Archaeological Models

Archaeological models have been used for decades in CRM to guide planning efforts and help make management decisions. A major advantage of using archaeological models in creative mitigation is that they can leverage available data, expert knowledge, and insight to address the challenges faced today in CRM. These include not having the time and resources to know where important resources are located, how important they are, what condition they are in, where they are vulnerable to adverse effects, and what to do to investigate and preserve them. Archaeological models can thus be used to guide decisions about where and how to apply creative mitigation efforts. For example, archaeological models can be used to

• • Depict the archaeological sensitivity of broad areas to planners and the public, without having to disclose individual site locations;

• • Estimate the cost of inventory, evaluation, and/or data recovery efforts and decide on appropriate investigative methods;

• • Identify where particular types of sites of interest to managers, researchers, or the public are likely to be located and where sites are likely to be surficial or buried;

• • Identify where additional effort is needed or resources are vulnerable to ongoing or anticipated disturbance processes;

• • Develop historic contexts and identify factors that influenced settlement and land use through time;

• • Design sampling strategies for research, inventory, condition assessments, and archaeological testing programs; and

• • Integrate cultural and natural resource management concerns in landscape-level planning, synthesis, and conservation.

Most CRM practitioners are familiar with archaeological locational models (ALMs). ALMs predict the archaeological sensitivity of a study area, usually by dividing it into a series of sensitivity zones, such as those of low, moderate, or high sensitivity (Figure 1). A key advantage of an ALM is that it can predict where sites, or sites of a particular type, are likely to be located in unsurveyed or undersurveyed areas, allowing planners and developers to manage risk, minimize adverse effects, and assess the costs and benefits of project alternatives early in the planning process (Altschul et al. Reference Altschul, Sebastian and Heidelberg2004, Reference Altschul, Klein and Sebastian2005; Green et al. Reference Green, Altschul, Heilen, Cushman, Homburg, Trampier, Hayden and Lerch2012; Ingbar et al. Reference Ingbar, Sebastian, Altschul, Hopkins, Eckerle, Robinson, Finley, Hall, Hayden, Rohe, Seaman, Taddie and Thompson2005; Judge and Sebastian Reference Judge and Sebastian1988; Kamermans et al. Reference Kamermans, van Leusen and Verhagen2009).

FIGURE 1. Archaeological locational model integrating surface and subsurface models created for Eglin Air Force Base, Florida (Green et al. Reference Green, Altschul, Heilen, Cushman, Homburg, Trampier, Hayden and Lerch2012).

Models can be based on theory, empirical relationships, simulation, or a combination of these approaches. Empirical models generally do not explain archaeological phenomena. They describe them objectively based on statistical analysis of a set of variables and cases using a variety of techniques. Many ALMs make use of logistic regression analysis, which results in an equation that estimates the probability across a study area of a site being found in a particular location. In recent years, machine learning has proved to be a powerful approach to developing models in archaeology and other fields, using algorithms such as random forests (Breiman Reference Breiman2001; Green et al. Reference Green, Altschul, Heilen, Cushman, Homburg, Trampier, Hayden and Lerch2012; Prasad et al. Reference Prasad, Iverson and Liaw2006). A variety of statistics, such as true positive rate, false positive rate, area under the receiver operating characteristics curve, gain, and gain over random, are used to test model performance (Kvamme Reference Kvamme, Judge and Sebastian1988b).

Many ALMs make use of proxy variables that are theoretically related to the behaviors and affordances leading to site formation. Proxy variables quantify access to resources and landscape affordances, such as topographic position, exposure to wind or solar insolation, visibility, plant habitats, agriculturally productive soils, lithic sources, or transportation systems. Variables can be calculated at a variety of scales that align with the multiple scales at which people interact with landscapes. Experimental variables can also be created to isolate specific affordances that influenced behavior and site formation (Heilen and Windingstad Reference Heilen and Windingstad2017; McDonald et al. Reference McDonald, Bullard, Britt, Ruiz, Caldwell, Ehlen and Harmon2004; Nuninger et al. Reference Nuninger, Verhagen, Bertoncello and Barba2014; Verhagen and Drăguţ Reference Verhagen and Drăguţ2012).

Variables that predict aspects of mobility can also be useful in predicting site location. For example, White and Barber (Reference White and Barber2012; see also White Reference White, White and Surface-Evans2012, Reference White2015) found that major urban centers in Oaxaca, Mexico, were located in areas predicted to contain the most intensively traveled routes in the region. To model site location in Naval Air Weapons Station China Lake, an arid area of the Mojave Desert, where mobility and land use was tethered to springs, Windingstad and I calculated least-cost paths between several hundred spring locations in the mountains surrounding our study area and used the results to estimate the intensity of travel along path segments (Heilen and Windingstad Reference Heilen and Windingstad2017). Path proximity and use intensity proved important in predicting the location of resource processing sites and temporary camps that may have been used during travel between springs. In a similar fashion, historical-period transportation routes and route intersections can be useful in predicting historical period site location, as transportation networks often influence the spatial configuration of feature systems and economic interactions (Hardesty Reference Hardesty1988; Heilen et al. Reference Heilen, Altschul, Reddy, Heckman and Norris2015; Verhagen et al. Reference Verhagen, Nuninger, Groenhuijzen, Verhagen, Joyce and Groenhuijzen2019).

Variables used in modeling can be further leveraged in creative mitigation efforts to learn something new about sites and their environment, such as how sites are related to each other and their environment. In the same way that landscape ecologists experiment with landscape metrics to study the ecological dimensions of landscapes (Cushman et al. Reference Cushman, McGarigal and Neel2008), variables developed as part of modeling efforts can be used to study how landscape characteristics influence settlement behavior, land use, interaction, sustainability, and demography.

Improvements in the quality and resolution of multispectral satellite imagery and remote sensing data have allowed researchers to experiment with another empirical approach in predicting archaeological site location, referred to as “direct detection” modeling (Chen et al. Reference Chen, Comer, Priebe, Sussman, Tilton, Comer and Harrower2013; Comer and Harrower Reference Comer and Harrower2013; Megarry et al. Reference Megarry, Cooney, Comer and Priebe2016). Direct detection modeling uses image classification methods to discretely identify possible sites, much in the way that models of natural resources use spectral signatures to identify particular minerals or plant communities in the production of forestry, critical habitat, or geologic maps. In a similar fashion, the Arizona Army National Guard is exploring methods to rapidly identify resource vulnerabilities and minimize site disturbance through the use of remote sensing data, natural resource indicators, and machine learning techniques. The goal is to predict the presence or absence of subsurface cultural deposits in areas where ground disturbance or other adverse effects are anticipated, using minimally invasive techniques (Manney et al. Reference Manney, Cushman, Kaszta, Wan, Heilen, Rego and Klucas2020).

In contrast to empirical models, deductive models attempt to explain archaeological phenomena by using theoretical expectations to predict patterns of land use and mobility and testing how well those predictions reproduce observed patterns. Deductive models include using foraging theory and ecological principles to model the temporal and spatial distribution of valued ecological resources, seasonally and with environmental change, to predict where activities resulting in site formation would have been concentrated (Reeder-Myers Reference Reeder-Myers2014; Verhagen and Whitley Reference Verhagen and Whitley2012; Whitley Reference Whitley, Thompson and Thomas2013, Reference Whitley2016). Deductive approaches have also been used to anticipate the location, depth, and thickness of buried cultural deposits (Tucker, Lancaster, Gasparini, and Bras Reference Tucker, Lancaster, Gasparini, Bras, Harmon and Doe2001; Tucker, Lancaster, Gasparini, Bras, and Rybarczyk Reference Tucker, Lancaster, Gasparini, Bras and Rybarczyk2001; Zeidler Reference Zeidler2001). Models that combine empiricism with theory and insight are likely to be more convincing to stakeholders than are purely deductive or empirical models alone (Green et al. Reference Green, Altschul, Heilen, Cushman, Homburg, Trampier, Hayden and Lerch2012). Simulation modeling could bridge the gap between empirical and deductive approaches (Axelrod Reference Axelrod and Rennard2006; Kohler and Varien Reference Kohler and Varien2012; Romanowska et al. Reference Romanowska, Crabtree, Harris and Davies2019). Simulation models start with a set of simplified formal rules for how a cultural system, economic activity, or other process operates and then simulate how those rules play out in a dynamic system under varying conditions, using approaches such as agent-based modeling (Perreault and Brantingham Reference Perreault and Brantingham2011; Premo Reference Premo2006, Reference Premo, Mesoudi and Aoki2015). Simulation modeling has been applied to a wide array of archaeological problems but has thus far been used rarely in developing ALMs, although ALM applications show promise and can be richly informative (Davies et al. Reference Davies, Romanowska, Harris and Crabtree2019; Kohler and Varien Reference Kohler and Varien2012).

While ALMs have been the main focus of modeling efforts in CRM, several types of models address additional management needs. These include buried site potential models (Bullard et al. Reference Bullard, McDonald, Sabol, Schumer, Bacon, Baker, Caldwell, Dalldorf, Jenkins, Kreidberg and Minor2011; Eckerle et al. Reference Eckerle, Ingbar, Taddie, Finley, Drews, Hopkins, Seddon, Roberts and Ahlstrom2011; Green et al. Reference Green, Altschul, Heilen, Cushman, Homburg, Trampier, Hayden and Lerch2012; Heilen and Windingstad Reference Heilen and Windingstad2017; Meyer et al. Reference Meyer, Young and Rosenthal2010, Reference Meyer, Rosenthal, Byrd and Craig Young2011; Monaghan and Lovis Reference Monaghan and Lovis2005), significance models (Cushman and Sebastian Reference Cushman and Sebastian2008; Doelle et al. Reference Doelle, Barker, Cushman, Heilen, Herhahn and Rieth2016; Fatoríc and Seekamp Reference Fatoríc and Seekamp2018; Heilen et al. Reference Heilen, Sebastian, Altschul, Leckman and Byrd2012, Reference Heilen, Altschul, Reddy, Heckman and Norris2015; McManamon et al. Reference McManamon, Doershuk, Lipe, McCulloch, Polglase, Schlanger, Sebastian and Sullivan2016; Sebastian Reference Sebastian, Sebastian and Lipe2009), vulnerability models (Anderson et al. Reference Anderson, Bissett, Yerka, Wells, Kansa, Kansa, Noack Myers, Carl DeMuth and White2017; Dore and Wandsnider Reference Dore, Wandsnider, Mehrer and Wescott2006; Heilen et al. Reference Heilen, Altschul and Lüth2018; Nicu Reference Nicu2017a, Reference Nicu2017b; Reeder-Myers Reference Reeder-Myers2015), agricultural suitability models (Gillreath-Brown and Bocinsky Reference Gillreath-Brown and Bocinsky2017; Gillreath-Brown et al. Reference Gillreath-Brown, Nagaoka and Wolverton2019; Homburg et al. Reference Homburg, Heilen, Leckman, Murrell and Vierra2014, Reference Homburg, Casey, Heilen and Vanderpot2017), and cultural landscape models (Barrett et al. Reference Barrett, Dingwall, Gaffney, Fitch, Huckerby, Maguire, Dingwall and Gaffney2007; Fairclough Reference Fairclough, Palang and Fry2003; Heilen and Altschul Reference Heilen and Altschul2019). Each of these kinds of models has a place in creative mitigation efforts by forecasting the nature and distribution of the heritage resources so that they can be effectively identified, categorized, sampled, preserved, and managed and their values can be enjoyed by researchers, traditional communities, economic sectors, and the public. In other words, models can provide a framework for guiding decisions on how, where, and when to undertake creative mitigation efforts.

Modeling is best viewed as a process that begins with the development of a baseline model that is iteratively refined with improvements in data, techniques, knowledge, and insight (Altschul Reference Altschul, James Judge and Sebastian1988; Green et al. Reference Green, Altschul, Heilen, Cushman, Homburg, Trampier, Hayden and Lerch2012). The accuracy and precision of ALMs have improved in recent years with the advent of geographic information science, advances in statistical procedures and information systems, and the increasing availability and quality of environmental data used in establishing correlations and making predictions (Kvamme Reference Kvamme and Schiffer1989, Reference Kvamme1999; Lock and Stančič Reference Lock and Stančič1995; Mehrer and Wescott Reference Mehrer and Wescott2006). During their use in CRM, ALMs have fallen in and out of favor, largely due to concerns that models can be inaccurate and result in mistakes, such as in a case where a model predicts the absence of a site when, in fact, a site is present. Another disadvantage is that the available data used to develop models are often generated over the course of many disjointed projects undertaken without a cohesive sampling strategy, methodology, or research focus (Verhagen and Whitley Reference Verhagen and Whitley2012). Variation in the quality, completeness, and standardization of available data (Ortman et al. Reference Ortman, Varien and Gripp2007) can introduce bias and error in model results. Deficiencies in CRM data, whatever their source, can lead to bias and error whether they are used in modeling or in other applications. Is it modeling that is at fault or the underlying data? If the latter, what can be done to improve the utility of existing data and the quality of data collected in the future?

Concerns regarding data quality and representativeness can be addressed as part of a modeling project. Modeling, for example, may lead to important information about where there are data gaps, where methods have worked or failed, where there is interobserver error, where greater standardization or detail is needed in recording data, and how disparate datasets or typologies can be folded into a common ontology (Doelle et al. Reference Doelle, Barker, Cushman, Heilen, Herhahn and Rieth2016; Ortman et al. Reference Ortman, Varien and Gripp2007). Modeling can lead to the compilation, organization, and synthesis of data and information that, prior to modeling efforts, existed as isolated datasets and dots on a map. In planning creative mitigation efforts, modeling can help to connect the dots and generate meaningful information, insight, and dialogue through goal-oriented interaction with cumulative CRM datasets. Moreover, using Section 110 responsibilities to conduct stratified sample survey or eligibility evaluations can improve model performance and improve knowledge of regional site distributions and characteristics. Instead of using CRM data mostly as an administrative record of individual sites and projects, modeling and synthesis of CRM data can be used in creative ways to advance research, preservation, and public benefit. CRM practitioners can learn from what has failed or worked in the past in order to improve preservation outcomes.

Combining Modeling Approaches for Priority Planning and Long-Term Preservation

My colleagues and I have proposed that several different types of models could be used together to help decision makers plan for mitigating the effects of climate change on heritage resources (Heilen et al. Reference Heilen, Altschul and Lüth2018). Climate change and development are likely to damage or destroy large numbers of important resources, many before they are discovered and recorded. For example, Anderson and colleagues (Reference Anderson, Bissett, Yerka, Wells, Kansa, Kansa, Noack Myers, Carl DeMuth and White2017) estimated that more than 19,000 recorded sites in the southeastern United States will be underwater as a result of sea level rise and coastal flooding by the end of the century, conservatively assuming a 1 m rise in sea level. The majority of resources that will be adversely affected by sea level rise have yet to be identified and recorded, let alone made subject to research or preservation efforts (Erlandson Reference Erlandson2008, Reference Erlandson2010). Another study used the state database for Georgia (Williams et al. Reference Williams, Turck and Chamblee2010) and National Oceanic and Atmospheric Administration models of coastal inundation to show not only that large numbers of recorded and unrecorded sites on the Georgia coast will be affected in the near future by coastal flooding and erosion (see, e.g., Alexander et al. Reference Alexander, Robinson, Jackson, McCabe and Crass2008; Robinson et al. Reference Robinson, Alexander, Jackson, McCabe and Crass2010) but that the most vulnerable sites are also the kinds of sites that are most highly valued by stakeholders (Heilen and Altschul Reference Heilen and Altschul2016). Due to the scale of anticipated effects and the incomplete state of survey, my colleagues and I recommended that site location, buried site potential, significance, and vulnerability be modeled to identify hot spots where important resources are vulnerable to adverse effects (Heilen et al. Reference Heilen, Altschul and Lüth2018). This information could then be used in creative mitigation efforts to prioritize identification, preservation, and research of vulnerable sites before the opportunity is lost.

Another impending threat to heritage research and preservation is urban sprawl and infrastructure development. Modeling of time series satellite data has predicted that urban sprawl in the southeastern United States will expand by two to three times over the next five decades, causing major impacts to habitats and ecosystems that support threatened and endangered species and provide essential ecosystem services (Terando et al. Reference Terando, Costanza, Belyea, Dunn, McKerrow and Collazo2014). Cultural resources will also be affected. Altschul and I used time series land cover data (1974–2008; Natural Resources Spatial Analysis Lab 2020) and CRM data for the state of Georgia to examine the relationship between development and site condition (Heilen and Altschul Reference Heilen and Altschul2016). Between 1974 and 2008, 2.4 million acres of land in Georgia were converted to light urban (or suburban) development, and another 470,000 acres were converted to high-intensity urban development. Analysis of the correspondence between land cover and preservation status for recorded sites showed that site condition has deteriorated with increased development (Table 1). Retrodiction analysis showed that documented sites have increasingly fallen within urbanized areas since 1974, particularly within areas of low-intensity urban development (Figure 2). This trend is likely to continue in the coming decades.

FIGURE 2. Retrodiction of land use category for sites recorded in Georgia in the 2015 state cultural resource management database, based on the Georgia Land Use Trend data (Heilen and Altschul Reference Heilen and Altschul2016).

TABLE 1. Percent Deviation from Expected Number of Sites in a Land Use Category, according to the Preservation Status of Archaeological Sites Recorded in Georgia.

To mitigate adverse effects and foster long-term preservation, creative mitigation efforts should be integrated at a landscape level with natural resource management approaches. A framework that could facilitate an integrative approach to creative mitigation is the Landscape Conservation Cooperative, established by the U.S. Department of the Interior in 2009. The program was designed to foster effective conservation and management of landscapes and seascapes that will sustain cultural and natural resources over the long term. The Landscape Conservation Cooperative is a cross-jurisdictional, collaborative framework that is intended to employ the best available science along with experience and traditional knowledge in landscape conservation (National Academies of Sciences, Engineering, and Medicine 2016). Thus far, the program has placed little emphasis on integrating cultural resources into landscape-level management initiatives and has focused primarily on natural resource concerns. The ways that natural resource management efforts identify critical habitats and conservation priorities at a landscape level are closely aligned with archaeological modeling frameworks, however. Archaeological modeling is a key way to proactively integrate cultural and natural resource planning at a landscape level (Heilen and Altschul Reference Heilen and Altschul2019). Leveraging and revising state databases using a problem-oriented approach that folds consideration of cultural resources into landscape conservation design will also be important (Murtha et al. Reference Murtha, Lawres, Mazurczyk and Brown2019). Doing so would allow cultural and natural resources to be considered together in landscape-level conservation and management and provide an arena for collaboration and cross-fertilization among social science and environmental disciplines.

Efforts to integrate and synthesize CRM, natural resource management, and traditional knowledge are rare but are certainly possible. For example, the Coalition for Archaeological Synthesis is sponsoring two proof-of-concept projects that are investigating coupled human-natural systems to address pressing scientific and public policy concerns (Altschul and Kintigh Reference Altschul and Kintigh2019). The first, referred to as the ArchaeoEcology Project, is organized by Stefani Crabtree of Pennsylvania State University. The project is integrating archaeological, ethnographic, ecological, and other data in a series of case studies to reconstruct deep-time ecological networks that will be used to investigate how humans can sustainably interact with the ecosystems on which they depend. The second project, organized by Evan Larson of the University of Wisconsin–Platteville, addresses how fires are managed in the borderlands region of the United States and Canada (Larson et al. Reference Larson Evan, Kipfmueller and Johnson2020). Current approaches to fire management rely on the concept of “wilderness,” ignoring the pivotal role that humans have traditionally played in managing ecosystems. The result of the current management approach has been larger, more frequent, and more destructive forest fires. To learn how people have managed fire in the past, the project is investigating long-term interactions among people, landscapes, and fire through collaborative synthesis of traditional knowledge with archaeological, ethnographic, forestry, and tree-ring data. CRM data are being used to reconstruct past land use and human-environment interactions, and their use in the project has also inspired new approaches to combining traditional and scientific perspectives in support of more effective stewardship (Altschul and Kintigh Reference Altschul and Kintigh2019).