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

TL;DR

This work tackles the high computational cost of running Earth System Models (ESMs) by introducing DiffESM, a diffusion-based emulator that downscales monthly ESM outputs to daily sequences conditioned on monthly means. Trained on a small set of realizations across forcings, DiffESM generates daily temperature and precipitation with spatiotemporal statistics closely matching the target ESM and visible extreme-event behavior. The approach is validated under unseen scenarios and on a different ESM, showing robust generalization and realistic variability without mode collapse. When paired with a low-cost monthly emulator, DiffESM enables rapid large ensembles for robust risk assessment of climate extremes, reducing computational demands while preserving critical event statistics.

Abstract

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.

Figures (9)

• Figure 1: Time series of generated, Held Out 1 and Held Out 2 sets of daily temperature (T) and precipitation (PR) at the three locations, covering the period 2080-2100. The x-axis shows integer values indexing the days that span the period 2080-2100 (4 weeks per month).
• Figure 2: As Figure \ref{['fig:time-series']} but after subtracting the seasonal cycle.
• Figure 3: Global Kolgomorov-Smirnov Tests for Precipitation and Temperature. Values exceeding 0.10 are displayed as the same color.
• Figure 4: Autocorrelation and partial autocorrelation functions of daily time series of temperature (T) at the three locations (after subtracting the seasonal cycle). Generated data ACFs and PCFs along the left column can be compared to those of the Held Out 1 and Held Out 2 sets, along the middle and right column respectively. Each pair of rows corresponds to one of the three locations, with Novosibirsk at the top, Hawaii in the middle, and Melbourne at the bottom.
• Figure 5: Like Fig. \ref{['fig:ACF_T']}, for daily precipitation time series.