple crops of broomcorn and foxtail millet, leading to abandonment of sites in southeastern Tibet (68, 69). The introduction of two foreign crops that thrive in lower temperatures, wheat and barley, allowed farming to once again became viable on the margins of the Plateau (Fig. 3). These models also have implications for past trade-route dynamics because they reveal that at some sites the crops present could not have been grown in situ but must have been brought in from lower elevations. Rapidly increasing temperatures in the Himalayas are now threatening the livelihoods of thousands of smallholder barley farmers and yak pastoralists (70). Understanding how humans in this area adapted their subsistence regimes to periods substantially warmer than today may prove crucial for ensuring sustainable food resources on the Tibetan Plateau.

Fig. 3.

(Upper) Mean probability of being in the thermal growing niche on the Tibetan Plateau. Millets (blue) declined precipitously between 4,000 and 3,500 calibrated years B.P., whereas wheat and barley (red) remained largely in the niche. Lines are at the mean probability for the Marcott et al. (35) northern hemisphere temperature reconstruction; shaded areas represent the 1σ confidence intervals around this reconstruction. (Lower) Radiocarbon dates on the Tibetan Plateau. Shaded areas in background are summed probability density distributions, whereas lines in the foreground represent a nonparametric phase model on ¹⁴C ages via Gaussian Mixture density estimation; blue: dates from millets; red: dates from wheat and barley; gray/black: all available dates (after ref. 72).

Farmers and Climate in the Prehispanic United States Southwest.

With abundant paleoclimatic archives and numerous well-preserved and precisely dated archaeological sites, the upland Southwest has long served as a laboratory for understanding the interactions between climate and farmers (71, 72). Reconstructions of potential annual maize production for southwestern Colorado—telling because of the heavy reliance on maize in the Puebloan diet (73)—began to appear in the early 1980s (74). By the early 1990s, GIS technology allowed annual potential production estimates by 4-ha cells in some areas, suggesting that reductions in maize production during the 13th-century exodus from the Four Corners were perhaps not so severe as previously believed (75).

For this and another densely occupied portion of the Pueblo world along the northern Rio Grande of New Mexico, water–year precipitation and growing degree days have recently been retrodicted using a method from quantitative genomics to choose the best combinations of tree-ring chronologies as predictors (31). Thresholding these reconstructions at the thermal and moisture requirements for maize, locations where it could grow without irrigation for the last 2,000 y were mapped, helping explain many population movements in the Pueblo region. From the mid-1100s to the mid-1200s, for example, southwestern Colorado was less favorable for farming than the northern Rio Grande in New Mexico, the probable destination for many of the Colorado emigrants in the mid-to-late 1200s.

Now, facilitated by high-performance computing, similar methods have identified the spatial extent of the annual maize dry-farming niche for a study area enclosing all those portions of the United States Southwest for which we have tree-ring–dated archaeological sites (76). Episodes of large-scale construction and social codification peaking in the mid-600s, mid-800s, early 1100s, and mid-1200s (just preceding the complete depopulation) each ended with reductions in extent of the maize dry-farming niche. The degree of culture change experienced as these episodes terminated is roughly proportional to the size of the maize niche reductions in each case.

These niche reconstructions can be compared with 21st-century drought projections for the Southwest from GCMs (34) (Fig. 4). This approach permits direct comparison between the droughts faced by Pueblo peoples and those likely to be experienced in these same areas in coming decades. Under the business-as-usual scenario, the Southwest will likely undergo drought that dwarfs any experienced by prehispanic Pueblo people. The historic dimension of this comparison adds value to the GCM-enabled projections by putting the effects of projected climate change into a human perspective. Future conditions in this region are likely to be far worse than those precipitating the collapse of the Chacoan regional system in the mid-AD 1100s and the complete departure of farmers from the northern Southwest in the mid-late AD 1200s.

Fig. 4.

Maize dry-farming niche and PDSI reconstructions in the United States Southwest. The solid black line is the percentage of all 30-arc second cells in the maize dry-farming niche; the dotted black line is the percentage of cells that contain any tree-ring date within this period; and the blue line with gray shaded area is the percentage of cells that have a tree-ring date in the plotted year or in any of the previous 3 y with 95% confidence interval (after figure 2C in ref. 76). The green line is the PDSI moisture balance reconstruction from the North American Drought Atlas (NADA). The red line and gray shaded 95% confidence interval are the multimodel PDSI means averaged across 17 CMIP5 models (after figure 1 in ref. 34). All series are smoothed using a 21-y center-aligned Gaussian filter with a 5-y SD.

Climate-induced changes in plant-growth patterns arguably played a critical role in when and where humans began farming in the first place. Richerson et al. (77) postulate that lower carbon dioxide (CO₂) levels, colder temperatures, and generally more arid conditions during the late Pleistocene reduced plant production sufficiently to make it impossible for humans to begin cultivation. Experimental work will soon enable calibration of mechanistic models to evaluate such all-or-nothing scenarios. Experimental growing of teosinte, the wild progenitor of maize, under late-glacial conditions does result in significantly lower productivity but also (unexpectedly) induces teosinte to respond with a number of maize-type traits (78). Similar experiments with other crop plants will become foundational to archaeological explanations of culture change in the next decade. Future models should seek to integrate other key factors that influence plant distribution, such as soils, changing CO₂ levels, and low-frequency variability (particularly in temperature) not well captured by some proxies (e.g., tree rings). Careful coupling of experimental and archaeological data, and climatically driven crop-production models, are essential for improving our understanding of the Neolithic and the growth in population (79) and sociopolitical scale it entrained (80). It is only through such studies that we can hope to really understand the advent of the Neolithic, because they allow us to rigorously test such recent suggestions as the proposal that the uptake of farming is not because of its initially higher productivity relative to foraging but because of changes in institutions supporting the introduction of farming, such as adopting the concept of private property (81).

Micro Models: Humans with Agency.

So, although several studies show how changing climates affected cultigen distribution and productivity, they are silent on other key questions, such as social responses to climate-driven resource variability and human effects on the environment. We would like to know, for example, the extent to which mid-Holocene impact on land cover by farmers led to gradual increases in atmospheric CO₂ and methane (CH₄) levels, perhaps warranting an “early Anthropocene” (82).

Population estimates and pollen records derived from archaeological sites, and historic records, have been used to estimate the extent of human impact on the vegetation landscape. HYDE (83) and KK10 (84) models anthropogenically induced land-cover change over the course of the Holocene. A comparison of their results (85) demonstrates the widespread and profound ecological changes brought on by even relatively small human populations more than three millennia ago, underscoring the importance of archaeology as the main source of knowledge on these societies.

Surface-process models (sometimes combined with agent-based models) have suggested that in semiarid areas with significant topographic relief, even limited settlement by Neolithic farmers or pastoralists can have outsized impacts on local and downstream dynamics of intermittent watercourses, initiating arroyo/barranco formation that reduced agricultural productivity by eroding fields in some areas and burying them in others (86). These processes can happen even without climate change, although they can be exacerbated by climate change.

Many contemporary approaches to modeling human–environment interactions draw on methods developed to study complex systems. Such approaches seek to understand how (possibly heterogenous) individuals interact to lead to systemic changes that in turn affect the individuals themselves (87). Two main complex systems approaches used by archaeologists are network applications and agent-based modeling. Both enable characterization of feedbacks between human systems and environmental variables, and can accommodate (and study the roles of) heterogeneity and noise.

Trophic networks quantify the presence and absence of feeding links among species in a given system. In a recent study, Yeakel et al. (88) reconstructed 6,000 y of species’ occurrence and their trophic networks in ancient Egypt, demonstrating how aridification pulses, hunting, and habitat competition by growing populations affected species composition, and how species extinctions were dynamically destabilizing for the Egyptian ecosystem. Both anthropogenic and climatic disturbances led to the extinction of larger-bodied species; stability decreased markedly around 3000 BC, and further declined around 2000 BC (in both instances following aridification pulses) before falling precipitously in the last 150 y, resulting in a contemporary community highly sensitive to additional perturbations.

Another trophic network analysis of faunal assemblages from archaeological sites formed by dense populations of prehispanic maize farmers in southwestern Colorado shows that climate change and human-mediated deforestation together reduced habitat for certain species (elk and snowshoe hare) (89), perhaps contributing to the 13th-century depopulation of this region. However, archaeological research also shows that human predation is not always destabilizing. Dunne et al. (90) studied the structural role of humans in highly detailed reconstructions of marine food webs surrounding Sanak Island, Alaska. Despite 10,000 y of predation, Aleut fishing and sea-mammal hunting did not generate long-term extinctions; relatively small human populations, prey switching, and connections of prey to vast population reservoirs afforded ecosystem stability. Although a traditional zooarchaeological study would certainly note the presence or absence of species (91), food-web approaches additionally allow us to examine the effects of species additions and removals on the ecosystem as a whole (92). These approaches also allow for detection of keystone species via centrality measures (93). Finally, by displaying the patterns in large datasets, they provide a source for inferences about the processes creating the food web and allow for synthesis of large amounts of information (90).

Agent (or individual) -based modeling provides another framework for analyzing human interactions with a landscape or an ecosystem, which permits modeling human mobility as well as various economic, social, and political behaviors. This type of modeling is especially useful for structured exploration of the paths by which new “higher-level” entities (such as villages or polities) might emerge from interactions among lower-level agents (such as households) while taking climatically mediated changes in landscape productivity into account. Archaeologists developing such models generally combine empirical research (to derive target patterns to be explained) with modeling to assess the capacity of various processes to reproduce those patterns. Such approaches therefore allow archaeologists to assess causal pathways—explanations—for the patterns seen in the record.

Relatively precise demographic reconstructions, chronologies, climate proxies, and a great deal of archaeological research allow us to glimpse the complexity of causes for the depopulation of the northern United States Southwest during the 13th century. Initial models developed for the United States Southwest (94, 95) tended to demonstrate that farmers in these uplands were exquisitely sensitive to precipitation-mediated variability in maize production before intensification of water-control technologies after ∼AD 1300.

One response to this dependence was development of food sharing between households (96). Food-sharing models strongly suggest that interhousehold exchanges of maize and meat permitted larger populations to be supported, encouraged people to aggregate in villages (especially in periods of high production), and stabilized reliance on domesticated turkey that became important in the northern Southwest in the 11th century AD following depression of deer populations. Tacking between results from maize-niche modeling, agent-based models, trophic network modeling, and empirical research allows us to suggest that long-term regional population growth, depression of important wild resources (especially deer), extreme dependence on just one cultigen, climate-induced migration into portions of the central Mesa Verde region that were highly susceptible to drought, religious, and probably political balkanization, conflict, and perhaps hostile encounters with mobile foragers just entering the Southwest from the north together created the context for its final abandonment (41). We should expect to find similar complexity in cases of climate-related culture change in other times and places where conditions cannot be viewed so clearly (97).

In pastoral Mongolia, Rogers et al. (98) have tied surviving environmental instability at varying temporal scales (from immediate effects to thousands of years) to social adaptations and the health of mixed-species herds at the suprahousehold level. Surprisingly, perhaps, the wealth of a household is not a good predictor of its survival in periods of environmental instability. Rather, the connectedness of pastoralists in social networks, as well as the strength of hierarchical control, ensures survival, although optimal decisions for the group may be detrimental for certain households, putting group-level and individual-level decisions occasionally at odds.

Conclusions

Challenges facing scientists today include not only understanding how environmental changes of climatic or anthropogenic origin affected the well-being of humanity over the longue durée, but also how our interactions with the environment create dependencies extending into the future (99). Computational, model-based approaches (100) to archaeology allow us to accommodate—in a structured way—knowledge of ancient climatic variability at ever-higher temporal and spatial resolution by providing linkages between this variability and the resources on which humans relied. These approaches additionally allow archaeologists to both account for and study the effects of human feedbacks on natural systems, opening a path for rigorous analysis of how changes in climate—paired with the decisions that humans make—lead to cascades of socio-natural change.

Taken together, these emerging approaches move archaeology well beyond merely noting coincidences of social and climatic change. Causal pathways suggested by models become new targets for empirical investigation. The immediate goal—within grasp for a few well-studied regions—is to provide accurate accounts of the role of climate change and variability in the successes and failures of the societies we study by assessing the weights on the multiple pathways through which climatic effects may be felt in the context of the factors at play along each. On a much more distant horizon, forward modeling of many possible worlds from arbitrary starting points, compared with the trajectories realized by history, will help us evaluate our human capacity to induce and successfully react to climatic changes.

When the demographic and social effects of past climatic variability are well understood through historical or archaeological research, coupling models of past climatic variability with models for climate futures adds value to both (34). In fact the interpretability of forward-looking climate models (and in some cases their skill) depends in part on the quality of paleoclimatic and archaeological records for the same area. Ironically though, as archaeologists renew their interest in past climates and develop increasingly powerful methods to study them, climate change itself is actively undermining our efforts. In many places coastal sites are rapidly eroding as sea levels rise (Fig. 5); thousands of sites (from Mesolithic shell middens to Iron Age brochs, Norse settlements, and military structures from two world wars) are at risk in coastal Scotland alone (101). Indeed, computational modeling is now also being applied to study how changing sea levels are affecting archaeological sites and to predict locations of future impact (102, 103). Rising sea levels, however, are not the only threat. Permafrost thawing is leading to the sudden and rapid loss of well-preserved climate archives in organic middens in Greenland (104) and throughout the Arctic. Increasing temperatures and drought stress in the United States Southwest are driving higher tree mortality and more wildfire (105), damaging the many archaeological sites in these forests.

Fig. 5.

Images of a late 16th-century site at Brora, northeastern Scotland, where coal heated seawater for extraction of salt. (Left) A volunteer-based excavation recording the saltpans just before a big storm washed them away (Right). Photographs by (Left) John Wombell and (Right) Penny Paterson, and images courtesy of Scotland's Coastal Heritage at Risk Project.

As we face our uncertain future, having well-calibrated models of thousands of years of ancestral trials should help us more knowledgeably manage our numbers, our ecosystem interactions, and in general avoid past errors. From both the perspective of how soon we would like to draw on the findings of such models, and how complete the archive on which they are based will be, we need to move rapidly.