Tents and Tukuls: Lessons from the Development of AMALGAM

Figure 1: Satellite imagery of destroyed tukuls in the Abyei Region, disputed border area of Sudan and South Sudan. Image (c) 2014 DigitalGlobe, Inc.

Satellite imagery of destroyed tukuls in the Abyei Region, disputed border area of Sudan and South Sudan. Image (c) 2014 DigitalGlobe, Inc.

The path of innovation is rarely, if ever, a straight line. Where you end up often bears little to no resemblance to where you initially thought your initial idea would take you. This is one of the many lessons we at the Signal Program on Human Security and Technology (Signal Program) at the Harvard Humanitarian Initiative (HHI) learned as a result of our participation in the Mass Atrocity Prevention Tech Challenge.


Our submission, AMALGAM (Automated Mass Atrocity ALGorithmic Analysis Methodology), was originally intended as a platform where satellite imagery analysts looking at evidence of alleged mass atrocities could share the raw data of their findings with each other. These individual results could be aggregated and verified on the platform with the help of open source algorithms embedded within the system.


The first feature that we had intended to focus on was different types of tents present in satellite images of refugee and IDP camps. The long-term goal of creating a platform for analyzing tents and other similar temporary structures used by displaced people was to attempt to estimate changes in the size of these populations.


What AMALGAM has become, however, is something else entirely. AMALGAM, now called the “Tukul Detector”, has evolved into a working prototype that can automatically identify and count tukuls in certain conditions.


What is a tukul, you may be asking?


A tukul is a traditional daub and wattle hut commonly found in several countries throughout East and Central Africa, including Sudan, South Sudan, Ethiopia, Central African Republic, and Somalia. When civilian populations are attacked in many African nations, one of the clearest indicators that an attack on these communities may have occurred is the apparent destruction of these buildings.


Our team refined our concept with the help of a HHI Fellow, Mike Hughes, currently a doctoral candidate in computer science at Brown University. It was Mike’s unique skills and experience that allowed us to realize that the hardest thing we had hoped to do with AMALGAM, counting tukuls, was more feasible than we had first thought.


The Signal Program is currently in discussions with NGOs and international agencies about testing the Tukul Detector in various operational contexts. We hope to have several trial runs of the Tukul Detector in the next few months as we work to produce a beta version for general use.


Meanwhile, our work around temporary structures, which began with AMALGAM, has now turned into something else entirely. Rather than trying to count tents and other emergency shelters with a computer, the Signal Program is developing a typing guide, or analyst’s handbook, to facilitate the identification of these objects in satellite imagery and on-the-ground.


While we are proud that the AMALGAM concept won third prize in the evidence collection category of the tech challenge, the Signal Program is even prouder of the way the contest forced us to think in a more dynamic and interdisciplinary way about innovation. The ongoing process of invention and iteration that the tech challenge helped support has proven to be the greatest reward.