Bayesian Spatio-temporal Additive Modeling of Severe Food Insecurity Dynamics Across Africa

Abstract Spatio-temporal analysis is a powerful tool for exploring geo-referenced data containing space and time information. The models are often visualized through maps to represent the spatial dependence and temporal correlation over time. Therefore, this study aims to investigate the country-level determinants and spatio-temporal dynamics of severe food insecurity across 52 African countries over the period 2015–2021. The study employed Bayesian spatio-temporal additive models, including the classical parametric trend model, spatiotemporal ANOVA model, dynamic nonparametric trend model, and space-time interaction nonparametric trend model. The estimations were carried out using R-INLA. Among the fitted models, the Bayesian spatio-temporal additive model with a Type I interaction demonstrated the best overall fit for the dataset. The findings show evidence that severe food insecurity was significantly spatially dependent (τ_θ^2 = 2705.77) and temporally correlated (τ_α^2 = 10.75) across the continent. The spatio-temporal interaction term (τ_δ^2= 29,438.77) also exhibits high precision, suggesting that the interaction between space and time contributes relatively little additional variability as compared to spatial and temporal components. Model-based estimates were mapped to examine the continent's geographic disparities and temporal variability. The temporal analysis at the continental scale showed a significant and sustained upward trend in severe food insecurity over the study period, with most countries experiencing rising rates. The spatial analysis also revealed that the rate of vulnerabilities varied by geographic location, with countries such as the Democratic Republic of Congo, Central African Republic, South Sudan, Kenya, Ethiopia, Libya, Algeria, Nigeria, Niger, Mali, Burkina Faso, Angola, and Zimbabwe consistently and persistently experiencing a high rate of severe food insecurity throughout much of the study periods. Furthermore, the study identified that malaria incidence, climate change, livestock production and investment inflow had statistically significant linear fixed effects on the severe food insecurity rate. In contrast, cereal import dependence, Greenhouse Gas (GHG) emissions, dietary energy supply, dietary protein supply, gross domestic product (GDP), unemployment, inflation, and caloric loss exhibited statistically significant intricate, dynamic and spatially varying nonlinear influences on the severe food insecurity. Our findings underscore the need for multi-sectoral, adaptive policies integrating health, agriculture, climate, and economic planning. Governments should prioritize malaria prevention, climate adaptation, livestock development, investment promotion and macroeconomic stability while tailoring responses to country-specific contexts. Keywords: Spatio-temporal Additive Models, Spatial Effects, Temporal Effects, Space-time Interaction, INLA, Severe Food Insecurity, Africa