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

Surface albedo feedback (SAF) amplifies warming in northern high latitudes, affecting the Arctic climate system, ecosystems, infrastructure, and global trade routes. However, Earth system models (ESMs) exhibit large uncertainties in SAF projections, complicating future Arctic warming estimates. Here, we develop a machine-learning method based on emergent constraints (ECs) and use in-situ observations to constrain SAF projections over Arctic land regions. Our approach leverages a physical relationship between historical albedo-temperature dynamics (1985–2014) and future SAF (2070–2099) across ESM ensembles. The constrained SAF is reduced by 0.29–0.52 W m-2 K-1 across emission scenarios, with uncertainties decreased by 45–55% compared to unconstrained projections. These findings enhance confidence in regional climate projections, offering more precise insights for climate adaptation and policy in vulnerable high-latitude communities. This study uses machine learning and ground data from the Arctic to constrain land albedo feedback. The projected feedback is reduced by 0.29–0.52 W m⁻² K⁻¹ and uncertainty by 45–55%, suggesting that climate models overestimate this key warming amplifier.