UC San Diego

Abstract. Ancient air trapped in ice core bubbles has been paramount to developing our understanding of past climate and atmospheric composition. Before air bubbles become isolated in ice, the atmospheric signal is altered in the firn column by transport processes such as advection and diffusion. However, the influence of low-permeability layers and barometric pumping (driven by surface pressure variability) on firn air transport is not well understood and is not readily captured in conventional one-dimensional (1-D) firn air models. Here we present a two-dimensional (2-D) trace gas advection–diffusion–dispersion model that accounts for discontinuous horizontal layers of reduced permeability. We find that layering or barometric pumping individually yields too small a reduction in gravitational settling to match observations. In contrast, when both effects are active, the model's gravitational fractionation is suppressed as observed. Layering focuses airflows in certain regions in the 2-D model, which acts to amplify the dispersive mixing resulting from barometric pumping. Hence, the representation of both factors is needed to obtain a realistic emergence of the lock-in zone. In contrast to expectations, we find that the addition of barometric pumping in the layered 2-D model does not substantially change the differential kinetic fractionation of fast- and slow-diffusing trace gases. Like 1-D models, the 2-D model substantially underestimates the amount of differential kinetic fractionation seen in actual observations, suggesting that further subgrid-scale processes may be missing in the current generation of firn air transport models. However, we find robust scaling relationships between kinetic isotope fractionation of different noble gas isotope and elemental ratios. These relationships may be used to correct for kinetic fractionation in future high-precision ice core studies and can amount to a bias of up to 0.45 °C in noble-gas-based mean ocean temperature reconstructions at WAIS Divide, Antarctica.