EU regions face 7.6 percent agriculture growth hit from heat
A new ECB working paper estimates that extreme heat waves and droughts could reduce annual agricultural growth by up to 7.6 percentage points in EU regions. The study uses machine learning models to predict economic impacts across 1,117 regions from 2002 to 2022.
Machine learning maps climate risks
The study develops climate-augmented machine learning (ML) models, specifically Random Forest and XGBoost, to predict real growth in per capita value added.
These models combine economic indicators with high-frequency climate data, including heat waves and various drought types.
The ML approach significantly improves predictive accuracy, particularly in climate-sensitive sectors like agriculture, where heat wave and drought indicators provide meaningful predictive power.
Simulations of extreme combined heat and drought scenarios suggest that agricultural annual growth could fall by 1.9 to 7.6 percentage points in most regions.
This highlights the ML models' ability to better reflect complex climate–economic interactions compared to traditional linear models, supporting their use for early warning and policy planning.
Uneven impacts across sectors and regions
While agriculture faces substantial losses, the industrial sector experiences smaller reductions, averaging -0.75 percentage points, and manufacturing remains broadly stable, with an average impact of -0.11 percentage points.
Impacts are more pronounced in eastern Europe and the Baltic states for both agriculture and industry.
These heterogeneous effects likely stem from differences in exposure, adaptive capacity, irrigation infrastructure, and the prevalence of indoor production environments.
The greater vulnerability of agriculture is consistent with limited adaptive capacity and rain-fed systems, whereas manufacturing's resilience may be due to indoor production buffering against extreme weather.
The findings underscore the importance of integrating climate information into short-term economic monitoring tools for targeted adaptation strategies.
Beyond linear models, a clearer picture
This study provides a crucial methodological leap by demonstrating the power of machine learning to capture complex, non-linear climate-economic interactions.
While the findings confirm agriculture's high vulnerability, the nuanced regional and sectoral insights offer a more granular basis for targeted policy interventions.
The paper underscores the urgent need for robust, data-driven tools to navigate Europe's increasing climate risks.