Computational Modelling Group

How sensitive is ocean model utility to resolution?

Started
7th January 2013
Research Team
Maike Sonnewald
Investigators
Kevin Oliver

A horizontal section along the 2021 m level showing potential density (kg m−3) in December 1978. The contour shows the 1027.8 isopycnal. Figure a shows the smooth nature of ORCA1, while figure b highlights the more variable nature of ORCA12.

One of the most intriguing problems in recent ocean modeling research is the impact of varying model resolution on model accuracy. Increasing model resolution one includes more of the important processes. However, the increase in accuracy with resolution is unlikely to be linear. Thus, as computational cost increases with resolution, a critical assessment of achieved benefits is prudent. Here we analyse a suite of realistic and compatible global ocean model runs from coarse (1o, ORCA1), eddy-permitting (1/4o, ORCA025) and eddy resolving (1/12o, ORCA12) resolutions. Comparisons of steric height variability (varSH) highlight changes in ocean density structure, revealing impacts on mechanisms such as downwelling and eddy energy dissipation. We assess vertical variability using the covariace of the deep and shallow varSH. Together with assessing isopycnal movements, we demonstrate the influence of deep baroclinic modes and regions where the barotropic flow sheds eddies. Significant changes in the deepwater formation and dispersion both in the Arctic and Antarctic are found between resolutions. The varSH increased from ORCA1 to ORCA025 and ORCA12, particularily in the Southern Ocean and Western Boundary Currents. However, there is no significant covariance between the surface and deep in ORCA1, while ORCA025 and ORCA12 show significant covariance, implying an important missing energy pathway in ORCA1. Comparing ORCA025 and ORCA12 we see significant differences in eddy energy dissipation. We assess the impact of varying model resolution on the mean flow, discussing implications to dissipation pathways on model accuracy, with reference to stochastic parameterisation schemes.

Categories

Physical Systems and Engineering simulation: Climate, Earth surface dynamics, Oceanography, Turbulence, Wave propagation

Algorithms and computational methods: FFT, Multi-core, statistical analysis

Simulation software: NEMO

Visualisation and data handling software: HDF5

Software Engineering Tools: Vim

Programming languages and libraries: C, Fortran, Python

Computational platforms: HECToR, Linux

Transdisciplinary tags: Complex Systems