DiffESM: Conditional Emulation of Temperature and Precipitation in Earth
System Models with 3D Diffusion Models
Abstract
Earth System Models (ESMs) are essential tools for understanding the
interaction of the human and Earth systems. One key application of these
models is studying extreme weather events, such as heat waves or high
intensity precipitation events, which have significant socioeconomic
consequences. However, the computational demands of running a sufficient
number of simulations to robustly characterize expected changes in these
hazards, and therefore provide a strong basis to analyze the ensuing
risks, are often prohibitive. In this paper we demonstrate that
diffusion models – a class of generative deep learning models – can
effectively emulate the spatio-temporal trends of ESM daily output.
Trained on a handful of runs, 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 of
temperature and precipitation that have statistical characteristics
close to the ESM output. This approach requires only a small fraction of
the computational resources that would be needed to run a large ensemble
under any scenario of interest. We evaluate model behavior over a range
of scenarios, time horizons and two ESMs, using a number of extreme
metrics, including ones that have been long established in the climate
modeling and analysis community. Our results show that the samples
produced by 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.