Category Archives: Crowdsourcing

My brother Brice just sent me a very interesting study that combines satellite imagery and field reporting to analyze Kenya’s 2008 election violence. The peer-reviewed piece is entitled “Violence and Exodus in Kenya’s Rift Valley, 2008: Predictable and Preventable?” and was pub- lished in the Journal of East African Studies.

Given the use of satellites to monitor the referendum in Sudan, this blog post reviews the methodology and insights gained from the Kenya analysis. I’ll do this by providing key excerpts from the study along with my own commentary. This case study is of particularly interest to me since I was in Kenya the time and because that was when the first Ushahidi platform was launched. For more information on the use of satellite imagery to document human rights abuses, I highly recommend Amnesty International’s Science for Human Rights Explorer.

I wasn’t aware how much scrambling for information was going on in the humanitarian community:

“Over the first days, and then weeks following the December election, information about the outbreak and extent of violence was fragmented and difficult to access. Even those tasked with responding most rapidly to violence and displacement faced problems in interpreting information that was frequently distorted by rumour and misinformation.”

Interesting to know that humanitarians were facing some of the same challenges as crowdsourcing presents. Would using SwiftRiver have made a difference to try and assess the validity of the information they were collecting?

“In the early days of January 2007, UN agencies and other humanitarian bodies had numerous sources reporting that tens of thousands of people had been displaced and dozens killed across the country, yet details on the extent, location, and chronology of the violence were hard to establish, making it difficult for these agencies to plan an effective response.”

Note the need for location and time-stamped information. Would drawing on reports from the Ushahidi platform have helped? See my co-authored study on Crisis Mapping Kenya’s Election Violence. That said, this was the first time that Ushahidi was deployed in Kenya so the reports may not have been of the highest quality.

“The Kenya Red Cross Society (KRCS), for example, implemented the election contingency plan it had put in place prior to the December polls, but staff could not confirm reports of violence, and could not deliver essential food and relief items to those people displaced by the fighting because of roadblocks mounted by protestors. Even after carrying out a helicopter assessment mission on 1 January 2008, the KRCS still found it difficult to present an overall picture of the location and timing of the violence.”

So UN agencies in Kenya turned to satellite imagery.

“In response to the challenges facing them in January 2007, UN agencies in Kenya asked UNOSAT to produce a series of maps showing the likely location of election-related violence in the west of the country. UNOSAT has a variety of satellite imaging data available to them, but one tool used to map conflict situations is data on active fires. Fire plays an important role in forcing people from their homes and terrorizing local populations, so the location of active burn sites in a conflict zone offers a reasonable indicator of where violence and displacement is occurring.”

“Indeed, upon examining available fire data from Kenya for 27 December to 3 January, staff at UNOSAT noticed unusual patterns of fires on tea plantations—areas where fire is never normally employed for agricultural management. They then carried out further analysis, and created maps of areas where, according to a chronological and spatial evaluation of the fire data, it was ‘probable that a majority of detected fires are directly or indirectly linked to the civil unrest’.”

“The result was five maps covering a portion of Rift Valley Province from Nakuru to Kitale, as well as the eastern edges of Nyanza and Western Provinces. Map 1 provides an aggregate view of all active fire locations from 27 December 2007 to 3 January 2008.”

“Maps 2-5 show fires on specific days during that period. Each of the diamonds on the maps represents an area of a square kilometre that contained an active fire location at the time a satellite passed overhead. Fires generally have to cover an area of about fifty square metres to be noticed by this technology, though intensity can affect this. The colouring in the background, on the other hand, is a function of the relative clustering of active fire locations—purely a tool to direct the map-reader and not an indication of fire intensity.”

“Apart from demonstrating the geographical dimensions of the arson and conflict occurring in the area, the maps also begin to provide a general chronology of events into which more specific accounts from witness testimonies and other sources can be integrated. UNOSAT’s four chronological maps (Maps 2-5) cover the majority of the eight day period: 27-28 December, 29-30 December, 1 January, and 2!3 January, and provide powerful visuals of how events unfolded.”

“It is important to bear in mind that gaps in data collection occurred on 31 December and 2 January, and that satellite imagery captures what is happening in a particular fraction of a second—data acquisition times generally occurred around 10:30 a.m., 1:30 p.m., and 11 p.m. Kenyan local time. Bearing this in mind, it is possible using the maps to begin to understand the broad pattern of the escalation of the violence over the period.”

The authors of the study point out some limitations:

“These maps are visually compelling, but we should note that they ‘hide’ important dimensions of the violence—on a map all fires look the same. Violence in urban areas, for instance, differed markedly from that in rural areas, and these maps do not represent this difference. Another example is how little these maps reveal about the increasingly serious situation in Mt Elgon.”

This limitation is inherent to static maps, not so for live crisis maps that are interactive and dynamic.

“The mark of a single fire in the southern part of the district included on two of the maps does not stand out from the other fire locations. However, we know from other sources that violence in Mt Elgon continued to increase in severity after the elections. If violence was occurring on the Chebyuk land settlement schemes on Mt Elgon at this time, then it did not involve fires of sufficient magnitude to be detected by this satellite technology.”

“The mapping of fires can therefore tell us only part of the story. Cohesive explanations of specific situations can only begin to emerge if we triangulate the evidence provided by the maps with other kinds of information. This research is continuing, but at this stage we can offer a preliminary analysis that highlights several significant points:

• Even before the first wave of violence in the Rift Valley was sparked by the announcement of the presidential poll result on 30 December 2007, conflict had already broken out in some areas over the two days between the closure of the polls and the announcement of the presidential result. This correlates with evidence in media and human rights reports that some majimboist activists planned violence after the election regardless of the outcome of the vote.

• Over the hours following Kivuitu’s announcement of Kibaki’s victory, violence broke out in several different locations across the province, some of this undoubtedly a spontaneous reaction to the alleged ‘theft’ of the election, and targeted against persons associated with the PNU and its allies. However, many other attacks were evidently planned and orchestrated. Kikuyu-settled areas of Eldoret were ablaze within two hours of Kibaki’s re-election, armed Kalenjin men arriving in lorries to carry out the attacks. These assaults were not confined to ‘aliens’, but included attacks upon properties owned by ex-president Moi and his close Kalenjin associates, including Nicholas Biwott, whose KANU party had made an electoral pact with PNU.

• The locations of this first major wave of violence in the first week of January show a clear spatial pattern: the outbreaks were invariably in places where non-indigenous populations were living. The targets of this violence were predominantly Kikuyu and Kisii communities, who were identified as PNU supporters. Though many attacks were murderous, the main purpose was to ‘chase away’ the victims. By 6 January, the Kenya Red Cross estimated a national figure of 211,000 persons internally displaced in violence since 30 December, the vast majority of these being within Rift Valley.

• The violence accordingly coalesced in two types of location: the first was larger and smaller towns, where populations are ethnically more mixed and where businesses are concentrated—for example the rapid upsurge of conflict in and around Eldoret. The second was on rural settlement schemes, where land has been purchased or leased by farmers from a wide range of ethnic groups—for example, Burnt Forest, Ndalat, and the Molo area of Nakuru District. The settlement schemes at Burnt Forest, the scene of dreadful violence in the 1990s, were completely cut off by road barricades by the morning of 1 January, impeding the work of relief agencies, in what was clearly an organized and coordinated assault.”

This study clearly shows the added value of combining satellite imagery analysis with reporting from the ground. This analysis was all carried out retroactively, however. To this end, lets hope that the Satellite Sentinel Project, which I blogged about here, and Sudan Vote Monitor, which uses the Ushahidi platform, will be sharing information to allow for near real-time integrated analysis.