Nonlinear economic damages from compound heat and drought events

Abstract

Europe is increasingly exposed to heatwaves and droughts, yet their short-term, sector-specific economic impacts remain difficult to quantify in predictive frameworks. We develop climate-augmented models for regional real growth in per capita gross value added across 1117 EU regions over 2002–2022, combining standard economic indicators with high-frequency climate variables capturing heatwaves and multiple drought types across temporal horizons. Using mixed-frequency designs and machine learning (ML) approaches, Random Forest and XGBoost, we show that climate predictors improve predictive accuracy relative to linear benchmarks in the climate-sensitive agricultural sector. In contrast, other sectoral aggregations exhibit more limited gains, and climate-augmented specifications do not consistently outperform their economic-only counterparts. Heatwave indicators contribute robustly to predictive performance, while the importance of drought varies by sector. Simulations of an extreme compound heat-drought scenario using XGBoost suggest that agricultural growth (NACE A) declines by 1.93–7.36 percentage points in 99% of regions, whereas industry (NACE B–E) experiences smaller losses, with manufacturing (NACE C) emerging as relatively resilient. Our exercise indicates that ML models better capture nonlinear, seasonal, and spatial climate-economic interactions, highlighting the value of climate-augmented predictive modelling for early warning, regional fiscal planning, and targeted adaptation.