Can we monitor climate and food crisis risks in real time?
According to the 2021 Global Report on Food Crises (GRFC), 155 million people in 55 countries and territories in 2020 faced acute food insecurity and were in need of urgent assistance, the highest number since this type of global reporting started. The economic consequences of the COVID-19 pandemic was an important driver of the recent surge, conspiring with the structural drivers of food crises -- conflict, economic shocks, and weather extremes. These drivers are mutually reinforcing and are expected to intensify in the coming years. Continued environmental degradation and more intense and frequent weather extremes will put greater pressure on food systems, shock local economies (e.g., through food supply shortages and food price spikes), and increase the risk of conflict over resources. There is thus high risk that many more people will face acute food insecurity in the coming years. The rural poor will be disproportionately affected as they heavily depend on natural assets for their livelihoods.
The role of climatic shocks as “threat multipliers” is increasingly recognized. In November 2021, for instance, in an unprecedented communiqué, the African Union Peace and Security Council called for immediate action to mitigate the climate security nexus in Africa, reiterating the importance of “…adopting a climate-sensitive planning dimension in peacekeeping and post-conflict reconstruction and development efforts to prevent any relapse to armed conflicts in fragile communities, while ensuring its incorporation in national and continental early-warning activities”.
Preventative action requires adequate early warning systems that alert to any heightening of the associated risks. This will require robust, solid, and timely data and evidence-based analysis to inform policymaking and actions that will build resilience to weather extremes, help avoid food crises, and reduce the risk of conflict. This note takes stock of some recent promising approaches that could help come to real-time monitoring of food crisis risks, focusing on climate-related risk factors. Below we summarize the efforts by several agencies and academia, including risk models and climate data developed by The World Bank, the World Food Program (WFP), IFPRI’s Food Security Portal (FSP) with the NASA HARVEST, the CGIAR Climate Security Crisis Observatory, and several US-based universities.
In an effort to bolster innovation around food security analytics, The World Bank explored the use of machine learning methods for the early detection of food crisis outbreaks. It developed a machine learning model for 21 countries to monitor food crisis risks at sub-national levels using food price data, conflict data, and satellite imagery on important weather and climate-related variables. To explore how such tools can be used to guide anticipatory financing decisions during times of elevated food crisis risks, together with the Food Security and Nutrition Analysis Unit (FSNAU) and the United Nations Office for the Coordination of Humanitarian Affairs (UN-OCHA), it ran a pilot real-time monitor of food crisis risks and estimated the need for financing from the UN’s Central Emergency Response Fund to address expected impacted from the 2018-2019 drought in Somalia. The study found that many major crisis events could be forecasted based on drought indicators. In a continuation of the modeling efforts, Balashankar et al. (2021) show that the predictions of acute food insecurity risks can be further improved by combining the satellite and food price data and using artificial intelligence methods that enable natural language processing of news streams. This study finds that news mentions of weather conditions preceding food crises and that news mentions of pests and diseases provide early warning signals for crop failures before they are visible in vegetation indices. A second advancement is outlined in another World Bank paper entitled Stochastic Modeling of Food Insecurity. The study uses the Predicting Food Crises database to develop a predictive, statistical model that identifies drivers of food insecurity and can be used for multi-year scenario analysis and risk assessment. From 1,670 candidate variables, the 30 most important drivers are selected. Of these, the weather-related variables include rainfall, normalized difference vegetation index (NDVI), evapotranspiration, and evaporative stress index. These weather-related variables are found to be the best predictors of food crises that occurred in contexts facing moderate food insecurity levels in the recent past, while conflict-related variables are better predictors of food crises in areas facing high initial levels of food insecurity. Food price shocks appear as early signals of imminent crises in both contexts. The latter finding hints at the importance not only of real-time monitoring of food prices, but also of timely assessments of how climate-related shocks impact food prices.
The World Food Program (WFP) is undertaking important work on nowcasting food insecurity, reflected inter alia in the paper Nowcasting food insecurity on a global scale The WFP uses a machine learning approach to predict the prevalence of people with insufficient food consumption, in order to identify where urgent humanitarian assistance is required. The model has strong predictive power in anticipating low levels of food consumption. The approach identifies weather-related variables, such as anomalies in rainfall and the NDVI, as main drivers of declines in food consumption in vulnerable contexts, such as Ethiopia.
The IFPRI-facilitated Food Security Portal has developed tools for tracking food crisis risk, including the control panels for food price and food crisis risk monitoring and the Early Warning Systems Hub. These tools bring together existing data on the drivers of food crises and the alerts produced by various early warning systems. In addition, the Food Security Portal is collaborating with the World Bank and NASA HARVEST to improve World Bank’s forecasting model, utilizing high-frequency data on daily forecasted NDVIs at high-image resolution. A recent study conducted by the Food Security Portal and Cornell University finds that while applying CNN ( Convolutional Neural Network)s to read Google imagery has potential, access to such imagery is still limited and lacks accuracy to reliably predict levels of poverty among poor communities. Instead, the study finds greater predictive power of changes in levels of consumption and asset holdings when using estimates of the NDVI at 250-meter resolution in the cases of Malawi, Nigeria, Rwanda, Tanzania, and Uganda. The NDVI is seen as a meaningful indicator of poverty in such contexts, as most poor depend on agriculture and the NDVI signals crop health and prospects for good or bad harvests. As such, this approach opens the possibility of real-time, dynamic poverty (and food insecurity) mapping at low cost.
The CGIAR’s Climate Security Crisis Observatory brings together CGIAR research and evidence on the “climate security nexus”, providing real-time data to inform humanitarian action and sustainable development policies. The Observatory pays special attention to how climate shocks influence the risk of conflict, how conflict undermines resilience to climate shocks and how these factors affect land, water, and food systems.
Researchers at the University of Texas and University of Illinois model three poverty measures as functions of weather and price fluctuations, controlling for geographic variation and seasonal trends in Malawi. They used weather-related variables such as precipitation and soil quality along with market prices and demographic data to measure the reduced coping strategies index (rCSI), the household dietary diversity score (HDDS), and the food consumption score (FCS). To extend this work to real-time assessment, they applied a statistical regression model to predict rCSI, HDDS, and FCS and obtain predictive accuracy of between 65% and 88% for the case of Malawi. In 2021, University of Illinois at Urbana Champaign developed a prototype model to predict food insecurity in Malawi, Tanzania, and Uganda. The model was estimated using data on precipitation, temperature, market prices, soil, quality, and geographic indicators. Weather-related variables including the first day of rain, length of dry spells, and others were based on calendars of growing seasons of the Famine Early Warning Network (FEWS NET). The study finds that food prices, assets, and weather-related variables all have strong predictive power in explaining levels of food insecurity.
As the 2021 Global Report on Food Crises points out, the drivers of food crises often co-exist and reinforce one another. Without diminishing the importance of the other drivers, we focused here on climate variability to highlight its central role in food crisis risk since climate change-related weather events affect other drivers such as food price shocks and conflict.
Existing early warning systems like FEWS NET and the Integrated Food Security Phase Classification (IPC) rely on in-person data gathering and administrative data on food insecurity indicators. While generally leading to good identification of food security outcomes, this type of data gathering is costly and time-consuming, and the information obtained is mainly post-factum. Great progress is being made to come to more timely and cost-effective real-time monitoring of the underlying risk factors. This should help strengthen existing early warning and early action systems and inform preventative action. While promising, much more collaborative work in this direction is needed to come to effective, credible, and actionable approaches.
Brendan Rice is a Research Analyst with Markets, Trade, and Institutions Division in the International Food Policy Research Institute (IFPRI). Rob Vos is the Director of the Markets, Trade, and Institutions Division at IFPRI. Soonho Kim is a Senior Data Manager with Markets, Trade, and Institutions Division in the IFPRI. Grazia Pacillo is a Senior Economist co-leading CGIAR Climate Security FOCUS/Climate change resilience, food security and agriculture in the Alliance of Bioversity International and CIAT. Bo Pieter Johannes Andree is an Economist at the World Bank. Michael Humber is an Assistant Research Professor in the department of geographical science at the University of Maryland. Betina Dimaranan is a Senior Research Coordinator with Markets, Trade, and Institutions Division in the IFPRI. Yanyan Liu is a Senior Research Fellow with Markets, Trade, and Institutions Division in the IFPRI. Peter Läderach is the Leader of the Climate-Smart Technologies and Practices Flagship of the CGIAR Research Program on Climate Change, Agriculture and Food Security in the Alliance of Bioversity International and CIAT. Boram Min is an Intern with Markets, Trade, and Institutions Division in the IFPRI. John Keniston is a Faculty Specialist in the department of geographical science at the University of Maryland.
Originally published on the International Food Policy Research Institute's (IFPRI) Food Security Portal.