Skip static interpolation step if input grid isn't a unit sphere #1259
Conversation
This function uses the sum of cellArea for all owned cells to approximate the surface area of the sphere. This function returns true if the sum of cellArea is close to unit sphere surface area. Otherwise this function returns false. cellAreas are simple to sum, while direct checks of distance of cell centers to the origin could be more expensive.
If static_interp is repeated, the mesh is scaled to unrealistic dimensions. This can cause obvious science issues but can also cause issues for running the init_atmosphere core. With unrealistic distances, the process to assign soil categories may request too much memory and cause the init_atmosphere core to fail.
| @@ -51,7 +51,6 @@ subroutine init_atm_setup_case(domain, stream_manager) | |||
| type (domain_type), intent(inout) :: domain | |||
| type (MPAS_streamManager_type), intent(inout) :: stream_manager | |||
|
|
|||
|
|
|||
There was a problem hiding this comment.
I'd recommend not changing whitespace here.
| !> | ||
| !> This routine will use the global sum of cell areas to | ||
| !> approximate the surface area of the sphere. If this surface area | ||
| !> is within a factor of 4*pii, the mesh is likely still a unit |
There was a problem hiding this comment.
The statement "within a factor of 4pii" could be misinterpreted to mean that the surface area is +/- 4pii of something.
Perhaps it would be better to reframe this description slightly to say that the routine returns true if it determines that the mesh is defined on a unit sphere -- which is the critical point -- and to then follow up with some notes that due to inexactness of arithmetic, the routine actually checks that the total sphere area is not larger than twice the area of a unit sphere.
|
|
||
| surfArea = 0.0_RKIND | ||
| g_surfArea = 0.0_RKIND | ||
| tol = 2.0_RKIND |
There was a problem hiding this comment.
This could be a Fortran parameter (i.e., a const).
| type (dm_info), intent(in) :: dminfo | ||
| type (mpas_pool_type), intent(inout) :: mesh | ||
|
|
||
| ! Output vars |
There was a problem hiding this comment.
I'd suggest Return value rather than Output vars.
|
@mgduda, these fixup commits should address your comments. I'll rebase them if they're accepted. |
This PR adds a function to the init_atmosphere core to quickly check if a grid is likely a unit sphere and uses the function to throw an error if the static interpolation step is about to be repeated.
Most cases for the init_atmosphere model assume the input grid is a unit sphere and involve calculations to scale up to a new radius. Unlike the other cases, the case to generate real-data initial conditions is typically run a few times in sequence with different options. It's possible users could cause unintuitive issues by doubly scaling the input grid unless they take great care when editing namelist.init_atmosphere.
This PR protects against such errors. Now, running the init_atmosphere model to generate real-data initial conditions and attempting the static interpolation step with a non-unit sphere input grid will cause the model to abort with an error message.