Journal Article

Old Dog, New Trick: Reservoir Computing Advances Machine Learning for Climate Modeling

Bretherton, Christopher S.

2023

Overview

Physics‐informed machine learning (ML) applied to geophysical simulation is developing explosively. Recently, graph neural net and vision transformer architectures have shown 1–7 days global weather forecast skill superior to any conventional model with integration times over 1,000 times faster, but longer simulations rapidly degrade. ML that achieves high skill in both weather and climate applications is a tougher goal. This Commentary was inspired by Arcomano et al. (2023, https://doi.org/10.1029/2022GL102649), who show impressive progress toward that goal using hybrid ML, combining reservoir computing (RC) to a coarse‐grid climate model and coupling to a separate data‐driven RC model that interactively predicts sea‐surface temperature. This opens new horizons; where will the next ML breakthrough come from, and is conventional climate modeling about to be disrupted? Plain Language Summary Many new research groups are making rapid progress in applying diverse machine learning (ML) methodologies to weather forecasting and climate modeling. These new approaches could make simulations that are 1,000x faster than conventional approaches for the same fidelity. One successful approach for weather forecasting has been replacing an entire conventional global atmospheric model with a ML emulator, but so far the spatial patterns of long‐term average temperature and precipitation simulated by this approach develop large biases. An alternate new approach, hybrid reservoir computing, combines the conventional model with a form of ML that remembers the recent atmospheric evolution. It produces better climate simulations, including realistic El‐Nino/La Nina variability, but with much less spatial detail. This opens new horizons; where will the next ML breakthrough come from, and is conventional climate modeling about to be disrupted? Key Points Arcomano et al. (2023, https://doi.org/10.1029/2022GL102649) combined reservoir computing (RC) with a coarse‐grid climate model for data‐driven ocean‐coupled simulations By building long‐term memory into predictions, RC nearly removes climate bias Challenges remain with interpretability and scalability to fine‐scale prediction that new machine learning approaches may soon surmount

• Publisher: John Wiley & Sons, Inc,Wiley
• Subject: Atmosphere evolution / Atmospheric evolution / Atmospheric models / Climate / Climate and weather / climate modeling / Climate models / Computation / El Nino phenomena / Global weather / La Nina / Learning algorithms / Machine learning / Modelling / Neural networks / Physics / reservoir computing / Simulation / Sulfuric acid / Surface temperature / Weather / Weather forecasting