PseudospectralNet: Toward Hybrid Atmospheric Models for Climate Simulations

Maximilian Gelbrecht, Milan Klöwer, Niklas Boers

Journal of Advances in Modeling Earth Systems · 01 October 2025 · Massachusetts Institute of Technology (MIT), Potsdam Institute for Climate Impact Research (PIK), Technical University of Munich (TUM), University of Oxford

Paper PDF

In one sentence. PseudospectralNet couples a quasi-geostrophic dynamical core with a UNet in spectral and grid space, improving short-term skill and long-term stability over pure ML models.

Abstract

Abstract Recent machine learning (ML) models have shown great success for weather prediction tasks, suggesting that atmospheric dynamics can in principle be learned from data. However, such approaches suffer from instability or drift in long integrations and are hence usually not suited for climate simulations. Incorporating physical knowledge promises to alleviate such shortcomings. Here, we present PseudospectralNet (PSN), an architecture for a hybrid atmospheric model that combines a quasi‐geostrophic physics‐based dynamical core with a data‐driven core based on an UNet. Our architecture transforms between grid and spectral space at every time step and therefore mimics the pseudospectral solution approach many intermediate‐complexity atmospheric models follow. Neural networks are separately defined in the spectral and grid space and are combined with physics‐based dynamical cores in each of these spaces. We train PSN separately on data from quasi‐geostrophic models, primitive equation models, and reanalysis data. We use this model to study the effect of adding physics‐based cores to ML models on forecast error and numerical stability. We find in both regards that adding the physics‐based dynamical core to our model helps short‐term predictability and long‐term numerical stability of the hybrid model.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.