Toward skillful forecasting of super El Niño events using a diffusion-based westerly wind burst parameterization

Abstract

Forecasting super El Niño remains challenging, partly due to poor representation of westerly wind bursts (WWBs). We developed an artificial intelligence-based denoising diffusion probabilistic model (DDPM) to skillfully parameterize WWBs, capturing their joint modulation by oceanic and atmospheric processes. The DDPM-based scheme effectively captures observed WWBs’ characteristics (e.g., frequency, intensity, and spatial center). When implemented in the Community Earth System Model, it outperforms both the control (CTRL, without WWBs parameterization) and conventional warm pool eastern edge (WPEE)-dependent parameterization in predicting intensity and seasonal phase-locking for super El Niños (1982/83, 1997/98, 2015/16). This improvement stems from DDPM’s realistic WWBs representation, correcting CTRL and WPEE’s biases of overly weak and westward-shifted winds during El Niño growth. Consequently, DDPM produces more realistic eastern Pacific sea surface temperature anomaly warming patterns. These findings underscore WWB's accuracy as key to super El Niño prediction and demonstrate machine learning’s potential for WWB's parameterization.