AI-Based Assessment of Food Security and Climate Change: Multi-Scenario Analysis for Africa

Abstract

Food security is tightly linked to climate change, with conflicts, pandemics, and extreme events compounding risk. Sub-Saharan Africa is especially affected by this situation. This study develops an AI-based framework to quantify how climate variability and market dynamics shape food security in Africa and to translate these relations into forward-looking indicators. A compile multi-source panel have for African countries that integrates climate variables, undernourishment, production, yield, and price data. After cleaning, normalization, and context-aware imputation, we construct a Food Security Vulnerability Index (FSVI) that synthesizes availability, access, and stability through principal components and an autoencoder. Panel models approach combines machine learning algorithms, clustering, scenario experiments, and attribute imputation to capture structural drivers and short-term shocks. The findings indicate that warming exerts the strongest negative pressure on maize yields, while wheat is more sensitive to rainfall anomalies. Moderate CO2 fertilization cannot offset heat stress in some regions. High import dependence on wheat and rice increases access risks during price increases.Vulnerability clusters concentrate in parts of the Sahel, the Horn of Africa, and Southern Africa, where climate exposure coincides with limited input use, storage constraints, and inland transport frictions. Projections indicate that yield growth will lag behind demand in many subregions, unless targeted interventions are implemented. Validation demonstrates strong agreement with observations (Pearson r ≈ 0.81) and meaningful concordance with outcome-based indices, supporting the early warning utility. In scenario analyses, yield changes for maize/teff clusters range from -1.5% to -5.0% under Dry/Hot conditions and +1.5% to +5.0% across six crops under Wet/Warm conditions, with an FSVI increase of roughly +0.3 to +0.6 in Dry/Hot regimes. Overall, the framework provides a practical and interpretable tool for prioritizing interventions and tracking vulnerability as data ecosystems evolve.