Changes in 1958–2019 Greenland Surface Mass Balance are Attributable to both Greenhouse Gases and Anthropogenic Aerosols

Abstract. Greenland Ice Sheet (GrIS) mass loss is a main contributor to rising Global Mean Sea Level (GMSL), exhibiting decadal variability due to surface mass balance (SMB) changes. Greenhouse gases (GHG) have long been identified as a key driver of GrIS mass loss through warming-induced runoff. However, there has not been a formal attribution of historical GrIS SMB changes to GHG and the potential role for other forcings such as anthropogenic aerosols (AAER). Here, we use the Community Earth System Model version 2 large ensemble and single forcing large ensemble (CESM2-LE and CESM2-SFLE) to formulate a detection and attribution analysis for historical GrIS SMB changes. We show that the decadal variability of SMB is forced by historical radiative forcing attributable to both GHG and AAER through their forced changes of runoff. This highlights that, in addition to the frequently mentioned GHG, AAER also contributes to SMB changes during the historical period. GHG influences GrIS runoff mainly through long-term radiatively-forced warming, while AAER influences it through the decadal variability of atmospheric circulation that projects onto a Greenland blocking pattern, leading to relative cooling from cyclonic circulation over Greenland pre-1980 and relative warming from anti-cyclonic circulation thereafter. The attribution of SMB, and specifically runoff, to AAER has a lower signal-to-noise ratio (S/N) than the attribution to GHG due to both a weaker signal and wider confidence intervals. The lower S/N in attributing runoff changes to AAER is partly due to a smaller temperature response in AAER than in GHG and partly due to a mean state temperature dependency of the runoff sensitivity. In simulations with only AAER, the climate is colder than in simulations with all forcings or only GHG, leading to more time below freezing when temperature variations do not affect runoff as much. We resolve this issue by comparing simulations with all forcings with simulations in which everything-but-AAER is changing, thereby stressing the need to account for mean state dependencies when conducting detection and attribution with single forcing simulations.