Submitted by Sander Houston, 510 the Netherlands Red Cross
22 Jun 2021 , last updated 22 Jun 2021

Forecasting changes in the state of food security with machine learning

Food insecurity is a growing concern due to man-made conflicts, climate change, and economic downturns. Forecasting the state of food insecurity is essential to be able to trigger early actions. To measure the actual state of food insecurity, expert and consensus-based approaches, and surveys are currently used. Both require substantial manpower, time, and budget. Joris Westerveld (currently Data Scientist at TNO Defense, Security and Safety, formerly with 510 and Utrecht University) was the lead author of a peer-reviewed paper, published in April 2021, that explored the use of machine learning to forecast the transitions in the state of food security. This blog post summaries the key findings from the study and highlights the relevance for the anticipatory action community. Additional authors are Marc van den Homberg, Gabriela Nobre, Dennis van den Berg, Aklilu Teklesadik, and Sjoerd Stuit.
 

Machine learning model            

The paper introduces a machine learning model to forecast monthly change in the state of food security in Ethiopia, at a spatial granularity of livelihood zones, and for lead times of one to 12 months, using open-source data. The change in the state of food security is represented by the differences in Integrated Food Security Phase Classification Data from one month compared to another.     

 

Predicting food security

From 19 categories of datasets, 130 variables were obtained and used as predictors of the transition in the state of food security. The predictors represent changes in climate and land, market, conflict, infrastructure, demographics and livelihood zone characteristics. The most relevant predictors are found to be food security history and surface soil moisture. Overall, the model performs best for forecasting deteriorations and improvements in the state of food security compared to other machine learning algorithms and baselines.              

The proposed method performs at least twice as well as the best baseline for a deterioration. The model performs better when forecasting long-term (7 months) compared to short-term (3 months). Combining machine learning, Integrated Phase Classification ratings from monitoring systems, and open data can add value to existing consensus-based forecasting approaches as this combination provides longer lead times and more timely updates.             

 

This graphic shows a summary of the paper Forecasting changes in the state of food security with machine learning
Visual summary of the paper

Next steps

“I am excited that this project shows the added benefit of using AI and open data to assist and support the humanitarian sector. The Red Cross (or more general: humanitarian organizations) can now explore ways to pilot the model as part of food security and/or drought anticipatory action and replicate the model to other countries as most of the data on the predictors is also available for other countries”, says Joris Westerveld, the leading author of the study. “510 is very happy that we have created a scientific evidence-base for our food security forecasting model. This will enable us to better support Red Cross National Societies that work on forecast-based financing for food insecurity”, says Marc van den Homberg, Scientific Lead Data for Disaster Management at 510. Read the full article here.

 

The link with the Forecast-based Financing for Food Security project

The forecasting model of “transitions in the state of food security” was produced in alignment with the Forecast-based Financing for Food Security (F4S) project. The F4S project was concluded in May 2021 and aimed at developing information that enables the triggering of early actions to reduce the risk of food insecurity in Ethiopia, Kenya, and Uganda. For achieving this aim, the F4S project centered its developments around three pillars: (1) Forecasting key drivers of food insecurity, (2) Collection of local evidence and (3) Evaluation of cash transfer mechanism. All information obtained through the F4S pillars attempted to address challenges inherent to decision-making based on forecasting information, with a special focus on the implementation of ex-ante cash transfers. “What I found most exciting about the F4S project, was the opportunity to link science and practice to deliver insights on how early action can be designed and delivered based on the perspective of beneficiaries in combination with the support of predictive analytics”, says Gabriela Nobre, coordinator of the project.

If you are interested in learning more on how the F4S project has contributed to the development of new insights for the anticipatory action community, check out the F4S final report. The F4S project was born from a partnership among the Vrije Universiteit Amsterdam, the 510 initiative from the Netherlands Red Cross, the Climate Hazard Center at the University of California, Santa Barbara, and the ICHA/Kenya Red Cross. The project received grant through the World Bank Challenge Fund with funds from the Global Facility for Disaster Reduction and Recovery, the Foreign, Commonwealth & Development Office, and the Centre for Global Disaster Protection.

 

Title image © Uganda Red Cross Society