Arteries of Destruction

Map showing the destroyed structures found along a portion of the Yei-Koboko Road in South Sudan. Approximately 2000 structures were visibly destroyed by fire between 27 December 2016 and 26 January 2017 over the 27 square kilometer area analyzed. The orange icons indicate areas where large active fires were detected over the same time frame using the Visible Infrared Imaging Radiometer Suite (VIIRS) sensor. Analysis by Amnesty International

After increasing reports of large influxes of refugees entering Uganda from the Yei and Kajo-Keji regions of South Sudan, an investigatory challenge presented itself: why, specifically, are they fleeing now and how should one document those causes?

These refugees — among the over 1.5 million who’ve fled South Sudan — are bringing stories of violence from intense fighting, kidnapping, rape, and killings by armed forces. People are fleeing the country through remote routes to avoid any encounters on the main roads with these armed forces. With refugees avoiding the main arteries from Yei to Uganda, it is a reasonable first place to look for the security threats that have forced so many from their homes, and created the worst displacement crisis on the continent.

One of the roads from Yei to Uganda is the Yei-Koboko Road, also know as the Kaya highway, since it passes through the border town of Kaya. A small section of road, stretches from Ambo to Kaya over 18 kilometers. This became the focus area. Along that road, satellite imagery and other remotely sensed data sources covering 27 square kilometers were analyzed for signs of damage and destruction that might shine a light onto what is happening to these communities.


The first data analyzed was high resolution imagery captured on 27 December 2016 and 26 January 2017. Over the 27 square kilometers, approximately 2000 structures had been destroyed by fire between the two dates. These structures are pinpointed on the overview map above. Two main hotspots for damage, Ambo and Kaya, are the largest villages in the focus area, though the damage is not exclusive to those locations. Other than the obvious damage, little to no other change is visible, including changes that relate to human presence.

Ambo Village

Ambo village before the destruction on 27 December 2016 (top) and one month later, on 26 January 2017 (bottom), after a fire razed much of the village. Images: © CNES 2017, Distribution AIRBUS DS (top), © 2017 DigitalGlobe, Inc. (bottom)

Kaya Village

Kaya village before the destruction on 27 December 2016 (top) and one month later, on 26 January 2017 (bottom), after a fire razed much of the village. Images: © CNES 2016, Distribution AIRBUS DS (top), © 2016 DigitalGlobe, Inc. (bottom)

The so-called “Big Data Revolution” means investigators are more flush than ever with data. Working on remote regions of the world, I’ve often been faced with situations where data coverage of a specific geography is limited — to the point it can be frustrating to map the accurate location of a town or village. This data vacuum makes it difficult — at best — to conduct thorough investigations, and determine the ‘who, what, where, and when’ of grave human rights abuses. But as more and more data becomes accessible, these challenges decrease in many ways.

In this particular situation, while I had a good measure of ‘what’ — the destruction of some 2000 structures — I wanted to test the use of other remotely sensed data to narrow down the ‘when’ of the destruction in the focus area.

Included on the overview map above is the VIIRS 375m Active Fire Data. This sensor, aboard the NASA/NOAA Suomi National Polar-orbiting Partnership (SNPP) satellite, is able to detect fires. Based on the high-resolution satellite imagery, the 2000 structures in the Ambo-Kaya corridor were visibly destroyed by fire between 27 December 2016 and 26 January 2017 and it was possible the VIIRS sensor picked it up. Choosing the VIIRS data points in the focus area during the 30 day time frame, 36 active fires were found. When narrowed down to the highest populated areas, Ambo and Kaya, four points were found to be closest to the areas of most destruction. The dates fires were detected in Ambo were on 16 and 17 January 2017 and in Kaya on 21 January 2017.

VIIRS Active Fire Data is helpful, but in areas of the world where seasonal burning of fields is common and fields make up a large portion of the landscape, verification of what might actually be burned is needed. The VIIRS data then led me to the the next source of data to create a visual timeline for verification of events.

The company Planet has the ultimate goal of mapping the entire globe daily. Though this goal has not been fully realized yet, the coverage over remote areas of the world is quite good. Armed with the dates of fires detected by VIIRS, the Planet imagery coverage over Ambo before the dates were from 14 January 2017 and after the dates, on 18 January 2017. Comparing these two dates using false-colored infrared imagery to highlight the burned areas, it is easy to see much of the village was burned within those four days.

Ambo Village

According to the VIIRS Active Fire Data, two large fires were detected close to Ambo village — shown above with the flame icons — on 16 and 17 January 2017. This information allows researchers the ability to narrow down the time frame of the destruction. Image: PlanetScope imagery ©2017 Planet Labs, Inc.
Using the near-infrared band of three meter resolution imagery, the expansion of burned land in the Ambo area is dramatically visible when comparing the false color images from 14 January 2017 (left) and 18 January 2017 (right). False-colored infrared imagery highlights the healthy vegetation in red colors while recently burned areas appear black and gray. The yellow square shows the newly burned areas in Ambo village. Images: PlanetScope imagery ©2017 Planet Labs, Inc.

In Kaya village, the VIIRS sensor detected fires on 21 January 2017. Planet coverage over this location was from before the fire date on 18 January 2017 and after on 27 January 2017. The Planet imagery shows the area was not burned on 18 January 2017, but is visibly burned on 27 January 2016, narrowing down the time frame of major destruction in Kaya to nine days (eight if the high resolution imagery from 26 January 2017 is included).

Kaya Village

According to the VIIRS Active Fire Data, two large fires were detected in Kaya village on 21 January 2017. This information allows researchers the ability to narrow down the time frame of the destruction. Image: PlanetScope imagery ©2017 Planet Labs, Inc.
The same false-colored infrared imagery is used to compared the changes in burned land in the Kaya area. The yellow square shows the area burned in Kaya village between 18 January 2017 (left) and 27 January 2017 (right). Images: PlanetScope imagery ©2017 Planet Labs, Inc.

What does this tell us?

This information gives a visual time frame of events related to the razing of approximately 2000 structures in a 27 square kilometer area. The major destruction in Ambo occurred within a four day time period, likely over 16 and 17 January 2017. The major destruction in Kaya, happened within an eight day window, likely on 21 January 2016.

While this degree of intensive precision is welcome from an investigatory perspective, there remains much extensive research to do. Refugees are fleeing from all areas around Yei and Kajo-Keji which I estimate as over 5000 square kilometers of land. With over 2000 structures damaged in only 27 square kilometer area, what does the rest of the region look like? While it may be tempting to extrapolate, to do so would relegate the crimes occurring in the region to statistics, rather than grave breaches of law that require not only precise documentation, but justice and accountability.