PNNL

Abstract

Arctic amplification (AA) refers to the phenomenon whereby the Arctic warms faster than the global average in response to increasing greenhouse gas concentrations. AA is often quantified as the ratio of Arctic-mean to global-mean surface air temperature (SAT) trends. Based on observations from 1980-2022, the annual-mean AA reached a value of 4.2 when the Arctic is defined as the region north of 70°N. While climate models robustly simulate the presence of AA, they rarely capture its observed magnitude. Sweeney et al. (2023) recently suggested that much of the discrepancy between modeled and observed annual-mean AA arises from internal variability. However, AA shows large seasonality, so the model-observation discrepancy also varies with season. We find that spring (March-May, MAM) experiences the largest discrepancy: observed AA is 4.2, while the multi-model mean AA is only 2.7. This raises several key questions: (1) What is the role of internal variability in the observed spring AA? (2) How does simulated spring AA compare to observations when internal variability is removed? and (3) If internal variability is significant, what mechanisms drive it? To answer these questions, we modified the machine learning algorithm developed by Sweeney et al. (2023) and trained it on simulated multi-decadal seasonal SAT and sea level pressure (SLP) trend maps. This allows us to isolate the contribution of internal variability to springtime Arctic and global-mean SAT trends. After removing the influence of internal variability, we reconcile the modeled-observed discrepancy in springtime AA. The contribution of internal variability to recent springtime Arctic surface warming, as estimated by machine learning, is supported by an independent dynamic adjustment approach. We further identify an atmospheric circulation pattern in observations that is related to this internally driven spring Arctic warming. Further work is warranted to better understand the processes and mechanisms behind this circulation pattern.