Climate models stand as the foundation of our understanding and projecting of global climate change. Yet, their persistent biases constrain both historical accuracy and future projections. Among these biases, the discrepancy in simulating global sea surface temperatures and the misrepresentation of key phenomena like the El Nino-Southern Oscillation (ENSO) pose significant challenges.
In a recent study, led by Prof. HUANG Gang and his team with the Institute of Atmospheric Physics at the Chinese Academy of Sciences, researchers have made progress in addressing this issue. They utilized an unsupervised learning model (an unsupervised learning model learns from data without explicit one to one labels）called CycleGAN, and devised a method to reduce biases in the Community Earth System Model version 2 (CESM2)'s simulation of daily sea surface temperatures (SST).
Their findings, published in
Advances in Atmospheric Sciences, showcase the potential of CycleGAN in not only correcting climatological biases but also substantially enhancing the dynamic modes such as ENSO, Indian Ocean Dipole (IOD), and extreme SST events.
Prof. HUANG's team demonstrated that CycleGAN largely mitigates the excessive westward bias in ENSO SST simulations. This bias, a primary contributor to errors in simulating the Northwest Pacific anomalous anticyclone and East Asian summer monsoon, has proven resistant to traditional correction methods such as quantile mapping.
Prof. Huang remarks, "Our study unveils a novel approach for correcting global SST simulations and underscores its efficacy in enhancing the representation of key dynamic modes. This paves the way for similar methodologies to further refine crucial dynamic modes and the regional climate systems they influence, such as the East Asian monsoon."
