Abstract

Global climate models are central tools for understanding past and future climate change. The assessment of model skill, in turn, can benefit from modern data science approaches. Here I will present recent work on causal discovery algorithms as a novel approach for process-oriented climate model evaluation [1,2].

Following an introduction to the concept of causal discovery, I will move on to key scientific implications of this new approach when applied to global sea level pressure datasets. Using causal networks learned from meteorological reanalysis data (as a proxy for observations) and from CMIP5 climate model output, I demonstrate that climate models which better reproduce the observed causal information flow also better reproduce important precipitation patterns over highly populated areas such as the Indian subcontinent, Africa, East Asia, Europe, and North America. In addition, the method identifies expected model interdependencies due to shared development backgrounds of many climate models. Finally, I find that causal network metrics provide stronger relationships for constraining precipitation projections under climate change than traditional model evaluation metrics. Such emergent relationships highlight the potential of causal discovery approaches to constrain longstanding uncertainties in climate change projections.

[1] Nowack P, Runge J, Eyring V, Haigh JD. Causal networks for climate model evaluation and constrained projections. Nature Communications 11, 1415 (2020). [2] Runge J, Nowack P, Kretschmer M, Flaxman S, Sejdinovic D. Detecting and quantifying causal associations in large nonlinear time series datasets. Science Advances 5, eaau4996 (2019).