Atlantic Ocean over one_degree_simulation

[uganzedo1974]

Hi ;

Thanks for your work!! Excellent.
I have a question that worked in the previous version, but not in this one. I'm referring to the example mediterranean_simulation_with_ecco_restoring.jl
or
one_degree_simulation.jl. If we narrow the domain to the west, keeping it as the Atlantic Ocean, with the North-South narrowing, the same problem doesn't appear.

[ Info: Time: 0 seconds, Iteration 0, Δt 1.500 minutes, max(vel): (5.70e-01, 4.30e-01, 8.36e-02), max(T): 25.11, min(T): 2.43, wtime: 1.723 minutes
[ Info: ... simulation initialization complete (1.374 minutes)
[ Info: Executing initial time step...
[ Info: ... initial time step complete (1.369 minutes).
[ Info: _time = 1000.0, iteration = 100: NaN found in field u_ocean. Stopping simulati_on

The time jumps to 1000 and the same error occurs, and it's not a matter of initial data. Why is this happening now? It didn't happen before, and we could run limited simulations in the Atlantic or Mediterranean.

Thank you, and sorry for the inconvenience.
Unai

using Oceananigans
using Oceananigans.Units
using Dates
using CUDA
import ClimaOcean

arch = GPU()
grid = LatitudeLongitudeGrid(arch,
size = (1440, 560, 10),
halo = (7, 7, 7),
longitude = (180, 360),
latitude = (-70, 70),
z = (-3000, 0))

bathymetry = ClimaOcean.regrid_bathymetry(grid) # builds gridded bathymetry based on ETOPO1
grid = ImmersedBoundaryGrid(grid, GridFittedBottom(bathymetry))

Build an ocean simulation initialized to the ECCO state estimate version 2 on Jan 1, 1993

ocean = ClimaOcean.ocean_simulation(grid)
start_date = DateTime(1993, 1, 1)
set!(ocean.model,
T=ClimaOcean.Metadatum(:temperature; date=start_date, dataset=ClimaOcean.ECCO2Daily()),
S=ClimaOcean.Metadatum(:salinity; date=start_date, dataset=ClimaOcean.ECCO2Daily()))

Build and run an OceanSeaIceModel (with no sea ice component) forced by JRA55 reanalysis

atmosphere = ClimaOcean.JRA55PrescribedAtmosphere(arch)
coupled_model = ClimaOcean.OceanSeaIceModel(ocean; atmosphere)
simulation = Simulation(coupled_model, Δt=5minutes, stop_time=30days)
run!(simulation)

[simone-silvestri]

Hello, are you sure it is not a problem of initial conditions? Your model breaks quite quickly so it suggests that there is some issues with the initial profiles.
Try printing out the model initial state and look if there are issues in the initial tracer distribution.
A sign that something is wrong is salinity equal to zero in some cells

[uganzedo1974]

Hello;
I have tested it for several days with full domains Longitude 0 to 360. The model works well but when reducing the domain length the same error always appears.
time = 1000.0, iteration = 100: NaN found in field u_ocean. Stopping simulation
I've used data from the previous version, and now it doesn't work. In some cases, the model makes a bit of progress if you use a lower At=10, but then it jumps to Time=1000 and throws an error.

[glwagner]

Can we convert this to a discussion? I am not sure if there is a problem with source code --- instead here we are wrestling with producing a setup that runs stably, if I understand correctly.