Climate provides weak power to predict African violence

Two boys from the Local Defence Unit (LDU) in Kitgum, Northern Uganda, whose job is to protect the people at a refugee camp from attacks and kidnappings by the Lord’s Resistance Army, which has been responsible for much violence in East Africa in recent years. Credit: John & Mel Kots/Flickr

Rainfall and temperature changes are linked to conflict in East Africa, but have less power to predict violence than links to political, economic and geographical factors. That’s according to one of the most detailed studies into climate-violence relationships yet, done by John O’Loughlin from the University of Colorado, Boulder, and his team-mates. “Fears of climate wars across Africa are exaggerated,” John told me. “Any effect of climate change will likely be localized and subject to other conditions.”

One political argument made about our changing climate is that it will bring more violence, particularly in Africa. For example, in 2009, US President Barack Obama told the United Nations that a warming world represents an “urgent, serious, and growing threat” because “more frequent drought and crop failures breed hunger and conflict”. But, with a background of looking in detail at where violence happens and having studied African conflicts before the 1990s, John was concerned that the evidence didn’t back such statements. That’s even though other researchers have reported statistical links between climate and conflict. “I believe that previous studies were limited by data problems and also that the policy discussions were not connected to the research findings,” John explained.

John’s team described how they fixed this in a paper published in the research journal Proceedings of the National Academy of Sciences of the USA on Monday. Previous attempts to study climate-violence have either lacked detail, looking at data per country, per year, or have been too narrowly focussed to allow generalisations, they wrote. John added that this type of work needs detailed data on both conflicts and all the factors that might predict it. Collecting the information needed on conflicts as well as political, health, location and other possibly predictive data was the biggest challenge in the work, he said.

Putting violence on the map

The researchers focused their efforts on nine countries in East Africa, stretching from Eritrea in the north to Burundi in the south, broken up into 402 100 km x 100 km grids. For the conflict data his team turned to the Armed Conflict Location and Event Dataset (ACLED), which contains the locations of 16,359 violent events for East Africa from 1990 to 2009. A group of undergraduates at John’s university also trawled online information sources to fill in details of individual violent conflicts in East African countries in 1990-1997. For the 13 possibly predictive factors other than climate, such as infant mortality, distances to roads and political rights, the researchers used other existing databases and maps.

The distribution of ACLED violent events for 5 year periods in the nine countries of the study area. The larger the red circles, the more violent that grid square in that period.The devastating civil conflicts in Rwanda (early 1990s) and Burundi (throughout the study period), the Ethiopia–Eritrea border war (1998–2000), the diffusion of violence into eastern Democratic Republic of Congo (DRC) from Rwanda, the Lord’s Resistance Army war in northern Uganda and surroundings, the Ethiopian invasion of Somalia in mid-2006, the civil conflict in Somalia involving the Al-Shabab militias, and the Kenyan electoral violence of early 2008 are easy to see on the maps. Border areas adjoining the nine countries of study (in Sudan, Zambia, DRC, Malawi, and Mozambique) are included in the analysis and have also seen violence. Copyright National Academy of Sciences.

The climate data were the easiest to get, John said, thanks in part to help from teammates from the US National Centre for Atmospheric Research (NCAR), also in Boulder. Comparing temperature and rainfall changes from the long term average over six month periods against conflict data using common statistical methods did find some links. Wetter periods decrease the risk of violence, though drier and normal periods show no effects. Much warmer periods than normal raise the risk of violence, though average and cooler temperatures have no effect. These effects pass the key test of statistical significance that scientists look for, however John underlined that this doesn’t show how strong the link is.

Modest power

University of Boulder, Colorado’s John O’Loughlin. Credit: University of Boulder, Colorado

John’s team therefore went on to cross-check how well each set of data they collected predicted violence. Their checks first looked at how often each factor predicted conflicts correctly. They then examined how important the factors are when combined with others to create a prediction. If removing one factor from the combination worsens the accuracy more than when any of the others are removed it is the most important. Removing rainfall and temperature had almost no effect on predictive power, while population and distance to the capital, border and roads were much more powerful. “Compared to the socio-demographic and geographic variables, the climate effects are ‘modest’,” John said.

The researchers worked hard to ensure that this set of results was strong, breaking down their findings by country and year, and finding large variations between them. They also looked at times when a climate pattern called El Niño was in effect, and the impacts of extreme weather. Another key test was replacing their conflict data set with one collected by Uppsala University in Sweden. “The basic argument holds – that climate effects are modest but measurable,” John said. He now hopes that these findings will be remembered when people include climate change’s link to violence in talks. “Policy discussions about the effects of climate change on conflict are simplistic compared to the complexity of the relationships and the variegated experiences across Africa.”

Journal reference:

O’Loughlin, J., Witmer, F., Linke, A., Laing, A., Gettelman, A., & Dudhia, J. (2012). Climate variability and conflict risk in East Africa, 1990-2009 Proceedings of the National Academy of Sciences DOI: 10.1073/pnas.1205130109