The fighting between David Yau Yau’s forces and the SPLA in the Pibor area over the past few months has me thinking about environmental factors influencing conflict dynamics in South Sudan. Jonglei’s swamps along the borderlands between Murle and Nuer territory have always proved volatile areas – perhaps more so now than ever since the early-mid 1990s, when the White Army was last on the move and a rinderpest epidemic had killed off many of the area’s cows, pushing cattle herding youth to raid other areas to regain lost wealth. Conflict analysts in South Sudan have long pointed to seasonal trends in conflict: Armed conflict heats up in the dry season because conventional army can access the swamps where militias or rebel groups are active; ethnic conflict and cattle raiding also increases in the dry season because of the accessibility of specific areas.
By 2011 I had spent several months over a course of three years compiling what I consider a fairly comprehensive dataset on media-reported armed conflicts in Sudan and South Sudan. I downloaded the Sudan Armed Conflict Location Event Dataset (from the ACLED website) as a skeleton, deleted all duplicate events and selected only those qualified as “armed conflict” between groups of people rather than armed crimes. I pored over books, journal articles, NGO/IGO reports from Small Arms Survey and others, and hundreds of news articles in local and international news sites to verify reports and add others that were not included in the dataset. Breaking down the data, I found no correlation between the season (wet vs. dry) and the number of reported conflicts. However, as we will see below, dry years tended to have more conflict than wet years.
Looking at overall trends, the mean center of conflict shifted northward from 2005-2011 and the directional ellipse centered along an axis through the region in which Dinka, Nuer and Murle ethnic areas meet. Reported conflicts within 40 km of the Dinka-Nuer ethnic boundary (much of this is along the Warrap-Unity border; see the full paper below for other map sources) rose significantly between 2005 and 2011. For those of you who have not visited this border area, much of it is a flat floodplain interspersed with areas of higher ground on which villages are set. Especially along the Warrap-Unity border, the floodplains are miles wide and create a lot of difficulty in accessing the opposite area during the wet season or in wetter years. I proceeded to investigate whether there was some correlation between the surface water coverage and the number of conflicts.
After organizing and mapping the conflict events I had recorded, I downloaded 250m MODIS 8-day composite images from the end of each wet season, selecting the clearest two images for each year from between September 20 and October 20 – basically taking the two clearest images that were taken as soon as possible after the rainy season cloud cover cleared enough to allow what appeared to be an accurate ground cover map. I used known bodies of water to reclassify imagery to show flooded areas, and checked this against 15m Landsat imagery, which for the most part lined up pretty accurately.
Based on this satellite imagery, we can get an idea of how much the surface water coverage at the end of the rainy season fluctuates each year, indicating the amount of rain that has fallen on that area in the late wet season and how much water is left in the toich for cattle herds to utilize during the drier months.
After classifying the imagery for all of South Sudan, I calculated the area of South Sudan that had surface water remaining at the end of each rainy season, providing an idea of overall dry years vs. wet years. I then compared each year’s “wetness” with the number of conflicts that year for every year since 2005, since that is when we could begin to differentiate ethnic conflict and Southern rebellions that had ethnic dynamics from the Sudanese Civil War in the media reports. Keep in mind that from 2005-2011 is only 7 years, a very small population for a statistical study. Nevertheless, there was a decent negative correlation between surface water coverage and the number of conflicts during the year, indicating that conflict is more likely to occur during dry years than during wet years. This makes sense for a couple of reasons: 1) floods keep warring groups separated; 2) dry years bring pastoralist groups into closer proximity as they search for water for their herds; and 3) it could also be the case that conflict reporting is poor during wet years due to inaccessibility.
Taking it to the next spatial level, I broke it down by county and found that the counties in which armed conflict correlates negatively with surface water coverage were highly clustered, mostly in counties along the Dinka-Nuer-Murle ethnic boundaries.
That makes sense because those areas between Abiemnhom and Bor Counties are largely flooded during wet years and therefore are less accessible. The stats indicate that the trend toward conflict in these areas is significantly clustered enough that it is nearly impossible that it is random, which indicates that we could predict that conflict is more likely to happen in these areas during dry years, rather than just during the dry season.
Because of the small sample size, in 2011 when I worked on all of this, I was unable to find seasonal trends. With a few more years of data tucked away, I think there is good potential for updating this study. What I determined from the study is that there is an apparent relationship between the number of armed conflicts each year and the wetness or dryness of certain areas of South Sudan. I also found that the data for water and conflict locations could be meaningful compared based on a theory about accessibility related to pastoralist groups herding cows toward water sources during dry years, and water creating barriers between groups during wet years. Using MODIS composites seemed to work well for determining surface water coverage. I hope this provides a good basis on which to build a future study of environmental influences on conflict.
You can download the whole study below.