Food Insecurity Projections for Anticipatory Action: Comparative Spatiotemporal Analysis of FEWS NET and the IPC in Somalia

Abstract. Anticipatory Action (AA) has become a central pillar of contemporary humanitarian action, aiming to reduce the impacts of climate-related hazards by enabling preventive interventions before crises fully materialize. Food insecurity projections play a critical role in AA, particularly in the context of droughts. Two internationally operating systems, FEWS NET and the Integrated Food Security Phase Classification (IPC), provide widely used assessments and projections of food insecurity. Practitioners are often required to choose between these frameworks when designing AA trigger systems, a challenge that was recently amplified by USAID funding disruptions and the temporary outage of FEWS NET, highlighting the vulnerability of critical humanitarian data streams and the need for informed substitution. This study presents a systematic spatiotemporal comparison and projection skill assessment of FEWS NET and IPC food insecurity data for Somalia between 2017 and 2025. Using harmonized geospatial time series, we evaluate differences in historical assessments and projections, the frequency of projected crisis conditions, and projection accuracy relative to subsequent current assessments. Results indicate strong agreement between the two systems in their assessments of current conditions, but notable divergence in projections: FEWS NET consistently projects slightly higher levels of food insecurity and substantially more frequent Emergency and Famine conditions than the IPC. Projection accuracy is moderate for both systems and characterized by a predominance of positive bias, particularly for higher food insecurity classes. This reflects a precautionary forecasting approach that is appropriate in humanitarian contexts but must be explicitly considered when defining trigger sensitivity and resource allocation in anticipatory action. Overall, the findings demonstrate that FEWS NET and IPC projections are not directly interchangeable despite their shared classification scale, underscoring the need for dataset-specific trigger calibration.

 Reproducibility review available at: https://doi.org/10.53962/pdw8-n5v0