Bayesian weighting of climate models based on climate sensitivity

Area of Science:

• Climate science
• Earth system science
• Climate modeling

Background:

• Climate model ensembles can produce biased projections due to 'hot models' with high climate sensitivities.
• Existing methods like model emulators or culling have limitations in addressing this bias.

Purpose of the Study:

• To implement Bayesian Model Averaging (BMA) to address the 'hot model' problem without excluding models.
• To produce unbiased posterior probability distributions of model weights using multiple lines of evidence.

Main Methods:

• Utilized Bayesian Model Averaging (BMA) as a statistical framework.
• Constructed model weights based on multiple lines of evidence for Earth's climate sensitivity.

Main Results:

• BMA yields updated multi-model ensemble projections of end-of-century global mean surface temperature increases: 2°C (SSP1-2.6) and 5°C (SSP5-8.5).
• These BMA-derived estimates are lower than those from a simple multi-model mean of the CMIP6 ensemble.
• The approach retains some weight on low-probability models, acknowledging potential extreme climate sensitivity values.

Conclusions:

• Bayesian Model Averaging offers a robust framework for unbiased climate change projection.
• This method integrates scientific evidence effectively, providing more reliable future climate scenarios.