Kraulich et al. (2025) The impact of aerosol forcing on the statistical attribution of heatwaves

Identification

• Journal: Weather and Climate Extremes
• Year: 2025
• Date: 2025-09-13
• Authors: Florian Kraulich, Peter Pfleiderer, Sebastian Sippel
• DOI: 10.1016/j.wace.2025.100803

Research Groups

• Leipzig University, Institute for Meteorology, Leipzig, Germany

Short Summary

This study demonstrates that the standard statistical method for attributing heatwaves, which relies solely on global mean temperature, produces significant biases in regions with strong aerosol trends. Incorporating regional aerosol optical depth as an additional covariate in the Generalized Extreme Value distribution model substantially reduces these biases and improves heatwave return period estimates.

Objective

• To assess how missing regional aerosol concentration changes influence the results of statistical extreme event attribution studies for heatwaves.
• To evaluate whether adding aerosol optical depth (AOD) as an additional covariate to the location parameter improves the statistical Generalized Extreme Value (GEV) model and, consequently, statistical extreme event attribution results for heatwaves.

Study Configuration

• Spatial Scale: Global analysis with a focus on four selected mid-latitude regions: Iberian Peninsula (8°W to 3°W, 38°N to 43°N), Germany (8°E to 13°E, 48°N to 53°N), Midwestern United States (88°W to 83°W, 38°N to 43°N), and China (110°E to 115°E, 31°N to 36°N).
• Temporal Scale: Climate model simulations from 1850 to 2100 (historical forcing until 2014, then SSP3-7.0 scenario). Analysis period for GEV fitting is 1900–2050. Trends are analyzed for 1979–2025. Probability ratios are calculated between 1979 and 2025.

Methodology and Data

• Models used:
  • Community Earth System Model version 2 (CESM2-LE) large ensemble (100 members) with Modal Aerosol Module (MAM4).
  • CESM2 single forcing large ensemble (SFLE) (15 members) with evolving greenhouse gases only (GHG-only).
  • Generalized Extreme Value (GEV) distribution model for extreme value statistics.
  • R-package ‘extRemes’ version 2 for GEV parameter estimation using maximum likelihood estimation.
• Data sources:
  • Climate model simulations (CESM2-LE, CESM2-SFLE GHG-only).
  • Annual maximum daily average surface air temperatures (Tx1d).
  • June-August (JJA) seasonal mean aerosol optical depth at 550 nm (AOD) (December-February (DJF) in the Southern Hemisphere).
  • Global Mean Temperature (GMT) as a covariate.
  • GHG-forced response and anthropogenic aerosol (AAER) forced response from SFLE as alternative covariates.
  • Dataset links: https://doi.org/10.26024/kgmp-c556 (CESM2-LE), https://doi.org/10.26024/yw4w-1w27 (CESM2-SFLE).

Main Results

• The 'standard method' (GEV with GMT covariate only) exhibits substantial biases in heatwave return period estimates in regions and periods characterized by strong regional aerosol changes (e.g., North America, Central and Eastern Europe, East Asia).
• These biases can lead to an underestimation of the probability ratio (PR) of a 100-year event by up to a factor of 2 when comparing periods like 1979 and 2025 (e.g., Germany: PR 20.3 vs. 29.9; Midwestern US: PR 18.0 vs. 37.5).
• The observed biases in the 'standard method' are significantly weaker or absent in GHG-only simulations, indicating that regional aerosol forcings are the primary cause.
• Including Aerosol Optical Depth (AOD) as an additional covariate in the GEV model ('GMT+AOD method') significantly reduces these biases and improves the GEV fit, particularly in regions with strong aerosol trends.
• The 'GMT+AOD method' leads to more accurate estimates of return levels (e.g., reducing bias by up to 0.9 °C in the Midwestern US during 1950-2000) and probability ratios.
• Likelihood ratio tests confirm that the GEV model performance is significantly improved by adding AOD in regions such as central and eastern North America, central and eastern Europe, and East Asia.
• The Root Mean Squared Error (RMSE) between individual GEV mean estimates and the ensemble Tx1d mean decreases in these regions (e.g., Germany by 10%, Midwestern US by 37%, and China by 15%) when AOD is included.
• An alternative 'GHG+AAER method' using single forcing forced responses also shows widespread improvements, suggesting that using forced responses instead of GMT can be recommended almost everywhere when SFLE simulations are available.

Contributions

• Systematically assesses the impact of regional aerosol trends on statistical extreme event attribution of heatwaves using large ensemble climate model simulations.
• Introduces and validates an improved statistical attribution framework by incorporating regional aerosol optical depth (AOD) as an additional covariate in the Generalized Extreme Value (GEV) model.
• Quantifies the substantial biases of the 'standard method' (GMT-only covariate) in regions with strong aerosol changes and demonstrates how the proposed method significantly reduces these biases.
• Provides guidance on geographical regions where it is crucial to account for regional aerosol trends in heatwave attribution studies.
• Highlights the importance of considering regional, non-GHG forcings for accurate climate change attribution, especially for extreme events.

Funding

• climXtreme project (German Federal Ministry of Education and Research, Phase 2, project PATTETA, grant number 01LP2323C).
• Project ‘Artificial Intelligence for Enhanced Representation of Processes and Extremes in Earth System Models’ (AI4PEX; grant agreement 101137682, funded by the EU’s Horizon Europe program).

Citation

@article{Kraulich2025impact,
  author = {Kraulich, Florian and Pfleiderer, Peter and Sippel, Sebastian},
  title = {The impact of aerosol forcing on the statistical attribution of heatwaves},
  journal = {Weather and Climate Extremes},
  year = {2025},
  doi = {10.1016/j.wace.2025.100803},
  url = {https://doi.org/10.1016/j.wace.2025.100803}
}