Journal Article

DiffESM: Conditional Emulation of Temperature and Precipitation in Earth System Models With 3D Diffusion Models

Bassetti, Seth, Tebaldi, Claudia, Hutchinson, Brian, Kravitz, Ben

2024

Overview

Earth system models (ESMs) are essential for understanding the interaction between human activities and the Earth's climate. However, the computational demands of ESMs often limit the number of simulations that can be run, hindering the robust analysis of risks associated with extreme weather events. While low‐cost climate emulators have emerged as an alternative to emulate ESMs and enable rapid analysis of future climate, many of these emulators only provide output on at most a monthly frequency. This temporal resolution is insufficient for analyzing events that require daily characterization, such as heat waves or heavy precipitation. We propose using diffusion models, a class of generative deep learning models, to effectively downscale ESM output from a monthly to a daily frequency. Trained on a handful of ESM realizations, reflecting a wide range of radiative forcings, our DiffESM model takes monthly mean precipitation or temperature as input, and is capable of producing daily values with statistical characteristics close to ESM output. Combined with a low‐cost emulator providing monthly means, this approach requires only a small fraction of the computational resources needed to run a large ensemble. We evaluate model behavior using a number of extreme metrics, showing that DiffESM closely matches the spatio‐temporal behavior of the ESM output it emulates in terms of the frequency and spatial characteristics of phenomena such as heat waves, dry spells, or rainfall intensity. Plain Language Summary Ideally, to study how damaging phenomena like heatwaves, droughts and downpours will change in the future under global warming, we would want a large number of climate model runs producing many realizations of climate futures that we can analyze and from which the new characteristics of climate extremes can be quantified. Currently, emulators can rapidly generate simulations of future climate, but often to relatively low frequencies, as decadal, annual or monthly output at best in most cases, which is insufficient for studying extreme events that occur on a daily timescale. We show how it is possible to train a machine learning model to produce daily series of temperature or precipitation from monthly averages, thus facilitating a more robust investigation into how extreme events will change in the future. Key Points Earth system models (ESMs) are key devices for understanding how human actions will affect the future global climate Computational demands prevent us from running them for more than a handful of scenarios. ESM emulators are most commonly limited to the monthly frequency We present DiffESM as a data‐driven emulator of ESMs that closely matches the spatiotemporal distributions of ESMs at daily frequency

• Publisher: John Wiley & Sons, Inc,American Geophysical Union (AGU)
• Subject: Algorithms / Climate / Climate and human activity / Climate change / Climatic analysis / Deep learning / Diffusion / diffusion model / Diffusion models / Dry spells / Earth / Earth system model emulation / Emissions / Extreme weather / extremes / Future climates / generative model / Heat waves / Heavy precipitation / Machine learning / Mean precipitation / Neural networks / Precipitation / Radiative forcing / Rainfall intensity / temperature / Weather forecasting