ENSO-ASC 1.0.0: ENSO deep learning forecast model with a multivariate air–sea coupler

Abstract. The El Niño–Southern Oscillation (ENSO) is an extremely complicated ocean–atmosphere coupling event, the development and decay of which are usually modulated by energy interactions between multiple physical variables. In this paper, we design a multivariate air–sea coupler (ASC) based on graph using features On basis coupler, ENSO deep learning forecast model (named ENSO-ASC) proposed, whose structure adapted to characteristics dynamics, including encoder decoder for capturing restoring multi-scale spatial–temporal correlations, two attention weights grasping different strengths start calendar months varied effects variables in amplitudes. addition, datasets same resolutions used train model. We firstly tune performance optimal compare it with other state-of-the-art models. Then, evaluate skill from contributions predictors, effective lead time months, spatial uncertainties, further analyze underlying mechanisms. Finally, make predictions over validation period 2014 2020. Experiment results demonstrate that ENSO-ASC outperforms Sea surface temperature (SST) zonal wind crucial predictors. correlation Niño 3.4 index 0.78, 0.65, 0.5 within 6, 12, 18 respectively. From heat map analyses, also discover common challenges predictability, such as forecasting skills declining faster when making forecasts through June–July–August errors being more likely show up western central tropical Pacific Ocean longer-term forecasts. can simulate strengths, forecasted SST patterns reflect obvious Bjerknes positive feedback mechanism. These indicate effectiveness superiority our predicting analyzing dynamic mechanisms sophisticated way.