Changing with the times: From agricultural potential to spatially explicit reconstructions of past land use

Abstract

Past land-use reconstructions are a key tool for studying long-term human ecodynamics and addressing pressing questions about the origins and evolutionary dynamics of the Anthropocene. In particular, agricultural landcover reconstructions are vital for understanding long-term human-environment dynamics. Most past agricultural land-use models, however, rely heavily on modelling assumptions, make limited use of known archaeological site locations to constrain or inform their estimates and tend to be limited to general estimates of percentages of plant types within catchments around pollen trapping lakes. The lack of information outside catchment areas and low spatial resolution even within catchments constrain the utility of these models. To address this problem, we propose a new approach that combines archaeological predictive modelling with pollen-based agricultural landcover reconstructions to produce more accurate, spatially explicit past landcover estimates. Here, we present the results of a case study deploying the new approach to produce improved past landcover maps for a region in the Western Taurus Mountains, southwestern Turkey. The study area surrounds Sagalassos, an antique urban centre with a regional settlement history encompassing nine millennia. We produced five archaeological predictive models for the study region using the ‘Locally Adaptive Models of Archaeological Potential’ (LAMAP) method spanning the Hellenistic through Late Ottoman period. We then combined those predictive surfaces with ‘Regional Estimates of VEgetation Abundance from Large Sites’ (REVEALS) pollen landcover reconstructions for the same periods based on pollen from sediment cores extracted from three catchments within the study area. Lastly, we compared the resulting hybrid landcover models to the archaeological record using data not used to make the predictions. We found that the hybrid landcover model aligned very well with the known extents of agricultural land-use from the study area. These results indicate that the proposed approach is a viable way to combine pollen-based landcover models with archaeological data and produce more accurate, empirically-based landcover reconstructions reflecting real human activity in the past.

Keywords

agricultural potential land use predictive modelling

Introduction

In recent decades, the Eastern Mediterranean has experienced a progressive decline in agricultural production and loss of agricultural land as a result of climate change (Alrteimei et al., 2022). Based on current trends, projections of future agricultural loss for the region are dramatic (Kairis et al., 2022; Zittis et al., 2022). Turkey in particular has experienced severe effects of climate change on its proportion of agricultural land, being exacerbated by specific cropland management and land-use strategies including a rapidly urbanizing countryside (Gosling et al., 2011; Karapinar and Özertan, 2020; Özdoğan, 2011; Ozkan and Akcaoz, 2002). The current entanglement of regional climate change and local environmental degradation driven by the impact of land-use practices does not mark the first time when such conditions have necessitated considerations of reorganization of productive land use. The deep-time nature of archaeological datasets provides a unique opportunity to contribute to the discussion on the nature of sustainability of land-use practices in the context of environmental change and pressures such as reduced water availability and aridification, loss of topsoil and erosion, and shifting growing seasons (Jacobson, 2022; Silva et al., 2022). This paper aims to contribute to this special issue dedicated to resilience and sustainability in socio-ecological systems by presenting a method for reconstructing past cropland for examining the reciprocally influential relationship between environmental change and human land-use behaviour in a mountainous Eastern Mediterranean agriculturally productive landscape.

Regional climate oscillations, changes in water availability, land modification and/or degradation, the introduction of innovations in agricultural technologies, crop variety and management strategies and other factors affect agricultural potentials and land use. Present and future land-use behaviour may well be informed by the success and failure of adaptive systems of the past with regard to sustainability. Here we will present a novel and scalable interdisciplinary approach to combining archaeological and ecological methods for reconstructing past land-use practices as represented by cereal cropland distribution. Fundamental to any consideration of the relationship between social and environmental phenomena is a basic understanding of the conditions surrounding each component with an adequate degree of reliability. Fields concerned with the archaeological and paleoenvironmental record inherently present fragmentary information from which to interpret, and therefore the elaboration of these data through extrapolative and interpolative means is often a key element within this process. Beginning in the middle of the twentieth century and steadily increasing in prevalence since then, computational methods for the processing, elaboration and analysis of data have furthered possibilities and efficiencies (Chisholm, 1962; Gattiglia, 2015; Higgs and Vita-Finzi, 1972; Hodder and Orton, 1976; Huggett, 2022; Kohler, 1988; Warren, 1990; Willey, 1953, 1956). Despite these innovations, or sometimes encouraged by them, certain underlying paradigms have persisted that continue to orient research design towards either environmentally deterministic or socially contingent models of human-environmental interaction (Arponen et al., 2019; Hodder and Hutson, 2003; Kristiansen, 2019).

When it comes to the reconstruction of the past environmental conditions which feature in these considerations, this issue is also pervasive. Paleoenvironmental models tend to use the available abiotic and biotic evidence to reconstruct environments around archaeological sites in discrete slices of time and correlate those conditions with archaeological evidence under causal assumptions (Contreras, 2016; Hodder and Hutson, 2003). The ecological record and the archaeological record are treated as separate but parallel lines of evidence with points of interaction rather than as interwoven and mutually dependent histories. Large-scale paleoenvironmental reconstructions that account for the agency of social actors at the foundational level have for the most part failed to materialize in a substantial way. In this paper, we will outline a novel methodology for paleoenvironmental landcover reconstruction – cropland, specifically – which offers a possible remedy to this problem, using the case of the 1450 km² Sagalassos study region located in southwestern Turkey (Figure 1).

Figure 1.

Location of study region in the Western Taurus mountains, pollen sampling areas mention: (1) Ağlasun valley, (2) Gravgaz marsh and (3) Bereket valley.

This paper will begin by presenting our methodological aims, followed by a general consideration of modelling the spatial distributions of past vegetation, our methods, and the first results of the alternative approach being offered here. The methodology will be applied to the partial reconstruction of the vegetative regime of the paleoenvironment of the study area during five archaeological periods: Achaemenid-Hellenistic (546–25 BCE), Roman Imperial (25 BCE–300 CE), Late Antique (300–700 CE), Byzantine (700–1200 CE) and Late Ottoman (1700–1921 CE) – specifically the distribution of past cropland in three valley catchment systems (Figure 1: pollen sample areas 1–3). The paper will conclude by evaluating the results of the case study and the wider potential of the methodology to contribute to the discipline of archaeology and paleoenvironmental studies.

Aims

This research aims to introduce a methodology for the reconstruction of landcover that is both strongly data-driven and firmly-rooted in the traditions of social behaviour observable in the archaeological record. Through these reconstructions, we offer a tool for assessing trends in human impact on the environment, human-environment interaction, and the resilience and sustainability of land-use systems in an area of interest over time. As a necessary step in the creation of these novel landcover reconstructions, this research undertook the creation of reliable models of archaeological potential trained with known-site sets comprised of agricultural settlements for the study region (see below). These aimed to facilitate the reconstruction of cropland by modelling suitability of land parcels for agriculture through time, based on the known-site set.

Modelling spatial distributions of past vegetation

The relationships that exist between the settlement patterns observed so far in the Sagalassos study region and the surrounding environments represent the basis for the modelling efforts that follow. As will be discussed in further detail below, the environmental variables that factor into these particular models comprise mainly topographical and geomorphological physiognomies, but the range of possible characteristics relating to the landscape that might be employed in another context or to suit a different research agenda are numerous. These include any evidence of the climatic and environmental conditions available, which in the case of archaeological contexts is through indirect means via proxy indicators, such as sedimentological and stratigraphical indicators, and biotic indicators like pollen, phytoliths and faunal remains.

Methods for modelling spatial distributions of past landcover are less numerous than those for non-spatially explicit quantitative reconstruction of climate variables. Yet changes in landcover are essential indicators of human impact on an environment, climate change and landscape affordances. Changes in landcover due to human land uses have been shown to have drastic effects on local ecosystems (Galicia et al., 2007; Lambin et al., 2003), and to be a strong indicator of shifting population pressures (Beach et al., 2006) and degrees of agricultural intensification (Geist and Lambin, 2001). Understanding the relationships between landcover, human land use and environmental consequences is a topic relevant to a diverse range of research fields. For these reasons, the interest in modelling changes in landcover based on proxy evidence of past vegetation has persisted over the past few decades (Chang-Martínez et al., 2015).

The majority of efforts to model past landcover and landcover change have been concerned with estimating the role of changing landcover on the atmospheric concentration of greenhouse gases. They have adapted a variety of statistical methods, and generally use archaeological and historical datasets to map population densities and vegetation types over large super-regional or global surfaces (e.g. Foley et al., 2005; Goldewijk, 2001; Horvath et al., 2019; Houghton, 1999; Pirzamanbein et al., 2014). Models such as the historic database of the global environment (HYDE; Goldewijk, 2001) and the anthropomorphic land cover change database (ALCC; Pongratz et al., 2008) use historic data on population size and locations with estimations of the land requirements to meet the agricultural demands for specific societies assuming a linear relationship to predict the extents and general distribution of croplands. Non-linear models drawing on similar datasets have also been created where agricultural and pastoral land is assigned based on climate and soil characteristics (e.g. Kaplan et al., 2009), but the predictions of landcover for specific parcels remain low resolution and macro in scale.

On the other hand, models that are primarily concerned with the human-environment relationship itself, with landcover change as a constituent factor, are less common and tend to more often employ agent-based approaches (e.g. Axtell et al., 2002; Griffin and Stanish, 2007; Kohler et al., 2012; Murphy, 2012). These reconstructions tend to operate at higher resolutions and on smaller scales. The inputs in these cases are usually archaeological datasets and they attempt to model the land-use behaviour of individuals or communities over short periods of time due to the complexity of the datasets. Agent-based models tend to focus on interactions between agents, within a given set of environmental parameters, and model changes in the landscape as a result of those interactions. These models tend to be spatially limited, as the specific parameters governing interactions are not applicable across large and/or heterogeneous areas.

Existing models for reconstructing past landcover therefore tend to suffer from some combination of limitations including low resolution (e.g. super-regional or global models), short time durations, small spatial range and poor performance in diverse landscapes. In light of these limitations, a new method is presented here that utilizes fossil pollen assemblages and predictive modelling based on the archaeological record to spatially distribute past landcover as a result of interaction between people and landscapes of affordances across large areas and significant durations time.

The multidisciplinary nature of archaeological datasets provides opportunities for wide-ranging multifactorial approaches to the reconstruction of past landcover. The material record of human activity in a landscape provides essential insight into interpreting environmental histories and histories of land use. For case studies where environmental records can be integrated with a substantial record of human occupation, the affordances of natural landscapes can be recontextualized within the scope of human ingenuity to better understand their economic and productive potentials. The suitability of a landscape to accommodate forms of human subsistence is not fixed, but is dynamic through time, changing with variables such as technological and social innovation, human impact and climate. These factors are essential to consider when evaluating the productive potential of past landscapes through a modern lens.

Methods for reconstructing past landscapes to better understand how they have been altered by processes such as erosion are useful for providing a sample dataset through which at least some part of landscape change over time can be quantified. High quality results are often quite spatially limited though, and results from one area are not easily extrapolated to others due to the diversity and complexity of local environmental inputs. Archaeological site data, on the other hand, is abundant, easily obtained and aggregated. With a substantial enough sample size of archaeological site data, the changing suitability of landscapes to accommodate various forms of human land use through time can be reliably quantified and interpolated over large areas. Training predictive models with site data specifically indicative of agricultural land-use practices can be an effective way of reconstructing the past agricultural potential of landscapes within the economic and subsistence frameworks of the past. Where data on the presence of vegetation taxa are also available, these two categories of information can be used together to reconstruct the suitability of landscapes for certain classes of landcover.

Methods

Substantial and reliable fossil pollen records exist for three valley catchments within the Sagalassos study area: Ağlasun (~1120 m a.s.l.), Gravgaz (~1220 m a.s.l.) and Bereket (~1435 m a.s.l.) (Figure 1). Using the ‘Regional Estimates of VEgetation Abundance from Large Sites’ (REVEALS) model (see below) to translate these pollen data into vegetation estimates, the proportions of total vegetation comprising crop species could be estimated for these valleys. Subsequently, ‘Locally Adaptive Models of Archaeological Potential’ (LAMAP) predictive surfaces indicating the agricultural potential of these areas were used to reconstruct the distribution of cropland based on the suitability of land parcels as informed by the extensive history of land use documented within the study area (Carleton et al., 2012, 2017; Rondeau et al., 2022; Vandam et al., 2021; Willett et al., 2022). These predictive cropland reconstructions can form a key component of the reconstruction of the overall vegetative landscape as they provide a basis around which to allocate the taxa present in a pollen assemblage founded on observable patterns of past human behaviour.

Step 1 – Translating pollen data into vegetation estimates

In order to translate the pollen data derived from the extant assemblages into quantitative vegetation estimates, including vegetation percentage of cereals, we employed the REVEALS model technique. The resulting percentages of vegetation comprising cereals were in turn used as proxies for the total cropland area of the three valley catchments. The REVEALS model is a tool for reconstructing past regional vegetation composition using pollen data within Sugita’s (2007a, 2007b) Landscape Reconstruction Algorithm (LRA). The LRA is a two-step approach: the REVEALS model (Regional Estimates of Vegetation Abundance from Large Sites; Sugita, 2007a) estimates the regional vegetation by using a generalized form of Davis’ (1963) R-value model. In the next step, local vegetation within the relevant source area of pollen (RSAP) of the target sites can be estimated using the LOVE (Local Vegetation Estimates; Sugita, 2007b) model. REVEALS was developed to reconstruct regional vegetation composition within 50–100 km using pollen data from large lakes, and has successfully been evaluated against modern and historical vegetation in Sweden (Hellman et al., 2008), Switzerland (Soepboer et al., 2010), and Norway (Hjelle et al., 2015). However, data from large lakes are not available in the study region at a close enough distance to draw meaningful conclusions given the geographical diversity of the area. Fortunately, previous studies have demonstrated that a group of small sites may also be used to estimate regional vegetation with REVEALS (Fyfe et al., 2013; Hoevers et al., 2022; Mazier et al., 2012; Trondman et al., 2016), although standard errors of REVEALS estimates will generally be slightly larger when doing so (Trondman et al., 2016). Within the scope of this research, we thus aimed to reconstruct past regional vegetation composition for three valley catchments in the Sagalassos study region using pollen data from multiple small sites.

To accomplish this, the REVEALS.v5 (Sugita, unpublished) software package was used. The maximum spatial extent of regional vegetation was set to 50 km, and atmospheric conditions were set to be neutral. Standard errors on the vegetation abundances were calculated in REVEALS based on the variance and covariance of pollen productivity estimates and pollen counts per sample site (Sugita, 2007a). Eight detailed and well-dated pollen records from the Sagalassos study region were available and appropriate for use in the analyses (Table 1). The REVEALS model was run for the three valley catchments separately: Ağlasun (three sites; Vermoere, 2004), Gravgaz (three sites; Bakker et al., 2013; Vermoere et al., 2002) and Bereket (two sites; Bakker et al., 2013; Kaniewski et al., 2007). Pollen type parameters for the study area were available from Bakker (2012; Table 2). Of the constituent taxa within the assemblage, reliable pollen productivity estimates (PPE) are available only for Asteraceae, Cerealia, Juniperus, Pinus, Poaceae and Quercus coccifera. Consequently, only these pollen taxa were included in the analyses.

Table 1.

Overview of the fossil pollen study sites in the Sagalassos region.

Site no.	Site name	Core name	References
1	Ağlasun1	Core6	Vermoere (2004)
2	Ağlasun2	Core12	Vermoere (2004)
3	Ağlasun3	Core13	Vermoere (2004)
4	Gravgaz1	G96	Vermoere et al. (2002)
5	Gravgaz2	G99	Vermoere et al. (2002)
6	Gravgaz3	SA06EP1	Bakker et al. (2013)
7	Bereket1	BKT1 and BKT2	Kaniewski et al. (2007)
8	Bereket2	SA09JBDrill2	Bakker et al. (2013)

Table 2.

Fall speed of pollen (FSP; m s⁻¹) and pollen productivity estimates (PPE) for the pollen types used in the REVEALS analyses; based on Bakker (2012).

Pollen types	FSP (m/s)	PPE
Asteraceae	0.051	0.83
Cerealia	0.06	0.02
Juniperus	0.016	0.07
Pinus	0.045	0.57
Poaceae	0.038	0.02
Quercus coccifera	0.024	1

Radiocarbon ages for the samples were calibrated using the IntCal13 calibration curve (Reimer et al., 2013) and Oxcal 4.3 software (Ramsey, 2009). Age-depth models were calculated for each pollen record using the clam.R package version 2.2 (Blaauw, 2010). The pollen data range between ca. 2550 and 100 cal BP. Individual pollen samples from each site were aggregated into 50-year windows of time for the period 2550–1800 cal BP, and into 200-year windows for the period 1800–200 cal BP. These aggregated time windows result in larger pollen sums and reduce errors. The outputs of the REVEALS models were aggregated to correlate with the archaeological data; that is, one average value for each archaeological time period was calculated.

Results of vegetation estimates

The results for the three study areas are shown in Table 3 (in summary) and Table 4 (in detail). Table 3 provides the estimates of the total cropland area for each of the study areas per archaeological time period.

Table 3.

Summary estimates of the mean cropland area for the three catchments per archaeological period.

Period	Mean cropland area (percentages)
Period	Gravgaz	Bereket	Ağlasun
Achaemenid-Hellenistic	0.12	0.09	0.25
Roman Imperial	0.18	0.18	0.25
Late Antique	0.27	0.06	0.18
Byzantine	0.10	0.05	0.24
Late Ottoman	0.09	0.09	0.38

Table 4.

Detailed estimates of the mean cropland area for the three catchments per archaeological period.

Period	Age cal BP	Mean age (cal a BP)	Gravgaz		Bereket		Aglasun
Period	Age cal BP	Mean age (cal a BP)	Cropland area (percentage)	Mean cropland area (percentage)	Cropland area (percentage)	Mean cropland area (percentage)	Cropland area (percentage)	Mean cropland area (percentage)
Archaic	2650-2600	2625	NA	NA	0.00	0.00	na	NA
Archaic	2600-2550	2575	NA	NA	0.00	0.00	na	NA
Achaemenid-Hellenistic	2550-2500	2525	0.08	0.12	0.00	0.09	na	0.25
	2500-2450	2475	0.25		0.13		na
	2450-2400	2425	0.02		0.16		na
	2400-2350	2375	0.03		0.00		0.31
	2350-2300	2325	0.09		0.00		na
	2300-2250	2275	0.07		NA		0.14
	2250-2200	2225	0.08		0.00		na
	2200-2150	2175	0.18		0.16		na
	2150-2100	2125	0.16		0.12		0.20
	2100-2050	2075	0.27		0.23		0.33
Roman Imperial	2050-2000	2025	0.20	0.18	0.09	0.18	na	0.25
	2000-1950	1975	0.10		0.12		na
	1950-1900	1925	0.07		0.27		na
	1900-1850	1875	0.22		0.18		0.21
	1850-1800	1825	0.31		0.24		0.30
Late Antique	1800-1600	1700	0.27	0.27	0.08	0.06	0.00	0.18
	1600-1400	1500	0.16		0.06		0.28
	1400-1200	1300	0.37		0.04		0.26
Byzantine	1200-1000	1100	0.11	0.10	0.03	0.05	0.17	0.24
Byzantine	1000-800	900	0.09	0.10	0.08	0.05	0.32	0.24
Late Ottoman	800-600	700	0.08	0.09	0.09	0.09	0.35	0.38
	600-400	500	0.10		0.06		0.44
	400-200	300	0.05		0.05		0.44
	200-0	100	0.14		0.16		0.28

Step 2 – Predictive surfaces of agricultural potential

To predict which land parcels were most suitable for crop cultivation for each period, individual LAMAP predictive surfaces were created from training datasets with this specific purpose in mind. For a detailed explanation of the LAMAP method see Carleton et al. (2012, 2017). Briefly, though, the approach uses multivariate empirical cumulative distribution functions of quantitative landscape variables from circular sampling areas centred on known-site locations (training data) to estimate the empirical probability of observing a cell around a given known-site that is similar to a target cell (a cell for which a prediction is required). For a given target cell, this probability is estimated for all sites in the training set and then these probabilities are weighted by distance to the target cell and combined using the law of total probability. Effectively, this means a LAMAP estimate is the probability of observing a location like a given target cell – in terms of the relevant landscape traits – across all known sites in the training dataset. This procedure is repeated for all cells in a region of interest, or study area, in order to produce a LAMAP surface. The LAMAP values range between 0 and 1, where 0 indicates a very low probability and 1 a very high probability. The method has similarities to MaxEnt (another predictive model commonly used for species distribution modelling). LAMAP differs from MaxEnt and other predictive models in two key ways: (1) it treats sites like areas instead of points and is, therefore, capable of accounting for intra-site variability; and (2) it assumes that nearby known-sites should be weighted more heavily in predictions than distant ones, leveraging spatial autocorrelation (Tobler’s First Law of Geography) in both ecological traits and human land-use to improve predictions. Following earlier promising results of the validation of the LAMAP method for predicting archaeological potential in the Sagalassos research area with predictive surfaces trained with the entire dataset of archaeological periods (Vandam et al., 2021; Willett et al., 2021), the surfaces created for predicting agricultural potential relied on a specific subset of site data.

30 years of archaeological surveying in the Sagalassos study region (Vandam et al., 2019) revealed 704 individual sites at 289 geographical locations. These sites described in the Sagalassos site database were individually interpreted based on all available documentation (e.g. site reports, field notes, photographs, drawings of artefacts and structural remains) as to whether the material record indicated that agriculture was likely taking place in the close vicinity of the site and that the consideration of suitability for cultivation had been a primary factor in the selection of the site location. Therefore, site classes that may have accommodated agricultural activity nearby, but likely had other more preeminent motivations for the selection of site location, such as religious centres and fortifications, were omitted. Sites which indicated the nature of activity happening at the location was primarily pastoral in nature rather than cultivation were also omitted. Site classes included in the training dataset were those interpreted as (at least semi-) permanently occupied settlements such as farmsteads and villages that revealed material evidence associated with settled life and agriculture, such as presses, basins, wells, and domestic pottery. This resulted in a sample size of 237 sites for training the predictive models, divided by archaeological period. Earlier periods known from the study area were not included due to the beginning of the adequately dated pollen records at ca. 2550 cal BP. Importantly, these data were used in earlier research to validate the LAMAP predictions for this region and that work involved substantial ground-truthing, constituting a strong test of robustness and predictive utility (Vandam et al., 2021, Willett et al., 2022). Iterative cross-validation procedures, like leave-one-out or k-fold approaches, are intended to simulate out-of-sample predictions for estimating out-of-sample predictive utility, which is unnecessary here since previous validation analyses used real new predictions (novel ground survey results gathered specifically to test the LAMAP models employed in the present work). Thus, we were able to use all 237 sites for generating model predictions for this study without the need to re-run cross-validation analyses.

The agricultural potential of each 50 × 50 m land parcel within the 1450 km² study region was determined individually for all five periods by calculating its degree of congruence with the set of known sites from that period based on five landscape variables: elevation, slope, aspect, convexity and proximity to drainage. These are the same set of landscape variables that proved effective in an earlier validation survey (Willett et al., 2021). The similarity between the previously undocumented portions of land and the known set – that is, their relative suitability for the same human activities – were represented by continuous similarity measures between 0 and 1, from low potential to high, on the output predictive surfaces (Figure 2).

Figure 2.

Predictive surfaces of agricultural potential. Lighter gradients equal higher potential.

Step 3 – Spatially reconstructing cropland distribution

Pollen assemblages may reflect vegetative composition over a limited range that varies based on taxa, collection technique and taphonomy. To account for this, the cropland percentages for the three valleys were not extrapolated beyond their topographic boundaries, and the cropland reconstructions were confined to within the valley extents of Ağlasun, Gravgaz and Bereket. To account for the limited number of sites in one or more of the valleys during certain archaeological periods, in particular Gravgaz which had not been previously intensively surveyed, the complete predictive surfaces for the full study area were clipped to the extents of the valleys rather than making individual surfaces for each valley. This allowed for maximization of the sample size for each period and mitigation of standard error.

To translate the predictive surfaces into a tool for the reconstruction of cropland distribution, the land parcels with the highest agricultural potential, that is, those with the greatest degree of congruency with the value of landscape variables measured for the known-site set of agricultural settlements and therefore most suitable parcels for cultivation based on land-use history, must be identified in equal proportion to the cropland percentages predicted by the REVEALS model. For example, in the case of the Ağlasun valley during the Achaemenid-Hellenistic period, the mean cropland area was estimated to be 25% (Tables 3 and 4), therefore only the 25% most suitable land parcels based on the model’s prediction are rendered in the final surface. This rendering is accomplished through a reclassification of the continuous similarity measures. The degree of congruence is reclassified using ‘quantiles’ created by binning the original continuous similarity measures into four bins representing the total number of pixels (i.e. land parcels), with only the two bins containing the highest values equal to the mean cropland area estimation being rendered and those with lower values not being rendered. The two bins rendered contain equal portions of pixels, representing those with ‘high’ and the ‘highest’ potential suitability for cropland. Splitting the most suitable pixels into two bins was a presentation choice and they could also be rendered into more or fewer gradients as long as the total number remains proportional to the mean cropland area estimation.

Results

The Ağlasun Valley

The Ağlasun valley was well represented in the sample set of known sites due to the extensive history of archaeological prospection in the vicinity of Sagalassos (Vandam et al., 2019). The predicted distributions of croplands there (Figure 3) appear largely consistent with other indicators of past land uses that were not included as predictor variables in the modelling, such as constructed terraces, irrigation channels and other water infrastructures, as well as current cultivation status. The directional aspect of the predicted distributions – mostly concentrated on the south-facing slopes and in the centre of the valley – is also consistent with current land uses, and demonstrates that the known-site set reliably reflects the preference for maximum sunlight exposure held by farming communities. Over time, the highest potential parcels cluster further downslope towards the valley bottom, likely in response to the movement of soil due to erosion from the higher elevations (García-Ruiz et al., 2013).

Figure 3.

Predictive surfaces of cropland distribution in the Ağlasun valley. Orange parcels represent high potential, red represent highest potential.

The Gravgaz Valley

In the valley including Gravgaz marsh, intensive archaeological survey activities have so far been limited to the validation survey of the initial LAMAP models (Vandam et al., 2021), resulting in low overall survey coverage and therefore low representation in the known-site set. Nevertheless, the predicted cropland distributions (Figure 4) largely agree with the results of the human induced soil erosion models of Van Loo et al. (2017), situating more cultivation on the higher slopes along the rim of the valley during the earlier periods and shifting towards the valley bottom as the soil eroded, peaking during the Late Antique at 27% mean cropland area with most of the highest potential parcels at the lower elevations. At the higher elevations, south facing aspects are again favoured where the slope is less steep at the northern end of the valley, but conversely face inward towards the valley centre in positions where the outward south facing slopes are steeper. A full intensive archaeological survey of the valley is necessary to increase the reliability of these results.

Figure 4.

Predictive surfaces of cropland distribution in the Gravgaz valley. Orange parcels represent high potential, red represent highest potential.

Bereket Valley

The estimated mean area of past cropland in Bereket is the lowest overall between the three valleys. The maximum extent of cropland is estimated to have occurred during the Roman Imperial period, reaching 18% of the total area of the valley. The cropland distribution predictions (Figure 5) allocate much of the highest potential area during this time to the flanks of the slopes to the south and east overlooking the valley. During the Late Antique, Byzantine and Late Ottoman periods, the highest potential areas for cropland cluster more in the bottom of the valley and its branches to the east, and especially along the course of the stream running north-south through the centre of the valley. Interestingly, very little cropland is distributed to the western side of the valley, perhaps due to the steepness of the slopes there and the narrowness of the spaces between them.

Figure 5.

Predictive surfaces of cropland distribution in the Bereket valley. Orange parcels represent high potential, red represent highest potential.

Discussion and conclusions

The principle of archaeological potential relies on the human decision-making processes of the past to inform the production of predictive surfaces. Using this principle as the foundational component for landcover reconstructions has not previously been attempted, yet it presents a framework within which process – for example, long-term ecological variation indicated through environmental proxies – and historic contingency, in the form of the known record of human decision-making in the study area, can be integrated interdependently. As a result, the human-environment relationship can be interpreted from a perspective that is both non-environmentally deterministic and strongly driven by environmental and archaeological datasets, accounting for the ‘historical dialect’ (Hodder and Hutson, 2003) between physical conditions and the social relations of transformation. This represents the broader aim of the methodological approach that is presented here.

Though myriad techniques for translating environmental proxy datasets such as fossil pollen assemblages into reconstructions of the wider conditions of past landscapes have been developed in recent decades, the majority of these methods result in largely abstract quantitative reconstructions that are not spatially explicit. Those methods that are spatially explicit, such as some landcover distribution techniques, often suffer from limitations of low resolution, poor performance in heterogeneous landscapes, and either a lack of consideration for human agency or the affordances of landscapes.

The method proposed here for modelling the suitability of land parcels for the distribution of estimated mean cropland area based on fossil pollen assemblages overcomes many limitations of other approaches. The calculations underlying LAMAP cope particularly well with geographical diversity, as they are locally adaptive in nature and suitability is always treated relatively across the area of interpolation. This allows a large and broad dataset describing land-use history to be scaled down to very specific landscape units if there is limited information on the past vegetation, or scaled up if there are data available on a wide territory. The reliability of these predictive surfaces has so far been demonstrated at resolutions as high as 50 × 50 m, which corresponds well with the incremental scale of expansion through land clearance and planting of some traditional agricultural field systems (Doolittle, 1984: 127). Distributing the mean cropland area values estimated by the REVEALS model in this way provides a new dimension of high-resolution spatial explicitness to an already established technique for the estimation of vegetation composition from fossil pollen data. Most importantly, using this method, the reconstruction of the target landscape is directly informed by the record of past land-use behaviours of communities living in the specific study area.

Further potential for the enhancement of cropland distribution predictions may reside in the inclusion of additional datasets relating to the suitability of land parcels for cultivation based on observations of human-built and natural variables such as terraces, soil depth and chemistry, springs and water infrastructures, and current plant species composition. Significant representation of the specific study area from which the pollen samples come in the known-site training dataset is also ideal for the best results. Reconstruction of cropland distributions using this method is intended as a component of a wider approach to the reconstruction of past landscapes, and can comprise a key framework around which to orient such additional research.

Footnotes

Acknowledgements

The authors would like to thank the Sagalassos Project survey team, the Max Planck Society, the Belgian American Educational Foundation, Research Foundation – Flanders (FWO), C1 Research Fund – KU Leuven, the Research Center for Anatolian Civilizations (ANAMED) and the American Research Institute in Turkey (ARIT).

Author contributions

Patrick Willett: Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Project administration; Resources; Supervision; Validation; Visualization; Writing – original draft; Writing – review & editing

W. Christopher Carleton: Formal analysis; Methodology; Resources; Software; Validation; Visualization; Writing – original draft; Writing – review & editing

Nils Broothaerts: Data curation; Formal analysis; Resources; Visualization; Writing – original draft; Writing – review & editing

Ralf Vandam: Data curation; Funding acquisition; Investigation; Project administration; Resources; Supervision; Validation

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: from the Belgian American Educational Foundation, Research Foundation – Flanders (FWO), C1 Research Fund – KU Leuven, the Research Center for Anatolian Civilizations (ANAMED), the American Research Institute in Turkey (ARIT), and the Max Planck Society.

ORCID iDs

Patrick T. Willett

Nils Broothaerts

References

Alrteimei

Ash’aari

Muharram

(2022) Last decade assessment of the impacts of regional climate change on crop yield variations in the Mediterranean region. Agriculture 12(11): 1787.

Arponen

Dörfler

Feeser

, et al. (2019) Environmental determinism and archaeology. Understanding and evaluating determinism in research design. Archaeological Dialogues 26(1): 1–9.

Axtell

Epstein

Dean

, et al. (2002) Population growth and collapse in a multiagent model of the Kayenta Anasazi in Long House Valley. Proceedings of the National Academy of Sciences 99(Suppl. 3): 7275–7279.

Bakker

(2012) Late-Holocene vegetation dynamics in a mountainous environment in the Territory of Sagalassos, Southwest Turkey (Late Roman till present). Dissertation, Katholieke Universiteit Leuven, Leuven, Belgium.

Bakker

Paulissen

Kaniewski

, et al. (2013) Climate, people, fire and vegetation: New insights into vegetation dynamics in the Eastern Mediterranean since the 1st century AD. Climate of the Past 9(1): 57–87.

Beach

Dunning

Luzzadder-Beach

, et al. (2006) Impacts of the ancient Maya on soils and soil erosion in the central Maya Lowlands. Catena 65(2): 166–178.

Blaauw

(2010) Methods and code for ‘classical’ age-modelling of radiocarbon sequences. Quaternary Geochronology 5(5): 512–518.

Carleton

Cheong

Savage

, et al. (2017) A comprehensive test of the Locally-Adaptive Model of Archaeological Potential (LAMAP). Journal of Archaeological Science: Reports 11: 59–68.

Carleton

Conolly

Ianonne

(2012) A locally-adaptive model of archaeological potential (LAMAP). Journal of Archaeological Science 39(11): 3371–3385.

10.

Chang-Martínez

Mas

J-F

Torrescano Valle

, et al. (2015) Modeling historical land cover and land use: A review from contemporary modeling. ISPRS International Journal of Geo-Information 4(4): 1791–1812.

11.

Chisholm

(1962) 1962: Rural Settlement and Land Use. London: Hutchinson.

12.

Contreras

(2016) Correlation is not enough: Building better arguments in the archaeology of human-environment interactions. In: Contreras

(ed.) The Archaeology of Human-Environment Interactions. New York, NY: Routledge, pp.17–36.

13.

Davis

(1963) On the theory of pollen analysis. American Journal of Science 261(10): 897–912.

14.

Doolittle

(1984) Agricultural change as an incremental process. Annals of the Association of American Geographers 74(1): 124–137.

15.

Foley

DeFries

Asner

, et al. (2005) Global consequences of land use. Science 309(5734): 570–574.

16.

Fyfe

Twiddle

Sugita

, et al. (2013) The Holocene vegetation cover of Britain and Ireland: Overcoming problems of scale and discerning patterns of openness. Quaternary Science Reviews 73: 132–148.

17.

Galicia

García

Gómez

, et al. (2007) Cambio de uso del suelo y degradación ambiental. Ciencia 58(4): 50–59.

18.

García-Ruiz

Nadal-Romero

Lana-Renault

, et al. (2013) Erosion in Mediterranean landscapes: Changes and future challenges. Geomorphology 198: 20–36.

19.

Gattiglia

(2015) Think big about data: Archaeology and the Big Data challenge. Archäologische Informationen 38: 113–124.

20.

Geist

Lambin

(2001) What drives tropical deforestation. LUCC Report Series 4: 116.

21.

Goldewijk

(2001) Estimating global land use change over the past 300 years: The HYDE database. Global Biogeochemical Cycles 15(2): 417–433.

22.

Gosling

Dunn

Carrol

, et al. (2011) Climate: Observations, Projections and Impacts. Devon: Met Office Hadley Centre.

23.

Griffin

Stanish

(2007) An agent-based model of prehistoric settlement patterns and political consolidation in the Lake Titicaca Basin of Peru and Bolivia. Structure and Dynamics 2(2). https://doi.org/10.5070/SD922003290

24.

Hellman

Gaillard

Broström

, et al. (2008) The REVEALS model, a new tool to estimate past regional plant abundance from pollen data in large lakes: Validation in southern Sweden. Journal of Quaternary Science 23(1): 21–42.

25.

Higgs

Vita-Finzi

(1972) Prehistoric economies: A territorial approach. In: Higgs

(ed.) Papers in economic prehistory. Cambridge: CUP, pp.27–36.

26.

Hjelle

Mehl

Sugita

, et al. (2015) From pollen percentage to vegetation cover: Evaluation of the Landscape Reconstruction Algorithm in western Norway. Journal of Quaternary Science 30(4): 312–324.

27.

Hodder

Hutson

(2003) Reading the Past: Current Approaches to Interpretation in Archaeology. Cambridge: Cambridge University Press.

28.

Hodder

Orton

(1976) Spatial Analysis in Archaeology. Cambridge: Cambridge University Press.

29.

Hoevers

Broothaerts

Verstraeten

(2022) The potential of REVEALS-based vegetation reconstructions using pollen records from alluvial floodplains. Vegetation history and archaeobotany 31(5): 525–540.

30.

Horvath

Halvorsen

Stordal

, et al. (2019) Distribution modelling of vegetation types based on area frame survey data. Applied Vegetation Science 22(4): 547–560.

31.

Houghton

(1999) The annual net flux of carbon to the atmosphere from changes in land use 1850–1990. Tellus B: Chemical and Physical Meteorology 51(2): 298–313.

32.

Huggett

(2022) Is less more? Slow data and datafication in archaeology. In: Critical archaeology in the digital age, proceedings of the 12th IEMA visiting scholar’s conference, Buffalo, NY, 6–7 April 2019. Los Angeles, CA: UCLA Cotsen Institute of Archaeology Press, pp.97–110.

33.

Jacobson

(2022) Archaeological evidence for community resilience and sustainability: A bibliometric and quantitative review. Sustainability (Switzerland) 14(24): 16591.

34.

Kairis

Karamanos

Voloudakis

, et al. (2022) Identifying degraded and sensitive to desertification agricultural soils in Thessaly, Greece, under simulated future climate scenarios. Land 11(3): 395.

35.

Kaniewski

De Laet

Paulissen

, et al. (2007) Long-term effects of human impact on mountainous ecosystems, western Taurus Mountains, Turkey. Journal of Biogeography 34(11): 1975–1997.

36.

Kaplan

Krumhardt

Zimmermann

(2009) The prehistoric and preindustrial deforestation of Europe. Quaternary Science Reviews 28(27–28): 3016–3034.

37.

Karapinar

Özertan

(2020) Yield implications of date and cultivar adaptation to wheat phenological shifts: A survey of farmers in Turkey. Climatic Change 158(3–4): 453–472.

38.

Kohler

(1988) Predictive locational modelling: History and current practice. In: Judge

Sebastian

(eds) Quantifying the Present and Predicting the Past: Theory, Method, and Application of Archaeological Predictive Modeling. Denver, CO: US Bureau of Land Management, pp.19–59.

39.

Kohler

Cockburn

Hooper

, et al. (2012) The coevolution of group size and leadership: An agent-based public goods model for prehispanic Pueblo societies. Advances in Complex Systems 15(01n02): 1150007.

40.

Kristiansen

(2019) Who is deterministic? On the nature of interdisciplinary research in archaeology. Archaeological Dialogues 26(1): 12–14.

41.

Lambin

Geist

Lepers

(2003) Dynamics of land-use and land-cover change in tropical regions. Annual Review of Environment and Resources 28(1): 205–241.

42.

Mazier

Gaillard

M-J

Kuneš

, et al. (2012) Testing the effect of site selection and parameter setting on REVEALS-model estimates of plant abundance using the Czech Quaternary Palynological Database. Review of Palaeobotany and Palynology 187: 38–49.

43.

Murphy

(2012) Exploring complexity with the Hohokam Water Management Simulation: A middle way for archaeological modeling. Ecological Modelling 241: 15–29.

44.

Özdoğan

(2011) Modeling the impacts of climate change on wheat yields in Northwestern Turkey. Agriculture, Ecosystems & Environment 141(1–2): 1–12.

45.

Ozkan

Akcaoz

(2002) Impacts of climate factors on yields for selected crops in the Southern Turkey. Mitigation and Adaptation Strategies for Global Change 7: 367–380.

46.

Pirzamanbein

Lindström

Poska

, et al. (2014) Creating spatially continuous maps of past land cover from point estimates: A new statistical approach applied to pollen data. Ecological Complexity 20: 127–141.

47.

Pongratz

Reick

Raddatz

, et al. (2008) A reconstruction of global agricultural areas and land cover for the last millennium. Global Biogeochemical Cycles 22(3). https://doi.org/10.1029/2007GB003153

48.

Ramsey

(2009) Bayesian analysis of radiocarbon dates. Radiocarbon 51(1): 337–360.

49.

Reimer

Bard

Bayliss

, et al. (2013) IntCal13 and Marine13 radiocarbon age calibration curves 0–50,000 years cal BP. Radiocarbon 55(4): 1869–1887.

50.

Rondeau

Carleton

Collard

, et al. (2022) Does the Locally-Adaptive Model of Archaeological Potential (LAMAP) work for hunter-gatherer sites? A test using data from the Tanana Valley, Alaska. PLoS One 17(3): e0265597.

51.

Silva

Coward

Davies

, et al. (2022) Developing transdisciplinary approaches to sustainability challenges: The need to model socio-environmental systems in the Longue Durée. Sustainability (Switzerland) 14(16): 1–20.

52.

Soepboer

Sugita

Lotter

(2010) Regional vegetation-cover changes on the Swiss Plateau during the past two millennia: A pollen-based reconstruction using the REVEALS model. Quaternary Science Reviews 29(3–4): 472–483.

53.

Sugita

(2007a) Theory of quantitative reconstruction of vegetation I: Pollen from large sites REVEALS regional vegetation composition. The Holocene 17(2): 229–241.

54.

Sugita

(2007b) Theory of quantitative reconstruction of vegetation II: All you need is LOVE. The Holocene 17(2): 243–257.

55.

Trondman

A-K

Gaillard

M-J

Sugita

, et al. (2016) Are pollen records from small sites appropriate for REVEALS model-based quantitative reconstructions of past regional vegetation? An empirical test in southern Sweden. Vegetation History and Archaeobotany 25: 131–151.

56.

Van Loo

Dusar

Verstraeten

, et al. (2017) Human induced soil erosion and the implications on crop yield in a small mountainous Mediterranean catchment (SW-Turkey). Catena 149: 491–504.

57.

Vandam

Kaptijn

Willet

, et al. (2019) The countryside-Where are the people? In: Poblome J (ed.) Meanwhile in the Mountains: Sagalassos. Istanbul: Yapı Kredi Yayınları, pp.262–270.

58.

Vandam

Willett

Carleton

, et al. (2021) Get Going with Survey Legacy Data. The 2018 and 2019 Archaeological Survey Results from the Sagalassos Archaeological Research Project. In: Steadman

McMahon

(eds) The Archaeology of Anatolia: Recent Discoveries (2018–2020). Cambridge: Cambridge Scholars Press, pp.312–321.

59.

Vermoere

(2004) Holocene Vegetation History in the Territory of Sagalassos (Southwest Turkey): a Palynological Approach. Turnhout, Belgium: Brepols, 2004.

60.

Vermoere

Bottema

Vanhecke

, et al. (2002) Palynological evidence for late-Holocene human occupation recorded in two wetlands in SW Turkey. The Holocene 12(5): 569–584.

61.

Warren

(1990) Predictive modelling in archaeology: A primer. In: Allen

KMS

Green

EZubrow

EBW

(eds) Interpreting Space: GIS and Archaeology. London: Taylor &Francis, pp.90–111.

62.

Willett

Carleton

Torun

, et al. (2022) Modeling archaeological potentials in Southwest Anatolia. A tool for planning sustainable futures at ancient Sagalassos. In: Garstki

(ed.) Critical Archaeology in the Digital Age. Los Angeles, CA: Cotsen Press, pp.67–79.

63.

Willey

(1953) Prehistoric settlement patterns in the Virú; Valley, Peru. Bureau of American Ethnology Bulletin 155: 1–453.

64.

Willey

(1956) Prehistoric Settlement Patterns in the New World. New York, NY: Wenner-Gren Foundation; New York, NY: Johnson Reprint Corporation.

65.

Zittis

Almazroui

Alpert

, et al. (2022) Climate change and weather extremes in the Eastern Mediterranean and Middle East. Reviews of Geophysics 60(3): e2021RG000762.