Testing NeuralGCM's capability to simulate future heatwaves based on the 2021 Pacific Northwest heatwave event

TL;DR

This paper evaluates NeuralGCM, an AI-based climate model, for simulating extreme heatwaves like the 2021 Pacific Northwest event, comparing its performance to traditional physics-based models and analyzing its strengths and limitations.

Contribution

It demonstrates NeuralGCM's ability to accurately simulate specific heatwave events and generate stable future projections, highlighting its potential and current limitations.

Findings

NeuralGCM accurately replicates the 2021 heatwave.

It produces stable mid-century climate projections.

It underestimates warming amplitude due to missing land feedbacks.

Abstract

AI-based weather and climate models are emerging as accurate and computationally efficient tools. Beyond weather forecasting, they also show promise to accelerate storyline analyses. We evaluate NeuralGCM's ability to simulate an extreme heatwave against the Energy Exascale Earth System Model (E3SM), a physics-based climate model. NeuralGCM accurately replicates the targeted event, and generates stable and realistic mid-century projections. However, due to the absence of land feedbacks, NeuralGCM underestimates the projected warming amplitude compared to physics-based model references.