Combining global climate models using graph cuts.

Area of Science:

• Climate Science
• Computational Methods
• Data Analysis

Background:

• Global Climate Models (GCMs) are crucial for climate projections, often combined into Multi-Model Ensembles (MMEs) to reduce uncertainty.
• The Multi-Model Mean (MMM) is a common MME approach, but it assigns uniform weights across all locations, potentially overlooking local model performance.
• Existing methods for combining climate models may not fully leverage spatial information or local model accuracy.

Purpose of the Study:

• To propose and evaluate a new method for reconstructing climate variable means from MMEs.
• To address the limitation of uniform model weighting in traditional approaches like MMM.
• To improve the accuracy and spatial consistency of climate projections derived from MMEs.

Main Methods:

• A novel approach utilizing a computer vision technique, graph cuts, is introduced.
• This method selects the most appropriate model for each grid point based on local performance.
• The approach also ensures overall spatial consistency of the reconstructed climate fields.

Main Results:

• The graph cuts approach demonstrated lower biases compared to the Multi-Model Mean (MMM).
• Spatial continuity of the climate fields was preserved, similar to existing methods.
• Validation against ERA5 reanalyses and a perfect model experiment confirmed the method's effectiveness.

Conclusions:

• The proposed graph cuts method offers a superior alternative to MMM for reconstructing climate means from MMEs.
• This technique effectively incorporates local model performance and spatial consistency for more reliable climate projections.
• The findings suggest a significant advancement in utilizing MMEs for climate change research.