Exploring viscosity space in an eddy‐permitting global ocean model: Is viscosity a useful control for numerical mixing?

A generic shortcoming of constant-depth (or “z-coordinate”) ocean models such as MOM5 and Nucleus for European Models of the Ocean (NEMO) is a tendency for the advection scheme to produce unphysical numerical diapycnal mixing, which may exceed the explicitly parameterized mixing based on observed physical processes. Megann (2018, https://doi.org/10.1016/j.ocemod.2017.11.001) estimated the effective diapycnal diffusivity in the Global Ocean Version 5.0 (GO5.0) 0.25° global implementation of the NEMO model and showed that this was up to 10 times the explicit diffusivity used in the model's mixing scheme and argued that this was at least partly caused by large transient vertical velocities on length scales comparable to the horizontal grid scale. The current UK global NEMO configuration GO6, as used in the Global Coupled Model version 3.1 (GC3.1) and UK Earth System Model (UKESM1), is integrated in forced mode at 0.25° resolution with a range of viscosity parameterizations. In the present study, the effective diffusivity is evaluated for each integration and compared with the explicit value from the model mixing scheme, as well as with that in the control (using the default viscosity). It is shown that there is a strong correspondence between lower viscosity and enhanced numerical mixing and that larger viscosities lead to a marked reduction in the unrealistic internal temperature drift seen in the control configuration, without incurring excessive damping of the large-scale circulation, mixed layer depths, or sea ice cover. The results presented here will inform the choices made in global ocean configurations used in climate and Earth System models following the sixth Coupled Model Intercomparison Project (CMIP6)