How do recent extremes compare to past events and are there regional long term trends? Thus, data analysts of DKRZ have developed a method that can effectively reconstruct incomplete observation data on climate extremes by using methods of artificial intelligence (AI). The results of the study were published in the internationally renowned journal “Nature Communications” at the end of October 2024.
Data gaps because of traditional statistical methods
The analysis of past climate extremes is complicated by the fact that existing datasets of observed extremes generally exhibit spatial gaps and inaccuracies due to inadequate spatial extrapolation. This problem arises from traditional statistical methods used to account for the lack of measurements, particularly prevalent before the mid-20th century.
The study demonstrates how AI can effectively reconstruct sparse observational data of European climate extremes (warm and cold days and nights) and reveal spatial trends across the time span from 1901 to 2018 that is not covered by most reanalysis datasets. The analysis shows that the AI method surpasses established statistical methods such as Kriging. The reconstruction is based on transfer learning with Earth System Model data e.g. large data amounts from the Coupled Model Intercomparison Project CMIP6. The computations used the GPU part of DKRZ’s HPC system “Levante”.
The AI reconstructed dataset reveals quantitative evidence for hot and cold extremes in the early 20th century and sheds a new light on the evolution of these extremes. The dataset is provided to the climate community for a better characterization of climate extremes and to improve risk management and policy development.
Contact for scientific information:
Ètienne Plesiat, Data analyst at DKRZ: plesiat@dkrz.de
Original publication:
Étienne Plésiat, Robert J. H. Dunn, Markus G. Donat & Christopher Kadow: Artificial intelligence reveals past climate extremes by reconstructing historical records, Nature Communications volume 15, Article number: 9191 (2024), https://doi.org/10.1038/s41467-024-53464-2
