Transferring climate change physical knowledge.

Area of Science:

• Climate Science
• Machine Learning Applications
• Earth System Modeling

Background:

• Reliable climate projections are crucial for adaptation and mitigation strategies.
• Current Earth system models possess significant uncertainties, limiting projection accuracy.
• Existing methods struggle to capture the climate system's nonlinear complexity.

Purpose of the Study:

• To reduce the uncertainty spread in 21st-century global surface air temperature projections.
• To leverage Machine Learning for optimal merging of Earth system model simulations and historical observations.
• To improve regional temperature pattern accuracy in climate projections.

Main Methods:

• Utilized a Transfer Learning approach within Machine Learning.
• Integrated knowledge from global temperature maps simulated by Earth system models.
• Incorporated observed historical temperature data to train the model.

Main Results:

• Achieved an uncertainty reduction of over 50% compared to state-of-the-art methods.
• Demonstrated improved regional temperature patterns in projections.
• Provided narrower projections uncertainty crucial for climate adaptation.

Conclusions:

• Machine Learning, specifically Transfer Learning, offers a powerful tool to enhance climate projection reliability.
• The developed method effectively reduces uncertainty and improves regional detail in temperature projections.
• This approach is vital for informing effective climate adaptation and mitigation planning.