tomchor / CSI_insights

Supporting code for "Insights into the mixing efficiency of submesoscale Centrifugal-Symmetric instabilities"

This repository contains the code necessary for generating the data in the paper "Insights into the mixing efficiency of submesoscale Centrifugal-Symmetric instabilities". The code is fully written in Julia, the main package is Oceananigans.jl, and simulations output NetCDF files.

Organization

• Files Project.toml and Manifest.toml ensure a reproducible Julia environment that should generate identical results to those in the paper, apart from the randomness in the initial noise and machine imprecision.

• File surfjet_np.jl is the main script (it defines the grid, model, simulation, and finally runs the simulation)

• File diagnostics.jl is an auxiliary file that contains diagnostics that will be written to the output netcdf files

• File jetinfo.jl is an auxiliary file that contains the parameters for each simulation

Running the simulations

Before running the simulations you'll need to download and install the necessary packages (which are described in Project.toml and Manifest.toml). This is done automatically by Julia(instructions here) and should not be done manually!

After the packages are properly installed, in order to properly run the simulations you'll need to specify a few options. The command below has an example:

julia --project surfjet_np.jl --simname=PNN_CIfront1 --factor=1

The flag --project is necessary to start Julia with the package versions defined in Project.toml and Manifest.toml.

surfjet_np.jl is the main script.

--simname=PNN_CIfront1 is the name of the simulation you want to run. The prefix PNN means it should be run with a periodic, nonperiodic, nonperiodic. The other option is FNN which means flat, nonperiodic, nonperiodic (which runs a 2D simulation). The rest of the name needs to be defined in the file jetinfo.jl.

--factor=1 is the "coarsening factor". Setting a value of 1 runs a simulation with full resolution. Factor of 2 halves the grid spacing. Factor of 4 divides the spacings by 4 and so on.

This code will run either on a CPU or on a GPU, however, the full (i.e. factor=1) simulations are large (~130 million points) so they are too large to run on a CPU and a GPU is needed.