AI-Driven Weather Forecasts to Accelerate Climate Change Attribution of Heatwaves

Abstract

Anthropogenic climate change (ACC) is driving an increase in the frequency, intensity, and duration of heatwaves (HWs), making the rapid attribution of these events essential for assessing climate-related risks. Traditional attribution methods often suffer from selection bias, high computational costs, and delayed results, limiting their utility for real-time decision-making. In this study, we introduce a novel artificial intelligence (AI)-driven attribution framework that integrates physics-based ACC estimates from global climate models with state-of-the-art AI weather prediction (AIWP) models. We apply this approach to four HWs across different climatic regions using two AIWP models (FourCastNet-v2 and Pangu-Weather) and one hybrid AI-physics model (NeuralGCM). Our results show that AIWP models accurately predict HW intensity and spatial patterns, capturing key synoptic features such as persistent high-pressure ridges. The attribution analysis reveals a robust ACC signal in all four events and a good agreement across models. Results from the hybrid model (NeuralGCM) suggest that the intensification of HWs due to ACC can largely be inferred from the atmospheric state a few days prior to the event, while sea surface temperature forcing becomes increasingly relevant at longer lead times and in specific regions. This study demonstrates that AI-based attribution enables near real-time and anticipatory assessment of HWs, offering a scalable and computationally efficient alternative to conventional methods. By providing timely and consistent attribution of extreme heat events, this approach enhances our ability to anticipate climate risks and inform adaptation strategies in a rapidly warming world.