Thesis Overview
The world is in an escalating food crisis.
Only through a deeper understanding of how food insecurity drives instability can we better secure our societies against the looming climate catastrophe.
My master’s thesis, titled ‘Unfed Unrest: Drought, Food Insecurity and Rioting in Sub-Saharan Africa’, is a small part of a growing field of quantitative research looking at links between the environment and human conflict. It served as the final part of a Master of International Affairs, specialising in Security and Sustainability, that I completed at the Hertie School of Governance in Berlin.
An overview of my major findings, as well as my research methods, are provided below, or alternatively, the full text can be downloaded at the link below.
Drought: A Driver of Human Conflict?
There is broad agreement that climate change represents a severe threat to human civilisation, through the increased frequency of natural disasters such as cyclones, droughts and wildfires. Nonetheless, the second-order effects that this could have on levels of human conflict are still up for debate.
Within this research space, much focus has been put on the impacts of drought, including an emerging finding across several studies that drought can lead to an increase in rioting. Past research has suggested that drought leads to water shortage, bringing people into conflict over scarce resources.
Food Riots: An Alternative Hypothesis
This study theorises that instead, the observed increase in rioting following drought could follow a more complex pathway. Rainfall shortage during drought leads to reduced crop yields, and subsequent food insecurity in countries with a dependence on subsistence/smallholder farming. The grievances accumulated due to this food insecurity then lead to rioting.
The impact of this is expected to be higher in Sub-Saharan Africa compared to other parts of the globe, due to the region’s reliance on subsistence or smallholder farming. Thus, the study examines 8 Sub-Saharan African countries, broken down in 120 disaggregated First-Level Administrative Subdivisions.
In the context of Sub-Saharan Africa, agricultural seasonality is an important factor. The highest likelihood of riots is theorised to occur during the so-called ‘lean season’ - the period of the agricultural year prior to harvest when food supplies are at their lowest within Sub-Saharan Africa.
Data & Methodology
-
Due to data processing constraints, the study is limited to a subset of eight Sub-Saharan African countries, distributed across multiple geographic, climatic and cultural zones. These countries included: Kenya, Tanzania, Botswana, Zimbabwe, Mozambique, Zambia, Nigeria and Ethiopia. Together the countries contain a total of 120 regions suitable for analysis, yielding 14400 individual data points in the raw dataset.
-
To operationalise the independent variable, agricultural drought, remote sensing is used to collect data on large areas of agricultural land. To achieve this, the study makes use of Normalised Difference Vegetation Index (NDVI) data from the Moderate Resolution Imaging Spectroradiometer (MODIS) instrument aboard several NASA satellites. NDVI is a measure of vegetation health obtained through a relatively simple calculation derived from measurements of red and near-infrared (NIR) wavelengths of light.
-
To derive relative crop health information from this data, these raster images are clipped to the geometries of farmlands within selected regions. These farmland geometries are obtained from Global Food Security-Support Analysis Data at 30 m (GFSAD30) a USGS/NASA project to provide high-resolution global cropland data. The GFSAD30 data is derived from a machine learning model applied to imagery from that Landsat 8 satellite, and is available as tiled raster images. This farmland raster data is vectorised and then used to clip the separate NDVI anomaly rasters, with all other non-farmland areas discarded . Then, for each month of each study period year, mean NDVI figures are calculated for the farmland areas and then compared with the 20-year (2000-2020) average for the same areas, to derive a mean NDVI anomaly figure for each month of the study period. These spatial and temporal data reductions result in monthly NDVI data points capturing the deviation in crop health from the long-term average or alternatively serve as a measure of agricultural drought.
-
Data on the dependent variable, riots in Sub-Saharan Africa, is obtained from the Armed Conflict Location & Event Data Project (ACLED). The ACLED database contains data on the locations, dates, actors and number of fatalities of all reported political violence and protest events around the world. To account for the fact that the population of the regions selected for study varied significantly, the number of riots each month is normalised for population. Population data is obtained from census data, national statistics agencies, as well as the UN. As reliable yearly population counts are not undertaken in the study regions, two population counts are found within, or slightly outside of the study period, and a linear approximation of population change is derived. From this linear approximation, yearly population counts are generated, and rioting is normalised to a final statistic of ‘Riots per 100,000 people’.
Key Findings
No evidence is found for a statistically significant relationship between agricultural drought and a decrease in rioting when applied to the year as a whole. However, when applied to two data subsets focusing on those months during the lean season, and those outside of it, a more complex picture is established.
Highly statistically significant results are found demonstrating that agricultural drought increases the frequency of riots during lean season months, with a one standard deviation decrease in mean growing season NDVI anomaly resulting in a 12.26 percent increase in monthly riot frequency.
During months outside of the lean season, the effect is reversed, with the same one standard deviation decrease in mean growing season NDVI anomaly reducing the frequency of riots by 15.92 percent.