Modeling food crises: Looking at a complex problem through two lenses

Food insecurity is complex problem with profound and long-lasting humanitarian consequences. Last year, more than 820 million people were undernourished and at least 130 million were estimated to be in food crisis. Although humanitarian organizations frequently react swiftly and effectively once a crisis is declared, crisis declarations are made based on increased mortality which means irreversible damage has already been done.

Getting a head start on anticipating looming food crises is the common goal of two recent PRWP publications. The work explores the capacity of readily observable data and statistical models to provide risk estimates at alternative time horizons and with different geographic detail. “Predicting Food Crises”, predicts local outbreaks of food crises through random forests (RF). “Stochastic Modeling of Food Insecurity” models the distribution of a country’s population across different levels of food insecurity using panel vector-autoregressions (PVARs). Both papers use Integrated Food Security Phase Classification (IPC) system reported by FEWS NET, for 1162 districts in 21 countries since 2009. Both papers generate predictions of food insecurity using a set of covariates including remote-sensing data on environmental factors relating to food production; the incidence of violent conflict; and market signals in the form of food price dynamics.