Global Precipitation Correction Across a Range of Climates Using CycleGAN

• Published in: Geophysical research letters Vol. 51; no. 4
• Main Authors: McGibbon, J., Clark, S. K., Henn, B., Kwa, A., Watt‐Meyer, O., Perkins, W. A., Bretherton, C. S.
• Format: Journal Article
• Language: English
• Published: Washington John Wiley & Sons, Inc 28.02.2024; Wiley
• Subjects: Atmospheric models; Bias; bias correction; Climate change; Climate models; Climatology; cyclegan; Distribution functions; Diurnal cycle; Diurnal variations; Environmental impact; gan; Generative adversarial networks; Global precipitation; Impact prediction; Machine learning; Mean precipitation; Morphing; neural network; Precipitation; Precipitation patterns; Precipitation variations; Sea surface; Simulation; Systematic errors
• ISSN: 0094-8276; 1944-8007
• DOI: 10.1029/2023GL105131

Accurate precipitation simulations for various climate scenarios are critical for understanding and predicting the impacts of climate change. This study employs a Cycle‐generative adversarial network (CycleGAN) to improve global 3‐hr‐average precipitation fields predicted by a coarse grid (200 km) atmospheric model across a range of climates, morphing them to match their statistical properties with those of reference fine‐grid (25 km) simulations. We evaluate its performance on both the target climates and an independent ramped‐SST simulation. The translated precipitation fields remove most of the biases simulated by the coarse‐grid model in the mean precipitation climatology, the cumulative distribution function of 3‐hourly precipitation, and the diurnal cycle of precipitation over land. These results highlight the potential of CycleGAN as a powerful tool for bias correction in climate change simulations, paving the way for more reliable predictions of precipitation patterns across a wide range of climates. Plain Language Summary Using CycleGAN, a machine learning technique, we can remove key biases in precipitation simulated by a fast, coarse‐grid atmospheric model. This method morphs maps of the output precipitation to match typical characteristics of a slower but more accurate fine‐grid configuration, correcting systematic errors in both long‐term average spatial precipitation patterns and 3‐hourly precipitation variations. It retains skill in intermediate climate states unseen in training, making it a useful tool for climate change simulations. Key Points A Cycle‐generative adversarial network can debias precipitation across a range of climate forcings The model is able to debias data from intermediate forcings not present in training data One model is able to debias across climates without being explicitly told the input climate