AI-Based Anomaly Detection for Extreme Event Attribution: An Analysis of European Heatwaves

Abstract

Formal extreme event attribution traditionally relies on large, computationally intensive climate-model ensembles,which often hampers the provision of information in real-time attribution. To overcome this, we propose a lightweight, inter-pretable framework that couples unsupervised anomaly detection with Bayesian deep learning enabling near-real-time attri-bution without statistical assumptions or costly climate model simulations. A convolutional variational autoencoder (VAE) istrained on daily 2 m temperature fields from the CMIP6 HadGEM3-GC31-LL Hist-NAT experiment (1850–2020) and earlyERA5 reanalysis (1940–1980). The VAE achieves a spatial-mean correlation of ∼ 0.9 over land and exhibits robust skill indiscriminating warm days from non-extreme conditions across the full domain. Spatial gradients of the VAE’s reconstructionmean-squared error and Kullback–Leibler divergence yield three threshold-free metrics quantifying local departures frompre-industrial conditions. We examine their relationship with more statistical metrics like the counterfactual exceedanceprobability p ex , which measures the chance that an event drawn from the factual climate exceeds any event expected underthe counterfactual scenario. Across the training period, these metrics exhibit a strong monotonic anti-correlation with p ex ,demonstrating that VAE loss gradients encode meaningful dynamical–thermodynamic information. A Bayesian multi-layerperceptron calibrated solely on these metrics reproduces grid-point p ex with a mean spatial correlation of 0.92, deliveringexceedance probabilities alongside interquartile ranges. Case studies of four exceptional European heatwave summers (2010,2018, 2019, and 2022) show that our framework accurately captures spatio-temporal patterns, assigns near-zero probabilities( p̂ ex < 0.05 ) to record-breaking temperatures, and yields return-period estimates consistent with large-ensemble attributionstudies–all without assuming a parametric distribution or using global-mean temperature predictors. This novel approachprovides an efficient tool for operational attribution of extreme events under a changing climate by quantifying changes inthe probability of occurrence on daily maps, which would offer valuable insights for the development of proactive regionalplanning and effective adaptive strategies. Motivated by the growing demand for actionable climate-risk information, theproposed protocol-ready framework supports decision-making by delivering quantitative, uncertainty-aware estimates ofchanging current risk levels.