From Palmyra to the Euphrates: Tracing trade routes as networks

This post summarizes parts of a paper presented at the Connected Past conference in Paris, April 24 this year, and parts of a paper offered together with my colleague Professor Jørgen Christian Meyer at the ARAM conference in Oxford on July 14.

The challenge we’re dealing with is tracing the ancient caravan route from the city of Palmyra to the Euphrates. This part of the Near East is inaccessible to archaeologists, and has been so for a long time. Most existing research dates back to the 1930s, when Antoine Poidebard and Aurel Stein surveyed the route from the air and on the ground, from the Syrian and Iraqi sides of the border respectively.

The network analysis is only part of the wider case study, which besides publsihed archaeological work also considers GIS modelling, satellite images, ethnographic accounts, travel descriptions, and the physical environment of the Syrian Desert. Pending peer review, the study will be published in a forthcoming volume of the journal ARAM.

Step one of the Network Analysis was to locate all known archaeological sites in the relevant part of the Syrian Desert (below). This was done on basis of archaeological reports as well as British, French, German and Soviet Maps of the region, cross-checked with Google Earth, the Corona Atlas of the Middle East and Bing Maps. These were plotted in Google Earth, and then imported into Arcmap. Still, considering the scarcity of past archaeological work, we had no idea whether there were not also other sites out there, that might equally well have been stations on the trade route.

We decided to approach this by looking at hydrology. If you want to move through the Desert with a caravan, you’ll need to know where to find water. Utilizing 1:100 000 maps imported as overlays into Google Earth we plotted all hydrological features in Google Earth. Imported into ArcGis they look like this (below). Altogether there are 244 of them, wells, springs, cisterns. Can we trust that they were the same in antiquity? To a large extent we think we can. The climate has not changed much. Most wells draw on groundwater and are placed at the bottom of wadis, seasonal watercourses that were the same in antiquity as they are today. Finally, our experience from the area North of Palmyra, where we did survey for four years, is that these wells and cisterns are associated with pre-Islamic pottery, and have thus been in continuous use by the nomadic population.

This, however, still did not enable us to trace the route. In order to do that, we turned to network approaches.

First, I added a 20km buffer to all hydrological features (below). 40 km is a long day’s march for people and camels alike, and wherever two circles intersect, you can reasonably walk or ride from one point to another within a day, if you know your way of course. You see here that the region close to Palmyra has a high density of wells. Also, close to the Euphrates, the availability of water is good. Whereas in the middle, you have stretches of up to 100 kilometer without perennial water sources. This is a strong argument that a caravan route needed to be created and maintained, and this was something that the Palmyrenes needed to deal with in their period, regardless of the actual age of the ruins that early explorers in the Syrian Desert visited.

(The map also shows the routes proposed by Poidebard and Stein as well as the theoretical cost path suggested by Arcmap).

I then wanted to turn this into a network. This I did in Arcmap, by automatically creating lines from all hydrological points to all other hydrological points within 40 kilometers. I then exported all points and lines as spreadsheets, keeping information on geographic location intact. These I imported into the Graph software Gephi, using the Geo-layout algorithm plug in developed by Alexis Jacomy. Below you can see what the result looked like.

I did the same with the archaeological sites identified by earlier scholarship. Here, inspired by Cyprian Broodbank and Anna Collar’s use of Proximal Point Analysis, I added the minimum number of edges needed in order to connect nodes to their closest neighbors on all sides. This, I admit, is probably the weakest point of the analysis, as it involved a certain amount of personal judgment.

I then merged my two networks by combining the spreadsheets. This is the result, with nodes sized according to betweenness centrality. We see very clearly how the areas with good access to water, were connected by places where we find archaeological evidence in the nature of defensive structures or inscriptions, and that these nodes act as gateways, that serve to integrate the network.

Calculating shortest paths proved not to be so useful, because there are so many nodes very close to each other and because this treats minor cisterns in the same manner as large fortified stations and major wells, but the measure of betweenness-centrality gives a very good indication that there are some places you simply need to go if you want to have something to drink on your way from Palmyra to Hit. By the way, the shortest path from Palmyra to Hit is 11, coinciding very with recorded travel times of 10 to 14 days. Indicating that the proximal point approach works fairly well.

So, in conclusion. What did we learn from this, and what did Network approaches contribute with?

In terms of the identifying the trade route, it seems safe to say that Poidebard was correct in Syria and Stein was correct in Iraq. What did we contribute with then. Well, while they followed tracks on the ground, there was no guarantee that there were not other tracks around that they never saw. We have made their conclusion testable, by showing that there simply was no other feasible route if you wanted to go the whole way between Palmyra and Hit in the dry season. In that way the question about the date of the ruins in the desert becomes less important, because whether there were fortifications there or not in the Roman period, the network layout shows that the Palmyrenes needed to pass through this places.

Case study: The social networks of client-rulers in the Roman Near East

John the Baptist before Herod Antipas,
Albrecht Dürer 1509. Source: Wikipedia

This week I went to the annual meeting of the UK Classical Association, which was hosted by the University of Nottingham this year. Colleagues Leonardo Gregoratti (Durham) and Eran Almagor (Ben Gurion University of the Negev) organized a session on "the Eastern Client States", where I took part. Client states in this context refer to polities in the Near East, that held a large degree of autonomy and were rued by local princes, but which were part of the Roman Empire or the Parthian Empire. Herod the Great, king of Judea 37-4 BCE and universally infamous due to the infanticide ascribed to him in the Gospel of Matthew, is perhaps the most famous of these rulers. In fact there were many of them, and even if there is a clear tendency towards direct and centralized rule over time, the Roman Empire always remained a patchwork of cities, tribes, and principalities with varying degree of autonomy, although there was never any doubt that the real power was in Rome and later in Constantinople.

The client rulers are one of the cases I am studying, with the aim of better understanding the fabric of Near Eastern society in the Roman period. In time I plan to make a study of them for the whole period of Roman rule in the Near East, but for the presentation in Nottingham I started in an end, and attempted a network analysis of the system in the first century BCE and the first century CE. Below is a short summary of my preliminary ideas and finds. Comments and advice on how I could develop this are greatly appreciated.

I started by plotting 163 members of ruling dynasties in the Near East from 63 BCE (the start of Roman Rule) until 125 CE and the 369 ties of full siblinghood, marriage and descent between them in Excel. The entries were based on Richard Sullivan's invaluable prosopographical articles for the Aufstieg und Niedergang der Römischen Welt, bolstered with information from classical encyclopedias. The resulting spreadsheet was saved in csv-format and easily imported into the open-source graph visualization software Gephi using this great tutorial from University of Wisconsin Green Bay Digital Humanities blog. After some time spent identifying and correcting errors in my database that became evident during the import-process, i got this unprocessed graph (below). It does not immediately make much sense. The thick lines represent connections between individuals sharing more than one tie, in effect people marrying their siblings, an unusual, but not unheard-of practice among royals at the time.

The next step was to find a good way of visualizing the whole network. I used the force atlas 2 algorithm in order to arrange nodes and edges in a pattern where they did not overlap. Then I assigned the different dynasties different colors, based on the dynasties people were borne into (as opposed to those they married into. This I did by assigning different series of node id's to different dynasties in my spreadsheet, for instance all individuals belonging to the Herodian dynasty got an id-numer starting with 3. In this way I was able to easily filter out all members of this dynasty in Gephi. Now the network looked like this:

Here, the network is organized according to dynasty, showing the different connections of marriage, descent and siblinghood for the period from 75 BCE until 150 AD and colored after which dynasty people were born into. On one hand of course this is problematic, because dead and living people are included in the same network, on the other hand it is useful, as dynastic connections were used for claims to legitimacy as well as territory and position, and it helps us see which families were important local players and who were more marginal. In that sense it gives us a more comprehensive picture than the stemma we usually look at when we study dynastic networks.

Then I wanted to see how the network changed over time. The problem with this is that we don't have secure information about when all the people in question actually lived. I've tried to overcome this by assigning them quarter centuries when they were politically active, either as rulers or simply as marriage partners and parents. Some were active in dynastic politics for almost 75 years, others only briefly. By assigning each period a unique value in Gephi and using the software's partition feature I was able to create time-series of the network. I've made a short movie of these (below).

Here individuals have been sorted into overlapping 25-years intervals, according to the periods in which they were active. Some of them were political figures for three quarters of a century, others only briefly The slides show how the different dynasties engaged with each other over time, making it possible to discuss questions of integration, fragmentation and marginalisation. It shows very well, for instance how the Herodian dynasty of Judea emerges as the regional power-broker in the late first century BC, and how Armenia is constant arena of dynastic competition, where different dynasties vie for influence. Dynasties such as Emesa never really becom important, while Commagene and Cappadocia remained in the game, but were marginalized over time.

Next, I used the really useful Geo Layout algorithm developed by Alexis Jacomy in order to arrange my nodes according to geographical position (which I had included in my spreadsheet). Now all nodes belonging to the same dynasty were gathered in one point, and thus indiscernible, but instead the geographical development of the network over time became visible:

In this example we no longer see the individuals, but ties between the different dynastic centres instead. They move slightly because the scale of the network varies over time. In terms of geography, we have three main clusters, centered on Anatolia, Northern Mesopotamia and Judea, with Armenia and Commagene as not only the geographical, but also the main dynastical links between them. Also this allows us to look at interaction across the so-called border between the Parthian and Roman worlds or spheres of influence. Doing this, we see that these networks are geographically very expansive, spanning from Mauretania and Rome in the West, to Ctesiphon in Mesopotamia in the east at their greatest. We also see that the great rivals of the Romans, the Arsacid dynasty, by way of Parthia and Media Atropatene, are active participants in the dynastic networks of the Roman Near East, although they seem to become less important over time.

I had great fun while trying to model the client king system, but I also found it scholarly very rewarding. More on this at a later stage, but network perspectives allow us to move the focus from the imperial center to a multiplicity of peripheral points of view. Each of the 163 individuals in the network were at the centre  of their own world, and approaching them as a social network allows us to appreciate this in a different manner. At this stage this is very much work in progress, and I'll continue to develop the technical as well as the scholarly side of this in months to come.

Thanks to the audience and my co-panelists in Nottingham for a good discussion!