Causal networks for climate model evaluation and constrained projections.

Area of Science:

• Climate Science
• Data Science
• Meteorology

Background:

• Global climate models are essential for understanding climate change.
• Assessing climate model skill can be improved with data science.
• Current evaluation methods may not fully capture model performance.

Purpose of the Study:

• To apply causal discovery algorithms for process-oriented climate model evaluation.
• To develop objective causal network fingerprints for model assessment.
• To explore the potential of causal networks in constraining climate change projections.

Main Methods:

• Applied causal discovery algorithms to sea level pressure data from climate model simulations and meteorological reanalyses.
• Generated causal networks (fingerprints) to represent model behavior and observations.
• Utilized network metrics for model evaluation and comparison.

Main Results:

• Climate models with fingerprints closer to observations showed better reproduction of precipitation patterns over populated regions.
• Identified interdependencies among climate models stemming from shared development.
• Network metrics demonstrated stronger relationships for constraining precipitation projections than traditional metrics.

Conclusions:

• Causal networks provide an objective pathway for process-oriented climate model evaluation.
• Improved model evaluation using causal networks can lead to more reliable precipitation projections.
• This approach has the potential to reduce longstanding uncertainties in climate change projections.