Abstract. Ice cores are an important record of the past surface mass balance
(SMB) of ice sheets, with SMB mitigating the ice sheets' sea level
impact over the recent decades. For the Antarctic Ice Sheet (AIS),
SMB is dominated by large-scale atmospheric circulation, which
collects warm moist air from further north and releases it in the form
of snow as widespread accumulation or focused atmospheric rivers on
the continent. This suggests that the snow deposited at the surface of
the AIS should record strongly coupled SMB and surface air temperature
(SAT) variations. Ice cores use δ18O as a proxy for SAT
as they do not record SAT directly. Here, using isotope-enabled
global climate models and the RACMO2.3 regional climate model, we
calculate positive SMB–SAT and SMB–δ18O annual
correlations over ∼90 % of the AIS. The high spatial
resolution of the RACMO2.3 model allows us to highlight a number of
areas where SMB and SAT are not correlated, and we show that wind-driven
processes acting locally, such as foehn and katabatic effects, can
overwhelm the large-scale atmospheric contribution in SMB and SAT
responsible for the positive SMB–SAT annual correlations. We focus in
particular on Dronning Maud Land, East Antarctica, where the ice
promontories clearly show these wind-induced effects. However, using
the PAGES2k ice core compilations of SMB and δ18O of
Thomas et al. (2017) and Stenni et al. (2017), we obtain a
weak annual correlation, on the order of 0.1, between SMB and
δ18O over the past ∼150 years. We obtain an
equivalently weak annual correlation between ice core SMB and the SAT
reconstruction of Nicolas and Bromwich (2014) over the past ∼50 years, although the ice core sites are not spatially co-located
with the areas displaying a low SMB–SAT annual correlation in the
models. To resolve the discrepancy between the measured and modeled
signals, we show that averaging the ice core records in close spatial
proximity increases their SMB–SAT annual correlation. This increase
shows that the weak measured annual correlation partly results from
random noise present in the ice core records, but the change is not
large enough to match the annual correlation calculated in the
models. Our results thus indicate a positive correlation between SAT
and SMB in models and ice core reconstructions but with a weaker value
in observations that may be due to missing processes in models or some
systematic biases in ice core data that are not removed by a simple
average.