When stratified ocean currents flow across steep underwater topography, they can generate internal lee waves. These waves help transfer tidal energy into turbulence and mixing, making them an important part of the ocean's internal energy pathway.
However, even high-resolution ocean models cannot resolve every turbulent motion and must rely on parameterizations for unresolved vertical mixing. A new study published in Atmospheric and Oceanic Science Letters shows that if this parameterized mixing becomes too strong, it can interfere with wave motions that the model is already capable of resolving.
Using the widely applied K-profile parameterization (KPP), the researchers simulated stratified tidal flow over a supercritical sill. The standard KPP configuration produced strong surface- and bottom-boundary-layer mixing near the sill crest. This weakened local stratification and strongly suppressed the downstream lee-wave signal.
"The key message is not that ocean models should use less mixing everywhere," says Prof. Xuefeng Zhang from Hainan Tropical Ocean University, China, corresponding author of the study. "The problem arises when parameterized boundary-layer mixing becomes strong enough to alter the stratification in the wave-generation region before a resolved lee wave can fully develop."
The lee-wave response was particularly sensitive to the critical bulk Richardson number, Ric. Reducing Ric from 0.3 to 0.03 weakened excessive boundary-layer mixing, partly restored stratification near the sill crest, and produced a much clearer downstream wave signal. By comparison, changes in the empirical turbulent-shear coefficient, Cv, had a much smaller effect within the tested range.
Additional experiments showed that the excessive damping came mainly from the KPP surface and bottom boundary-layer components rather than from its interior mixing formulation. Yet completely removing boundary-layer mixing was not a practical solution. In a realistic shallow-sea simulation, doing so caused insufficient upper-ocean mixing and larger temperature errors. Reducing Ric instead improved the simulated thermal structure while retaining the mixing needed by the regional ocean.
A supplementary inverse physics-informed neural network analysis also recovered a parameter pair close to the reduced-Ric experiment, providing an additional consistency check within the synthetic sensitivity ensemble.
The findings suggest that turbulence parameterizations should be evaluated together with model resolution and the physical processes being explicitly resolved. As ocean models move toward finer scales, excessive subgrid mixing may otherwise smooth out the very dynamics that higher resolution is intended to capture.
