Correcting Climate Model Sea Surface Temperature Simulations with Generative Adversarial Networks: Climatology, Interannual Variability, and Extremes

Climate models are vital for understanding and projecting global climate change and its associated impacts. However, these models suffer from biases that limit their accuracy in historical simulations and the trustworthiness of future projections. Addressing these challenges requires addressing internal variability, hindering the direct alignment between model simulations and observations, and thwarting conventional supervised learning methods. Here, we employ an unsupervised Cycle-consistent Generative Adversarial Network (CycleGAN), to correct daily Sea Surface Temperature (SST) simulations from the Community Earth System Model 2 (CESM2). Our results reveal that the CycleGAN not only corrects climatological biases but also improves the simulation of major dynamic modes including the El Niño-Southern Oscillation (ENSO) and the Indian Ocean Dipole mode, as well as SST extremes. Notably, it substantially corrects climatological SST biases, decreasing the globally averaged Root-Mean-Square Error (RMSE) by 58%. Intriguingly, the CycleGAN effectively addresses the well-known excessive westward bias in ENSO SST anomalies, a common issue in climate models that traditional methods, like quantile mapping, struggle to rectify. Additionally, it substantially improves the simulation of SST extremes, raising the pattern correlation coefficient (PCC) from 0.56 to 0.88 and lowering the RMSE from 0.5 to 0.32. This enhancement is attributed to better representations of interannual, intraseasonal, and synoptic scales variabilities. Our study offers a novel approach to correct global SST simulations and underscores its effectiveness across different time scales and primary dynamical modes.