A comparison of two causal methods in the context of climate analyses

• DOI: 10.5194/egusphere-2023-2212
• Publication Date: 2023-10-05T10:07:30Z
• Authors: David Docquier; Giorgia Di Capua; Reik V. Donner; Carlos A. L. Pires; Amélie Simon; Stéphane Vannitsem
• Keywords: Spurious relationship; Causation; Independence; Causal model; Strengths and weaknesses; Causality

Abstract. Correlation does not necessarily imply causation, and this is why causal methods have been developed to try disentangle true links from spurious relationships. In our study, we use two methods, namely the Liang-Kleeman information flow (LKIF) Peter Clark momentary conditional independence (PCMCI) algorithm, apply them four different artificial models of increasing complexity one real-case study based on climate indices in North Atlantic Pacific. We show that both are superior classical correlation analysis, especially removing links. LKIF PCMCI display some strengths weaknesses for three simplest models, with performing better a smaller number variables, being best larger variables. Detecting fourth model more challenging as system nonlinear chaotic. For indices, present similarities differences at monthly time scale. One key identifies Arctic Oscillation (AO) largest driver, while El Niño-Southern (ENSO) main influencing variable PCMCI. More research needed confirm these links, particular including methods.