Note: For those who haven't been keeping up with the most recent posts, this is part 4 of a fairly extensive series on network analysis techniques and the human geography of Turkey. In case you missed them, the first three posts can be found here: Part 1 (The Free Syrian Army), Part 2 (Ethnolinguistics of Turkey), Part 3 (Borrowing Analytic Techniques: Populations, Predictions And What Physics Tells Us About The Movement Of Alawites).
 |
| Figure 1. Simulating Population Movement Source: CASOS *ORA Software |
My ethnolinguistic project was no different. I quickly realized that I would have to do far more than "at least mention" the current crisis in Syria. In fact, it quickly became the focal point of the investigation.
Conceptually, the problem was simple.
How to turn Turkey, a 1000 mile-long country with 81 provinces and 957 districts, into a network. Why did I want to make it a network? So I could use it as an architecture through which I could simulate refugee population flow.
And the answer is going to surprise you (I know it surprised me).
If you are a tabletop wargamer, you will likely already have an answer to this. Ever played Showdown? Then you know what I'm talking about.
For those unfamiliar with classic tabletop wargames, they tend to overlay hexagonal grids onto geographic landscapes in order to create a game board. Showdown turns the India-Pakistan border, for example, into a hexagonal grid system on which players explore the possibilities surrounding a hypothetical, near-future, nuclear Indo-Pakistani war.
Intelligence analysts also have a way of assessing terrain of strategic importance within the context of war: The well-known analytic method Intelligence Preparation of the Battlespace. With this method, intelligence analysts look at everything from climate to elevation to road/rail infrastructure to land cover/use in order to determine the most strategic plan of attack or invasion.
Finally, brilliant scientists are using a methodology similar to mine in South Africa to predict elephant migratory patterns and calling it "resistance mapping." [1]
I, however, wanted to take a more organic approach to building the network architecture. And what better way to do so than to use the pre-existing administrative boundaries of Turkey? (A particularly ironic decision given that the original goal of the project was to replace all administrative boundaries with ethnolinguistic ones).
This is when the 957 districts of Turkey because 957 nodes in a network, and the two to seven district border crossings for each of these 957 districts became a link in the network. It looked something like Figure 2.
 |
| Figure 2. Derived Turkey Network Source: CASOS *ORA Software and ArcGIS |
Now that the nodes and links were established, there was the issue of link weights (if all links were weighted equally, the simulation would spread through the network equally in all directions, and that would be rather useless and not very predictive).
The link weights were derived for each district border crossing from six factors on a scale of one to three. The factors were a combination of the ethnolinguistic mapping, the gravity model analysis and IPB factors:
1. Terrain
2. Roadways
3. Railways
4. Expanding, neutral or decreasing ethnolinguistic group
5. Pre-existing Syrian refugee presence
6. Gravity model output (influence on area from abroad in the next 12 - 24 months)
A higher link weight was indicative of increased difficulty a refugee would experience in moving from one district to the next. A lower link weight indicated that the route would be easier to navigate for reasons reflected in the factors above. [2]
The simulation began in the five nodes representative of the seven most frequently-trafficked border crossings as determined by the UN. [3] It iterated 19 times and moved through Turkey revealing some encouraging findings along the way (See Figure 3).
 |
| Figure 3. Simulated Syrian Refugee Population Movement Prediction Over 19 Weeks Source: CASOS *ORA SOftware |
 |
| Figure 4. Known Syrian Refugee Presence in Turkey Source: UNHCR (November 2013) |
 |
| Figure 5. Known locations of Syrian refugees. Size of dot indicates number of camp inhabitants. Source: UNHCR (November 2013) |
An important element in validating the simulation is analyzing the degree to which it tracks with reality. The simulation ran directly from the border districts between Turkey and Syria and tracked with the known locations of Syrian refugees (See Figures 4 & 5). When the image of the known locations is overlaid with the image of the simulation it looks like this:
 |
| Figure 6. Simulation Overlay on Known Syrian Refugee Presence |
The two provinces outlined in white (Nigde and Aksaray) are where the Turkish government subsequently indicated that it would construct new refugee camps to relieve camps operating at capacity within the region. More confirmatory evidence.
Now fast forward to two months later. The date is 18 February and, according to the UNHCR, the refugee population has moved further into the provinces indicated in Figure 7. Reality is unfolding within the lines of the simulation from which analytic predictions were drawn back in December 2013.
 |
| Figure 7. Starred Provinces (Left to Right): Konya, Karaman, Mersin, Batman (Yes, that is correct, there is a Turkish province called Batman) |
All of this, from methods and process to final product analysis, led me to Istanbul (or just outside, as it were) where stories from Syrian refugees both challenged and confirmed my predictions.
Be sure to check out the next installment, What The Syrians Had To Say: Views from Istanbul.
***
[1] The article, Computational Tools in Predicting and Assessing Forced Migration, employs much the same methodology as my approach, but utilizes an artificial hexagonal grid as opposed to the more organic network architecture that I used. It is an excellent read!
Edwards, S. (2008). Computational tools in predicting and assessing forced migration. Journal of Refugee Studies, 21(3), 347-359.
[2] These numbers were later inverted for the purposes of the simulation as, in a network, a higher link weight value indicates a stronger relationship. The numbers were inverted so that, the lower the link weight, the more difficult it would be for a refugee to move there and the less likely the simulation would be to move into that region as a result.
[3] Five nodes represented seven border crossings because each node represented one district. Two of the seven most frequently-trafficked border crossings were located in the same district as another border crossing, therefore only five nodes were necessary.
