Abstract The Earth system model is central to understanding climate change and informing policy decisions. However, current models, with their coarse resolution and inherent biases, limit the utility of climate simulations. In this study, we introduce a multivariate generative downscaling model (MVGDM) that downscales global climate simulations from a 100 km to a 25 km resolution, while simultaneously correcting climate simulation biases. The MVGDM provides accurate simulations by refined resolution and addressing biases in three variables: sea surface temperature (SST), 2‐m temperature (T2M), and 500‐hPa geopotential height (Z500), reducing climatological biases by 72%, 79%, and 71%, respectively. Additionally, the MVGDM mitigates the common westward bias in El Niño‐Southern Oscillation (ENSO)‐related SST anomalies and significantly improves the simulation of the Indian Ocean Dipole. With high‐resolution simulations and improved representation of multi‐scale physical processes, the MVGDM substantially enhances the simulation of climate extremes, and shows some potential for improving future climate projections.

Read original article

Read original article