SMSTAV not reproducible across MPI task counts

[mgduda]

The SMSTAV field produced by the Noah scheme is not reproducible across MPI task counts for grid cells with snow/ice land use index. Below is a plot showing the SMSTAV field for an arctic domain run with 36 MPI tasks (left) and 144 MPI tasks (right).

[mgduda]

The issue seems to be that when we set SMSTAV(I,J)=SOILW at L.1471 in module_sf_noahdrv.F, the value of SOILW may come from a previous iteration of the I-loop at L.731 when the grid cell is permanent snow/ice. One potential solution may be to move the initializations at L.661-L.665 inside the I-loop.

[mgduda]

@barlage For some reason, GitHub won't let me add you as an Assignee for this Issue.

[dudhia]

I see that SMSTAV is set over sea ice for itimestep=1. However if sea ice changes during the run, this won't work. Also if sea ice is treated as permanent ice(?) it may not be initialized. Could that be what is happening. However, this is a diagnostic this problem shouldn't change other results.

On Thu, Apr 4, 2019 at 5:24 PM Michael Duda notifications@github.com wrote:

@barlage https://github.com/barlage For some reason, GitHub won't let me add you as an Assignee for this Issue.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/wrf-model/WRF/issues/868#issuecomment-480097752, or mute the thread https://github.com/notifications/unsubscribe-auth/ARGf_AtzhiNwiu4CHVyVlSPjK_oYqoAlks5vdompgaJpZM4cd4a1 .

[mgduda]

The irreproducibility in my example is over Greenland, so I don't think this is an issue of how sea ice is treated.

As a side note, ncview appears to be inverting the y-axis, which leads to a confusing view of the Arctic.

[dudhia]

It seems to be that SOILW is not set for glacial ice (ICE=-1) which may apply to Greenland. Setting it at the top as mgduda suggested would work, or set it in the ICE=-1 block.

[dudhia]

Having said that SOILW is set at the top, so I don't see how SMSTAV does not get its value.

[mgduda]

I'm dealing with this issue in MPAS-Atmosphere as well. What are the chances we could coordinate a fix in the WRF v4.3.1 release and in the MPAS-Atmosphere v7.2 release?

[weiwangncar]

@mgduda We should consider it. So what would the best domain and time of year to reveal this issue in WRF?

[mgduda]

I think any domain that contains permanent snow and ice should be sufficient to demonstrate the problem.

[weiwangncar]

@mgduda What day are you looking at?

[weiwangncar]

@mgduda I see it: 2019-04-04.

[mgduda]

@weiwangncar I think any initial time should work, though. I ran into the same problem today with an MPAS-A simulation initialized 2010-10-23.

[weiwangncar]

@mgduda I've created a PR to fix the problem in WRF. Please comment.