Citation
Biogeography of neandertals

Material Information

Title:
Biogeography of neandertals site distribution patterning in the Southern Italian Middle Paleolithic
Alternate title:
Site distribution patterning in the Southern Italian Middle Paleolithic
Creator:
Knox, Kelsey Isabel
Language:
English
Physical Description:
1 electronic file (154 pages). : ;

Subjects

Subjects / Keywords:
Neanderthals -- Italy ( lcsh )
Mousterian culture -- Italy ( lcsh )
Stone implements -- Italy ( lcsh )
Mousterian culture ( fast )
Neanderthals ( fast )
Stone implements ( fast )
Italy ( fast )
Genre:
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )

Notes

Review:
Most research on the Middle Paleolithic in southern Italy has focused on Mousterian sites in the region of Apulia. This research has been extensive and productive; therefore it stands to reason that other, less researched regions of southern Italy, including Basilicata, Calabria, and Campania, hold potential to produce more Mousterian sites. This paper uses ecocultural niche modeling and predictive site modeling to explore site distributions and the past Mousterian niche as predicted from site occurrence data and environmental variables. This Mousterian niche is then projected onto the landscape. This projection produces a potential site distribution map when modeling modern site niche against modern environmental variables, and produces a spatial distribution map when modeling the past niches with past environmental layers. Results indicate that producing potential Mousterian site distributions is a successful exercise, which can identify areas with in under researched regions of southern Italy that have a high probability of presence of sites. Additionally, a general description of the Mousterian niche in glacial and interglacial periods is produced, and a potential biogeographic variable, the Apennine Mountains, is identified.
Thesis:
Thesis (M.A.) University of Colorado Denver
Bibliography:
Includes bibliographic references,
System Details:
System requirements: Adobe Reader.
General Note:
Department of Anthropology
Statement of Responsibility:
by Kelsey Isabel Knox

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
921892896 ( OCLC )
ocn921892896
Classification:
LD1193.L43 2015m K66 ( lcc )

Downloads

This item has the following downloads:


Full Text
BIOGEORAPHY OF NEANDERTALS: SITE DISTRIBUTION PATTERNING IN
THE SOUTHERN ITALIAN MIDDLE PALEOLITHIC
by
KELSEY ISABEL KNOX
B.A., University of California, Los Angeles, 2012
A thesis submitted to the
Faulty of the Graduate School of the
University of Colorado in partial fulfillment
of the requirements for the degree of
Masters of Arts
Anthropology Program
2015


2015
KELSEY ISABEL KNOX
ALL RIGHTS RESERVED


This thesis for the Master of Arts degree by
Kelsey Isabel Knox
has been approved for the
Anthropology Program
by
Tammy Stone, Chair
Julien Riel-Salvatore
Jamie Hodgkins
June 9, 2015
11


Kelsey Isabel Knox (M.A., Anthropology)
Biogeography of Neandertals: Site Distribution Patterning in the Southern Italian Middle
Paleolithic
Thesis directed by Professor Tammy Stone
ABSTRACT
Most research on the Middle Paleolithic in southern Italy has focused on
Mousterian sites in the region of Apulia. This research has been extensive and
productive; therefore it stands to reason that other, less researched regions of southern
Italy, including Basilicata, Calabria, and Campania, hold potential to produce more
Mousterian sites. This paper uses ecocultural niche modeling and predictive site
modeling to explore site distributions and the past Mousterian niche as predicted from
site occurrence data and environmental variables. This Mousterian niche is then
projected onto the landscape. This projection produces a potential site distribution map
when modeling modern site niche against modern environmental variables, and produces
a spatial distribution map when modeling the past niches with past environmental layers.
Results indicate that producing potential Mousterian site distributions is a successful
exercise, which can identify areas with in under researched regions of southern Italy that
have a high probability of presence of sites. Additionally, a general description of the
Mousterian niche in glacial and interglacial periods is produced, and a potential
biogeographic variable, the Apennine Mountains, is identified.
The form and content of this abstract are approved. I recommend its publication.
Approved: Tammy Stone


ACKNOWLEDGEMENTS
Thanks to my advisor, Julien Riel-Salvatore, and my friends and family for all the
support and guidance.
IV


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION.......................................................1
II. THE DISTRIBUTION OF MOUSTERIAN SITES IN ITALY.......................4
Neandertals in Italy: An Overview...................................4
Prehistoric Archaeology in Italy: A Historical Review...............6
Italian Geography: Apulia, Campania, Basilicata, and Calabria......12
Explaining the Pattern of Sites: Discussion........................22
III. ECOLOGICAL NICHE MODELING:
HUMAN AND NON-HUMAN SPECIES........................................27
Development of the Niche Concept...................................28
Applications and Roles of GIS Technologies.........................30
Theoretical Foundations of Ecological Niche Modeling (ENM)........31
Ecocultural Niche Modeling (ECNM)..................................38
Predictive Site Modeling...........................................43
Extra Theoretical Considerations...................................47
Hypotheses.........................................................51
IV. RESEARCH METHODS....................................................53
Known Sites in Southern Italy......................................53
Environmental Data Layers..........................................57
Software: GARP with Best Subsets...................................70
V. RESULTS.............................................................78
Data and Model and Preparation.....................................78
Selection of Outputs...............................................79
v


Jackknife Test of Variables.......................................80
The Niche of Mousterian Sites in Southern Italy...................83
The Niche of Mousterian Sites in the Last Glacial Maximum........92
The Niche of Mousterian Sites in the Last Interglacial Period....97
Comparison of the Niche Models...................................101
VI. DISCUSSION........................................................105
Occupied Niche versus Actual Niche...............................105
Model Similarity.................................................108
Model Generalization.............................................Ill
Mousterian and Neandertal Sites as a Homogenous Assemblage.......114
Biogeographic Variable: The Apennine Mountains...................117
VII. CONCLUSIONS AND FUTURE DIRECTIONS.................................119
WORKS CITED............................................................123
APPENDIX...............................................................137
A: Geologic Age, Soils, Land Cover Values........................137
B: Comparison of Niche Values....................................141
vi


LIST OF TABLES
TABLE
1. Known Sites in Southern Italy........................................55-56
2. Environmental Layers....................................................60
3. Jackknife of Environmental Layers.......................................82
4. Ni che Overl ap Stati sti c s...........................................89
5. Niche Breadth Statistics................................................90
vii


LIST OF FIGURES
FIGURE
1. Distribution of Mousterian Sites in Southern Italy.........................5
2. Mountains of Italy........................................................13
3. El evati on of S outhem Italy.............................................14
4. Landsat Image of Southern Italy...........................................15
5. Soil Erosion Risk in Southern Italy.......................................15
6. Climate of Southern Italy.................................................16
7. Population Density of Southern Italy......................................17
8. Southern Italian Provinces and Regions..................................21
9. Hypothetical Niche........................................................30
10. Environmental Envelope....................................................32
11. Theoretical Foundations of Ecological Niche Models........................34
12. Geographic Space and Environmental Space..................................36
13. Ecocultural Niche Modeling................................................71
14. Selecting Best Models with GARP Best Subsets Protocol.....................76
15. Modem Niche at 4km Resolution.............................................85
16. Modem Niche at 1km Resolution.............................................87
17. Productive Areas for Future Research Identified...........................92
18. Glacial Niche at 4km Resolution...........................................94
19. Glacial Niche with -80m Sea Level.......................................96
20. Interglacial Niche at 1km Resolution.....................................100
21. The European Neandertal Niche............................................102
viii


22. Model Extrapolation......................................................108
23. Similarity of the Modeled Niche..........................................110
24. Productive Areas for Future Research Identified..........................110
25. Comparison of the Niche Models...........................................116
26. Biogeographic Barrier of Southern Italy..................................118
IX


LIST OF ABBREVIATIONS
ABBREVIATIONS
ENM: Ecological Niche Modeling
ECNM: Ecocultural Niche Modeling
AUC: Area Under the Curve
ROC: Receiver Operating Characteristics
GCM: General Circulation Models
GARP: Genetic Algorithm for Rule-Set Prediction
LGM: Last Glacial Maximum
LIG: Last Interglacial Period
OIS: Oxygen Isotope Stage
x


Supplementary Materials: Microsoft Access Database Table with information about
known southern Italian Mousterian sites, titled Known Southern Italian Mousterian
Sites is included. Each row represents a Mousterian depositional layer, and the
information recorded includes a general description of lithic and non-lithic tools, counts
of lithic and non-lithic tools, fauna, raw material, symbolism, structures, cultural
attribution, site type, discovery, excavation and publication dates, and finally a list of all
sources referencing the particular site. A references cited document is also included, titled
References of Known Southern Italian Mousterian Sites. Geographic coordinates are
presented separately in the text (Table 1).
xi


CHAPTERI
INTRODUCTION
This thesis applies ecocultural niche modeling (ECNM) to investigate potential
distributions of Mousterian sites in southern Italy, as well as to investigate the Neandertal
niche in glacial and interglacial periods, as modeled based on Mousterian sites. Because
of the uneven distribution of Mousterian sites, southern Italy is productive study area to
explore how applications of ECNM can be applied to model the Mousterian niche during
various climatic regimes. Additionally, this method and study area is well suited to
determine if the modern niche Mousterian sites can be modeled and connected to ideas of
predictive site modeling, to produce a potential site distribution map, in order to direct
future research into unknown sites in the region. Overall, this research aims to explore
how sites might be patterned against modem and past environmental variables, and how
this patterning can be used to understand where sites are located on the landscape today,
and where they were located in the past.
The distribution of Mousterian sites in southern Italy can also be seen as function
of historic and geographic variables, that has led to an explosion of research in the region
of Apulia, while other southern Italian provinces of Campania, Calabria and Basilicata
have been largely overlooked. Because Apulia has produced many Mousterian sites, it is
possible these under-researched regions contain more sites than previously thought.
These geographical and historical factors, and how they have produced the distribution of
known sites in southern Italy, are considered in Chapter II.
1


Chapter III reviews the theoretical background of ECNM, which draws heavily on
ecological niche modeling (ENM), but focuses on modeling human species, as opposed
to non-human and considers culture in the modeling process. A brief literature review of
ECNM as applied to archaeological data is reviewed in this chapter. Additionally
theoretical considerations of predictive site modeling are mentioned, and concepts of
ECNM and predictive site modeling are explicitly connected to facilitate interpretation of
modeling outputs as site distribution maps.
In Chapter IV the research methods deployed in this study are reviewed and
explained in detail. Geographic coordinate collection is explained, and the environmental
data layers used in the modeling are described. Finally, data manipulation, data
processing, and the parameters of the modeling are explained.
Results are reviewed in Chapter V. Visual outputs in the form of site and spatial
distribution maps are presented, as well as a discussion of the more important
environmental variables that effected the model. Areas that may be productive locations
for future research are identified. Additionally, results of niche overlap and niche breadth
statistics are reported, with a comparison of the modeled European Neandertal niche from
previous studies and the modeled southern Italian Mousterian niche from this study.
Chapter VI first discusses how similarities between all the models, identified both
through visual comparison and through statistics of niche overlap and niche breadth,
indicate that the model was ultimately successful. Additionally, generalizations about the
glacial and interglacial niches are made; including how the niche may have shifted as
environmental conditions change. Also in this chapter, the suitability of considering
Neandertal and Mousterian sites as a homogenous unit when doing ECNM is reviewed.
2


Finally, the Apennine Mountains are considered as a potential biogeographical barrier in
southern Italy.
Chapter VII reviews conclusions of the project, and suggests future directions for
research; including expansion of the presence point collection area and the modeling
extent to all of Italy, and potentially other refugia areas in Europe. Additionally,
explorations of the biotic niche of the region, in the Eltonian sense, are considered, along
with applications of the modeled niche to the Italian island of Sicily.
3


CHAPTER II
THE DISTRIBUTION OF MOUSTERIAN SITES IN SOUTHERN ITALY
Level and intensity of archaeological research varies across geographic space.
Much of this variation can be linked to historical processes that define some areas as
more attractive and viable for archaeological study than others, as well as natural features
that make some areas more or less difficult to explore. Southern Italy is no exception.
Neandertals in Italy: An Overview
Both Neandertal, sites with human remains, and Mousterian, sites with only
Mousteiran lithic technology, archaeological sites have been found in Italy. Generally in
Italy, Mousterian sites span from oxygen isotope stage (OIS) 8 through OIS 3; (Mussi
2001) however, there are no known Mousterian sites in Italy from OIS 8, and they are
sporadic until OIS 5 (Mussi 2001:101). Lithic technologies found at these Italian
Mousterian sites are generally recognized to follow the frameworks established from the
French archaeological record by Francois Bordes (Bordes and Sonneville-Bordes 1970);
researchers have found the Typical Mousterian, Denticulate Mousterian and Quina
Mousterian, although Italian archaeologists often subsume the Denticulate Mousterian
and the Quina Mousterian in to broader category of the Charentian Mousterian. A
Pontinian industry has also been defined, which is thought to be a local adaptation of the
Quina Mousterian to small pebbles of chert that are one of the only sources of raw
material in the region (Mussi 2001: 104). Additionally, Italian researchers have identified
the Quinson Mousterian, characterized by small sized tools and carinated forms with a
dihedral ventral face and Clactonian notches. This typology is used to date sites, as
evidence of the technology is considered absent after OIS 5 (Mussi 2001: 105). A final or
4


evolved Mousterian, characterized by a decrease in sidescrapers, an increase in notches
and denticulates, and the presence of Upper Paleolithic tool types including blades, has
also been identified at Italian Mousterian sites.
Overall, with the exception of Sicily and Sardinia, there are over 350 Mousterian
sites (Milliken 2000:11) found in all areas of the country, including along the coast, in the
interior valleys, and on the high mountains. Southern Italy, the focus of this study, has
approximately 110 sites (Milliken 2000). Although generally southern Italy contains a
large number of Mousterian sites, as Figure 1 demonstrates, these sites are mostly
concentrated in the region of Apulia. Here, historical processes in the development of
Italian archaeology, in addition to geographic features, including natural, social and
political, are reviewed in an attempt to explain the patterning of Mousterian sites in
southern Italy, including the high concentration of sites in Apulia, and lesser
concentrations in Campania, Basilicata and Calabria (Figure 1).
Figure 1: Distribution of Mousterian sites in Southern Italy.
5


Prehistoric Archaeology in Italy: A Historical Review
Knowledge of the history of Italian archaeology, including trends in theories and
methods, is essential to understanding why Italian archaeologists have undertaken
research on Mousterian and Middle Paleolithic sites in specific areas. In this way, a
historical review of Italian archaeological history can help to explain the present day
distribution of known sites in southern Italy.
Like all of modern Western archaeology, Italian archaeological thought is derived
from ideas first developed in the Italian Renaissance (Guidi 1987; Mussi 2001), including
antiquarianism, comparative methods, and classification of the past into stages. The
fashion of antiquarian collections was born in Tuscany during the second half of the 15th
century (Guidi 1987: 237) and the first comparative method was used by Michele Mercati
to conclude that stone tools, thought by the Romans to be tools of mythical heroes and
Medieval Italians to be the tip of lightening blots, were actually produced by flint
percussion during a time before iron was used (Mussi 2001: 6; Guidi 1987: 237). In 1725,
Giambattista Vico proposed that human kind developed in stages, from the Age of the
Savage to the Age of Gods, of Heroes and of Men. Together, these concepts developed
into evolutionary-chronological theories in Italy, which continue to play an important role
in the study of prehistory today (Guidi 1987: 238). Despite being the birthplace of the
first archaeological theorizing and research, from the 17th to the 19th centuries, Italy
lagged behind other European countries in the development of archaeological thought.
These events included the advent of the reactionary and oppressive climate of the
Counter-Reformation (Guidi 1987: 238), the late formation of a centralized state in Italy,
and scholars focus on classical archaeology (Guidi 1987: 238).
6


In most other European countries, the birth of prehistoric studies was tied to the
emergence of the bourgeoisie, a social class that developed with the Industrial Revolution
and the creation of centralized states. The Italian state, unlike other European countries, is
relatively young, and previous to unification consisted of various independent smaller
states or kingdoms that had long and diverse histories (Guidi 1996c: 108). In 1861, with
the unification of the Italian state, and thus the full participation of Italy in the Industrial
Revolution and the development of the bourgeoisie, interest in prehistory resurged. After
unification, the history of Italian archaeological research can be divided into five phases:
Phase 1: 1860-1900, Phase 2: 1901-1921, Phase 3: 1922-1945, Phase 4: 1946-1970, and
finally, Phase 5: 1971-present (Guidi 2010).
Phase 1: 1860-1900
Local Italian antiquarians, drawing on the positivist ideas and concepts of
uniformitarianism and evolution in order to understand the past, drove archaeology in the
1860s. In the early stages of this period, Italian scholars were focused on reconstructing
the Italian Bronze age (Guidi 1987, 2010; Palma di Cesnola 1991). Also at this time,
foreign scholars were using paleontology to attempt to define periods of the Paleolithic at
Italian and other European archaeological sites.
Later, as the Italian state gained more centralized control, archaeology shifted
from antiquarians to the Roman School and its founder, Luigi Pigorini. The Roman
school coalesced with the appointment of Luigi Pigorini as the chair of Paleoethnology
(Italian term for prehistory) in 1877 in Rome. Pigorini attempted to centralize all Italian
research, and he and his students claimed almost absolute autonomy with regard to the
research on prehistory, including the terminology, the evolution of cultures, and methods
7


of study and the theoretical foundations (Palma di Cesnola 1991: 12). They argued for
ethnography as the only way to understand the prehistoric past and theorized that most
cultures evolved unilinearly and locally. Overall, the theoretical focus of this school was
cultural-historical. Additionally, the Roman School rejected the French classification
system of the Paleolithic, particularly the Upper Paleolithic, instead preferring to consider
the Italian Paleolithic as completely separate from other Paleolithic chronologies of
Europe. In fact, they claimed that there was no Upper Paleolithic in Italy, instead the Late
Mousterian evolved directly into the early Neolithic (Guidi 1987, 2010; Mussi 2001;
Palma di Cesnola 1991).
By 1900, Pigorini had successfully consolidated almost all archaeological
research in Italy, and prevented the formation of any more chairs of archaeology outside
of Rome. Despite this, scholars from the south like Ridola in Bascilicata, Taramelli in
Sardinia, and Paolo Orsi in Sicily began a long and profitable series of work which was
to prove important in the construction of the chronological and cultural framework of
prehistoric southern and insular Italy. (Guidi 1987)
Phase 2: 1901-1920
This period was characterized by fighting between the Roman and Florentine
Schools of archaeology, and their associated theoretical backgrounds. While Paolo
Mantegazza founded the Florentine School around the same time as the Roman School
(Mussi 2001; Guidi 1987, 2010; Palma di Cesnola 1991), it was not until the early
1900s, when Aldobrandini Mochi and G.A. Blanc became prominent within the
Florentine School that the conflict developed. The Florentine scholars took a more
naturalistic approach to understanding prehistory, which focused on the study of humans
8


in the context of the environment, and attempted to fit Italian chronology to broader
chronologies derived from continental Europe (Palma di Cesnola 1991). Through
excavations of Grotta Romanelli, these scholars were able to concretely establish the
presence of the Upper Paleolithic in Italy (Mussi 2001: 9). Unfortunately, because of
infighting between the two schools, this whole period was marked by a decline in
methods and increased isolation of Italian scholars from the rest of archaeological
research in Europe (Guidi 2010: 15).
Phase 3: 1922-1945
The fascist period in Italian history also had a marked influence on Italian
prehistoric archaeology. Emphasis shifted to classical archaeology as the Fascist Party
attempted to demonstrate continuity between their government and ancient Rome. The
shift toward classical archaeology was also driven by the development of the cultural
hegemony of [Bemedetto] Corce (Guidi 1987: 242). Croce argued against evolution, and
argued that there was no way to understand the pre-written past, and therefore efforts to
do so should be abandoned. The Roman School was most affected by these changing
perspectives on prehistory, while the Florentine School continued their naturalist path by
attempting to understand the geological-environmental setting of the evolution of the
industries of the Italian Paleolithic (Guidi 1987: 242). Unfortunately, their connections
with the larger European community continued to wane as scholars became convinced
Italy was unique, an idea fostered by fascist ideals. Despite the separation between the
two schools, the excavations at the Arene Candide in Liguria during this period were the
first collaboration between a classical scholar, Luigi Bemabo Brea and naturalistic
scholar, Luigi Cardini. This excavation slowly began the unification of culture-history
9


ideas of the Roman school and naturalistic ideas of the Florentine school, and was the
first precursor of modem prehistoric Italian archaeology (Guidi 1987, 2010).
Phase 4: 1946-1970
After the end of the World War II, and with the waning of power of the Fascist
Party, there was a sudden national revival in all fields of social and cultural life, including
prehistoric archaeological research. Many prehistory chairs were created throughout the
Italian university system. This period in the United States saw the rise of 1960s
archaeological revolution, or new archaeology; however, in Italy, adoption of this
theoretical framework did not extend beyond the implementation of statistical analyses in
lithic studies (Guidi 2010). Croces lingering influence continued to support a climate of
opposition to functional approaches and opposition to experimental archaeology that the
New Archaeology advocated. Additionally, the trend of isolationism continued as most of
Europe began using the Bordes system of lithic classification, while Italian scholars
focused on the Laplace method (Mussi 2001: 11).
Also at this time, the general idea of Italy as an isolated peninsula at the bottom
of Europe, even the recipient rather than the instigator of change, (Barker and Hodges
1981: 2) which was encouraged by cross dating Italian prehistory sites from other areas
of Europe was disrupted by the development of radiocarbon dating. This led to Italian
scholars considering models of functional stability and internal evolution as drivers of
culture change rather than invading cultures from the north (Barker and Hodges 1981: 2).
Ultimately, this may have encouraged further isolation.
10


Phase 5: 1970-present
Significant change occurred during the 1970s as a group of northeastern Italian
scholars emerged who associated themselves with theoretical approaches and up-to-date
excavation methods from English archaeologists (Guidi 1987, 2010). However, despite
the use of new theories and methods, in the tradition of the past a significant portion of
the new work was still cultural-historically focused. The use of mathematics and personal
computers that began in the 1960s continued to spread, as well as the appointment of
prehistoric archaeologists into various State Offices for Antiquities throughout the
country. Additionally, amateur archaeologists saw another resurgence; in fact amateurs
located many of the new prehistoric sites. In the 1980s, archaeological approaches
became localized within regions and areas of Italy. Southern Italian archaeologists
developed a practical approach based on data publications and the organization of local
museums and congresses, and a group of Campanian archaeologists developed a
postprocessual approach. However, most of the research on the Paleolithic was, and
continues, to be done by northern archaeologists. These archaeologists began to use a
mainstream approach, which focuses on differences between northern Italy and the rest
of the country, as well as a processual approach with a strong interest in Middle Range
theory and intensive computer applications (Guidi 2010: 18).
Finally the 1990s saw a slight shift in some circles toward an interest in lithic
technology, microwear analysis, zooarchaeology and taphonmy (Milliken 2000: 13).
The current challenge of Italian archaeology is the integration of the historical approach
with the archaeological, in addition to attempting to integrate local and regional
archaeological trends (Bietti 1991).
11


Italian Geography: Apulia, Campania, Basilicata, and Calabria
Geography, including natural, social and political factors, also inherently plays a
role in site distribution. Italy has extreme differences in geography between each region,
as well as within some regions, and so it is likely these factors can help explain the
present day distribution of sites.
Italy is a long, narrow, extremely mountainous peninsula. It extends the Alps at
47 degrees north, to Sicily at 37 degrees north, at the same latitude of Algeria, Tunisia,
and Syria. The unique characteristics.. .are better expressed by numbers: within an area
of 300,000 km sq., there are 125,000 km sq of hills, and 100,000 km sq of mountains,
encircled by some 9,000 km of marine coasts (Mussi 2001:1). The terrain of Italy
translates into three types of elevation, mountain areas above 1,000 meters with valleys at
various elevations throughout, plains which consists of level or gently sloping land below
approximately 300 meters, and the intermediate hill areas between the two (Cole 1966:
20). Italy is also a zone of tectonic activity including volcanoes and earthquakes. This
tectonic activity has contributed to the development of the mountain ranges of Italy
including the Alps in the north of the country and the Apennines which run in
northwestern-southeastern parallel ranges through most of the peninsula of Italy (Mussi
2001: 2; Figure 2). Due to the presence of the Apennines, east-west movement across the
country is difficult; however due to valleys, north-south movement through the peninsula
is comparatively easier.
12


Figure 2: Mountains of Italy: Italian Alps and Apennine Mountains. Modified from
www.shiney7.co.uk/Gustavline.html
The Apennines make movement difficult throughout all four regions, however
this far south they are made of many distinct mountain areas and do not form a
continuous range. Additionally, in the south these mountains do not usually reach higher
than 2,000 meters, and generally are around 1,000 meters in elevation (Cole 1966: 214;
Figure 3).
A long history of human habitation, dense settlements and the practice of
agriculture have significantly altered the vegetation of southern Italy (Figure 4). Ancient
Roman settlements first started the process of land reclamation and deforestation; these
processes continue to the modem day. Large grazing herds have altered mountain and
flatland environments since the Late Middle Ages in central and southern Italy.
13


Currently, the vegetation of southern Italy falls into the Apennine-type; the typical tree is
the oak, while olive, oleander, and pine are found in more coastal areas. The foothills are
characterized by oak and pine. Mountain locations and areas of higher elevation still
preserve ancient mountain forest. Beech woods are still present in Calabria and Apulia
and areas of deforestation are now been replaced by a scrub bush (Ermoli and di Pasqule
2002: 212). The combination of human modification of the landscape and mountainous
terrain with steep valleys means the whole region is at risk of erosion (Figure 5).
Figure 3: Elevation of Southern Italy. From Hydrosheds 1km.
14


Figure 4: Landsat Image of Southern Italy. ESRI images.
Figure 5: Soil Erosion Risk in Southern Italy. From the European Soil Database From
Grimm et al. 2003, Pangos et al. 2012.
15


Climate is affected by multiple factors, but as a rule there are overall increasing
temperatures and decreasing precipitation from north to south (Mussi 2001: 5). However,
the mountains and sea are never far away from any area of land, which causes in large
variations of local climates and annual rainfall levels, both between and within the
regions (Cole 1966; Mussi 2001). For example, on the west coast rainfall is usually
around 800 mm per year, on the east annual precipitation is approximately 700 mm per
year, while areas in a rain shadow, such as the Tavoliere plain in Apulia, only receive 500
mm of rain per year (Mussi 2001, Figure 6).
Figure 6: Climate of Southern Italy. Annual precipitation and annual mean temperature.
From BioClim.
16


Southern Italy was a single political unit before unification, the Kingdoms of
Naples, and because of this Campania, Apulia, Bascilicata and Calabria together form a
reasonably distinct geographic region (Cole 1966:214). Currently, and throughout time, it
is the poorest area of Italy, with the lowest per capital income in the whole country, and
the lowest quintiles on the development index (Faini et al. 1993). The population is
concentrated in the city of Naples and the region of Apulia (Figure 7). Additionally,
population can be considered as a proxy for development, more populated areas of
southern Italy tend to have more infrastructure, more development of industry, and a
higher annual income. Transportation throughout the area is difficult; roads are generally
adequate; however the railway system suffers from one-way tracks and difficulties
navigating the mountainous areas (Cole 1966).
Figure 7: Population Density of Southern Italy. Can also be viewed as proxy for
development. From the GIS Data Depot at geocomm.com.
17


Apulia
Apulia is by-passed by the Apennine Mountains, and so is distinct from the other
southern Italian regions. It consists mainly of low limestone plateaus and plains with little
elevation changes and so is one of the least mountainous regions of Italy (Cole 1966:
216). However, the Gargano Peninsula contains its own, structurally distinct mountain
formation (Cole 1966: 214). This formation is found in the Foggia province, and it
contrasts with the other five provinces of the region, Barietta-Andria-Trani, Bari,
Taranto, Brindisi and Lecce (Figure 8), all of which have no mountain formations and are
generally lacking in elevation change (Cole 1966). Much of the land is used for
agriculture or shepherding, and there is some industry in the larger cities (Cole 1966).
This region has the second highest population in southern Italy, with approximately 4.1
million people (Comuni Italiani 2004), and most of the population is concentrated in the
cities.
Apulia has one of the densest concentrations of known Mousterian sites in Italy.
In this region, there are approximately 90 sites, almost as many sites as the provinces of
northern or central Italy combined (Milliken 2000: 42-43). The first collections of
Mousterian artifacts from Apulia were compiled by amateur archaeologists in the mid-
1800s, and the first book addressing the Paleolithic of the region, titled Preistoria della
Puglia, was published in 1914 by Antonio Jatta. Beginning in the early 1900s, more
systematic excavations by P.E. Stati and E. Regalia were carried out; these were then
taken over by the scholars from the Roman school, such as Pigorini, Antonielli, and
Rellini, as well as those from the naturalist school, including A. Mochi, G. A. Blanc, and
Romanelli (Palma di Cesnola 1967). In the early 1930s, Rellini, R. Battaglia, and E.
Bamgaertei carried out extensive surveys and excavations in the Gargano Peninsula.
18


Starting in the 1950s and through the 1970s, excavations at known sites and surveys to
identify new sites had spread to all areas of Apulia through the research of scholars such
as A. Palma di Cesnola, Borzatti von Lowenstem, Cardini and various others (Palma di
Cesnola 1967). Recent research has mostly focused on revisiting these previously
excavated sites to determine more accurate stratigraphy, or to better investigate an older
and deeper stratigraphic layer.
Campania
Campania contains large regions of the Apennine Mountain range; approximately
35% of the province consists of mountains, while 51% consists of hills, and 14% consists
of plains. However, despite this large amount of mountain range, the plains are
exceptionally fertile and agriculture is extremely important and productive, in this region.
Additionally, there are extensive arable valleys and agriculturally productive hills in the
provinces of Benevento and Avellino (Cole 1966: 221; Figure 8). This is the only
province in southern Italy with a large concentration of industry, centered in the city of
Naples. Campania has a population of approximately 5.8 million (Comuni Italiani 2004),
which makes it the most populous region in the south. However much of this population,
approximately 1 million, is concentrated in Naples (Comuni Italiani 2004). The mountain
areas of Campania are generally not occupied at all, while in the hill areas populations
densities are extremely high (Cole 1966: 221-3).
This region has approximately 10 known Mousterian sites, considerably less than
Apulia. Unlike Apulia where the sites are evenly distributed across the landscape, the
majority of Mousterian sites in Campania are located along the coast. For the most part,
these sites were identified and excavated in the 1970s, by many of the same scholars who
19


excavated a large number of the cave sites in Apulia. Generally, the inner areas of
Campania have been ignored in terms of archaeological research (Radmilli 1978).
Basilicata
This region is the least known of all the regions in Italy, both to the Italians and to
foreigners. Almost two thirds of Bascilicat is classified as mountains while the rest is
classified as hill (Figure 8). There is intense erosion in this province due to natural factors
such as floods and human-created deforestation. Seismic activity is also high in this
province. The western provience, Matera, contains high limestone massifs, with karst
phenomena on the high plateaus, while the eastern provience, Potenza, consists of newer
land of high hills and alluvial soils (Cole 1966). The province of Potenza contains
mountains with deep, narrow valleys that are not particularly productive for crops.
Villages are extremely isolated and there is only one city. This is the poorest area of Italy
(Cole 1966: 224-5). On the other hand, Matera is at a lower elevation, less rugged, and
there is a relatively large area of coastal lowland that allows for more extensive
agriculture. Neither province has any significant industrial development, although some
is present. Population is at approximately 600,000, making Basilicata the least populated
region in the south.
Basilicata has approximately four known Mousterian sites. The areas of the
provience that have been the subject of research are generally along the coast, or
extremely close to the shared borders of the provience, at the expense of the inner areas.
Calabria
Calabria is even longer and narrower than the rest of the Italian peninsula. High
mountains and proximity to the sea dominate much of the geography of this region (Cole
20


1966). There is one large valley and most of the mountains steeply slope down to small
coastal or alluvial plains at their feet; these steep slopes continue at least several hundred
meters below sea level (Cole 1966: 225). Erosion from seasonal torrential rivers is
common in Calabria. Due to railways, the west coast of the province is more connected to
the rest of Italy when compared to the east; however travel is difficult in both areas. The
majority of land use is agricultural, and the remaining land consists of some of the largest
forested areas in Italy (Falcucci et al. 2007). Population of this region is larger than
Basilicata, but only about half as large as Apulia, with approximately two million people
occupying five provinces (Figure 8). Five Mousterian sites have been located within this
region, and they, like sites in Campania, are distributed in coastal areas or along borders
of the province.
Figure 8. Southern Italian Provinces and Regions. From the GIS Data Depot at
geocomm.com
21


Explaining the Pattern of Sites: Discussion
Distribution of Neanderthal sites through southern Italy can be seen as a function
of both historical and geographic variables working in concert to create disparities in the
level of research into the Middle Paleolithic of the regions of Apulia, Campania,
Basilicata and Calabria.
Generally, Italian archaeological theories should be discussed as trends rather
than real traditions of research. These trends have been deployed in various ways within
various political climates in Italy, particularly in regards to nationalism. In the early
periods of Italian archaeology, the pioneers of Italian prehistory were all professionals
from northern Italy. They were also all members of the same upper social class and
northern regional population that was responsible for the unification of the various states
and kingdoms on the Italian peninsula. In original synthesis of Italian history, written by
these early archaeologists, the development of prehistoric Italy was explained by the
spread southward, during the Bronze Age, of various waves of northern populations who,
superimposing themselves on the natives of Neolithic origin, invented new forms of
settlements, such as lake dwellings (Guidi 1996b: 112). The newly unified Italian
government paid particular attention to this version of prehistory, likely because it
reinforced the narrative of Italian unification. In contrast, during fascist times, the trend
shifted from prehistoric archaeology to classical, as the Fascist Party attempted to show
continuity from Roman times to their regime to legitimize their rise to power (Guidi
1996b: 113). After the 20 years of centralizing fascist regime archaeological trends have
moved toward the development of local traditions. The combination of a lack of
centralized push for Paleolithic research combined with the more recent development of
22


local archaeological traditions may help explain the dispariety in levels of Mousterian
and Middle Paleolithic research between and within regions of southern Italy.
The basis of known Mousterian site distribution also can be directly tied to ideas
of Italian uniqueness, both in terms of regions and country, and a unilinear evolutionary
bias. To define culture, Italian archaeologists draw on cultural-historical theories;
cultures are archaeological facies, characterized by industrial assemblages made up of
particular types of artefacts which represent normative ideas and mental templates of past
people (Milliken 2000: 11). They also draw on the naturalistic perspective to argue that
changes in the cultural facies can be seen as changes in industrial assemblages; however
in Italy this change is only conceptualized as unilinear (Bietti 1991: 261; Milliken
2000:11). Therefore, concepts of interstratification are not considered viable to
understand the archaeological record, and it is generally assumed that each region has its
own local development of Mousterian assemblages. So, comparisons between two
regions to help understand a lesser-studied region are not common (Bietti 1991, Mussi
2000: 12).
These concepts of uniqueness and unilinear evolution also lead to a focus on the
establishment of a reference or type site (Mussi 2000:12) which is then used to
interpret the chronological order of all other sites in the region. Because of the and
importance of site types, Italian archaeologists typically focus on excavating deep
trenches at cave and rockshelter sites, at the expense of both horizontal excavation and
systematic survey. This has led to a serious bias in many regions, like Calabria,
Campania and Basilicata, where most of the sites found are located in caves or
rockshelters, as well as bias against research in areas lacking this type of stratified type
23


site, like Basilicata and Calabria (Mussi 2000:12). Additionally, northern archaeologists,
who were the first to adopt more modem archaeological techniques during Phase 5, have
undertaken the few systematic planned surveys that have been done in Italy, and have
generally concentrated efforts in the north (Mussi 2000:12).
Finally, much of the archaeological work in the south, outside of Apulia has
focused on the Classical Period, likely due to a combination of a holdover of this focus
from Phase 3 and local tradition. For example, there is a whole series of Archaeology in
South Italy articles from the Archaeological Reports journal, which ran from the 1970s
through the 1980s that only focuses on Classical Archaeology (e.g. Ridgway 1982).
Geographic variables are also important to explain the distribution of known sites;
some of the most important variables are relief, population and development. Apulia,
with its high density of sites has one of the lowest average elevations of southern Italy,
which makes it an attractive region for archaeological surveys, especially when compared
to the other provinces. Additionally, the karstic areas mean caves are common, which
also attracts archaeologists to research in the region. The higher, fairly evenly distributed
population of the region in addition to the high focus on agriculture increases chance of
encounters with Mousterian sites across the province. Finally, the productive agriculture
and levels of industry means more construction projects are undertaken where sites may
be found, and there is more money available to excavate and research those sites.
Combined, all these factors may be driving where archaeological research has been done.
Site pattering in Campania follows similar patterns; there are more sites in this
region than Basilicata and Calabiria because there are more people, as well as large areas
of plains and more development due to industry and productive agriculture. However,
24


areas with relief where people have not settled are not usually investigated, and surveys
within the region are rare, likely due to difficulties of surveying at elevation, and a focus
on Classical Archaeology (Arthur 1991).
Basilicata and Calabria can be discussed together do to similarities in their
geography. Both have intense mountainous relief that dominates the landscape, and both
have less productive agriculture, little to no industrialization and a smaller population
than the other two provinces. Thus, it is not unexpected that the Middle Paleolithic
research in these regions has been neglected. Surveys that have been undertaken are small
and rare (e.g. Hodder 1984, Small, et al. 1998,) and have not been productive in
producing Middle Paleolithic sites.
One could make the argument that the sites Campania, Basilicata and Calabria are
lacking in Mousterian sites because Neandertals were present in these regions. This is not
likely the case. First, Mousterian sites have been found at high elevations in Armenia
(Pinhasi et al. 2011) and Romania (Hoffecker and Cleghom 2000), and have also been
found throughout the eastern Italian Alps (Milliken 2000: 45). Italy, particularly southern
Italy, was also a refugia during glacial periods in the Pleistocene (Hewitt 2000).
Therefore Neandertals could have potentially occupied the area at varying densities
throughout the species time period. Finally, the presence of Acheulean sites in
Campania, Basilicata and Calabria, at higher frequencies than Neanderthal sites in
Basilicata and Calabria, (Lefevre et al. 2010; Piperno and Tagliacozzo 2001; Radmili
1978), demonstrate that Pleistocene sediments are not completely eroded away in these
regions.
25


Past and modem politics combined with geographic variables can help explain
why certain regions in southern Italy have experienced an extensive amount of research
(Apulia), others have a middling amount of research (Campania), and still others
(Basilicata and Calabria) have been ignored. It is important to consider why sites are
patterned so that present day archaeologists can move past simplistic assumptions that
conclude all sites have been found in a region, or that sites are not present in a region.
Further research should take into account these deficits in Mousterian research in
southern Italy, and should be directed toward identifying potential areas within southern
Italy that might be productive in terms of finding more sites to broaden our picture of
Neandertals and Mousterian site locations in Italy. Therefore, distribution of sites in
southern Italy provides a potentially productive study area to explore how applications of
ECNM can be used to model the southern Mousterian niche and potential spatial
distribution of Mousterian sites during various climatic regimes, as well as to determine if
an assemblage composed of all known Mousterian sites can be modeled and connected to
ideas of predictive site modeling to produce a potential site distribution map.
26


CHAPTER III
ECOLOGICAL NICHE MODELING: HUMAN AND NONHUMAN SPECIES
Ecoloigcal Niche Modeling is a method of species distribution modeling which
estimates the actual or potential geographic distribution of a species. The modeling
begins with the characterization the environmental conditions in which the species can
live, or the niche, and then continues by identifying that suitable environment in other
locations within the research area. A spatial distribution map is produced as a visual
representation of the niche on the landscape. This method was first developed to
understand distributions and niches of non-human species; however, recently Banks et al.
(2006) reconceptualized ENM as ecocultural niche modeling (ECNM) and applied it to
human species to understand the niche of humans in the past. This chapter reviews these
concepts, including their methods and their theoretical backgrounds, and also considers
predictive archaeological site modeling as another method of modeling distributions of
past peoples.
Unlike ENM and ECNM, which seek to model niche, predictive archaeological
site modeling seeks to model the distribution of sites on the landscape. Despite differing
theoretical goals, all three methods are related through a use of similar inputs and through
a smiliar modeling process. This section will review how ENM, ECNM, and predictive
site modeling are connected. By connecting these three methods, this study aims to
interpret ecocultural niche outputs as both potential site distributions and potential past
spatial distributions to explore productive areas for future research in southern Italy, and
to explore Mousterian site distributions during glacial and interglacial periods in southern
Italy, respectively.
27


Development of the Niche Concept
As implied in the name, ENM heavily relies on the niche concept, and as such,
ENM cannot proceed without an explicit definition of the term (Araujo and Guisan 2006:
1679; Guisan and Thuiller 2005: 998; Pearson 2010). This section reviews the variation
in definitions of niche and how these defintions have varied thorugh time.
Species distribution maps have long been used by scientists to study the range a
species occupies, however prior to the development of ENM methods they were either
shaded outline maps, which were created by experts extrapolating between and beyond
known sightings of species, or they were simple dot maps depicting where the species
had been observed. However, these methods tended to over-predict or under-predict,
respectively, the range of the species of interest (Anderson, et al 2002). In an attempt to
rectify this issue, ecologists began to look at the relationship between environment and
the distribution of species (Soberon and Peterson 2005: 1). This relationship was first
recognized in the 1800s, with the work of Charles Darwin (Chase and Leibold 2003),
and today, it is termed niche. Currently, the quantification of such species-environment
relationships [niche] represents the core of predictive geographical modeling in ecology
(Guisan and Zimmermann 2000: 148), and is the ultimate goal of ENM and ECNM.
Grinnell (1917) was the first to use the niche concept in a research program; he
defined it as the set of environmental requirements in which a species can survive, and
outside of which the species cannot survive. Elton (1927) elaborated on this, and defined
the niche of an animal as its place(s) in the biotic environment, its relations to food and
enemies (11). This definition of niche does not assume the organisms within the niche
are passive; instead it views organisms as active agents in creating their niche through
interactions and alterations of their physical environment (Elton 1927: 11; Guisan and
28


Thuiller 2005: 998). Hutchison (1957) further defined the niche concept as an n-
dimensional hypervolume (416) made up of abiotic and biotic variables, the conditions
and relationships of which describe the environment in which an organisms lives (Figure
9). Along with this new definition, Hutchison (1957) also conceptualized and quantified
the differences between the fundamental niche and the realized niche. The fundamental
niche includes all the potential areas that could support the species in question, as
opposed to the realized niche, which is made up of the areas within the fundamental niche
where the species is actually present. Realized niche has the potential to be smaller than
the fundamental niche due to competition between species, or other confounding factors
(Guisan and Thuiller 2005; Soberon and Peterson 2005; Walton 2009). Finally, Chase
and Leibold (2003) propose a refined definition of niche, which includes Grinellian and
Eltonian definitions, as well as the fundamental and realized niche of Hutchinson. They
define niche as
the joint description of the environmental conditions that allow a species to
satisfy its minimum requirements so that the birth rate of the local population is
equal to or greater than its death rate along with the per capita effect of that
species on these environmental conditions. (15).
Definitions of niche in ENM depend on the research question(s) being asked, and
what ENM is actually quantifying. This paper draws on the Grinellian and fundamental
definitions of niche, as discussed later in this section.
29


Factor X
Figure 9: Hypothetical Niche. Hypothetical depiction of a three-dimensional value (three
factors) within Hutchisons (1957) ^-dimensional hypervolume niche. The area within
the cube represents the total available amount of each factor, while the area within the
sphere represents the amount of each factor needed for a given species to survive, i.e. its
niche. From Chase and Leibold 2003:9.
Applications and Roles of GIS technologies
As ENM aims to quantity the relationships between species distribution and
environment, GIS technologies, and associated software have been useful in the
successful implementation of the method. Use of GIS allows researchers to gather
appropriate environmental layers representing real world conditions, and process them
into the correct format for application in ENM and ECMN. Additionally, GIS allows the
projection of the models onto the landscape, and can turns jumbled collection of rules for
where niche is located into an actual, understandable visual representation of niche by
identifying areas of the environment where those rules apply. These GIS technologies
have been applied to understand a wide variety of research topics in biology including
guiding field surveys to find known and unknown species (Bourg et al. 2005; Guisan et
al. 2006; Raxworthy et al. 2003), predicting effects of climate change and species
invasions (Berry et al. 2002; Hannah et al. 2005), exploring speciation mechanisms and
30


species delimitation (Graham et al. 2004; Raxworthy et al. 2007), comparison of
paleodistribution and phylogeography (Hugall et al. 2002), assessing disease risk
(Peterson et al. 2006; Peterson, Benz and Papes 2007) and more (list from Pearson
2010:62 and Guisan and Thiuller 2005). Additionally, the ENM method has been
modified as ECNM for application to human species through the modeling of
archaeological sites, reviewed in the Ecocultural Niche Modeling section later in the
chapter.
Theoretical Foundations of Ecological Niche Modeling (ENM)
At its most basic, ENM draws on environmental factors represented in spatial
layers to produce potential habitats, and then identifies, or projects, those environments
on to the landscape. The model does this by stacking environmental raster layers, and
then placing a layer with planametric (x,y) presence data on top of this stack. Based on
the environmental values at each presence point, the model is able to describe that
species niche as the environmental envelope or the environmental tolerance ranges,
which include the environmental and biological conditions that allow a species to survive
in an area (Figure 10). It is this environmental envelope that defines the niche, which is
projected on the landscape to identify areas of high or low potential for species presence.
Since the niche is being modeled, not just where presence points are found, the model is
able to extrapolate other areas of the study region which fall within the environmental
envelope. Once the projection is complete, the environmental tolerances can be described
with descriptive statistics such as the mean, mode, and standard deviation, and analyzed
through methods such as jackknifing, to determine which are most important to a species
(Banks et al. 2011; Beeton et al. 2013; Walton 2009).
31


Species
Occurrence
Record
pH
Elevation
Sand
Precipitation
Imagery
Figure 10: Environmental Envelope. The species environmental envelope will identify
the value for each variable at points of presence data. From Walton 2009: 38.
Given the various definitions of niche, researchers have often struggled with
determining the correct definition to use, and have also struggled to determine which type
of niche is being modeled through the ENM and ECNN process. Resolving these issues
has been the focus of many theoretical papers (Araujo and Guisan 2006; Guisan and
Thuiller 2005; Soberon 2007; Soberon and Peterson 2005).
Soberon and Peterson (2005) review the theoretical foundations of the ENM
method, in an effort to clarify the issues above. They identify four different classes of
variables which all effect niche, all which must be accounted for in order to completely
describe the niche of a species. These include abiotic conditions, biotic conditions,
accessible regions for dispersal, and finally evolutionary capacity of populations
(Soberon and Peterson 2005: 2). Abiotic conditions are the physiological limits on
species ability to persist in an area, including variables such as climate and physical
environment, while biotic conditions include those factors that make up the set of
32


interactions with other species that can affect the species ability to maintain population.
Accessible regions help to differentiate between the fundamental and the realized niche
of a species. The evolutionary capacity of a species is important to address potential
change in niche; however this is difficult to model, and is usually removed from
modeling and assumed to have a negligible effect for most purposes (Soberon and
Peterson 2005: 2).
Since evolutionary capacity is not of major interest to the modeling process, we
can assume the species that is the focus of modeling will be present at the given point or
area where the other three conditions meet (Figure 11). The first condition that must be
met is region A, which includes the regions where abiotic conditions are favorable for
species presence. Region B, or condition two, represents the suit of biotic variables,
including the presence of needed resources and absence of competitors, and finally
condition three, or region M indicates the region where species are actually able to
access; this can be constrained by factors such as mountain ranges (Soberon and Peterson
2005: 3). Region A, or the suite of abiotic variables, can also be thought of as
fundamental niche, in the Hutchisonian sense, while region B, the suite of biotic
variables, can be thought of as a niche in the Eltonian sense. The overlap between region
A and region B represents a world in which (1) abiotic conditions are suitable for
positive population growth of the species; and (2) required mutualists are present and the
suite of competitors, predators and diseases present will not prevent positive fitness
(Soberon and Peterson 2005: 3). The intersection of A, B, and M is termed P, or the
perfect area (niche) where populations should be found. Soberon and Peterson (2005)
argue that most ecological niche models produce results that describe some aspect of the
33


fundamental niche, which is appropriate for looking at large scaled models and
overarching distributions. This is useful for applying ENM as ECNM to at Mousterian
and other archaeological sites, as the biotic variables and human effect on the niche is
extremely complicated both in modeling and in the production of representative
environmental layers to be used in the modeling process. This study specifically aims to
model fundamental niche, however because not all abiotic variables were considered, it
likely only includes a portion of that niche. Therefore, this study discusses the potential
fundamental niche of Neandertal and Mousterian sites, as well as potential locations
where we would expect to find sites within this niche.
A Abiotic or the B = Biotic or the
Fundamental Niche Realized Niche
Figure 11: Theoretical Foundations of Ecological Niche Models. It is the intersection of
abiotic (A), biotic (B) and movement (M) in which the perfect (P) environmental
conditions exist. Adapted from Soberon and Peterson 2005 and Walton 2009: 37.
34


Geographic and Environmental Space
Another aspect of the theoretical background of EMN is the concepts of
geographic and environmental space. Understanding these aspects can help illuminate the
type of niche being modeled and can help facilitate interpretation of results.
Geographic space is the actual distribution of where a species is located. It is
represented by the first panel in Figure 12, and typically is referenced using x and y
coordinate data so that it can be located in the real world. The pluses represent observed
presence points, the grey color is the actual occupied niche, and the solid lines represent
the potential occupied niche. A and B represent considerations that can cause actual
occupied niche to be smaller than potential occupied niche; these include factors such as
a lack of observed presence points in an area that species actually occupies, represented
by area A, or areas such as B, which the species could occupy but does not due to
movement constraints or interactions with other species.
The environmental space of ENM is represented by the right panel in Figure 12,
and is the conceptual space defined by the environmental variables to which the species
responds (Pearson 2010: 59). Environmental space can be considered similar and
equitable to Hutchinsons fundamental niche, and is represented by the solid black line
surrounding area E, as is called actual niche Also present in environmental space is the
occupied niche, represented in grey in the right panel, which is similar to Hutchinsons
(1957) realized niche, which is defined as the area of the actual niche where the species is
actually living. As shown in the map, there are areas in this occupied niche where there
may not be any presence points, area D in the figure, and the occupied niche may not be
the same size as the actual niche because of movement constraints and interactions
between species. Occupied niche can also fail to represent actual niche if bias in presence
35


point location prevents the modeled niche from predicting into areas which are expected
to be occupied.
The patterning of movement, occupied niche, potential occupied niche,
competitive exclusions, and location of presence points all alter how the environmental
space is conceptualized. This is demonstrated by the differences between the panels in
Figure 12, and also how the projection of a species distribution model is represented back
onto the landscape from its conception in environmental space. Therefore, biases in the
data, such as sampling bias in presence point data, can have an adverse effect on the
environmental space ENMs are able to produce. Additionally, because of these two
differing theoretical and actual spaces, the model can produce two different types of
outputs; the first type of output aims to verify the actual distribution of species, while the
goal of second type is to produce potential habitats. To successfully interpret ENM
outputs, these concepts must be kept in mind.
Geographical space Environmental space
+ Observed species occurrence record
Actual distribution (left panel)/Occup*ed niche (right panel)
C J Potential distribution (left panell/Fundamental niche (right panel)
Figure 12: Geographic and Environmental Space. Illustration of the relationship between
a hypothetical species distribution in geographic space and environmental space. From
Pearson 2010: 60.
36


Genetic Algorithm for Rule Set Prediction (GARP)
An algorithm typically performs the modeling aspect of the ENM process. A large
variety of algorithms can be used in ENM and ECNM; the nature of the research question
and the nature of the data being modeled both drive algorithm selection. For this
particular research question, the Genetic Algorithm for Rule-Set Prediction or GARP
(Stockwell and Peters 1999) was chosen to model Mousterian site distribution because it
is appropriate to the study at hand, and for consistency with previously published studies.
When ENM was first developed and applied, most researchers used multivariate
statistical analysis to determine and describe a species niche, with inconsistent results
(Stockwell and Noble 1992). Much of that inconsistency stemmed from unreliable and
variable data, both environmental layers and presence points make applications of
multivariate statistical methods difficult (Stockwell and Noble 1992). This problem is
exemplified in the logistical regression approach to species distribution modeling, where
static rules are produced that must be applied to all the data, regardless of the quality of
presence point or type of environmental layer. In contrast, GARP, as a machine learning
technique, is able to produce various different types of rules which can only be applied
when certain conditions are met. When the conditions of application are met by a subset
of the data, the rule is applied. When they are not, the algorithm selects another, more
appropriate rule (Stockwell and Peters 1999: 146). This method allows varying quality of
data and varying types of data (i.e. both categorical and continuous) to be modeled
together.
The GARP algorithm is appropriate to this particular study area because of its
demonstrated ability to extrapolate niche into un-sampled areas of the study region at
high probability levels (Peterson, Papes and Eaton 2007). A previous study that evaluated
37


the strength of models, Elith et al. (2006), ranked GARP as one of the lowest performing
algorithms on the basis of area under the curve (AUC) values. However, this study was
focused on the algorithms ability to successfully model heavily sampled regions, rather
than the ability to project modeled niche into new areas. Peterson, Papes and Eaton
(2007) demonstrated that in cases where presence point sampling was not evenly
distributed throughout the study region, GARP was more successful at predicting new
regions of distribution than other algorithms. Also, Peterson, Papes and Eaton (2007) and
Peterson et al. (2008) review the use of the AUC value as a form of model evaluation,
and conclude that AUC values are generally underestimated in GARP models due to the
way the curve is produced. As this study seeks to project species distribution models into
under-sampled areas to predict where new Mousterian sites may be found, GARP is the
appropriate choice. Additionally, applications of ECNM, discussed in the next section,
typically apply GARP; for consistency this study applies the same algorithm.
Ecocultural Niche Modeling (ECNM)
ECNM, or the application of ENM to human species, was first presented by
Banks et al. (2006), was developed from a series of workshops specifically organized to
explore the application of ENM to humans and archaeological data. The method fo
ECNM is based on the idea that
prehistoric human populations were influenced by climate change and resulting
environmental variability, and developed a wide variety of cultural mechanisms to
deal with these conditions. In an effort to understand the influence of
environmental factors on prehistoric social and technical systems, there is a need
to establish methods with which to model and evaluate the rules and driving
forces behind these human-environment interactions (Banks et al. 2006: 69).
Just like predictive species modeling, ECNM combines the concept of the niche with
geospatial modeling tools to predict the occurrence of species on a landscape through
38


georeferenced site locality data and sets of spatially explicit environmental data layers
(Lozier et al. 2009:1); however due to culture and its role in human adaptation to the
environment, the niche is conceptualized as a both related to environmental and cultural
variables, or ecocultural. Additionally, cultures are considered to be human adaptations to
specific environments, and so differing cultures are assumed to have differing niches
(Banks et al. 2006).
Identification of the geography and variability of past culturally coherent human
groups is critical to understanding the complex mechanisms that have shaped the
interactions among genetics, linguistics, cultural affiliation and climate (Banks et al.
2006: 69). The goal of the application of ENM to human-environment interaction through
ecocultural niche modeling (ECNM) is to turn archaeological description into prediction
- to model the ecology of human and hominin population in the past (Banks et al.
2006:69). Typically, ECNM models reconstruct geographic distributions of
archaeological defined populations by determining the ecocultural niches those
populations inhabit.
ECNM can be applied anywhere there is occurrence data of archaeological sites
and reconstructed environmental data. It has been applied all over the world including,
Asia and North and South America, but its greatest florescence is in Europe. This is
likely due to a long history of archaeological and paleoenvironmental research that has
created strong data sets and quantified many environmental variables, thus allowing
robust implementation of ENMs to human species. Banks, dErrico, Peterson, Vanhaeren
(2008) use ECNM to identity the habitable areas of Europe during the Last Glacial
Maximum (LGM), as well as to compare the ecological niches and geographical
39


distributions of the Solutrean and the Epigravettian. The results of the EMN match
archaeological data of human extent during the LGM, and the models also suggest that
the Solutrean and Epigravettien cultures occupied different niches and thus can be
considered cultural adaptations to the environment. ECNM has also been used to
demonstrate that differences between the Proto-Aurignacian and Early Aurignacian is due
to an ecological niche expansion between the two periods (Banks et al. 2013). This study
is the first demonstration of how a cultural adaptation was used to expand a human
populations ecological niche. Banks, dErrico, Peterson, Kageyama et al. (2008) use
ECNM to look at Neandertal and anatomically modern human populations in Europe
during the Paleolithic. These researchers consider two hypotheses for the contraction of
the Neandertal range, either as a response to changing climate, or as a result of
competition with expanding anatomically modem human populations. Modeling
indicates that the second hypothesis is the more likely explanation. North and South
American case studies have focused on Paleoindian dispersal and adaptation. For
example, Gilliam et al. (2007) uses occurrence data of Clovis points in North America to
identify the niche characteristics of where the Clovis culture is found, and then
extrapolates that information to Asia in an attempt to determine the source areas of pre-
Clovis culture. Case studies from Central Asia include examination of hominin presence
during the late Pleistocene in order to identify the abiotic characteristics that best predict
species distribution during interglacial and glacial periods (Beeton et al. 2013: 1).
40


Paleobiogeographical and Archaeological Considerations
The theoretical background that applies to ENM also applies to ECMN. Terms
must be defined and explicitly connected to ecological theory, including the definition of
niche being used and the type of niche being modeled. However, there are some
challenges that apply specifically to ECMN, including modeling the past, and biases with
environmental layers and with archaeological presence data
ECNM is based on the idea that the past can be modeled, either by using
environmental data reconstructed to approximate past conditions, or through hind-casting
where a model calibrated with data for the present is used to predict the range of past
species in modern environments. ECNM must consider concepts of niche stability, niche
evolution and ecological niche shifts. Niche stability only takes into account the
similarity of climatic conditions that allow a single species to persist through time
(Nogues-Bravo 2009:522), however the ideas of niche evolution and ecological niche
shifts demonstrate that niches are not always stable. In ECNM, the stasis or dynamism of
the niche allows interpretations of human adaptations to the environment in the form of
cultural adaptations, but they can also pose problems when attempting to model the past
if they are not considered.
Environmental predictors also pose problems. Not only does the researcher have
to be aware of the type, amount, and relationship of climatic variables, in ECNM the
researcher must also be aware that environmental variables can be inherently biased due
to their generally coarse-grained outputs of climatic scenarios generated from General
Circulation Models (GCMs). GCMs simulate the general circulation of the atmosphere
and oceans of the Earth through the assumption of certain initial conditions (Varela et al.
41


2011), including carbon dioxide levels, temperature, winds, cloud cover etc., and run a
large number of iterations until they stabilize and produce a picture of climate in the past.
An additional problem that comes from the use of GCMs is the focus on abiotic variables.
As discussed earlier in the chapter, biotic variables do play a role in species distribution,
but ECNM are forced to operate under the assumption that abiotic variables explain most
of the response. This problem can be overcome by selecting abiotic variables that are
most likely to affect the species physiology and thus most likely to affect species
persistence. In addition, the GCMs produce very coarse-grained environmental data
layers. To make these layers fit the scale of the data required for the research question, a
process of downscaling is required, which is discussed in more detail in Chapter IV. The
more downscaling required, less likely the modeled variables are accurate. In ECNM,
data must be even more closely scrutinized, biases that cannot be accounted for should be
explicitly stated, and processes behind the derivation of variables should be explained.
Data preparation is more complex in ECMN than ENM due to spatiotemporal
considerations; archaeological and fossil occurrence data inherently include more biases
than non-human and modem species occurrence data (Varela, et al. 2011: 452). The first
bias relates to absence data. In ECNM, absence data is almost never available due to the
fact that archaeological data is typically not collected in this manner. This causes spatial
bias within data, with some periods are either over or underrepresented due to
taphonomic processes. These same taphonomic processes that produce spatial bias also
produce temporal bias; the older the remains the fewer that are generally present (Varela
et al. 2011). Bias in the data can also be produced due to collectors bias, such as
differences between survey efforts within and between countries, differences in the
42


amount of attention paid to specific remains and time periods, and difference in
taxonomic identification. The combination of these biases can cause samples to fail to
represent the full population of the past, and so as a consequence the relationships
between the occurrence data and the predictors may offer a flawed picture of species
responses to the environment (Varela et al. 2011: 454). Additionally, dating provides
another source of imprecision within the data set; many occurrences of sites are undated,
and archaeological dating has its own issues and biases (Banks et al. 2006). All of these
biases must be accounted for within the dataset before the data can be applied in an
ECNM model. Often times bias within the sample cannot be fixed, and researchers have
to proceed knowing their model may be inherently flawed due to incomplete knowledge.
Predictive Site Modeling
Predictive site modeling is another method to model the human-environmental
interaction, but this method, unlike ECNM, draws on human behavioral ecology (HBE)
as well as settlement theory as opposed to the concept of niche. Additionally, instead of
modeling species distribution, it is specifically aimed at modeling distribution of sites.
Predictive site modeling and its theoretical background are important concepts to this
study because the method demonstrates that sites themselves can be predicted on the
landscape. This provides the basis for the hypothesis that if the niche of modem day sites
is modeled in ECNM programs, the resulting spatial distribution map will also be a site
distribution map, which will facilitate the identification of productive potential areas of
research.
43


Theoretical Background
Sites pattern against the landscape with respect to environmental and cultural
variables. This patterning can be explained by evolutionary ecology, specifically the
framework of HBE. Evolutionary ecology is defined as the study of adaptive design in
behavior, life history, and morphology (Bird and OConnell 2005: 144). In the
framework of evolutionary biology, behavior is adaptive when it tracks environmental
variability in ways that enhance an individuals inclusive fitness, defined most generally
as its propensity to survive and reproduce (Bird and OConnell 2005: 145) More
specifically, HBE studies the fitness related behavioral trade-offs that humans face in
particular socioecological contexts. These theories of human behavior allow us to relate
cultural diversity of the environment in a consistent theoretical fashion.. .by uniting an
ecological with an evolutionary perspective (Kelly 2007:39). HBE can be used to
hypothesize and to model the conscious and unconscious decisions and tradeoffs humans
make in specific natural and social environments in order to maximize or optimize their
goals. These goals are usually considered to be survivorship or reproductive potential
(Kelly 2007).
As applied to understanding location choices of archaeological sites, HBE
consideres the the decision of where to place a camp as based on tradeoffs between
locations in the natural and social environment in order to place a site where it will be
best promote survivorship or reproductive potential. These decisions involve a complex
analysis of the interplay between many different food resources, as well as nonfood
resources like water, firewood, terrain, other human groups, predation and more (Kelly
2007). In this way, exploring why hunter-gatherers place their sites within a region with
44


respect to environmental variables, through the lens of HBE, can help permit
reconstructions and evaluation of economic activities at a site (Kelly 2007; Shermer and
Tiffany 1985: 228).
As an archaeological method, predictive modeling was first applied in the 1960s
with the rise of the processualist framework; however its ultimate roots lie with the
beginnings of settlement archaeology and in the work done by Gordon Willey (1953) in
the Viru Valley. From this research, Willey concluded that settlement patterns not only
reflect the natural environment, but also the technology of the culture, and the various
institutions of social interaction and control with the culture maintained (Willey 1953:1).
In other words, settlement archaeology has the potential to aid understanding of the
economic, social and political organization of past societies (Trigger 2006: 377). These
concepts of settlement archaeology combined with the processual focus on modeling,
quantitative methods, and importance of environment in site choice (Trigger 2006)
stimulated the first statistical modeling of potential archaeological site location to
unexplored areas (Kvamme 2006; Verhagen and Whitley 2012).
This theoretical perspective in explaining site location choice is supported by
various studies from around the world that conclude that, human economies, settlement
patterns, and social relationships (Wise 2000: 141) are significantly related to
environment and environmental change (Kellogg 1987; Fry et al. 2004; Schermer and
Tiffany 1985; Thomas and Bettinger 1976). In other words, the placement of settlements
can be viewed as a strategy for attaining economic, social and political ends (Jochim
1981: 151), and because of this, site patterns lend themselves to modeling.
Site predictive modeling is based on the environmental characteristics of sites and
45


prior knowledge or hypotheses of where humans liked to place their sites. A
generalization of site location is produced via multivariate statistical methods and then
other geographic locations where this generalization exists are identified. These models
are then able to provide information on interactions between variables, to demonstrate
significance of relationships between variables, to test hypothetical interactions between
variables, and to expand on the strengths of the relationships modeled. Fundamentally, in
this process, sites are assumed to be nonrandomly distributed with respect to the
environment (Kvamme 1992: 23).
Again, these predictive site models, connected with potential distribution models
of Mousterian sites in southern Italy, allow us to consider outputs of the ECNM as
potential site distribution models, not just spatial distribution models of potential niche.
Ultimately, due to ECNM modeling the niche of sites, these concepts are not as far apart
as they may first appear.
A Note on Terminology
Ecocultural niche modeling and spatial distribution models are both widely
and often interchangeably used in the literature; spatial distribution models can be
defined as the projection of the ecological niche onto the landscape (Franklin 2009),
while ENM and ECNM is the process of producing descriptions of the niche. Many
authors refer to ENM as a subset of the former. By using concepts of spatial distribution
modeling, a wider range of questions can be asked using the same processes and
methods, and a visual map can be produced (Guisan and Thuiller 2005). Throughout this
paper, the term ECNM is used to discuss the actual modeling process, as well as to
highlight the theoretical basis of the model. The resulting outputs, or projections, are
46


typically referred to as potential spatial distribution models. However, when the modern
day niche of the site is being modeled, and when the ECNM outputs are being connected
to predictive site modeling, the term potential site distribution model is used.
Extra Theoretical Considerations
All three of these methods for reconstructing the past have been subjected to
various critiques that mostly stem from postprocessual concerns. This section reviews
these critiques and refutes some of the postprocessual claims against using modeling to
understand the past. Inductive versus deductive modeling is discussed, as well as the use
of social versus environmental variables, and finally the benefits of modeling in
archaeological research are reviewed.
Inductive versus Deductive Modeling
ECNM and predictive species modeling are often critiqued for starting with data,
and then applying inductive modeling to build hypotheses, rather than beginning with
archaeological theory, then formulating a hypothesis based on that theory, and finally
using the model to test that hypothesis, or deductive modeling (Kvamme 2006: 13;
Vergen and Whitely 2011). Authors have argued that through deductive modeling, social
variables can be accounted for, such as viewsheds, as well has how environmental
variables like topography and site location interact with the ranges of human-exploited
species to explain what factors drive where humans place their sites on the landscape.
This is considered to be in opposition to inductive modeling which typically focuses on
environmental variables, and only considers culture after the modeling is complete. The
deductive model also draws on ideas of place or a culturally defined local that acts as a
medium for action and is part of human experience and reality (Lock and Harris 2006:
47


43). Thus, the deductive model is explicitly cultural, and tends to focus on regional biotic
variables, which ties it to ideas of Eltonian niche. These similiarities mean deductive
modeling suffers from the same sorts of issues as attempts to model Eltonian niche in
ENM, ECNM and site predictive modeling, including difficulty of producing raster
representations of complex, interconnected biotic variables, compounded with the issues
of modeling cultural variables.
In recent years it has been demonstrated that the extreme dichotomy between
deductive and inductive modeling is mostly a function of how each is defined. In
practice, a combination of deductive and inductive modeling is often applied, and it can
be argued that the decision to model, as well as the selection of environmental variables,
comes from a body of theory (Kvamme 2006). In this specific case, deductive reasoning
provides the underlying framework; the study draws on HBE to explain why sites are
patterned across the landscape, and then uses the theory to formulate a hypothesis that
Mousterian sites in southern Italy should be patterned against specific environmental
variables because Neandertals selected location based on their environmental envelope.
Then, inductive modeling is applied to describe the niche, including which variables have
the most effect on the niche, and to produce a potential site distribution of that niche
projected onto the landscape. In this way, deductive and inductive reasoning are used in a
recursive manner to improve all parts of the modeling process. Finally, deductive and
inductive reasoning can be complementary becuase once important abiotic variables have
been defined through the inductive process, more complex biotic and cultural variables
can begin to be modeled through the deductive process.
48


Social versus Environmental Variables
Although site location decisions involve a combination of social and
environmental considerations, for various reasons, most hunter-gatherer researchers use
environmental variables to explain site location. First, they make the assumption that the
most basic and the most important decisions people made were associated with the
environment (Kolher and Parker 1986: 400). Finally, environmental data, in the form of
maps, both paper and digital, is relatively easy to obtain, while social variables are not.
This focus on environmental variables is the basis for postprocessual critique of
predictive modeling; various critiques have discussed issues in predictive modeling
including environmental determinism, ignoring archaeological concepts of the individual
and of place, the inductive process of the model itself, and the inability of modeling to
accurately account for human behaviors or site formation processes (Wheatly 2000;
Whitely 2004a; Whitely 2004b).
However, in this case the nature and age of the sample makes social modeling
almost impossible. Chronological resolution in terms of dating Middle Paleolithic sites is
not high. Many sites, particularly ones excavated in southern Italy before the
development of radiocarbon dating, are generally only assigned to a Wurm period (Wurm
I: 100-70 kya; II: 70-50 kya; or IIP 50-30 kya), on the basis of zooarchaeological
analysis. Because of this, the modeling in this thesis is focused on trends and patterns in
the record that are consistent through 100,000 years. At scales this large, individual and
group preference about site location becomes less important. Application of social
variables to models of the distant past is also difficult because of the inability to
reconstruct the whole settlement system; without this whole picture, complete hypotheses
about the systemic context of Neandertals and Mousterian sites are difficult. Areas with
49


high vegetation, high depositional or erosional environments, and intensive human
modification, like southern Italy, may never reach a level where a complete systemic
context can be produced. In these regions it is typically the case that existing site records
obtained through haphazard archaeological reconnaissance over the past century must be
employed (Brandt et al. 1992: 207) when doing archaeological research. Because of all
these factors, archaeologists focus on environmental variables when doing ECNM.
The Power of Predictive Models
Despite postprocessual critiques, these modeling techniques can be very
productive. Predictive models are always simplifications of reality because subtle social
determinants of location are probably at work in all settlement systems (Kohler and
Parker 1986: 401) but usually the models are run without consideration for these social
factors. Instead of being a weakness, this simplification of reality through modeling can
be extremely useful to point out what archaeological remains fit, and which do not fit,
with in the expected pattern. These outliers can point researchers in new and fruitful
directions of research, including cultural considerations in an attempt to explain why the
pattern is not consistent between all remains. Modeling also reduces a problem to its key
elements and allows researchers to address and explain those elements one at a time; as
Bird and OConnell (2005) state: one of the most effective ways of eliminating
problematic answers and identifying and pursuing more promising ones (10).
Additionally, it is recognized that modeling was designed to look at large scale human-
environmental interactions which is the purpose of this study. Overall, the constraints of
modeling, including simplification of real life and the marginalization of concepts such
50


as the individual, symbolism and culture, when deployed to answer appropriate research
questions and used with appropriate data can be extremely effective.
Hypotheses
ECNM usually attempts to explain the systemic context as opposed to the
archaeological; this project however, seeks to do both. As discussed in the background
chapter, large areas of southern Italy have been neglected in terms of Paleolithic
archaeological research, for both historical and geographic reasons. Due to these
processes, the picture of Mousterian sites in Calabria, Campania and Bascilicata, and to a
lesser extent in Apulia, are skewed toward coastal caves with deep stratigraphy. It is
because of this incomplete picture of Mousterian sites in southern Italy that inductive
modeling of the archaeological sites can be extremely powerful, both to produce spatial
distribution maps from modeling based on past environmental variables and site
distribution models based on modem day environmental variables.
Goal 1: To determine if modeling the niche of Mousterian sites against modern
variables can produce an effective and useful model of potential site distribution, and if
that model identifies new areas of southern Italy that have the potential to be future sites
of productive research. By drawing on predictive site modeling from other archaeological
contexts and interpreting species distribution models from ECNM as site distribution
models, all presence data can be used in the modeling process without having to consider
issues of correct dating, variable excavation methods and other paleobiogeographical and
archaeological biases. This will allow a reconstruction of Mousterian site distribution
with less error derived from bias in the inputs, as well as allow modeling with a large
sample of sites. In this way, limited resources for survey in a difficult terrain can be
51


applied in the most effective way, and a more complete picture of the archaeological
context of southern Italy will be produced.
Goal 2: To model the systemic context of the glacial and interglacial Mousterian
niche in southern Italy as a heuristic device to explore where the southern Italian
Mousterian niche may have been during warmer or colder periods. In addition, the
modeling will provide a description of the Mousterian niche based on past abiotic
variables. It is hypothesized that comparisons of the past niche to modern will support
accuracy of potential site distribution models produced from modern day variables.
Goal 3: The final aim of this study is to explore the suitability of considering
Neandertal and Mousterian sites as a homogenous unit when doing ECNM. The southern
Italian modeled niche will be compared to other modeled niches of Middle Paleolithic
hominins in Europe (Banks, dErrico, Peterson, Kageyama, et al. 2008) to determine if
modeling at different scales and in different locations produces different results of the
predicted Mousterian niche.
Overall, this study seeks to explore a region of Italy in which Paleolithic
archaeological research has been relatively neglected, but which may hold great potential
for adding to knowledge of Neandertal and Mousterian lifeways in southern Italy.
52


CHAPTER IV
RESEARCH METHODS
The purpose of this research is to explore the potential of ECNM in modeling the
southern Italian Mousterian niche and modeling potential spatial and site distributions in
southern Italy. To this end, methods of this study involved the gathering information and
geographic coordinates of known Mousterian sites from southern Italy, determining
which environmental variables would be most important in modeling the Mousterian
niche, gathering raster data layers representing those variables, and finally calibrating the
algorithm and the computer program to implement the algorithm.
Known Sites in Southern Italy
Data on known sites in southern Italy was collected through an extensive survey
of previously published site reports and literature from the late 1800s through the
present. An article by Milliken (2000), titled The Neanderthals of Italy, provided a
starting list of Mousterian sites for the regions of Apulia, Calabria, Basilicata, and
Campania. As many articles as possible were collected for each site through the Auraria
Campus Interlibrary Loan System, however, the final list of sites used in this research is
slightly different from the list included in Milliken (2000) because some articles were not
available for circulation or were located out of the country. Data on sites was collected
and collated into a Microsoft Access table where each row represents a depositional layer
(Supplementary Materials: Microsoft Access Database), and ultimately, a data set of 114
sites, each with one or more depositional layers, was produced. Subsets of this database
were used in each modeling process (Table 1). Information recorded includes geographic
coordinates, general description of lithic and non-lithic tools, counts of lithic and non-
53


lithic tools, fauna, raw material, symbolism, structures, cultural attribution, site type,
discovery, excavation and publication dates, and finally a list of all sources referencing
the particular site. The full table and references can be found in the supplementary
materials.
Geographic coordinate information was recorded from the articles themselves
when possible. Additionally, official caving websites, (Federazione Speleogica Puglia
and Federazione Speleogica Campana) provided coordinates for the location of caves
throughout Apulia and Campania. Google maps provided the location for well-known
sites at points of interest. An archaeological database associated with The Stage 3 Project
(Davies 1996-2015) also provided geographic coordinates for some sites. Finally
descriptions within the texts themselves often times allowed estimation of the placement
of sites on the landscape. All x,y data was collected in the WGS 84 coordinate system
when possible, and converted if not. Cave site locations are more secure and precise, and
most, especially in Apulia, have error ranges of less than one-kilometer, while most of
the open air sites were lacking specific coordinate data, and therefore coordinates are less
precise, and error in site location can range up to four-kilometers (see Table 1 for source,
accuracy, geographic coordinates and modeling subsets).
Known biases within this data set include a focus on stratified cave sites, due to
historical trends in Italian archaeological research (Chapter II). This could lead to the
fundamental niche being under-predicted in environmental space because of the range of
sites is not wide enough; because of this, known presence points were compared with the
predicted niche to confirm that the model is successfully extrapolating to new areas (see
Chapter VI for discussion). Additionally, to account for the variation of accuracy in
54


known site location the model was produced at a one-kilometer and a four-kilometer
resolution with modem variables, and at a four-kilometer resolution for glacial and
interglacial models.
Tablel: Known Sites in Southern Italy. Geographic coordinates of each site, subsets of
sites used to produce each model, including number of layers if applicable, source of
geographic coordinates, region the site is from and site type. Supplemental materials
abbreviated as Supp. Mat. Continued on next page.
Site Longitude Latitude modern (4km) modern (1km) glacial interglacial Source Confidence Region Site Type
1 Grotta di CavalJo 17.981 40.155 X X X(3 layers) X (8 layers) Stage 3 high Apulia Cave
2 Grotta di Lardoni 17.0934 41.0154 X X X (1 layer) fspulia.com high Apulia Cave
3 Grotta di Giganti 18.335 39.808 X X (1 layer) lspulia.com high /Apulia Cave
4 La rala dell'Eiephante 18.335 39.808 X fspulia.com high Apulia Cave
5 Grotta delle Tre Porte 18.335 39.808 X X(1 layer) lspulia.com high Apulia Cave
6 Grotta Titti 18.339 39.807 X X (1 layer) fspulia.com low Apulia Cave
7 Grotta del Bambino 18.335 39.808 X X (1 layer) fspulia.com high Apulia Cave
8 Grotta Romanclli 18.4334 40.0161 X X X (2 layers) X (2 layers) stage 3 high Apulia Cave
9 Grotta Zinzulusa 18.4316 40.0121 X X X (1 layer) googlcmaps high Apulia Cave
10 Grotta (Torre) dcllAlto 17.979 40.144 X X X(8 layers) articles (see supp. mat.) high Apulia Cave
11 Grotta (Torre) dell Uluzzo 17.981 40.159 X X articles (see supp. mat.) high Apulia Case
12 Grotta Mario Bernardini 17.953 40.176 X X X(l layer) fspulia.com high /Apulia Cave
13 Grotta di Serra Ckora 17.936 40.203 X X X (3 layers) googlcmaps high Apulia Cave
14 Grotta di Santa Croce 16.4691 41.177 X X X (1 layer) fspulia.com low Apulia Cave
1$ Grotta della Mura 17.3072 40.9473 X X X (1 layer) fspulia.com low Apulia Cave
16 Grotta (Riparo) Marcello Zei 18.019 39.995 X X |5 layers) articles (see supp. mat.) high Apulia Cave
17 Grotta Montani 18.288 39.85 X X (6 layers) articles (see supp. mat.} high Apulia Cave
18 Grotta di Capelvcncre 17.98 40.142 X X X (11 layers) lspulia.com high Apulia Cave
19 Grotta B di Spagnoli 17.1482 40.8754 X X X (3 layers) lspulia.com high Apulia Cave
20 Grotta dell'Angelo 17.9347 40.4198 X X googlcmaps high Apulia Cave
21 Grotta del Vagno 16.4007 41.0531 X X fspuiia.com high /Apulia Cave
22 Grotta di Orimini 17.2228 40.7017 X X X (1 layer) lspulia.com high Apulia Cave
23 Grotta di liuzzo C 17.934 40.18 X X X (2 layers) lspulia.com high /Apulia Cave
24 Torre delTAlloTIuzzo 17.98 40.14 X X fspulia.com high Apulia Cave
2$ Grotta Grande di Cinlo 18.3851 39.8378 X X XII layer) fspulia.com high Apulia Cave
26 Grotta dell Vencri 18.1144 40.0703 X X X(l layer) fspulia.com high Apulia Cave
27 Grotta Santa Maria di Agnano 17.584 40.727 X X X (1 layer) stage 3 high Apulia Cave
28 Grotta l.amalunga 16.5875 40.8719 X X fspulia.com high Apulia Cave
29 Grotta San Pietro 16,7936 40.6225 X X fsputia.com high Apulia Cave
30 Grotta Paglicci 15.6264 41.6671 X X X (2 layers) stage 3 high Apulia Cave
31 Grotta di Torre Moscia 17.3161 40.8232 X X lspulia.com high Apulia Cave
32 Grotta di Striarc 18.4357 40.0247 X X X(1 layer) fspulta.com high Apulia Cave
33 Grotta di San Giacinto 17.0835 40.9701 X X fspuiia.com high Apulia Case
34 Grotta Masseria del Monte 17.144 41.001 X X X (1 layer) fspulia.com high Apulia Cave
35 Oscurusciuto 16.76 40.58 X X X (11 layers) articles (sec supp. mat.) medium Apuha Cave
36 Caverna di Porto Rosso 17.3077 40.947 X X (1 layer) lspulia.com low Apulia Cave
37 Grotta di Cala Corvine 17.2647 40.9735 X fspulia.com low Apulia Cave
38 Grotta di Cala Comincia 17.32 40.94 X fspulia.com low Apulia Cave
39 FondoCattie 18.31 40.13 X X X(l layer) articles (see supp. mat.) tried Apulia Cave
40 Santa Caterina 17.989 40.141 X X googlcmaps high Apulia Open
41 LaGattarela 16.179 41.836 X articles (sec supp. mat.) high Apulia Open
42 Forcsta Umbra 16 41.81 X X googlcmaps med Apulia Open
43 Lesina 15.35 41.86 X X articles (see supp. mat.) med Apulia Open
44 Proloappcnninico di Spigolizzi 18.23 39.86 X X articles (sec supp. mat.) med Apulia Open
45 Sorgcnti di Irchio 15.8102 41.8727 X X X11 layer) X (1 layer) googlcmaps med Apulia Open
46 Ingarano 15.5068 41.7836 X X X (1 layer) googlcmaps med Apulia Open
47 Melendungo 18.34 4027 X X articles (see supp. mat.) med Apulia Open
48 Serra di S. Eleutcrio 18.15 40.06 X articles (see supp. mat.) low Apulia Open
49 Sannicandro Garganico 15.57 41.83 X X X (1 layer) articles (see supp. mat.) med Apulia Open
50 Conlrado S. Michele 15.57 41.83 X X articles (see supp. mat.) med Apulia Open
55


Table 1 Cont.
Site Longitude latitude modern (4km) modern (1km) glacial interglacial Source Confidence Region Site Type
51 ContradaCardinale 15.57 41.83 X X articles (see supp. mat.) med Apulia Open
52 Pianidi S. Vito 15.89 41.71 X X X (1 layer) articles (see supp. mat.) high Apulia Open
53 Sant'Andrea 18.42 40.29 X articles (see supp. mat.) low Apulia Open
54 Monticello 17.38 40.69 |X X articles (see supp. mat.) med Apulia Open
55 Marina Franca 17.38 40.69 X X articles (see supp. mat.) high Apulia Open
56 Torre Bianca 17.45 40.33 X articles (see supp. mat.) low Apulia Open
57 Torre Canne 17.46 40.833 X articles (sec supp. mat.) low Apulia Open
58 Gallieo 11 17.87 40.69 X articles (see supp. mat.) low Apulia Open
59 Papalucio 17.64 40.5 X articles (see supp. mat.} med Apulia Open
60 San Ermete 18.31 40.l5jx articles (see supp. mat.) low Apulia Open
61 San Giovanni Lo Pariete 17.6646 40.5004 X X archeocluboria.com high Apulia Open
62 Pappada' 17.67 40.5 X articles (sec supp. mat.) low Apulia Open
63 Monti 17.66 40.51 X articles (see supp. mat.) low Apulia Open
64 Monte Saot'Angelo 15.94 41.71 X X googlcmaps med Apulia Open
65 Papalucio 17.65 40.45 X X googlcmaps med Apulia Open
66 Torre Testa 17.87 40.69 X googlemaps med Apulia Open
67 Basin of Sant'F.gidio 15.7709 41.7268 X articles (see supp. mat.) low Apulia Opot
68 FUehino 15.33 41.9 X articles (see supp. mat.) low Apulia Open
69 Camerata 15.32 41.87 X articles (sec supp. mat.) low Apulia Open
70 Maccbia a Marc 16 41.95 X articles (see supp. mat.) low Apulia Open
71 Masseria Malepezza 18.42 40.25 X articles (see supp. mat.) low Apulia Open
72 Donna Lucrezia 17.5 40.61 X X articles (see supp. mat.) med Apulia Opai
73 Montagnulo 17.51 40.62 X X googlcmaps med Apulia Open
74 Masseria Le Fiattc 17.59 40.37 X X googlcmaps med Apulia Open
75 Rocavccchia 18.42 40-29 X X articles (sec supp. mat.) med Apulia Open
76 Atella Basin 15.66 40.9 X articles (sec supp. mat.) low Basilicata Open
77 Grotta di Torre Nave 15.77 39.93 X X X(1 layer) articles (see supp. mat.) med Calabria Cave
78 Scoglio di San Giovanni 15.774 39.834 X X articles (see supp. mat.) med Calabria Cave
79 Riparo Torre Talao 15.79 39.81 X X googlcmaps High Calabria Cave
80 Grotta Fiumkello 15.6981 39.9985 X X X(1 layer) articles (see supp. mat.) low Calabria Cave
81 Grotta Cicchetti 16.62 40.61 }X articles (see supp. mat.) low Calabria Cave
82 Grotta Pipislrelli 16.63 40.61 X X X(1 layer) articles (see supp. mat.) med Calabria Cave
83 Grotta Cuparo 14.34 40.61 X X articles (see supp. mat.) med Campania Cave
84 Grotta lertanto 14.34 40.61 X X articles (see supp. mat.) med Campania Cave
B5 Grotta dcllo Scoglione 14.345 40.585 X X articles (see supp. mat.) med Campania Cave
86 Grotta Taddeo 15.394 39.998 X X <1 layer) googlcmaps low Campania Cave
87 Grotta Tina 15.394 40.001 [X X(1 layer) articles (see supp. mat.) low Campania Cave
88 Riparo Silhar 15.394 39.998 X articles (sec supp. mat.) low Campania Cave
89 Grotta della Cala 15.394 40 X X (4 layers) articles (see supp. mat.) low Campania Cave
90 Riparo del Poggio 15.395 40 X X(1 layer) X (4 layers) fscampania.com low Campania Cave
91 Grotta del Poggio 15.395 40 X X(1 layer) X (2 layers) fscampania.com low Campania Cave
92 Grotta di Porto Infmchi 15.34 40.008 X fscampania.com low Campania Cave
93 Grotta di Santa Maria 15.4242 39.9941 X X X (1 layer) fscampania.com high Campania Cave
94 Riparo Difesa 15.394 39.998 X articles (see supp. mat.) low Campania Cave
95 Grotta dell'Acqua 15.4249 39.9987 X X fscampania.com high Campania Cave
96 Grotta Massetta 15.42 40 X X articles (see supp. mat.) med Campania Cave
97 Grotta Grande di Scario 15.476 40.066 X X X(1 layer) X (1 layer) fscampania.com high Campania Cave
98 Riparo del Molare 15.49 40.065 X X(1 layer) X (2 layers) fscampania.com low Campania Cave
99 Grotta Visco 15.3004 40.0304 X X fscampania.com high Campania Cave
100 Grotta dei Ciavolc 15.2918 40,0239 X X fscampania.com high Campania Cave
101 Grotta di Castclcivita 15.2094 40.4957 X X X (2 layers) fscampania.com high Campania Cave
102 PalinuroCapriozzi 15.29 40.04 X googlcmaps low Campania Open
103 Cala Bianca 15.4135 39.9971 X X googlemaps high Campania Open
104 Pacstum 15.01 40.42 X X googlcmaps med Campania Opai
105 Serino 14.8606 40.857S|x X articles (see supp. mat.) high Campania Open
106 Montemiletto 14.89 41.02 X X googlcmaps med Campania Open
107 Colic Pietra 15.42 40 X 1 articles (sec supp. mat.) low Campania Open
108 Val Pescara , 15.42 40 X articles (see supp. mat.) low Campania Open
109 Rio Maggio 15.42 40 X articles (see supp. mat.) low Campania Open
110 Grotta del Noglio 15.4208 39.9929 X articles (sec supp. mat.) low Campania Open
111 San Francesod'Archi 15.67 38.13 X 1 X(1 layer) articles (see supp. mat.) low Campania Open
112 CapeCalonne 17.19 39.03 X articles (see supp. mat.) low Campania Open
113 TomDino 15.78 39.84 X X(1 layer) articles (see supp. mat.) low Campania Open
114 Contrada ianni 16.02 38.58 X articles (see supp. mat.) low Campania Open
56


Environmental Data Layers
Environmental data layers were collected in raster format from various open
sources at both one-kilometer resolution, or 30 arc seconds, and four-kilometer
resolution, or 2.5 arc minutes. A raster file is composed of cells and each cell is given a
value, which holds the data that the particular raster file represents. The resolution of the
raster is based on the size of the cell; a 30 arc second raster means that each cell
represents one-kilometer of land, while a 2.5 arc minute raster means that each cell
represents approximately four-kilometers of land. Smaller resolved rasters are more
accurate to real life as they make less generalization about the data they are representing,
but depending on how the raster is being used larger cell rasters are often required.
Therefore, because ECNM modeling must be done at the scale of the presence data, or
the modeling runs the risk of predicting the Mousterian niche into unoccupied areas, both
of these resolutions were necessary to model the set of presence points with a smaller
error range and those with a larger
All processing of raster layers was done in arcGIS 10.1 (ESRI), and all layers
were converted to a raster format if necessary, clipped to the same extent, projected to the
same coordinate system, and processed to the same cell size, all of which is required by
the modeling program. Any resampling or reprojecting of raster data sets for continuous
data used the cubic convolution method, which determines the new value of a cell based
on the fitting of a smooth curve through the values 16 nearest input cells. The nearest
neighbor method, which reprojects rasters based on the nearest neighbor, was used for
categorical data.
57


Most data was downloaded in the World Geodetic System (WGS) 84 geographic
coordinate system, and data that could not be downloaded in this coordinate system were
converted to WGS 84. All the data layers were projected into a custom projection, Italy
Conformal Conic, created in arcGIS using guidelines outlined in Price (2012). This
projection is based on the Lambert Conformal Conic projection, which preserves area,
shape, and distance, and is used for regional mapping of smaller countries in middle
latitudes. The datum and geographic coordinate system of the Lambert Conformal Conic
projection is WGS 84, which is ideal since most of the layers did not have be projected
into a new coordinate system, which can significantly decrease accuracy. A custom
projection was created because Southern Italy does not fit easily into most commonly
used projections; the region lies within three UTM zones, and the Mario Monte
projections divide the country into east and west sections rather than north and south.
With this custom projection, there is no distortion of coordinate values, and slope and
other raster values could be correctly produced.
A suite of environmental variables which produce patterning in archaeological
sites by limiting the environment in which humans can survive were included in the
modeling process, with a focus on limiting factors (Table 2). These limiting factors are
defined as factors which control species eco-physiology, like temperature, water, and soil
composition, as opposed to disturbances, all types of perturbations affecting
environmental systems, natural or human induced, or resources which are defined as all
compounds that can be assimilated by organisms (Guisan and Thuiller 2005: 994). These
abiotic layers were also chosen to facilitate the modeling of the fundamental niche in the
Grinellian sense. Additionally a literature review of previous ECNM demonstrates that
58


these types variables are most commonly deployed when modeling the niche of human
species (Chapter II). Previous studies have identified some layers that seem to have a
large effect on human niches, including elevation, access to water, temperature and
rainfall. This study made sure to include these types of layers, with the addition of other
layers that describe modem environmental conditions, which may contribute to where
Mousterian sites are found on the southern Italian landscape today (Banks et al. 2011;
Brandt et al. 1992; Kvamme 1992; Wise 2000). Finally, due to limitations of past
environmental reconstructions, and the difficulty of producing such models, only a subset
of the layers discussed below were used in modeling the glacial and interglacial
Mousterian niche.
59


Table 2: Environmental Layers. List of all environmental layers used in modeling, their
source organization, anc website from which the layer was obtained.
Data Layer Source Website
Elevation Bio clim http://www.worldclim.org
Aspect Bio clim http://www.worldclim.org
Slope Bio clim http://www.worldclim.org
Flow Accumulation Bio clim http://www.worldclim.org
Flow Direction Bio clim http://www.worldclim.org
Distance from Populated Places Geo comm http://data.geocomm.com/catalog/IT/index.html
Distance from Water Mapcruzin http://www.mapcruzin.com/download-free-arcgis-shapefiles.htm
Geologic Age HYDRO IK USGS https: //lta. cr .u sgs. gov/H YDRO1K
Land cover European Environmental Agency http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-3
Soils European Soil Database http://eusoils.jrc.ec.europa.eu/ESDB Archive/ESDB/Index.htm
Annual Mean T emp Bio clim http://www.worldclim.org
Mean Diurnal Range Bio clim http://www.worldclim.org
Isothermality: modem Bio clim http://www.worldclim.org
Temp Seasonality Bio clim http://www.worldclim.org
Max Temp of Wamiest Month Bio clim http://www.worldclim.org
Min Temp of Coldest Month Bio clim http://www.worldclim.org
Temp Annual Range Bio clim http://www.worldclim.org
Mean Temp of Wettest Quarter Bio clim http://www.worldclim.org
Mean Temp of Driest Quarter Bio clim http://www.worldclim.org
Mean Temp of Warmest Quarter Bio clim http://www.worldclim.org
Mean Tem of Coldest Quarter Bio clim http://www.worldclim.org
Annual Precipitation Bio clim http://www.worldclim.org
Precipitation of Wettest Month Bio clim http://www.worldclim.org
Precipitation of Driest Month Bio clim http://www.worldclim.org
Precipitation Seasonality Bio clim http://www.worldclim.org
Precipitation of Wettest Quarter Bio clim http://www.worldclim.org
Precipitation of Driest Quarter Bio clim http://www.worldclim.org
Precipitation of Warmest Quarter Bio clim http://www.worldclim.org
Precipitation of Coldest Quarter Bio clim http://www.worldclim.org
Modern Environmental Data and Mousterian Site Location
To investigate Goal 1, as discussed in the Hypotheses section (Chapter III), this
study applies modern environmental data to model the present day niche of Mousterian
sites in southern Italy. This is in contrast to previous application of the ECNM methods
which investigate the niche of humans through environmental resconstructions of the
Pleisotcene (Banks et al. 2006; Banks et al. 2011; Banks, dErrico, Peterson, and
Kageyama et al. 2008; Beeton et al. 2013). Through this application of modern
environmental data to model the modern niche of Mousterian sites, it is hoped that the
60


modeling process will produce a theoretical niche of modem day site location, as well as
a site distribution map of where Mousterian sites are expected to be located in Southern
Italy.
Modeling with modern variables also avoids potential error and bias in ECNM in
two ways. The first is a removal of error associated with using reconstructed
environmental data layers. Past variables are generated through GCMs, and there are
various circulation models, all of which produce slightly different outputs as a coarse
resolution (200-100 km). Additionally, each downscaling method to convert the outputs
of the GCMs from 200-100 km resolution to 1km or 4km resolution is different, and has
the potential to produce multiple outputs from the same inputs. GCMs also require
modeling of the atmosphere of large regions in the past to produce estimations of
environmental variables. Modeling the atmosphere requires considerations of complex
interactions between wind, heat transfer, radiation, relative humidity surface hydrology,
levels of carbon dioxide and many more factors, all of which must be reconstructed to the
past. While techniques have been successful, they still likely contain sources of error,
particularly when they are downscaled. Modem day environmental layers do have their
own accuracy problems, especially at larger raster resolutions, however they avoid many
of the assumptions which researchers are forced to make to model past environment.
Therefore, using modem variables, and interpreting the output as where sites may be
found on the modern day landscape as opposed to the niche of Neandertals based on
Mousterian sites, a source of error in ECNM can be avoided.
Also, by using modem day variables, the whole suite of southern Italian
Mousterian sites can be modeled together, removing some bias in occurrence data while
61


preserving sample size. Previous studies, (Banks et al. 2006; Banks et al. 2011; Banks
dErrico, Peterson, and Kageyama et al. 2008; Beeton et al. 2013) have modeled small
slices of time to look at the niche of a particular culture. Therefore, they rely on absolute
dating methods to confirm that each site included in the modeling process is actually
within the time period they are studying. Because of the difficulties associated with
dating many of these Pleistocene sites, sample sizes used in ECNM usually end up
smaller than samples used in ENM. Collectors bias must also be considered as sites
which have absolute dates tend to be cave sites. While modeling the concept of the niche
allows ECNM to avoid only modeling the location of potential cave sites, it can still
produce a bias in the modeled niche. By using modern variables, and explicitly modeling
potential locations where sites may be found on the landscape in the present day, this
study is able to use all the present points in southern Italy, which increases sample size,
and removes some bias by including open air sites in the dataset.
Elevation, Slope, and Aspect
Elevation has been demonstrated to be an important factor that can limit
archaeological site location (Kvamme 1992; Wise 2000), and most ECNM and ENM
takes this variable into account. Closely associated with elevation is slope, also an
important variable in site location since hunter-gatherers typically prefer to camp in areas
with low slope values (Kvamme 1992:25). Slope also increases erosion, so sites on slopes
have a larger chance of eroding away. Aspect plays a potential role in determining site
location, as various aspects can help control temperature at a site (Kvamme 1992: 26).
Slope at a one-kilometer resolution was downloaded from Data Downloads
section of the USGS HydroSHEDS website (HydroSHEDS 2010). This data set was
62


produced from elevation data collected by a space shuttle flight for NASAs Shuttle
Radar Topography Mission, and has gone through automated processing to make the
rasters more effective at modeling hydrology by the Conservation Society Program of the
World Wildlife Fund (HydroSHEDS 2010). At this resolution, data must be downloaded
by tile to reduce file size, and so tiles n35e015_dem_bil.zip, n40e010_dem_bil.zip, and
n40e015_dem_bil.zip were downloaded from the Europe, Southwest Asia section of the
website. These tiles were then merged into one raster in arcGIS using the mosaic to new
raster tool (Data Management tools > Raster > Raster Dataset > Mosaic to New Raster).
Then, using the Project Raster Tool (Data Management > Projections and
Transformations > Raster > Project Raster), the raster was projected into Italy Conformal
Conic, using cubic convolution resampling. Finally, the Extract by Mask Tool (Spatial
analyst > Extract > Extract by Mask) was used to constrain the extent of the raster to
southern Italy.
From this elevation raster, slope in degrees, or rise over run, was produced, in
addition to an aspect data layer which is essentially slope direction, and describes the
direction of maximum rate of change in elevations between each cell and its eight
neighbors. The Slope and Aspect tools (Spatial Analyst > Surface) were used to produce
each respective layer.
Elevation at a four-kilometer resolution was downloaded from the BioClim
website (BioClim). This raster layer was also derived from the Shuttle Radar Topography
Mission, and the same processes and tools, without the Mosaic Raster Tool since this data
did not need to be downloaded in tiles, were used to reduce the extent of the elevation
63


layer to southern Italy as well as to project the data. Again, from this dataset, slope and
aspect were produced through Spatial Analyst tools.
Flow Accumulation and Flow Direction
Water is another important environmental variable in determining where hunter-
gatherers placed their sites, and previous studies have included these variables (Banks,
dErrico, Peterson, Kageyama et al. 2008; Banks, dErrico, Peterson, Vanhaeren et al.
2008; Brandt, et al. 1992; Kvamme 1992; Wise 2000: 26). Therefore, flow accumulation
and flow direction layers were used to model how the Mousterian sites relate to water in
the study area.
Flow direction and accumulation at a one-kilometer resolution were also available
for download from the USGS Hydrosheds website (HydroSHEDS 2010). Each layer was
clipped to the extent of the elevation layer using the extract by mask tool, as well as
projected into the Italy Conformal Conic projection.
The flow direction layer at the four-kilometer resolution was produced in arcGIS
from the elevation raster layer using the Flow Direction tool (Spatial analyst >
Hydrology) and then by using a combination of the newly produced flow direction raster
layer and the elevation layer, a flow accumulation layer was produced using the Flow
Accumulation Tool (Spatial Analyst > Hydrology).
Distance from Populated Places and Distance from Major Waterways
Both of these layers are of interest when modeling and predicting the distribution
of Mousterian sites in modern environments as opposed to past environments, and so
were not included in the glacial and interglacial niche modeling process, but were
included in the modem model. Distance to populated places allows consideration of
64


modern day modification on the landscape to be included in the modeling process. This
layer was specifically included to explore the role of Mousterian site locations to areas of
dense modern human occupation, or in other words, if sites were more likely to be found
near or further away from cities and densely populated towns. The Distance from major
waterways layer is important in terms of erosion and deposition processes that could
prevent sites from being found in modern day contexts (Banks, et al. 2011; Brandt, et al.
1992: 273; Kvamme 1992).
The populated places network was downloaded in vector format from
GeoCommunity (GIS Data Depot 1995-2014); it is available for download on the Italy
nationwide data administration and political boundaries page of the website. To remove
the populated places polygons, from the network of other vector files, the layer of interest
was first saved as a shapefile. The major waterways vector polyline data set was
downloaded from MapCruzin (Free GIS Shapefiles, Software, Resources and Geography
Maps 1996-2015), which provides free GIS shapefiles. This specific layer was produced
from open street map, and can be found in the Italy category in the Free Shapefiles page
of the website.
Both the distance from populated places and the distance from major waterways
layers were created with the Euclidean Distance tool (Spatial Analysis > Surface tools).
This tool produces a raster from a vector layer by calculating, for each cell, the Euclidean
distance to the source, in this case either the populated place or the river depending on
the shapefile layer. While each raster was created, the cell size was set to the same
resolution as the four-kilometer or one-kilometer elevation layer, and after creation each
raster layer was clipped to the extent of the elevation raster.
65


Geologic Age
Geologic age of the underlying bedrock is important to determine where cave
sites might be found, and is also important as a general environmental description that
could constrain location of other sites. Values of geologic age are generalizations of the
original UNESCO age classes by the U.S. Geoglogical Survey (Appendix A).
This raster is derived from a vector layer of geological age downloaded from the
HYDRO IK dataset, which can be found at the USGS Earth Explorer interface under
digital elevation (HYDROIK 2012). It is part of a raster and vector data set that was
developed at the United States Geological Surveys Earth Resources Observation and
Science (EROS) Center, and is available from the HYDRO IK dataset. To process this
layer, it was first converted into a raster using the Polygon to Raster tool (Conversion
tools > To Raster). After conversion, the extract by mask tool was used to convert the
raster to the correct cell size, correct extent and correct projection, all based on the
elevation layer both of the one and the four-kilometer resolution.
Land Cover
The land cover raster layer was obtained to test if modern day land cover
contributes to the location of present day sites on the landscape. The layer was developed
by the Corine Program of the European Environment Agency and processed by the
European Topic Centre on Land Use and Spatial Information at a one-kilometer
resolution (CORINE Land Cover). The data was downloaded in the EPSG: 3035
coordinate system, and converted into the custom projection using the Project Raster tool.
Since this raster was downloaded at the one-kilometer resolution to begin with, it only
required a clip to the correct extent for use in the one-kilometer modeling process.
66


Using the extract by mask tool and basing all changes on the four-kilometer
elevation layer dataset, the raster was also clipped and the cells were resized to the
correct resolution. Because the resolution of the layer went from one-kilometer to four-
kilometers, the details and the accuracy were decreased due to the inherent nature of
raster data sets. Additionally, and unavoidably, the coastline details were lost in this
transformation, meaning that many presence points were lacking environmental
information for this layer. If sites lacking land cover data could be moved and still remain
within the four-kilometer error range and resolution, then they were moved; however,
many could not be moved. Therefore, models that were run with this environmental layer
had a smaller sample size because the modeling process leaves out sites that have null
data values.
Soils
This raster layer comes from the European Soil Database. The data at one-
kilometer resolution is available without charge for noncommercial use, in the ETRS:
3035 projection (European Soil Database 1996-2014; Liederkerke et al. 2006; Pangos
2012). Values and soil types are derived from the World Reference Base for Soil
Resources, which is the current international standard taxonomic soil classification
system. As this data set was downloaded at one-kilometer resolution, this layer was
processed to the correct projection and extent using the extract by mask tool for the one-
kilometer modeling.
For the four-kilometer resolution, the data layer was converted to the correct
projection, cell size and extent using the extract by mask tool. Again, like the land cover
layer, this raster faced unavoidable issues with a decrease in coastal resolution and a
67


considerable number of sites that could not be moved to the raster layer within the four-
kilometer error range; therefore, like the land cover layer, models created with this layer
use a smaller subset of presence points.
Modern Climate
Climate variables were downloaded from the BioClim website (BioClim). They
consist of a suite of 19 climatic variables specifically produced to provide biologically
meaningful data for ENM, and are derived from monthly temperature and rainfall data.
The bioclimatic variables represent annual trends (e.g., mean annual
temperature, annual precipitation) seasonality (e.g., annual range in temperature
and precipitation) and extreme or limiting environmental factors (e.g.,
temperature of the coldest and warmest month, and precipitation of the wet and
dry quarters) (BioClim).
These climate layers include data compiled by the Global Historical Climatology
Network (GHCN), the Food and Agriculture Organization of the United Nations (FAO),
the World Meteorological Organization (WMO), the International Center for Tropical
Agriculture (CIAT) and various other regional databases. See Table 2 for a complete list
of variables within this dataset.
This data was available at both a four-kilometer and one-kilometer resolution
(Hijman et al. 2005). After downloading they were each projected to the custom
projection and clipped to the extent of the study area using the elevation layer.
Past Climate
This same suite of biologically meaningful climatic data has been produced for
various times in the past, also available on the BioClim website. Of interest to this study
are the models for the Last Interglacial (LIG) Period (140,000-120,000 years ago), and
the Last Glacial Maximum (LGM) (-22,000 years ago). These past climate models are
68


downscaled climate data from simulations derived from Global Climate Models (CGMs),
and were produced through the Coupled Model Intercomparison Project Phase 5
(CMIP5). Generally, GCMs simulate weather in the past through the assumed
atmospheric concentrations of greenhouse gasses; unfortunately, due to the complexity of
modeling and limited computer memory, most modeling takes place at approximately 2
or 3 degree raster grids (200-100km). Therefore for modeling at smaller scales the
information must be downscaled. There are also a variety of downscaling models;
however all of them combine modern climatic data with the CGM outputs to create more
fine-grained layers. (For more detailed information on methods see Hijmans et al. 2005).
Because not all presence points also had climatic information, the Mousterian
niche in past climates was modeled at a four-kilometer resolution so that sites which are
less securely located on the landscape could be included in the modeling process. This
was necessary to preserve an appropriately large sample size (Papes and Gaubert 2007,
Pearson et al. 2007). Therefore, the Mousterian niche in past climate was only modeled at
four-kilometer resolution. All 19 bioclimatic layers representing the LIG were produced
by Otto-Bliesner et al. (2008) at a one-kilometer resolution and are available for
download from the BioClim website. Once they were downloaded, they were processed
in arcGIS to the correct projection, cell size and extent. In the case of the LGM, the
BioClim variables were available at a four-kilometer resolution. Once they were
downloaded, they were processed to the correct projection and extent with the same
methods as the other layers
69


Software: GARP with Best Subsets
As discussed in Chapter II, ECNM uses presence data, environmental layers, and
a modeling algorithm to produce a site distribution or a spatial distribution map. As
visually represented by Figure 13, ENM combines presence data and environmental data,
and then using the environmental values at each of the known sites, the algorithm
generalizes the niche of the species being modeled. This produces a theoretical niche in
environmental space, and finally, the niche is projected into geographic space, producing
a raster output where areas with a high likelihood of allowing the modeled species to
survive are indicated in red, while areas in which species presence is unlikely are
modeled in blue. In other words, the model identifies the range of environmental
variables in which a species can survive, the niche, and then identifies other locations on
the landscape which are within this range, producing a geographic projection of the past
niche or site location.
For the algorithm, this project uses GARP (Stockwell and Peters 1999) to model
the Mousterian niche. Additionally, this study uses the best subset protocol for GARP,
developed and explained by Anderson et al. (2003) as a method to select the best
algorithm outputs. All algorithms and models were implemented in openModeller
software for ENM.
70



Figure 13: Ecocultural Niche Modeling. (1) species occurrence points, (2) environmental
layers, (3) modelling algorithm, (4) model in the environmental space, (5) model
projection, with red indicating higher suitability values and blue lower suitability. From
openmodeller.sourceforge.net.
When producing species distribution model outputs, GARP develops a description
of the potential niche in the form of a set of rules based on biological and geographic
variables that describe the potential geographic distribution of a given species. This set of
rules describing site locations is developed through a process of evolutionary
refinement. In this process, combinations of environmental variables are considered like
combinations of genes; the ones that are invalid or deleterious, by producing habitats
outside of the Mousterian range, are removed from the modeling (Stockwell and Peters
1999; Walton 2009), just as deleterious genetic combinations are removed from
populations through natural selection.
Before the modeling begins, the presence points are divided in half (GARP
Desktop Users Manual); the first set is used as training data and the second set is used
as test data. With data that lack absence points, such as this data set, GARP produces
71


pseduoabsence data by taking a random sample of background points in the modeling
space, which then become part of the training data set. GARP takes the environmental
values at each training dataset point, as well as the values at nearby cells, and based on
these values generates one of four different types of statistical rules. The rules are atomic,
range, negated range or logit types, and use an if.. .then format. Atomic rules are the
most simple, as they use a single environmental value:
if elevation =10, then Mousterian habitat = present.
Range rules are very similar to atomic, however they deploy a range of data values to
make statements such as:
if elevation = 10-20, then Mousterian habitat may be present.
Negated range rules use the same principle as range rules, however they use a range of
variables that constrain Mousterian habitat to exclude areas from being considered part of
the habitat;
if elevation > 20, then Mousterian habitat is not present.
Finally, logit rules, which are an adaptation of logistic regression models, can also be
applied. This type of rule production is applied to determine the probability of
Mousterian habitat being present at a particular point (Bergen et al. 2007; Stockwell and
Peters 1999; Walton 2009).
A first population of rules is produced based on the training dataset. Then, genetic
concepts, like crossover, mutation and selection are applied to this population of rules to
produce new generations of rules, or iterations of the model. These new rules are then
tested against the training data set. Crossover, where two rules exchange values or range
of variables, and mutation, where a value is randomly changed to a new value, both
72


introduce variation into the population of rules. Then, selection is used to decide which
rules are most predictive. If a new variant of the rule is highly predictive, the algorithm
continues to modify it until it is as predictive as possible. New iterations of the model,
alternatively thought of as generations of rules, continue to be produced until a fixed
number of iterations is reached, in this case 1000, or until changes in the new iterations
fall below a set percentage, in this case .01% (Banks et al. 2011; Walton 2009; Anderson
et al. 2003). Due to the elements of randomness in the modeling algorithm, even with the
exact same inputs no two models will ever be exactly the same.
Finally, the population of rules, which describes the niche of the species, is
extrinsically verified against the test data to avoid overfitting and to check for accuracy,
and then areas of similar habitat in the modeling space are identified. Overfitting is
defined as when a model too closely follows the data at the expense of being able to
extrapolate and identify areas of high probability of Mousterian presence in under-
sampled regions. Finally, a raster is produced which uses a color ramp to demonstrate
areas with a high probability inclusion in the Mousterian niche, as well as those areas
with a lower probability of inclusion. A good model is able to both correctly classify
known presence points as present, and is able to identify other areas in the study region
that are included in the potential niche (Desktop GARP Users Manual; Stockwell and
Peters 1999; Pearson 2010; Walton 2009).
Results of these extrinsic tests, are used to describe a models predictive potential
as well as to compare it to other models, and are reported as receiver operating
characteristic (ROC) graphs, AUC values and omission and commission error values.
ROC graphs illustrate the performance of a binary (presence/not presence in this case)
73


classifier system and are created by plotting the correct classification rate against the
false positive rate at various threshold settings, and the AUC value is produced from this
graph. This value represents the probability that a random point that is chosen will be
correctly placed in an area of the geographic space with a high probability of species
presence. Therefore, if the AUC is .7, then there is a 30% chance that a known presence
location will be incorrectly classified as absent. A confusion matrix is also provided as an
output of these extrinsic model tests; percentages of omission error, false negatives, and
commission error, false positives, are reported. In modeling exercises like this one, where
there is only presence data, only an omission error percentage is provided (Anderson et
al. 2003; Stockwell and Peters 1999; Walton 2009), as there are no values with which to
produce a commission error.
Best Subsets Protocol
The validation tests discussed in the previous section are extrinsic to the model;
however, as discussed by Anderson et al. (2003), applications of both extrinsic and
intrinsic tests produce the best models. This is because modifications of GARP for
presence only data can cause a fitness ridge rather than a fitness peak... solutions along
that ridge differ dramatically in error composition as well as qualitative aspects of the
geographic prediction with errors at one end of the ridge including a great deal of
commission errors and the other of omission (Anderson et al. 2003:20). Therefore, the
best subsets procedure, which is a method of intrinsic testing, as described by Anderson,
et al. (2003) was used in this modeling process, to make sure that outputs of the modeling
process are within the best area of the fitness ridge (Figure 14).
74


In these graphs, the X axis represents the commission index, or the percent of
geographic space predicted as present, and the Y-axis is the omission error, or the amount
of known presence points which are predicted as absent by the model. Panel A within
Figure 14 demonstrates the range of models which the GARP algorithm can produce. The
smaller boxes demonstrate how different types of error affect the modeled niche; high
omission error is inaccurate because known presences are predicted as absent, zero
omission error with low commission index is not desirable because it causes overfitting
and zero omission errors but high commission index causes over generalization. Panel B
demonstrates that models with a higher than 10% omission error are definitively bad
because they incorrectly classify known presence points. The last panel C shows the
fitness ridge produced by the model, where the middle region of this ridge includes the
best models because they prevent overfitting and overprediction. Models within the green
circle are all summed to one output spatial distribution map.
The user of the model can set parameters that force the modeling program to only
produce outputs within the desired range of internal omission error and commission
index, or percent of area predicted present. These parameters include a number of runs, in
this case 20 (Anderson et al. 2003; Banks et al. 201 l;Walton 2009), 10% maximum
acceptable omission error, and a 50% commission error. Therefore, the algorithm
produces 20 models, all of which are under 10% omission error. Then, the 10 models
with a commission index that are closest to the mean (50%) of the commission index are
selected and the outputs are summed into one map. This selects models which strike the
best balance between overfitting and overpredicting, and so likely best represents the
actual niche.
75


100
Region of
the best
models
Figure 14: Selecting Best Models with GARP Best Subsets Protocol. From Enrique
Martinez-2007.
76


openModeller
All modeling in this paper was performed with openModeller desktop software,
which is open source and free software package, available from
openmodeller.sourceforge.net. There are a wide range of algorithms which can be used
for ENM purposes, and each time a new one is developed or updated it is typically
released as a new software program that can only handle a certain format of data inputs
and only produces a certain format of data outputs. The openModeller software was
developed to solve these issues through building a single computing framework capable
of handling different data formats and multiple algorithms that can be used in potential
distribution modelling (Munoz et al. 2011: 111). The use of this software allows for
quicker data preparation, allows multiple models to be run at once, and allows a string of
models to be lined up in preparation to run. This decreased user time spent preparing data
and monitoring the progress of the program, so more experiments could be run in a
shorter amount of time.
Each model was run with the GARP with Best Subsets new openModeller
implementation. The algorithm run in openModeller is the second implementation of the
GARP algorithm, based on the original code by David Stockwell (Stockwell and Nobel
1992; Stockwell and Peters 1999), and modified by Ricardo Scachetti Pereira to run in
open source software.
77


CHAPTER V
RESULTS
This research project was designed to test the efficacy of using ecocultural niche
modeling to produce potential site distribution maps of Mousterian sites in southern Italy,
or in other words, to model the niche of Mousterian sites based on modern day variables.
Most studies which model the Mousterian niche focus on describing and modeling the
past niche (Chapter II) but this study also attempts to connect site distribution to broader
theories of spatial distribution modeling through the use of modem data to model the
niche of southern Italian Mousterian sites. Additionally, this study seeks to explore niches
and species distributions of the southern Italian Mousterian sites in interglacial and
glacial periods.
Data and Model Preparation
Once all presence points were collected, confidences in accuracy were assigned to
the planametric coordinates. High confidence refers to those sites which have a
confidence of a few meters, medium confidence refers to those sites which could have an
error range of one-kilometer or less, and low confidence refers to those sites which could
have an error range of one-kilometer to four-kilometers (Table 1). The subset of sites
classified as high or medium confidence sites were collected into one excel file, and all
sites, high, medium, and low classification, were collated into another excel file. Each file
was saved as a text (tab) delimited file with the appropriate layout and column names for
modeling in openModeller. Additionally, a subset of presence points was selected based
on climatic information, either interglacial or glacial, and saved as its own file for
78


modeling. As data was recorded for each Mousterian layer at the site of interest, each
layer was treated as its own distinct presence point.
Environmental data layers were processed to the same spatial extent, cell
resolution, and custom projection, Italy Conformal Conic, based on the WGS 84 datum.
For a more detailed discussion see Chapter IV. The Desktop GARP algorithm with best
subsets with Open Modeler implementation was used to model the data; hard omission
error was set at 10%, iterations were set at 1,000, and convergence was set at .01, as per
previous ECNM studies (Anderson et al. 2003; Banks et al. 2011; Walton 2009).
Selection of Outputs
As a major goal of this project was exploratory data analysis, as well as testing the
efficacy of using modem variables to model the Mousterian niche, approximately 30
models were ran which produced 50 maps to explore how various sets of data, both
environmental and presence, changed the model outputs. Not all of these outputs
produced useful results, and so outputs that are reported on and discussed below were
chosen based on a combination of highest AUC values, lowest omission error and a
robust number of input presence points (Pearson 2010; Peterson, et al. 2008; Phillips et
al. 2006). Researchers have demonstrated that AUC values are not always the strongest
methods with which to evaluate models; however, it is a more robust method to compare
results of outputs that were produced with the same algorithm. (Banks, dErrico,
Peterson, Kageyama et al. 2008; Peterson et al. 2008) Additionally omission error was
taken into account to avoid overfitting.
Consideration of the number of inputs is important because smaller inputs have a
greater chance of producing biased models, and model performance tends to decrease
79


with very small sample sizes. Various studies have indicated that model performance
drops at n<10, 15, or 30 data points (Banks, dErrico, Peterson, Kageyama et al. 2008,
Stockwell and Peterson 2002, Wisz, et al. 2008), especially with a 50% partition for
training and model testing, where only half of the total presence points would be used to
produce the model. Therefore, those subsets of data which dropped below 30 points were
thrown out. Additionally, the large resolution of some of the models, four-kilometers,
means that some cells included multiple sites. Each of the points in the same cell provides
the model with the same exact environmental inputs, and so the largest sample sizes were
chosen to provide the most variety in environmental inputs to the algorithm.
Additionally, to confirm that these models are performing at a better than random
chance an exact one-tailed binomial test was performed on all outputs considered below
(Anderson et al. 2002; Pearson 2010; Phillips et al. 2006). Eising values produced in the
confusion matrix of each output, the probability of having at least t(l-r) successes out of t
trials, each with a probability of a was tested, where t = the number of test data points,
r = the omission rate, and a = the proportionally predicated area, or the percentage of
cells present (Pearson 2010). All models predicted significantly (p<01) better than
random, which is also supported by AUC values, which are all greater than .5.
Jackknife Test of Variables
Each set of data was run through the jackknife function in the openModeller
interface. This was done to find the subset of layers which provides the highest AUC
value and smallest omission error, and also to prevent overfitting of the model and thus
allow for extrapolation to under sampled areas. This function runs the model the same
number of times as there are layers, removing one layer each time and then reporting the
80


effect of each layer removal on the model. Results are reported as a percent accuracy of
the model when each specific layer is removed (Table 3). Layers that are more important,
and have more effect on the model outputs, have lower percent accuracy values when
they are removed from the modeling process.
Once a subset of significant layers was identified based on low percent accuracy
values from the jackknife function, the same model was run again with only the subset of
layers to explore if accuracy and usefulness of the model improved. If the model saw
improvement in terms of decreased omission error and increased AUC value, this model
was chosen as the most effective model and reviewed below. When it was not, the two
outputs were compared. Due to the stochastic nature of the algorithm, which means the
same outputs will not be reproduced even with the same inputs; slight differences in
model accuracy are not usually important. When two models are similar, as they are in all
cases, they are both considered to determine how, if at all they differ, and what the cause
of the differences may be, through both the color ramp raster and through a fixed
thresholding of the outputs, where a value of 50 or higher is considered as part of the
Mousterian niche (Pearson 2010).
81


Table 3: Jackknife of Environmental Layers: Accuracy of the each model without the
specific layer. Bolded numbers are the important layers that were modeled as subsets.
Stars indicate the most important environmental layers to each moc el.
4km 1km Glacial Interglacial
Elevation 90.2439 80.0000 72.4138 77.4194
Aspect 85.3659 65.7143 79.3103 67.7419
Slope 87.8049 77.1429 79.3103 74.1935
Dist. to Populated PI. 87.8049 68.5714 N/A N/A
Dist. to Rivers 85.3659 80.0000 N/A N/A
Flow Accumulation 90.2439 65.7143 68.9655* 58.0645*
Flow Direction 85.3659 80.0000 82.7586 77.4194
Geological Age 85.3659 68.5714 72.4138 87.0968
Land cover 87.8049 80.0000 N/A N/A
Soils 82.9268 77.1429 N/A N/A
Annual Mean T emperature 85.3659 68.5714 79.310 64.5161
Mean Diurnal Range 87.8049 71.4286 75.8621 58.0645*
Isotherm ality: modem 92.6829 71.4286 72.4138 83.8710
T emperature Seasonality 92.6829 65.7143 72.4138 67.7419
Max T emp of W armest Month 87.8049 71.4286 75.8621 51.6129*
Min T emp of Coldest Month 90.2439 68.5714 72.4138 83.8710
T emperature Annual Range 92.6829 68.5714 72.4138 74.1935
Mean Temp of Wettest Quarter 85.3659 71.4286 68.9655* 83.8710
Mean Temp of Driest Quarter 90.2439 80.0000 75.8621 61.2903
Mean Temp of Warmest Quarter 85.3659 77.1429 75.8621 80.6452
Mean Temperature of Coldest Quarter 85.3659 71.4286 79.3103 80.6452
Annual Precipitation 82.9268 68.5714 72.4138 83.871
Precipitation of Wettest Month 87.8049 68.5714 72.4138 61.2903
Precipitation of Driest Month 90.2439 71.4286 68.9655* 80.6452
Precipitation Seasonalitv 85.3659 77.1429 82.7586 80.6452
Precipitation of Wettest Quarter 90.2439 74.2857 72.4138 80.6452
Precipitation of Driest Quarter 85.3659 62.8571 82.7586 80.6452
Precipitation of Warmest Quarter 90.2439 71.4286 79.3103 58.0645*
Precipitation of Coldest Quarter 87.8049 71.4286 72.4138 80.6452
82


The Niche Mousterian Sites in Southern Italy
These first two models were run to determine if considering all Mousterian sites
from the region as one set of data can produce an acceptable model of the Mousterian
niche based on modem variables, and thus produce a potential site distribution. This test
was performed at both a four and one-kilometer resolution to determine if the differing
scales of data as well as the differing sets of presence point inputs produced a similar
patterning in the potential distribution of sites.
The Modern Niche at Four-kilometer Resolution
The four-kilometer resolution model only includes the subset of variables
identified in the jackknife as the most important, which includes aspect, distance to
rivers, flow direction, geologic age, soils, annual mean temperature, mean temperature of
wettest quarter, mean temperature of warmest quarter, mean temperature of coldest
quarter, annual precipitation, precipitation seasonality, and precipitation of driest quarter
(Table 3). Reduction of the variables increases the AUC value by .03, from .82 to .85 and
decreases omission error from 1.4% to 0% when compared to modeling done with all the
variables.
While modeling with the subset leads to improvement in the model, this
improvement is still small, and so modeling outputs with all the variables are compared
to the subset. Overall, the figures look extremely similar, especially when a fixed
threshold at 50 was applied to determine where sites are present and where they are not
(Figure 15). Major differences can be seen in the Tavoliere Plain, which is a depositional
plain with thick Holocene deposits (De Santis et al. 2010). This large gap is likely driven
by a combination of climate and soils. Due to the Holocene deposits, no sites have been
83


found in this region, which means that the suite of environmental variables present are
interpreted by the model as beyond the range of Mousterian site inputs values, or outside
of the niche. As shown by the subset model, when the climatic variables are reduced, the
model is able to better predict into this region.
Using summary statistics within ArcGIS, it is possible to generally characterize
the modeled site distribution output based on some of the more important variables. In
this case, the two most important variables were soil and annual precipitation. Soil types
from the region are varied (Appendix A), but the range of the modeled niche does not
include soil types that are classified as towns. The most common soil type in the niche is
eutric cambisol; or a young soil with relatively little horizon development. Annual
precipitation ranges from 46.5 centimeters to 94.0, out of a possible range of 44.9 to
116.5. Very low and very high values of annual precipitation are not conducive to
predicting Neandertals and Mousterian sites as present. Interestingly, some of the lowest
values are in the Tavoliere Plain, which could help to explain the large gap in the model
outputs.
84


Figure 15: Modern Niche at 4km Resolution. Mousterian sites included in the modeling
process are represented. The Tavoliere Plain is circled on the upper panels.
The Modern Niche at One-kilometer Resolution
The one-kilometer resolved outputs are again very similar (Figure 16), however a
model produced with a the subset of important variables identified through the jackknife
test, including aspect, distance to populated places, flow accumulation, geologic age,
annual mean temperature, annual precipitation, precipitation of wettest month,
precipitation of driest quarter, temperature seasonality, minimum temperature of the
coldest month and temperature annual range, sees a decrease in AUC value from .84, as
85


modeled with all variables, to .82. Omission error remains at 0%. The subset model is
interesting because it predicts more area present in the peninsula of Calabria, which likely
better represents actual Mousterian range given that the site of San Fransico di Archi is
located in that region (Figure 16).
Jackknifing variables demonstrate that flow accumulation, precipitation of driest
quarter, and temperature seasonality are the variables whose removal decreases accuracy
of the model the most. Precipitation of the driest quarter ranges from 3.5 centimeters to
11.9, with a mean of 7.2 centimeters and a standard deviation of 1.6. Values for the
whole area reach up to 22.4 centimeters ; the model suggests that areas which are the
wettest in the dry season are not suitable for the presence of Mousterian sites. This
pattern could be due to the fact that the highest values of rainfall in the dry season are
associated with highest elevations, so altitude or another associated variable are
preventing Mousterian sites from being present in areas with the highest levels of
precipitation. Values of flow accumulation, which is a measure of the upstream
catchment area where the values represent the amount of upstream area draining into
each cell, range from a low of 1 to a high of 30,866 in the niche, while the highest of the
region is 35,810. However the mean of the niche is approximately 97. Therefore, in this
model, sites are usually found present in areas with less catchment, which could driven
by where Neandertals were placing their sites, or could be due to taphonomic processes if
areas with high catchment destroy archaeological remains. Temperature seasonality also
has an effect on the patterning of sites. Seasonality, which is calculated from the standard
deviation of the lower temperature and the highest, multiplied by 100 in the niche ranges
from 4935 to 6372. In the region the highest seasonality is 6461, therefore, the
86


Mousterian niche is not modeled into areas that experience the most extremes in
temperature between seasons.
Figurel6: Modem Niche at 1km Resolution. Site of San Francesco dArchi was not
included in the modeling process. Other sites represented by black dots were included in
the modeling process.
87


The one-kilometer prediction layer was produced as a way to test the outputs from
the four-kilometer resolution to confirm that the model is producing consistent outputs at
differing resolutions and with differing sets of presence points. In addition, as a goal of
this project was to produce a map which would assist survey efforts into under-
researched regions, the one-kilometer output was produced as a more resolved prediction
map to direct areas of future research.
Overall, statistical comparisons of niche overlap and niche breadth in ENMTools
version 1.2 (Warren et al. 2009, 2010) suggest the one-kilometer and four-kilometer
models are extremely similar. EMNTools is a freely downloadable software is meant to
facilitate quantitative comparison of environmental niche models (Warren et al. 2010:
607) and was used to calculate statistical comparisons between results. Niche overlap is
measured through Schoeners (1968) D and by a measure derived from Hellinger distance
called / (Schoener 1968), both these measurements are obtained by comparing the
estimates of habitat suitability from the output files produced the ECNM. In other words,
once an output raster layer with probability of site presence has been produced, with
values ranging from 0-100, two outputs can be compared by prediction value at each cell
to determine similarity. To apply this statistic to the one-kilometer resolved output and
the four-kilometer output, the four-kilometer output cell size had to be decreased to the
same cell size as the one-kilometer output in arcGIS. Results indicated that the two
niches are similar. The D statistic was lower, at .58, while the / statistic suggested that the
overlap of the models was .81 (Table 4). There are some differences in the model,
particularly in the Tavoliere Plain and the peninsula of Calabria, which is what the
statistics are likely representing. Additionally, it must be remembered when comparing
88


Full Text

PAGE 1

BIOGEORAPHY OF NEA N DERTALS: SITE DISTRIBUTION PATTERNING IN THE SOUTHERN ITALIAN MIDDLE PALEOLITHIC by KELSEY ISABEL KNOX B.A., University of California, Los Angeles, 2012 A thesis submitted to the Faulty of the Graduate School of the University of Colorad o in partial fulfillment of the requirements for the degree of Masters of Arts Anthropology Program 2015

PAGE 2

2015 KELSEY ISABEL KNOX ALL RIGHTS RESERVED

PAGE 3

ii This thesis for the Master of Arts degree by Kelsey Isabel Knox has been approved for the Anthropology Program by Tammy Stone, Chair Julien Riel Salvatore Jamie Hodgkins June 9 2015

PAGE 4

iii Kelsey Isabel Knox (M.A., Anthro pology ) Biogeography of Neandertals: Site Distribution Patterning in the Southern Italian Middle Paleolithic Thesis directed by Professor Tammy Stone ABSTRACT Most researc h on the Middle Paleolithic in s outhern Italy has focused on Mousterian sites in the region of Apulia. This research has been extensive and productive; therefore it stands to reason that other, less researched regions of southern Italy, including Basilicata, Calabria, and Campania, hold potential to produce more Mousterian sites This paper uses ecocultural niche modeling and predictive site modeling to explore site distributions and the past Mousterian niche as predicted from site occurrence data and environmenta l variables. This niche is th en projected onto the landscape. This projection produce s a potential site distribution map when modeling modern site niche against modern environmental variables and produce s a spatial distribution map when modeling the past niche s with past environmental layers. Results indicate that producing potential Mousterian site distributions is a successful exercise, which can identify areas with in under researched regions of southern Italy that have a high probability of presence of sites. Additionally, a general description of the Mousterian niche in glacial and interglacial periods is produced, and a p otential biogeographic variable, the Apennine Mountains, is identified. The form and content of this abstract are approved. I recommend its publication Approved: Tammy Stone

PAGE 5

iv ACKNOWLEDGEMENTS Thanks to my advisor, Julien Riel Salvatore, and my friends and family for all the support and guidance.

PAGE 6

v TABLE OF CONTENTS CHAPTER I. INTRODUCTION . 1 II. THE DISTRIBUTION OF MOUSTERIAN SITES IN ITALY ..4 Neandertals in Italy: An Overview ..4 Prehistoric Archaeology in Italy: A Historical Review ... Italian Geography: Apulia, Campania, Basilicata, and Calabria 12 Explaining t 22 III. ECOLOGICAL NICHE MODELING: HUMAN AND NON HUMAN SPECIES 27 Development of the Niche Concept 28 Applications and Roles of GIS Technologies.. Theoretical Foundations of Ecological Niche Modeling (ENM)... 31 Ecocultural Niche Modeling (ECNM). ..38 Predictive Site Modeling 43 Extra Theoretical Considerations 47 Hypotheses .51 IV. RESEARCH METHODS 53 Known Sites in Southern Italy Environmental Data Layers 57 Software: GARP with Best Subsets 70 V. RESULTS Data and Model and Preparation 78 Selection of Outputs 79

PAGE 7

vi Jackknife Test of Variables The Niche of Mousterian Sites in Southern Italy The Niche of Mousterian Sites in the Last Glacial Maximum 92 The Niche of Mouste rian Sites in the Comparison of the Niche Models VI. DISCUSSION 105 Occupied Niche v ersus Actual Niche Model Similar M odel Generalization Mousterian and Neandertal S ites as a Homogenous Assemblage 4 Biogeographic Variable: The Apennine Mountains VII. CONCLUSIONS AND FUTURE DIRECTIONS 119 WORKS CITED.. 123 APPENDIX .. 137 A: Geologic Age, Soils, Land C over V alues 137 B: Comparison of Niche Values 141

PAGE 8

vii LIST OF TABLES TABLE 1. Known Sites in Southern Italy 56 2. Environmental Layers 0 3. Jackknife of Environmental Layers 82 4. Niche Overlap Statistics .89 5. Niche Breadth Statistics .90

PAGE 9

viii LIST OF FIGURES FIGURE 1. Distribution of Mousterian Sites in S outhern Italy 2. Mountains of Italy 3. 4. Landsat Image of Southern Italy 5. Soil Erosion Risk in Southern Italy 6. Climate of Souther 7. Popul ation Density of S 8. Southern Italian Provinces and Regions ...........21 9. Hypothetical Niche 10. Environmental Envelope 11. Theoretical Foundations of Ecological Niche Models 12. Geographi c Space and Environme nt 13. Ecocultural Niche Modeling 14. Selecting Best Models with GA RP Best Subsets Protocol 15. Modern Niche at 4km Resolution 16. Modern Niche at 1km Resolution 17. Productive Areas for Future Research Identified 18. Glacial Niche at 4km Resolution ...94 19. Gl acial Niche with 80m 20. Interglacial Niche at 1km Resolution 21. The European Neandertal Niche

PAGE 10

ix 22. 108 23. Similarity of the Modeled Niche 24. Productive Areas for Future Research Identified 25. Comparison of the N 26. Biogeographic Barrier of Southern Ital y 8

PAGE 11

x LIST OF ABBREVIATIONS ABBREVIATIONS ENM: Ecological Niche M odeling ECNM: Ecocultural Niche Modeling AUC: Area Under the C urve ROC: Receiver Operating C haracteristics GCM: General Circulation Models GARP: Genetic Algorithm for Rule Set Prediction LGM: Last Glacial Maximum LIG: Last Interglacial Period OIS: Oxy gen Isotope S tage

PAGE 12

xi Supplementary Materials : Microsof t Access Database Table with information about known so Each row represents a Mousterian depositional layer, and the information recorded includes a general description of lithic and non lithic tools, counts of lithic and non lithic tools, fauna, raw material, symbolism, structures, cultural attribution, site type, discovery, excavation and publication dates, and finally a list of all sources referencing the particular site A references cited document is al so included, titled Geographic coordinates are p resented separately in the text (Table 1)

PAGE 13

1 CHAPTER I INTRODUCTION This thesis applies ecocultural niche modeling (ECNM) to investigate potential distributions of Mousterian sites in southern Italy, as well as to investigate the Neandertal niche in glacial and interglacial periods as modeled based on Mousterian sites Because of the uneven distribution of Mousterian sites, southern Italy is productive study area to explore h ow applications of ECNM can be applied to model the Mousterian niche during various climatic regimes Additionally, this method and study area is well suited to determine if the modern niche Mousterian sites can be modeled and connected to ideas of predict ive site modeling to produce a potential site distribution map in order to direct future research in to unknown sites in the region. Overall, this research aims to explore how sites might be patterned against modern and past environmental variables, and h ow this patterning can be used to understand where sites are located on the landscape today, and where they were located in the past. The distribution of Mousterian sites in southern Italy can also be seen as function of historic and geographic variables that has led to an explosion of research in the region of Apulia, while other southern Italian provinces of Campania, Calabria and Basilicata have been largely overlooked. Because Apulia has produced many Mousterian sites, it is possible these under resea rched regions contain more sites than previously thought. These geographical and historical factors and how they have produced the distribution of known sites in southern Italy are considered in Chapter I I.

PAGE 14

2 Chapter II I reviews the theoretical background of ECNM which draws heavily on ecological niche modeling (ENM) but focuses on modeling human species as opposed to non human and considers culture in the modeling process. A brief literature review of ECNM as applied to archaeological data is reviewed in this chapter Additionally theoretical considerations of predictive site modeling are mentioned and concepts of ECNM and predictive site modeling are explicitly connected to facilitate interpretation of modeling outputs as site distribution maps. In Chapter IV the research methods deployed in this study are re viewed and explained in detail. Geographic coordinate collection is explained, and the environmental data layers used in the modeling are descri bed. Finally, data manipulation, data processing and the parameters of the modeling are explained Results are reviewed in Chapter V. Visual outputs in the form of site and spatial distribution maps are presented, as well as a discussion of the more importan t environmental variables that e ffected the mod el. Areas that may be productive locations for future research are identified. Additionally, results of niche overlap and niche breadth statistics are repor ted, with a comparison of the modeled European Neandertal niche from previous studies and the modele d sou thern Italian Mousterian niche from this study. Chapter VI first discusses how similarities between all the models, identified both through visual comparison and through statistics of niche overlap and niche breadth indicate that the model was ultima tely successful. Additionally, generalizations about the glacial and interglacial niches are made; including how the niche may have shifted as environmental conditions change. Also in this chapter the suitability of considering Neandertal and Mousterian sites as a homogenous unit when doing ECNM is reviewed.

PAGE 15

3 Finally, the Apennine Mountains are considered as a potential biogeogra phical barrier in southern Italy. Chapter VII reviews conclusions of th e project, and suggests future directions for research; including expansion of the presence point collection area and the modeling extent to all of Italy, and potentially other refugia areas in Europe. Additionally, explorations of the biotic niche of the region in the Eltonian sense, are considered, along with applications of the modeled niche to the Italian island of Sicily.

PAGE 16

4 CHAPTER II THE DISTRIBUTION OF MOUSTERIAN SITES IN SOUTHERN ITALY Level and intensity of archaeological research varies across geographic space. Much of this variation can be linked to historical processes that define some areas as more attractive and viable for archaeological study than others as well as natural features that make some area s more or less difficult to explore. Southern Italy is no exception. Neandertals in Italy: An Overview Both Neandertal, sites with human remains, and Mousterian, sites with only Mousteiran lithic technology, archaeological sites have been found in Italy. Generally in Italy, Mousterian sites span from oxygen isotope stage ( OIS ) 8 through OIS 3; ( Mussi 2001) however, t here are no known Mousterian sites in Italy from OI S 8, and they are sporadic until OI S 5 (Muss i 2001:101). L ithic technologies found at these Italian Mousterian sites are generally recognized to follow the frameworks established from the French archaeological record by Francois Bordes (Bordes and Sonneville Bordes 1970); researchers have found the Typical Mousterian, Denticulate M ousterian and Quina Mousterian, although Italian ar chaeologists often subsume the D enticulate Mousterian and the Quina Mousterian in to broader category of the Charentian Mousterian A Pontinian indust ry has also been defined, which is thought to be a loca l adaptation of the Quina Mousterian to small pebbles of chert that are one of the only sources of raw material in the region (Mussi 2001: 104). Additionally Italian researchers have identified the Quinson Mousterian, characterized by small size d tools an d carinated forms with a dihedral ventral face and Clactonian notches. This typology is used to date sites, as evidence of the techn ology is considered absent after OIS 5 (Mussi 2001: 105). A final or

PAGE 17

5 evolved Mousterian, characterized by a decrease in side scrapers, an incre ase in notches and denticulates, and the presence of Upper Paleolithic tool types including blades has also been identified at Italian Mousterian sites Overall with the exception of Sicily and Sardinia, there are over 350 Mousterian sites (Milliken 2000:11 ) found in all areas of the country, including along the coast, in the interior v alleys, and on the high mountains S outhern Italy, the focus of this study has approximately 110 sites (Milliken 2000 ). Although generally southern Ita ly contain s a large number of Mousterian sites as Figure 1 demonstrates these sites are mostly concentrated in the region of Apulia. Here, historical processes in the development of Italian archaeology, in addition to geographic features, including natur al, social and political, are review ed in an attempt to explain the patterning of Mousterian sites in southern Italy, including the high concentration of sites in Apulia and lesser concentrations in Campania Basilicata and Cala bria (Figure 1). Figure 1: Distribution of Mousterian sites in Southern Italy.

PAGE 18

6 Prehistoric Archaeology in Italy: A Historical Review Knowledge of the history of Italian archaeology including trends in theories and methods, is essential to understanding why Italian archae ologists have undertaken research on Mousterian and Middle Paleolithic sites in specific areas. In this way, a historical review of Italian archaeological history can help to explain the present day distribution of known sites in s outhern Italy. Like all of modern Western archaeology, Italian archaeological thought is derived from ideas first developed in the Italian Renaissance (Guidi 1987; Mussi 2001), including antiquarianism, comparative methods, and classification of the past into stages. The fashion of antiquarian collections was born in Tuscany during the second half of the 15 th century (Guidi 1987: 237) and t he first comparative method was used by Michele Mercati to conclude that stone tools, thought by the Romans to be tools of mythical heroes and Medieval Italians to be the tip of lightening blots, were actually produced by flint percussion during a time befo re iron was used (Mussi 2001: 6; Guidi 1987: 237). In 1725, Giambattista Vico proposed that human kind developed in stage s, from the Age of the Savage to the Age of Gods, of Heroes and of Men. Together, these concepts developed into evolutionary chronological theories in Italy, which continue to play an important role in the study of prehistory today (Guidi 1987: 238). Despi te being the birthplace of the first archaeological theorizing and research from the 17 th to the 19 th centuries, Italy lagged behind other European countries in the development of archaeological thought. These events included y and oppressive climate of the counter 238), the late formation of a centralized state in Italy, archaeology (Guidi 1987: 238).

PAGE 19

7 In most other European countries, the birth of prehistoric studies was tied to the emergence of the bourgeoisie, a social class that developed with the Industrial R evolution and the creation of centralized state s The Italian state, unlike other European countries, is relatively young, and previous to unification consiste d of various independent smaller states or kingdoms that had long and diverse histories (Guidi 1996c: 108). In 1861, with the unification of the Italian state, and thus the full participation of Italy in the Industrial Revolution and the development of the bourgeoisie, inte rest in prehistory resurged After unification, the history of Italian archaeological research can be divided into five phases: Phase 1: 1860 1900, Phase 2: 1901 1921, Phase 3: 1 9 22 1945, Phase 4: 1946 1970, and finally, Phase 5: 1971 pre sent (Guidi 2010) Phase 1: 1860 1900 Local Italian antiquari ans, drawing on the positivist ideas and concepts of uniformitarianism and evolution in order to understand the past, drove archaeology in the In the early stages of this period, Italian scholars were focused on reconstruct ing the Italian Bronze age (Guidi 1987, 2010; Palma di Cesnola 1991). Also at this time, foreign scholars were using paleontology to attempt to define periods of the Paleolithic at Italian and other European arch aeological sites. Later, as the Italian state gained more centralized control, archaeology shifted from antiquarians to the Roman School and its founder, Luigi Pigorini. The Roman school coalesced with the appointment of Luigi Pigorini as the chair of Pale oethnology (Italian term for prehistory) in 1877 in Rome Pigorini attempted to centralize all Italian research, and he and his students claimed almost absolute autonomy with regard to the research on prehistory, including the terminology, the evolution o f cultures, and methods

PAGE 20

8 of study and the theoretical foundations (Palma di Cesnola 1991: 12). They argued for ethnography a s the only way to understand the p rehistoric past and theorized that most cultures evolved unilinearly and locally. Overall, the theo retical focus of this school was cultural historical. Additionally, the Roman School rejected the French classification system of the Paleolithic, particularly the Upper Paleolithic, instead preferring to consider the Italian Paleolithic as completely sepa rate from other Paleolithic chronologies of Europe. In fact, they claimed that there was no Upper Paleolithic in Italy, instead the Late Mousterian evolved directly into the ea rly Neolithic (Guidi 1987, 2010; Mussi 2001; Palma di Cesnola 1991). By 1900 P igorini had successfully consolidated almost all archaeological research in Italy, and prevented the formation of any more chairs of archaeology outside of Rome. Despite this, scholars from the south like Ridola in Bascilicata, Taramelli in Sardinia, and P to prove important in the construction of the chronological and cultural framework of Phase 2: 1901 192 0 This period was characterized by fighting between the Roman and Florentine Schools of ar chaeology, and their associated theoretical backgrounds. While Paolo Mantegazza founded the Florentine School around the same time as the Roman School ( Mussi 2001; Guid i 1987, 2010; Palma di Cesnola 1991), it was not until the early within the Florentine School that the conflict developed. The Florentine scholars took a more naturalistic approach to understa nding prehistory, which focused on the study of humans

PAGE 21

9 in the context of the environment, and attempted to fit Italian chronology to broader chronologies derived from continental Europe (Palma di Cesnola 1991). Through excavations of Grotta Romanelli, thes e scholars were able to concretely establish the presence of the Upper Paleolithic in Italy (Mussi 2001: 9). Unfortunately because of infighting between the two schools, this whole period was marked by a decline in methods and increased isolation of Itali an scholars from the rest of archaeological research in Europe (Guidi 2010: 15). Phase 3: 1922 1945 The fascist period in Italian history also had a marked influence on Italian prehistoric archaeology. Emphasis shifted to classical archaeology as the Fasci st P arty attempted to demonstrate continuity between their government and ancient Rome The shift toward classical archaeology was also driven by the development of the ion, and argued that there was no way to understand the pre written past, and therefore efforts to do so should be abandoned The Roman S chool was most affected by these changing perspectives on pr ehistory, while the Florentine S chool continued their natur alist path by environmental setting of the evolution of the with the larger European community continued to wane as scholars became convinced Italy was unique an idea fostered by fascist ideals Despite the separation between the two schools, the excavations at the Arene Candide in Liguria during this period were the first collaboration between a classical scholar Luigi Berna bo Brea and naturalist ic scholar, Luigi Cardini. This excavation slowly began the unification of culture history

PAGE 22

10 ideas of the Roman school and naturalistic ideas of the Florentine school, and was the first precursor of modern prehistoric Italia n archaeolog y (Guidi 1987, 2010 ). Phase 4: 1946 197 0 After the end of the World War II, and w ith the waning of power of the Fascist P arty, there was a sudden national revival in all fields of social and cultural life, including prehistoric archaeological research. Many prehistory chairs were created throughout the Italian university system. This period in the United States saw the rise of 196 adoption of this theoretical framework did not extend beyond the implementation of st atistical analyse s in lithic studies ( Guidi 2010). influence continued to support a climate of opposition to functional approaches and opposition to experimental archaeology that the New Archaeology advocate d. Additionally, the t rend of isolationism continued as most of Europe began using the Bordes system of lithic classification, while Italian scholars focused on the Laplace method (Mussi 2001: 11). ted peninsula at the bottom (Barker and Hodges 1981: 2) which was encouraged by cross dating Italian prehistory sites from other areas of Europe was disrupted by the development of radioc arbon dating. This led to Italian scholars considering models of functional stability and internal evolution as drivers of culture change rather than invading cultures from the north (Barker and Hodges 1981: 2). Ultimately, this may have encouraged further isolation.

PAGE 23

11 Phase 5: 1970 present Significant change occurred during the 1970s as a group of northeastern Italian scholars emerged who associated themselves with theoretical approaches and up to date excavation methods from English archaeolog ists (Guidi 1987, 2010). H owever despite the use of new theories and methods in the tradition of the past a significant portion of the new work was still cultural historical ly focused The use of mathematics and personal computers that began in the 1960s continued to spread, as well as the appointment of prehistoric archaeologists into various State Offices for Antiquities throughout the country. Additionally, a mateur ar chaeologists saw another resurgence ; in fact amateurs located many of the new prehistoric sites. became localized within regions and areas of Italy. Southern Italian archaeologists developed a ications and the organization of local museums and congresses, and a group of Campanian archaeologists developed a postprocessual approach. However, most of the research on the Paleolithic was and continues to be done by northern archaeologists. These ar chaeologists began to use a focuses on differences between northern Italy and the rest theory and intensive computer applications (Guidi 2010: 18). technology, microwear analys is, zooarchaeology and taphonmy The current challenge of Italian archaeology is the integration of the historical approach with the archaeological, in addition to attempting to integrate local and regional archaeological trends (Bietti 1991).

PAGE 24

12 Italian Geography: Apulia, Campania, Basili cata, and Calabria Geography, including natural, social and political factors, also inherently plays a role in site distribution. Italy has extreme differences in geography between each region, as well as within some regions, and so it is likely these fact ors can help explain the present day distribution of sites. Italy is a long, narrow, extremely mountainous peninsula. It ext ends the Alps at 47 degrees north, to Sicily at 37 degrees north, at the same latitude of Algeria, Tunisia, of 300,000 km sq., there are 125,000 km sq of hills, and 100,000 km sq of mountains, encircled by some 9,000 km of marine coasts (Mussi 2001:1). The terrain of Italy translates into three type s of elevation, mountain areas above 1,000 meters with valleys at various elevations throughout, plains which consists of level or gently sloping land below approximately 300 meters, and the intermediate hill areas between the two (Cole 1966: 20). Italy is also a zone of tectonic activity includi ng volcanoes and earthquakes. T his tectonic activity has contributed t o the development of the mountain ranges of Italy including the Alps in the north of the c ountry and the Apennines which run in northwestern southeastern parallel ranges through most of the peninsula of Italy (Mussi 2001: 2; Figure 2) Due to the presence of the Apennines, east west movement across the country is difficult; however due to valleys north south movement through the peninsula is comparatively easier.

PAGE 25

13 Figure 2 : Mountains of Italy: Italian Alps and Apennine Mountains. M odified from www.shiney7.co.uk/Gustavline.html The Apennines make movement difficult throughout all four regions, however this far south they are made of many distinct mountain areas and do not form a continuous range Additionally, in the south these mountains do not usually reach higher than 2,000 meters, and generally are around 1,000 mete rs in elevation (Cole 1966: 214; Figure 3) A long history of human habitation, dense settlements and the practice of agriculture have significantly altered the vegetation of s outhern Italy (Figure 4). Ancient Roman settlements first started the process of land reclamation and deforestation; these processes continue to the modern day. Large grazing herds have altered mountain and flatla nd environments since the Late Middle A ges in central and southern Italy. Italian Alps Apennines

PAGE 26

14 Currently, the vegetation of souther n Italy falls into the Apennine type ; the typical tree is the oak, while olive, oleander, and pine are found in more coastal areas. The foothills are characterized by oak and pine. Mountain locations and areas of higher elevation still preserve ancient mountain forest. Beech woods are still present in Calabria and Apulia and areas of deforestation are now been replace d by a scrub bush ( Ermoli and di Pasqule 2002: 212 ) The combination of hum an modification of the landscape and mountainous terrain with steep valleys means the whole region is at risk of erosion (Figure 5). Figure 3: Elevation of S outhern Italy. From Hydrosheds 1km.

PAGE 27

15 Figure 4 : Landsat Image of S outhern Italy ESRI images. Figure 5 : S oil Erosion Risk in S outhern Italy. From the European Soil Database From Grimm et al 2003, Pangos et al 2012

PAGE 28

16 Climate is affected by multiple factors, but as a rule there are overall increasing temperatures and decreasing precipitation from north to south (Mussi 2001: 5) However, the mountains and sea are never far away from any area of land which causes in large variations of local climates and annual rainfall levels both between a nd within the regions (Cole 1966; Mussi 2001) For example, on the west coast rainfall is usuall y around 800 mm per year, on the east annual precipitation is approximately 700 mm per year, while areas in a rain shadow, such as the Tavoliere plain in Apulia only receive 500 mm of rain per year (Mussi 2001 Figure 6) Figure 6: Climate of Southern Italy. Annual precipitation and annual mean temperature From BioClim

PAGE 29

17 Southern Italy was a single political unit before unification, the Kingdoms of Naples, and because of this Campania, Apulia Bascilicata and Calabria together form a reasonably distinct geographic region (Cole 1966:214 ). Currently, and throughout time, it is the poorest area of Italy, with the lowest per capital income in the whole country, and the lowest quintiles on the development index ( Faini et al. 1993 ). The population is concentrated in the city of Naples and the region of Apulia (Figure 7). Addition ally, population can be considered as a proxy for development, more populated areas of southern Italy tend to have more infrastructure, more development of industry, and a higher annual income. Transportation throughout the area is difficult; roads are gen erally adequate ; however the railway system suffers from one way tracks and difficulties navigating the mountainous areas (Cole 1966). Figure 7: Popul a tion D ensity of Southern Italy. Can also be viewed as proxy for development From the GIS Data Depot at geocomm.com.

PAGE 30

18 Apulia Apulia is by passed by the Apennine Mountains, and so is distinct from the other southern Italian regions. It consists mainly of low limestone plateaus and plains with little elevation changes and so is one of the least moun tainous regions of Italy (Cole 1966: 216). However, the Gargano Peninsula contains its own, structurally dist inct mountain formation (Cole 19 66: 214). This formation is found in the Foggia province, and it contrasts with the other five provinces of the region, Barletta Andria Trani, Bari, Taranto, Brindisi and Lecce ( Figure 8 ) all of whic h have no mountain formations and are generally lacking in elevation change ( Cole 1966). Much of the land is used for agriculture or shepherding, and there is s ome industry in the larger cities ( Cole 1966 ). This region has the second highest population in southern Italy, with approximately 4.1 million people (Comuni Italiani 2004), and most of the population is concentrated in the cities. Apulia has one of the densest concentrations of known Mousterian sites in Italy. In this region, there are approximately 90 sites, almost as many sites as the provinces of northern or central Italy combined (Milliken 2000: 42 43) The first collections of Mouste rian artifacts from Apulia were compiled by amateur archaeologists in the mid 1800s and the first book addressing the Paleolithic of the region, titled Preistoria della Puglia was published in 1914 by Antonio Jatta. Beginning in the early 1900s, more sys tematic excavations by P.E. Stati and E. Regalia were carried out ; these were then taken over by the scholars from the Roman school, such as Pigorini, Antonielli, and Rellini, as well as those from the naturalist school, including A. Mochi, G.A. Blanc, and Romanelli (Palma di Cesnola 1967). In the early 1930s, Rellini, R. Battaglia, and E. Bamgaertei carried out e xtensive surveys and excavation s in the Gargano Peninsula.

PAGE 31

19 Starting in the 1950s and through the 1970s, excavations at known sites and surveys to identify new sites had spread to all areas of Apulia through the research of scholars such as A. Palma di Cesnola, Borzatti von Lowenstern, Cardini and various others (Palma di Cesnola 1967). Recent research has mostly focused on revisiting these previous ly excavated sites to determine more accurate stratigraphy, or to better in vestigate an older and deeper stratigraphic layer. Campania Campania contains large regions of the Apennine Mountain range ; approximately 35% of the province consists of mountain s while 51% consists of hill s, and 14% consists of plains However, despite this large amou nt of mountain range, the plains are exceptionally fertile and agriculture is extremely important and productive, in this region Additionally, there are extensive a rable valleys and agriculturally productive hills in the provinces of Benevento and Avellino ( Cole 1966: 221 ; Figure 8 ). This is the only province in southern Italy with a large concentration of industry, centered in the city of Naples Campania has a popu lation of approximately 5.8 million (Comuni Italiani 2004), which makes it the most populous region in the south. H owever much of this population, approximately 1 million, is concentrated in Naples (Comuni Italiani 2004) The mountain areas of Campania are generally not occupied at all, while in the hill areas populations densities are extremely high (Cole 1966: 221 3). This region has approximately 10 known Mousterian sites, considerably less than Apulia U nlike Apulia where the sites are evenly distribut ed across the landscape, the majority of Mousterian sites in Campania are located along the coast. For the most part, t hese sites were identified and excavated in the 1970s, by many of the same scholars who

PAGE 32

20 excavated a large number of the cave sites in Apulia. Generally the inner areas of Campania have been ignored in terms of archaeological research (Radmilli 1978 ). Basilicata This region is the least known of all the regions in Italy, both to the Italian s and to foreigners. Almost two thirds of Bascilicat is classified as mountains while the rest is classified as hill (Figure 8 ). There is intense erosion in this province due to natural factors such as floods and human created deforestation. Seismic activi ty is also high in this province. The western provience, Matera, contains high limestone massifs, with karst phenomena on the high plateaus, while the eastern provience, Potenza, consists of newer land of high hills and alluvial soils ( Cole 1966). The prov ince of Potenza contains mountains with deep, narrow valleys that are not particularly productive for crops. Villages are extremely isolated and there is only one city. This is the poorest area of Italy (Cole 1966: 224 5). On the other hand, Matera is at a lower elevation, less rugged, and there is a relatively large area of coastal lowland that allows for more extensive agriculture N either province has any sig nificant industrial development, although some is present. Populati on is at approximately 600,000 making Basilicata the least populated region in the south. Basilicata has approximately four known Mousterian sites. T he areas of the provience that have been the subject of research are generally along the coast or extremely close to the shared borders of the provience, at the expense of the inner areas. Calabria Calabria is even long er and narrow er than the rest of the Italian peninsula. High mountains and proximity to the sea dominate much of the geogr aphy of this region (Cole

PAGE 33

21 1966 ). Th ere is one large valley and most of the mountains steeply slope down to small coastal or alluvial plains at their feet; these steep slopes continue at least several hundred meters below sea level (Cole 1966: 225). Erosion from seasonal torrential rivers is common in Calabria. Due to railways, the west coast of the province is more connected to the rest of Italy when compared to the east; however tra vel is difficult in both areas. The majority of land use is agricultural, and the rema ining land consists of some of the largest forested areas in Italy ( Falcucci et al 2007 ). Population of this region is larger than Basilicata, but only about half as large as Apulia with approximately two million people occupying five provinces (Figure 8) Five Mousterian site s have been located within this region, and they like sites in Campania, are distr ibuted in coastal areas or along borders of the province Figure 8 Southern Italian Provinces and Regions From the GIS Data Depot at geocomm.com

PAGE 34

22 Explaining the Pattern of Sites: Discussion Distribution of Neanderthal sites through southern Italy can be seen as a function of both historical and geographic variables working in concert to create disparities in the level of research into the Middle Pal eolithic of the regions of Apulia Campania, Basilicata and Calabria. Generally, Italian archaeological theories should be discussed as trends rather than real traditions of research. These trends have been deployed in various ways within various political climates in Italy, particularly in regards to nationalism. In the early from northern Italy. They were also all members of the same upper social class and northern regional population that was responsible for the unification of the various states and kingdoms on the Italian peninsula. In original synthesis of Italian history, written by aly was explained by the spread southward, during the Bronze Age, of various waves of northern populations who, superimposing themselves on the natives of Neolithic origin, invented new forms of The newly unified Italian government paid particular attention to this version of prehistory, lik ely because it reinforced the narrative of Italian unification In contrast during fascist times, the trend shifted from prehistoric archaeology to classical, as the Fascist Party attempted to show continuity from Roman times to their regime to legitimize their rise to power (Guidi 1996b: 113). After the 20 years of centralizing fascist regime archaeological trend s have moved toward the development of local traditi ons. The combination of a lack of centralized push for Paleolithic research combined with the more recent development of

PAGE 35

23 local archaeological traditions may help explain the dispariety in levels of Mousterian and Middle Paleolithic research between and wit hin regions of southern It aly The b asis of known Mousterian site distribution also can be directly tied to ideas of Italian uniqueness, both in terms of regions and country, an d a unilinear evolutionary bias To define culture, Italian archaeologists dra w on cultural historical theories ; particular types of artefacts which represent normative ideas and mental templates of past people They also dra w on the naturalistic perspective to argue that changes in the cultural facies can be seen as changes in industrial assemblages; however in Italy this change is only conceptualized as unilinear (Bietti 1991: 261; Milliken 2000:11). Therefore, concepts of i nterstratification are not considered viable to understand the archaeological record, and it is generally assumed that each region has its own local development of Mousterian assemblages. So, comparisons between two regions to help understand a lesser stud ied region are not common (Bietti 1991, Mussi 2000: 12). These concepts of uniqueness and unilinear evolution also lead to a focus on the interpret the chronological order o f all other sites in the region. Because of the and importance of site types, Italian archaeologists typically focus on excavating deep trenches at cave and rockshelter sites, at the expense of both horizontal excavation and systematic survey. This has led to a serious bias in many regions, like Calabria, Campania and Basilicata, where most of the sites found are located in caves or rockshelters, as well as bias against research in areas lacking this type of stratified type

PAGE 36

24 site, like Basilicata and Calabri a (Mussi 2000:12). Additionally, northern archaeologists, who were the first to adopt more modern archaeological techniques during Phase 5, have undertaken the few systematic planned surveys that have been done in Italy, and have generally concentrated eff orts in the north (Mussi 2000:12). Finally, much of the archaeological work in the south, outside of Apulia has focused on the Classical Period, likely due to a combination of a holdover of this focus from Phase 3 and local tradition. For example, there is Archaeological Reports journal, which ran from the 1970s through the 1980s that only focuses on Classical A rchaeology (e.g. Ridgway 1982). Geographic variables are also important to explain the distribution of known sites; some of the most important variables are relief, population and development. Apulia with its high density of sites has one of the lowest averag e elevations of southern Italy, which makes it an attractive region for archae ological surveys, especially when compared to the other provinces. Additionally, the karstic areas mean caves are common, which also attracts archaeologists to research in the region. The higher, fairly evenly distributed population of the region in additi on to the high focus on agriculture increases chance of encounters with Mousterian sites across the province Finally, the productive agriculture and levels of industry means more construction projects are undertaken where sites may be found, and there is more money available to excavate and research those sites. Combined, all these factors may be driving where archaeological research has been done. Site pattering in Campania follows similar patterns; there are more sites in this region than Basilicata and Calabiria because there are more p eople, as well as large areas of plains and more development due to industry and productive agriculture. However,

PAGE 37

25 areas with relief where people have not settled are not usually investigated, and surveys within the region are rare, likely due to difficulties of surveying at eleva tion, and a focus on Classical A rchaeology (Arthur 1991). Basilicata and Calabria can be discussed together do to similarities in their geography. B oth have intense mountainous relief that dominat es the landscape, and both have less productive agriculture, little t o no industrialization and a smaller population than the other two provinces. Thus, it is not unexpected that the Middle Paleolithic research in these regions has been neglected. Surveys that have been undertaken are small and rare (e.g. Hodder 1984, Small et al 1998 ) and have not been productive in producing Middle Paleolithic sites. One could make the argument that the sites Campania, Basilicata and Calabria are lacking in Mousterian sites because Neandertals were present in these regions T his is not likely the case. First, Mousterian sites have been found at high elevations in Armenia (Pinhasi et al 2011) and Romania (Hoff ecker and Cleghorn 2000 ), and have also been found throughout the eastern Italian Alps (Milliken 2000 : 45 ). Italy, particularly southern Italy, was also a refugia during glacial periods in the Pleistocene (Hewitt 2000). Therefore Neandertals could have potentially occupied the area at varying de nsities thr time period. Finally, the presence of Acheulean sites in Campania, Basilicata and Calabria, at higher frequencies than Neanderthal sites in Basilicata and Calabria, (Lefevre et al 2010; Piperno and Tagliacozzo 2001; Radmili 1978), dem onstrate that Pleistocene sediments are not completely eroded away in these regions

PAGE 38

26 Past and modern politics combined with geographic variables can help explain why certain regions in so uthern Italy have experienced an extensive amount of research (Apuli a), others have a middling amount of research (Campania), and still others (Basilicata and Calabria) have been ignored. I t is important to consider why sites are patterned so that present day archaeologists can move past simplistic assumptions that conclud e all sites have been found in a region, or that sites are not present in a region. Further research should take into account these deficits in Mousterian research in southern Italy, and should be directed toward identifying potential areas within southern Italy that might be productive in terms of finding more sites to broaden our picture of Neandertals and Mousterian site locations in Italy. The refore, distribution of sites in southern Italy provides a potentially productive study area to explore how applications of ECNM can be used to model the southern Mousterian niche and potential spatial distribution of Mousterian sites during various climatic regi me s, as well as to determine if an assemblage composed of all known Mousterian sites can be modeled and connected to ideas of predictive site modeling to produce a potential site distribution map.

PAGE 39

27 CHAPTER II I ECOLOGICAL NICHE MODELING: HUMAN AND NONHUMAN SPECIES E coloigcal N iche M odeling is a method of species distribution modelin g which th e actual or potential geogr The modeling begins with the characterization the environmental conditions in which the species can live, or the niche, and then continues by identifying that suitable environment in other locations within the research area. A spatial distribution map is produced as a visual representation of the nic he on the landscape. This method was first developed to understand distributions and niches of non human species; however, recently Banks et al (2006) reconceptualized ENM as ecocultural niche modeling (ECNM) and applied it to human species to understand the niche of humans in the past. This chapter reviews these concepts, including their methods and their theoretical backgrounds, and also considers predictive archaeological site modeling as another method of modeling distributions of past peoples. Unlike ENM and ECNM, which seek to model niche, predictive archaeological site modeling seeks to model the distribution of sites on the landscape. Despite differing theoretical goals, all three methods are related through a use of simila r inputs and through a sm iliar modeling process T his section will review how ENM, ECNM, and predictive site modeling are connected. By connecting these three methods, this study aims to interpret ecocultural niche outputs as both potential site distributions and potential past sp atial distributions to explore productive areas for future research in southern Italy, and to explore Mousterian site distribution s during glacial and interglacial periods in southern Italy, respectively.

PAGE 40

28 Development of the Niche Concept As implied in the name, ENM heavily relies on the niche concept, and as such, ENM cannot proceed without an explicit definition of the ter m (Araujo and Guisan 2006: 1679; Guisan and Thuiller 2005: 998; Pearson 2010 ). This section reviews the variation in definitions of niche and how these defintions have varied thorugh time. Species distribution map s have long been used by scientists to study the range a species occupies however prior to the development of ENM methods they were either shaded outline maps which were created by experts extrapolating between and beyond known sightings of species or they were simple dot maps depicting where the species had been observed. However, these methods tended to over predict or under predict, respectively the range of the species of interest (Anderson, et al 200 2 ) In an attempt to rectify this issue, ecologists began to look at the relationship between environment and the distribution of species (Soberon and Peterson 2005 : 1). This relationship was first Darwin (Chase and Leibold 2003), and t oday, Currently the of such species environment relationships [niche] represents the core of predictive geographical model ing in ecolog (Guisan and Zimmermann 2000 : 148 ), and is the ultimate goal of ENM and ECNM. Grinnell (1917 ) was the first to use the niche concept in a research program; he defined it as the set of environmental requirements in which a species can surviv e, and outside of which the species cannot survive Elto n (1927) elaborated on this, and defined the n iche of an animal as its place(s) in the biotic environment, its relat ions to food and definition of niche does not assume the organisms within the niche are passive; instead it views organisms as active agents in creating their niche through interactions and alterations of their physical environment ( Elton 1927: 11 ; Guisan and

PAGE 41

29 Thuiller 2005: 998 ). Hutchison ( i otic and b i otic variables, the conditions and relationships of which describe the environment in which an organisms lives (Figure 9). Along with this new definiti on, Hutchison (1957) also conceptualized and quantified the differences between the fundamental niche and the realized niche. The fundamental niche includes all the potential areas that could support the species in question, as opposed to the realized nich e, which is made up of the areas within the fundamental niche where the species is actually present Realized niche has the potential to be smaller than the fundamental niche due to competition between species or other confounding factors ( Guisan and Thui ller 2005 ; Soberon and Peterson 2005; Walton 2009). Finally, Chase and Leibold (2003) propose a refined definition of niche, which includes Grinellian and Eltonian definitions, as well as the fundamental a nd realized niche of Hutchinson. They define niche as satisfy its minimum requirements so that the birth rate of the local population is equal to or greater than its death rate along with the per capita effect of that species Definitions of niche in ENM depend on the research question(s) being asked, and what ENM is actually quantifying This paper draws on the Grinellian and fundamental definitions of niche, as discussed later in this section.

PAGE 42

30 Figure 9: Hypothetical Niche. Hypothetical depiction of a three dimensional value (three n dimensional hypervolu m e niche. The area within the cube represents the total available amount of each factor, while the area within the sphere represents the amount of each factor needed for a given species to survive, i.e. its niche. From Chase and Le ibold 2003:9 Applications and Role s of GIS technologies As ENM aims to quantity the relationships between species distribution and environment GIS technologies and associated software have been useful in the successful implementation of the method Use of GIS allows researchers to gather appropriate environmental layers representing real world conditions, and process them into the correct format for application in E NM and ECMN Additionally, GIS allows the projection of the models onto the landscape, and can turns jumbled collection of rules for where niche is located int o an actual, understandable visual r epresentation of niche by identifying areas of the environment where those rules apply. These GIS technologies have been applied to understand a w ide variety of research topics in biology including guiding field surveys to find known and unknown species (Bourg et al 2005; Guisan et al 2006; Raxworthy et al. 2003), predicting effects of climate change and species invasions (Berry et al 2002; Hannah et al 2005), exploring speciation mechanisms and

PAGE 43

31 species delimitation ( Graham et al 2004; Raxwo rthy et al 2007), comparison of paleodistribution and phylogeog ra p hy (Hugall et al. 2002), assessing disease risk (Peterson e t al 2006; Peterson, Benz and Papes 2007) and more (list from Pearson 2010 :62 and Guisan and Thi u ller 2005) Additionally, the ENM method has been modified as ECNM for application to human species through the m odeling of archaeological sites, reviewed in the Ecocultural Niche Modeling section later in the chapter. Theoretical Foundations of Ecological Nich e Modeling (ENM) At its most basic, ENM draws on environmental factors represented in spatial layers to produce potential habitats, and then identifies, or projects, those environments on to the landscape. The model does this by stacking environmental rast er layers, and then placing a layer with planametric (x,y) presence data on top of this stac k. Based on the environmental values at each presence point, the model is able to describe that species tolerance ranges which include the environmental and biological conditions that allow a species to survive in an area (Figure 10) It is this environmental envelope that defines the niche which is projected on the landscape to identify areas of high or low potential for species presence. Since the niche is bein g modeled, not just where presence points are found the model is able to extrapolate other areas of the study region which fall within the e nvironmental envelope Once the projection is complete, the environmental tolerances can be described with descriptive statistics such as the mean, mode, and standard deviation, and analyzed through methods such as jackknifing, to determine which are most important to a species ( Banks et al. 2011; Beeton et al. 2013; Walton 2009).

PAGE 44

32 Fi gure 10: Environmental Envelope. the value for each variable at points of presen ce data. From Walton 2009: 3 8. Given the various definitions of niche, researchers have often struggled with determining the correct definition to use, and have also struggled to determine which type of niche is being modeled thr ough the ENM and ECNN process. Resolving these issues has been the focus of many theoretical papers ( Araujo and Guisan 2006 ; Guisan and Thuiller 2005 ; Soberon 2007 ; Soberon and Peterson 2005 ). Soberon and Peterson (2005) review the theoretical foundations of the ENM method in an effo rt to clarify the issues above. They identify four different classes of variables which all effect niche, all which must be accounted for in order to compl etely describe the niche of a species These include abiotic conditions, biotic conditions, accessibl e regions for dispersal, and finally evolutionary capacity of populations (Sob eron and Peterson 2005: 2). Abio tic conditions are the physiological limits on species ability to persist in an area, including variables such as climate and physical environment while biotic conditions include those factors that make up the set of

PAGE 45

33 interactions with other species that can affect the species ability to maintain population. Accessible regions help to differentiate between the fundamental and the realized niche of a species. The evolutionary capacity of a species is important to address potential change in niche; however this is difficult to model, and is usually removed from modeling and assumed to have a negli gible effect for most purposes (Soberon and Peterson 2005: 2). Since evolutionary capacity is not of major interest to the modeling process, we can assume the species that is the focus of modeling will be present at the given point or area where the o ther three conditions meet (Figure 11 ). The first condition that must be met is region A, which includes the regions where abiotic conditions are favorable for species presence. Region B, or condition two, represents the suit of biotic variables, including the presence of needed resources a nd absence of competitors, and finally condition three, or region M indicates the region where species are actually able to access ; this can be constrained by factors such as mountain ranges (Soberon and Peterson 2005: 3). Region A, or the suite of abiotic variables, can also be thought of a s fundamental n iche, in the Hutchisonian sense, while region B, the suite of biotic variables, can be thought of as a niche in the Eltonian sense. The overlap between region A ) abiotic conditions are suitable for positive population growth of the species; and (2) required mutualists are present and the suite of competitors, predators and disease s (Soberon and Peterson 2005: 3). The int ersection of A, B, and M is termed P, or the perfect area (niche) where populations should be found. Soberon and Peterson (2005) argue that most ecological niche models produce results that describe some aspect of the

PAGE 46

34 fundamental niche, which is appropriat e for looking at large scaled models and overarching distributions. This is useful for applying ENM as ECNM to at Mousterian and other archaeological sites, as the biotic variables and human effect on the niche is extremely complicated both in modeling and in the production of representative environmental layers to be used in the modeling process. This study specifically aims to model fundamental niche, however because not all abiotic variables were considered, it likely only includes a portion of that niche Therefore, this study discuss es the potential fundamental niche of Neandertal and Mousterian sites as well as potential locations whe re we would expect to find sites within this niche. Figure 11 : Theoretical Foundations of Ecological Niche Models. It is the intersection of abiotic (A), biotic (B) and movement (M) in which the perfect (P) environmental conditions exist. A dapted from Sobe ron and Peterson 2005 and Walton 2009: 37.

PAGE 47

35 Geographic and Environmental Space Another aspect of the theoretical background of EMN is the concepts of geograp hic and environmental space. U nderstanding these aspects can help illuminate the type of niche being modeled and can help facilitate interpretation of results Geographic space i s the actual distribution of where a species is located. It is represented by the first panel in Figure 12 and typically is referenced using x and y coordinate data so that it can be located in the real world. T he pluses represent observed presence points the grey color is the actual occupied niche, and the solid lines represent the potential occupied niche. A and B represent considerations that can cause actual occupied niche to be smaller than potential occupied niche; these include factors such as a la ck of observed presence points in an area that species actually occupies represented by area A or areas such as B, w hich the species could occupy but does not due to movement constraints or interactions with other species. The environmental space of ENM is represented by the right panel in Figure 12, and is the species : 59) Environmental space can be considered similar and equitable to niche, and is represented by the solid black line surrounding area E as is called actual niche Also present in environmental space is the occupied niche, represented in grey in the right panel, (1957) realized niche, which is defined as the area of the actual niche where the species is actually living. As shown in the map, there are areas in this occupied niche where there may not be any presence points, area D in the figure and the occupied niche may not be the same size as the actual niche because of movement constraints and interactions between species. Occupied niche can also fail to represent actual niche if bias in presence

PAGE 48

36 point location prevents the modeled niche from predicting into areas which are expected to be o ccupied. The patterning of movement, occupied niche, potential occupied nic he, competitive exclusions, and location of presence points all alter how the enviro nmental space is conceptualized. This i s demonstrated by the difference s between the panels in Figure 12 and also how the projection of a species distribution model is represented back onto the landscape from its conception in environmental space. Therefore, biase s in the data such as sampling bias in presence point data, can have an adverse effect on the environmental space ENM s are able to produce. Additionally, b ecause of these two differing theoretical and actual spaces, the model can produce two different types of outputs ; the first type of output aims to verify the actual distribution of species, while the goal of second type is to produce potential habitats. To successfully interpret ENM outputs, these concepts must be kept in mind. Figure 12 : Geographic and Environmental Space. Illustration of the relationship between a hypothetical environmental space From Pearson 2010 : 60

PAGE 49

37 Genetic Algorithm for Rule Set Prediction (GARP) An algorithm typically performs the modeling aspect of the ENM process A large variety of algorithms can be used in ENM and ECNM ; the nature of the research question and the nature of the data being modeled both dr ive algorithm selection. For this particular research question, the Genetic Algorithm for Rule Set Prediction or GARP (Stockwell and Peters 1999 ) was chosen to model Mousterian site distribution because it is appropriate to the study at hand, and for consistency with previously published studies. When ENM was first developed and applied, most researchers used multiv ariate statistical analysis to determine (Stockwell and Noble 1992 ). Much of that inconsistency stemmed from unreliable and variable data both environmental layers and presence points make applications of multivariate statistical methods difficult (Stockwell and Noble 1992 ). This problem is exemplified in the logistical regression approach to species distribution modeling, where static r ules are produced that must be applied to all the data, regardless of the quality of presence point or type of environmental layer. In contrast, GARP, as a machine learning technique, is able to produce various differen t types of rules which can only be applied when certain conditions are met When the conditi ons of application are met by a subset of the data, the rule is applied. When they are not, the algorithm selects another, more appropriate rule (Stockwell and Peters 1999: 146). This method allows varying quality of data and varying types of data (i.e. bo th categorical and continuous) to be modeled together. The GARP algorithm is appropriate to this particular study area because of its demonstrated ability to extrapolate niche in to un sampled areas of the study region at high probability levels (Peterson Papes and Eaton 2007). A previous study that evaluated

PAGE 50

38 the strength of models, Elith et al (2006), ranked GARP as one of the lowest perform ing algorithms on the basis of area under the curve (AUC) values. However, this study was focused on the algorithms ability to successfully model heavily sampled regions, rather than the ability to project modeled niche into new areas. Peterson Papes and Eaton (2007 ) demonstrated that in cases where presence point sampling was not evenly distribut ed throughout the stu dy region, GARP was more successful at predicting new regions of distribution than other algorithms. Also, Peterson Papes and Eaton (2007) and Peterson et al (2008) review the use of the AUC value as a form of model evaluation, and conclude that AUC valu es are generally underestimated in GARP models due to the way the curve is produced. As this study seeks to project species distribution models into under sampled areas to predict where new Mousterian sites may be found, GARP is the appropriate choice. Add itionally, applications of ECNM discussed in the next section typically apply GARP; for consistency this study applies the same algorithm Ecocultural Niche Modeling (ECNM ) ECNM, or the application of ENM to human species, was first presented by Banks et al (2006), was developed from a series of workshops specifically organized to explore the application of ENM to humans and archaeological data. The method fo ECNM is based on the idea that ge and resulting environmental variability, and developed a wide variety of cultural mechanisms to deal with these conditions. In an effort to understand the influence of environmental factors on prehistoric so c i a l and technical systems, there is a need to establish methods with which to model and evaluate the rules and driving forces behind these human environment interactions et al 2006: 69) Just like predictive species modeling, ECNM combines the concept of the niche with geospatial modeling tools to predict the occurr ence of species on a landscape through

PAGE 51

39 georeferenced site locality data and sets of spatially explicit environmental data layers (Lozier et al 2009:1); however due to culture and its role in human adaptation to the environment, the niche is conceptualized as a both related to environmental and cultural variables, or eco cultural. Additionally, cultures are considered to be human adaptations to specific environments, and so differing cultures are assumed to have differing niches (B anks et al 2006). groups is critical to understanding the complex mechanisms that have shaped the interactions among genetics, linguistics, c ultural affiliation and climat e 2006: 69) The goal of the application of ENM to human environment interaction through ecocultural to model the ecology of human and hominin population in the p 2006:69). Typically, ECNM models reconstruct geographic distributions of archaeological defined populations by determining the ecocultural niches those populations inhabit. ECNM can be applied anywhere there is occurrence data of archaeological sites and reconstructed environmental data. It has been applied all over the world including Asia and North and South America, but its greatest florescence is in Europe. This is likel y due to a long history of archaeological and paleoenvironmental research that has created strong data sets and quantified many environmental variables, thus allowing robust implementation o f ENMs to human species. Banks (200 8) use ECNM to identity the habitable areas of Europe during the Last Glacial Maximum (LGM), as well as to compare the ecological niches and geographical

PAGE 52

40 distributions of the Solutrean and the Epigravettian. The results of the EMN match archaeological data of human extent during the LGM, and the models also suggest that the Solutrean and Epigravettien cultures occupied different niches and thus can be considered cultural adaptations to the environment. ECNM has also been used to demonstrate that differences between the Proto Aurignacian and Early Aurignacian is due to an ecological niche expansion between the two periods (Banks et al 2013). This study is the first demonstration of how a cultural adaptation was used to expand a human niche. Banks ) use ECNM to look at Neandertal and anatomically modern human populations in Europe during the Paleolithic. These researchers consider two hypotheses for the contraction of the Neandertal r ange, eithe r as a response to changing climate, or as a result of competition with expanding anatomically modern human populations. Modeling indicates that the second hypothesis is the more likely explanation. North and South American case studies have focused on Pal eoindian dispersal and adaptation. For example, Gilliam et al (2007) uses occurrence data of Clovis points in North America to identify the niche characteristics of where the Clovis culture is found, and then extrapolates that information to Asia in an attempt to determine the source areas of pre during th e late Pleistoc ene in order to identify the ab i o tic characteristics that best predict species 2013 : 1 ).

PAGE 53

41 Paleobiogeographical and Archaeological Considerations The theoretical background that applies to ENM also applies to ECMN. Terms must be defined and explicitly connected to ecological theory, including the definition of niche being used and the type of niche being modeled. However, there are some challenges that apply specifically to ECMN, including modeling the past, and biases with environmental layers and wit h archaeological presence data ECNM is based on the idea that the past can be modeled either by using environmental data reconstructed to approximate past conditions, or thr oug h hind casting where a model calibrate d with data for the present is used to predict the range of past species in modern environments ECNM must consider concepts of niche stability, niche into account the (Nogues Bravo 2009:522), however the ideas of niche evolution and ecological niche shifts demonstrate that niches are not always stab le. In ECNM, the st asis or dynamism of the niche allows interpretations of human adaptations to the environment in th e form of cultural adaptations, but they can also pose problems when attempting to model the past if they are not considered. Environmental predictors also po se problems. Not only does the researcher have to be aware of the type, amount, and relationship of climatic variables, in ECNM the researcher must also be aware that environmental variables can be inherently biased due to their generally coarse grained ou tputs of climatic scenarios generated from General and oceans of the Earth through the assumption of

PAGE 54

42 2011) including carbon dioxide levels, temperature, winds, cloud cover etc and run a large number of iterations until they stabilize and produce a picture of climate in the past. An additional problem that comes from the use of GCMs is the focus on ab io tic variables. As discus sed earlier in the chapter, biotic variables do play a role in species distribution but ECNM are forced to operate under the assumption that ab io tic variables explain most of the response. This problem can be overcome by selecting ab io tic variables that a re most likely to affect the species physiology and thus most likely to affect species persistence. In addition, the GCMs produce very coarse grained environmental data layers. To make these layers fit the scale of the data required for the research questi on, a process of down scaling is required, which is discussed in more detail in Chapter IV. The more down scaling required, less likely the modeled variables are accurate In ECNM, data must be even more closely scrutinized, biases that cannot be accounted for should be explicitly stated, and processes behind the derivation of variables should be explained. Data preparation is more complex in ECMN than ENM due to spatio temporal considerations; a rchaeological and fossil occurrence data inherently include mor e biases than non human and modern species occurrence data (Varela et al 2011: 452). The first bias relates to absence data. In ECNM, absence data is almost never available due to the fact that archaeological data is typically not collected in this manne r This causes spatial bias within data, with some periods are either over or underrepresented due to taphonomic processes. These same taphonomic processes that produce spatial bias also produce temporal bias; the older the remains the fewer that are gener ally present (Varela et al differences between survey efforts within and between countries difference s in the

PAGE 55

43 amount of attention paid to specific remains and time periods, an d difference in taxonomic identification. The combination of these biases can cause samples to fail to between the occurrence data and the predictors may offer a flawed p icture of speci es responses to the environment 2011: 454) Additionally, dating provides another source of imprecision within the data set; many o ccurrences of sites are undated, and archaeological dating has its own issues and biases (Banks et al 2006). All of these biases must be accounted for within the dataset before the data can be applied in an ECNM model. O ften times bias within the sample cannot be fixed, and researchers have to proceed knowing their model may be inherently fla wed due to incomplete knowledge. Predictive Site Modeling Predictive site modeling is another method to model the human environmental interaction, but this method, unlike ECNM draws on human behavioral ecology (HBE) as well as settlem ent theory as oppose d to the concept of niche. Additionally, instead of modeling species distribution, it is specifically aimed at modeling distribution of sites. Predictive site modeling and its theoretical background are important concept s to this study because the method d emonstrates that sites themselves can be predicted on the landscape. This provides the basis for the hypothesis that if the niche of modern day sites is modeled in ECNM programs, the resulting spatial distribution map will also be a site distribution map, which will facilitate the identification of productive potential areas of research.

PAGE 56

44 Theoretical Background S ites pattern against the landscape with respect to envir onmental and cultural variables T his patterning can be explained by e volutionary ecology, specifically the framework of HBE. the study of adaptive design in behavior, life history, and morpholo ( 144). In the framework of evo lutionary biology, behavior is when it tracks environmental as its prop ensity to survive and reproduce 145 ) More specifically, HBE studies the fitness relate d behavioral trade offs that humans face in cultura l diversity of the environment in ecological with an evolutionary pe HBE can be used to hypothesize and to model the conscious and unconscious decisions and tradeoffs humans make in specific natural and social environments in order to maximize or optimize their goals. These goals are usually cons idered to be survivorship or reproductive potential (Kelly 2007). As applied to understanding location choices of archaeological sites, HBE consideres the the decision of where to place a camp as based on tradeoffs between locations in the natural and soci al environment in order to place a site where it will be best promote survivorship or reproductive potential. These decisions involve a complex analysis of the interplay betwee n many different food resources as well as nonfood resources like water, firewood, terrain, other human groups, predation and more (Kelly 2007). In this way, explo ring why hunter gatherers place their sites within a region with

PAGE 57

45 respect to environmental variables, through the lens of HBE, can help permit reconstruct ions and evaluation of economic activities at a site ( Kelly 2007 ; Shermer and Tiffany 1985: 228 ). As an archaeological method, predictive modeling was first applied in the 1960s with the rise of the processualist framework; however its ultimate roots lie w ith the beginnings of settlement archaeology and in the work done by Gordon Willey (1953) in the Viru Valley. From this research, Willey concluded that settlement patterns not only reflect the natural environment, but also the technology of the culture, an institutions of social interaction and control with the culture maintained I n other words, settlement archaeology has the potential to aid understanding of the economic, social and political organization of past societies ( Trigger 2006: 377). These c oncepts of settlement archaeology combined with t he processual focus on modeling, quantitative methods, and importance of environment in site choice (Trigger 2006) stimulated the first statistical modeling of potential archaeolog ical site location to unexplored areas ( Kvamme 2006 ; Verhagen and Whitley 2012 ). This theoretical perspective in explaining site location choice is supported by various studies from around the world that patterns environment and environmental change ( Kellogg 1987; Fry et al 2004 ; Schermer and Tiffany 1985; Thomas and Bettinger 1976 ) 1981: 151), and because of this site patterns lend themselves to modeling. Site predictive modeling is based on the envi ronmental characteristics of sites and

PAGE 58

46 prior knowledge or hypothese s of where humans liked to place their site s A generalization of site location is produced via multivariate statistical methods and then other geographic locations where this generalizatio n exists are identified. These models are then able to provide information on interactions between variables, to demonstrate significance of relationships between variables, to test hypothetical interactions between variables, and to expand on the strengt hs of the relationships modeled. Fundamentally, in this process, sites are assumed to be nonrandomly distributed with respect to the environment (Kvamme 1992: 23). Again, these predictive site models, connected with potential distribution models of Mouste rian sites in southern Italy, allow us to consider outputs of the ECNM as potential site distribution models, not just spatial distribution models of potential niche Ultimately, due to ECNM modeling the niche of sites, these concepts are not as far apart as they may first appear. A Note on Terminology niche modeling and spatial distribution models are both widely and often interchangeably used in the literature; spatial distribution models can be defined as the projection of the ecological niche onto the landscape (Franklin 2009), while ENM and ECNM is the process of producing descriptions of the niche. M any authors refer to ENM as a subse t of the former. By using concepts of spatial distribution modeling, a wider range of questions can be asked using the same processes and methods, and a visual map can be produced (Guisan and Thuiller 2005). Throughout this paper, the term ECNM is used to discuss the actual modeling process, as well as to highlight the theoretical basis of the model. The resulting outputs, or projections, are

PAGE 59

47 typically referred to as potential spatial distribution models However, when the modern day niche of the site is be ing modeled, and when the ECNM outputs are being connected to predictive site modeling, the term potential site distribution model is used. Extra Theoretical Considerations All three of these methods for reconstructing the past have been subjected to vari ous critiques that most ly stem from postprocessual concerns T his section reviews these critiqu es and refutes some of the post processual claims against using modeling to understand the past. Inductive versus deductive modeling is discussed, as well as the use of social versus environmental variables, and finally the benefits of modeling in archaeological research are reviewed. In duc tive versus Deductive M odeling ECNM and predictive species modeling are often critiqued for starting with data, and then applying inductive modeling to build hypotheses, rather than beginning with archaeological theory, then formulating a hypothesis bas ed on that theory, and finally using the model to test that hypothesis or deductive modeling (Kvamme 2006: 13; Vergen and Whitely 2011). Authors have argued that t hrough deductive modeling, social variables can be accounted for, such as viewsheds as well has how environmental variables like topography and site location interact with the ranges of human exploited species to explain what factors drive where humans place their sites on the landscape. This is considered to be in opposition to inductive modeli ng which typically focuses on environmental variables, and only considers culture after the modeling is complete. The ly defined local that acts as a medium for action and is part of human experien

PAGE 60

48 43). Thus the deductive model is explicitly cultural, and tends to focus on regional biotic variables, which ties it to ideas of Eltonian niche. These similiarities mean deductive modeling suffers from the same sorts of issues as attempts to model Eltonian niche in ENM ECNM and site predictive modeling including difficulty of producing raster representations of complex, interconnected biotic variables, compounded with t he issues of modeling cultural variables. In recent years it has been demonstrated that the extreme dichotomy between deductive and inductive modeling is mostly a function of how each is defined. In practice, a combination of deductive and inductive model ing is often applied, and it can be argued that the decision to model, as well as the selection of environmental variables comes from a body of theory (Kvamme 2006). In this specific case, deductive reasoning provides the under lying framework; the study d raws on HBE to explain why sites are patterned across the landscape, and then uses the theory to formulate a hypot hesis that Mousterian sites in s outhern Italy should be patterned against specific environm ental variables because Neandertals selected locati on based on their environmental envelope Then inductive modeling is applied to describe the niche, including which variables have the most effect on the niche, and to produce a potential site distribution of that niche projected onto the landscape. In th is way, deductive and inductive reasoning are used in a recursive manner to improve al l parts of the modeling process. Finally deductive and inductive reasoning can be complementary becuase once important ab i otic variables have been defined through the in ductive process, more complex biotic and cultural va riables can begin to be modeled through the deductive process

PAGE 61

49 Social versus Environmental Variables Although site location decisions involve a combination of social and environmental considerations, for various reasons, most hunter gatherer researchers use environmental variables to explain site location. First, t hey make the assumption that the most bas ic and the most important decisions people made were associated with the environment (Kolher and Parker 1986: 400). Finally, environmental data, in the form of maps, both paper and digital, is relatively easy to obtain, while social variables are not. This focus on environmental variables is the basis for postprocessual critique of predictive modeling; various critiques have discussed issues in predictive modeling including environmental determinism, ignoring archaeological concepts o f the individual and of the inductive process of the model itself, and the inability of modeling to accurately account for human behaviors or site formation processes ( Wheatly 2000; Whitely 20 04a; Whitely 2004b ). However, in this case the nature and age of the sample m akes social modeling almost impossible. Chronological resolution in terms of dating Middle Paleolithic sites is not high. Many sites, particularly ones excavated in southern Italy before the development of radiocarbon dating are generally only assigned to a Wurm period (Wurm I: 100 70 kya; II : 70 50 kya; or III : 50 30 kya ), on the basis of zooarchaeological analysis. Because of this, the mod eling in this thesis is focused on trends and patterns in the record that are co nsistent through 100,000 years. A t sc ales this large individual and group preference about site location become s less important. A pplication of social variables to models of the distant past is also difficult because of the inability to reconstruct the whole settlement system; without this w hole picture, complete hypothese s about the systemic context of Neandertals and Mousterian sites are difficult. Areas with

PAGE 62

50 high vegetation, high depositional or erosional environments, and intensive human modification, like southern Italy, may never reach a level where a complete systemic context can be produced. In these regions it is typically the case that existing site records obtained through haphazard archaeological reconnaissance over the past century must be employed (Brandt et al 1992: 207) when d oing archaeological research. Because of all these factors, archaeologists focus on environmental variables when doing ECNM. The Power of Predictive Model s Despite post processual critiques, these modeling techniques can be very productive. Predictive mod els are always simplifications of reality because s ubtle social determinants of location are probably at work in all settlement systems ( Kohler and Parker 1986: 401) but usual ly the models are run without consideration for these social factors Instead o f being a weakness, t his simplification of reality through modeling can be extremely useful to point out what archaeological remains fit, and which do not fit, with in the expected pattern. These outliers can point researchers in new and fruitful direction s of research including cultural considerations in an attempt to explain why the pattern is not consistent between all remains. Modeling also reduces a problem to its key elements and allows researchers to address and explain those elements one at a time; as most effective way s of eliminating problematic answers and identifying a nd ). Additionally, it is recognized that modeling was designed to look at large scale human environme nt al interactions which is the purpose of this study Overall, the constraints of modeling marginalization

PAGE 63

51 as the individual, symbol ism and culture, when deployed t o answer appropriate resea rch questions and used with appropriate data can be extremely effective. Hypotheses ECNM usually attempt s to explain the systemic context as o pposed to the archaeological; this project however, seeks to do both As discussed in the back ground chapter, large areas of s outhern Italy have been neglected in terms of Paleolithic archaeological research, for both historical and geographic reasons. Due to these processes, the picture of Mousterian sites in Calabria, Campania and Bascilicata, and to a lesser ex tent in Apulia are skewed toward coastal caves with deep stratigraphy. It is because of this incomplete picture of Mousterian sites in southern Italy that inductive modeling of the archaeological sites can be extremely powerful, both to produce spatial di stribution maps from modeling based on past environmental variables and site distribution models based on modern day environmental variables. Goal 1: To determine if modeling the niche of Mousterian sites against modern variables can produce an effective and useful model of potential site distribution and if that model identifies new areas of southern Italy that have the potential to be future sites of productive research. By drawing on predictive site modeling from other archaeological contexts and inter preting species distribution models from ECNM as site distribution models, all presence data can be used in the modeling process without having to consider issues of correct dating, variable excavation methods and other paleobiogeographical and archaeologi cal biases. This will allow a reconstruction of Mousterian site distribution with less error derived from bias in the inputs, as well as allow modeling with a large sample of sites. In this way, limited resources for survey in a difficult terrain can be

PAGE 64

52 applied in the most effective way, and a more complete picture of the archaeologica l context of southern Italy will be produced. Goal 2: To model the systemic context of the glacial and interglacial Mousterian niche in southern Italy as a heuristic device to explore where the southern Italian Mousterian niche may have been during warmer or colder periods. In addition, the modeling will provide a description of the Mousterian niche based on past ab io tic variables. I t is hypothesized that comparisons of the past niche to modern will support accuracy of pote ntial site distribution models produced from modern day variables. Goal 3: The final aim of this study is to explore the suitab ility of considering Neandertal and Mousterian sites as a homogenous unit whe n doing ECNM. The southern Italian modeled niche will be compared to other modeled niches of Middle Paleolithic hominins in Europe t al. 2008 ) to determine if modeling at different scales and in different locations pr oduces different results of the predicted Mousterian niche Overall, this study seeks to explore a region of Italy in which Paleolithic archaeological research has been relatively neglected, but which may hold great potential for adding to k nowledge of Ne andertal and Mousterian lifeways in southern Italy.

PAGE 65

53 CHAPTER IV RESEARCH METHODS The purpose of this research is to explore the potential of ECNM in modeling the southern Italian Mousterian niche and modeling potential spa tial and site distributions in s outhern Italy. To this end, methods of this study involved the gathering information and geographic coordinates of known Mousterian sites from s outhern Italy, determining which environmental variables would be most important in modeling the Mousterian ni che, gathering raster data layers representing those variables, and finally calibrating the algorithm and the computer program to implement the algorithm. Known Sites in Southern Italy Data on known sites in s outhern Italy was collected through an exte nsive survey starting list of Mousterian sites for the regions of Apulia, Calabria, Basilicata, and Campania. As many articles as possible were collected for each site through the Auraria Campus Interlibrary Loan System, however, the final list of sites used in this research is slightly different from the li st included in Milliken (2000) because some articles were not available for circulation or were located out of the country. Data on sites was collected and collated into a Microsoft Access table where each row repr esents a depositional layer ( Supplementary Materi als : Microsoft Access Database ), and u ltimately, a data set of 114 sites each with one or more depositional layers, was produced. S ubsets of this database were used in each modeling process (Table 1 ) Information recorded includes geographic coordinates, general description of lithic and non lithic tools, counts of lithic and non

PAGE 66

54 lithic tools, fauna, raw material, symbolism, structures, cultural attribution, site type, discovery, excavation and publication dates, and finally a list of all sources referencing the particular site The full table and references can be found in the supplementary materials. Geographic coordinate information was recorded from the articles themselves when possible. Additionally, official caving websites, ( Federazione Speleogica Puglia and Federazione S peleogica Campana ) provided coordinates for the location of caves throughout Apulia and Campania. Google maps provided the location for well known sites at points of interest. An a rchaeological database associated with The Stage 3 Project ( D avies 1996 2015 ) also p rovided geographic coordinates for some sites. Finally descriptions within the texts themselves often times allowed estimation of the placement of sites on the landscape. All x,y data was collected in the WGS 84 coordinate system when possible, and converted if not. Cave site locations are more secure and precise, and most, especially in Apulia, have error ranges of less than one kilometer while most of the open air sites were lacking specific coordinate data, and therefore coordinates are less pre cise, and error in site location can range up to four kilometers ( see Table 1 for source, accuracy, geographic coordinates and modeling subsets ) Known biases within this data set include a focus on stratified cave sites, due to historical trends in Italian archaeological research (Chapter I I ). This could lead to the fundamental niche being under predicted in environmental space because of the range of sites is not wide enough; because of this known presence points were compared with the predicted niche to confirm that the model is successfully extrapolating to new areas (see Chapter VI for discussion) Additionally, to account for the variation of accuracy in

PAGE 67

55 known site location the model was produc ed at a one kilometer and a four kilometer resolution with modern variables, and at a four kilometer resolution for glacial and interglacial models. Table1 : Known Sites in Southern Italy Geographic coordinates of each site, subsets of sites used to prod uce each model, including number of layers if applicable source of geographic coordinates, region the site is from and site type. Supplemental materials abbreviated as Supp. Mat. Continued on next page.

PAGE 68

56 Table 1 Cont

PAGE 69

57 Environmental Data Layers Environmental data layers were collected in raster format from various open sources at both one kilometer resolution, or 30 arc seconds, and four kilometer resolution, or 2.5 arc minutes. A raster file is composed of cells and e ach cell is given a value, w hich holds the data that the particular raster file represents. The resolution of the raster is based on the size of the cell; a 30 arc second raster means that each cell represents one kilometer of land, while a 2.5 arc minute raster means that each cell represents approximately four kilometers of land. Smaller resolved rasters are more accurate to real life as they make less generalization about the data they are representing, but depending on how the raster is being used larger cell rasters are often req uired. Therefore, because ECNM modeling must be done at the scale of the presence data, or the modeling runs the risk of predicting the Mousterian niche into unoccupied areas, both of these resolutions were necessary to model the set of presence p oints wit h a smaller error range and those with a larger All processing of raster layers was done in arcGIS 10.1 (ESRI), and all layers were converted to a raster format if necessary, clipped to the same extent, projected to the same coordinate system, and process ed to the same cell size, all of which is required by the modeling program. Any resampling or reprojecting of raster data sets for continuous data used the cubic convolution method, which determines the new value of a cell based on the fitting of a smooth curve through the values 16 nearest input cells. The nearest neighbor method, which reprojects rasters based on the nearest neighbor, was used for categorical data.

PAGE 70

58 Most data was downloaded in the World Geodetic System (WGS) 84 geographic coordinate syste m and data that could not be downloaded in this coordinate system were converted to WGS 84. All the data layers were projected into a custom projection Italy Conformal Conic, created in arcGIS using guidelines outlined in Price (2012). This projection is based on the Lambert Conformal Conic projection, which preserves area, shape and distance, and is used for regional mapping of smaller countries in middle latitudes. The datum and geographic coordinate system of the Lambert Conformal Conic projection is WGS 84, which is ideal since most of the layers did not have be projected into a new coordinate system, which can significantly decrease accuracy. A custom projection was created because Southern Italy does not fit easily into most commonly used projections; the region lies within three UTM zones, and the Mario Monte projections div ide the country into east and west sections rather than north and south. With this custom projection, there is no distortion of coordinate values, and slope and other raster values could be correctly produced. A suite of environmental variables which produ ce patterning in archaeological sites by limiting the environment in which humans can survive were included in the modeling process, wi th a focus on limiting factors (Table 2). These limiting factors are defined as factors which control species eco physiol ogy, like tempe rature, water, and soil composition, as opposed to disturbances, all types of perturbations affecting environmental sys tems, natural or human induced, or resources which are defined as all compounds that ca n be assimilated by organisms (Guis an and Thuiller 2005: 994). These abiotic layers were also chosen to facilitate the modeling of the fundamental niche in the Grinellian sense. Additionally a literature review of previous ECNM demonstrates that

PAGE 71

59 these types variables are most commonly deplo yed when modeling the niche of human species (Chapter II). Previous studies have identified some layers that seem to have a large effect on human niches, including elevation, access to water, temperature and rainfall. This study made sure to include these types of layers, with the addition of other layers that describe m odern environmental conditions which may contribute to where Mousterian sites are found on the southern Italian landscape today (Banks et al 2011 ; Brandt et al 1992; Kvamme 1992; Wise 200 0). Finally, due to limitations of past environmental reconstructions, and the difficulty of producing such models, only a subset of the layers discussed below were used in modeling the glacial and interglacial Mousterian niche.

PAGE 72

60 Table 2: Environmental La yers List of all environmental layers used in modeling, their source organization, and website from which the layer was obtained. Data Layer Source Website Elevation Bio clim http://www.worldclim.org Aspect Bio clim http://www.worldclim.org Slope Bio clim http://www.worldclim.org Flow Accumulation Bio clim http://www.worldclim.org Flow Direction Bio clim http://www.worldclim.org Distance from Populated Places Geo comm http://data.geocomm.com/catalog/IT/index.html Distance from Water Mapcruzin http://www.mapcruzin.com/download free arcgis shapefiles.htm Geologic Age HYDRO 1K USGS https://lta.cr.usgs.gov/HYDRO1K Land cover European Environmental Agency http://www.eea.europa.eu/data and maps/data/corine land cover 3 Soils European Soil Datab ase http://eusoils.jrc.ec.europa.eu/ESDB_Archive/ESDB/Index.htm Annual Mean Temp Bio clim http://www.worldclim.org Mean Diurnal Range Bio clim http://www.worldclim.org Isothermality: modern Bio clim http://www.worldclim.org Temp Seasonality Bio clim http://www.worldclim.org Max Temp of Warmest Month Bio clim http://www.worldclim.org Min Temp of Coldest Month Bio clim http://www.worldclim.org Temp Annual Range Bio clim http://www.worldclim.org Mean Temp of Wettest Quarter Bio clim http://www.worldclim.org Mean Temp of Driest Quarter Bio clim http://www.worldclim.org Mean Temp of Warmest Quarter Bio clim http://www.worldclim.org Mean Tem of Coldest Quarter Bio clim http://www.worldclim.org Annual Precipitation Bio clim http://www.worldclim.org Precipitation of Wettest Month Bio clim http://www.worldclim.org Precipitation of Driest Month Bio clim http://www.worldclim.org Precipitation Seasonality Bio clim http://www.worldclim.org Precipitation of Wettest Quarter Bio clim http://www.worldclim.org Precipitation of Driest Quarter Bio clim http://www.worldclim.org Precipitation of Warmest Quarter Bio clim http://www.worldclim.org Precipitation of Coldest Quarter Bio clim http://www.worldclim.org Modern Environmental Data and Mousterian Site Location To investigate G oal 1, as discussed in the Hypothese s section (Chapter III) t his study applies modern environmental data to model the present day niche of Mousterian sites in southern Italy. This is in contr ast to previous application of the ECNM methods which investigate the niche of humans through environmental resc onstructions of the Pleisotcene (Banks et al 2006; Banks et al 2011; and Kageyama et al 2008; Beeton et al. 2013 ). Through this application of modern environmental data to model the modern niche of Mousterian sites, it is hoped that the

PAGE 73

61 modeling process will produce a theoretical niche of modern day site location, as w ell as a site distribution map of where Mousteri an sites are expected to be located in Southern Italy. Modeling with modern variables also avoids potential error and bias in ECNM in two ways. The first is a removal of error associated with using reconstructed environmental data layers. Past variables ar e generated through G CM s, and t here are various circulation models, all of which produce slightly different outputs as a coarse resolution (200 100 km). Additionally, each downscaling method to convert the outputs of the GCMs from 200 100 km resolution to 1km or 4km resolution is different, and h as the potential to produce multiple outputs from the same inputs. GCMs also require modeling of the atmosphere of large regions in the past to produce estimations of environmental variables. Modeling the atmosphere requires considerations of complex interactions between wind, heat transfer, radiation, relative humidity surface hydrology, levels of carbon dioxide and many more factors, all of which must be reconstructed to the past. While techniques have been successful, they still likely contain sources of error, particularly when they are downscaled. Modern day environmental layers do hav e their own accuracy problems, especially at larger raster resolutions, however they avoid many of the assumptions which researchers are forced to make to model past environment. Therefore, using modern variables, and interpreting the output as where sites may be found on the modern day landscape as oppo sed to the niche of Neandertals based on Mousterian sites, a source of error in ECNM can be avoided. Also, by using modern day variables, the whole suite of southern Italian Mousterian sites can be modeled together removing some bias in occurrence data while

PAGE 74

62 preserving sample size. Previous studies, ( Banks et al 2006; Banks et al 2011; Banks 2008; Beeton et al. 2013) have model ed small slices of time to look at the niche of a particular culture. Therefore, they rely on absolute dating methods to confirm that each site included in the modeling process is actually within the time period they are studying. Because of the difficulties associated with dating many of these Pleistocene sites, sample sizes u sed in ECNM usually end up smaller than samples used in ENM. sites which have absolute dates tend to be cave sites. While modeling the concept of the niche allows ECNM to avoid only modeling the location of potential cave sites, it can still produce a bias in the modeled niche. By using modern variables, and explici tly modeling potential locations where sites may be found on the landscape in the present day, this study is able to use all the present points in southern Italy, which increases sample size, and removes some bi as by including open air sites in the dataset Elevation, Slope, and Aspect Elevation has been demonstrated to be an important factor that can limit archaeological site l ocation (Kvamme 1992; Wise 2000 ), and most ECNM and ENM takes this variable into account. Closely associated with elevation is slo pe, also an important variable in site location since hunter gatherers typically prefer to camp in areas with low slope values (Kvamme 1992:25). Slope also increases erosion, so sites on slopes have a larger chance of eroding away. Aspect plays a potential role in determining site location, as various aspects can help control temperature at a site (Kvamme 1992: 26). Slope at a one kilometer resolution was downloaded fr om Data Downloads section of the USGS HydroSHEDS website (HydroSHEDS 2010). This data set was

PAGE 75

63 produced Radar Topography Mission, and has gone through automate d processing to make the rasters more effective at mode ling hydrology by the Conservation Society Program of the World Wildlife Fund (HydroSHEDS 2010). At this resolution, data must be downloaded by tile to reduce file size, and so tiles n35e015_dem_bil.zip, n40e010_dem_bil.zip, and n40e015_dem_bil.zip were downloaded from the Europe, Southwest Asia section of the w ebsite. These tiles were then merged into one raster in arcGIS using the mosaic to new raster tool (Data Management tools > Raster > Raster Dataset > Mosaic to New Raster). Then, using the Project Raster Tool (Data Management > Projections and Transformati ons > Raster > Project Raster), the raster was projected into Italy Conformal Conic, using cubic convolution resampling. Finally the Extract by Mask Tool ( Spatial analyst > Extract > Extract by Mask) was used to constra in the extent of the raster to s outh ern Italy. From this elevation raster, slope in degrees, or rise over run, was produced, in addition to an aspect data layer which is essentially slope direction, and describes the direction of maximum rate of change in elevations between each cell and it s eight neighbors. The Slope and Aspect tools ( Spatial Analyst > Surface ) were used to produce each respective layer. Elevation at a four kilometer resolution was downloaded from the BioClim website (BioClim) This raster layer was also derived from the Shuttle Radar Topography Mission, and the same processes and tools, without the Mosaic R aster Tool since this data did not need to be downloaded in tiles, were used to reduce the ex tent of the elevation

PAGE 76

64 layer to s outhern Italy as well as to project the dat a. Again, from this dataset, slope and aspect were produced through Spatial Analyst tools. Flow Accumulation and Flow Direction Water is another important environmental variable in determining where hunter gatherers placed their sites, and previous studie s have included these variables ( Banks Errico, Peterson, Kageyama et al 2008 ; Banks et al 2008 ; Brandt et al 1992; Kvamme 1992 ; Wise 2000: 26 ). Therefore, flow accumulation and flow direction layers were used to model how the Mousterian sites relate t o water in the study area. Flow direction and accumulation at a one kilometer resolution were also available for download from the USGS Hydrosheds website (HydroSHEDS 2010). Each layer was clipped to the extent of the eleva tion layer using the extract by mask tool, as well as projected into the Italy Conformal Conic projection. The flow direction layer at the four kilometer resolution was produced in arcGIS from the elevation raster layer using the Flow Direction tool (Spat ial analyst > Hydrology) and then by using a combination of the newly produced flow direction raster layer and the elevation layer a flow accumulation layer was produced using the Flow Accumulation Tool (Spatial Analyst > Hydrology). Distance from Populated Places and Distance from Major Waterways Both of these layers are of interest when modeling and predicting the distribution of Mousterian sites in modern environments as opposed to past environments and so were not included in the glacial and inte rglacial niche modeling process, but were included in the modern model. Distance to populated places allows consideration of

PAGE 77

65 modern day modification on the landscape to be included in the modeling process. This layer was specifically inclu ded to explore the role of Mousterian site locations to areas of dense modern human occupation or in other words, if sites were more likely to be found near or further away from cities and densely populated towns. The Distance from major waterways layer i s important in t erms of erosion and deposition processes that could prevent sites from being found in moder n day contexts (Banks et al 2011; Brandt et al 1992: 273; Kvamme 1992). The populated places network was downloaded in vector format from GeoCom munity (GIS Data Depot 1995 2014) ; it is available for download on the Italy nationwide data administration and political boundaries page of the website. To remove the populated places polygons, from the network of other vector files, the layer of interest was first saved as a shapefile. The major waterways vector polyline data set was downloaded from MapC ruzin (Free GIS Shapefiles, Software, Resources and Geography Maps 1996 2015) which provides free GIS shapefiles. This specific layer was produced from open street map, and can be found in the Italy category in the Free Shapefiles page of the website. Both the distance from populated places and the distance from major waterways l ayers were created with the Euclidean Distance tool (Spatial Analysis > Surface tools). This tool produces a raster from a vector layer by calculating, for each cell, the Euclidean ver depending on the shapefile layer. While each raster was created, the cell size was set to the same resolution as the four kilometer or one kilometer elevation layer, and after creation each raster layer was clipped to the extent of the elevation raster

PAGE 78

6 6 Geologic Age Geologic age of the underlying bed rock is important to determine where cave sites might be found, and is also important as a general environmental description that could constrain location of other sites. Values of geologic age are generaliz ations of the original UNESCO age classes by the U.S. Geoglogical Survey ( Appendix A ). This raster is derived from a vector layer of geological age downloaded from the HYDRO 1K dataset, which can be found at the USGS Earth Explorer interface under digital elevation (HYDRO1K 2012) It is part of a raster and vector data set that was developed at the United States ation and Science (EROS) Center, and is available from the HYDRO 1K dataset. To process this layer, it wa s first converted into a raster using the Polygon to Raster tool (Conversion tools > To Raster) After conv ersion, the extract by mask tool was used to convert the raster to the correct cell size, correct extent and correct projection, all based on the ele vation layer both of the one and the four kilometer resolution. Land Cover The land cover raster layer was obtained to test if modern day land cover contributes to the location of present day sites on the landscape The layer was developed by the Corine Program of the European Environment Agency and processed by the European Topic Centre on Land Use and Spatial Information at a one kilometer resolution (CORINE Land Cover) The data was downloaded in the EPSG: 3035 coordinate system, and converted into the custom projection using the Project Raster tool. Since this raster was downloaded at the one kilometer resolution to begin with, it only required a clip to the correct extent for use in the one kilometer modeling process.

PAGE 79

67 Using the extract by mask tool and basing all changes on the four kilometer elevation layer dataset, the raster was also clipped and the cells were resized to the correct resolution. Because the resolution of the layer went from one kilometer to four kilometers, the details and the acc uracy were decreased due to the inherent nature of raster data sets. Additionally, and unavoidably, the coastline details were lost in this transformat ion, meaning that many presence points were lacking environmental information for this layer. If sites la cking land cover data could be moved and still remain within the four kilometer error range and resolution, then they were moved; however, many could not be moved. Therefore, models that were run with this environmental layer had a smaller sample size beca use the modeling process leaves out site s that have null data values. Soils This raster layer comes from the European Soil Database The data at one kilometer resolution is available without charge for noncommercial us e, in the ETRS: 3035 projection ( Eur opea n Soil Database 1996 2014; Liederkerke et al 2006 ; Pangos 2012 ). Values and soil types are derived from the World Reference Base for Soil Resources, which is the current international standard taxonomic soil classification system. As this data set was downloaded at one kilometer resolution, this layer was processed to the correct projection and extent using the extract by mask tool for the one kilometer modeling. For the four kilometer resolution, the data layer was converted to the correct projection, cell size and extent using the extract by mask tool. Again, like the land cover layer, this raster faced unavoidable issues with a decrease in coastal resolution and a

PAGE 80

68 considerable number of sites that could not be moved to the raster layer within the fou r kilometer error range; therefore, like the land cover layer, models created with this layer use a smaller subset of presence points. Modern Climate Climate variables were downloaded from the BioClim website (BioClim) They consist of a suite of 19 clima tic variables specifically produced to provide biologically meaningful data for ENM and are derived from monthly temperature and rainfall data. The bioclimatic variables represent an nual trends (e.g., mean annual t emperature, annual precipitation) seasonality (e.g., annual range in temperature and precipitation) and extreme or limiting environmental factors (e.g., temperature of the coldest and warmest month, and precipitation of the wet and ( BioClim ). These climate layers include d ata compiled by the Global Historical Climatology Network (GHCN), the Food and Agriculture Organization of the United Nations (FAO), the World Meteorological Organization (WMO), the International Center for Tropical Agriculture (CIAT) and various othe r reg ional databases. See T able 2 fo r a complete list of variables within this dataset. This data was available at both a four kilometer and o ne kilometer resolution (Hijman et al 2005). After downloading they were each projected to the custom projection and clipped to the extent of the study area using the elevation layer. Past Climate This same suite of biological ly meaningful climatic data has been produced for various times in the past, also available on the BioClim website. Of interest to this study are the models for the Last Interglacial (LIG) Period (140,000 120,000 years ago), and the Last Glacial Maximum (LGM) (~22,000 years ago). These past climate models are

PAGE 81

69 downscaled c limate data from simulations derived from Global Climate Models (CGMs), and were produced through the Coupled Model Intercomparison Project Phase 5 (CMIP5). Generally, GCMs simulate weather in the past through the assumed atmospheric concentrations of gree nhouse gasses; unfortunately, due to the complexity of modeling and limited computer memory, most modeling takes place at approximately 2 or 3 degree raster grids (200 100km) Therefore for modeling at smaller scales the information must be downscaled. The re are also a variety of downscaling models; however all of them combine modern climatic data with the CGM outputs to create more fine grained layers. (For more detailed information on methods see Hijmans et al 2005). Because not all presence points also had climatic information, the Mousterian niche in past climate s was modeled at a four kilometer resolution so that sites which are less securely located on the landscape could be included in the modeling process. This w as necessary to preserve an appropriately large sample size (Papes and Gaubert 2007, Pearson et al 2007). Therefore, the Mousterian niche in past climate was only modeled at four kilometer resolution. All 19 bioclimatic layers representing the LIG were pr oduced by Otto Bliesner et al (2008) at a one kilometer resolution and are available for download from the BioClim website. Once they were downloaded, they were processed in arcGIS to the correct projection, cell size and extent. In the case of the LGM, t he BioClim variables were available at a four kilometer resolution. Once they were downloaded, they were processed to the correct projection and extent with the same methods as the other laye rs

PAGE 82

70 Software: GARP with Best Subsets As discussed in Chapter II, ECNM uses presence data, environmental layers, and a modeling algorithm to produce a site distribution or a spatial distribution m ap. As visually represented by F igure 13, ENM combines presence data and environmental data, and t hen using the environmental values at each of the known sites, the algorithm generalizes the niche of the species being modeled. This produces a theoretical niche in environmental space, and finally, the niche is projected into geographic space, producing a raster output where areas with a high likelihood of allowing the modeled species to survive are indicated in red, while areas in which species presence is unlikely are modeled in blue. In other words, the model identifies the range of environmental varia bles in which a species can survive, the niche, and then identifies other locations on the landscape which are within this range, producin g a geographic projection of the past niche or site location. For the algorithm, t his project uses GARP ( Stockwell an d Peters 1999 ) to model the Mousterian n iche Additionally, this study uses the best subset protocol for GARP, developed and explained by Anderson et al. (2003) as a method to select the best algorithm outputs. All algorithms and models were implemented in openModeller software for ENM

PAGE 83

71 Figure 13 : Ec ocultural Niche Model ing (1) species occurrence points, (2) environmental layers, (3) modelling algorithm, (4) mo del in the environmental space, (5) model projection, with red indicating higher suitability values and blue lower suitability. From openmodeller.sourceforge.net. When producing species distribution model outputs, GARP develops a description of the potential niche in the form of a set of rules based on biological and geographic variables that describe the potential geographic dist ribution of a given species. This set of rules describing site locations is developed tions of environmental variables are considered like outside of the Mousterian range, are removed from the mode ling (Stockwell and Peters 1999; Walton 2009), just as d eleterious genetic combinations are removed from populations through natural selection. Before the modeling begins, the presence points are divided in half (GARP Desktop a

PAGE 84

72 pseduoabsence data by taking a random sample of background points in the modeling space, which then become part of the training data set. GARP takes the environmental v alues at each training dataset point, as well as the values at nearby cells, and based on these values generates one of four different types of statistical rules. The rules are atomic, range, negated range or logit types, and use an format. Atomic rules are the most simple as they use a single environmental value: Mousterian Range rules are very similar to atomic, however they deploy a range of data values to make statements such as: if elevation = 10 20, then Mousterian Negated range rules use the same principle as range rules however they use a range of variables that constrain Mousterian habitat to exclude areas from being considered part of the habitat; if elevation > 20, then Mousterian Finally, logit rules, which are an adaptation of logistic regression models, can also be applied This type of rule production is applied to determine the probability of Mousterian habitat being present at a particular point ( Bergen et al 2007 ; Stockwell and Peters 1999 ; Walton 2009 ). A first population of rules is produced based on the training dataset. Then, genetic concepts, like crossover, mutation and selection are applied to this populatio n of rules to produce new generations of rules, or iterations of the model. These new rules are then tested against the training data set. Crossover, where two rules exchange values or range of variables, and mutation, where a value is randomly changed to a new value, both

PAGE 85

73 introduce variation into the population of rules. Then, selection is used to decide which rules are most predictive. If a new variant of the rule is highly predictive, the algorithm continues to modify it until it is as predictive as poss ible. New iterations of the model, alternatively thought of as generations of rules, continue to be produced until a fixed number of iterations is reached, in this case 1000, or until changes in the new iterations fall below a set percentage, in this case .01% (Banks et al 2011 ; Walton 2009; Anderson et al. 2003). Due to the elements of randomness in the modeling algorithm, even with the exact same inputs no two models will ever be exactly the same. Finally, the population of rules, which describes the nic he of the species, is extrinsically and then areas of similar habitat in the modeling space are identified. Overfitting is defined as when a model too closely follows the data at the expense of being able to extrapolate and identify areas of high probability of Mousterian pre sence in under sampled regions. Finally, a raster is produced which uses a color ramp to demonstrate areas with a high probability inclusion in the Mouster ian niche, as well as those areas with a lower probability of inclusion. A good model is able to both correctly classify known presence points as present, and is able to identify other areas in the study region that are included in the p otential niche (Des ktop GARP U ser anual ; Stockw ell and Peters 1999; Pearson 2010 ; Walton 2009). R esults of these extrinsic tests, as well as to compare it to other models, and are reported as receiver operating characte ristic (ROC) graphs, AUC values and omission and commission error values. ROC graphs illustrate the performance of a binary (presence/not presence in this case)

PAGE 86

74 classifier system and are created by plotting the correct classification rate against the false positive rate at various threshold settings, and the AUC value is produced from this graph. This value represents the probability that a random point that is chosen will be correctly placed in an area of the geographic space with a high probability of species presence. Therefore, if the AUC is .7, then there is a 30% chance that a known presence location will be incorrectly classified as absent. A confusion matrix is also provided as an outp ut of these extrinsic model tests; percentages of omission error, false negatives, and commission error, false positives, are reported. In modeling exercises like this one, where there is only presence data, only an omission error percentage is provided ( A nderson et al. 20 03; Stockwell and Peters 1999 ; Walton 2009 ), as there are no values with which to produce a commission error. Best Subsets Protocol The validation tests discussed in the previous section are extrinsic to the model; however, as discussed by Anderson et al (2003), applications of both extrinsic and intrinsi c tests produce the best models. This is that ridge differ dram atically in error composition as well as qualitative aspects of the geographic prediction with errors at one end of the ridge including a great deal of commission e rrors and the other of omission 2003:20) Therefore, the best subsets pr ocedure, which is a method of intrinsic testing, as described by Anderson et al (2003) was used in this modeling process, to make sure that outputs of the modeling (Figure 14)

PAGE 87

75 In these graphs, the X axis represents the commission index, or the percent of geographic space predicted as present, and the Y axis is the o mission error, or the amount of known presence points which are predicted as absent by the model. Panel A within Figure 14 demonstrates the range of models which the GARP algorithm can produce The smaller boxes demonstrate how different types of error affect the modeled niche; high omission error is inaccurate because known presences are predicted as absent, zero omission error with low c ommission index is not d esirable because it causes overfitting and zero omission errors but high commission index cause s over generalization. Panel B demonstrates that models with a higher than 10% omission error are definitively bad because they incorrectly classify known presence poin ts. The last panel C shows the fitness ridge produced by the model, where the middle region of this ridge includes the best models because they prevent overfitting and overprediction. Models within the green circle are all summed to one output spatial distribution map. The user of the model can set parameters that force the modeling program to only produce outputs within the desired range of internal omission error and commission index, or percent o f area predicted present These parameters include a number of runs, in this case 20 ( Anderson et al 2003 ; Banks et al. 2011 ; Walton 2009 ), 10% maximum acceptable omission error, and a 50% commission error. Therefore, the algorithm produces 20 models, all of which are under 10% omission error. Then, the 10 models with a commission index that are closest to the mean (50%) of the commission index are selected and the outputs ar e summed into one map. This selects models which strike the best balance between ov erfitting and overpredictin g, and so likely best represents the actual niche.

PAGE 88

76 Figure 14 : Selecting Best M odels with GARP Best Subsets Protocol F rom Enrique Martinez 2007. A. B. C.

PAGE 89

77 openModeller All modeling in this paper was performed with openModeller desktop software, which is open source and free software package, available from openmodeller.sourceforge.net. There are a wide range of algorithms which can be used for ENM purposes, and each time a new one is developed or updated it is typically released as a new software program that can only handle a certain format of data inputs and only produces a certain format of data outputs. The openModeller software was developed to solve these issues thr of handling different data formats and multiple algorithms that can be used in p otential distribution modelling 2011 : 111 ). T he use of this software allows for q uicker data preparation, all ows multiple model s to be run at once, and allows a string of models to be lined up in preparation to run. This decrease d user time spent preparing data and monitoring the progress of the program, so more experiments could be run in a shorter amount of tim e. Each model was run with the GARP with Best Subsets new openModeller implementation The algorithm run in openModeller is the second implementation of the GARP algorithm, based on the original code by David Stock well (Stockwell and Nobel 1992; Stockw ell and Peters 1999), and modified by Ricardo Scachetti Pereira to run in open source software.

PAGE 90

78 CHAPTER V RESULTS This research project was designed to test the efficacy of using ecocultural niche modeling to produce potential site distribution maps of Mousterian sites in s outhern Italy, or in other words, to model the niche of Mousterian sites based on modern day variable s. Most studies which model the Mousterian niche focus on describing and modeling the past niche (Chapter II) but this study also attempts to connect site distribution to broader theories of spatial distribution modeling through the use of modern data to model the niche of southern Italian Mousterian sites Additionally, this study seeks to explore niches and spec ies distributions of the southern Italian Mousterian sites in interglacial and glacial periods Data and Model P reparation Once all presence points were collected, confidences in accuracy were assigned to the planametric coordinates. High confidence refer s to those sites which have a conf idence of a few meters, medium confidence refers t o those sites which could have an error range of one kilometer or less, and low confidence refers to those sites which could have an error range of one kilometer to four kilometer s (Table 1). The subset of s ites classified as high or medium confidence sites were collected into one excel file, and all sites, high, medium, and low classification, were collated into another excel file. Each file wa s saved as a text (tab) delimi ted file with the appropriate layout and column names for modeling in openModeller Additionally, a subset of presence points was selected based on climatic information, either interglacial or glacial, and saved as its own file for

PAGE 91

79 modeling. As data was recorded for each Mousterian layer at the site of interest, each layer was treated as its own distinct presence point. Environmental data layers were processed to the same spatial extent, cell resolution, and custom projection, Italy Conformal Conic, based on the WGS 84 datum. For a more detailed discussion see Chapter IV The Desktop GARP algorithm with best subsets with Open Modeler implementation was used to model the data; hard omission error was set at 10%, iterations were set at 1,000, and conver gence was set at .01, as per previous ECNM studies ( Anderson et al. 2003; Banks et al 2011 ; Walton 2009). Selection of O utputs As a major goal of this project was exploratory data analysis, as well as testing the efficacy of using modern variables to mod el the Mousterian niche, approximately 30 models were run which produced 50 maps to explore how various sets of data, both environmental and presence, changed the model outputs. Not all of these outputs produced useful results, and so outputs that are reported on and discussed below were chosen based on a combination of highest AUC values, lowest omission error and a robust number of in put presence points (Pearson 2010 ; Peterson et al 2008 ; Phillips et al. 2006 ). Researchers have demonstrated that AUC values are not always the strongest methods with which to evaluat e models ; however, it is a more robust method to compare results of outputs that were produced with the same algorithm. (Banks, Peterson, Kageyama et al. 2008; Peterson et al. 2008 ) Additionally omission error was taken into account to avoid overfitting. Consideration of the number of inputs is important because smaller inputs have a greater chance of producing biased models, and model performance tends to decrease

PAGE 92

80 with very small s ample sizes. Various studies have indicated that model performance drops at n<10, 15, or 30 data points ( Banks et al 2008 Stockwell and Peterson 2002, Wisz et al 2008 ), especially with a 50% partition for training and model testing, where only half of the total presence points would be used to produce the model. Therefore, those subsets of data which dropped below 30 points were thrown out. Additionally, the large resolut ion of some of the models, four kilometers, means that some cells included multiple sites Each of the points in the same cell provides the model with the same exact environmental inputs, and so the largest sample sizes were chosen to provide the most vari ety in environmental inputs to the algorithm. Additionally, to confirm that these models are performing at a better than random chance an exact one tailed binomial test was performed on all outputs considered bel ow (Anderson et al 2002; Pears on 2010; Phillip s et al 2006 ). Using values produced in the confusion matrix of each output, the probability of having at least t (1 r ) successes out of t trials, each with a probability of a was tested, where t = the number of test data points, r = the omission rate, and a = the proportionally predicated area, or the percentage of cells present (Pearson 2010) All models predicted significantly (p<.01) better than random, which is also supported by AUC value s which are all greater than .5. Jackknife Test of Var iables Each set of data was run through the jackknife function in the openModeller interface. This was done to find the subset of layers which provides the highest AUC value and smallest omissi on error, and also to prevent overfitting of the model and thus allow for extrapolation to under sampled areas. This function runs the model the same number of times as there are layers, removing one layer each time and then reporting the

PAGE 93

81 effect of each layer removal on the model. Re sults are reported as a percent accuracy of the model when each specific layer is removed (Table 3). Layers that are more important, and have more effect on the model outputs, have lower percent accuracy values when they are removed from the modeling proce ss. Once a subset of significant layers was identified based on low percent accuracy values from the jackknife function, the same model was run again with only the subset of layers to explore if accuracy and usefulness of the model improved. If the model saw improvement in terms of decreased omission error and increased AUC value, this model was chosen as the most effective model and reviewed below. When it was not, the two outputs were compared. Due to the stochastic nature of the algorithm, which means t he same outputs will not be reproduced even with the same inputs; slight differences in model accuracy are not usually important. W hen two models are similar, as they are in all cases, they are both considered to determine how, if at all they differ, and w hat the cause of the differences may be, through both the color ramp raster and through a fixed thresholding of the outputs, where a value of 50 or higher is considered as part of the Mousterian niche (Pearson 2010 ).

PAGE 94

82 Table 3: Jackknife of Environme ntal Layers : Accuracy of the each model without the specific layer. Bolded numbers are the important layers that were modeled as subsets Stars indicate the most important environmental layers to each model. 4km 1km Glacial Interglacial Elevation 90.2439 80.0000 72.4138 77.4194 Aspect 85.3659 65.7143 79.3103 67.7419 Slope 87.8049 77.1429 79.3103 74.1935 Dist. to Populated Pl. 87.8049 68.5714 N/A N/A Dist. to Rivers 85.3659 80.0000 N/A N/A Flow Accumulation 90.2439 65.7143 68.9655 58.0645 Flow Direction 85.3659 80.0000 82.7586 77.4194 Geological Age 85.3659 68.5714 72.4138 87.0968 Land cover 87.8049 80.0000 N/A N/A Soils 82.9268 77.1429 N/A N/A Annual Mean Temperature 85.3659 68.5714 79.310 64.5161 Mean Diurnal Range 87.8049 71.4286 75.8621 58.0645 Isothermality: modern 92.6829 71.4286 72.4138 83.8710 Temperature Seasonality 92.6829 65.7143 72.4138 67.7419 Max Temp of Warmest Month 87.8049 71.4286 75.8621 51.6129 Min Temp of Coldest Month 90.2439 68.5714 72.4138 83.8710 Temperature Annual Range 92.6829 68.5714 72.4138 74.1935 Mean Temp of Wettest Quarter 85.3659 71.4286 68.9655 83.8710 Mean Temp of Driest Quarter 90.2439 80.0000 75.8621 61.2903 Mean Temp of Warmest Quarter 85.3659 77.1429 75.8621 80.6452 Mean Temperature of Coldest Quarter 85.3659 71.4286 79.3103 80.6452 Annual Precipitation 82.9268 68.5714 72.4138 83.871 Precipitation of Wettest Month 87.8049 68.5714 72.4138 61.2903 Precipitation of Driest Month 90.2439 71.4286 68.9655 80.6452 Precipitation Seasonality 85.3659 77.1429 82.7586 80.6452 Precipitation of Wettest Quarter 90.2439 74.2857 72.4138 80.6452 Precipitation of Driest Quarter 85.3659 62.8571 82.7586 80.6452 Precipitation of Warmest Quarter 90.2439 71.4286 79.3103 58.0645 Precipitation of Coldest Quarter 87.8049 71.4286 72.4138 80.6452

PAGE 95

83 The Niche Mousterian Sites in Southern Italy These first two model s were run to determine if considering all Mousterian sites from the region as one set of data can produce an acceptable model of the Mousterian niche based on modern variables, and thus produce a potential site distribution. This test was performed at both a four and one kilometer resolution to determine if the differing scales of data as well as the differing sets of presence point inputs produced a similar patterning in the potential distribution of sites. The Modern Niche at Four kilometer Resolution The four kilometer resolution model only includes the subset of variables identified in the jackknife as the most important, which includes aspect, distance to rivers, flow direction, geologic age, soils, annual mean temperature, mean temperature of wettest quarter, mean temperature of warmest quarter, mean temperature of coldest quarter, annual precipitation, precipitation seasonality, and precipitation of driest quarter (Table 3) Reduction of the variables increases the AUC value by 0 3, from .82 to .85 and decreases omission error from 1.4% to 0% when com pared to modeling done with all the variables. While modeling with the subset leads to improvement in the model, this improvement is still small, and so modeling outputs with all the variables are compared to the subset. Overall, the figures look extremely similar, especially when a fixed threshold at 50 was applied to determine where sites are present and where they are not (Figure 15) Major differences can be seen in the Tavoliere Plain which is a depositional plain with thick Holocene deposits (De Sant is et al 2010). This large gap is likely driven by a combination of climate and soils. Due to the Holocene deposits, no sites have been

PAGE 96

84 found in this region, which means that the suite of environmental variables present are interpreted by the model as be yond the range of Mousterian site inputs values or outside of the niche. As shown by the subset model, when the climatic variables are reduced, the model is able to better predict into this region. Using summary statistics within ArcGIS, it is possible to generally characterize the modeled site distribution output based on some of the more important variables. In this case, the two most important variables were soil and annual precipitation. Soil types from the region are varied (A ppendix A ), but the range of the modeled niche does not include soil types that are classified as towns. The most common soil type in the niche is eu tric cambisol; or a young soil with relatively little horizon development. Annual precipitation ranges from 46.5 centimeters to 94.0, out of a possible range of 44.9 to 116.5. Very low and very high values of annual precipitation are not conducive to predicting Neandertals and Mousterian sites as present. Interestingly, some of the lowe st values are in the Tavoliere P lain, which c ould help to explain the large gap in the model outputs.

PAGE 97

85 Figure 15 : Mo dern Niche at 4 km R esolution. Mousterian sites included in the modeling process are represented. The Tavoliere Plain is circled on the upper panels. The Modern Niche at One kilometer Resolution The one kilometer resolved outputs are again very similar (Figure 16) however a model produced with a the subset of important variables identified through the jackknife test including aspect, distance to popu lated places, flow accumulation, geologic age, annual mean temperature, annual precipitation, precipitation of wettest month, precipitation of driest quarter, temperature seasonality, minimum temperature of the coldest month and temperature annual range, s ees a decrease in AUC value from .84, as

PAGE 98

86 modeled with all variables, to .82. Omission error remains at 0%. The subset model is interesting because it predicts more area present in the peninsula of Calabria, which likely better represents actual Mousterian range given that the site of San Fran sico di Archi is located in that region (F igure 16 ). Jackknifing variables demonstrate that flow accumulation, precipitation of driest quarter and temperature seasonality are the variables whose removal decreases accuracy of the model the most. Precipitation of the driest quarter ranges from 3.5 centimeters to 11.9, with a mean of 7.2 centimeters and a standard deviation of 1.6. Values for the whole area reach up to 22.4 centimeters ; the model suggests that area s which are the wettest in the dry season are not suitable for the presence of Mousterian sites This pattern could be due to the fact that the highest values of rainfall in the dry season are associated with highest elevations, so altitude or another asso ciated variable are preventing Mousterian sites from being present in areas with the highest levels of precipitation. Values of flow accumulation, which is a measure of the upstream catchment area where the values represent the amount of upstre am area drai ning into each cell, range from a low of 1 to a high of 30,866 in the niche while the highest of the region is 35,810. However the mean of the niche is approximately 97. Therefore, in this model, sites are usually found present in areas with less catchmen t, which could driven by where Neandertals were placing their sites, or could be due to taphonomic processes if areas with high catchment destroy archaeological remains Temperature seasonality also has an eff ect on the patterning of sites. Seasonality, wh ich is calculated from the standard deviation of the lower temperature and the highest, multiplied by 100 in the niche ranges from 4935 to 6372. I n the region the highest seasonality is 6461, t herefore, the

PAGE 99

87 Mousterian niche is not modeled into areas that e xperience t he most extremes in temperature between seasons. Figure16 : Modern Niche at 1km R included in the modeling process. Other sites represented by black dots were i ncluded in the modeling process

PAGE 100

88 The one kilometer prediction layer was produced as a way to test the outputs fro m the four kilometer resolution to confirm that the model is producing consistent outputs at differing resolutions and with differing sets of presen ce points. In addition as a goal of this project was to produce a map which would assist survey effort s into under researched regions, the one kilometer output was produced as a more resolved prediction map to direct areas of future research. Overall, st atistical comparisons of niche overlap and niche breadth in ENMTools version 1.2 (Warren et al. 2009, 2010) suggest the one kilometer and four kilometer models are extremely similar. EMNTools is a freely downloadable software is meant to et al 2010: 607) and was used to calculate statistical comparisons between results. Niche overlap is D and by a measure derived from Hellinger distance called I (Schoener 1968) both these measurements are obtained by comparing the estimates of habitat suitabilit y from the output files produced the ECNM In other words, once an output raster layer with probability of site presence has been produced, with values r anging from 0 100, two outputs can be compared by prediction value at each cell to determine similarity. To apply this statistic to the one kilometer resolved output and the four kilometer output, the four kilometer output cell size had to be decreased to the same cell size as the one kilometer output in arcGIS. Results indicated that the two niches are similar. The D statistic was lower, at .58, while the I statistic suggested that the overlap of the models was .81 (Table 4). There are some differences in the model, particularly in the Tavoliere P lain and the penins ula of Calabria, which is what the statistics are likely representing Additionally, it must be remembered when comparing

PAGE 101

89 statistics that the lower resolution was produced at a four kilometer out put, meaning that there are nuances in the one kilometer resolved map which the four kilometer overlooks. Some of the difference between the values of the statistic outputs particularly in the D statistic, may also be due to this consideration (Warren et al. 2009). Table 4: Niche Overlap Statistics

PAGE 102

90 Like niche overlap testing, niche breadth testing also indicates that these two site distribution patterns can be considered similar Niche breadth is the variety of resources or habitats used by a given species. In other applications of this concept, the utilization of each resource within a species niche is determined, and then two species are compared, but for the purposes of determining similarity between ENCM outputs, ENMT ools compares values (ranging from 0 100) of the predictions of niche suitability that are produced from the ecological modeling process as raster files. So while these outputs are not directly comparable to statistics produced with these methods outside o f the smoothness of the distribution of habitat suitability scores in geographic space for two or 2010 ). Two values are produced as outputs from this test, for more details see Co lwell (1971) and Levins (1968). I n both cases the values were almost exact ly the same for the two maps, the B1 statistic was .76 for the four kilometer resolution, and .75 for the one kilometer resolution; the B2 statistic was .98 f or both the four and one kilometer resolution (Table 5). Additionally, the results of this test are maximized, or closer to 1 than to 0, which indicates a broad niche (Colwell 1971) Table 5: Niche Breadth Statistics B1 (inverse concentration) B2 (uncertainty) glacial 0.513412 0.953521 interglacial 0.613071 0.958943 4 km modern 0.76197 0.977571 1km modern 0.750516 0.983212

PAGE 103

91 A visual inspection of these two maps demonstrates similarity between the models as well as indicates areas which have great potential to direct future research in to the region (F igure 17 ). Although thresholding does provide a way to avoid overfitting ( Banks et al 2011 ), it is also useful to inspect areas of high probability of distribution as well as low proba bility areas predicted by the color ramp raster, so that research resources can be allocated in a scaled way to each region. An unexpected amount of high probabilit y cells are predicted in the e astern region of Basilicata, suggesting that future research in this area has the potential to be extremely productive. Additionally, it is helpful to have areas within the inner portion of Campania and the western portion of Basilicata that are identified as hot spots, so that the least productive areas can be avoi ded to save time and money in a survey process. This same process of identification of productive areas is also useful in Calabria. In all mountainous regions, overlaying these maps with other local maps representing factors like erosion, seasonality of la nd cover and other modern factors can direct where future research sh ould take place as well as identity areas where Mousterian sites may be at risk Finally, the four kilometer resolution map, and to some extent the one kilometer i dentify areas in the Tavoliere P lain where potential ground penetrating survey techniques could be most effectively applied to search for Mousterian sites in the region.

PAGE 104

92 Figure 17: Productive Areas for Future Research Identified. Circles represent areas where fut ure research may be productive in modern niche modeled at 4km (left) and 1km (right) resolution. The Niche of Mousterian Sites in the Last Glacial Maximum This test attempts to explore patterning of the southern Italian Mousterian niche in glacial periods. This model is intended to be a heuristic device to examine the Mousterian niche during colder climatic regimes. This is because, a s discussed in the methods section, all glacial sites were modeled together, without focu s on the nuances and fluctuations of climate withi n and between each time period. Additionally this model was produced with environmental variables reco nstructed to represent the LGM, which is a time when Neandertals were no longer present on the landscap e. Therefore, these model outputs cannot be interpreted as modeling the actual Mousterian niche but can be useful to discuss general trends during glacial periods

PAGE 105

93 T he subset of presence data used in this model was collected from the information gathered on all the sites of the region. Unfortunately, not all the sites had availab le climate data, so this model, was produced at a four kilometer resolution so that sites with less accurate planametric data could be include in the modeling process Sites were classified as glacial based on the climatic and dating information collected in the Microsoft Access Database (supplementary materials) A jackkn ife of available data layers indicated that variables which contributed most to model accuracy included eleva tion, geologic age, flow accumulation, precipitation of wettest month, precipitation of driest month, precipitation of wettest quarter, precipitation of coldest quarter, isothermality, temperature seasonality, min temperature of coldest month, temperature annual range, and mean temperature of wettest quarter ( T able 3) The site distribution model which was produced with thes e layers can be seen in ( F igure 18 ) and with an AUC of .98 and only 13.7% of cells predicted as present at a 50% threshold, seems to be suffering from overfitting. The model with a subset of important variables expands the presence of Mousterian sites further north, with only a slight decrease of AUC value to .95, a 0% omission rate, and 22% of cells predicted at a fixed threshold of 50 and so likely better represents the true distribution.

PAGE 106

94 F igure 18: Glacial Niche at 4km Resolution. Both models, particularly the model using all the variables (Figure 18) seem to be suffering from an inability to predict sites on the coastline exposed by falling sea levels This is likely due to a couple of confounding factors; first, Neandertals were not present at 22,000 years ago, and so no sites were produced during t he environmental conditions along the coast and second, sea levels were not as low during earlier glacial periods. To explore these factors further, the niche is s hown projected against a n elevation raster at

PAGE 107

95 80 meters below current sea level, which is an approximate reconstruction at the beginning of the Stage 4 glacial period (Figure 19 ) The raster was pro duced using the Spatial Analyst map algebra raster calculator, to produce a DEM where sea level is dropped to 80 instead of 0 meters. While raising s ea level reduces area which is unpredicted overall, the model is still not extrapolating the niche into the Gargano Peninsula. Because none of the sites use d in the modeling process are present along these exposed coasts the suite of environmental variabl es are not identified as within the range of Neanderthal environmental tolerance. As discussed in the environmental versus geographic space section, and d iscussed further in Chapter VI, this coul d be because the occupied niche is not fully representing the actual, or fundamental niche. Additionally, it could be due to h arsher climates during the LGM, and it is possible that Neandertals would not have been able to occupy these areas if they were present at 22,000 years ago and so Mousterian sites would not be present F uture modeling using more accurate glacial climatic data may expand the niche into this region.

PAGE 108

96 Figure 19. Glacial N nich e at 80m Sea L evel. Gargano Peninsula circled in white. Unlike the models produced to specifically model site distribution, modeling cold sites against LGM variables allows some discussion o f variables that may have been affecting the Mousterian niche in colder periods. Out of the variables that the jackknife identified as important, the three variables of f low accumulation, mean temperature of the wettest quarter, and precipitation of driest month are identified as most important. While values in the predicted niche include flow accumulations from 0 to the highest possible, 132, the most common values are 32 and then 16, and the least are 5 and 12. So, for the most part, summary statistics are indicating that the glacial the Mousterian niche is typically found in areas of low flow accumulation, which is consistent with the one kilometer model. Precipitation o f the driest month also affects model accuracy, out of a possible range of .7 to 4.9 centimeters; Mousterian sites are only present in regions of 1 cm to 4.1 cm. Therefore, it is likely again that areas with the driest and wettest dry

PAGE 109

97 month s are not suitable for Mousterian site presence While the driest does make sense, again, like in modern times, some of the wettest areas are associated with higher elevations, which could be preventing Neandertals from exploiting those areas. Finally, mean temperatures of the wettest quarter are also driving Mousterian site pattern ing. Values here range from 9 to 15 degrees Celsius out of a possible 5.3 to 16.3. Again, the Neandertals are not occupying the extreme of the range. Here, the mean is 4.6 and t he standard deviation is 2.5 degrees, so for the most part Mousterian sites are found in areas between 2.1 and 7.1 degrees in the wettest season of the year. Despite being modeled against cold variables, niche overlap (Table 4) between the modern four kil ometer niche and the LGM niche is high; the D value comes in at .79, while the I value is even higher at .91. This patterning is likely because southern Italy is a refugia area, so Neandertals were living in t he area through glacial periods. Also, Italy ha s a relatively small area which constrains where Neandertals could live. Niche breadth (Table 5) i n the LGM is lower, B1: .51, B2: .95 which supports the idea of Mousterian range contracting during colder periods; however, it must also be noted that the L GM had more climatic extremes than any period Neandertals actually occupied, so actual niche m ay have been broader. The Niche of Mousterian Sites in the Last Interglacial Period This last test models the niche of Mousterian si tes during interglacial periods in s outhern Italy based on climatic variables reconstructed to represent the LIG Again, this is a heuristic device to explore what the potential Mousterian niche might look like in s outhern Ital y during an interglacial period

PAGE 110

98 The su bset of presence data used to model these sites was produced from the tion from the access database Unlike the LGM modeling, where none of the sites were actually occupied during the climatic regime being modeled, s ome of the presence points were probably occupied during this time period, but due to coarseness of the dating and to preserve sample size, all sites with an interglacial classification were combined. A jackknife of available data layers demonstrated that the layers driving the potential distribution were aspect, flow accumulation, annual mean temperature, precipitation of wettest month, precipitation of warmest quarter, mean diurnal range, temperature seasonality, max temperature of warmest month, mea n te mperature of driest quarter ( Table 3 ). Modeling with only these variables produced a strong output with a n AUC value of .88 and a 0% omission error, in comparison to modeling with all the variables that produced a n AUC value of .9 8, with an omission error 1.6%. While th e AUC value did drop here, a value of .98 is extremely high, and so the model with all the variables may have been overfitting. The geographic range of the high probability areas in the outputs also suggests this; the model with all the variables does not predict any Mousterian presence in the Calabrian peninsula, while the m odel with the reduced variables does predict that Mousterian sites would be present in this region (Figure 20) The jackknife procedure ident ified four environmental variables that drive the di stribution more than the others; flow accumulation, preci pitation of the warmest quarter, mean diurnal range and the max temperature of the warmest month. Given that flow accumulation also drives patterns in the LGM and one kilometer models, this is not surprising. Values for this variable in southern Italy range from 0 to 170, but only go to

PAGE 111

99 140 in the niche. Within the niche, the majority of the values of flow a ccumulation are 0, and the mean is 4 It ma y be possible to say that Neandertals were choosing to place their sites in areas with low flow accumulation; however, taphonomy coul d be playing a role in eroding sites away from these areas, so it should not be assumed that sites were not present here. A gain, as the discussion from the geographic versus environmental space demonstrates, this could just be a function of the inputs, however it is important to note the different theoretical assumptions we can make when using diff erent subsets of data In s ou thern Italy, values for precipitation of th e warmest quarter range from 4.5 cm to 22 .7 cm. Th e range of Mousterian sites in s outhern Italy includes precipitation levels from 5.1 to 13.7 cm. In this modeled niche, the mean diurnal range, which is a value representing the mean monthly temperature times the maximum annual temperature minus the minimum annual temperature, ranges from 6.8 to 10.0, out of a possible 5.1 to 11.9. The maximum temper ature of the warmest month ranges from 29.0 to 35.6 degrees Celsius with a mean of 33, out of a possible 17.9 to 36.5 degrees Celsius. Again, like the other models, the two ends of the extremes are avoided for all values. Also v ery high values of precipit ation of the warmest quarter are found in the most northern area of the environmental raster layer, which is beyond the limits of the region being modeled so no presence points come from that region ; therefore Mousterian sites would not be modeled to the largest extent possible However, it does suggest, as does the distribution of modeling where red areas do not continue up to the edge of the raster, that southern Italy may have required a separate niche adaptation for Neandertals in the area. M ore testin g using the whole of Ital y is necessary to say for sure.

PAGE 112

100 Niche overlap tests (Table 4) between the modern four kilometer site distribution prediction and the i nterglacial predictions are extremely high, likely because interglacial periods in southern Ital y are very similar to modern climates, only slightly warmer and wetter as a comparison of environmental layers indicates (appendix B) and general climate reconstructions of the LIG suggest (Kukla et al. 2002). The I value is .97, and the D value is .92. These slight differences suggest that climates of the LIG did affect the Mousterian niche, altering it slightly from where the modern sites are expected to be located Niche breadth values are .61 (B1) and .96 (B2) (Table 5 ), and implcations of which will be discussed in Chapter VI in the context of the other measures of niche breadth. Figure 20 : Interglacial Niche at 4 km R esolution

PAGE 113

101 Comparison of the Niche Models The niches modeled by this study were compared to a previous study, Banks, rtal Extinction by Competitive The previous study examines how the Neanderthal and the anatomically modern human (AMH) niches change d through time as Europe was colonized AMH. term middle events separated by two Greenland Interstadials (pre H4, 43.3 4 0.2 kyr cal BP) the Heinrich event 4 (H4, 40.2 38.6 kyr cal BP) and the Greenland Interstadial 8 (GI8, 38.6 36.5 kyr cal BP). Through this modeling, the researchers identify the pre H4 Mousterian niche in areas with a mean temperature range of 1 to 12 deg rees Celsius and with annual precipitation of less than 109 5 cm per year in the first climatic period. Geographically, in this time period, Mousterian sites are modeled as occupying most of the region, but are predicted at lower levels in the south of Ita ly when compared to Eastern Europe and area s of France and Spain (Figure 21 ). In the H4 event, the Mousterian niche includes mean annual temperatures of 0 10 degrees Celsius, and less than 730 mm of annual precipitation. In this model, presence throughout southern Ital y is highly predicted (Figure 21 ). Finally, during the GI8 event, Mousterian range significantly contracts, which the authors conclude is due to competition with AMH populations. Here, the temperature range is 6 14 degrees Celsius, and annual precipitation remains less than 73.0 cm per year. In this model, southern Italy is predic ted as present at medium levels (Figure 21). Overall, temperature variables are the most important bioclimatic variables driving this model.

PAGE 114

102 F igure 21 The European Neandertal Niche Grid squares with 1 5 of 10 models predicting presence of suitable conditions are indicated in grey, grid squares with 6 9 models in agreement are depicted in pink, and squares with all 10 models in agreement are indicated in red. Archaeological site loca tions are indicated with circles. Panel A is pre H4, panel C is H4, and panel E is GI8. From ico, Peterson, Kageyama et al. 2008 : 3972

PAGE 115

103 Models of the past southern Italian interglacial niche produced in this study include annual mean temper atures that range from 12 to18 degrees Celsius, and precipitation values which range from 53.4 to116.1 cm per year. Additionally, the model identified precipitation of the warmest quarter as the most important variable. The g lacial niche in southern Italy is defined by annual mean temperatures which range from 5 to 13 degrees Celsius, and precipitation values which range from 36.7 to 66.9 cm per year with the most important bioclimatic variable s of mean temperature of the wettest month and precipitation of the driest month. The LGM niche modeled for southern Italy falls Kageyama et al. (2008) for the GI8 period, and is similar to the pre H4 and H4 periods, although the maximum temperature in the modeled Mousterian niche is higher in the southern Italian models than in the pre H4 and H4 European models. However, the interglacial niche includes temperature and precipitation values that are beyond the range of the niche modeled for all of Europe. These differences can be attributed to the models using slightly different climatic data from different time periods to produce the modeled niche Overall, identification of differences between these modeled niches is difficult, because the y each use differing sets of data and differing modeling scales It is particularly difficult when viewed in the context of this study where different outputs are produced with different subsets of presence points used in modeling, for example the four kilometer resolution and the one Kageyama et al. (2008) are modeling the entirely of Eur ope at a 50 km resolved raster, and use radiocarbon dating to constrain the presence point assemblage sites to a specific

PAGE 116

104 time period. In contrast, this study models a much smaller area, but does not attempt to parse out temporal differences between site a ssemblages. Despite these issues, comparisons can be made. The first comparison is between the variables identified as the temperature variables are identified as driving the m odel. This is likely due to northern Europe including a hard limit on where sites could be found due to glaciations, while southern Italy does not have the same temperature constrains, and so availability of water becomes important to accuracy of the model The second comparison is the geographic extents predicted present. These do differ, most obviously in the scale; this study models the niche at a one and four Kageyama et al. (2008) model the niche at a 50 kilometer resolution. Because of this, study is lacking (Figure 23), and some of the models seem to predict the Mousterian niche as present at low levels in areas o f southern Italy which, in this study, are associated with a high likelihood of Mousterian niche, most noticeably the peninsula of Apulia. However, these low level predictions are similar, if less resolved. A discussion of what these results indicate for the usefulness of considering spatial and temporal aggregations of sites for the modeling process will be reviewed in Chapter VI.

PAGE 117

105 CHAPTER VI DISCUSSION Overall, these results can help direct future research on Mousterian sites in southern Italy, as we ll as produce generalizations of the southern Italian Mousterian niche and the environmental variables that drive the niche patterning in the region. The model is a first step toward further in vestigations of both the archaeological and systemic context of Mousterian site distributions in southern Italy. Occupied Niche versus Actual Niche Before moving to a discussion of what the modeled niche suggests about Mousterian site locations, it is important to determine what type of nich e these models are representing, and how the type of niche produced may constrain interpretations of the models. The goal of this study was to model the fundamental or actual, niche in the Grinellian sense, using ab i otic variables However, it is possible that due to biases within the presence point data set, these models do not fully represent the actual niche, but instead model the occupied niche as defined by the presence point inputs (Chapter III; Figure 12). The Tavoliere Plain in all the output maps the northern extent of the output maps, and the Gargano peninsula and new coa stlines during glacial periods all hint toward the possibility that the full distribution of the actual niche is not being modeled. The Tavoliere Plain is one area where Mousterian sites would be expected, but are not predicted in the glacial, interglacial and modern one kilometer resolved models and predicted at low levels in the modern four kilometer resolution. Although this area does not receive much annual precipitat ion, it does have year round rivers, little elevation

PAGE 118

106 changes and fertile soil. Additionally, compared to the rest of southern Italy, it is relatively close to o ne of the only sources of high quality lithic raw material, which is found in the Gargano Penin sula (Milliken 2000). It is likely that Neandertals were exploiting this area, but Mousterian sites have not been found on the plain due to extensive deposition o f newer sediments from multiple rivers (De Santis et al. 2010) Therefore, the actual niche i s expected to extend into this region, but the model excludes this area due to b iases in the presence points. This same issue is present in th e glacial model in the Gargano P eninsula and along the coastline uncovered when sea level decreases (Figure 21) In all other models coasts are typically identified as high probabilit y areas of Mousterian sites and the Gargano Peninsula would have had raw material, forests, and year round water (Milliken 2000) Therefore it is likely that both these areas were within the Mousterian actual niche, but are outside of the occupied niche due to sample bias. Finally, this same issue may be present in the northern extent of the output maps, which, besides the four kilometer resolution, are predicted as abs ent. Again, while portions of this area are expected to be within the actual niche, the occupied niche is slightly smaller due to the suite of presence points included in the modeling process and due to the research question which does not focus on modelin g central Italian Mousterian site distribution. This area could also indicate, however, that the suite of sites found in southern Italy has a different environmental envelope than the Italian sites directly north, which could mean a slightly different adap tation was needed for Neandertals to colonize the area. Future ECNM studies that include all of Italy can hel p to illuminate this issue.

PAGE 119

107 Despite these limitations and considerations, the model does successfully exptrapolate the Mousterian niche or potentia l site locati ons into new areas beyond the presence points While the model is only predicting the occupied niche, Figure 22 demonstrates that all four models are successfully able to extrapolate beyond presence point location and so this modeled occupied niche is most likely a sizable p ortion of the actual niche Areas such as the Tavoliere Plain, in which Mousterian sites are expected to be present, and the Calabrian peninsula, in which they are modeled as present in the interglacial and glacial models, but are predicted with low levels of probability in the modern outputs could be productive locations to begin attempts to correct bias in the presence point sample. These areas are likely predicted at low levels d ue to taphonomic processes, deposition and erosion respectively, thus the presence point input sample is biased against the Tavoliere Plain and the Calabrian peninsula By directing efforts to find sites in these areas as well as the higher probab ility areas as discussed later in the chapter rese archers may be able to get closer to the actual Mousterian niche in southern Italy.

PAGE 120

108 Figure 22 : Model Extrapolation. Each model with a fixed threshold applied at a value of 50 to demonstrate ability of the models to extrapolate niche to new areas. Produced in ESRI arcGIS 10.1. Model Similarity Similarity between all the models is demonstrated thr ough both a visual inspection of the site distribution outputs ( Figure 22 ) and through niche overlap and breadth statistics ( Tables 4 and 5). This similarity demonstrates that modeling the niche of Mousterian sites against modern environmental variables an d then projecting that niche to under sampled areas is an effective way to identify high probability areas of Mousterian site presence (Figure 23 ) A combination of these maps with recent representation of the state of localized factors such as erosion, la nd cover, land use etc.

PAGE 121

109 will help to identify where sites will most likely be found (Figure 24) and w ill allow researchers to conduct highly targeted surveys in an attempt to rectify historical and geographic issues which have led to a n uneven distribution of sites Additionally, through the production of modern niche of Mousterian sites, the full set of presence point data was included in the modeling process and issues of timescale and variable excavation methods did not have to be considered thus removing an aspect of bias from the model. Niche breadth (Table 5) also supports the success of the models ; interglacial breadth values are slightly higher than the breadth at the LGM, and slightly lo wer than modern breadth values, which is consist ent since the modern day outputs are modeling with all the sites, while the glacial and interglacial are only model ing sites with specific climate s Breadth statistic values a lso agree with the fact that Mousterian range was likely constricted during glaci al periods and expand ed during interglacial periods. This statement is made with the caveat that modeling with environmental variables reconstructed to a less severe glacial period like one Neandertals would have inhabited may produce a value closer or even equal to the interglacial breadth value. Therefore, while this pattern holds for modeling with these time periods, other studies may find the Mousterian nich e was just as broad during glacial periods.

PAGE 122

110 Figure 23 Similarity of the Modeled Niche. Figure 24 Productive Areas for Future Research Identified.

PAGE 123

111 Model Generalization Variables that were identifi ed as the most important with the jackknife procedure and thus also identified as drivers of the model outputs can help produce some gener alizations of the Mousterian niche in the study area Due to the stochastic nature of the algorithm, the different subsets of presence points and environmental layers included in the mode l, the different mod eling scales, and the relative importance of all the variables to modeling process (Table 2) it is unsurprising the models do not share the same specific important environmental layers. However, overarching generalizations can be made about the layers that were identified as important. In the modern outputs, aspect is important in the four kilometer and one kilometer resolution models, as well as geologic age, annual mean temperature and annual precipitation T his is consistent with site location as op posed to the past niche being modeled ; however caves strongly pattern against geology, and the modern models do include more cave sites as inputs tha n the glacial and interglacial models so it is possible that the difference derives from this association rather than a difference between the outputs. Additionally, an index of seasonality was also identified in both modern models. Jackknifing identified precipitation seasonality as important to the four kilometer model, and temperature seasonality as import ant to the one kilometer. While the same variable was not identified these models demonstrate Mousterian sites in southern Italy are generally not fo und in areas where seasonality, either in precipitation or in temperature, is high.

PAGE 124

112 Variables that are related, for example slope, aspect and elevation do not show in combination for the important variables in any of the tests. Also, a variable modeling relationship to water flow accumulation, flow direction or distance to water, was identified as im portant in all the models, and either flow accumulation or flow direction was identified as most important. The modern four kilometer resolution was the outlier because distance to water source and flow accumulation were identified as important, which may be a function of a larger sample size and thus more varied environmental inputs. The modern one kilometer resolved model identified flow direction as most important, which is likely a fu nction of scale and sample size, and also may be due to the relationsh ip between flow accumulation and flow direction (Chapter IV flow direction raster layers are produced through a combination of a flow accumulation and elevation layers). As the glacial and interglacial models also identified flow accumulation as important in where sites are located on the landscape, the models indicate that flow accumulation is the most important layer when modeling southern Mousterian sites and how they relate to water, both in modern times and the past This relationship could be due to site placement choices, taphonomic processes, or a combination of both. Some variable related to precipitation levels is also common to all models Climate variables that constrained the LGM niche mostly describe precipitation levels, except the mean temp erature of the wettest quarter which despite measuring temperature is also related to precipitation suggesting that during colder and dryer glacial periods, the amount of rainfall was the constrai ning factor in the Mousterian niche. A comparison of annual precipitation levels for the modern four kilometer glacial and interglacial periods indicates that the glacial period niche had the areas with the lowest

PAGE 125

113 amount of annual precipitation, 36.7 cm per year as opposed to 46.5 in modern times and 54.3 in inte rglacial periods. Interglacial variables are a combination of temperature, precipitation and seasonality, perhaps suggesting that in this time period water was less of a co nstr aining factor, and seasonality and tem perature became more important as drivers of this niche model. Elevation was not an important factor for the modern models or the interglacial models; instead the jackknife indicated that aspect played a more important role. However, in the glacial model elevation was identified as one of the more important variables. Elevation in this niche ranged up to 1670 meters, while in the other two dis tribution models it only reached between 1000 1100 meters. There are instances of Mousterian sites at high elevations; a few sites in Italy on the Monte Gennaro and Monte Pellecchia are above 1200 meters above sea level (Milliken 2000: 45), there are Mousterian sites at 1300 meter elevations in Romania ( Hoffecker and Cleghorn 2000) and Jovk 1 in Armenia has evidence for Middl e Paleolithic hominins living at 2040 m eters above sea level (Pinhasi et al. 2011) Therefore, it is possible Neandertals were exploiting areas of high elevation in southern Italy as well This pattern could be due to the constraining factor of preci pitati on in glacial periods, discussed in the previous paragraph. T hroughout this region of Italy higher elevations are associated with higher precipitation so during water stressed times Neandertals may have increased t heir range to higher elevations to take a dvantage of this association. During less water stressed times, enough rainfall could have been present at lower elevations, so it may not have been necessary for the niche to expand to these elevations. However, the southern Italian Mousterian site presen ce point sample is bias ed against sites at elevation, so to explore if

PAGE 126

114 this is a real pattern or a function of the occupied niche as smaller than the actual niche, modeling the niche of all of Italy with the entirety of Italian Mousterian sites could be be neficial (Chapter VI). An overall comparison of ranges of all variables used in the modeling process indicates that the Mousterian niches do not include extreme highs or lows of a ny environmental factors (A ppendix B ). Additionally the jackkni fe values from all models (T able 3 ) suggest that there is no one variable that is consistently most important, or one that is consistently unimportant. This could be due to either a variable not being considered in this study or more likely, it is due to the human ability to adapt to environment through culture. Mousterian and Neandertal Sites as a Homogenous Assemblage Through an examination of similarities and differences between the past niche modeled in this study r son, Kageyama et al. (2008), this section explores the appropriateness of considering Mousterian sites as one large assemblage in modeling, or if smaller, regional populations of sites should be considered. The model produced ageyama et al. (2008) is extremely similar to the model produced in this study (Figure 25) ; t he modeled LGM niche falls within the range of the niche s identified for all of Europe. These two models are not exactly the same, the southern Italian interglacial niche includes temperature and precipitation values that are beyond the range of the niche modeled for all of Europe, likely attributable to the warmer and wetter climates during the last interglacial period. Differences a re also present in the variables identified as the most important; while the

PAGE 127

115 ageyama et al. (2008) paper identified temperature as most important to the models, this study identified precipitation as most important for the LGM a nd a combination of temperature and precipitation factors as most important for the LIG. These differences are likely tied to the size of the study area. In the Banks, ageyama et al. (2008) study, all of Europe was modeled which means temperature would have likely had a large effect on explaining the distribution of sites. In southern Italy, there are none of these areas, and so other bioclimatic variables became most important to patterning and are no long concealed by overarching pa tterns in geography Overall, through a visual comparison, these studies identify similar areas of southern Italy as occupied; despite some discrepancies in int ensity of occupation predicted, and despite the less resolved outputs in the Banks paper due to modeling being done at 50km Th e comparison of these outputs, despite the differing scale of study and differing assemblages of sites, suggests that for ECNM purposes, considering all Mousterian sites as one assemblage is effective. Smaller areas and smal ler resolution of data gives more finely tuned results, which can be seen in the resolution of outputs between this study and terson, Kageyama et al. (2008). However, lower resolution models are appropriate and useful to answer research questions which seek to model the niche over large geographic spaces Again, despite the differences in scale and resolution both of these studies lump sites, Ban et al. (2008) lump on the basis of geogr aphy, while this study lumps temporally. Therefore, it is possible that modeling with temporally and regionally constrained subsets may produce extremely effective models to answer mo re specific research questions and may idenfity local

PAGE 128

116 patterns and trends within the Neandertal and Mousterian niche which are obscured by modeling with larger assemblages of data. F or now t he assumption that Mousterian sites can be modeled together temporally and spatially can be considered effective for answering current rese arch questions. Figure 25. Comparison of the Niche Models. The European Neandertal niche (right) as interglacial (top left) and the glacial niche (bottom left) modeled by this study.

PAGE 129

117 Biogeographic Variable : The Apennine Mountains Fin ally, in all four of the outputs, Mousterian presence is predicted as low or absent through the Apennine Mountains (Figure 26 ). This mountain range may represent a biogeographical barrier that hindered movements and connections between Mousterian populations on the east and west coasts of southern Italy, potentially creating slightly distinct coastal populations. Interestingly, the degree of this separation seem s to vary with climatic period; more separation is present in the interglacial periods than in glacial. This isolation, and hypothesized local evolution, may help to explain why Italian Paleolithic archaeologists have identified typological groups, such as the Quinsonian, in some areas of Italy but not others, and may help to explain some of the variation between sites in the Mousterian record of the region (Milliken 2000). Genetic studies done in the future may also support the idea of the Apennine Mountains as a biogeographic variable if slight genetic differences between the eastern and we stern coast al populations and /or the populations to the north are identified. Finally as sites are better dated, it may be beneficial to model temporally and spatial ly constrained past niche for the east and west coasts of southern Italy to further explor e this biogeographic variable.

PAGE 130

118 Figure 26 Biogeographic Barrier of Southern Italy. All models produced with biogeographic barrier of Apennine Mountains circled. A. modern one kilometer resolution, B. modern four kilometer resolution; C. glacial, D. interglacial

PAGE 131

119 CHAPTER VII CONCLUSIONS AND FUTURE DIRECTIONS Overall, modeling site and spatial distributions of the Mousterian niche in southern Italy is a successful exercise although independent survey testing is needed to confirm this conclusion All the models consistently suggest the same areas of southern Italy have a high likelihood of Mousterian site presence, indicating that modeling the niche of Mousterian sites based on modern day variables is successful Three area s of high probability of presence of sites were identified, including eastern Basilicata, hotspots throughout the mountains of western Basilicata and Campania, and area s into the Calabrian peninsula (Figure 23) Although independent survey is required to d emonstrate the success of the modeling and the accuracy of the site distribution maps, it is hoped that future survey in the region will make use of these maps to target areas with high probability of site presence. In this way, regions that have been over looked in terms of archaeological research, due to geographic and historical factors, can be explored for archaeological sites and limited survey resources can be effectively utilized. Additionally, generalizations about the Mousterian niche in glacial a nd interglacial periods were produced. All models indicated that the Mousterian niche does not include extremes of all variables, and in glacial periods, precipitation is the commonly most important environmental factor, while in interglacial periods, the most important factors are a combination of precipitation and temperature. Also a potential biogeographical barrier between the east and west coasts, the Apennine Mountains was identified

PAGE 132

120 Finally, comparisons of the Mousterian niche in southern Italy to modeled European Mousterian niche demonstrate similarity between the two, albeit with some differences, which suggest that consistent niche predictions are produced through various aggregations of Mousterian presence p oints at differing scales. However, this does not preclude successful model production with more spatially and temporally constrained study areas from being effective; and as knowledge of Neandertals and Mousterian sites increases and research questions be come more specific, it is possible this type of modeling will begin to be applied. It also may be effective in further exploring the role of the Apennine Mountains as a biogeographical barrier in southern Italy, among other questions. Along those lines, f urther direction of this research a t a smaller scale can potentially begin to reconstruct the more complex Eltonian niche of Mousterian sites, or at least aspects of it. The Stage 3 Project (2010) is one such project that may prove useful for future studie s researching smaller scale human environmental interactions in southern Italy and throughout the rest of Europe. The objectives of The Stage 3 P roject include s the determination of the paleoclimate and paleoenvironmental base of OIS 3, as well as the effe cts of the rapidly oscillating climate on the humans occupying Europe during this time. To accomplish this, researchers involved with the project have produced high quality downscaled GCMs representing bioclimatic variables from this stage, as well as a be ta model of past vegetation and land cover. Additionally, an archaeological and a faunal database was produced including Middle and Upper Paleolithic European sites with dates and geographic coordinates (Davies 1996 2015).

PAGE 133

121 This method also holds potential to explore the potential distribution of Mousterian sites and the niche of Neandertals in to the rest of Italy beyond the study area including Sicily, and to other refugia areas in Europe. P roducing a site distribution model for the whole of Italy holds p otential to produce an even more resolved output. By increasing the sample of sites beyond southern Italy, more sites will be accurately located on the landscape, and modeling can be done at an even smaller raster cell size, allowing targeted survey of pot ential Mousterian site locations throughout Italy. Additionally, including the rest of Italy in the modeling process can test the suggestion from some of the raster outputs, particularly the interglacial, that southern Italy required a separate adaption fr om areas directly to the north, as the niche based on southern Mousterian sites does not seem to extend at high levels beyond the study area. Finally, expansion of the study area to all of Italy can test the association of Mousterian sites and higher eleva tions during glacial period by including sites above 1000 meters elevation in the presence point subset. Also, Sicily holds great potential for applications of future ECNM studies. Currently, no Mousterian sites have been found on this island, despite abundant local material, varied ecotones (Robb 2007), and presence of year round water sources (Free GIS Shapefiles, Software, Resources and Geography Maps 1996 2015). Additionally, Lower Paleolithic sites have been found on the island, so this area has po tential for human habitation, especially as a refugia area during glacial periods. Currently, and likely for signfiacnt times in the Middle Paleolithic, Sicily is separated from the mainland of Italy by the Straight of Messina, which is three kilometers wi de in the north and 90 meters deep, and reaches 10 km wide in the south with a maximum depth of 250 meters.

PAGE 134

122 During glacial periods in the Middle Paleolithic, sea levels dropped to approximately 80 meters below current levels. This suggests that at some poi nts in time, Mousterian populations could have easily crossed to the island on foot or even been accidentally swept there as in the case of Homo floresensis I t is possible that sites have not been found in this region due to bias against Paleolithic arch aeology on the island driven by the various historical and geographic factors discussed in chapter II in the context of Southern Italy In this case, ECNM could be a very productive tool to identify areas where sites have a hig h probability of being locat ed by extrapolating the modeled niche from mainland southern Italy to the island. Overall, applications of the ECNM can begin to determine if the lack of Mousterian sites on Sicily is an actual archaeological pattern, or a function of bias. T his rese arch provides interesting insights into where Mousterian sites may be located on the southern Italian landscape, and the ways in which researchers can explore site locations through GIS technologies and modeling methods. Further studies along these lines w ill continue to facilitate understanding of both the systemic and archaeological context of Mousterian sites in southern Italy, and throughout the rest of Europe.

PAGE 135

123 WORKS CITED Anderson, R. P., D. Lew, and A. T. Peterson 2003 Evaluating predictive models of species distribution: criteria for selecting optimal models. Ecological Modeling 162: 211 232. Anderson, R, P., M Gomez Laverde and A. T Peterson 2002 Geographical distributions of spiny pocket mice in South Ame rica: insights from predictive models. Global Ecology and Biogeography 11: 131 141 Araujo, M B. and A Guisan 2006 Five (or so) challenges for species distribution modeling. Journal of Biogeography 33: 1677 1688 Arthur, P. 1991 Romans in northern Campania: settlement and land use around the Massico and the Garigliano Basin British School at Rome Banks, W E., F . Zilhao 2013 Human climate interaction during the Early Upper Paleolithic: testing the hypothesis of an adaptive shift between the Proto Aurignacian and the Early Aurignacian. Journal of Human Evolution 64: 39 55 Banks, W E., T Aubry, F . Zilho, A Lira Noriega and A. T Peterson 2011 Eco cultural niches of the Badegoulian: Unraveling links between cul tural adaptation and ecology during the Last Glacial Maximum in France. Journal of Anthropological Archaeology 30(3):359 374 Banks, W E., F . Peterson, M Kageyama, A Sima, M Sanchez Goni 2008 a Neanderthal Extinction by Competitive Ex clusion. PLOS one 3: 1 8 Banks, W E., F . Peterson, M Vanhaeren, M Kageyma, P Sepulchre, G Ramstein, A Jost and D l Lunt. 2008 b Human ecological niches and ranges during the LGM in Europe derived from an application of eco cultural niche modeling. Journal of Archaeological Science 35: 481 491 Banks, W E., F . L. Dibble, L Krishtalka, D West, D. I. Olszewski, A. T Peterson, D.G. Anderson, J.C. Gillam, A. Montet White, M. Crucifix, C. W. Marean, M. Sanchez Goni, B. Wo hlfarth, and M. Vanhaeran. 2006 Eco Cultural Niche Modeling: new tools for Reconstructing the geography and ecology of past human populations. PaleoAnthropology 4:68 83

PAGE 136

124 Barker, G. and R. Hodges. 1981 Archaeology and Italian society: prehistoric, Roman and medieval studies. British Archaeological Reports Beeton, T A., M. M. Glantz, A K. Trainer, S. S. Ternirbekov, and R M. Reich. 2013 The fundamental hominin niche in late Pleistocene Centr al Asia: a preliminary refugium model. Journal of Biogeography 1 16 Bergen, K. M., A.M., Gilboy, and D.G. Brown. 2007 Multi dimensional vegetation structure in modeling avian habitat. Ecological Informatics 2: 9 22. Berry, P M T P Dawson, P A Harrison and R G Pearson 2002 Modelling potential impacts of climate change on the bioclimatic envelope of species in Britain and Ireland. Global Ecology and Biogeography 11(6):453 462. Biddittu, I. 1987 Il Musteriano della Grotta dei Lardoni al Rip agnola, Polignano a Mare (Bari). Atti della XXV riunione scientifica; preistoria e protostoria della Puglia centrale 25: 119 127. Bietti, A. 1991 Normal science and paradigmatic biases in Italian hunter gatherer prehistory. In Perspectives on the past: Theoretical biases in Mediterranean hunting gathering research Eds G. Clark, pps 258 281. BioClim WorldClim Global Climate Data. Electronic document. http://www.worldclim.org accessed May 3, 2015. Bird, D. W and J. F 2006 Behavioral ecology and archaeology. Journal of Archaeological Research 14(2):143 188 Bordes, F. and D. Sonneville Brodes. 1970 The significance of variability in Palaeolithic assemblages. World Archaeology 2(1): 61 73. Bourg, N. A ., W. J McShea and D. E Gill 2005 Putting a CART before the search: successful habitat prediction for a rare forest herb. Ecology 86(10): 2793 2804.

PAGE 137

125 Brandt, R., B. J Groenewoudt and K. L Kvamme 1992 An experiment in archaeological site location: modeling in the Netherlands using GIS techniques. World Archaeology 24(2):268 282. Cole, J. P. 1966 Italy: An Introductory Geography Praeger Colwell, R. K. and D. J. Futuyma. 1971 On the measurement of niche breadth and overlap. Ecology : 567 576. Comuni Italiani 2004 Prometheo. Electronic document, http://en.comuni italiani.it.html accessed December 2, 2014 CORINE Land Cover European Environmental Agency. Electronic document. http://www.eea.europa.eu/publications/COR0 landcover accessed May 3, 2015 Davies, W. 1996 2015 Stage 3 P roject. Archaeological Datab a se. Electronic document. http://www.esc.cam.ac.uk/research/research groups/oistage3, accessed May 3, 2015. De Santis, V., M. Caldara, T. de Torres, and J.E. Ortiz. 2010 Stratigraphic units of the Apulian Tavoliere plain (southern Italy): Chronology, correlation with marine isotope stages, and implications regarding vertical movements. Sedimentary Geology 228(3) : 255 270. Delle Rose, M. and V. Marras. Ambienti e climi dei Neandertaliani del Salento. From carsismo.it. Di Cesnola, A. P. 1991 Il paleolitico superiore in Italia: introduzione allo studio Garlatti e Razzai. Elith, J., C. H. Graham, R. P. Anderson, M. Dudk, S. Ferrier, A. Guisan, R. J. Hijmans, F. Huettmann, J. R. Leathwick, A. Lehmann, J. Li, L. G. Lohmann, B. A. Loiselle, G. Manion, C. Moritz, M. Nakamura, Y. Nakazawa, J. McCoverton, A. T. Peterson, S. J. Phillips, K. Richardson, R. Scac hetti Pereira, R. E. S chapire, J. Sobern, S. Williams, M. S. Wisz, N. E. Zimmermann and M. Araujo occurrence data. Ecogeography 29: 129 151

PAGE 138

126 Elton, C. S 1927 Animal ecology Univers ity of Chicago Press. Ermoli, E.R. and G. di Pasquale. 2002 Vegetation dynamics of south western Italy in the last 28 kyr inferred from pollen analysis of a Tyrrhenian Sea core. Vegetation History and Archaeobotany 11(3): 211 220. ESRI 2011 ArcGIS Desktop: Release: 10.1. Redlands, CA: Environmental Systems Research Institute. ETOPO1 Global Relief Model 2015 National Oceanic and Atmospheric Administration, National Centers for Environmental Information. Electronic document. http://www.ng dc.noaa.gov/mgg/global/global.html accessed May 3, 2015. European Soil Database. 1995 2013 European Soil Portal Soil Data and Information Systems. Electronic document. http://eusoils.jrc.ec.europa.eu/esdb_archive/ESDB/Index.htm accessed May 5, 201 5. Faini, R., G. Galli, and C. Giannini. 1993 Finance and development: the case of Southern Italy. In Finance and Development: Issues and Experience Edited by A. Giovanni. pp. 158 213. Falcucci, A., L. Maiorano, and L. Boitani. 2007 Changes in land use/land cover patterns in Italy and their implications for biodiversity conservation. Landscape Ecology 22(4): 617 631. Federazione Speleolgica Campana Catasto Grotte delle Campania. Electronic document http://www.fscampania.it/catasto.htm a cces sed May 3, 2015. Fe derazione Speleologica Pugliese. 1977 2015 Catasto Grotte. Electronic document. http://www.fspuglia.it/elenco%20grotte%20al%202014.03.01.htm, accessed May 3, 2015. Franklin, J. 2009 Mapping species distributions: spatial inference and prediction Cambridge University Press.

PAGE 139

127 Free GIS Shapefiles, Software, Resources and Geography Maps 1996 2015 MapCruzin. Electronic document. http://www.mapcruzin.com/ accessed May 3, 2015. Fry, G. L A B Skar, G Jerpsen, V Bakkestuen and L Erikstad 2004 Locating archaeological sites in the landscape: a hierarchical approach based on landscape indicators. Landscape and Urban Planning 67(1): 97 107. GARP Users Manual and Technical Manual. 2015 Electronic document. http://www.nhm.ku.edu /desktopgarp/UsersManual.html accessed May 5, 2015. Gillam, J. C., D. G. Anderson, and A. T. Peterson 2007 A Continental scale Perspective o n the Peopling of the Americas: Modeling Geographic Distributions and E cological Niches of Pleistocene Populati ons. Current Research in the Pleistocene 24: 86 90 Giusti, M. 1979 La Grotta di Capelvenere a S. Caterina Nardo (Campagna di scavo 1974). 1: 19 39. 1980 La Grotta di Capelvenere a S. Caterina Nardo (campagna di scavo 1975). 2: 77 85. GIS Data Depot 1995 2014 Geocommunity. Electronic document, http://data.geocomm.com/ accessed May 3, 2015. Graham, C H, S R Ron, J C Santos, C J Schneider and C Moritz 2004 Integrating phylogenetics and environmental niche models to explore speciation mechanisms in dendrobatid frogs. Evolution 58(8): 178 1 1793.

PAGE 140

128 Guidi, A 1987 The development of prehistoric archaeology in Italy. A short review. Acta archaeologica 58: 237 246. a 1996 Nationalisms without a Nation: The Italian case. Nationalism and archaeology in Europe 108 118 b 1996 Processual and Post processual trends in Italian Archaeology. Theoretical and methodological problems 29 36. 2010 The Historical Development of Italian Prehistoric Archaeology: A Brief Outline. Bulletin of the History of Archaeology 20(2). Guisan, A. and N E. Zimmermann 2000 Predicitive habitat distribution models in ecology. Ecological Modeling 135: 147 186. Guisan, A and W Thuiller 2005 Predicting species distribution: offering more than simple habitat models. Ecology Letters 8: 9 93 1009 Guisan, A O Broennimann, R Engler, M Vust, N G Yoccoz, A Lehmann and N. E Zimmermann 2006 Using niche based models to improve the sampling of rare species. Conservation biology 20(2):501 511. Hannah, L., G. Midgley, G. Hughes and B. Bomhard 2005 The view from the Cape: extinction risk, protected areas, and climate change. BioScience 55(3):231 242. Hewitt, G 2000 The genetic legacy of the Quaternary ice ages. Nature 405: 907 913. Hijmans, R.J., S.E. Cameron, J.L. Parra, P.G. Jo nes and A. Jarvis. 2005 Very high resolution interpolated climate surface from global land areas. International Journal of Climatology 25: 1965 1978. Hodder, I. and C. Malone. 1984 Intensive survey of prehistoric sites in the Stilo region, Calabria. Proceedings of the Prehistoric Society 50: 121 150

PAGE 141

129 Hoffecker, J. F., and N. Cleghorn. 2000 Mousterian hunting patterns in the northwestern Caucasus and the ecology of the Neanderthals. International Journal of Osteoarchaeology 10(5): 368 378. Hugall, A., C. Moritz, A. Moussalli and J. Stanisic 2002 Reconciling paleodistribution models and comparative phylogeography in the Wet Tropics rainforest land snail Gnarosophia bellendenkerensis (Brazier 1875). Proceedings of the National Academy of Sciences 99(9):6112 6117. Hutchinson, G. E. 1957 Concluding remarks, Cold Spring Harbor Sump. On Quantitaive Biol. 22, Population Studies: Animal Ecology and Demography. Cold Spring Hrbor, Biol. Lab., 415 427 HYDRO1K 2012 United States Geological Survey, U.S. Department of the Interior. Electronic document. https://lta.cr.usgs.gov/HYDRO1K accessed May 5, 2015. HydroSHEDS 2010 Conservation Society Program of the World Wildlife Fund. Electronic document. http://hydrosheds.cr.usgs.gov/index.php accessed May 5, 2015. Jochim, Michael A 1976 Hunter gatherer subsistence and settlement: a predictive model Academic Press New York. Kellogg, Douglas 1987 Statistical relevance and site locational data. American Antiquity : 143 150. Kelly, Robert L 2013 The lifeways of hunter gatherers: The foraging spectrum Cambridge University Press. Kohler, Timothy A and Sandra C Parker 1986 Predictive models for archaeological resource location. Advances in archaeological method and theory :397 452.

PAGE 142

130 Kukla, G. J., B. L. Bender, J. de Beauleiu, G. Bond, W.S. Broecker, P. Cleveringa, J.E. Gavin, T.D. Herbert, J. Imbre, J. Jouzel, L. D. Keigwin, K. Knudsen, J.F. McManus, J. Merki, D.R. Muhs, H. Muller, R.Z. Poore, S. C. Porter, G. Seret, N.J. Shackleton, C. Turner, P.C. Tzedakis, and I.J. Winograd. 2002 Last Interglacial climates. Quaternary Research 58(1): 2 13. Kvamme, K 1992 A predictive site location model on the High Plains: An example with an independent test. The Plains Anthropologist : 19 40. 2006 There and back again: Revisiting archaeological locational modeling. In GIS and archaeological site location modeling, edited by Mehrer, Mark W., and Konnie L. Wescott.pp. 3 40. Boca Raton, FL. Lambeck, K. and J. Chappell. 2001 Sea level change through the last glacial cycle. Science 292(5517): 679 686. Lefvre, D., J. Raynal, G. Vernet, G. Kieffer, and M. Piperno. 2010 Tephro stratigraphy and the age of ancient Southern Italian Acheulean settlements: The sites of Loreto and Notarchirico (Ven osa, Basilicata, Italy). Quaternary International 223: 360 368. Levins, R. 1968 Evolution in Changing Environments: Monographs in Population Biology Princeton University Press, Princeton, New Jersey, USA. Lock, G. and T. Harris. 2006 Enhancing archaeological predictive modeling. In Gis and archaeological site location modeling edited by Mehrer, Mark W., and Konnie L. Wescott. Pp. 41 62. Boca Raton, FL. Lozier, J.D., P. Aniello, and M.J. Hickerson 2009 Predicting the distribution of Sasquatch in western North America: anything goes with ecological niche modeling. Journal of Biogeography 36:1623 1627 Martinez Myers, E. 2007 Inter model variation. Best subsets selection. Paper presented at the 4 th GBIF Ecological Niche Modelling Workshop. Available at http://www.ksib.pl/enm2007/index.php?id=documents Martini, F. and L. Sarti. 1974 Grotta Tina (Marina di Camerota, prov. di Salerno). Rivista di scienze preistoriche 341.

PAGE 143

131 Martini, F., B. Sala, G. Bartolmei, M. Tonon, and L. Catta ni. 1974 La Grotta Tina a Marina di Camerota (Salerno). Bullettino di paletnologia italiana 81: 27 79. Milliken, S. 2000 The Neanderthals in Italy. The Accordia Research Papers 8: 12 82 \ Oxford Journal of Archaeology 26(4): 331 358. Munoz, M.E.S., R. Giovanni, M.F. Siqueira, T. Sutton, P. Brewer, R.S. Pereira, D.A.L. Canbos, and V.P. Canhos. 2009 op modeling. GeoInformatica 15(1): 111 135. Mussi, Margherita. 2001 Earliest Italy: an overview of the Italian Paleolithic and Mesolithic Springer, 2001. Nogues Bravo, D. 2009 Predictin g the past distribution of species climatic niches. Global Ecology and Biogeography 18: 521 531 OpenModeller Overview 2015 Sourceforge. Electric document. http://openmodeller.sourceforge.net/index.html accessed May 5, 2015. Otto Bliesner, B.L., S. J. Marshall, J.T. Overpeck, G.H Miller, and A. Hu. 2006 Simluating Arctic climate warmth and icefield retreat in the last interglaciation. Science 311 (5768): 1751 1753. Panagos, P. 2006 The European soil database. GEO: connexion 5 (7): 32 33. Panag os P., M. Van Liedekerke, A. Jones, and L. Montanarella 2012 European Soil Data Centre: Response to European policy support and public data requirements. Land Use Policy 29(2): 329 338. Papes, M. and P. Gaubert. 2007 Modelling ecological niches from lower numbers of occerances: assessment of the conservation status of poorly known viverrids (Mammalia, Carnivora) across two continents. Diversity and Distributions 13(6): 890 902.

PAGE 144

132 Pearson, R. G 2010 istribution modeling for conservation educators and practitioners. Lessons in conservation 3:54 89. Pearson, R. G ., C. J Raxworthy, M. Nakamura and A T. Peterson 2007 Predicting species distributions from small numbers of occurrence records: a test cas e using cryptic geckos in Madagascar. Journal of Biogeography 34(1):102 117. Peterson, A.T., R R. Lash, D. S Carroll and K. M Johnson 2006 Geographic potential for outbreaks of Marburg hemorrhagic fever. The American Journal of Tropical Medicine and H ygiene 75(1): 9 15. Peterson, A T B W Benz and M 2007 Highly pathogenic H5N1 avian influenza: entry pathways into North America via bird migration. PLoS One 2(2):e261. Peterson, A T., M . Eaton 2007 Transferability and model eval uation in ecological niche modeling: a comparison of GARP and Maxent. Ecography 30(4):550 560. Peterson, A.T., M and J Sobern 2008 Rethinking receiver operating characteristic analysis applications in ecological niche modeling. Ecological modelling 213(1):63 72. Peterson, T. A., J. Soberon, R. G. Pearson, R. P. Anderson, E. Martinez Meyer, M. Nakamura and M. B. Araujo 2011 Ecological Niches and Geographic Distributions Princeton University Press, Princeton. Peterson, A. T. 2001 nic he modeling. The Condor 103: 599 605 Petronio, C. 1994 Melpignano (Lecce), Cava Nuzzo. Taras 14: 28 29. Petronio, C., E. Billia, L. C. Barbato, G. di Stefano, M. Mussi, S. J. Parry, R Sardella, and M. Voltaggio. 1995 The Late Pleistocene fauna from Ingarano (Gargano, Italy): biochronological, palaeoecological, paleoethnological and geochronological implications. Italiana 34(3): 333 339.

PAGE 145

133 P hillips, S. J., R. P. Anderson, and R. E. Schapire 2006 Maximum entropy modeling of species geographic distributions. Ecological Modeling 190: 231 259 Piccinno, A and F. Piccinno. 1972 Melendugno (Prov. di Lecce). Rivista di Scienze Preistoriche 27: 443. 1976 S. Eleuterio (Prov. di Lecce). Rivista di scienze preistoriche 31: 300 301. Pinhasi, R., et al. 2011 Middle Palaeolithic human occupation of the high altitude region of Hovk 1, Armenia. Quaternary Science Reviews 30(27): 3846 3857. Price, M. 2012 Mastering ArcGIS McGraw Hill, New York NY 5 th edition Radmilli, A.M. 1978 Guida della preistoria italiana Sansoni Editore. Raxworthy, C. J, E. Martinez Meyer, N. Horning, R. A Nussbaum, G. E Schneider, M. A Ortega Huerta and A T. Peterson 2003 Predicting distributions of known and unknown reptile species in Madagascar. Nature 429: 837 841. Raxworthy, C J C M Ingram and R G Pearson 2007 Species delimitation applications for ecological niche mode ling: a review and empirical evaluation using Phelsuma day gecko groups from Madagascar. Systematic Biology 56: 907 923. Ridgway, D. 1982 Archaeology in South Italy, 1977 81. Archaeological Reports 28:63 83. Robb, J. 2007 The early Mediterranean Village: agency, material culture and social change in Neolithic Italy Cambridge University Press, New York. Schermer, S. J and J. A Tiffany 1985 Environmental variables as factors in site location: An example from the upper Midwest. Midcontinental Jo urnal of Archaeology : 215 240.

PAGE 146

134 Shuttle Radar Topography Mission (SRTM) 2015 National Aeronautics and Space Administration. Electronic document. http://www2.jpl.nasa.gov/srtm/ accessed May 3, 2015. Small, A., C. Small, I. Campbell, M. Miackinnon, T. Prowse, and C. Sipe. 1998 Field survey in the Basentello Valley on the Basilicata Puglia border. Echos du monde classique 42: 337 368 Sobern, J. 2007 Grinnellian and Eltonian niches and geograph ic distributions of species. Ecology letters 10(12):1115 1123. Soberon, J. and A.T. Peterson distributional areas. Biodiversity Informatics 2: 1 10 Stage 3 Project. 2010 Department of Earth Sciences University of Cambridge. Electronic document. http://www.esc.cam.ac.uk/research/research groups/oistage3 accessed May 6, 2015. Stockwell, D. R. B., and D. P. Peters. 1999 The GARP modelling system: Problems an d solutions to automated spatial prediction. International Journal of Geographic Information Systems 13:143 158. Stockwell, D R. B. and A. T. Peterson 2002 Effects of sample size on accuracy of species distribution models. Ecological Modeling 148: 1 13 Stockwell, D. R. B. and I. R Noble 1992 Induction of sets of rules from animal distribution data: a robust and informative method of data analysis. Mathematics and computers in simulation 33(5):385 390. Stockwell, D Original creator of Gen etic Algorithm for Rule set Production (GARP) at ERIN Unit of Environment Australia and enhanced at the San Diego Supercomputer Center. Svenning, J. C Flogaard, K. A. Marske, and D. Nogues Bravo 2011 Applications of species distribution modeling to p aleobiology. Quaternary Science Reviews 30: 2930 2947

PAGE 147

135 Thomas, D. H., R. L Bettinger, B. W Hatoff, A. Leventhal and L Williams 1976 Prehistoric pion ecotone settlements of the upper Reese River Valley, central Nevada. Anthropological papers of the AMNH 53( 3 ): 267 363 Trigger, B. G. 2006 A History of Archaeological Thought Cambridge University Press, New York. Van Liederkerke, M., A. Jones, and P. Panagos. 2006 ESDBv2 Raster Library a set of rasters derived from the European Soil Database distribution v2.0. Published by the European Commission and the European Soil Bureau Network. Varela, S., J. M. Lobo, and J. Hortal 2011 Using Species distribution models in paleobiogeography: A matter of data, predictors and concepts. Palaeogeograph y, Palaeoclimatology, Palaeoecology 310: 451 463 Verhagen, P. and T. G Whitley 2012 Integrating archaeological theory and predictive modeling: a live report from the scene. Journal of Archaeological Method and Theory 19(1):49 100. Waltari, E., R. J. Hijmans, A. T. Peterson, A. S. Nayari, S. L. Perkins, and R. P. Guralnick 2007 Locating Pleistocene Refugia: comparing phylogeographic and ecological niche model prediciti ons. PLoS one 7: 1 11 Walton, E. M 2009 Ecological niche modeling as a conservat ion tool to predict actual and potential habitat for the bog turtle, Glyptemys muhlenbergii Ph.D. dissertation, Department of Geography, The University of North Carolina at Greensboro. Warren, D. 2010 New version of ENMTools up. Electronic document. http://enmtools.blogspot.com/2010/10/new version of enmtools up.html accessed May 6, 2015. Warren, D., R.E. Glor, and M. Turelli. 2009 ENMTools User Manual v1.0. 2010 ENMTools: a toolbox for comparative studies of environmental niche models. Eco graphy 33(3): 607 611.

PAGE 148

136 Wheatley, D 2000 Spatial technology and archaeological theory revisited. BAR INTERNATIONAL SERIES 845:123 132. Whitley, T G a. 2004 Causality and cross purposes in archaeological predictive modeling. BAR INTERNATIONAL SERIES 1227:236 239. b. 2004 Spatial Variables as Proxies for Modeling Cognition and Decision Making in Archaeological Settings: A Theoretical Perspective. Internet Archaeology 16(2). Willey, G. R 1953 Prehistoric settlement patterns in the Viru Valley, Per u Wise, A. L 2000 Building theory into GIS based landscape analysis. BAR INTERNATIONAL SERIES 845: 141 148. Wisz, M.S., R. J. Hijmans, J. Li, A.T. Peterson, C.H. Graham, and A. Guisan. 2008 Effects of sample size on the performances of species distribution models. Diversity and Distributions 14(5): 763 773.

PAGE 149

137 APPENDIX A Geologic Age, Soils and Land C over Layer Values Geologic Age Key The geologic age generalized by USGS and the original UNESCO age classes are as follows: 1. Q= q, qh, qh`m, qh`g, qh`fg, qh`f, qh`e, qh`t, qh`r, qp, qp`m, qp`g, qp`fg, qp`f, qp`e 2. Qv = V q, B qh, B qp 3. T = q+ng, m,ng, m4, m3, m3+2, pg, m2, m1 4. Tv = V m, B m 5. CzMzv = Vq t, tq t, B*q t, aV*q t, V`t*q t, L`t*q t 6. CzMzi = G ne, M ne, O ne, U ne, a ne 7. TK = m+c 8. Mz = ms 9. MzPz = ms+pl, t+p 10. K = c, c2, c1 11. KJ = c+j 12. J = j, j3, j2, j1 13. JTr = j+t 14. Tr = t, t3, t2, t1 15. Pz = pl, pl2, pl1, h+d 16. Pzv = V*p pr, L*p pr, B*p pr, aV*p pr 17. Pzi = Mpl, Opl, Upl, aMpl 18. P = p, p2, p1 19. C = h, h2, h2`h, h1, h1`h 20. D = d, d3, d2, d1 21. S = s 22. SO = s+o 23. O = o 24. Cm = cb 25. Pz = pl 26. Pzm = Paleozoic age color or label with metamorphic fill patterns present 27. Pzl = pl1, s+cb 28. Pzu = pl2 29. PzpCm = eo, s+eo, cm+eo, pl+eo, pl1+eo 30. PzpCmm = eo, s+eo, cm+eo, pl+eo, pl1+eo (metmamorphic fill patterns present) 31. pCm = eo, pl+pr, pr, prA, prB, prC, prD, G pr2, G prA, G prB, M pr2, M prA, G pr1, G prC, G prD, M prC, I pr, O pr, U pr, aM pr 32. Metamorphic rocks (Mzm,Pzm) were defined by metamorphic fill patterns on source map sheets.

PAGE 150

138 Soils 1. Key Soils with thick organic layers: Histosols (HS) 2. Soils with strong human influence Soils with long and intensive agricultural use: Anthrosols (AT) Soils containing many artefacts: Technosols (TC) 3. Soils with li mited rooting due to shallow permafrost or stoniness Ice affected soils: Cryosols (CR) Shallow or extremely gravelly soils: Leptosols (LP) 4. Soils influenced by water Alternating wet dry conditions, rich in swelling clays: Vertisols (VR) Floodplains, tidal marshes: Fluvisols (FL) Alkaline soils: Solonetz (SN) Salt enrichment upon evaporation: Solonchaks (SC) Groundwater affected soils: Gleysols (GL) 5. Soils set by Fe/Al chemistry Allophanes or Al humus complexes: Andosols (AN) Cheluviation and chilluviation: Podzols (PZ) Accumulation of Fe under hydromorphic condition s: Plinthosols (PT) Low activity clay, P fixation, strongly structured: Nitisols (NT) Dominance of kaolinite a nd sesquioxides: Ferralsols (FR) 6. Soils with stagnating water Abrupt textural discontinuity: Planosols (PL) Structural or moderate textural discontinuity: Stagnosols (ST) 7. Accumulation of organic matter, high base status Typically mollic: Chernozems (CH) Transition to drier climate: Kastanozems (KS) Transition to more humid climate: Phaeozems (PH) 8. Accumulation of less soluble salts or non saline substances Gypsum: Gypsisols (GY) Silica: Durisols (DU) Calcium carbonate: Calcisols (CL) 9. Soils with a clay enriched subsoil Albeluvic tonguing: Albeluvisols (AB) Low base status, high activity clay: Alisols (AL) Low base status, low activity clay: Acrisols (AC) High base status, high activity clay: Luvisols (LV)

PAGE 151

139 High base status, low activity clay: Lixisols (LX) 10. Relatively young soils or soils with little or no profile development With an acidic dark topsoil: Umbrisols (UM) Sandy soils: Arenosols (AR) Moderately developed soils: Cambisols (CM) Soils with no significant profile development: Regosols (RG)

PAGE 152

140 Land Cover Key 1. Continuous urban fabric 2. Discontinuous urban fabric 3. Industrial or commercial units 4. Road and rail networks and associated land 5. Port areas 6. Airports 7. Mineral extraction sites 8. Dump sites 9. Construction sites 10. Green urban areas 11. Sport and leisure facilities 12. Non irrigated arable land 13. Permanently irrigated land 14. Rice fields 15. Vineyards 16. Fruit trees and berry plantations 17. Olive groves 18. Pastures 19. Annual crops associated with permanent crops 20. Complex cultivation patterns 21. Land principally occupied by agriculture with areas of natural vegetation 22. Agro forestry areas 23. Broad leaved forest 24. Coniferous forest 25. Mixed forest 26. Natural grasslands 27. Moors and heathland 28. Sclerophyllous vegetation 29. Transitional woodland shrub 30. Beaches dunes sands 31. Bare rocks 32. Sparsely vegetated areas 33. Burnt areas 34. Glaciers and perpetual snow 35. Inland marshes 36. Peat bogs 37. Salt marshes 38. Salines 39. Intertidal flats 40. Water courses

PAGE 153

141 APPE NDIX B Comparison of Niche Values

PAGE 154

142