Learning to correct climate projection biases

Paper in Journal of Advances in Modeling Earth Systems, 2021

Baoxiang Pan¹, Gemma Anderson¹, Andre Goncalves¹, Donald Lucas¹, Celine Bonfils¹, Jiwoo Lee¹, Yang Tian¹, Hsi-yen Ma¹
Lawrence Livermore National Lab¹

Overview

This work develops the Regularized Adversarial Domain Adaptation (RADA) methodology to correct historical climate projection biases. RADA is built on the generative adversarial nets (GANs, Goodfellow et al. 2014), adding various constraints to guarantee phyiscal coherency. The general idea is illustrated as follows:

In the top, GAN is applied to generate cat samples. In the bottom, we make use of the adversarial learning idea to match samples from a source domain (climate simulations), to corresponding samples in a target domain (climate observations). This line of research is often named domain adaptation in the machine learning literature. We can not apply supervised learning, since climate is a chaotic system, preventing us from obtaining sufficient paired observation and simulation data to identify and correct climate model biases.

To guarantee physical coherency, we translate the three requirements Ehret et al. (2012) proposed for a ``perfect'' climate model bias corrector into three regularization terms to the adversarial domain adaption process, namely cycle consistency (Zhu et al. 2017), dynamical consistency (Pan et al. 2019), and dynamical dependency. The final model is illustrated as follows:

Below we provide guidance for applying RADA to correct the Community Earth System Model version 2 (CESM2, Danabasoglu et al. 2020) daily precipitation projection over the contiguous United States (CONUS).

Getting started

Obtain the histotical climate simulation and observation data

The Community Earth System Model version 2 (CESM2) historical simulation data are available through https://esgf-node.llnl.gov/projects/cmip6/.

The CPC unified gauge-based analysis of daily precipitation data are available through https://psl.noaa.gov/data/gridded/data.cpc.globalprecip.html.

The ECMWF atmospheric reanalysis of the 20th century (ERA-20C) data are available through https://www.ecmwf.int/en/forecasts/datasets/reanalysis-datasets/era-20c.

Cut, regrid, and normalize the data:

math -script Data_Processing.m

Train the dynamical regularization module

math -script Dynamical_Regularization.m

Train the RADA bias corrector

math -script RADA.m

Compare with the baselines

math -script Baseline/Baseline_n.m

Evaluation

math -script Evaluation.m

Acknowledgments

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under contract DE-AC52-07NA27344. Lawrence Livermore National Security, LLC. The views expressed here do not necessarily reflect the opinion of the United States Government, the United States Department of Energy,or Lawrence Livermore National Laboratory. This work was supported by LLNL Laboratory Directed Research and Development project 19-ER-032. This document is released with IM tracking number LLNL-JRNL-817982.