Climate, Food, and Conflict: A New Approach to Risk Assessment
In a recent one-year pilot project for the Food and Agriculture Organization (FAO) in Rome, Uncharted Waters* have developed and tested an innovative approach to understanding and predicting climate-related conflict risks. The project specifically focuses on how changes in our food systems, driven by climate variability, might contribute to conflict risk in vulnerable regions.
Why This Matters
The spread of conflict coupled with climate and economic stressors is pushing millions of people to the brink, with acute food insecurity set to increase in both magnitude and severity according to a recent FAO report. Evidence of a direct relationship between climate, food insecurity and conflict, however, is patchy and incomplete. While we know that environmental degradation and climate change can affect resource availability and threaten livelihoods, the link to conflict is far from straightforward. Understanding these complex relationships is crucial for organizations like FAO to develop "conflict-sensitive" climate adaptation strategies in the agriculture sector.
Enter CoPro: A New Tool for Risk Assessment
At the heart of our pilot project is CoPro, a machine learning model designed to predict conflict probability. What makes CoPro unique is its ability to process diverse spatial data alongside conflict records to forecast the likelihood of future conflicts. The model doesn't just look at traditional conflict indicators – it also incorporates climate and food system data, making it particularly valuable for understanding how climate change might affect conflict risk through its impact on agriculture and food security.
Our simulated probability of conflict for the reference period, and for the year 2022.
The Data Challenge
A critical aspect of this project was the selection and integration of indicators. We identified 38 unique indicators across five categories: conflict, community, governance, climate & water, and food. Ten of these indicators specifically relate to food systems, covering different dimensions of food security from availability to stability.
We faced an interesting challenge when working with agricultural data. We compared three different sources of data: national crop production statistics from FAOSTAT, remote sensing data on vegetation health from FAO-ASIS, and crop yield model simulations from the LPJmL model. Each source has its strengths and limitations. For instance, while FAOSTAT provides comprehensive national-level production data, affected by both climate and socio-economic impacts, it is made available with a two-year lag. Remote sensing offers near real-time data but currently only covers cereal crops. In large parts of Africa root crops like cassava and fruits and vegetables crops like beans and plantains form an important part of the diet. The LPJmL model has no lag and can be used for future risk projections, but needs improvement in capturing year-to-year variations in yield in Sub-Saharan Africa. Interestingly, only in a few regions across Africa did these data sources agree on the years with the lowest yields.
What We Learned
Our pilot project revealed several crucial insights about predicting conflict risk:
1. Definition Matters: How we define conflict significantly affects the model's performance. When looking at conflicts with 50 or more fatalities, the model struggled due to limited data points. However, it performed much better when including smaller-scale conflicts (those with one or more fatalities), partly because there was more data to learn from.
2. History Matters: For smaller-scale conflicts, previous conflict in an area emerged as a dominant predictor. This aligns with existing research suggesting that conflict often breeds more conflict.
3. Food Systems Matter: When we removed agricultural indicators from the model, it relied more heavily on existing conflict to predict future conflict. While this increased the model's recall (ability to identify actual conflicts), it reduced precision (more false positives). This suggests that food system indicators help refine our understanding of conflict risk.
4. Contex Matters: The relationship between climate, food systems, and conflict varies significantly by region and context. What might trigger conflict in one area might not have the same effect in another.
Looking Forward
Our pilot project demonstrates both the potential and challenges of using machine learning to understand climate-related conflict risks. While CoPro shows promise in identifying areas of elevated conflict probability, its success depends heavily on how it's configured and what data is available.
The findings highlight the importance of having reliable, up-to-date agricultural data. We found that crop simulations could be improved by better accounting for local farming practices, such as planting dates and growing periods. This would be particularly valuable in Sub-Saharan Africa, where agricultural data is often limited.
Practical Implications
For organizations like FAO, this work provides a foundation for developing more sophisticated early warning systems that consider both climate and food system impacts. However, we emphasize that the model should be seen as one tool among many for understanding conflict risk. Its predictions should be interpreted alongside other analyses and local knowledge.
Our pilot project also underscores the need for better data collection and monitoring systems, particularly in vulnerable regions. Improving our ability to track and understand changes in food systems could significantly enhance our capacity to anticipate and prevent conflicts.
Conclusion
While this one-year pilot project has demonstrated the feasibility of incorporating food system indicators into conflict risk assessment, it also highlights the complexity of the challenge. As climate change continues to affect food systems globally, tools like CoPro could become increasingly valuable for organizations working to prevent and mitigate conflicts. The next steps will involve refining the model, improving data quality, and testing its applications in different contexts.
Our work for FAO represents an important step forward in understanding how climate change, food systems, and conflict intersect. It provides a foundation for future research and practical applications in this critical area of study.
* Uncharted Waters teamed up with Sophie de Bruin and Jannis Hoch, developers of the CoPro model.