Artificial intelligence reveals past climate extremes by reconstructing historical records

Area of Science:

• Climate Science
• Artificial Intelligence
• Data Science

Background:

• Understanding climate extremes and risks requires historical context, but current observational datasets have spatial gaps and inaccuracies.
• Traditional statistical methods struggle with spatial extrapolation, especially for data before the mid-20th century.
• Existing reanalysis datasets often lack coverage for the extensive historical period needed for comprehensive analysis.

Purpose of the Study:

• To reconstruct historical observations of European climate extremes (warm and cold days/nights).
• To leverage artificial intelligence and Earth system model data for improved climate data.
• To provide a comprehensive dataset for better climate risk characterization and policy development.

Main Methods:

• Utilized artificial intelligence, specifically transfer learning, to reconstruct climate extreme observations.
• Employed data from the Coupled Model Intercomparison Project Phase 6 (CMIP6).
• Developed a novel method that surpasses conventional statistical techniques and diffusion models.

Main Results:

• Successfully reconstructed European climate extremes (warm/cold days and nights) from 1901-2018.
• The AI-based reconstruction method demonstrated superior performance compared to traditional techniques and diffusion models.
• Revealed spatial trends and filled data gaps across an extensive historical time span.

Conclusions:

• The developed AI method provides a valuable tool for reconstructing historical climate extremes.
• The resulting dataset enhances the characterization of climate extremes and associated risks.
• Improved data facilitates better climate risk management and informs policy decisions.