Climate Change and Regional Economic Growth in the EU

 This study, titled "Beat the heat, the role of heat waves and droughts in regional EU economies," provides a comprehensive analysis of how extreme climate events influence regional economic output across the European Union. Developed as part of the European Central Bank's Working Paper Series, the research aims to bridge the gap in quantifying short-term economic effects of heat waves and droughts, which are increasingly frequent and widespread in Europe.

Study Overview and Scope

The study develops climate-augmented predictive models to forecast real growth in per capita Gross Value Added (GVA) across 1,117 EU regions (at the NUTS-3 level) from 2002 to 2022. It focuses on three key sectors:

• Agriculture (Sector A).
• Industrial Aggregate (Sectors B-E, including mining, manufacturing, energy, and water supply).
• Manufacturing as a standalone sector (Sector C).

The methodology integrates traditional economic indicators with high-frequency climate data from the Copernicus European Drought Observatory, capturing dimensions such as heat waves, meteorological droughts (precipitation deficits), hydrological droughts (low river flows), and agricultural droughts (soil moisture and vegetation stress).

Context: Heat Waves and Droughts in EU Regional Economies

The sources emphasize that Europe is entering a period of compounding climate risks where extreme events like the record-breaking 2022 drought illustrate how hot and dry conditions can cascade across sectors.

• Broader Macroeconomic Impact: Evidence suggests major droughts can depress regional output by up to 3 percentage points for as long as four years after the event. Furthermore, harvest failures have significant inflationary effects, explaining roughly 30% of medium-term euro area inflation volatility.
• Sectoral Disruptions: Beyond agriculture, droughts disrupt river transport (e.g., the Rhine), curtail hydroelectric and thermal power production, and increase costs for water-dependent manufacturing.

Key Findings on Sectoral Vulnerability

The study reveals that the economic impact of these climate stressors is highly sector-dependent and operates through nonlinear channels.

• Agriculture: This is the most sensitive sector. Simulations of extreme compound events (heat and drought) suggest agricultural annual growth could fall by an average of 4.54 percentage points across most regions.
• Industry and Manufacturing: These sectors are less affected overall. The industrial aggregate experiences moderate simulated reductions (average -0.75 pp), while manufacturing remains largely stable (average -0.11 pp). This resilience is attributed to the "buffering role" of indoor production environments, which protect manufacturing from direct exposure to extreme heat and drought.

Regional Heterogeneity and Adaptive Capacity

The sources highlight a stark divide in how different EU regions respond to these climate shocks:

• Eastern Europe and the Baltic States: These regions face the most severe simulated losses in both agriculture and industry. This vulnerability is linked to a reliance on rain-fed agriculture, less developed irrigation infrastructure, and higher energy-intensity in production.
• Southern Europe: Despite being prone to chronic heat and water scarcity, this region often exhibits more muted short-term impacts in agriculture. This is attributed to higher structural adaptation, such as the widespread use of irrigation systems that can buffer initial losses.

Methodological Significance

A core conclusion of the study is that Machine Learning (ML) models (specifically Random Forest and XGBoost) significantly outperform linear models in predicting outcomes for climate-sensitive sectors. This is because ML can better capture the complex nonlinear interactions and seasonal nuances of climate shocks—such as the spatial propagation of a heat wave from a neighboring region—that traditional linear frameworks often miss.

The methodological approach described in the sources represents a shift from traditional causal inference toward predictive modeling, specifically designed to capture the localized and complex nature of climate shocks across 1,117 EU regions. By integrating high-frequency climate data with annual economic indicators, the study seeks to quantify short-term impacts that are often obscured in country-level analyses.

The core components of this approach include:

1. Spatial and Sectoral Granularity

The study operates at the NUTS-3 level, which allows researchers to account for the highly localized nature of heat waves and droughts. This granularity is essential because economic impacts vary significantly depending on a region's exposure, sectoral composition, and adaptive capacity. The analysis focuses on three specific NACE sectors:

• Agriculture (Sector A).
• Industrial Aggregate (Sectors B-E, including energy and water supply).
• Manufacturing (Sector C).

2. Machine Learning vs. Linear Benchmarks

A central finding of the methodological exercise is that nonlinear Machine Learning (ML) models—specifically Random Forest and XGBoost—consistently outperform traditional linear regression in climate-sensitive sectors.

• Nonlinear Dynamics: The sources argue that the economic effects of climate extremes operate through nonlinear channels that linear models cannot represent.
• Regularization: ML models are better equipped to handle multicollinearity, which is a frequent challenge when climate indicators (like heat and drought) are highly correlated.

3. Reconciling Data Frequencies

To bridge the gap between daily climate observations and annual economic growth data, the study utilizes three distinct feature-engineering pipelines:

• Yearly Aggregation: Summarizing daily data into annual medians and standard deviations to capture both central tendencies and volatility.
• Principal Component Analysis (PCA): Extracting singular features from climate indicators to mitigate multicollinearity.
• Mixed Data Sampling (MIDAS): Applying lag weights to monthly observations to capture the gradual onset or dissipation of extreme events. Interestingly, the study found that simple annual aggregation often performed as well as or better than more complex extraction techniques, particularly in short panels where complex models risk over-fitting.

4. Accounting for Compounding and Spillovers

The methodology specifically addresses the compounding nature of climate risks across three dimensions:

• Temporal Compounding: Incorporating event duration into severity indices.
• Geographic Compounding: Using spatial averages of climate features from neighboring regions to capture spillover effects and shared climate systems.
• Event Compounding: Creating interaction terms (specifically for heat waves and droughts) to model the unique economic impact of simultaneous stressors.

5. Model Interpretability and Simulation

To make these "black box" ML models transparent, the researchers use SHAP (SHapley Additive exPlanations) values. This technique identifies which predictors—economic or climatic—are driving the model's output without assuming a specific functional form.

Finally, the study applies its best-performing model (XGBoost with yearly aggregation) to simulate extreme scenarios, such as a repeat of the 2022 drought. These simulations provide a "stress test" for regional economies, revealing how a uniform climate shock would result in heterogeneous economic outcomes based on regional resilience and infrastructure.

The study utilizes a multifaceted set of climate indicators sourced from the Copernicus European Drought Observatory to capture the diverse dimensions of climate stress affecting EU regional economies. These indicators are essential for characterizing the intensity, duration, and spatial extent of extreme events that disrupt economic output.

Core Climate Indicators

The researchers categorized these indicators into four primary groups to reflect different environmental and economic impacts:

• Heat Waves: Measured by the Heat and Cold Waves Index (HCWI), which monitors daily minimum and maximum temperatures against 30-year climatological thresholds to identify events lasting at least three consecutive days.
• Meteorological Droughts: Captured by the Standardised Precipitation Index (SPI) across three horizons:
  • SPI01 (1-month): Indicates immediate effects such as soil moisture deficits.
  • SPI06 (6-month): Reflects deficits in streamflow and reservoir storage.
  • SPI12 (12-month): Highlights long-term reductions in groundwater and reservoir recharge.
• Hydrological Droughts: Monitored via the Low-Flow Index (LFI), which detects periods of unusually low river discharge compared to long-term thresholds.
• Agricultural Droughts: Assessed using the Soil Moisture Anomaly (SMA) and the Fraction of Absorbed Photosynthetically Active Radiation (fAPAR). SMA measures deviations in soil moisture, while fAPAR tracks anomalies in vegetation photosynthetic activity, which are critical for predicting crop stress.

Compounding and Spatial Indicators

Recognizing that climate risks rarely occur in isolation, the study developed indicators to capture compounding effects and spatial spillovers:

• Compound Events: Researchers created a specific feature to account for the simultaneous occurrence of heat waves and droughts by interacting the scaled HCWI and SPI indicators.
• Geographic Compounding: To account for the spatial propagation of climate shocks, the study incorporates regional averages of climate features from neighboring areas. This allows the models to capture how a drought or heat wave in one region might influence economic activity in another through shared ecosystems or trade routes.

Observed Trends in EU Regions

Data from these indicators reveal significant shifts in the European climate landscape over the study period (2002–2022):

• Increasing Heat Stress: Most European regions have seen a rise in the average number of annual heat wave days, with the most pronounced increases in the Spanish Mediterranean, southern Italy, and parts of northern and eastern Europe.
• Intensifying Dryness: Indicators like SPI01 and LFI suggest a trend toward increased dryness, particularly in central Europe, while agricultural drought (fAPAR) has shown significant increases in western and northern Europe over the last decade.

Sector-Specific Sensitivity

The predictive power of these indicators varies by economic sector. Agriculture is most sensitive to HCWI and short-term meteorological drought (SPI01, SPI06), which directly impact crop yields and vegetation health. In contrast, the Industrial Aggregate and Manufacturing are more influenced by hydrological and long-term drought (LFI, SPI12), which affect water-dependent processes like cooling for power plants and inland shipping.

The sources provide a detailed sectoral impact analysis of how heat waves and droughts influence 1,117 EU regional economies, focusing on three specific classifications: Agriculture (Sector A), the Industrial Aggregate (Sectors B-E, which includes mining, manufacturing, energy, and water supply), and Manufacturing (Sector C) as a standalone subset.

The analysis reveals that these extreme climate events do not affect sectors uniformly, with impacts ranging from severe disruptions to notable resilience.

1. Agriculture (Sector A): The Most Sensitive Sector

Agriculture is identified as the most vulnerable sector to climate extremes.

• Magnitude of Impact: Simulations of extreme compound heat and drought events, similar to conditions in 2022, suggest agricultural annual growth could fall by an average of 4.54 percentage points (pp) relative to a no-anomaly benchmark.
• Regional Variation: Losses are most severe in Eastern Europe (averaging -5.36 pp), particularly in regions like Romania, Bulgaria, and the Czech Republic. In contrast, Southern Europe exhibits more muted short-term impacts.
• Drivers of Vulnerability: The high sensitivity in Northern and Eastern Europe is linked to a heavy reliance on rain-fed systems and limited irrigation infrastructure. Conversely, Southern Europe’s resilience is attributed to structural adaptation, such as extensive irrigation systems that buffer initial output losses.

2. Industrial Aggregate (Sectors B-E): Moderate and Heterogeneous Effects

The broader industrial sector experience smaller but still meaningful reductions in economic output.

• Magnitude of Impact: The simulated compound event results in an average reduction of 0.75 pp in real GVA per capita growth.
• Exposure Channels: This aggregate is more affected than manufacturing alone because it includes activities with high outdoor or infrastructure-related exposure, such as extraction, utilities, and water infrastructure.
• Regional Vulnerability: Again, Eastern Europe and the Baltic states face more visible negative effects, which the sources suggest may reflect a combination of higher energy intensity, aging infrastructure, and less developed adaptive capacity.

3. Manufacturing (Sector C): A "Buffer" of Resilience

Manufacturing shows significant resilience compared to other climate-exposed sectors.

• Magnitude of Impact: Simulated losses remain broadly stable, averaging only -0.11 pp across the EU.
• Indoor Production Protection: The sources attribute this resilience to the predominance of indoor production environments, which protect labor productivity and industrial processes from the direct physical stress of extreme heat and drought.
• Economic Significance: Despite these small percentage declines, the sources note that manufacturing accounts for 16.9% of total EU GVA (compared to 1.89% for agriculture). Therefore, even minor percentage drops can translate into significant absolute economic losses.

4. Broader Context: Compounding Factors and Inflation

Beyond direct production losses, the sectoral analysis highlights several cascading economic effects:

• Energy Intensity: While manufacturing is energy-intensive, the study suggests that direct climatic exposure and sectoral sensitivity (like being outdoors) are more significant drivers of short-term GVA loss than energy-product use alone.
• Inflationary Pressures: Agricultural shocks are a major source of volatility, with harvest failures explaining roughly 30% of medium-term euro area inflation volatility.
• Supply Chain Disruptions: Droughts propagate through production networks via disruptions to river transport (e.g., the Rhine), water-dependent manufacturing, and electricity generation, which can increase costs and constrain output across multiple sectors.

The 2022 Extreme Scenario Simulation serves as a "stress test" within this study, designed to quantify how a uniform climate shock—modeled after the record-breaking heat waves and droughts of 2022—would impact economic output across diverse EU regions. Using a high-performing XGBoost model, the researchers compared a "no-anomaly" benchmark against a scenario where every region was subjected to extreme tail percentiles of climate stress (e.g., the 99th percentile for heat waves and the 1st percentile for soil moisture).

The simulation highlights a stark divide in sectoral vulnerability and regional resilience across the European economy:

Sectoral Losses: Agriculture vs. Industry

• Agriculture (Sector A): This sector faces the most devastating simulated impacts, with an average reduction in real GVA per capita growth of 4.54 percentage points (pp). In some regions, losses reached as high as 7.36 pp, illustrating the extreme sensitivity of food production to compounding heat and moisture stress.
• Industrial Aggregate (Sectors B-E): The impact on the broader industrial sector is more moderate, averaging a 0.75 pp reduction. These losses are driven by climate-sensitive infrastructure, such as water-dependent power plant cooling and extraction activities.
• Manufacturing (Sector C): This sub-sector remains remarkably stable, with an average loss of only 0.11 pp. The sources attribute this resilience to the buffering role of indoor production environments, which shield labor and processes from direct heat exposure.

Regional Heterogeneity and Adaptation

A critical finding of the 2022 simulation is that the same climatic shock produces vastly different economic outcomes based on a region's adaptive capacity:

• The Eastern European Vulnerability: Regions in Romania, Bulgaria, and the Czech Republic face the most severe losses (averaging -5.36 pp in agriculture). This is linked to a heavy reliance on rain-fed agricultural systems and less developed irrigation and water infrastructure.
• The Southern European Buffer: Despite being prone to chronic heat, regions in Italy and Greece show more muted short-term agricultural impacts (averaging -3.9 pp). This resilience is credited to structural adaptation, specifically the widespread use of irrigation systems that can temporarily offset drought conditions.

Broader Economic Context

The sources emphasize that even the "modest" percentage declines in manufacturing must be viewed through the lens of economic scale. Since manufacturing accounts for 16.9% of total EU GVA (compared to less than 2% for agriculture), even a fractional percentage drop represents economically significant absolute losses.

Ultimately, the 2022 simulation demonstrates that while the European economy has some "indoor" defenses, compounding climate risks—where heat and drought interact—can cascade through energy systems and supply chains, potentially exacerbating regional wealth disparities if adaptation infrastructure is not prioritized.

The study concludes that heat waves and droughts exert material, nonlinear, and sector-dependent influences on regional economic activity in the EU. By integrating high-frequency climate data with economic indicators, the researchers provide a framework for real-time monitoring and targeted policy interventions,.

Key Conclusions of the Study

• Predictive Value of Machine Learning (ML): A primary conclusion is that nonlinear ML models (Random Forest and XGBoost) significantly outperform traditional linear benchmarks in climate-sensitive sectors like agriculture,,. These models are better at capturing the complex, compounding nature of climate shocks—such as simultaneous heat and dryness—that linear models often miss,.
• Extreme Sectoral Sensitivity: Agriculture is the most vulnerable sector, with simulations suggesting that an extreme event like the 2022 drought can reduce annual growth in real per capita Gross Value Added (GVA) by an average of 4.54 percentage points across EU regions,.
• The "Indoor" Resilience of Manufacturing: Manufacturing shows notable resilience, with average simulated losses of only -0.11 pp,. This is attributed to the "buffering role" of indoor production environments, which shield labor and industrial processes from the direct physical stress of extreme heat,.
• Regional Wealth and Adaptation Disparities: The study finds that climate shocks exacerbate regional disparities. Eastern Europe and the Baltic states face more severe industrial and agricultural losses due to higher energy intensity and less developed irrigation,,. Conversely, Southern Europe exhibits more muted short-term agricultural impacts because of extensive structural adaptation, such as irrigation systems,.

Policy Implications and Recommendations

The sources highlight several critical avenues for policymakers to enhance regional economic resilience:

• Integration into Monitoring Tools: The findings underscore the importance of integrating high-frequency climate information into short-term economic monitoring tools used by central banks and fiscal authorities,.
• Early-Warning Systems: Improved climate-augmented modeling can underpin early-warning systems, allowing for more proactive management of upcoming climate-induced economic stress,.
• Targeted Adaptation Strategies: By identifying the specific sectors and regions most at risk, the research guides the design of targeted adaptation policies,. For instance, policymakers in Northern and Eastern Europe may need to prioritize irrigation infrastructure, while Southern Europe may need to address the "finite" nature of its current water buffering capacity.
• Regional Fiscal Planning: The models can inform regional fiscal planning, such as the allocation of contingency funds and intergovernmental transfers, to support regions hit by severe shocks.
• System-Wide Stress Testing: The sources suggest that future policy priorities should include coupling these sectoral predictions with financial-exposure data for system-wide stress testing to understand how climate shocks propagate through banking and financial systems.

Limitations and Future Directions

The researchers caution that their findings are predictive rather than causal, representing conditional associations rather than established structural parameters,. They suggest that future work should explore absolute thresholds—such as water requirements for cooling power plants—to better characterize the nonlinear impacts on manufacturing and energy-intensive industries. Additionally, the view expressed in the paper is that of the authors and does not necessarily represent the official position of the European Central Bank.