Non-Spatial Data and Modelling Multiscale Systems in Archaeology

1 Introduction

More and more people are accepting the severe nature of the effects that humans are having on climate and environments around the globe, and the precarious situation this has put them in (Shukla et al., 2019). Aspects of these developments are, however, neither new nor unique to the modern era, but the scale and potential to drive changes is now much larger than at any time in the past (Bell & Walker, 2005). Archaeology is unique in its position of being able to study the ways in which humans have interacted with the environment over long (millennial) timescales and to assess the multitude of ways in which humans have both caused and responded to change. The knowledge generated from these studies is vital if we are to contextualize modern day changes and understand the effects that different actions and reactions may have in the future (Boivin & Crowther, 2021; Rick & Sandweiss, 2020).

Modelling has become an essential part of the archaeologist’s toolkit for testing hypotheses and ideas on human–environment interactions. However, as archaeology is a multidisciplinary subject with roots in both the Sciences and Humanities, defining modelling is not an easy task. Various theoretical approaches and developments over the past decades have emphasized different goals and ideas within archaeology, as well as different methodologies for achieving these (e.g. Preucel & Mrozowski, 2010; Reitz & Shackley, 2012; Tilley, 1990). Consequently, definitions and purported purposes of modelling paralleled these developments (Nakoinz, 2018, pp. 103–108). Archaeology has borrowed not only methods but also theoretical frameworks from other disciplines (Johnson, 2010), affecting both how we approach the material studied and the modelling itself. However, the overall goal of providing “…simplified mapping for a special purpose” does appear to permeate this plurality of definitions (Nakoinz & Knitter, 2016, p. 31).

A central component in contemporary archaeological workflows is the geographical information system (GIS), something which has had a significant effect on the management and analysis of spatial data (e.g. Gillings, Hacıgüzeller, & Lock, 2020; Howey & Brouwer Burg, 2017; Landeschi, 2019). The GIS environment offers a number of easily accessible tools for not only modelling data but also structuring it. Computational models, such as those found in agent-based modelling or predictive modelling, are most often hierarchical in structure (Brouwer Burg, 2017); the complete or final model is in fact built upon a number of lower-level tasks. For example, when assessing the archaeological potential of a region on the basis of evidence for human impact, the environmental data are not used in its raw form (e.g. counts of pollen or insects, LiDAR point cloud) but instead are aggregated and interpreted, or simplified into models of landscape change and topography. These models then form subcomponents or backgrounds, which can be populated by agents, or used as the starting points for predicting site locations. There are, however, a number of risks associated with the uncritical use of such systems (Brouwer Burg, 2017; Buckland, Dell’Unto, & Pálsson, 2018; Price, Louys, Faith, Lorenzen, & Westaway, 2018). With easy access to large amounts of open environmental data, the uncritical point-and-click approach to GIS may unintentionally push the limits of assumptions underpinning the models, or even risk the outright misrepresentation of certain landscape features. Something as rudimentary as a digital elevation model (DEM) takes relatively few steps to create and as a result is seldom reflected upon with regard to the quality of the underlying data or model output. In turn, the DEM is used to create new models, such as viewsheds or past shoreline levels, where any potential flaws in the underlying data may be carried over (Verhagen & Whitley, 2012, pp. 55–58). To more accurately represent the palaeolandscape (past landscape), empirically robust topographical and vegetational modelling is also essential (e.g. Sugita, 2007a,b). For example, a DEM based on LiDAR or other elevation data may be a sound basis for modelling a modern landscape, but it will not account for any potential movement of sediments that may have occurred over time, and this would undoubtedly require additional empirical studies (such as sediment coring and geomorphology analysis) to substantiate, date, and recreate the past landscape (Scherer et al., 2021).

The aim of this article is to encourage the discussion of these lower-level modelling tasks, as it is apparent that they hold considerable potential and information, that can be used to further strengthen a study. With the steady advancement of GIS and an ever-expanding community of “amateur programmers” in archaeology (computer literate users not necessarily at a professional level), the field is well positioned to push boundaries and answer new (and old) questions. Not all modelling tasks need to be handled within a GIS software package, and not all modelling questions directly concern space. Despite this, there appears to be a tendency towards a more frequent use of spatial analysis and simulation in archaeology. This article will contextualize this with respect to the often accompanying reduction in focus on the non-spatial empirical modelling that lies behind many theories.

2 Modelling and GIS

It can be difficult to clearly separate models into singular categories, but generally they can be described as being either theoretical or empirical (Nakoinz, 2018). Put simply, theoretical models are based on various theoretical considerations and used to produce something of an “ideal case” scenario, whereas the empirical models are based on direct (or indirect) observations of the real world. Binford’s (1980) description of the collector–forager adaptation among hunter-gatherers, whilst based on his own studies of the Nunamiut and San peoples, is an example of a theoretical model. This model could be, and indeed has been (e.g. Bang-Andersen, 1996; Forsberg, 1985; Savelle, 1987), used to try to explain the observed settlement patterns of past hunter-gatherer groups. Nakoinz and Knitter (2016) argue the need for both empirical and theoretical models to generate new knowledge, as the theoretical model can potentially explain the observations made in the empirical one. Simulations could be placed as nestled in between these general categories, providing examples of how these types of model interact. Although the term is often used as a synonym for modelling in general, it is better described as a modelling approach; simulation makes use of theoretical models to generate artificial data, which can then be used to construct an empirical model (Nakoinz & Knitter, 2016, p. 233, Romanowska, Crabtree, Harris, & Davies, 2019). Computerized simulation has been likened to a “virtual laboratory” where a number of different hypotheses can be tested by implementing different sets of rules that will affect the emergent patterns (artificial data) (Cottineau, Reuillon, Chapron, Rey-Coyrehourcq, & Pumain, 2015). As is discussed by Whitley (2016), this is not dissimilar from experimental archaeology where recreating different lithic reduction or wood-working techniques will produce “patterns” that can be compared to their observed, empirical counterparts.

Modelling lets us test new ideas, but also return to previous theories and models to evaluate and further develop them. The latter is increasingly facilitated by easier access to large amounts of data, new empirical methods, and computing power. The Binfordian model has had a significant influence on the interpretation of the spatial organization of hunter-gatherers in much of Europe but has also seen considerable modification and development since its original publication (Binford, 1980; Preston & Kador, 2018; Ucko & Layton, 1999). In Sweden, functional analysis of artefact assemblages has been central to modelling settlement and mobility in the northern regions for the 8000–2000 BP period (Bergman, 1995; Forsberg, 1985). Whilst tentative interpretations suggest the hunter-gatherer groups sourced lithic raw materials from the Scandian foothills and mountain region, it may be difficult to substantiate such claims (Forsberg, 1985, pp. 271–273). The subject of procurement and provenance of lithic raw materials in this region needs to be approached carefully due to the past glacial impact on the landscape, and displacement of geological material (Stephens & Bergman Weihed, 2020). The moraine formations left behind by the glacier may include materials from the mountains, which was transported down towards the inlands, making them available for use by people as part of a more embedded lithic procurement strategy (Binford, 1979; Rosenberg, Gluhak, Kaufman, Yeshurun, & Weinstein-Evron, 2021). The petrographic character of a lithic artefact might thus point towards a geological origin in the mountain zone mistakenly assumed to be the location of quarry activity.

The author’s current research at Umeå University applies new techniques and methods to the characterization of quartzite points using non-destructive spectroscopy. Compared to other materials, like obsidian and flint, characterization of quartzite has seen less methodological development in Europe (Prieto, Yusta, & Arrizabalaga, 2019). With quartzite representing a major group of geological material utilized by hunter-gatherers within Northern Sweden, this material holds considerable potential for the further development of existing models of resource use. Characterization of lithics merely represents the first step, however, and subsequent spatial analysis requires the incorporation of both environmental and cultural data into the model. In building on these models, the aim is to infuse new life into the topic of raw material procurement, material flow, and spatial organization among hunter-gatherers within the region.

The treatment of space within archaeology, especially when analyzed within a digital environment, is a central topic for modelling (Gillings et al., 2020; Lock & Pouncett, 2017). An overt reliance on quantitative analysis of environmental data risks producing a deterministic model that fails to take human agency into consideration (Wheatley, 2004). Predictive modelling is an example of one such field that has been criticized for its environmentally deterministic view of the past, and for failing to take the behavioural differences with which hunter-gatherers view the landscape into account (Grøn, 2018). The critique has somewhat dissuaded its use in research, and as a consequence, this method has mainly become a tool for cultural resource management to focus resources in heritage management and surveying (Lausanne, Fedje, Mackie, & Walker, 2021; Verhagen & Whitley, 2012).

Working within GIS the user is restricted by what Lock and Pouncett (2017) refer to as “absolute space,” as the system is developed with a geographical representation of the world in mind. The (environmental and cultural) data archaeologists collect, along with their contexts, are also bound to this same “absolute” representation of space. Thus, to escape from the constraints of physical space, a more abstract modelling approach must be applied to the data. However, Lock & Pouncett (2017, p. 132) remark that any model of a more subjective “place” when added to GIS absolute representation of the world will become “…a ‘location’ defined by ‘co-ordinates’ with ‘attributes’…”. The incompatibility inherent in incorporating an informal understanding of space into a system that was developed with formal approaches in mind is readily apparent and has been wrestled with since GIS became fundamental to archaeological research (Lock, 1995).

Taking a step back from the GIS software and recognizing that modelling does not always entail the modelling of geographical data may be a way to opening up routes into more informal approaches. As a matter of fact, most archaeologists engage in this type of modelling without recognizing it when preparing data for spatial analysis, as will be discussed below.

3 “Models Within Models”

Modelling the spatial organization of Mesolithic hunter-gatherers is essentially an act of modelling a multiscale system. The use of the term “Multiscale” here is analogous to its use in biology, to highlight the complex nature of modelling human–environment relationships (Walpole, Papin, & Peirce, 2013). The results of actions taken by people in the past can be traced at different spatiotemporal scales, each contributing to the patterning observed today. These patterns were formed not only by the actions taken by people at different times and locations, but they also potentially occurred within environments much different to those in which the sites are presently situated. Furthermore, previous activity can influence subsequent action such that palimpsests of human–landscape interactions with subtle degrees of variation can occur in a single locale (e.g. Beneš & Zvelebil, 1999, p. 74).

3.1 Landscape-scale

Landscape-scale models incorporate data from multiple archaeological sites and often include palaeoenvironmental data from outside of the areas of excavations. The appropriate spatiotemporal scale and level of data aggregation of a model are largely dependent on the research questions for which it is designed to address (Buckland, Sjölander, von Boer, Mähler, & Linderholm, 2022, pp. 19–20). It is not (yet) feasible to manage the full context and complexity of each individual site included in a study of, for example, the Neolithic expansion in Northern Europe. Instead the site-level data are simplified, i.e. modelled, so as to facilitate (spatial and temporal) aggregation. Land cover reconstructions based on counts of pollen taxa simplify the data by grouping taxa into vegetation categories (e.g. Brouwer Burg, 2013) on the basis of an ecological understanding of the plants and their pollen (e.g. pollen dispersal rates and their expected representativity in sediments) (Sugita, 2007a). In a similar vein, archaeologists may infer site function on the basis of artefact assemblages and extrapolate these results to study the wider settlement system over a collection of sites (Forsberg, 1985; Gron & Sørensen, 2018); and they might trace the Neolithic expansion in a region using particular artefact types as proxies for a wider cultural phenomenon (Cortell-Nicolau, García-Puchol, Barrera-Cruz, & García-Rivero, 2022). This is exemplified in Figure 1, where each site in the analysis of a hypothetical settlement pattern is the result of one or more modelling tasks at different levels.

Figure 1

Illustration of the multiscale nature of modelling: (a) the hypothetical regional model of settlement patterns among hunter-gatherers is dependent on (b) inferences made at a local scale of site function, (c) and to assess function it is necessary to perform micro-scale analysis of different artefacts and data. In some cases, it is also necessary to move further down in scale to the (d) molecular level, for instance, to determine the petrographic character of a material. Each level simplifies data to some extent, introducing both new uncertainties, as well as information, to the higher-level model.

Aggregation is not merely limited to the aggregation of spatial data, however. As chronological change is central to archaeology, a common approach is to aggregate data by different “slices of time” (e.g. 100 years). Different methods have been developed to achieve this, but they also introduce their own assumptions, as discussed by Carleton and Groucutt (2020). Bayesian modelling is frequently used in the calibration of radiocarbon dates and construction of more precise chronologies, providing a means for visualizing change over time (Bayliss, 2009, Otarola-Castillo & Torquato, 2018). The flexibility of Bayesian modelling has also enabled a variety of ways of exploring spatiotemporal change, including resource exploitation (Béland et al., 2018), population dynamics (Crema & Kobayashi, 2020), and trade and exchange (Carrignon, Brughmans, & Romanowska, 2020).

Simulation approaches, whilst well established in palaeoclimate science (Haywood et al., 2019; Liu et al., 2009), are increasingly used to generate artificial data on past human activities, especially at large temporal and spatial scales (Verhagen, de Kleijn, & Joyce, 2021; Willmes et al., 2020). These may be based purely on the theoretical assumptions (Graham & Weingart, 2015) anchored in a number of empirical situations (i.e. calibrated; Won, Murali, Patrick, & David, 2016) or refined in an iterative process of simulation and data-model comparisons (Izumi & Bartlein, 2016).

Irrespective of the degree to which empirical or theoretical data are used in modelling or simulation, an increase in scale most often implies a trade-off in detail. It is not unusual for large-scale models to be based on data that have undergone a number of levels of abstraction before their incorporation (e.g. pollen and climate data; Strandberg et al., 2022). As discussed by Brouwer Burg (2017), no model is perfect, and a model at any scale always introduces more uncertainty into any higher-level model into which its results are incorporated. This is a necessary compromise when engaging in large-scale spatiotemporal analysis, and something to be considered at all stages in the modelling workflow, if not most importantly, when interpreting the results in terms of past human–landscape interactions.

3.3 Micro-scale

The term “micro-scale” is used here as a catch-all term for all types of analysis that are applied to individual finds and samples collected at a site. It thus represents the smallest data aggregation level (Figures 1 and 2) (Buckland et al., 2022, pp. 19–20). In many cases, this is the level at which “raw data” are recorded. Interpretations of raw material use, consumption practices, living environment, etc., can all be traced back to the treatment and analysis of individual samples. The same rigorous standard with which an archaeological site is evaluated should be applied when analyzing individual samples; just as a site is subject to post-depositional processes, so are the material remains (Huisman, 2009). An abundance of measurements and analyses can be undertaken on the material from a site, and the challenge for the archaeologist is to understand which of these are representative for the character of the studied material.

Figure 2

An illustration of the relationship between aggregation level and data resolution. With increasing aggregation level, such as in the reconstruction of past landscapes, the nuances of the underlying data become more vague (i.e. lower resolution). Lower aggregation levels typically produce larger volumes of data with high detail, for instance in spectroscopy where a single artefact might have multiple measurements linked to it.

In palaeoecological studies, the palynologist will rarely, if ever, count all the pollen grains present in a single fossil pollen sample. Due to the sheer number of pollen present in a single sample, it is not a reasonable time investment. Therefore, the palynologist will typically decide on a representative pollen sum for which the proportions between different species manifest themselves in a way that permits inferences to be made on vegetational development (Reitz & Shackley, 2012, pp. 276–283). This count is then used to represent the vegetational composition of a region for a certain point in time (Birks et al., 2016).

In lithic studies, petrographic characterization frequently includes spectroscopic or elemental analysis of the material. Whilst measuring on thin sections cut from artefacts provides more detailed information on the texture and composition of the material (e.g. Prieto et al., 2019), it is not always practical. Using a non-destructive approach and measuring on the surface of artefacts can still provide useful information (e.g. Sciuto, Linderholm, & Geladi, 2017), but it also raises different issues of sample representativity. Metamorphic rocks, such as quartzite, exhibit considerable heterogeneity, which can lead to different analytical results being obtained from different points on an artefact’s surface (Figure 3). The contact area between the probe and surface of the object can range from a few millimetres in dimension to a couple of centimetres, depending on the type of probe used. As the data collected are related to the parts of the surface of the artefact that the probe interacts with sampling can potentially be difficult (Magnusson, 1999). The analyst must be able to judge which parts of the object will provide a result which best represents the chemical character of the material as a whole. This is one of the reasons why repeated measurements on different parts of the object are desirable.

Figure 3

Quartzite points from Västerbotten County, Sweden. The petrographic heterogeneity varies significantly between the points, which needs to be considered when sampling for non-destructive analysis. Depending on what parts of the surface are sampled for spectroscopic analysis the results may differ significantly. Artefacts derive from the collections at Västerbottens museum (www.vbm.se). Photograph by the author.

Other types of instrumentation allow for sampling of the entire surface of the object at the same time. Spectral imaging works similarly to an ordinary digital camera, in that it captures a photograph of the object (Sciuto, 2018, pp. 33–39). It differs in that where an ordinary camera captures only the visible part of the spectrum (∼400–750 nm), a spectral camera also records other bands of the electromagnetic spectrum (e.g. infrared, ultraviolet). The types of instrumentation vary but most ultimately allow the user to analyze each pixel of the image of the artefact individually. This makes it possible to delimit different areas of the surface and compare their spectral features.

Regardless of the method used, the output of the spectroscopic analysis will be a multivariate dataset needing to be processed and modelled (Figure 4). Whilst the artefacts themselves can be linked to the geographical position of their discovery, the spectral data are generated from the interaction of the probe with the geological material. The features recorded in the data are a result of the petrogenetic environment in which the material originally formed, meaning that any inference of “space” is related to the geological source and not the archaeological activity per se (although the spectroscopic signal could potentially have been altered by environmental or anthropogenic processes). This relationship can be statistically tested by including reference samples from various sources in a multivariate model (e.g. Blomme, Degryse, Van Peer, & Elsen, 2012). Following this train of thought, a lithic artefact could thus be represented at, at least, two spatiotemporal scales at the same time. One of these would be represented by the geological material itself, the geographical position of the (potential) original source and its petrogenesis, which contributes to the chemical features recorded in the spectra (e.g. Hunt, 1977). The second would be the lithic artefact made from the material, which would move through a number of transformational steps, before its eventual deposition and later discovery (Conneller, 2011; Sellet, 1993).

Figure 4

A number of NIR measurements collected on different points of the same artefact, represented as spectra. Each measurement is represented by one line in the above diagram, where peaks indicate overtone regions of molecular vibration. Heterogeneous material might therefore exhibit differing levels of absorbance and produce slightly different sets of peaks depending on what part of the surface is measured. Image produced with R (Version 4.1.2; R Core Team, 2021) and the spectrolab package (Meireles, Schweiger, & Cavender-Bares, 2017).

The inference of large-scale organizational patterns is thus dependent on a number of interlinked models, with data being simplified and used in a number of analytical processes at different scales. What represents “data” for each level of analysis differs, and thorough documentation, or “metadata,” is necessary to contextualize the results. It is the responsibility of the archaeologist and analyst to be mindful of this and to be aware of the assumptions that are being made within each underlying model and how these represent various relationships in the large-scale spatial analysis.

4 Modelling as Visualization

In addition to testing hypotheses, models serve as a means of communicating research results. In this way, modelling is intrinsically linked to visualization, an area where GIS has been of considerable benefit. Whilst the results of an analysis can be represented numerically in the form of tables, summary statistics, and significance tests, it is far more common that archaeological data (including environmental data) are represented visually as maps at different geographical scales. Archaeologists are partial to the map-based view of the world, whether it be through the nodes and edges of a network (Pálsson, 2019) or a number of agents moving through a reconstructed environment (Baum et al., 2020). This view of the past, whilst perfectly suitable in many cases, can be somewhat restrictive. In his study of 18th century Iceland, Pálsson (2018) visualized land ownership by performing network analysis on surveyed farmsteads. Pálsson demonstrated that, whilst the network may give the appearance of a straightforward relationship between two farms, the actual relationship described in the historical documents could be of a much more dynamic nature (Pálsson, 2021). Whilst the network model provides a useful tool for analysis and visualization, deeper insights into human agency were gained through the underlying database, which manages the historical data. In other words, the networks represent a means of exploring the database and the information contained within it, rather than the database being a simple storage vessel for data (the same could be said for a simple spreadsheet).

The Icelandic case shares similarities with exploratory data analysis techniques found in multivariate statistics (Jebb, Parrigon, & Woo, 2017). As datasets grow and become increasingly complex, exploratory data analysis represents an invaluable method for generating hypotheses and making full use of the information stored within the data. This informal approach incentivises visual interaction with the data to explore and identify various points of interest, which can then be followed up with formal approaches. Exploratory data analysis is a valuable tool for analyzing the spectroscopic data used in petrographic characterization (Camizuli & Carranza, 2018; Sciuto, 2018). The spectral data represent a large multivariate quantitative dataset, with measurements recorded for a certain range of wavelengths of electromagnetic radiation. It has a tendency to include more information potential that can easily be comprehended without exploration and simplification. For instance, in near infrared (NIR) spectroscopy, the spectral features are related to combinations and overtones of molecular vibrations (Linderholm & Geladi, 2014a,b) (Figure 4). Through experimentation and comparison, certain features within the data have been linked to the different chemical characteristics of different materials (Hunt, 1977).

Whilst it may be possible to analyze a couple of spectra manually, once the dataset grows to be in the hundreds, it quickly becomes unfeasible. As in the Icelandic case (Pálsson, 2018) a more efficent approach is needed to cut through the data. Principal component analysis (PCA) is an example of an exploratory technique frequently used in chemometrics with spectral data (Geladi & Linderholm, 2020), but it also has a history of use within other areas of research, such as lithic morphometrics (e.g. Chacón, Détroit, Coudenneau, & Moncel, 2016; Forsberg, 1985) or archaeogenetics (Slatkin, 2016). The PCA models the variance within the data in an effort to reduce the complexity, from thousands of variables into a couple (usually 3–5), referred to as principal components (PC). In this way, it becomes possible to explore the structures in the data visually in the form of a biplot where the PCs form the axes, effectively creating a window into the multidimensional space (Eriksson, 2013). By plotting different PCs against each other, different groupings within the data, related to different spectral features, can be identified. Those features that can be related to significant chemical characteristics of the material will then form the basis of the classification of the objects (Sciuto, Geladi, La Rosa, Linderholm, & Thyrel, 2019). The process summarized above merely represents the first step in the study of raw material use among a hunter-gatherer population, i.e. the characterization of the geological material used. As the PCA represents an informal technique, the knowledge and expertise of the analyst is critical when exploring the PCA and identifying the clustering within the data. Whilst the process may not be streamlined to the same extent as when creating a DEM, this does not change the fact that in both cases, the same critical disposition should be taken in the evaluation of the resulting model.

Typically, the next step would be to classify the analyzed stone artefacts according to the groups identified within the PCA and examine the spatial distribution of these within GIS and its representaiton of the “geographical space”. Being able to study the spatiotemporal distribution of different material classes could prove useful in modelling the sourcing habits of Mesolithic hunter-gatherers; however, it would not communicate the potential uncertainties or nuance of the underlying petrographic model. In contrast to working within the “absolute space” described by Lock and Pouncett (2017), the axes in a PCA are defined by either quantitative data or qualitative concepts within “variable space.” The data are not divorced from geography, however, but can instead be used as an attribute such as in the case of Brandt, Szécsényi-Nagy, Roth, Alt, and Haak (2015). In their study of aDNA sequences from Europe during the Holocene, they visualized differences and similarities between different cultural and geographical groups as attributes in the PCA model. Whilst the different haplogroups are what drives the clustering within the “variable space” that is the PCA, visualizing the cultural and spatiotemporal entities in the symbology can aid in identifying different relationships and potential explanations for the clusters. The previous example of petrographic characterization can be approached in a similar way, for instance, grouping artefacts by typological category to examine how different tools might be related to different subgroups of raw material and variation over time.

Modelling data in the “variable space” is useful for exploring the structures of your data, and something most researchers will do to simplify it in preparation for spatial analysis. Concepts that may be difficult to manage within the formal environment, which the GIS often represents may be more manageable within this more informal representation of space.

5 Discussion

Whilst the term “modelling” may invoke certain technological and formal connotations, modelling in practice is in fact much more varied and reflective of archaeology as a whole (Nakoinz, 2018). In an effort to investigate various large-scale patterns of Mesolithic hunter-gatherers, archaeologists will inevitably engage in modelling at different scales. It may not be thought of as such, when creating a scatter plot based on the dimensions of a certain artefact type assemblage, but it is a model testing a relationship within your data. The abstracted results of the model can then be used in further studies of spatial patterns, thus adding to what is likely to be a chain of models. This article has been an attempt at highlighting this multiscale nature of modelling and the potential, as well as risks, this entails. The emphasis has been on GIS and digital approaches to modelling as these have undergone much development during the most recent decades. Modelling approaches can range from complex quantitative reconstructions and simulations on a landscape level to theoretical conceptual models describing workflows and relationships, all in an effort to gain a better understanding of the past.

Archaeology is, at its core, a destructive field dealing with a fragmentary material culture inventory that is fragmented further through excavation, in the sense that the physical context of the site will not be left post-excavation (unless strategies for minimizing impact are implemented). The documentation and collection of data is therefore vital, and as time goes on, these datasets continue to grow in size and new tools and methods are needed to store, structure, maintain, and analyze them. The issues associated with the management and dissemination of scientific data have been recognized by stakeholders, adding more pressure to make certain that published results are of a high quality (Wilkinson et al., 2016). The same care should be applied to the analysis and modelling of data as well. The interlinked nature of modelling makes it easy to introduce a number of uncertainties to the higher-level output of an analysis. To conduct research in an open and transparent manner, it is, therefore, critical that the analyst is mindful of and documents the manipulation of data at each step (Marwick et al., 2017).

GIS has been instrumental in the way in which archaeologists manage and interact with their data, and whilst a lot of the previously time-consuming tasks have been made more efficient, it may have come at the expense of insight into the way in which data are processed. When analyzing data within a point-and-click environment, it can be easy to lose track of exactly what is being done with the data. Likewise, whilst the layered structure of the GIS software has the benefit of isolating certain tasks and datasets, it is a somewhat inefficient structure for data management as you are essentially working with multiple isolated flat tables. Linking in outside resources, databases, and programming suites can do a lot to alleviate these issues. It also has the added benefit of providing the user with means of constructing more sophisticated queries based on their research question (e.g. Pálsson, 2021).

A core issue in debates on GIS and modelling is how more subjective and relative aspects of the past are handled (e.g. Gillings et al., 2020; Llobera, 2012; Lock, 1995; Lock & Pouncett, 2017; Whitley, 2017). GIS was developed with formal approaches in mind where “geographical space” is central to whatever question you are studying. The way in which it represents different environmental and cultural features is thus somewhat limited in comparison to the more dynamic, and sometimes subjective, real-world counterpart. There is a clear difference between the static and empirical representation of an environmental variable in a landscape reconstruction and how this might have been perceived by hunter-gatherers in the past. Computational simulation approaches hold an interesting potential in this aspect. Whilst they may not address some of the issues related to the representation of phenomenological features in geographical space, they do provide a means of investigating how such theoretical models express themselves in the landscape.

There is also a slight discrepancy between the level of detail at which environmental and archaeological data are recorded. Environmental analysis of samples collected in contexts meant to represent the surrounding environment (e.g. bogs and lakes) typically features high-resolution dating, sometimes at an annual scale. Compare this to the much more variable resolution of dating found at most archaeological sites, where an intact cultural stratigraphy is not always guaranteed and it can be difficult to link the dated material to different phases of human activity. In terms of spatial resolution, however, the inverse could be argued. Given the character of site formation and all the nuances linked to it, archaeologists have come to value spatial integrity and documentation highly. In contrast, environmental reconstructions are often at a more diffuse regional scale and strongly influenced by the nature of depositional contexts, which capture material (e.g. pollen, insects) from a wider, often difficult to define, area. There is, of course, significant variation between cases but it is nonetheless important to consider this contrast as the two are frequently intertwined in modelling.

Whilst the vision of a hyper-dynamic model, which is capable of visualizing spatiotemporal change at an extreme, high resolution might be tempting, the question is at what point such a model would stop being useful? One reason for the simplification of data is, after all, a way to manage the abundance of data, something that a highly precise model would undoubtedly produce. To build on the famous quote by Box and Draper (1987, p. 74), “…all models are wrong…how wrong do they have to be to not be useful?,” it would be prudent to also ask at what point does an accurate model stop producing meaningful information? That is to say, at what point will there be an “information overload”? A hypothetical model capable of visualizing environmental and cultural change at a second by second basis would likely make it extremely difficult to take in and process the information and may obscure the more general features of interest in the “noise” of short term variation. The phenomenon and changes archaeologists study are rarely at such high resolution, and the purpose of modelling is more often to understand general patterns than give accurate depictions of reality.

GIS represents a unique and integral part of the archaeological framework, and this is unlikely to change any time soon. It is important to recognize, however, that it also has its limitations and that at times it is necessary to look to other approaches. Rather than restricting the analysis within “geographical space,” certain spatial and temporal features can be represented in “variable space” in a meaningful way, as in the case of Brandt et al. (2015). It can be a way to avoid abstracting the data unnecessarily, as well as be more transparent about your research. Where GIS excels is at tasks that require geographical accuracy and manipulation of geographical data, but its strengths are extremely reliant on the quality of the underlying data (e.g. Buckland et al., 2018).

In summary, modelling is a complex and varied field where data are simplified for specific purposes. It can be easy to view it as a predominantly formal approach, but this may have more to do with the use of the term within certain areas of research. GIS has been of great benefit to modelling within archaeology, but it is at the same time limited by the ways in which it represents data. It is the responsibility of the analyst to consider whether the system is appropriate for the type of analysis they have in mind and to remember that there may be other means of exploring, testing, and visualizing their data. A model is only as strong as the sum of its components (data and models), and should any of these be flawed it will impact the overall performance. Instead of hurrying to engage in large-scale spatiotemporal analysis, it would thus be prudent to take a good look at the lower-level models, to ensure that they are of a good quality and represent the data accurately. Approaching the problem from a “non-spatial” perspective, with an exploratory methodology, could open up for innovative research that retains details, which could be lost through the simplification implicit in spatial and temporal aggregation.

Abbreviations

GIS

Geographical Information System

NIR

Near Infrared

PC

Principal Component

PCA

Principal Component Analysis