THE SYMBIOSIS OF LICHENOMETRY AND RADIOCARBON DATING: A BAYESIAN CHRONOLOGY OF ALPINE HUNTING IN COLORADO’S SOUTHERN ROCKY MOUNTAINS, USA

Abstract

Archaeologists keep a limited arsenal of methods for dating stone features at alpine sites. Radiocarbon (¹⁴C) dating is rarely possible, and it is common that dates do not accurately represent the activity of interest (stone feature construction). In this paper I review a legacy set of 89 ¹⁴C dates for stone driveline sites built by hunter-gatherers in Colorado’s Southern Rocky Mountains. I amend the sample of dates using chronometric hygiene and focus on dates with direct association to hunting features. I then present a newly calibrated set of 29 lichenometric dates for rock features at these sites and use hygiene protocols to remove inaccurate dates. Size-frequency lichenometry, though poorly known in archaeology, provides a way to date stone features indirectly by measuring the growth of long-lived lichens that colonize rock surfaces after construction events. Bayesian modeling of the combined set of dates suggests that the tradition of alpine game driving spans over 6000 years BP, with abundant use over the last 2000 years. Archaeologists must use multiple methods for dating stone features in alpine environments. This Bayesian analysis is a formal effort to combine lichenometry and ¹⁴C dating for archaeological interpretation.

INTRODUCTION

Hunter-gatherers regularly made use of lands above timberline throughout much of North America. At times, the alpine tundra served as a simple corridor for regional migrations through mountainous terrain, but in other instances nomadic peoples established specific places to carry out tasks on a more routine basis. Evidence for site reoccupation is well-expressed in the archaeology of stone drivelines in Colorado’s Southern Rocky Mountains. Native Americans built vast architectural networks of rock-walled fences, cairns, and hunting blinds along the open tundra of the Continental Divide to intercept migratory game such as bighorn sheep, elk, and deer (Benedict 1992). These sites resemble other examples of driveline architecture located throughout the world (Lemke 2021, 2022), including the inukshuk of Canada and Alaska as well as the “desert kite” phenomenon of the Levant, Arlo-Caspian region, and greater Arabian Peninsula (Benedict 2005a; Brink 2005; Crassard et al. 2015; Barge et al. 2016, 2022). In Colorado, there are more than 80 hunting sites with stone architecture above timberline (∼3500 m asl), which range in size and complexity, but the largest contain dozens of stone features used by communal hunting groups on a seasonal basis. The tradition of game driving is a critical variable for models of seasonal transhumance in the region (Black 1991; Benedict 1992, 1999; Brunswig et al. 2014), and the high density of sites suggests a local fluorescence of driveline technologies.

The chronology of driveline hunting in Colorado has never received formal analysis, but it is generally accepted that the tradition spans several millennia (Benedict 2005a:429). Artifacts documented during surface surveys indicate use of stone drivelines may have occurred as early as the Late Paleoindian period (e.g., Benedict 2000), but continued throughout the Archaic (7500–1800 BP) and Late Prehistoric (1800–400 BP) to the mid-to-late 1800s AD (Cassells 1995; LaBelle and Pelton 2013; Whittenburg 2017; Meyer 2021). The legacy set of radiocarbon (¹⁴C) dates mirrors this long span of use, but many of the dated samples lack clear association to game drive features. For several sites, lichenometric dating of stone walls helps to strengthen chronologies that suffer from a shortage of ¹⁴C dates (Benedict 1975a; Benedict 1985; Hutchinson 1990; Cassells 1995; Benedict 1996; LaBelle and Pelton 2013; Meyer 2021). The size-frequency method of lichenometry accounts for the growth rate of Rhizocarpon rhizocarpon (hereafter Rhizocarpon sp.) that colonize stone walls following construction events (Benedict 1985; Benedict 2009; Cassells 2012).

In this paper I build a chronology of alpine hunting in the Southern Rocky Mountains from the existing sample of ¹⁴C dates and lichenometric dates. I follow a basic chronometric hygiene protocol (sensu Graf 2009; Pettitt et al. 2003) to remove problematic dates that have poor association with features and insufficient numbers of thalli measurements for accurate lichenometric dating. I recalibrate new lichenometric dates for game drive walls using a recently revised version of the Colorado Front Range growth curve for Rhizocarpon sp. (Meyer 2021). I then use Bayesian chronological models to account for statistical uncertainty within the reliable set of dates, including samples with inbuilt age issues (Bronk Ramsey 2009a; Bronk Ramsey 2009b; Dee and Bronk Ramsey 2014). I conclude with several implications of the chronological model as well as recommendations for future research.

MATERIALS AND METHODS

Radiocarbon Dates

The complete set of ¹⁴C dates includes 89 dates from 17 driveline sites (Benedict 1975a, 1975b, 1978, 1979, 1996, 2000; Hutchinson 1990; Cassells 1995; Benedict and Cassells 2000; Brunswig 2005; LaBelle and Pelton 2013; Whittenburg 2017; Meyer 2021). The drivelines cover more than 38,000 km² in the Southern Rockies, but the Front Range massif and the Sawatch Range contain dense concentrations of dated sites (Figure 1). The spatial distribution of drivelines with absolute dates parallels the history and intensity of game drive research in the region as well as legitimate differences in site construction methods (Benedict 1992; LaBelle and Pelton 2013). Sites with dates include more than 40 driveline features on average, whereas undated sites may contain a single wall, blind, or cairn line suggesting construction by small hunting groups. Driveline features, particularly hunting blinds, serve as potential reservoirs to capture organic materials deposited during occupation events. Higher quantities of these features within individual sites improves chances for successful ¹⁴C dating. The chronology considered in this paper captures only one facet of game driving behavior in the region and potentially omits processes underlying the construction of smaller sites which are more difficult to date with absolute methods.

Figure 1 Distribution of hunting architecture above 3500 m asl in the Southern Rocky Mountains, Colorado. Inset LiDAR relief image (right) depicts the main intercept area at the Olson game drive, showing walls and blinds. Numbered sites include Trail Ridge (1), Flattop Mountain (2), Sawtooth (3), Blue Lake Valley (4), Murray, Hungry Whistler, 5BL68 (5), Arapaho Pass (6), Devil’s Thumb Pass (7-8), Devil’s Thumb Valley (9), Bob Lake (10), 5GA35 (11), High Grade (12), Olson (13), Water Dog Divide (14), 5CF499 (15).

The archaeological context of ¹⁴C dates varies widely between sites, as does the material sampled to produce dates. Dates on wood charcoal are the most abundant, consisting of bulk or single-grain samples of unidentified charred wood as well as individual burned twigs and needles identified to genus (Abies, Picea, and Pinus). Sample recovery methods include full or partial excavation of hunting blind pit floors (e.g., Cassells 1995), non-invasive soil coring of sediment within blinds (e.g., Benedict 1996), and excavation of external camps presumably associated with hunting features (e.g., Benedict 1978, 2000). The set of ¹⁴C dates includes samples from faunal remains as well, represented by elements of Ovis canadensis (bighorn sheep) and Odocoileus sp. (deer). Bones have been recovered from site surfaces but also within the floors and walls of hunting blind pits (Benedict 1975a; Hutchinson 1990; Cassells 1995; LaBelle and Pelton 2013; Meyer 2021).

Chronometric hygiene of the ¹⁴C dataset is not as simple as removing dates on wood charcoal with large standard deviations, though this issue is a factor. Both human and non-human agents introduced charcoal into alpine archaeological sites over time. Eolian deposition of forest-fire charcoal into sedimentary matrix is well-documented for alpine environments (e.g., Novák et al. 2010; Tinner et al. 2006), and this includes peak elevations along the Continental Divide (Benedict 2002). Benedict (2002) conducted a study in which he observed that culturally derived charcoal deposits often contain wood grains larger than 3 mm in diameter, but none of the published soil core information states the actual size of grains or weight of charcoal submitted for analysis. Because of this, Benedict (2002:35) questioned the validity of many ¹⁴C dates that he produced on charcoal flecks from alpine drivelines in the region, where he stated:

The depressed centers and low peripheral walls of prehistoric hunting blinds make the blinds excellent sediment traps, vulnerable to contamination by windblown charcoal. Several AMS dates recently obtained by coring blinds at Front Range game-drive sites (Benedict 1996, 2000; Benedict and Cassells 2000) are suspect for this reason.

Periglacial mass wasting skews stratigraphic sequences when sediment is available at sites. Charcoal can migrate through site deposits quickly because of continuous surface erosion, frost-heaving, and bioturbation. Several inverted sequences of charcoal dates at game drives suggest these natural processes are common (Hutchinson 1990; Cassells 1995; Benedict 2000; Brunswig 2005), and this is a globally documented issue for alpine environments in general (Payette and Gagnon 1985; Carcaillet 2001). The most reliable archaeological contexts for charcoal samples include thermal features identified by the cross-section of excavated hearths and related ash layers. Archaeologists have documented hearths within prepared hunting blind pit floors, which represent either pre-hunt or post-hunt use of features for non-hunting functions (Benedict 1975a; Hutchinson 1990; Cassells 1995; Benedict 2000; LaBelle and Pelton 2013). Many hunting blinds do not contain hearths and were built primarily to conceal hunters armed with arrows as well as atlatl darts (Figure 2).

Figure 2 Collapsed hunting blind at the High Grade game drive, Rollins Pass, Colorado. Native Americans constructed the blind by excavating a flat pit floor and stacking several courses of stone in the direction of the game intercept area.

The chronometric hygiene protocol that I followed is minimalistic by comparison to the standards of other studies which numerically scored individual dates (e.g., Douglass et al. 2019; Graf 2009; Napolitano et al. 2019). I simply emphasized that “clean” charcoal dates must at least have a clear association with hunting blinds and possibly represent an in situ burning event. The most reliable samples include 13 charcoal dates produced from hearths and ash layers in excavated hunting blinds (Table 1). The targeted event for charcoal dating (hearth ignition) does not directly relate to the use of stone drivelines, but the dates may at least be considered a minimum age for the hunting blinds which enclose hearth features. I also included dates made on charred materials from slotted-tube soil core samples in blinds (n=27) given the possibility that they could represent cultural burning events (Table 1). However, I emphasize here that some or all of the charcoal from soil probes could be windblown deposits or derive from some other context.

Table 1 List of 40 modeled ¹⁴C dates and feature contexts from alpine driveline sites in Colorado. See Appendix 1–4 for additional information about the complete set of radiocarbon dates.

Appendices 1–4 list the remaining dates from sites and features that I excluded from the analysis based on contextual issues, sample types, and large standard lab errors. This includes 11 charcoal dates from hunting blinds which were fully excavated and did not contain any thermal features (loose charcoal of unknown context), given that the excavations confirmed the absence of in situ burning in those specific blinds (Appendix 1). A total of 29 dates from general excavation areas were also removed (Appendix 2), including campsite locales from three sites (Trail Ridge, Devil’s Thumb Valley, and Hungry Whistler), due to lacking horizontal or stratigraphic association with hunting features. At the Devil’s Thumb Valley site (5GA3440), Benedict (2000:62–63) specifically described complications with the dates from the Area A campsite locality that are especially pertinent to their removal from this study:

Most charred material was from tree trunks and branches, but some was from roots that had smoldered belowground. Some was slaggy and vesicular, indicating quenching by water while sap was still stewing from the wood. Twenty-one radiocarbon dates, all attributable to wildfire, were obtained from Area A. The dates range from 9570 +/- 80 BP (Beta-122996) to 2880 +/- 60 BP (Beta-98414).

Many sites with driveline features show evidence for a variety of other non-hunting activities, including toolmaking and repair (e.g., Whittenburg 2017), plant processing (Cassells 1995; LaBelle and Pelton 2013), ceremonial fasting (Benedict 1987; Brunswig 2005), and residential behavior (Benedict 1978; Benedict 2000). These activities demonstrate the multi-faceted nature of alpine occupations, but ¹⁴C dates from these activities are not necessarily relevant for the chronology of driveline hunting. Benedict (2005a:429) mentioned one additional charcoal date (Beta-44747) from a hunting blind but the report does not include an associated site or feature, so I omitted it from analysis (Appendix 1). For dates on bone collagen, I rejected samples from site surfaces (n=5), given that natural death events can occur on sites without human predation (Appendix 3). Freeze-thaw cycles in the alpine may significantly alter bone surfaces when left exposed (Bertran et al. 2015), which limits the reliability of surface-collected bones for dating anthropogenic deposits.

Problematic charcoal dates with large errors can be informative for Bayesian modeling (see Hamilton and Krus 2018), but the dates should at least be accurate estimations of cultural events. In this study I eliminated charcoal dates with standard uncertainties greater than 100 years with missing information about sample selection and processing (Appendix 4). In the alpine, three-digit uncertainties for charcoal dates may be the result of mixed wood pieces with various degrees of inbuilt age (combinations of short-lived elements, old wood, and forest-fire debris), or perhaps very small sample quantities used in gas-proportional or liquid scintillation techniques prior to AMS ¹⁴C dating (Spriggs 1989; Graf 2009; Gragson and Thompson 2022). Several of the ¹⁴C dates were published in analyses of legacy collections which occurred decades after initial field collection, storage, and processing (e.g., LaBelle and Pelton 2013; Whittenburg 2017), which prohibited analysis of potential contaminants. The most reliable set of ¹⁴C dates includes a total of 40 dates from 11 sites after chronometric hygiene (Table 1), or roughly 45 percent of the composite set of dates.

Lichenometric Dates

Lichenometry is not a well-known method in archaeology, but it is frequently used by geologists to date the redeposition of rock substrata resulting from glacial processes (Bickerton and Matthews 1993; Roberts et al. 2010; Wiles et al. 2010), earthquakes (Emerman 2017), rockslides (Winchester and Chaujar 2002), flood discharges (Foulds and Macklin 2016), and other natural events. Benedict (1985, 1996, 2009) pioneered the use of the size-frequency method of lichenometry to date cultural stone features in the Front Range. Globally, there is abundant research on lichenometric dating with Rhizocarpon sp. and several key studies review the historical development and implementation of these numeric dating methods (Innes 1983; Loso and Doak 2006; Jomelli et al. 2007; Benedict 2009; Armstrong 2016), as well as criticisms leveled against specific types of lichenometric dating such as the maximum diameter technique (Jomelli et al. 2007; Osborn et al. 2015; Rosenwinkel et al. 2015).

Yellow members of Rhizocarpon sp. are crustose lichens that represent a symbiotic relationship between fungal mats and patches of algae (Figure 3). They quickly colonize exposed rock surfaces and make thalli with distinguishable edges. Individual thalli may grow for thousands of years at a near-linear rate if left undisturbed, making Rhizocarpon sp. a reliable lichen for numeric dating. The predictability of their growth depends largely on local climate variables (Loso and Doak 2006; Benedict 2009; Armstrong 2016), including effective moisture, snow cover, temperature, altitude, slope and aspect, sunlight, and air pollution as documented in recent studies (Armstrong 2016). When rocks are overturned because of natural or anthropogenic disturbances, existing colonies of yellow Rhizocarpons die off and their thalli spall from rocks within a period of about 10 years (Benedict 2009:151). Thalli may survive disturbance events if rocks are not completely overturned, and this prohibits the use of the largest observed thallus on cultural features for reliable dating. The growth and colonization rate of new colonists is highly distinguishable from survivors, however. Comprehensive random sampling of thalli diameters shows a significant negative log-linear relationship in the size and frequency of new thalli which grew after disturbance events (Benedict 2009: Figure 14; Loso and Doak 2006: Figure 2; Roberts et al. 2010:Figure 6). In the Colorado Front Range, Pearson’s correlation coefficient values typically average 0.99 for regression lines run through data points representing the size and frequency of new colonists (Benedict 2009). Survivor measurements, on the other hand, show a clear break from this negative log-linear relationship in thalli size and frequency; survivor diameters are too large and sample sizes are either too few or too abundant depending on the intensity of the rock disturbance event.

Figure 3 Yellow Rhizocarpon sp. thallus photographed by J. Benedict at Ouzel Lake in Rocky Mountain National Park, Colorado. (Please see online version for color figures.)

Numeric dating of lichen colonies requires a locally engineered age-growth calibration curve. Benedict (2009) constructed the Front Range growth curve for Rhizocarpon sp. by sampling large random sets of thalli diameters on disturbed rock substrata with associated ¹⁴C dates. The curve uses control points built from regression line slope values fit to post-disturbance colonist diameter measurements at each substratum, followed by associated radiometric or historically known ages for the disturbance events (Benedict 1985). Several researchers have used the calibration curve to produce age estimates for driveline features by measuring thalli that grow on stone walls and predicting numeric ages based on intercepts with the curve (Benedict 1985, 1996, 2000; Hutchinson 1990; Cassells 1995; Benedict and Cassells 2000; Benedict 2009; LaBelle and Pelton 2013; Meyer 2021). A revised version of the curve gives standard uncertainties for lichenometric dates in cal BP (Meyer 2021:Supplemental Materials 1), based on recalibration of ¹⁴C dates for each control point in the curve using IntCal20 (Reimer et al. 2020).

I present revised lichenometric dates for 29 features from nine alpine driveline sites (Figure 4, Table 2). The dates include both 1σ (68.3%) and 2σ (95.4%) credible ranges as well as median dates. The precision of the revised dates ranges widely, with 1σ uncertainty ranges spanning anywhere from 30 to 330 years and taking slightly asymmetrical probability distribution shapes. Wide probability ranges result from the shape of the curve and the coarse level of precision for conventional ¹⁴C dates used for the curve control points. I apply chronometric hygiene to the set of lichenometric dates by excluding dates that are potentially inaccurate. Benedict (2009: Figure 17) simulated the effect of low sample sizes on lichenometric dating accuracy by comparing the maximum discrepancy in predicted ages for contemporaneous walls at Arapaho Pass, suggesting that age estimates stabilize after about 1000 measurements. The reliable set of lichenometric dates includes 22 dates from seven alpine sites after removing walls and blinds with low sample sizes of Rhizocarpon sp. I supply individual date estimates and curve parameters using R code language in Supplementary Material 1.

Figure 4 Revised age-growth calibration curve for Rhizocarpon sp. in the Colorado Front Range. Blue crosses indicate curve intercepts using the slope of regression lines for Rhizocarpon sp. thalli growing on driveline wall features. Black dots represent curve control points (rounded to the nearest 10 years), based on recalibration from Meyer (2021).

Table 2 Complete list of size-frequency lichenometric dates from alpine driveline sites in Colorado, based on recalibration with the revised age-growth curve for Rhizocarpon sp in the Colorado Front Range (Meyer 2021).

Bayesian Modeling

Bayesian chronological modeling can improve our understanding of statistical uncertainty in calibrated date estimates (Buck et al. 1996; Bayliss 2009; Bronk Ramsey 2009a). This is especially important for the study of alpine drivelines, where the reliable set of ¹⁴C dates and lichenometric dates is small and the precision of dates is generally poor. Visual interpretation of dates or “eyeballing” is unlikely to provide a clear understanding of the spread of dated events or underlying patterns of site use over time (Hamilton and Krus 2018). Bayesian modeling improves interpretations of dated event uncertainties by accounting for departures in the ¹⁴C calibration curve with the use of prior information from archaeological context (Bronk Ramsey 2009a).

I calibrated and modeled all dates using Oxcal v.4.4 (Bronk Ramsey 2009a). The essential structure of Model A consists of a bounded Phase for the game drive tradition with nested overlapping subphases that correspond to individual sites. I applied a simple uniform Boundary to the beginning and end of the phase to characterize the probability distributions for unsampled events at the onset and conclusion of driveline hunting in the region. In this case, the uniform prior assumes that dated events from all game drive sites are a random selection of a uniformly distributed process which characterizes the overall chronology. This choice of boundary does not favor long or short phase lengths, nor does it impose significant bias on the presumed shape of the modeled probabilities; the uniform boundary is a “weakly” informative prior and is appropriate due to the poor understanding of processes influencing game drive construction, use, and reuse over time (Bayliss 2015; Taylor et al. 2017). I did not use additional boundaries on site subphases to avoid over-engineering the marginal posterior distributions of dated events. This was a practical choice given that some sites in the sample contain only one or very few reliable dates, which limits the effectiveness of site-by-site comparisons. There is no information to suggest site-specific factors in the overall temporality of the game drive tradition, but this could be explored with additional modeling studies.

Lichenometric dates were modeled alongside ¹⁴C dates (R_Dates) in site subphases using the C_Date command, based on the median of predicted ages and 1σ uncertainties. Two error terms were included with the lichenometric dates to account for the slightly asymmetrical shape of each probability distribution (e.g., C_Date(“Olson_Wall1”, calBP(1301), +142, –137)). I implemented a terminus ante quem (TAQ) constraint on the game drive phase using the Before command. The TAQ date of CE 1880 represents the forced removal of the White River Ute (Yamparika and Parianuche) and Tabeguache Ute from Colorado immediately following the Meeker Massacre (AD September–October 1979). Ethnohistoric accounts suggest Yamparika Ute hunting parties occupied areas in the Front Range up until AD 1875 (Brunswig 2020:145; Simmons 2000), but events following the Meeker Massacre would have restricted access to game drives in the region.

I included additional parameters in the Bayesian model to query the modeled chronology. The Interval command was used to construct a probability distribution for the duration of the game drive phase. This command modeled the spread of dated events and undated events between the phase boundaries, which provided an indication for the total extent of time that game drive sites were used in the region. I then implemented the KDE_Plot function using the default settings in OxCal to illustrate the underlying density of the marginal posteriors of events within the game drive phase. This method is a combination of a frequentist and Bayesian approach to summing modeled events, whereby Kernel Density Estimation (KDE) averages the Bayesian likelihoods and priors generated from the Markov chain Monte Carlo (MCMC) ensembles (Bronk Ramsey 2017). KDE smoothing of event densities within the Bayesian model helped to reduce artificial noise resulting from calibration effects, sample sizes, and the spread of posterior uncertainties.

I constructed a supplementary Bayesian model (Model B) which augments the primary model structure and includes additional information about dated samples, making it the preferred model for archaeological interpretation (Figure 5). The fundamental difference between Model B and Model A is that I did not apply ¹⁴C dates collected from soil core samples, which removed a significant number of dates from the analysis. Some of these soil core dates could represent in situ anthropogenic burning events, but there is no independent test available to prove or disprove this possibility. Model B applies a very strict interpretation of archaeological context, one which does not allow for guesswork about the origin of charcoal in hunting blinds when there is no additional information to evaluate each sample. The model only considers dates collected from clearly identifiable thermal features in hunting blinds and the sample of lichenometric dates.

Figure 5 Model A Baysian structure, including modeled start (green) and end (red) boundaries for the game drive tradition phase.

In Model B I combined several outlier models (Bronk Ramsey 2009b; Dee and Bronk Ramsey 2014), including an “old wood” model for dated samples with inbuilt age issues (bulk charcoal, other large wood fragments, or unknown samples) and a general outlier model for other dates (lichenometric, bone collagen, charred twigs and needles). A series of simulations were then used to examine the sensitivity of the Model B output (Griffiths 2014; Holland-Lulewicz and Ritchison 2021). I tested for model reproducibility by randomly sampling a range (population mean (μ) ± 1σ) within the marginal posteriors of dated events from Model B. This simulation method determines if the number and precision of dates distributed throughout the modeled phase is sufficient to produce a consistent model output when calendar dates and ranges are varied (Meadows et al. 2020:1275). The R_Simulate and C_Simulate commands were used to generate random ¹⁴C dates and lichenometric dates in place of expected calendar dates in each site subphase. These simulated dates were then included in a uniform phase model using the same priors and constraints as Model B.

I applied additional simulations to the existing set of dates from Model B (with flagged outliers manually removed) to reveal the effect of increasing sample sizes of high-precision ¹⁴C dates. This method examines how robust the model output is to the addition of new high-quality dates, and in effect, reveals the minimum number of new dates needed to increase precision for estimates of the start and end boundaries for the game drive phase (Holland-Lulewicz and Ritchison 2021). For this test I generated random sample sets of simulated ¹⁴C dates and constrained the date ranges for expected calendar dates based on the results of Model B. I provided a constant error of ± 20 years for each simulated ¹⁴C date, assuming lab errors with high precision. Each iteration of the model was run 10 times with the same simulated dates and actual Model B dates, but new random sets of simulated dates were generated with sequential iterations of the model. Samples sizes increased by 10 for each model iteration. I accepted the results of the simulation tests when sequential iterations produced start and end boundaries that exhibited diminishing returns for improving precision (Holland-Lulewicz and Ritchison 2021:276–277). Model code and simulation code are provided in Supplementary Material 2.

RESULTS

Modeled dates and ranges are presented in italics and rounded to the nearest five years, including 68.3% and 95.4% credible ranges as well as median dates (see Supplementary Materials 3 and 4 for complete model results). Model A passed with good agreement (A_model=102.4, A_overall=101.4) and the results suggest that alpine hunting with drivelines spanned a period of 5620–6020 years, beginning 6120–5750 cal BP and ending 145-70 cal BP (95.4% credible range). These ranges encompass the Early Archaic period (7500–5000 BP) and span through the Middle Archaic (5000–3000 BP), Late Archaic (3000–1800 BP), Late Prehistoric (1800–400 BP), as well as the Protohistoric (400–100 BP) (Gilmore 1999; Tate 1999). The KDE plot displays a highly non-random spread of events between the start and end boundaries, however. The distribution shows very low quantities of dates between 6000 and 2000 cal BP and then rises exponentially, suggesting the assumption of uniformity in modeled events is inaccurate (Figure 6). In practical terms, the results of Model A suggest that hunters may have used sites infrequently during an experimental period within the Early Archaic and again in the Middle Archaic. The latter end of the phase indicates abundant use of sites after 2000 years ago with the Archaic-Late Prehistoric transition.

Figure 6 KDE plots of the uniform phase model (Model A) and Bayesian model with outlier analysis (Model B). Gray crosses represent calibrated median dates of unmodeled events, and black crosses show the medians of modeled posteriors. Bars underneath modeled distributions represent 68.3 (upper) and 95.4 (lower) credible ranges for the start boundary (green) and end boundary (red) for the game drive tradition phase.

Figure 7 Modeled duration (interval) for the alpine game drive tradition in Colorado based on Model A and Model B results.

The results of Model B reflect the high density of dated events after 2000 years ago by downweighing the effect of outliers and soil core samples which may not be cultural in origin. The “old wood” outlier model shifts the modeled posteriors of problematic charcoal dates (bulk samples from hearths) towards non-outlier date estimates with better precision (charred twigs, burned needles, and bone). The values of the outlier shift are randomly selected by an exponential distribution calculated during MCMC ensembles which consider all the dates in the model (Bronk Ramsey 2009b; Dee and Bronk Ramsey 2014). Model B results suggest that driveline hunting started 2040–1575 cal BP and ended sometime between 155–65 cal BP (95.4% credible range), with an interval period of 1460–1930 years at the 2σ credible range (Figure 7). The ranges of these dates correspond to the end of the Late Archaic period and span several sequential periods in the Late Prehistoric era (Gilmore 1999; Tate 1999), including the Early Ceramic period (1800–800 BP) and Middle Ceramic (800–400 BP). The tradition ended during the Protohistoric period in northern Colorado (400–100 BP) when Euroamerican operations in the Colorado Front Range grew rapidly (Clark 1999). The output of the KDE plot shows an approximately uniform distribution, but there is a peak in the density of dated events between 700–650 BP which corresponds to the end of the Early Ceramic and the onset of the Middle Ceramic period (Figure 6).

Model Sensitivity

The initial set of simulations used random calendar dates and uncertainties in place of actual modeled events from Model B. Ten runs of the simulation revealed that the output of Model B is generally reproducible, but the low quantity and poor precision of dates towards the beginning of the phase causes greater variation in results than the end of the phase. Simulation runs favored solutions roughly one or two centuries younger than the Model B start boundary, with maximum estimates ranging 1950 cal BP to 1730 cal BP and minimum start dates between 1510 cal BP and 1355 cal BP (95.4% credible range). For the end boundary, simulation results consistently supported estimates from Model B. Repeated runs of the simulation produced a range of 195 cal BP and 145 cal BP for the maximum date of the ending boundary, and 70 to 65 cal BP for the minimum end (95.4% credible range). These results reflect the relative quality of dates for the youngest modeled events in the sample, but also the TAQ constraint on the end of the phase which restricts excessive spread in modeled uncertainties.

Additional simulation tests revealed effects from increasing sample sizes of randomly generated high-precision ¹⁴C dates. The results demonstrated that overall variance of Model B decreases with the addition of new ¹⁴C dates while model precision simultaneously increases (Figure 8). Differences between the maximum and minimum date estimate for the Model B start boundary shortened by roughly two centuries with as few as 10 new ¹⁴C dates, from 445 yrs to 250 yrs based on simulation averages (95.4% credible range). Precision of the start boundary steadily improved with sequential iterations but started to stabilize with simulations of 60 random ¹⁴C dates. The most precise simulations produced differences of 100 yrs for the maximum and minimum estimates of the start boundary, which occurred during simulations of 80 and 90 new ¹⁴C dates. Precision and variance of the modeled end boundary went essentially unchanged over the course of simulation runs. The difference between the maximum and minimum dates improved by 25 yrs with the addition of 70 new ¹⁴C dates (135–70 cal BP), but simulations favored solutions of 155–70 cal BP overall. These results further demonstrated that the younger end of the phase is well-sampled up to the TAQ constraint (AD 1880) on the ending boundary.

DISCUSSION

Model A displays a much broader temporal range for the duration of the game drive tradition phase than Model B (a difference of 4090 years between the median interval estimates of the two models). This is primarily due to the inclusion of dates from soil cores in Model A, which are treated as evidence for in situ anthropogenic fires. Model A reaffirms a long-held belief that the onset of the game drive tradition developed prominently throughout the Early Archaic period with the Mount Albion complex (Benedict 1978), and this was a process that continued with increasing frequency into the Late Prehistoric and Protohistoric periods. Model B, on the other hand, applies a stricter approach to chronological hygiene which focuses only on ¹⁴C samples from well-defined hearths inside hunting blinds and lichenometric dates on rock walls. Model B may be too strict, however, and the rejection of certain soil core dates from the chronology may falsely truncate the earlier age of the game drive tradition phase – especially dates representing the Early Archaic and Middle Archaic periods. Ultimately, the reader must choose which model best fits their own interpretation of the archaeological record given the immense variation in reliable sample types. I believe Model B is the most defensible result given its exclusive focus on cultural events.

Late Paleoindian-aged dates are notably absent from the modeled Bayesian chronologies in this paper, and it is worth exploring this issue with fine detail. Archaeologists have documented Late Paleoindian projectile points as surface finds from Flattop Mountain and Trail Ridge in Rocky Mountain National Park, including “Yuma”, Eden, Allen, and Foothills-Mountain complex types (Benedict 1996; Brunswig 2005; LaBelle personal communication). However, the most tantalizing and simultaneously problematic evidence for Late Paleoindian use of game drives comes from the Devil’s Thumb Valley site (Benedict 2000), which is the only game drive site that has produced Late Paleoindian-aged ¹⁴C dates.

The Devil’s Thumb Valley driveline consists of several short walls, cairn lines, and blinds that hunter-gatherers used to capture animals between a narrow bedrock trough as they travelled between a high mountain pass and a well-watered valley floor in the Indian Peaks Wilderness. The site connects with another ¹⁴C dated driveline, 5BL103, located beneath Devil’s Thumb Pass. Benedict (2000) collected a Foothills-Mountain projectile point fragment from the surface of the site in the tailings of a gopher hole near Blind 3, found at the terminal end of Drive Line C on the valley floor near spruce-fir krummholz and several wetland deposits. Benedict excavated the hunting blind and produced a ¹⁴C date of 2155 ± BP from a charcoal stain in the center of the feature, representing a Late Archaic-aged date. He continued his search for Paleoindian materials within several block excavations at Area A and Area B, which represented surface lithic scatters near the Foothills-Mountain point fragment. Benedict produced two ¹⁴C dates from a single hearth feature in Area B, dated 2250 ± 70 BP and 2160 ± 60 BP, again representing a Late Archaic age. I previously described the 21 non-cultural dates from Area A, which ranged from approximately 9000 to 3000 BP, and Benedict (2000:69) determined “At least six, and probably seven, wildfires affected this small tract of forest-tundra ecotone during the Holocene” based on his analysis of the clustering of the ¹⁴C dates.

When Benedict’s attempts to excavate and absolutely date the Late Paleoindian component at Devil’s Thumb Valley failed, he turned his attention to alternative dating methods. He applied a granodiorite weathering technique to the driveline walls which accounts for the “…grain by grain disintegration that causes an initially smooth rock to become rough to the touch.” (Benedict 2000:80). He proposed that the weathering profile of rock walls changed significantly after wall construction events, and that disturbances over time would be visible based on differential weathering between rock surfaces in the walls and the background slope of natural granite. It is unclear exactly how Benedict determined the degree of rock surface weathering quantitatively, other than noting a general appearance of the stones, and he did not provide usable data to reproduce his estimates that the Drive Line C wall dated between 12,000 and 10,000 years ago. He ultimately conceded that the granodiorite rate curve “…itself is preliminary, with too few control points to provide accurate dates in this early time range. Thus, construction during the Early Archaic Period cannot be completely excluded.” (Benedict 2000:81–82). Given the totality of the information at hand, it remains unclear whether the driveline is Late Paleoindian-aged. There is very little data in support of a testable hypothesis concerning the earliest use of the site, but Benedict’s attempts at dating did confirm a Late Archaic presence that is supported by ¹⁴C dates in hunting blinds as well as diagnostic projectile points found in several other lithic scatters closely adjacent to the blinds and walls (Benedict 2000:Figure 2.14, Figure 2.19).

The Devil’s Thumb Valley driveline is one of several quintessential sites showing the complexity of palimpsest deposits in the alpine tundra. Archaeologists must decide whether to link artifacts from different ages found on site surfaces to driveline features located nearby, or attempt dating of features directly via selective sampling. Ultimately, such practices in combination with one another may not yield perfectly digestible or expected results. Taphonomic bias is not a fully sufficient explanation for the lack of Late Paleoindian ¹⁴C ages in secure context at drivelines, however. The set of dates from Devil’s Thumb Valley proves that there is at least some degree of preservation of Late Paleoindian-aged materials near hunting features, suggesting that natural landscape changes over time did not completely erase organic materials from the early Holocene. Researchers have dated Late Paleoindian occupations elsewhere in the Southern Rocky Mountains at the subalpine/alpine ecotone (Pitblado 2000; Benedict 2005b; Brunswig and Doerner 2021), which demonstrate the importance of high-altitude landscapes for peoples living 10 ka-7500 BP. I do not question the technological capabilities of Late Paleoindian groups; sites of this age throughout the Rocky Mountain region show a diverse range of subsistence strategies involving communal hunting and intercept tactics (Morris 1990; Pitblado 1999; Kornfeld and Larson 2008; Lee and Puseman 2017). I do question whether there is sufficient evidence to support a hypothesis of Late Paleoindian-aged construction of drivelines in the Southern Rocky Mountains, given the results of chronological hygiene and Bayesian modeling. The modeled chronology presented in this paper is conservative and does not consider the age and affiliation of more than 60 other hunting sites with dry-laid features in the region, which could be much older, but these sites lack formal chronological research.

It is more difficult to rebuke the potential relationship between hunter-gatherers of the Early Archaic Mount Albion complex and stone drivelines in the Southern Rocky Mountains. In the Bayesian analysis of Model A, this time frame corresponds with the earliest portion of the game drive tradition phase which is supported by low quantities of ¹⁴C dates collected by soil cores in blinds from several sites in the sample. Mount Albion projectile points have been found across the surface of numerous game drives in the Southern Rockies (LaBelle and Pelton 2013), including the recently published High Grade site (Meyer 2019, 2021). It is remarkable, however, that Mount Albion points have not been reported in direct association with driveline intercept areas for sites with demonstrated survey coverage (e.g., Meyer 2019). Seven of the game drive sites in the sample have produced time-diagnostic artifacts within hunting blinds (Table 3), amounting to 55 projectile points and one glass trade bead. Whittenburg (2017:45, 48–49) reported a tentative Mount Albion point from Blind 573 at the 5GA35 site, but a ¹⁴C date of 3090 ± 250 BP from the feature is roughly 2000 years later than the accepted range for the Mount Albion complex. Other early projectile point styles from blinds include possible Duncan-Hanna types based on fragments found in Blind D-6 at the Sawtooth Game Drive (5BL55), which date regionally to the latter end of the Middle Archaic (Cassells 1995). The overwhelming majority of projectile points from hunting blinds date to the Early Ceramic period, represented by at least 37 Hogback phase corner-notched arrows. At the Murray site, Benedict (1975a) excavated 15 Hogback corner-notched points along with a thermal feature and other tools in Blind 1 (Figure 9), suggesting gear caching behavior, a discarded toolkit, a votive offering, or frequent site revisits during the Early Ceramic period.

Table 3 Time-diagnostic projectile points and unspecified point types collected during excavation of hunting blinds pits at alpine drivelines in Colorado.

¹ McKean or Duncan-Hanna dart (5350–2950 BP): (Metcalf 1973; Morris et al. 1985).

² Pelican Lake dart (3200–1720 BP): (Todd et al. 2001; LaBelle and Pelton 2013).

³ Hogback corner-notched arrow (1350–950 BP): (Nelson 1971; Benedict 1975b, 1975a).

⁴ Unspecified side-notched arrow.

⁵ Plains side-notched arrow (850–150 BP): (Kornfeld et al. 2016).

⁶ Plains tri-notched arrow (350–150 BP): (Reher and Frison 1980; Kornfeld et al. 2016).

⁷ Small white glass seed bead (115–100 BP): (Von Wedell 2011; Newton 2016).

⁸ Unspecified tri-notched arrow.

Figure 8 Variance of differences (yrs) between the maximum and minimum estimates (95.4% credible range) for the modeled start boundary of the game drive phase based on sequential simulation runs with increasing sample sizes of random ¹⁴C dates. The quantity of randomized dates increased by 10 for each iteration, and the simulations were run 10 times.

Figure 9 Early Ceramic period Hogback corner-notched projectile points from Blind 1 at the Murray site.

Hungry Whistler (5BL67) is both the type site for the Mount Albion tradition and the best possible case demonstrating the connection between Early Archaic hunter-gatherers and driveline sites in the region (Benedict 1978). The site is situated between three other driveline complexes, including the Murray site (5BL65) and 5BL68 which date to the Early Ceramic and Middle Ceramic periods, or roughly 3000 years later than Hungry Whistler. It is a unique site given that it is made up primarily of very low-lying and dispersed cairns as opposed to aggregated walls and hunting blinds like other formal driveline complexes. There is a v-shaped arrangement to the features like most drivelines, with a principal cairn wall that leads to a concentration of several tree islands at timberline. Only two hunting blinds were noted at the site, downslope of the primary concentration of features. Benedict (1978) did not directly date any of the hunting features, hence the exclusion of the dates from this analysis, but they did excavate two open lithic campsites in close vicinity to the game drive on flat benches at the western and eastern flanks of the drive system. In addition to Mount Albion complex materials, the team observed projectile points from later occupations including untyped stemmed, shouldered, and small lanceolate varieties comparable to McKean and Duncan-Hanna types (Benedict 1978:72).

Benedict produced five ¹⁴C dates from hearths and stains at the closest campsite to the principal cairn line at Hungry Whistler (roughly 30 m away) which spanned 5800–4010 BP (Benedict 1978:26), a roughly 2000-year period. Four of these dates have standard uncertainties greater than 100 years, which also excluded them from the Bayesian analysis in this paper. The team observed three cairns in the excavation area which were embedded near the contact between stratigraphic units containing the hearths (about 15 cm beneath the modern surface), suggesting a potential relationship between the driveline features, hearths, and artifacts. However, the excavation team also documented intense vertical mixing of deposits because of the Triple Lakes Stade Neoglaciation as well as late Neoglacial frost disturbance, visible through the reactivation of sorted nets and dilation cracks onsite (Benedict 1978:40). The extreme periglacial environment at Hungry Whistler allows for some speculation about spatial relationships between cultural materials, and it is worth introducing Benedict’s own interpretations to avoid biasing the reader. Regarding neoglacial frost disturbance (frost-heaving), Benedict (1978:73) stated:

Because of vertical mixing and the probability that each of several prehistoric groups visited the terrace made use of most of its limited level surface area, it is impossible to relate generalized butchering with grinding tools to specific projectile point styles or radiocarbon ages.

The team did demonstrate that cairns within the excavation area closely coincided with hearth features, but they also revealed that “with a single exception, hearths at the site were modified so strongly by early Neoglacial frost disturbance that their original characteristics could not be determined.” (Benedict 1978:45). Only the youngest basin hearth, dated to 4010 ± 90 BP (after Mount Albion), escaped significant modification by periglacial processes and this feature is in the same stratigraphic unit as the other dated hearths. I think it is reasonable to assume, based on the information presented, that the driveline features could post-date the Mount Albion component and perhaps relate to later components present on the site. Earlier in the report, Benedict (1978:10) also conceded that it is “…uncertain whether the wall and multiple cairn lines are part of a single drive unit, or whether different systems of different ages are superimposed.”

CONCLUSION

The Bayesian analysis, including Model A and Model B, overwhelmingly supports construction and use of stone drivelines beginning at the end of the Late Archaic period and continuing to the mid-to-late 1800s AD. This portion of the modeled chronology spans several regional technological traditions that are distinguishable based on changes in projectile point designs (Nelson 1971; Metcalf 1973; Morris et al. 1985; Gilmore 1999; Todd et al. 2001; Kornfeld et al. 2016), but also the emergence and proliferation of ceramic styles (Butler 1988; Owenby et al. 2021), rock oven technologies (Troyer 2014; Hedlund 2019), residential architecture (Cassells and Farrington 1986; Perlmutter 2015; Brunswig 2016), and mortuary practices (Gilmore 2008). Importantly, researchers have linked rapid changes in material culture to population expansion in northern Colorado at the end of the Archaic era (e.g., Gilmore 2008), and consequently, adjustments to landscape use stemming from population pressure between hunter-gatherer bands living in the Southern Rocky Mountains. The results of this study strongly suggest that the alpine game driving phenomenon is an additional development within a larger suite of socio-technological answers to demographic changes in the region. The game drive tradition most likely ended when the United States government began to enact legislations that forced Native Americans from their traditional hunting grounds at high altitudes (Simmons 2000; Brunswig 2020). More precise modeling of regional technological transitions and correlations with the game drive tradition will require chronological hygiene of the regional ¹⁴C dataset, implementing improved methods for summing ¹⁴C dates, and potentially revising chronologies by using Bayesian estimation as a statistically valid framework. Similar revisions of the regional climate record and its effect on high-altitude occupations may prove useful for interpreting the ebb and flow of alpine driveline construction and use over the last 2000 years (Benedict 1999).

New dating efforts at these sites must prioritize modern methods for ¹⁴C sample selection and processing as well as independent validation of lichenometric dating techniques. The results of the sensitivity analyses revealed that model precision can be markedly improved with the addition of as few as 10 new ¹⁴C dates from good context. Researchers should focus on measuring new random sample sets of Rhizocarpon sp. on walls with existing lichenometric dates to determine if dates are reproducible within an acceptable margin of statistical error. Researchers can revise the calibrated error of lichenometric dates with the addition of new control points in the age-growth calibration curve, and by redating existing control points. Archaeologists should also consider constructing independent calibration curves in other mountain ranges to better understand local climate effects on Rhizocarpon sp. growth over time, as well as the spatial range limits of accurate date predictions with calibration curves.

Additional prior information is needed to improve Bayesian modeling of alpine driveline sites. Specifically, archaeologists must prioritize establishing occupation sequences within individual sites which may be used as informative prior information to constrain date uncertainties and thus improve model precision. This will be a difficult task given that vertical separation of materials in alpine sites is often weak, mixed, or totally absent. The Devil’s Thumb Valley and Hungry Whistler sites are perfect examples of this persistent issue. Spatial statistical modeling and clustering algorithms may prove useful for grouping together sets of dated features based on functional relationships (i.e., clusters of hunting blinds and walls within sites). Such methods should be applied to surface distributions of artifacts, which may provide evidence to link projectile points to driveline intercept areas. Additionally, archaeologists should pursue new simulation studies involving comparisons with other technological traditions that span the Bayesian chronology presented here. Archaeologists may wish to explore new Bayesian models testing whether these weapon technologies overlapped in time at alpine driveline sites, or if intervals of site abandonment occurred between the development of new weapon technologies.