New Perspective on Neolithic sites in the Dubičiai Microregion Using a GIs spatial analysis

The article analyses the Dubičiai microregion, which is distinguished by the abundance and diversity of its stone age settlements, and presents the possibilities provided by GIs spatial analysis in reconstructing this microregion’s Neolithic landscape, distinguishing conditions that affected the selection of the stone age settlement sites, and determining the locations where the greatest possibility exists for the discovery of new, not yet known Neolithic settlements.


INTRODUCTION
The Dubičiai microregion (Varėna District Municipality) is distinguished by an abundance of diverse Stone Age settlements.This microregion began to be investigated archaeologically in the late 19 th century.Landlord Wandalin Szukiewicz conducted searches for Stone Age settlements and collected finds on the banks of the Rivers Ūla and Katra and on the shores of former Lakes Pelesa and Duba.The settlement descriptions and maps he presented have not only yielded valuable information about the archaeological heritage of the Dubičiai microregion, but have also allowed the landscape, which had previously existed and was greatly altered in the 20 th century, to be imagined and reconstructed (Шукевич В., 1893;Szukiewicz W., 1901).The archaeological excavation of Neolithic settlements and the search for new sites have been conducted intensively in the second half of the 20 th -early 21 st centuries.
The Dubičiai microregion has attracted the attention of not only archaeologists, but also geologists, who have devoted considerable space in their works to geomorphological, palaeogeographic, and palaeoeco-logic analyses of the Dubičiai microregion, have conducted palynological and diatom analyses, and have reconstructed the formation processes of the lakes (Stančikaitė et al., 2002;Balakauskas et al., 2012).Nevertheless, despite these diverse analyses, it is difficult to perceive and reconstruct the palaeolandscape in which the Neolithic inhabitants had settled.The sandy plains and peaty meadows currently found in the Dubičiai microregion make it difficult to imagine that large lakes had previously existed there (Fig. 1).A spatial analysis using GIS has not only helped to reconstruct the landscape but has also allowed natural environmental variables that affected the selection of habitation sites to be analyzed and the most favorable locations, i.e., those which hold the greatest possibility for the discovery of as yet unknown Neolithic settlements, to be determined.

REsEaRCH aREa
The Dubičiai microregion is on Lithuania's southeastern border with Belarus.On the west, the microregion is surrounded by continental dunes, most of which run parallel to the katra valley (from the vil-ISSN 1392-6748 DOI: http://dx.doi.org/10.15388/ArchLit.2016.17.10682 lage of Paramėlis), which drains the microregion.On the northeast-east and, in part, the south, the region is bound by the old highlands, which were heavily damaged by the last Weichselian glaciation (the Nemunas) and the foot of which is covered by sandy-gravelly layers deposited by glacial meltwater.The central part of the Dubičiai microregion, i.e., the Katra lowland, has been strongly affected by thermokarst processes, renewed in the late glacial period and furrowed by various-sized proglacial valleys, that run in diverse directions.The studied area covers approximately 186 km 2 in the watershed between the Rivers Ūla and katra on the edge of this sandy plain.The area is fairly level with elevations ranging from 121 to 179 m above sea level a.s.l.(Fig. 2) and is packed by thermokarst hollows and proglacial valleys.The aspect ratio analysis shows a fairly even distribution of the downslope direction in the microregion, but, probably owing to the flow direction of the glacial meltwater, somewhat more of the slopes are oriented to the south-southeast and north-northwest.The slope gradient ranges from 0 degrees in flat areas to 36.86 degrees in the N part of the studied area.The highest location in the microregion is a hill, which lies to the N-NW of Dubičiai village and is an erosional, hilly fragment of the ancient relief (Stančikaitė et al., 1999, p. 69).

RECONsTRUCTION OF THE BODIEs OF WaTER
Up until the late 19 th century, lakes were an important element of the Dubičiai microregion's landscape.In the late 19 th -early 20 th century, owing to the erosion of the Ūla river channel, part of the Katra river drainage basin was captured by Ūla River and the water from Lakes Matarai, Pelesa, and Duba gradually drained into Ūla River.Lake Duba alone shrank over 50 years from 221 ha in 1850 to 20 ha in 1900 (Česnulevičius, Švedas, 2010, p. 148).The drainage of the remnants of the old lakes was completed in 1958-1959 through melioration.A landscape of meadows and pine forests now predominates in the Dubičiai microregion; only some names of the Stone Age sites: Dubičiai Island (Lith.Dubičių salaitė), Margiai Island (Lith.Margių sala), and Lakeshore (Lith.Paežerys) allude to the big    The large lakes of Duba, Pelesa, and Matarai formed at the sites of chunks of ice in an active thermokarstic zone.During the peak of the last glaciation, the Weichselian (Nemunas), the studied area lay at the edge of two glacial lobes.In the vicinity of Merkinė, Nemunas River inundated the marginal moraine belt and the glacier's edge, its waters spreading out widely into the glacier-free territory, the eastern part of which became the present-day lowland of Katra River.
On the basis of the geomorphological, palynological, and diatom analyses, it was determined that up until the Allerød Oscillation a high water level of roughly 128-130 m a.s.l. had existed, but the aeolian processes and the universal drop in the water level that began in the Allerød Oscillation had caused the water level of Lakes Duba, Pelesa, and Matarai to fall to 125-126 m a.s.l.(Balakauskas et al., 2012, p. 122).The lowest recorded water level was in the Preboreal, but an abrupt rise is noticeable in the first half of the Boreal, even to as high as 130 m a.s.l.During 9520-9290 cal.bp, the water level again fell significantly to roughly 127 m a.s.l.(Balakauskas et al., 2012, p. 125).Marked changes in sedimentation are noticeable throughout the Atlantic.A rise in the water level is shown by the diatom assemblages, which display a predominance of planktonic species, from 8100 bp when the water level reached 129 m a.s.l.The previously discrete lakes merged and once again formed a single large eutrophic palaeobasin.Based on the results of the diatom analysis, the water level of Lake Pelesa remained high, at about 130 m a.s.l., in the Early Subboreal but later fell to 124-125 m a.s.l.(Stančikaitė et al., 2002, p. 403).
During the Stone Age, the lakes must have been one of the most important variables influencing the selection of habitation location in the Dubičiai microregion.The significant water level fluctuations that have been presented by geologists in their works are almost entirely indiscernible when looking at the location of the Stone Age settlements; Final Palaeolithic, Mesolithic, Early-Middle Neolithic, and Late Neolithic heritages have been discovered at the same place in the majority of the settlements.The incidence of settlements at an elevation of 126-138 m a.s.l., or roughly 130.5 m a.s.l. on average, in the Dubičiai microregion during the Neolithic when the lake's water level was 129-130 m a.s.l.signifies that part of the area of the settlements must have been under water.Therefore, the hypothesis that they could have been pile settlements should not be rejected.In analyzing a similar natural landscape in East Lithuania's Kretuonas microregion, which still has large lakes, it has also been observed that the majority of the settlements had been founded right beside the water despite the fact that they must have been seasonally inundated (Marcinkevičiūtė, Šatavičius, 2013, p. 560).It is likely that in order to mitigate the fluctuations in the water level, pile buildings were erected at places in the settlements in both the Kretuonas and the Dubičiai microregions.It should be noted that sand, in which wood and other organic materials survive especially poorly, predominates in the Dubičiai microregion.
On the other hand, it is not very likely that the large Late Neolithic-Early Bronze Age settlements of Margiai, Barzdis Forest, or Karaviškės were founded several hundred meters from the peaty shallow shores of Lake Duba in the second half of the Subboreal, when the water level of Lake Duba was roughly 124-125 m a.s.l.
The shores of the ancient Neolithic lakes, which predominantly had a water level of about 128 m a. s. l. (with a ±0.5 m seasonal fluctuation), were reconstructed (Fig. 2) on the basis of the results of the geological investigations, a digital elevation model, and the location of the Stone Age settlements known in the Dubičiai microregion.These bodies of water must have been flowing lakes that were well fed by rivers.During the last 150 years, the channel network has changed significantly, which has led to the disappearance of the huge lakes of Duba, Pelesa, and Matarai.A series of several mid-19 th -century events contributed to the erosion of the River Ūla: circa 1830, the beginning of a drop in the water level of Lake Duba, which belonged to the Katra catchment basin, the 1841 collapse of the Rudnia foundry dam, and the general effect of the erosion of the River Nemunas on the River Katra (Linkevičienė, 2009(Linkevičienė, , pp. 1238(Linkevičienė, -1239)).As a re-sult, the flow direction of the upper Katra changed, and its tributaries - the Nočia, Kaniavėlė, and othersbegan to discharge into the Ūla.Lakes Duba, Pelesa, and Matarai, which lay in the territory captured by the Ūla, disappeared in less than a 100 years.The landscape was further altered by Soviet melioration, which completed the area's drainage and converted the small rivers into melioration channels.A flow accumulation model created using SAGA GIS 3.0.0software on the basis of the digital elevation model was used to reconstruct the former channel network.

THE NEOLITHIC sITEs
In the 1960s, the systematic excavation of Stone Age settlements began in the Dubičiai microregion with Dubičiai 1, 2, 3 (Римантене, 1966), Barzdis Forest (Margiai 5) (Rimantienė, 1999a), and Margiai 1, 2 (Rimantienė, 1999b) settlements.The majority of them were non-stratified, multi-period sandy sites with an abundance of material from various Stone Age periods, mostly flint finds.For example, more than 1000 flint artifacts were discovered in 1 m 2 at margiai 1 settlement (Rimantienė, 1999b, p. 130).Some of the finds from this site are connected with the Final Palaeolithic-Mesolithic, but about half consist of Neolithic finds, including isolated pig, horse, and small ruminant (?) bone fragments and tools (stone hoes and small handstones), which should be associated with Late Neolithic agriculture (Rimantienė, 1999b, p. 158).At the turn of the 21 st century, the settlements of Karaviškės (Piličiauskas, 2012), Gribaša (Grinevičiūtė, 2002), Katra (Girininkas, 2000;Brazaitis, 2000), and Paramėlis (Šatavičius, 2005) were excavated.The Dubičiai microregion is still being archaeologically surveyed in the search for new settlements.A problem concerning names and numeration has occurred in registering this microregion's settlements.A large part of the Stone Age settlements registered by different archaeologists since the late 19 th century have several names: an object is listed under one name in the Register of Cultural Property, while the investigator has published it under another name or investigated the same object as a new, yet unknown settlement.In an effort to systemize the information from all of the investigations that have so far been con-ducted in the Dubičiai microregion, the earlier information identifying the same settlements by different names was reviewed and combined.In addition, the territory included in the settlement sites was checked on the basis of the information from the field evaluations and the physical relief maps.
70 Stone Age settlements occupying a total of 2.766 km 2 , i.e., 1.49% of the entire area in question, are currently known in the analyzed area.Of these, 63 have Neolithic cultural layers: 62 from the Early-Middle Neolithic and 44 from the Late Neolithic.The settlements vary greatly in size: from 4873 m 2 (Kajutis-Matarai 9) to 223550 m 2 (Margiai 5 (Barzdis Forest) Settlement) (Table 1).More than half of the settlements (including those in peat bogs) have been discovered during the last 20 years during field evaluations and surveys conducted by E. Šatavičius (Šatavičius, 2006).
An attempt was made to analyze the Early-Middle and Late Neolithic settlements separately, but the majority of the sites possess heritage from both periods.On the other hand, it must be acknowledged that the dating of the majority of the settlements is preliminary.Even in the more broadly excavated settlements, the more precise dating and cultural attribution of the discovered pottery and flint inventories is a cause for discussion.It should be pointed out that compared with the rest of Lithuania's territory, the highest concentration of settlements connected with the Globular Amphora and Corded Ware cultures occurs in this microregion (Brazaitis, 2005, pp. 222, 236).It would seem that these settlements, which were inhabited by nonlocal, immigrant farmers and animal breeders, should have been founded at new sites, which were perhaps separated by the natural landscape from the earlier Neolithic settlements.In actuality, the absolute majority of the heritage connected with the Globular Amphora and Corded Ware cultures has been discovered at settlements that had existed prior to these cultures.The location of the settlements in the microregion as well as an intrasite analysis of them (Marcinkevičiūtė, 2010) show the continuity of the settlements right up to the Bronze Age.The large areas occupied by the settlements or the fairly thick cultural layer discovered in them probably reflect less the population density and more a long habitation period when the area oc-cupied by the settlement was expanded in some direction in order to take advantage of a clean area uncontaminated by household waste or a periodic, perhaps seasonal, relocation to a nearby area and a subsequent return.The settlements were located at distances of 183-987 m from one another, the average distance being ~631 m.The impression has formed that the population in the Dubičiai microregion was fairly sedentary during the Neolithic, only low scale mobility being very likely.The absolute majority of the Stone Age settlements in the Dubičiai microregion are currently located in meadows or forests/forest margins.Therefore, the role of the former lakes in selecting the settlement sites has usually not been evaluated.Previously, the discovered flint inventory was often interpreted as agricultural or hunting tools, but fishing must have been a no less important food source.The new technologies probably spread from southern regions into the Dubičiai microregion via the rivers, perhaps even by (or more precisely, up) Katra River, which is connected with Nemunas.

ENVIRONMENTaL VaRIaBLEs
In order to investigate the conditions influencing the selection of Neolithic habitation sites, an analysis of the palaeolandscape was made.Nine environmental variables were created and examined using GIS applications: elevation above sea level, a terrain ruggedness index, the slope length and steepness factor (Ls factor), direct solar radiation, distance to the reconstructed lake shores and channel network, a topographic wetness index, altitude above the channel network, and a visibility index.The value of each environmental variable was analyzed in the Dubičiai microregion as a whole and in the area occupied by each Neolithic settlement.In order to evaluate and statistically substantiate that the selection of the settlement sites was not random, a layer of 137 background points covering all of the territory without any archaeological objects was additionally created alongside the 63 Neolithic settlements.Marine Geospatial Ecology Tools, which interface with the R statistical software, were plugged into ArcGIS and used to perform statistical calculations and create density histograms (Roberts et al., 2010).), and the environmental variable values at the settlement sites.

Table (continuation)
Lentelės tęsinys Based on the LiDAR data (©National Land Service under the ministry of Agriculture of the Republic of Lithuania), a digital elevation model (DEM) with a 10 x 10 m cell size (spatial resolution) was created using the ArcGIS 10.3 software.Based on the DEM, other environmental variable raster layers were cre-ated and examined.In analyzing the average elevation above sea level of the area occupied by each Neolithic settlement, it was determined that the Neolithic settlements occur from 128.74 m to 137.48 m a.s.l.The average elevation of all of the settlements was 130.75 m a.s.l.No differences were discerned in separately ana-
The terrain ruggedness index was calculated using ArcGIS focal neighbourhood statistics, an analysis of the changes in the height of all of the neighbouring cells in a radius of 100 m, and a raster calculator equation.The relief in Dubičiai microregion is fairly flat; therefore, values close to 0 predominate; only on the river banks and, in the north-eastern part of the analysed area, the ruggedness of the highlands was noted: from -9.54761 in hollows to +12.5759 at the highest places (Fig. 4).At the Neolithic settlement sites, the terrain ruggedness index ranges from -0.05 to +1.85 with an average of ~0.465.Based on the density histogram for the presence/absence of archaeological sites, the settlements were founded at locations with insignificant rises in elevation wherever the values predominating in the Dubičiai microregion approach 0 (Fig. 3B).
The slope length and steepness factor (Ls factor) is a measure of the sediment transport capacity of over-land flow and describes the effect of topography on soil erosion.With increasing slope steepness and slope length, the transport capacity of surface runoff rises.The Ls factor was calculated using the TAS GIS 2.0.9 software.In the analyzed area, low values of under 0.5 predominate, only on the N slopes of the highest hill in the Dubičiai microregion were values of over 100 obtained (Fig. 5).At the Neolithic settlement sites, the Ls factor values are from 0.026 to 1.534 with an average of ~0.376.The density histogram of the Ls factor values for the presence/absence of archaeological sites and the terrain ruggedness index show that the settlements had been founded at somewhat elevated locations with little slope length or steepness (Fig. 3C).
The ArcGIS software was used to calculate the direct solar radiation on the winter solstice, the shortest day of the year when the least sunlight reaches the northern hemisphere.The greatest direct solar radiation value (up to 89.862 Wh/m 2 ) was recorded on the S slopes of the highest hill in the Dubičiai microregion, while the N slopes of this hill received no direct solar radiation (values close to 0 Wh/m 2 ) (Fig. 6).Theoretically, owing to the warmth provided by the greatest solar radiation even during the winter, the

pav. Tiesioginės saulės apšvietos trumpiausią metų dieną žemėlapis. Sudarė E. Marcinkevičiūtė
most favorable place to live should have been the S slopes of the highest hill, but the long distance from bodies of water and the windiness at this high location determined that this locality was not inhabited.At the Neolithic settlement sites, the direct solar radiation values ranged from 26.36 to 45.25 Wh/m 2 with an average of ~35.56 Wh/m 2 .Based on the density histogram for the presence/absence of archaeological sites, no differences in the direct solar radiation were discernible (Fig. 3D).
In analyzing the distance from the Neolithic settlements to bodies of water based on the layers of the reconstructed lakes (Fig. 7) and rivers (Fig. 8), a distinct influence of bodies of water on the selection of a settlement site is discernible (Figs.9A, B).Of the 63 Neolithic settlements, 42 were within 150 m from a lake shore, 15 were 150-500 m from it, and only six were over 500 m away, but all of them were within 150 m of the channel network.
The topographic wetness index and the altitude above the channel network (vertical distance to the channel network) were calculated using SAGA GIS 3.0.0software.Both of these environmental variables

pav. Atstumo iki rekonstruotų ežerų krantų žemėlapis. Sudarė E. Marcinkevičiūtė
reflect the local water level and the site's wetness.Theoretically, the most favorable locations for founding settlements are considered to be well drained sites that do not flood when the water level fluctuates.The topographic wetness index shows fairly even surface drainage and low soil wetness (Fig. 10), but based on the altitude above the channel network, the bulk of the Dubičiai microregion is within 1 m of the local channel network (Fig. 11).At the Neolithic settlement sites, the topographic wetness index ranged from 5.92 to 10.89 with an average of ~7.41, the altitude above the channel network from 0.02 to 5.74 m with an average of ~1.15 m.Based on the density histogram for the presence/absence of archaeological sites, it was noticed that although the majority of the settlements were close to water, drier sites, where the topographic wetness index was somewhat lower (Fig. 12A) and the elevation above the water level was somewhat higher, were selected for settlements (Fig. 12B).
The visibility index, calculated using Whitebox GAT 3.3 software, allows the visibility of the entire analyzed microregion's landscape to be determined, i.e., the grid cells in the visibility index raster contain visibility   values that express the proportion of the area visible from that site.In the Dubičiai microregion, more than half of the analyzed territory is visible from the region of the highest hill in the N part and from the highlands in the S, part of which are in Belarusian territory (Fig. 13A).At the Neolithic settlement sites, the visibility index values range from 0.014 to 0.664 with an average of ~0.312.Based on the density histogram for the presence/absence of archaeological sites, it was noted that the locations with better visibility were selected for settlements (Fig. 12C).In order to evaluate the intervisibility of the Neolithic settlements, a visibility analysis was performed and the area potentially visible in a 5 km radius from each settlement in the Dubičiai microregion was analyzed.It was determined that from each settlement, a minimum of one neighboring settlement site was visible, but where the settlement density was greater, over 30 neighboring settlements could have been visible, including, for example, 39 from Margiai 5 (Barzdis Forest) Settlement and 41 from Karaviškės IV Settlement (Fig. 13B).It is, however, important to mention that in analyzing the visibility, vegetation was not evaluated.Dense forests greatly reduce visibility, but during the Neolithic, it is likely that forestless areas already existed.Based on the palynological data, signs of a fairly intensive forest burning are already noticeable for the mesolithic and especially the Neolithic (Stančikaitė et al., 2002, p. 406).
An attempt was also made to analyze another important natural environmental variable, i.e., the Dubičiai microregion's topsoil, which should reflect the most fertile locations best suited for agriculture as well as light, well-drained soil best suited to settlements.Unfortunately, the topsoil vector layer of this microregion has been created only fragmentally and owing to the lack of data, a thorough analysis of the incidence of topsoil was impossible.The study area belongs to large Aeolian formations that formed on a limnoglacial sand base.Slightly podzolized, sod-podzolic soil predominates in the region, and, at the former lake sites, peaty deep topsoil from the valley bogs.Based on the palynological data, the first grain of Cerealia pollen was discovered in sediments deposited circa 6500 bp in Lake Pelesa and circa 5900 bp in Lake Duba (Stančikaitė et al., 2002, p. 406).Nevertheless, the loamy sand that predominates on the surface was hardly favorable for the earliest practice of agriculture.It is more likely that first animal husbandry occurred in wet meadows and broad lake floodplains.

PREDICTIVE MODELLING
Predictive modelling is a complex of spatial-statistical methods used to determine the locations where the greatest possibility of discovering archaeological objects exists and to analyze the incidence of such loca-  tions in the natural environment.This is one of the GIS analyses that are very broadly employed, improved, and discussed in archaeology.Deductive or inductive approaches can be used.The former is based on theoretical assumptions about the location best suited for habitation, the archaeological site information being used only for testing purposes, while the latter compares known site data within a study area with datasets of environmental variables and then extrapolates the correlations to areas where no site information is available, usually by means of logistic regression.The logistic regression modelling method allows one to calculate the influence of each analyzed environmental variable in creating a predictive model, shows the probability for the presence/absence of archaeological sites, and allows one to check the model's reliability.The generalized additive model selected for the predictive modelling of the Dubičiai microregion is one of the logistic regression models, which can be fairly successfully used for the predictive modelling of archaeological sites (Łuczak, 2013;Marcinkevičiūtė, Šatavičius, 2013).This model is a non-parametric extension of the generalized linear model, a flexible and automated approach to identifying and describing non-linear relationships between variables and response terms.To create the model, the significance of each environmental variable is automatically calculated on the basis of the Akaike information criterion and the best suited variable combination selected, thereby improving the model's quality.Nine environmental variables were used for the modelling: elevation above sea level, the terrain ruggedness index, the slope length and steepness factor, direct solar radiation, the distances to lakes and rivers, the topographic wetness index, the altitude above the channel network, and the visibility index.In the beginning, a generalized additive model fitting 80% of the randomly selected data for the presence/absence of a site (training data) was created.It was determined that the six most significant variables were elevation above sea level, the distances to lakes and rivers, the terrain ruggedness index, the topographic wetness index, and the visibility index (pvalue < 0.0001), while the altitude above the channel network was less significant (p-value < 0.05), and direct solar radiation and the slope length and steepness factor were not significant.Only significant variables (first seven) were selected for the final model.The employed model was tested using the data from the remaining 20% of the sites (testing data).Its accuracy was determined using the area under the receiver operating characteristic curve (the area under ROC curve).The closer the value is to 1, the better the model's ability to predict the presence/absence of archaeological objects and the less random it is.The accuracy of the predictive model for Neolithic settlements in the Dubičiai microregion was very high as the area under ROC curve was 0.974.After the model fitting and testing procedure, a probability map for the occurrence of archaeological sites was generated (Fig. 14).Although seven environmental variables were employed in the predictive model, it is seen from the probability map that bodies of water had the greatest influence on the selection of habitation sites.Despite high probability prediction values, all predictive models have to be verified by a field survey and improved.Despite the reliability of the theoretical statistics, the results must be evaluated cautiously, because the variables selected for analysis reflect present-day or reconstructed natural conditions that could differ significantly from those that existed in the past.

CONCLUsIONs
Southeast Lithuania's Dubičiai microregion, which is distinguished by the abundance and diversity of its Stone Age settlements, especially Neolithic ones, was selected for the analysis.In ancient times, several large lakes, namely the Duba, Pelesa, and Matarai had existed in this region, but disappeared in the early 20 th century.
Despite the abundant and multidisciplinary scientific information regarding this region, information that would help in understanding the settlement patterns and the factors that may have influenced the selection of sites in the past is still lacking.In order to understand the incidence of Neolithic settlements in the palaeolandscape, GIS applications, as well as geological and archaeological data, were used to reconstruct the ancient bodies of water.
The settlements in the Dubičiai microregion were systemized and the extent of the area they occupied and their chronology were checked.There are 70 Stone Age settlements, 63 of them Neolithic, in the research area.Various environmental variables were analyzed at each settlement site.Any attempt to distinguish environmental variables characteristic of only final Palaeolithic-Mesolithic, Early-Middle Neolithic, or Late Neolithic settlements was groundless.many of the Neolithic settlements also had early layers and it would appear that the variables affecting the selection of habitation sites remained the same from the earliest period right through to the Bronze Age.
A palaeolandscape analysis was performed in order to study the conditions that influenced the selection of Neolithic settlement sites.In using density histograms to analyze the environmental variables at those locations where there were Neolithic settlements and those where there were none, the influence of the distance from bodies of water, the elevation above sea level, the terrain ruggedness index, the topographic wetness index, the visibility index, and the altitude above the channel network was noticed.
The significant environmental predictors were used to create a generalized additive model to predict the probability of a Neolithic site existing at a specific location.Predictive modelling is an important and useful method not only in the search for new archaeological sites, but also in the analysis of the settlement patterns and the environment.The very accurate predictive model and probability map that were created and tested reflect the especially large influence that bodies of water had in the selection of Neolithic habitation sites.acknowledgements I would like to thank Dr. Egidijus Šatavičius of the Department of Archaeology, Vilnius University for his advice on the palaeolandscape of the Dubičiai microregion and an opportunity to use unpublished survey information for the archaeological objects.I would also like to thank Anna Łuczak, PhD student at University of Wroclaw for very useful tips on the predictive modelling of archaeological sites when using a generalized additive model.

Fig. 1 .
Fig. 1.The currently known Stone Age sites in the Dubičiai microregion.Basis: a 2012 aerial photograph.For a list of the Stone Age settlements, seeTable 1. Composed by E. Marcinkevičiūtė.

Fig. 2 .
Fig. 2. A digital elevation model of the analyzed area.The reconstructed lakes and rivers are marked.Composed by E. Marcinkevičiūtė.

Fig. 3 .
Fig. 3. Density histograms of the environmental variables at locations with a Neolithic settlement present/absent: a) the elevation above sea level; b) the terrain ruggedness index; c) the slope length and steepness factor; d) direct solar radiation.Composed by E. Marcinkevičiūtė.

Fig. 7 .
Fig. 7. Map of the distance to the reconstructed lake shores.Composed by E. Marcinkevičiūtė.

Fig. 9 .
Fig. 9. Density histograms of the distance to lake shores (a) and the channel network (b) at locations with a Neolithic settlement present/absent.Composed by E. Marcinkevičiūtė.

Fig. 12 .
Fig. 12. Density histograms of the environmental variables at locations with a Neolithic settlement present/absent: a) the topographic wetness index; b) the altitude above the channel network; c) the visibility index.Composed by E. Marcinkevičiūtė.