Surface and Bottom Boundary Layer combination

[ymamoutos]

Greetings,

When I am trying to use LMD_SKPP and LMD_BKPP at the same
time I am experiencing a weird blowup. The model blows up
at the first step and according to log file, the point that the model blows up it's on the mask.
I also use PERFECT_RESTART option and I notice that my
mask at rho points is altering/changing. I have already check
my gridfile's depths for zeros, the initial and obc
files for NaN's and I didn't find any suspicious
(according to this thread viewtopic.php?f=1&t=3737)
I also tried to run with other obc combinations
(similar approach as here viewtopic.php?f=17&t=1755)
but again no luck. I am curious because I had already
used the same input data for other setups (different vertical
mixing schemes (MY25 and GLS) different horizontal and vertical
advection schemes) and my solutions were stable - with no blow ups -
for many years (almost decades). I am attaching the rho points mask
and the log.

Any help will be highly appreciated and thanks in advance.

Giannis

[kate]

That's odd. I had heard that skpp and bkpp together may still have bugs, but changing the land mask? Are you using the WET_DRY option?

[ymamoutos]

Hi Kate and greetings from Greece,

no I am not using WET_DRY option.
It is also weird because in another
setup - only the central and north
part of the grid I am using with
almost 1km resolution - BKPP and
SKPP works without any problem.
Any suggestions?

Again thanks for the answer.

Giannis

[kate]

I would watch it in the debugger. Pick a point where you know the mask changes and see where that happens in the code. Or use many print statements to track it down.

[ymamoutos]

I run it on the debugger (I am using gfortran)
and gave me that:

At line 123 of file lmd_swfrac.f90
Fortran runtime error: Index '0' of dimension 1 of array 'lmd_mu1' below lower bound of 1

That's interesting because, I am using WTYPE_GRID option. I 'll switch it of and see what is going to happen

Giannis

[ymamoutos]

Even more interesting. Turning off WTYPE_GRID
makes the model to run smoothly. Any ideas why
that is happening?

Giannis

[kate]

At line 123 of file lmd_swfrac.f90
Fortran runtime error: Index '0' of dimension 1 of array 'lmd_mu1' below lower bound of 1

When you were using WTYPE_GRID, how were you setting the Jwtype array? You are saying the Jerlov water time varies laterally - you have to set it somehow, whether through ANA_WTYPE or reading it from the grid file.

[ymamoutos]

Hi Kate

I am setting it through the wtype_grid
variable in my grid file. I had already used
the same grid with MY25 and GLS with
SOLAR_SOURCE and WTYPE_GRID options and
the model run without any problem.
I am attaching a figure with my wtype_grid
variable to see it.

[kate]

I think you need to give it a well-behaved value inside the land mask.

[ymamoutos]

Kate you are absolutely right. I change the values of
land - just set 2 and 3 all over the grid
- and the model run without blowing up. But I am still
curious. Why that happened with LMD and not with the
other schemes. Once more, many thanks Kate.

Giannis

[kate]

Kate you are absolutely right. I change the values of
land - just set 2 and 3 all over the grid
- and the model run without blowing up. But I am still
curious. Why that happened with LMD and not with the
other schemes. Once more, many thanks Kate.

Giannis

The call to lmd_swfrac only happens when SOLAR_SOURCE is defined. Were you defining it the other times?

[ymamoutos]

Yes I was defining SOLAR_SOURCE with the other vertical
mixing schemes (MY25 and GLS) and the water
type variable was read from the grid file - the
figure that I previously post.

Giannis

[kate]

I'm sure it would still fail with the bounds-checker turned on. You're overwriting some memory, but it will be different memory and perhaps not so important.

[ymamoutos]

Hi Kate and sorry for the late response

I run it at the debugger using -g -fbounds-check flag
(I am using gfortran 4.8.4) just for a day (480 timesteps)
and it didn't have any problem. The 2D and 3D fields as far
I can see looks ok. Any other suggestion?

Giannis

[kate]

This is with the bad old wtype and you want to know why it doesn't fail? Watch in the debugger to see if it really goes into lmd_swfrac. Or put it out of your mind and focus on your next real problem.