Machine-learning emergent constraints on surface albedo feedback over Arctic land regions

Area of Science:

• Climate Science
• Earth System Science
• Machine Learning Applications

Background:

• Surface albedo feedback (SAF) significantly amplifies Arctic warming, impacting climate, ecosystems, and infrastructure.
• Earth system models (ESMs) show substantial uncertainties in projecting SAF, hindering accurate Arctic warming estimates.

Purpose of the Study:

• To develop a machine-learning method using emergent constraints (ECs) to reduce uncertainties in SAF projections over Arctic land.
• To leverage in-situ observations and historical albedo-temperature dynamics to constrain future SAF.

Main Methods:

• Developed a machine-learning approach integrating emergent constraints (ECs).
• Utilized in-situ observations and historical albedo-temperature data (1985-2014) to establish physical relationships.
• Applied these relationships to constrain SAF projections (2070-2099) across ESM ensembles.

Main Results:

• Constrained SAF projections were reduced by 0.29-0.52 W m⁻² K⁻¹ across emission scenarios.
• Uncertainties in SAF projections were decreased by 45-55% compared to unconstrained estimates.
• The method established a physical link between historical and future SAF dynamics.

Conclusions:

• The machine-learning EC method successfully reduces SAF projection uncertainties.
• Enhanced confidence in regional Arctic climate projections is achieved.
• Provides more precise insights for climate adaptation and policy in high-latitude regions.