This article is published as part of
In many ways the Roman province of Baetica
is an ideal subject for exploring new
approaches to historic transport
geography. This is not due to the
completeness of its record (for it is
not), but because it provides a
remarkable breadth of pertinent data. This
paper, loosely based on a seminar hosted
by the Digital Classicist at King’s
College London, will briefly discuss the
results of applying some as-yet
relatively uncommon techniques to the
archaeology and documentary record of
transport in the area. It will then go
on to tackle some more general issues in
creating maps of movement in the past,
concluding that there is still much
theoretical work to be done, but that
the potential for discovering new
patterns in old data is great, and
indeed, ever growing. The main concept
that will be explored is that of a Node Network,
an abstract model of the interactions
between spatially separate locations.
This paper demonstrates the potential of
a standard relational database, coupled
with a GIS and Network Analysis software
package, to make a spatial argument
about the relative importance of key
towns within a transport network and
expose the constituent elements of that
argument in a formal, visual manner.
In many ways the
Roman province of Baetica is an ideal
subject for exploring new approaches to
historic transport geography. This is
not due to the completeness of its
record (for it is not), but because it
provides a remarkable breadth of pertinent data
(Sillières 1990, 9-16). This
paper, loosely based on a seminar hosted
by the Digital Classicist at King’s
College London, will briefly discuss the
results of applying some as-yet
relatively uncommon techniques to the
archaeology and documentary record of
transport in the area. It will then go
on to tackle some more general issues in
creating maps of movement in the past,
concluding that there is still much
theoretical work to be done, but that
the potential for discovering new
patterns in old data is great, and
indeed, ever growing.
The main concept
that will be explored is that of a Node Network,
an abstract model of the interactions
between spatially separate locations.
This paper demonstrates the potential of
a standard relational database, coupled
with a GIS and Network Analysis software
package, to make a spatial argument
about the relative importance of key
towns within a transport network and
expose the constituent elements of that
argument in a formal, visual manner.
Before describing the results, however,
it will be worthwhile to discuss here
some contextual issues that have
determined the techniques and
technologies adopted.
First, it is
important to distinguish between
transport networks and day-to-day
travel. The great majority of human
motion takes place in an extremely
complex and essentially unpredictable
fashion. Modern anthropological studies,
however, suggest that across virtually
all sedentary cultures it is a) limited
to approximately one hour a day, and b)
there is tendency to return to a place
of residence each day (Ausubel & Marchetti
2001, 20-22). This places a
hypothetical ‘upper limit’ on the
distance most individuals will travel
which we might loosely describe as
intra-site (by which is meant the ‘home’
location and its immediate environs).
Indeed, it is one of the restraining
factors on the growth of ancient cities,
which, despite occasionally being able
to marshal vast resources and being
densely populated, never grew more than
a few kilometres in diameter (Ausubel & Marchetti
2001, 20-22). Based on these facts
we shall here postulate that most inter-urban
transport takes place only under special
circumstances that require it (Turton
& Black 1998,
159-60). It is the routes used in this
type of long-distance transport, and the
specific conditions that gave rise to
them in Roman Baetica, that will form
the main subject of this report.
The primary source
of information that we can draw upon in
identifying such routes are the Roman
itineraries. Looking beyond the raw
lists of data that these documents
provide (which we shall address fully
later), we should remark upon two
further aspects of them. First of all,
they imply intentionality. In each case,
their authors are suggesting that under
rationale x, it
is preferable to travel from A to B by way of C. If we are
able to understand this rationale then
we can begin to make sense of the way
that people moved around the ancient
world. That said, in the majority of
cases we can currently only hypothesize
as to just what x
might be. Secondly, they are also
vectors, having a beginning and an end.
We might very reasonably surmise that
the journeys described could just as
well be undertaken in the opposite
direction, but this asymmetry is not
something we should entirely lose sight
of. As we shall see, it may give us a
crucial clue into the manner in which
these ancient travel guides were
created, and hence their strengths and
weaknesses as descriptions of the
network.
The chronological
development of the system is a further
area in which our evidence is sparse. It
is very difficult to distinguish, from a
transport perspective, between the
Tartessian, Phoenician, Punic,
republican and imperial periods, yet
surprisingly this is not as problematic
as might be thought. The sources upon
which this study was based are entirely
Roman (or post-Roman) so they reflect
the reality of the classical era (though
sadly shed little light on pre-Roman
development). We should nonetheless
expect, as with most transport
infrastructure throughout history, that
it was a response to, and often a
reinforcement of, the systems of
communication which had hitherto served
the area (White
& Senior 1983, p.
11). As such, whilst this study is
concerned specifically with Roman route
networks (a period itself spanning no
less than six centuries), many of the
routes we shall be discussing may well
have been used since the earliest times.
It should by no means be assumed that
the plains and valleys were a
With a handful of
exceptions, we have few material remains
of transport infrastructure outside of
urban settlements. The itineraries
however, along with a limited number of
well established routes from other
sources, will enable us to construct a
theoretical
network. It is to this sort
of data that Network Analysis can apply
a powerful set of tools to create
metrics showing the relative importance
of individual locations and routes
within their wider nexus.
The history of
Baetica’s territory is both long and
complex. It played host to a number of
literate cultures in antiquity,
beginning with the Tartessians On Tartessian literacy, see
Strabo 3.1.6 (Jones
1969).
Baetica itself, a
province created from the southerly part
of Hispania Ulterior by Augustus in the
late first century BC (Keay
1988, p. 49), is roughly
commensurate with modern Andalusia (see
Figure 1). Its primary importance to
Rome, and indeed many of its previous
inhabitants, was the valley of the River
Baetis (Guadalquivir), with its broad
plains and fertile soil, and the Mons
Marianus (Sierra Morena) mountain range
to its north, rich in precious metals.
The regional economy was not merely
dictated by its resources however, it
was also greatly affected by its
topography. The central valley is
virtually cut off to the North, South
and East by two significant mountain
ranges, the Sierra Morena and the
Cordillera Sub-Bética (Keay et al.
2000, p. 1). The natural
entry and exit point was therefore via
the large tidal estuary (the
Ruiz
Mata 2002 p. 158
(for the traditional account
see Strabo 3.5.5)
To complement this natural conduit, the province developed a complex network of roads. These are testified to by numerous milestones and a number of bridges throughout the region. They not only linked towns within the interior, but also connected them to neighbouring provinces, often providing guidance and sure footing through tortuous mountain valleys. On a more regional level, they would also have been necessary in enabling wheeled vehicles to transport local produce to central markets, or entrepôts from which they would be shipped to the wider empire.
Of particular
interest to us is the fact that the
Baetican economy centred on exports. In
contrast to many other regions whose
economic surplus appears to have been
derived either from services to the
military or limited regional trade,
Baetica produced and exported in large
volumes. The principle commodities were
metals (especially iron, gold, copper,
and silver), fish, wheat, wine, and most
importantly, olive oil (Ponsich
1998, 171-82). An interesting
aspect of these ‘industries’ is that
they appear to exhibit the economic
phenomena of A concept first expressed in
Ullman
1956.complementarity
Node Networks, a formal structure used in Graph Theory, are simply a number of entities, called nodes (or vertices), in real or abstract space that are linked together by lines, known as edges (or arcs, if directional). These may represent anything from molecules, to the World Wide Web, to social networks. In this case we are using them to represent route systems. Besides providing a useful and intuitive tool for describing such systems, they are also susceptible to mathematical analysis in order to ascertain the importance of individual nodes within a network.
Networks have been
used in a variety of ways to understand
connectivity in past cultures, and do
not of necessity require the use of
computer technology (see e.g. Pitts
1965), although modern processing
capacity undoubtedly makes analysis a
great deal faster and more practical.
Notable recent studies in their
application to ancient transport have
explored the autogeneration of networks
based on maritime trade and transport
factors in the Bronze Age Aegean (Evans et al
2007), and combining them
with Agent-Based Modelling to analyse
the Antonine Itineraries (Graham
2006). These approaches vary
greatly in the assumptions they make
about their base data and the techniques
they employ. It is therefore important
to remember that Network Analysis
closeness
and betweenness
centrality, and
will assume bi-directional links
(edges).
Closeness centrality can
be stated as the ease with which a
node can reach, or be reached by,
any other node on the network. It
is an index of how easily
accessible a node is to all the
other nodes in the network and is
a value between 0 (inaccessible)
and 1 (directly accessible in 1
step by all). Two graphs which
demonstrate this most clearly are
a simple star graph, in which the
central node has a closeness
centrality of 1.0, and a cycle
graph in which all the nodes will
have identical closeness
centralities. In a network of
vertices and lines, (V,L) the
function, cl(v), of the normalized
closeness of a vertex, v, is
formally defined as (Batagelj 2005):
Where d(v,u)
is the shortest path (or geodesic), in terms of nodes
traversed, between v and any
other node, u. These distances are
summed, and this value is then normalized by dividing by
the total number of vertices (n), – 1.
Normalization is important as it
enables us to compare this node’s
closeness with that of nodes on
other networks.
Betweenness centrality is
defined as the probability that a
node will be passed by traffic
travelling along the shortest
route between two other nodes on
the network. The index indicates,
not how easy it is to reach other
nodes, but the likelihood of it
being
Where: v is a
node in a network of vertices and
lines (V,L),
g
u,t is the
number of geodesics between each
possible pair of notes u, t, and
g
u,t(v)is
the number of geodesics between
each possible pair of nodes u, t that
pass through v.
Once again, the value is normalized
to a value between 0 and 1, this
time also to take into account the
fact that geodesics from u to t, and
from t to
u
will both be included in the
equation (hence (n-1) (n-2)).
Betweenness is the metric that interests us here most because it indicates which nodes have a higher degree of control over the network (Freeman 1977, 35-36). In a transport context, although this is not likely to be in the form of obstructing traffic, such key nodes have the potential to influence the way in which that traffic flows, perhaps in a very concrete fashion, as we shall see. They may also benefit from the increased degree of economic activity that is created by the confluence of separate linear routes (Pitts 1965, p. 15).
The first network we will look at is a series of linear routes known as the Antonine Itineraries, a list of both land and sea itineraries between towns throughout the empire attributed to an emperor Antoninus. Analysis of the locations recorded in the text suggests a date around the end of the third century. The ‘lost’ regions of Dacia and the Agri Decumates, between the Rhine and Danube, are notably absent, but on the other hand, Constantinople is generally referred to by its pre-Constantine toponym, Byzantium (Salway 2001, p. 39). Their exact function is unclear, though their internal structure suggests that several regional groups have been ‘stitched together’ to create a ‘global’ itinerary list (Salway 2001, p. 43). Despite providing an invaluable catalogue of information there are a number of surprising lacunae, and many routes do not follow the shortest path. Of the 225 routes described, thirteen are directly relevant (wholly or partially) to this study. The values of centrality we are generating are network dependent so it would not make sense to include all of them, but in this case they are short enough to decide on a case by case basis. In fact, a number of them stop either at, or close to, the limits of Baetica. The network generated is shown in Node Network 1.
What do our centrality indices tell us? Here the importance of a network is shown clearly, especially when displayed visually. Although closeness does not vary dramatically between sites, betweenness does, and it is no surprise to find that three of our provincial capitals dominate the graph below.
The Ravenna Cosmography written by an anonymous monk of Ravenna, is an attempt to compile a list of the all the towns in the known world at the end of the seventh century. The (corrupt) version we possess is a Latin translation from the Greek that dates from the ninth century. Though claiming to draw on a variety of ancient sources, recent scholarship suggests that it is based principally on the Peutinger Table – a spatial (though abstract) itinerary map of the ancient world dating from the mid-fourth century (Salway 2001, p. 28). Although a later medieval copy of the Peutinger Table has survived, frustratingly the westernmost section is missing, leaving the Ravenna Cosmography as our only guide to its contents. To further complicate matters, analysis of placenames on the map shows that their grammatical declension is not consistent, indicating that it in turn was compiled from a series of written itineraries. These undoubtedly came from several sources as they juxtapose towns destroyed by Vesuvius with those built in the reign of Constantine (Salway 2001, p. 44). An initial mapping creates a surprising result however – there appear to be three separate networks which do not interrelate. We might suppose, however that the compiler has made a mistake, perhaps by lifting sections out of an unknown source that links itineraries together. Almost all of the itineraries lead toward, but then stop prior to, important towns. If, however, we add in the most obvious candidate in each case (following (Sillières 1990 p. 32)), the network connects itself in a much more understandable fashion.
There is a noticeable difference to our first network however. Looking at centrality, although Hispalis is once again the key node, the other capitals do not appear in the top 15 at all.
In this network the chief axis is between Hispalis and Malaca (if we are right in taking the additional step from Aratispitani to the coast).
The strength of the
database system used to generate the
graphs is that it enables us to combine
multiple networks easily, adding or
removing itineraries and nodes as
desired. To investigate the entire known
transport system we can create a
‘super-network’ composed of the two
networks discussed above with three
further additions: The Via Augusta - the
principle imperial highway which is
known from a set of inscribed goblets,
the river transport network, and a route
of a single day’s journey is known both
from
Which nodes are most important for betweenness and closeness? The graph below leaves no doubt:
Despite the fact
that closeness values still do not vary
much, Astigi, Hispalis and Corduba are
considerably more important than any of
the other towns in terms of betweenness.
In other words, if our data can be taken as
broadly representative of the primary
transport routes in Baetica, they are
unquestionably the focal points of the
Baetican transport system. But if we are
to accept this as an at least partial
explanation for the locations of three
of the district capitals, then what of
Gades? So far, it has maintained fairly
low values in all of our networks. If
betweenness really plays such an
important role, then why was Gades,
rather than candidates such as Carteia,
which is more central within the
district, or Malaca, with better access
to the interior, chosen as capital?
This is a difficult question to answer, and indeed there may be a number of reasons. It may be as a result of its historic significance and development during the previous Phoenician and Punic periods. It has a deep water port but is also easily defensible from attacks from the mainland, a useful advantage during the civil wars of the republic. It is furthermore the principle harbour in the Bay of Cadiz and provides relatively easy access by boat to the other three capitals. Towns farther east would have to pass through the straits of Gibraltar, a passage which was periodically closed by bad weather. Yet a more convincing explanation might be provided by noting that the model we have been using for our land networks makes no reference to this mode of transport, and incorporating it would be a major challenge. Nonetheless, we can be certain that some patterns existed. Winds, tides and currents all interact with ship technologies in ways that restrict, or encourage the use of, different harbours (Evans 1988, p. 367). That is a topic well beyond the scope of this study, but one that would undoubtedly enlighten, and be enlightened by, Gades’s role within Baetica.
Having looked at
some of the structural aspects of the
network, we can now turn to the subject
matter of the sources themselves, and
specifically the
Starting with the
Antonine Itineraries, there is an
interesting correlation, at least within
the itineraries considered. Almost all
of them The exception being the
curious, single-step route
from Hispalis to Italica
The correlation is curious as the routes themselves do not necessarily respect provincial divisions, nor do they take the fastest route. We might also note that the number of non-capitals is actually very small – in fact there are only four: Italica (x1), Malaca (x2), Baesuris/Mouth of Anas (x2), and Castulo (x3). In fact, Baesuris and Castulo form the initial/terminal nodes of other itineraries, not included in our network, as well.
Turning to the Ravenna Cosmography, we see links between the capitals and frontiers once again:
The patterning is
similar to that of the Antonine
Itineraries, and with similar
‘frontier-towns’. Malaca features again
(x2), as does Castulo (x1). The new case
is Baesippo/Baelo, although in two cases
this is mentioned within the context of
the straits of Gibraltar “
What do our
previous results suggest about the idea
of key itinerary nodes with
is reflected
in the itineraries, then Astigi, a
district capital, does not appear to be
one of them. This is certainly not
strong proof against the theory, but it
suggests that a broader study would have
to be done to give further support, one
way or the other.
Being able to
break the network down by individual
itinerary, however, also gives us a
clearer idea of how each relates to the
other, and perhaps some further clues as
to the nature of their sources. For
example, some of the itineraries clearly
could be described at their origin,
taking a fairly direct route. Others
take diversions that might also be
permissible, especially when they go by
way of important towns such Astigi,
Acci, or Corduba. There are one or two
however that could surely not be
described on a public itinerary table.
Only the initial stages of the
Gades-Corduba Antonine Itinerary could
have been described on a
Yet the information we now possess may give us a new way of looking at the problem. The general shape of the overall network is surprisingly regular, and this regularity is also seen clearly in the Network Analysis diagrams. In fact it is remarkable how many similarities exist between the Ravenna Cosmography and Antonine Itineraries, and how well they complement each other, especially as the individual itineraries are quite different. Such similarity suggests, though it cannot prove, that the rationales behind them may reflect some wider reality. Including the river network and the extensions of Sillieres, the chief features (with starting points here chosen arbitrarily) seem to be
- Two central axes running North-South and East-West. The former running Emerita-Celti-Astigi-Antikaria-Malaca. The latter, Castulo-Corduba-Astigi-Hispalis-Baesuris.
- A separate circuit traces the main extents of the Gudalquivir Valley, running Hispalis-Urso-Antikaria-Corduba-Celti-Italica-Hispalis
- To the East, a route connects the end nodes of Malaca and Castulo, whilst another (not included in the database as it lies entirely outside of Baetica) joins Cordoba and Emerita. There is also a direct road between Emerita and Corduba, the provincial capitals.
- To the West, the
lacus prevents any direct land route between Malaca and Baesuris, but there is a route to Hispalis. There is also a route from Hispalis to Emerita.
If we draw a
schematic diagram of the principle route
network, and mark on the places in which
itineraries begin, an interesting
pattern emerges. All of the itineraries begin on
the boundaries of Baetica, except for
Corduba, the provincial capital. They
are particularly common in port towns as
well, notably those that are known to
have direct connections with other
provinces. Whilst it is difficult to
understand how they could have been
constructed from monumental
arrival at
a new province does not seem at all
far-fetched. Likewise, we would expect
such information to be available at its
capital. If the system did work in this
way it would provide an explanation for
Astigi’s absence from the list of
starting nodes, as well as explain its
fundamental role within the route system
as a whole (it is the central node of
the entire network).
We would be
foolish to jump to conclusions on the
basis of just one province, but the
co-incidence of starting points with
clear points of provincial interface is
one worthy of further investigation. If
we are to use the Roman itineraries then
we have to try and understand the
rationale behind them. The implication
of the study just described is that they
may well have been written by visitors to
Baetica, rather than the native
population. If that is the case, it is
no great wonder that the network appears
strangely symmetrical whilst the routes
meander, for they are plans made by
people who have had Baetica explained to
them.
The foregoing
discussion will hopefully have made
clear that there is a great deal more
potential in the use of Network Analysis
than can possibly be looked at within
the scope of this paper. On the one
hand, it has turned out to some
surprisingly strong results. It should
not surprise us that the district
capitals show a higher level of degree
centrality (direct links to other sites)
than other cities. Indeed, it would be
curious if they did not. It is also not
particularly unusual to see fairly even
levels of closeness within the network
as a whole, as there is a surprising
degree of symmetry and it is quite
strongly interlinked. What comes as a
greater surprise is the variation in
betweenness. This index shows to a high
degree of probability that the capitals
were chosen either as
As has been
reiterated throughout this discussion
however, such results must be approached
with caution. The network remains a
model founded on data that is both
corrupt and incomplete. It does not take
distances into account, nor links which
are not explicitly stated within the
record. It is also based on a subset of
data which, though not selected
arbitrarily, may conceivably not reflect
the ‘real’ social, political and
commercial structures of Baetica during
the Roman period. Until further work has
been done on the robustness of such
networks, i.e. the degree of change
resulting from the presence or absence
of nodes and links, then any conclusions
drawn are of interest only in so far as
clear correlations can be drawn with
independent factors (such as the
political status of sites). The network
model itself cannot be deemed reliable
in any of its particulars. But despite
all this, they are interesting results. Even if
they do not show the complete system of
routes used by the resident traders,
magistrates and other itinerants, they
may well reflect the thoroughfares
perceived to be important by external
visitors. This may go some way towards
explaining a layout which often seems to
make more sense from an abstract
perspective than it does on the ground.
One of the more exciting aspects of the work is that now the framework has been created, it is in fact very simple to extend. The dataset considered here is only a fraction of the itineraries that have come down to us. Both the Antonine Itineraries and the Ravenna Cosmography/Peutinger Table cover almost the entire ancient world, much of the latter in a visual format. There are numerous others, large and small. This is not to suggest that in creating a ‘super-itinerary’ we are mapping the Roman world. The problem with such approaches is that they tend to consolidate all the information into a unified network. But by introducing such data into a transparent framework we are able to see much more clearly the different sources, the starts and finishes, the virtually-certain, and the wild-speculation. In this way we may be able to glean some better understanding of the way itineraries worked and how they were used by travellers, rather than chasing the chimera of a complete knowledge of Roman transport structures.
A further
direction to explore is the
incorporation of distance into the
equation but there are difficulties to
be overcome – not least of which is what
distances to use as the concept of
distance requires some unpacking. Some
progress has been made towards this in
the application of ‘cost-distances’
which factor in the relative
difficulties in crossing different kinds
of terrain. Whilst this is clearly an
improvement on utilizing simple
Euclidean distances, it is still an
inadequate measure of transport cost.
First, transport is above all a value-based
activity. It involves a playoff between
the advantage gained and the cost (or
‘friction’) involved. This friction is
better understood as an equation
involving the two coefficients of time
and effort/expense. The value of either
may vary in both absolute and relative
terms but no-one moves things if it’s
not worth it.
Secondly, transport friction is not a single conceptual entity but composed of (at least) the following factors:
- Space. The actual distance involved.
- Environment. The natural surface over which transport takes place, affected by both physical and socio-cultural barriers (e.g. mountains and borders).
- Static technology. Infrastructure such as roads or jetties.
- Dynamic technology. Planes, trains and automobiles, or in our case, ships, barges, oxen, etc.
- Systems. The individual and collective behaviours through which the technologies are utilized. These include the abilities of the populace to use certain technologies, and the laws and norms that dictate how they do so.
- Load. The volume and nature of passengers and cargo (and dynamic technology) transported.
The challenge in
modelling transport is that there is,
and can be, no natural ‘friction’ for
any point of space, nor any fixed
‘price’ for any mode of transport.
Whilst all transport has a ‘cost’ it is
always a unique calculation. Though it
is possible to talk about ‘average
costs’, they may bear no relation to
actual realities of particular cases.
What we need to look for is the economic
See, for example, the behavior
of the consul L. Lucullus in
Appian, Iberike
51-55
Thirdly, the ubiquity and aggregative cost of transport makes it conducive to economies of scale. Transport technology, both static and dynamic, does not change frequently because the cost would be prohibitive and returns are only made in the long run. This is advantageous to the archaeologist, for transport networks have a heavy historical bias which will often be based on a wide range of interests, including many which are long since past. When considering such ‘legacy systems’, however, we must be careful to differentiate between the users and investors. Most infrastructure reflects the agenda of those wealthy and/or powerful enough to create or influence it and it will not necessarily represent the preferences of those who use it - it may just be the best means available for their requirements.
If we want to try and understand transport as anything more than historic contingencies of landscape we need to try and understand the forces at work behind it. Specifically we need to ask who was involved in the transport process, what their primary concerns were, and what obstacles and resources surrounded them. The inclusion of physical distances is just a single, if important, aspect of this.
The research presented was undertaken as
part of a Master’s dissertation under
the aegis of the Urban Connectivity in
Iron Age and Roman Southern Spain
Project, an AHRC-funded project based at
Southampton University In collaboration with the
Seville Office of Culture
and the Department of
Prehistory at Seville
University From UCIARSS website, http://www.arch.soton.ac.uk/Projects/projects.asp?ProjectID=64