I thank the Q10 organizers for inviting me to participate in the conference held in Bordeaux and for allowing me the opportunity to contribute to this volume. The research outlined and reviewed in this paper was supported in part by the European Research Council (FP7/2007/2013/ERC TRACSYMBOLS 249587) and the LaScArBx Cluster of Excellence-Agence Nationale de la Recherche (ANR-10-LABX-52). I also thank the anonymous reviewer whose comments served to improve the manuscript.

1 - Introduction

1One of the broad goals of archaeology is to describe and understand the behaviors of past populations by documenting archaeological cultural variability and tracing the cultural trajectories that we observe in the archaeological record. Since all cultures operate within an environmental context, it is important that we be able to identify and characterize the ecological niches that past populations exploited and take these ecological parameters into consideration when interpreting past cultural variability, especially across periods of climate change. In so doing, it becomes possible to better understand culture-environment relationships, disentangle the cultural and ecological mechanisms (constellations of factors and their processes of interaction) involved, and understand how they influenced cultural adaptations and documented cultural trajectories (d’Errico & Banks, 2013).

2This interest in culture-environment relationships is by no mean a recent phenomenon. Over the past several decades, anthropologists and archaeologists have employed a number of heuristic approaches to examine how cultures interact with their respective environments and how such contexts may have been implicated in the records of cultural adaptation and change that we study (Steward, 1972; Hardesty, 1972; Vayda & McCay, 1975; Hardesty, 1975; Binford, 2001; White, 2007). With respect to Paleolithic archaeology, a variety of studies have examined archaeological data in the context of climatic variability in order to infer how changing environmental frameworks may have influenced cultural adaptations and demography (Bocquet-Appel &Tuffreau, 2009; Lemmen et al., 2011; Discamps, 2014; Birks et al., 2015; Tallavaara et al., 2015; Pétillon et al., 2016). These approaches are of value and greatly aid our ability to identify the factors, be they cultural or environmental, that were potentially implicated in hunter-gather adaptations and cultural change through time. Examinations of culture-environment relationships that take the additional step of specifically considering past ecological niches and ecological niche dynamics are still relatively rare, however. By reconstructing and examining ecological contexts, as well as their variability over time, it becomes possible to explore potential links between cultural traits and ecological parameters thus aiding our ability to identify the long-term trends that played important roles in past processes of culture change (d’Errico & Banks, 2013).

2 - Eco-cultural niche modeling

3Over the past decade, a number of studies (Banks et al., 2009, 2011, 2013) have detailed an approach, termed Eco-Cultural Niche Modeling (ECNM), that can be used to explore the relationships between archaeological cultures and their paleoenvironments, as well as examine them across periods of climate change. ECNM integrates archaeological, chronological, and paleoclimatic data, via predictive algorithms developed within the ecological and biodiversity sciences, thereby providing a means to estimate the ecological conditions exploited by archaeological populations and evaluate the potential variability of these niches through time. Such an approach offers clear advantages over methods solely based on simple distributional data. For example, the archaeological distribution of past population may not completely or accurately reflect that population’s actual range because simply drawing a boundary around known occurrences will, more likely than not, tend to be an underestimation of the real range. Extrapolations of ranges between and beyond known occurrences depend of subjective knowledge and often overestimate a population’s distribution. Furthermore, plotting occurrences and comparing them to static backdrops of environmental information cannot accurately characterize the ecological niche exploited by a past human population.

4With ECNM methods, one is able to establish quantitatively whether cultural variability and behavioral changes observed in the archaeological record have an ecological basis, or if such shifts and population distributions were more influenced by non-ecological factors. In the case of the latter, detailed analyses of material culture remains and subsequent inferences pertaining to modes of cultural transmission and forms of social organization needed to maintain cultural traditions can allow us to better explore the structure and evolution of past cultural systems.

5In order to estimate a past ecological niche with available predictive algorithms or architectures, one must have a variety of detailed data. First, one needs the geographic coordinates of archaeological sites from which material culture assemblages diagnostic of a particular archaeological culture or technocomplex have been recovered. One also must have relatively high-resolution environmental data in the form of raster data layers. We have routinely chosen to use the outputs of high-resolution, regionally-zoomed paleoclimatic simulations produced by forcing either atmospheric or coupled ocean-atmosphere general circulation models with boundary conditions relevant to the time period or climatic event for which paleoclimatic conditions are being reconstructed. The outputs of these simulations can also be input into global vegetation models in order to estimate vegetation cover. Before being used to estimate ecological niches, these paleoenvironmental data must be critically evaluated against terrestrial and marine archives in order to determine whether they accurately reflect past conditions. For example, we found that while simulations of the LGM used in our investigations of the Solutrean archaeological culture were generally in close agreement with pollen data, they did slightly underestimate winter cooling over Western Europe (Banks et al., 2009, p. 2859). These cooling underestimations are also evident in reconstructed ranges of past permafrost distributions and associated environmental conditions (Andrieux et al., 2016). Thus, while paleoclimatic simulations may approximate past conditions relatively well, such accuracy cannot simply be assumed and continued efforts are needed to further improve them. Finally, analyses of archaeological chronological data are needed so that archaeological cultures can be accurately placed in time and correlated with specific paleoclimatic events (Banks, 2015), thereby allowing one to determine which paleoenvironmental data are the most appropriate for estimating a past niche.

6Numerous methods exist for reconstructing ecological niches (for a review, see Peterson et al., 2011). At a general level, they identify the environmental characteristics of known locations of where the population of interest has been observed (i.e., occurrence data: geographic coordinates where the study population has been observed) and employ a variety of methods to define these parameters’ relationships. These relationships are used to estimate an ecological niche in environmental space, which can then be projected geographically to determine its distribution. ECNM, thus, reconstructs or estimates a past ecological niche exploited by an archaeological culture, or distinct regional variant of that culture.

7While a detailed description of how various methods function is beyond the scope of this paper, it is worth noting that important differences exist, especially with respect to the types of occurrence data needed. Some architectures, such as Bioclim (Nix, 1986), work well with presence-only data. One potential limitation of Bioclim, though, is that it cannot extrapolate into combinations of environments not associated with the occurrence data. Other methods require absence data (i.e., locations that have been surveyed and at which the subject of study was not observed) and include Generalized Linear Models, Generalized Additive Models, as well as genetic algorithms (e.g. GARP; Stockwell & Peters, 1999). The latter is based on the development of linear string rule sets that are allowed to evolve through an iteractive process of training and testing. Because in many instances true absence data may not exist (e.g., in archaeology - issues related to sampling, site preservation and/or visibility), approaches exist with which pseudoabsences for the study area can be generated and used in the modeling process. Such practice expands the range of architectures that can be used to estimate an ecological niche when only presence data are available. Finally, there exist modeling methods, such as Maxent (Phillips et al., 2006), that use presence data in conjunction with background data (i.e., data pertaining to environmental variation across the study area). Maxent is based on the maximum entropy principle such that the generated probability distribution is closest to uniform and avoids assumptions not supported by the data. Its ability to extrapolate into combinations of environments unavailable during model training can be advantageous, but one challenge can be to avoid overfitting (i.e., excessive model complexity) due to its ability to fit extremely complicated response curves.

8Because no single predictive algorithm will be the best performer under all circumstances, the ecological community agrees that a combination of algorithms should be used (Araújo & New, 2007; Thuiller et al., 2009; Antunes, 2015) and that the best single or few should be used in a given situation (Qiao et al., 2015), based on the types of questions being investigated and types of available data (Araújo & Peterson, 2012). Once estimations are produced, there exist a variety of methods that serve to reduce excessive model complexity (Peterson et al., 2007), as well as those that allow niche estimations to be better characterized (Warren et al., 2010; Peterson et al., 2011).

9Another important capacity of these architectures is that they can be used to examine niches between time periods thereby allowing one to determine whether or not successive populations exploited different niches. By comparing the material cultures of two or more successive archaeological cultures and by taking into account environmental frameworks within which they operated, one can evaluate whether cultural innovations reflect a response to environmental fluctuations. This allows us to significantly improve our ability to understand the complexities of culture-environment relationships through time.

10With ecological niche modeling methods, what exactly is it that we are estimating? The concept of an ecological niche was formalized by Grinnel (1917) who defined it as the geographical expression of a species’ climatic and habitat requirements. The Grinnellian niche considers the roles that unlinked, non-consumable environmental variables play in the distribution of a species. Based on the Grinnellian concept, Hutchinson (1957) defined the fundamental niche, which represents the total range of environmental conditions within which a species or population can exist indefinitely. The factors and parameters that are implicated in the composition of an ecological niche, though, are quite diverse. It is for this reason that more recently, Soberón and Peterson (2005) proposed a conceptual framework that attempts to summarize these various factors. This framework, termed BAM (fig. 1), is composed of three broad classes of factors. The first is “A”, referred to as the “existing fundamental niche” by Peterson et al. (2011), represents the intersection of Hutchinson’s fundamental niche with the set of environments that are actually present on the landscape. Unit “B” represents variables that are dynamically linked to a species, which include food resources, as well as the influence of competitors and predators-factors that quantify how a species functions within its niche. Finally, “M” stands for the geographic areas that are accessible to a species or population via dispersal and that have been sampled such that occurrences could have been detected. It has been demonstrated that this factor is important to take into consideration when estimating niches (Barve et al., 2011). Since the factors associated with “B” are difficult to quantify and integrate into predictive modeling exercises, many modeling efforts neglect such interactions and allow them either to represent noise in the system or to manifest via correlations with A or M. The Eltonian Noise Hypothesis (Soberón & Nakamura, 2009) suggests that biotic interactions are only influential at finer spatial scales and will rarely be manifested in geographic effects. When this hypothesis holds true, the effects of biotic interactions are essentially averaged out over broad geographic scales, thereby making A = B (fig. 1), and when M has been incorporated into the niche modeling process, the resulting niche estimations could be considered, at a minimum, as estimating the intersection of the existing niche with regions that would have been accessible to the population under study. Research conducted by Kearney et al. (2010) suggests that correlative predictive methods are capable of effectively estimating the fundamental niche. A multitude of methods exist for characterizing estimated niches and determining under which circumstances and in which dimensions they underrepresent the fundamental niche (for a review, see Peterson et al., 2011).

Adapted from Soberón & Peterson (2005).

3 - ECNM example: the Proto and Early Aurignacian in Europe

11To provide an example of the application of this approach to the archaeological record, the results of a study conducted for the early part of the Upper Paleolithic are briefly presented here. The Aurignacian technocomplex is comprised of a succession of culturally distinct phases. Between its first two subdivisions, the Proto-Aurignacian and the Early Aurignacian, we see a number of shifts in technology and material culture. For example, we observe a change from single to separate reduction sequences for blade and bladelet production, along with other changes in stone tool typology and technology, the appearance of split-based antler points, as well as changes in symbolic material culture. Bayesian modeling of archaeologically reliable ¹⁴C determinations indicates that the cultural changes between the two phases are coincident with abrupt and marked climatic changes between Greenland Interstadials 10/9 and Heinrich Stadial 4. Following this chronological analysis, ECNM was used to quantitatively evaluate whether these material culture shifts are correlated with a significant shift between the niches exploited by these two broad populations (Banks et al., 2013). In other words, did these cultural changes have, at least in part, an ecological basis?

12Results indicate that the transition between the Proto-Aurignacian and the Early Aurignacian was associated with an expansion of the ecological niche (fig. 2). This was determined by first estimating the niches of the Proto-Aurignacian and also the Early Aurignacian phases, with the majority of variability within each of the two niches being primarily explained by temperature variables. Then, the Proto-Aurignacian niche was projected onto conditions associated with the Early Aurignacian (i.e., Heinrich Stadial 4), and the ability of this projected niche to predict Early Aurignacian occurrences was statistically evaluated. Results indicate that the observed failure of the projected Proto-Aurignacian niche to predict a large number of Early Aurignacian sites is significantly greater than one would expect by chance, thus indicating that the ecological niche expanded between these two phases. These shifts in both the eco-cultural niche and material culture are interpreted to represent an adaptive response to the relative deterioration of environmental conditions at the onset of Heinrich Stadial 4. This exemplifies a situation in which cultural flexibility was used by hunter-gatherer populations to quickly adapt to a rapid-scale and severe climatic fluctuation. Thus, the tendency observed among most animal species to conserve an ecological niche across relatively short time spans characterized by environmental variability (Peterson, 2011) does not appear to hold true for human hunter-gatherer populations in some situations. An issue that warrants further research is if and when the use of cultural solutions to expand the exploited ecological niche occurred with other hunter-gatherer cultures, as well as with other hominins (e.g., Neanderthals).

4 - Perspectives

13In attempts to understand culture-environmental relationships, regional cultural trajectories, and ultimately the mechanisms at work behind the variability that we observe in past material cultural records and their changes through time, ECNM represents an approach that can be extremely informative, especially when combined with other more traditional approaches (e.g., reconstructing technological chaînes opératoires, faunal studies, site function and activity analyses, etc.). With this approach, and resultant comparisons of regional cultural trajectories, we can begin to discern whether adaptive shifts and material culture changes were undertaken so that populations could continue to exploit the same niches in the face of environmental change (niche conservatism), or if such cultural changes were more routinely associated with niche shifts-patterns which are difficult to ascertain when environmental and related data are employed only as static, comparative backdrops. Furthermore, with this approach, one can identify instances in which cultural changes have no ecological basis. In essence, the integration of ECNM into research strategies aimed at better understanding the dynamics of culture change, and the implicated mechanisms, allows one to interrogate, rather than simply document, the archaeological record.