Closed eastjames closed 1 month ago
Tagging @msulprizio
We have seen similar patterns in other nested grid simulations of extremely high concentrations in a checkerboard pattern. It often has to do with the Courant number calculation. Others have been able to get around it by increasing the transport timestep and/or by increasing the nested grid buffer cells (default is 3 cells on each edge). You may also want to check the boundary conditions and/or met fields to see if anything stands out there as triggering this issue.
See also similar issues:
Halving the advection time step in geoschem_config.yml
solved the issue. This suggests the problem is most likely related to the CFL condition being violated, pointed out by @msulprizio and @hannahnesser in offline conversations.
Old:
#============================================================================
# Timesteps settings
#============================================================================
timesteps:
transport_timestep_in_s: 300
New:
#============================================================================
# Timesteps settings
#============================================================================
timesteps:
transport_timestep_in_s: 150
For sake of reproducing this issue, here's a script to run GEOS-Chem with the typical time step (300 s
) and the reduced time step (150 s
). Plotting the maximum methane concentration in any grid box in the domain shows that the more frequent transport time step prevents this issue (as James found out). It is an imperfect comparison, but interestingly, the 300 s
simulation took ~4 hours and the 150 s
simulation took only ~4.75 hours.
Name and Institution (Required)
Name: James East Institution: ACMG
Description of your issue or question
I ran into a bug background simulation for the jacobian runs. This simulation should have 0 emissions, but a pattern that looks like numerical instability shows up partway through the simulation. The CH4 concentrations blow up to unphysical values (~600,000 ppb)
The plots show simple average CH4 from the GEOS-Chem SpeciesConc files. The bug appears to happen in the stratosphere.
This is using lognormal errors and the
feature/2d_sensitivities
branch. The error is in the GEOS-Chem output and doesn't result from any IMI changes in that branch.