Modeling and Prediction of Large-Scale Climate Variability by Inferring Causal Structure

This study addresses how to model and predict large-scale climate variability, such as the El Nino-Southern Oscillation (ENSO). We introduce a framework for inferring the macroscale causal structure of the climate system using a spatial-dimension reduction and high-dimensional variable selection. The framework encodes the causal structure into a structural causal model, which captures the mechanisms and diversity of ENSO. It thus has a potential to reveal other physical processes within the climate system. The model predicts ENSO at a 1-month lead time with high accuracy, and the recursive predictions at multi-month leads are still reliable, even in a different climate state. The stand-alone oceanic experiments capture the observed oceanic response, proving the model's capability to predict large-scale climate variability using fragmentary information. This study demonstrates the potential for inferring causal structures to explain, model, and predict large-scale climate variability such as ENSO.