ThinkingEarth: Causal AI for Food Security

Taken from Jordi Cerdà-Bautista recent presentation on Use Case 3 at the CENTAUR workshop: 

Causality for Food Security Under Climate Change

When resources are limited, policymakers must know what causes food insecurity—not just what correlates with it. ThinkingEarth applies causal inference methods to estimate the impact of key drivers while accounting for shared confounders such as climate variability, conflict, market dynamics, displacement, and agricultural production.

Credit Access and Acute Food Insecurity in Somalia

We examine whether higher access to credit reduces the risk of acute food insecurity in Somalia. The outcome of interest is the share of the population in IPC Phase 3 or above (crisis level or worse), and the treatment is the share of the population with access to credit. To isolate the causal effect, we control for ENSO variability, rainfall anomalies, food prices, conflict events, displacement, and agricultural output.

Estimating the effect across multiple temporal aggregations—annual, seasonal, IPC-cycle, and monthly—we consistently find that districts with greater access to credit show 2–4% lower levels of acute food insecurity. This suggests that financial access strengthens resilience by enabling households to smooth consumption during shocks.

Climate Sensitivity and Food Price Spikes

We also investigate whether the impact of food price spikes depends on climate sensitivity. Regions are classified according to how vegetation responds to ENSO variability: in some areas ENSO reduces vegetation health (Negative regime), in others the link is weak, and in some it improves vegetation outcomes (Positive regime).

The results show that price spikes have a significantly stronger impact in drought-prone areas. In the Negative regime, price shocks are associated with a 6% increase in crisis-level food insecurity, compared to about 2% in the Weak and Positive regimes. This demonstrates that market shocks are more damaging where climate vulnerability is already high.

Food Groups and Micronutrient Adequacy

Finally, we will extend the framework to dietary quality in Nigeria, Ethiopia, and Sri Lanka. Building on predictive work from the World Food Programme, we move from correlation to causal estimation to determine which food groups provide the strongest protective effect against micronutrient inadequacy. By controlling for income, seasonality, and market access, we compare the relative impact of staples, fruits, vegetables, and animal-source foods on nutritional outcomes.

Core Message

ThinkingEarth shifts the focus from association to causation. By modelling Earth as an interconnected system and applying advanced causal AI, we identify which drivers truly influence food insecurity and where interventions can have the greatest impact.