Accurate Attribution and Seasonal Prediction of Climatic Anomalies Using Causal Inference Theory

Abstract

Abstract Using features based on correlation or noncausal dependence metrics can lead to false conclusions. However, recent research has shown that applying causal inference theory in conjunction with Bayesian networks to large-sample-size data can accurately attribute synoptic anomalies. Focusing on the East Asian summer monsoon (EASM), this study adopts a causal inference approach with model averaging to investigate causation of interannual climate variability. We attribute the EASM variability to five winter climate phenomena; our result shows that the eastern Pacific El Niño–Southern Oscillation has the largest causal effect. We also show that the causal precursors of the EASM variability are interpretable in terms of physics. Using linear regression, these precursors can predict the EASM one season ahead, outperforming correlation-based empirical models and three climate models. This study shows that even without large-sample-size data and substantial human intervention, even laymen can implement the causal inference approach to investigate the causes of climatic anomalies and construct reliable empirical models for prediction. Significance Statement We use causal inference theory to redesign the attribution procedure fundamentally and adjust a causal inference approach to commonly used climate research data. Our study shows that the causal inference approach can exhaustively reveal the causes of climatic anomalies with little human intervention, which is impossible for correlation-based studies. According to this attribution, one can construct models with a better predictive performance than the climate and correlation-based empirical models. Therefore, our causal inference approach will tremendously help both meteorologists and laymen (e.g., stakeholders and policymakers) accurately predict climate phenomena and reveal their interpretable causes. We recommend that it become a standard practice in attribution studies and operational prediction.