Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

GIS-based landform classification of Bronze Age archaeological sites on Crete Island

  • Athanasios V. Argyriou ,

    argyriou.nasos@gmail.com

    Affiliations School of Earth and Environmental Sciences, Centre for Applied Geosciences, University of Portsmouth, Portsmouth, PO1 3QL, United Kingdom, Laboratory of Geophysical—Satellite Remote Sensing & Archaeo-environment, Institute for Mediterranean Studies (I.M.S.) / Foundation of Research & Technology (F.O.R.T.H.), Melissinou & Nikiforou Foka 130, Rethymno, Greece

  • Richard M. Teeuw,

    Affiliation School of Earth and Environmental Sciences, Centre for Applied Geosciences, University of Portsmouth, Portsmouth, PO1 3QL, United Kingdom

  • Apostolos Sarris

    Affiliation Laboratory of Geophysical—Satellite Remote Sensing & Archaeo-environment, Institute for Mediterranean Studies (I.M.S.) / Foundation of Research & Technology (F.O.R.T.H.), Melissinou & Nikiforou Foka 130, Rethymno, Greece

GIS-based landform classification of Bronze Age archaeological sites on Crete Island

  • Athanasios V. Argyriou, 
  • Richard M. Teeuw, 
  • Apostolos Sarris
PLOS
x

Abstract

Various physical attributes of the Earth’s surface are factors that influence local topography and indirectly influence human behaviour in terms of habitation locations. The determination of geomorphological setting plays an important role in archaeological landscape research. Several landform types can be distinguished by characteristic geomorphic attributes that portray the landscape surrounding a settlement and influence its ability to sustain a population. Geomorphometric landform information, derived from digital elevation models (DEMs), such as the ASTER Global DEM, can provide useful insights into the processes shaping landscapes. This work examines the influence of landform classification on the settlement locations of Bronze Age (Minoan) Crete, focusing on the districts of Phaistos, Kavousi and Vrokastro. The landform classification was based on the topographic position index (TPI) and deviation from mean elevation (DEV) analysis to highlight slope steepness of various landform classes, characterizing the surrounding landscape environment of the settlements locations. The outcomes indicate no interrelationship between the settlement locations and topography during the Early Minoan period, but a significant interrelationship exists during the later Minoan periods with the presence of more organised societies. The landform classification can provide insights into factors favouring human habitation and can contribute to archaeological predictive modelling.

Introduction

During the Middle Minoan period of the Bronze Age period in Crete, there appears to have been a widespread increase of sites in low elevation areas suitable for arable farming. Systematic archaeological surveys across eastern and central Crete have examined the settlement dynamics and in all cases highlight a generalised population movement from high elevation areas with limited arable land to lower elevation areas, particularly plains favourable to cultivation and efficiency in irrigation [1]. Most of the studies hypothesized that this population movement tendency was caused by economic and political reasons but only a few of them considered the possibility that this tendency could be a result of environmental conditions and resource exploitation [2,3,4]. Mountainous sites at higher elevations (~600–800 m) seem not to attract further exploitation during the Middle Minoan (and were rarely re-occupied permanently during the later Minoan periods). That was perhaps due to the development of new practices for intensive agriculture which could not be applied in steeply sloping terrain [1]. As a result, people relocated to lower elevations (~300 m) and plains, which have high agricultural and irrigation potential to support population growth, as in Kavousi and Vrokastro districts [5,6].

This study aims to examine the hypothesis outlined above by studying settlement dynamics and landform characteristics during the Minoan periods, to check whether the population movement to lower elevation areas was a random tendency, or was interlinked with agricultural practices. It also examines the more general trend that the clusters of settlements followed, despite their local “micro-regional identity” [5]. Investigation of heterogeneity in geological and geomorphological properties can lead to the quantification of landscapes and to a better understanding of their complexity [7,8,9].

The Earth’s surface consist of large geomorphic features (e.g. plains, mountain ranges), through to smaller features (e.g. valleys) and their component landforms, such as valley slopes, floodplains and terraces [10]. Such landscape classification information can be of use for archaeological studies because the two sets of data can be interlinked to gain insights into the factors driving settlement evolution [11,12]. Topographic prominence was one of the first approaches to find application on archaeological studies while examination of other geomorphometric parameters followed (e.g. terrain ruggedness, amplitude of relief). Such was the case with archaeological studies of Cyprus, where an evaluation of the exposure of cultural heritage sites to natural hazards took place [13,14,15,16,17]. Most of these studies were mainly investigating the local or relative topographic position of archaeological sites in the landscape [18,19].

More recently, the development of computer technology, software packages and free access to datasets attracted the interest of analysts to develop computer algorithms to examine geomorphometric attributes and the topography of the Earth’s surface [20,21,22]. In particular, in recent years there has been an increase of interest in use of GIS-based analyses to classify landforms, over various scientific fields such as geomorphology, geology, agriculture [23,21]. Despite that, only a few studies exist with applications of archaeological interest [24,19]. The majority the studies consist of automated or semi-automated approaches, mostly evaluating homogeneous landscapes [25,26]. In contrast, this study examines both homogenous and heterogeneous landscapes. The information derived by terrain analysis, via DEMs and geomorphometrics, offers to analysts a powerful approach to describe the topographic position of features [27,21].

Understanding the relationship between the types of archaeological sites (e.g. settlements, defending sites or burial sites) and their surrounding landscape type is an important aspect of archaeological investigations. Landscape geomorphometrics can reveal insights into variations in the distribution of settlements over time [28,12]. Archaeological landscapes can be quantified by using the topographic position index (TPI) (or difference from mean elevation, DIFF), which can classify the landscape, both in terms of slope position and landform categories, into morphological classes based on the geomorphology [29,30,31]. In addition, the deviation from mean elevation (DEV) can be useful tool for geo-archaeological studies, because it highlights the subtle topographic features [32]. These two approaches are investigated in this study because they can be significantly enlightening from an archaeological perspective. The periods of the Early Minoan (5,600–4,000 BP), Middle Minoan (4,000–3,600 BP) and Late Minoan (3,600–3,000 BP) are the focus of this study [6,33,34].

Study area

Crete is located in the southern part of Greece and has a significant archaeological heritage. Evidence from stone tools reveals human presence on the island of Crete as early as 130,000 years ago. However, evidence for modern human presence dates to 10,000–12,000 years ago and it was not until the Neolithic period (8500–4900 BP) when the first signs of advanced agriculture appeared in the Aegean, to open the way for the subsequent emergence of the Minoan civilization (5,600 to 3,000 BP), which is considered as the birthplace of the earliest “high culture” in Europe [35,36,37] (Fig 1A). The Phaistos, Kavousi and Vrokastro districts were selected as case study sites because of their rich archaeological heritage (Fig 1B). The information derived from past archaeological surveys is sufficiently detailed for the analysis of variations in settlement locations during the Minoan periods [6,33,38]. Earlier prehistoric geomorphological conditions were not considered in this methodological framework, as a freely-available Digital Elevation Model (DEM) of recent years is being used in the analysis: further research is needed to determine palaeo-geomorphological features of the study region. In general, the landscape of the study region has remained stable since the earliest phases of human settlement, with the main terrain features being formed in the Pleistocene (i.e., mostly within the past 2 million years) [39]. Regarding palaeo-climatic conditions, at present no conclusive observations exist for Crete from Bronze Age to the present. The role of climatic fluctuations during that period is in general an unexplored field, although a few studies have attempted to evaluate climatic changes with societal developments in areas around eastern Mediterranean [40,41]. [34] reviewed existing indirect measurements of palaeo-climate, so-called climate proxies (e.g. stable isotopes, fossil microshells), with few climatic fluctuations (drought or wetter conditions) being observed in Crete during the Minoan period. As [34] describes, the existing data is unevenly distributed, with some periods having one or two proxies and others by seven or eight. That highlights the importance of integrating climate and environmental history because a relationship between minor Alpine glacial advances in Europe and periods of extreme weather in the Aegean seems to have existed in the study region [39].

thumbnail
Fig 1.

(a): The island of Crete, with red tones highlighting mountainous relief (LO: Lefka Ori; P: Psiloritis); (b) Crete shaded relief and contour intervals. The black boxes indicate the case study sites.

https://doi.org/10.1371/journal.pone.0170727.g001

Minoan land use patterns

During Early Minoan, in the case of the Kavousi-Vrokastro district, the settlements are preferably established on heterogeneous landscapes and many are found at higher elevations with complex rough terrain, mainly hills and ridges [6]. The Mirabello Bay area consists of small coastal valleys: most of the settlements seem to be placed on slopes rather than on the best (flat, lowland) agricultural land [42]. The site pattern was defensive, providing lines of sight between the settlements and out across the Mirabello gulf. Thus the Vrokastro site pattern may be in some accord with [43] suggestion that “people were forced to look for their safety in barely accessible places”. During this period it seems there was a dichotomy between occupying low-lying coastal sites and upland locations, indicating a tendency of the population to occupy diverse types of terrain, seeking to gain control of water sources and sites with the greatest visibility [6]. The evidence for settlement hierarchy during this period has been doubted [44], with clusters of settlements found around the arable coastal basins and not around large settlements.

The Middle Minoan is characterized by settlements in homogeneous landscapes, such as the Kavousi plains, which were well-suited for cultivation [38]. There appears to have been a movement of population into the lowlands and arable upland areas close to water resources, with an associated increase in number of settlements near the best agricultural land during Middle and Late Minoan [6,33,38]. [5] notes that the pattern of settlements exploitation in the Kavousi area is in accordance with the Argolid model during this period, with concentrated population around arable areas, similar to the expansion pattern of small farms inland in Chania district.

Between the Early and Middle Minoan, population growth occurred in the Gournia valley, with continuous expansion of settlements and coastal trading interests [6]. In addition, during the Middle Minoan there is exploitation of the landscape at relatively high elevations, between 450 m and 700 m above sea level. River terraces and bedrock of conglomerates and marl provided fertile soil for cultivation of the Gournia valley, along with fault fractures that provided water access via springs at higher elevation [6].

In the Late Minoan an overall reduction of the number of settlements is observed, especially in upland sites. Despite that, rural settlement is oriented to arable land and water resources, preferentially between 50–200 m elevation range, while a retreat from coastal zones (0–50m) seems to have occurred [6]. An event that might be able to explain this reduction of settlements is the impact of the Theran eruption, with evidence of a thick layer of tephra/ash found in the Mochlos area, near Gournia. Population centralization is also observed during the Late Minoan, for example at Pyrgos or Gournia. Some new large building structures during this period can be found (e.g. around the Istron river valley) which appear to be linked to trade routes, access to arable land and water resources [45].

Materials and methods

TPI and DEV analyses for various neighbourhood sizes

The landform classification was based on TPI and DEV analyses, using as an input dataset the free ASTER Global DEM (G-DEM: 30m pixel size) and the allocated archaeological sites for each Minoan period, as recorded by the archaeological surveys of [6], [33] and [38] (Figs 2 & 3). A buffer zone of 300m around each site was selected for the analyses, in order to evaluate the surrounding dominant landscape of each individual site. The TPI and DEV analyses were evaluated for various neighbourhood sizes, to select the optimum neighbourhood for the rest of the methodological framework. Such a decision depends on the degree that a particular neighbourhood size will represent, or separate better the various morphological classes in a more coherent and uniform way. The Fragstats free software was used to calculate the patch density (PD) and numbers of patches (NP), in order to check the fragmentation of the patches for the various morphological classes (Table 1). The NP equals to the number of patches of the corresponding patch type, as a measure of the extent of fragmentation of the patch type; and PD expresses in addition the number of patches on a per unit area basis [46].

thumbnail
Fig 2.

TPI or DIFF for EM, LM and MM period on Phaistos region, with six morphologic classes for the neighbourhood sizes: a) 150 m; b) 300 m; c) 600 m; d) 1200 m (see S1 Fig for Kavousi-Vrokastro region).

https://doi.org/10.1371/journal.pone.0170727.g002

thumbnail
Fig 3.

DEV for EM, LM and MM period on Phaistos region, with six morphologic classes for the neighbourhood sizes: a) 150 m; b) 300 m; c) 600 m; d) 1200 m (see S2 Fig for Kavousi-Vrokastro region).

https://doi.org/10.1371/journal.pone.0170727.g003

thumbnail
Table 1. Indices calculated by using free open source software Fragstats to characterize prior the shape and fragmentation of the patch types.

https://doi.org/10.1371/journal.pone.0170727.t001

TPI or DIFF (zo-z) measures the relative topographic position of the central point as the difference between the elevation of this central point and the mean elevation within a pre-determined neighbourhood [30,31]. The result from TPI is a positive value when the central point is situated higher than its average neighbourhood, while a negative value indicates a lower location than the average neighbourhood. This index can be used to classify the landscape into morphological classes [47] (Fig 2).

DEV () measures the relative topographic position of the central point as the TPI divided by the standard deviation of elevation (SD), within a predetermined neighbourhood, where SD measures the variability of the cell values in a DEM, around this central point, within the predetermined neighbourhood [30]. The results from DEV are: positive, when the central point is situated higher than its average neighbourhood; or negative, when the central point is situated lower than its average neighbourhood (Fig 3).

The exact difference in height between the central point and the average height of the neighbourhood can be highlighted by those analyses, with the range of the outcome values being dependant on the elevation differences in the DEM within the selected neighbourhood. The selection of the neighbourhood sizes were based on 5 out of 12 candidate radii from 100 to 2000m and the ones used to [19] study were decided as optimum for this study. The circular neighbourhood sizes considered within this study as more useful had radii of: 100 m, 300 m, 600 m, 1200 m and 2000 m. The application of a neighbourhood size is important during the analysis and is related to the landscape feature being analysed. In order to classify small features such as streams and valleys a small neighbourhood size need to be used. To identify large canyons or other large landscape features a large neighbourhood size should be selected during the analysis. When we are dealing with larger size of the neighbourhood and large differences in elevation values these conditions result to the wider range of the output values. Based on the above observations TPI was used instead of DEV for the subsequent analyses of slope position and landform classification, as it represents a more regional representation of the morphological classes (Figs 2 & 3).

Slope position classification for various neighbourhood sizes

In order to classify the landscape into slope position classes, the variation of TPI values by the slope at each point was examined. High TPI values determine the tops of the hills, while low TPI values highlight valley bottoms. TPI values close to 0 are observed over flat to mid-slope areas. In this study, a six-category slope position class (Table 2) was determined for each of the five neighbourhood sizes considered (100 m, 300 m, 600 m, 1200 m and 2000 m) (Fig 4). A 5° slope threshold value was considered for the discrimination of flat areas and mid-slope areas, while a TPI threshold value of ±1SD was considered for the identification of hilltops and valley bottoms. The six slope position classes, based on [31], are presented in Table 2.

thumbnail
Fig 4.

Slope position classification based on TPI of the case study sites of Phaistos, for EM, LM and MM periods, with six morphological classes for the neighbourhood sizes: a) 100 m; b) 300 m; c) 600 m; d) 1200 m; e) 2000 m (see S3 Fig for Kavousi-Vrokastro region).

https://doi.org/10.1371/journal.pone.0170727.g004

thumbnail
Table 2. Classification of the landscape into morphological classes, where SD is Standard Deviation.

https://doi.org/10.1371/journal.pone.0170727.t002

Landform classification for various neighbourhood sizes

The landform classification was based on TPI analysis. The TPI based landform classification can produce 10 landform classes: streams, mid-slope drainages, local ridges, valleys, plains, foot slopes, upper slopes, upland drainage, mid-slope ridges and high ridges [21] (Fig 5). Usually, two neighbourhood sizes are combined to offer detailed geomorphological information through the discrimination of complex landscape features, as a single neighbourhood size provides less information about the general shape of the landscape [21]. In this study, the two combined neighbourhood sizes in each case were: i) 100 m and 600 m; ii) 300 m and 1000 m; iii) 300 m and 2000 m and; iv) 600 m and 2000 m.

thumbnail
Fig 5.

Landform classification based on TPI of the case study sites of Kavousi-Vrokastro, for EM, LM and MM periods, with ten landform types for the combined neighbourhood sizes: a) 100 m and 600 m; b) 300 m and 1000 m; c) 300 m and 2000 m; d) 600 m and 2000 m (see S4 Fig for Phaistos region).

https://doi.org/10.1371/journal.pone.0170727.g005

The allocated settlements for each Minoan period in Vrokastro district provided an accuracy assessment of the combined neighbourhood sizes outcomes. Vrokastro district was selected to test the methodology on a more heterogeneous landscape relative to the less rough terrain of Phaistos region. The accuracy assessment consisted of the comparison between the coverage percentage of the individual landforms classes based on TPI analysis and the geological-geomorphological description of the settlements provided by the archaeological surveys of [6]. The Fragstats freeware was used to determine which of the combined neighbourhood sizes provided the best representation of the various landform types, relative to the rest of the tested neighbourhood sizes, regarding their shape and fragmentation characteristics. Various indices were examined via Fragstats (Table 1): i) shape index (SHAPE); ii) proximity index (PROX); iii) landscape shape index (LSI); iv) related circumscribing circle (CIRCLE); v) patch cohesion index (COHESION); vi) patch density (PD); vii) numbers of patches (NP) and; viii) perimeter-area fractal dimension (PAFRAC).

A spatial autocorrelation evaluation based on Moran’s I approach was carried out for the EM, MM, LM settlements and 100 random sites of the study area, to check whether the pattern expressed was clustered or random [48]. The derived z‐score or p‐value can indicate statistical significance, with a positive Moran's I index value indicating tendency toward clustering, while a negative Moran's I index value indicates tendency toward dispersion. The z‐score and p‐value calculations indicate whether the null hypothesis can be rejected. In this case, the null hypothesis states that features are randomly distributed across the study area. Clustering is a statistical classification technique for separating spatial information into relatively homogeneous groups: similarities are high between members belonging to a class, or cluster [48].

Results

TPI and DEV analyses for various neighbourhood sizes

The TPI and DEV outcomes are presented on Figs 2 and 3 for the various neighbourhood sizes (150 m, 300 m, 600 m, 1200 m and 2000 m), with six slope classes being discriminated (Table 3). The results show that TPI provides a generalised representation of topography, while DEV highlights subtle topographic variations, which is in accordance with the findings of [32]. This study focuses on the regional representation of the morphological classes so the generalised approach of TPI was selected, instead of DEV, for the analyses of slope position and landform classification evaluation. The calculated indices of NP and PD, revealed that DEV provides a more fragmented representation of the morphological classes than TPI, with the PD values indicating a higher density of patches for the individual classes (Table 3). The comparison of the NP and PD values between the various neighbourhood sizes, for both DEV and TPI, indicated that the 150 m neighbourhood size shows high fragmentation with a large number of patches and high density of patches. On the other hand, the 600 m and 1200 m neighbourhood sizes highlight a uniform composition of the morphological classes (steady decrease of NP and PD), which results to a high degree of generalization regarding the morphological classes, with particular features not visible at that scale, due to the generalization (Table 3). As a result, the 300m neighbourhood size has been selected as optimum for the rest of our analyses: it discriminates the various features with less fragmentation and diverse NP and PD values, without a high degree of generalization.

thumbnail
Table 3. Comparison of various neighbourhood sizes between DEV and TPI regarding fragmentation, to identify which neighbourhood size better highlights small or large geomorphological features.

https://doi.org/10.1371/journal.pone.0170727.t003

Slope position classification for various neighbourhood sizes

Slope position classification based on TPI for Early, Middle and Late Minoan periods consisted of six classes (Fig 4). The percentages of the slope position classification areal coverage, for the individual Minoan periods, are presented on Table 4. For valley bottoms, a decrease in the coverage percentage exists with increasing neighbourhood sizes, for all Minoan periods. For lower slopes, there is an increase of coverage with a neighbourhood size increase, for all Minoan periods. Gentle slopes, steep slopes and ridges have a steady variable coverage percentage for all Minoan periods. The two notable observations on those landform types are that: i) the upper slopes reveal a high decrease between 100 m and 300 m neighbourhood sizes, for all Minoan periods, but then a steady variation exists; ii) during the Early Minoan period, the ridges class coverage increases gradually with increased neighbourhood sizes; while during the Middle and Late Minoan periods there is a steady variation of ridges class coverage between the neighbourhood sizes. This might be a result that during Early Minoan period the settlements were developed on high hilltops, presumably for defensive purposes, so they were located on rougher terrain relative to Middle and Late Minoan. The neighbourhood size of 600m provides the better option to characterize the region, based on the indices statistics, visual interpretation of satellite imagery, in conjunction with information from the derived geomorphometric variables and the landscape shape/fragmentation indices (NP, PD, PROX and LSI) (Fig 4; Tables 4 & 5). The NP and PD indices indicate that lower neighbourhood size represents a fragmented representation of the morphological classes, with a higher density of patches for the individual classes. As the neighbourhood size increases, a uniform composition of the morphological classes (steady decrease of NP and PD) is observed. These indices, in conjunction with the rest of the indices, imply that the 600 m neighbourhood size is the optimum neighbourhood (Table 5). This particular neighbourhood size defines better the morphological classes, giving equal emphasis on regional and subtle topography, without exaggerating one or other morphological classes as it happens with the rest of the neighbourhood sizes.

thumbnail
Table 4. Slope position classes areal coverage (%) of the case study sites, for the Early, Middle and Late Minoan periods, regarding the six morphological classes for the selected neighbourhood sizes.

https://doi.org/10.1371/journal.pone.0170727.t004

thumbnail
Table 5. Indices used for the evaluation of the slope classification for the various neighbourhood sizes and the selection of the optimum, regarding their shape and fragmentation characteristics.

https://doi.org/10.1371/journal.pone.0170727.t005

Landforms classification for various neighbourhood sizes

The TPI based landform classification for the Early, Middle and Late Minoan periods consisted of ten different landform types, for combined neighbourhood sizes (Fig 5). The combined neighbourhood sizes of 300 m and 1000 m were optimal for characterizing the region, based on the Fragstats indices statistics and visual interpretation of satellite imagery (Table 6). Moreover, the study areas are located on rough terrain with abrupt changes of the topography, so a more generalised overview from the combined neighbourhood sizes of 300 m and 1000 m is considered as optimum for larger scale geo-archaeological interpretations. The more subtle topography determined by the combined neighbourhood sizes of 100 m and 600 m would be suitable for flat areas/benches, with the landform types being less fragmented. The histogram of landform elements percentage for the combined neighbourhood sizes of 300 m and 1000 m reveals the predominant landform types during each of the Minoan periods (Fig 6).

thumbnail
Fig 6. Histogram of landform elements (%) for the combined neighbourhood sizes 300 m and 1000 m.

(a) Kavousi- Vrokastro case study area; (b) Phaistos case study area.

https://doi.org/10.1371/journal.pone.0170727.g006

thumbnail
Table 6. Indices being used for the evaluation of the landform classification for the various combined neighbourhood size and the selection of the optimum neighbourhood, regarding their shape and fragmentation characteristics.

https://doi.org/10.1371/journal.pone.0170727.t006

For the Kavousi-Vrokastro district, two main highlights of the settlements location, from the Early Minoan towards the Late Minoan are: i) high increase on deeply incised streams, valleys and plains; ii) high decrease on open slopes, mid-slope drainages and mid-slope ridges (Fig 6). The settlement types and the number of settlements are presented in Table 7. For the Phaistos district, two main highlights of the settlements location from the Early to the Late Minoan are: i) decrease of high ridge and upper slope locations; ii) steady increase of locations in deeply incised valleys, plains and open slopes (Fig 6). The Phaistos district has less rough terrain than Kavousi-Vrokastro, so the tendency of populations moving to flatter landscape from the Early to Late Minoan, is more pronounced. On the other hand, there is a drop in the number of settlements found on high ridges, which indicates a tendency towards abandoning high-elevation settlements. These interpretations are in accordance with findings from the archaeological surveys of [6,33,38]. Based on Moran’s I analysis of the EM, MM, LM settlements and 100 random sites to check whether the pattern expressed is clustered or random, the outcomes from the Kavousi-Vrokastro study showed the following case study: i) for the 100 random points the z‐score was -0,75 and p‐value 0.45, indicating that the pattern is random; ii) similar observations exist for the EM settlements with the z‐score being 1.05 and p‐value 0.29. However, for the MM settlements the z‐score was 7,76 and p‐value 0, indicating that there is less than 1% likehood that this clustered pattern could be a result of random chance. Furthermore, with LM settlements the z‐score was 3.15 and p‐value 0,001, indicating a clustered pattern. Similar observations also exist for the Phaistos case study: i) for the 100 random points the z‐score was 1.47 and the p‐value was 0.14, indicating that the pattern does not appear to be significantly different than random; ii) similarly random pattern exists for the EM settlements with the z‐score being -0.98 and the p‐value being 0.32. On the other hand, for the MM settlements the z‐score was 5 and the p‐value was 0, indicating that there is less than 1% likehood that this clustered pattern could be a result of random chance; while for the LM settlements the z‐score was 2.84 and the p‐value was 0,0044, indicating a clustered pattern.

thumbnail
Table 7. Settlement types and number of settlements occurring over specific landform types.

https://doi.org/10.1371/journal.pone.0170727.t007

Discussion

This study has quantified the spatio-temporal variations of Early, Middle and Late Minoan settlement locations in the landscapes of Crete. A general trend is observed during the Early to Middle Minoan, with the population moving inland in search of arable land, most settlements initially being located within 1.5 km of the coast. During the Middle Minoan, there was an increase of settlements over low-gradient slopes, upper slopes and mesas in hilly terrain, where they remained during the Late Minoan, probably because of their strong defensive advantage (Fig 6).

The analysis of settlement areas indicates that during the Middle Minoan there was: i) an increase of settlements in deeply incised valleys, with people looking for access to perennial water supplies and; ii) a reduction of the number of settlements in shallow U-shaped valleys, mild slopes & mid-slope ridges, perhaps due to movement of population to more arable land on the plains and U-shaped valley floors (Fig 6) [6,33,38].

In Phaistos district, the Early Minoan settlements are located in heterogeneous landscapes, distributed sparsely over variable terrain. In the Middle Minoan the settlements are found over homogeneous landscapes, such as the plains. In the Late Minoan the largest settlements are found on the plains and lowlands. During the Middle to Late Minoan, the number of settlements decreased, especially at higher elevations; the settlements on the plains remained, while farming increased and population concentrated in the larger settlements [6,33,38].Based on the analysis of the area percentage of the Minoan settlements in Phaistos district, population increased within the deeply incised valleys, perhaps due to the need for perennial water, and on the plains and foot slopes, perhaps due to increased arable land access. However, population decreased on the upper slopes, mesas and high ridges, perhaps as a result of the observed movement to arable lowlands (Fig 6).

The accuracy assessment, comparing the findings of this study with the archaeological surveys of [6] (Table 8), shows a low percentage of no agreement (14%) while there is a substantial percentage of agreement (64% of moderate agreement and 22% of high agreement). That implies that the rapid low-cost geoinformatic techniques used in this study are a cost-effective technique that can complement conventional archaeological surveys.

thumbnail
Table 8. Accuracy assessment, whether there is an agreement or not, comparing the coverage percentage of the individual landforms classes, based on TPI, and the settlement geological-geomorphological description provided from the archaeological surveys of Hayden et al. (2004).

The qualitative agreement type was based on the overall areal percentage coverage of associated landform types: Low (<50%), Moderate (50–70%) and High (>70%).

https://doi.org/10.1371/journal.pone.0170727.t008

Conclusion

During the last decade the quantification of landform types has attracted the interest of the geoinformatic research community, with various studies using GIS-based approaches. The evaluation of geomorphometric datasets can be integrated with GIS techniques, highlighting the information within the interlinked geographic data. The extracted information can be particularly useful for landform classification, as this case study of Crete has illustrated. Such information is useful when linked to geospatial data about the distributions of archaeological sites over space and time.

This investigation of Phaistos, Kavousi and Vrokastro districts has produced valuable information regarding the distribution of settlements during the Early, Middle and Late Minoan periods. Based on the analysis of this study, a general trend is observed during the Early to Middle Minoan, with population moving from heterogeneous terrain at higher elevation, to the lowlands (Fig 6). During the Middle to Late Minoan, the population remained in the arable lowlands, with better organization and concentration in larger settlements.

This study was constrained by the limited amount of paleo-environmental data available for Crete, resulting in a knowledge gap that adds to the uncertainty associated with the interpretations made about the factors driving the variations in the Minoan settlement distributions. Nevertheless, the methodology presented here can provide useful spatio-temporal analyses, at district scales, for future studies to examine at local scales, with associated studies of palaeo-environmental conditions or archaeological predictive modelling. In addition, this study offers valuable information for further research, where socio-economic or political factors can be considered for settlement hierarchy assessments.

Supporting information

S1 Fig.

TPI or DIFF for EM, LM and MM period on Kavousi-Vrokastro region, with six morphologic classes for the neighbourhood sizes: a) 150 m; b) 300 m; c) 600 m; d) 1200 m.

https://doi.org/10.1371/journal.pone.0170727.s001

(TIF)

S2 Fig.

DEV for EM, LM and MM period on Kavousi-Vrokastro region, with six morphologic classes for the neighbourhood sizes: a) 150 m; b) 300 m; c) 600 m; d) 1200 m.

https://doi.org/10.1371/journal.pone.0170727.s002

(TIF)

S3 Fig.

Slope position classification based on TPI of the case study sites of Kavousi-Vrokastro, for EM, LM and MM periods, with six morphological classes for the neighbourhood sizes: a) 100 m; b) 300 m; c) 600 m; d) 1200 m; e) 2000 m.

https://doi.org/10.1371/journal.pone.0170727.s003

(TIF)

S4 Fig.

Landform classification based on TPI of the case study sites of Phaistos, for EM, LM and MM periods, with ten landform types for the combined neighbourhood sizes: a) 100 m and 600 m; b) 300 m and 1000 m; c) 300 m and 2000 m; d) 600 m and 2000 m.

https://doi.org/10.1371/journal.pone.0170727.s004

(TIF)

Acknowledgments

The authors would like to thank the anonymous reviewers for their comments that improved the quality of this manuscript.

Author Contributions

  1. Conceptualization: AA AS RT.
  2. Data curation: AA.
  3. Formal analysis: AA.
  4. Funding acquisition: AS.
  5. Investigation: AA.
  6. Methodology: AA AS RT.
  7. Project administration: AA.
  8. Resources: AA AS.
  9. Software: AA.
  10. Supervision: AA AS RT.
  11. Validation: AA.
  12. Visualization: AA.
  13. Writing – original draft: AA.
  14. Writing – review & editing: AA AS RT.

References

  1. 1. Muller Celka, S., Puglisi, D., and Bendali., F., «Settlement pattern dynamics and natural resources in MM-LMI Crete: the case of Malia», in Physis, L’environnement naturel et la relation homme-milieu dans le monde égéen protohistorique (Aegaeum 37). Edited by G. Touchais, R. Laffineur, and F. Rougemont, 2012: 431–440.
  2. 2. Chryssoulaki, S., «Minoan roads and guard houses–war regained», in R. Laffineur (ed.), Polemos. Le contexte guerrier en Égée à l’âge du Bronze, Aegaeum19, 1999, p. 75–85.
  3. 3. Nowicki, K., «The Historical Background of Defensible Sites in Crete: Late Minoan iiiC versus Protopalatial, » in Polemos, Le contexte guerrier en Egée a l’age du Bronze. Actes de la 7e Rencontre égéenne internationale (Aegaeum 19). Edited by P.P.Betancourt,V.Karageorghis, R.Laffineur, and W.D. Niemeier, 1999: 191–195.
  4. 4. Cunningham, T.F. and Driessen, J., «Site by site: Combining survey and excavation data to chart patterns of socio-political change in Bronze Age Crete», in J.F. Cherry and S.E. Alcock (eds.), Side by Side Survey: Comparative Regional Studies in the Mediterranean World, 2003, p. 101–113.
  5. 5. Haggis D.C., «Archaeological Survey at Kavousi, East Crete. Preliminary Report», Hesperia 65(4), 1996: 373–432.
  6. 6. Hayden, J.B., Dierchx, H., Harisson, W.M.G., Moody, J., Postma, G., Rackham, O., et al., «Reports on the Vrokastro Area, Eastern Crete, Volume 2: The settlement history of the Vrokastro area and related studies», University of Pennsylvania Museum of Archaeology and Anthropology, Philadelphiam, Pennsylvania, University Museum Monograph 119, 2004, p. 427.
  7. 7. Nieto LM., «Geodiversity: proposal of an integrative definition», Boletín Geológico y Minero 112(2), 2001: 3–12.
  8. 8. Gray M., «Geodiversity: Valuing and Conserving Abiotic Nature», John Wiley & Sons: Chichester; 2004, p. 434.
  9. 9. Panizza M, & Piacente S., «La geodiversità e una sua applicazione nel territorio emiliano», Il Geologo dell’Emilia-Romagna 29, 2008: 35–37.
  10. 10. Blaszczynski J.S., «Landform characterization with geographic information systems», Photogrammetric Eng. Remote Sens., 63 (2), 1997: 183–191.
  11. 11. Kvamme, K.L., 1992. «Terrain form analysis of archaeological location through geographic information systems», In: Lock, G., Moffett, J. (Eds.), Computer Applications and Quantitative Methods in Archaeology 1991. BAR International Series, 577. Tempus Reparatum, Oxford, p. 127–136.
  12. 12. Llobera M., 2001. «Building past landscape perception with GIS: Understanding topographic prominence», Journal of Archaeological Science, 28 (9): 1005–1014.
  13. 13. Warren R.E. and Asch D.L., «A Predictive model of archaeological site location in the Eastern Prairie Peninsula», In: Wescott K.L., Brandon R.J. (Eds.), Practical Applications of GIS for Archaeologists. A Predictive Modeling Toolkit. Taylor & Francis, London, 2000: 3–32.
  14. 14. Roughley, C., «Understanding the Neolithic landscape of the Carnac region: a GIS approach», In: Stancic, Z., Veljanovski, T. (Eds.), Computing Archaeology for Understanding the Past—CAA 2000—Computer Applications and Quantitative Methods in Archaeology. Proceedings of the 28th Conference, Ljubljana, April 2000. BAR International Series, 931, Archaeopress, Oxford, 2001: 211–218.
  15. 15. Bevan A. and Conolly J., «GIS, archaeological survey, and landscape archaeology on the island of Kythera, Greece», Journal of Field Archaeology, 29 (1–2), 2004: 123–138.
  16. 16. Fairén-Jiménez S., «British Neolithic rock art in its landscape», Journal of Field Archaeology, 32 (3), 2007: 283–295.
  17. 17. Agapiou A., Lysandrou V., Alexakis D.D., Themistocleous K., Cuca B., Argyriou A., et al., «Cultural heritage management and monitoring using remote sensing data and GIS: The case study of Paphos area, Cyprus», Computers, Environment and Urban Systems, Vol. 54, 2015: 230–239.
  18. 18. Christopherson, G.L., «Using ARC/GRID to calculate topographic prominence in an archaeological landscape», In: Esri User Conference Proceedings of the 23nd Annual Esri International User Conference July 7–11, 2003.
  19. 19. De Reu J., Bourgeois J., De Smedt P., Zwertvaegher A., Antrop M., Bats M., et al., « Measuring the relative topographic position of archaeological sites in the landscape, a case study on the Bronze Age barrows in northwest Belgium», Journal of Archaeological Science, 38, 2011: 3435–3446.
  20. 20. Pike, R.J., «A bibliography of geomorphometry, the quantitative representation of topography-supplement 3 (open- file report 99–140) », 1999, US. Geological Survey.
  21. 21. Tagil S. and Jenness J., «GIS-Based Automated Landform Classification and Topographic, Landcover and Geologic Attributes of Landforms Around the Yazoren Polje, Turkey», Journal of Applied Sciences, 8, 2008: 910–921.
  22. 22. Argyriou A.V., Teeuw R.M., Rust D., Sarris A., «GIS multi-criteria decision analysis for assessment and mapping of neotectonic landscape deformation: a case study from Crete», Geomorphology 253, 2016: 262–274.
  23. 23. Pracilio G., Smettem K., Bennett D., Harper R., Adams M., «Site assessment of a woody crop where a shallow hardpan soil layer constrained plant growth», Plant and Soil, 288, 2006: 113–125.
  24. 24. Patterson J.J., «Late Holocene land use in the Nutzotin Mountains: lithic scatters, viewsheds, and resource distribution», Arctic Anthropology, 45, 2008: 114–127.
  25. 25. MacMillan R.A., Pettapiece W.W., Nolan S.C., Goddard T.W., «A generic procedure for automatically segmenting landforms into landform elements using DEMs, heuristic rules and fuzzy logic», Fuzzy Sets and Systems, 113, 2000: 81–109.
  26. 26. Dragut L. & Blaschke T., «Automated classification of landform elements using object-based image analysis»,Geomorphology, 81, 2006: 330–344.
  27. 27. Speight J.G., «Landform», In: Australian soil and land survey field handbook, McDonald R.C., Isbell R.F., Speight J.G., Walker J. and Hopkins M.S. (Eds.), 2nd Edn. Inkata press, Melbourne, 1990: 9–57.
  28. 28. Tilley, C., «A Phenomenology of Landscape: Places, Paths and Monuments», Berg, Oxford, 1994.
  29. 29. Guisan A., Weiss S. B., Weiss A. D., «GLM versus CCA spatial modeling of plant species distribution», Plant Ecology, 143, 1999: 107–122.
  30. 30. Gallant J.C., and Wilson J.P., 2000. «Primary topographic attributes», In: Wilson J.P., Gallant J.C. (Eds.), Terrain analysis: principles and applications. Wiley, New York, pp. 51–85.
  31. 31. Weiss, A.D., «Topographic Position and Landforms Analysis», Poster Presentation. ESRI, San Diego, CA. Users Conference, 2001.
  32. 32. De Reu J., Bourgeois J., Bats M., Zwertvaegher A., Gelorini V., De Smedt P. M., et al., «Application of the topographic position index to heterogeneous landscapes», Geomorphology, 186, 2013: 39–49.
  33. 33. Watrous L. V., Hadzi-Vallianou D., Blitzer H., «The Plain of Phaistos. Cycles of social complexity in the Mesara region of Crete», Cotsen Institute of Archaeology at UCLA, Monumenta Archaeologica 23, 2004: 668.
  34. 34. Moody J., «Prehistoric and historic climate and weather in (East) Crete». In Molloy B. and Duckworth C., (eds), “A Cretan landscape through time: Priniatikos Pyrgos and Environs”, Archaeopress, BAR International Series 2634, 2014: 23–30.
  35. 35. Bowner B., «Humans: Raft out of Africa: Early hominids as ancient mariners: Hand axes excavated on Crete suggest sea crossings», Science News, Volume 177, Issue3, 2010, 14pp.
  36. 36. Chaniotis A., «Ancient Crete», In: Classics, Oxford University Press, 2010.
  37. 37. Wilford, J.N., «On Crete, New Evidence of Very Ancient Mariners», The New York Times, 2010.
  38. 38. Gesell, G., Preston-Day, L. & Coulson, D.E., W., «The Archaeological Survey of Kavousi region», INSTAP Academic Press, Philadelphia, Pennsylvania, Prehistory Monographs 16, 2004, p. 242.
  39. 39. Hayden, J.B., Moody, J., Rackham, O., «The Vrokastro survey project, 1986–1989», American School of Classical Studies at Athens, JSTOR,1992, p. 66.
  40. 40. Holmgren K., Gogou A., Izdebski A., Luterbacher J., Sicre M-A., Xoplaki E., «Mediterranean Holocene climate, environment and human societies», Quaternary Science Reviews, 136, 2016: 1–4.
  41. 41. Izdebski A., Pickett J., Roberts N., Waliszewski T., «The environmental, archaeological and historical evidence for regional climatic changes and their societal impacts in the Eastern Mediterranean in Late Antiquity», Quaternary Science Reviews, 136, 2016: 189–208.
  42. 42. Watrous, L. V, Xatzi-Vallianou, D., Pope, K., Mourtzas, N., Shay, J., Shay, T., et al., «A Survey of the Western Mesara Plain in Crete: Preliminary Report of the 1984, 1986, and 1987 Field Seasons», Hesperia 62, 1993: 191–248.
  43. 43. Nowicki K., «Some remarks on the Pre- and Protopalatial peak sanctuaries in Crete», AeA 1, 1994: 31–48.
  44. 44. Manning S.W., «The emergence of divergence: development and decline on Bronze Age Crete and the Cyclades». In Matthers C., and Stoddart S., (eds), “Developemnt and decline in the Mediterranean Bronze Age”, Sheffield archaeological monographs 8, 1994: 221–270.
  45. 45. Cadogan G., «Was there a Minoan landed gentry? », BICS, 18, 1971; 145–148.
  46. 46. Mc Garigal, K., Cushman, S.A., Neel, M.C., Ene, E., «Fragstats: Spatial Pattern Analysis Program for Categorical Maps», Computer software program produced by the authors at the University of Massachusetts: Amherst, 2002.
  47. 47. Jenness, J., Topographic position index (tpi_jen.avx) extension for ArcView 3.x. Jenness Enterprises, 2005.
  48. 48. Moran P.A.P., «Notes on Continuous Stochastic Phenomena». Biometrika. 37 (1), 1950: 17–23.