MPAS initialization temperature extrapolation and terrain mismatch

[JiliDong-NOAA]

When setting up the global MPAS test, NaN values in vertical velocity (w) were observed in the initial conditions. The issue was traced back to unrealistic temperatures generated during init_atmosphere, which subsequently caused problems in the hydrostatic balance calculation. Several factors appear to contribute to the unrealistic temperatures:

1. Our MPAS model top is set to ~80 km, which is higher than the first guess GFS background at some locations. Linear extrapolation of temperature was used (config_extrap_airtemp = 'linear') because the current 'lapse-rate' does not work properly for upper level extrapolation. However, the same linear extrapolation of temperature is also used for underground extrapolation near the surface . When a low-level temperature inversion is present, downward linear extrapolation can produce extremely cold and unrealistic surface temperatures (-300 °C in our case). Please see here for more details.
2. There is a terrain mismatch between GFS and MPAS, particularly over Antarctica Peninsula (Palmer Peninsula), which causes downward extrapolation over depth exceeding 1 km at some locations. More details are available here. The terrain difference appears from the source topography dataset used in GFS and MPAS. GFS uses a modified GMETD2010 dataset with corrections over the Antarctic Peninsula, while MPAS appears to use the original dataset. Discussions of the GFS terrain fix from @yangfanglin and George Gayno can be found here.

I have two questions:

About air temperature extrapolation:

We are currently using config_extrap_airtemp = 'constant' for temperature extrapolation. Would this be an appropriate choice for both upper-level and underground extrapolation?

I also noticed there is an open PR related to extrapolation options. Would this help address the issue described above? What would be the recommended extrapolation approach for this type of configuration?

About terrain mismatch:

Would it be acceptable to continue using the original GMETD2010 terrain dataset in MPAS without the Antarctic Peninsula corrections? If the terrain fixes are incorporated into MPAS, would they be expected to have any significant impact on forecasts?

[clark-evans]

Thanks for opening this issue, @JiliDong-NOAA! I've added a few assignees for awareness.

I don't see any issues with incorporating the GMETD2010 terrain dataset updates into MPAS. I would not expect significant forecast impact from doing so, although admittedly any such differences would be most likely to occur in regions where we at GSL have less experience in modeling/forecasting.

I had also noticed the open PR for temperature extrapolation. As best as I can tell, the existing extrap_airtemp = 2 code would use a -6.5 K/km lapse-rate when interpolating below ground (like in chgres), but it is not implemented when extrapolating above the top level in the input data. The modified code in that PR changes the below-ground interpolation to use the local lapse rate rather than a fixed -6.5 K/km at the same time that it enables extrapolation above the top level in the input data (where is also uses the local lapse rate). I think it's worth a test, but I'm also wondering if a simpler implementation of the -6.5 K/km formulation when extrapolating above the input data might be worth testing first?

[dustinswales]

@JiliDong-NOAA Thanks for opening this issue.
I'm not terribly familiar with core_init_atmosphere, all of the dycore implementation has been related to core_atmosphere.

In general, questions/issues/feature-requests/etc... are arising more and more frequently regarding core_init_atmosphere. Most of this will ultimately fall under the auspices of "workflow". For this reason, I added this to the MPAS dycore in the UFS Weather Model Project Board. Going forward I will try to be vigilant and link issues back to the project board to keep track.

[mdtoyNOAA]

@clark-evans et al. I'm not familiar with the extrapolation code, but I was wondering about the extrapolation at the model top. Using the -6.5 K/km lapse rate would be appropriate if the model top were in the troposphere, but what about when it's in the stratosphere -- then a positive lapse rate would be more appropriate.

[dustinswales]

FYI. Github incorrectly added this to the MPAS dycore in the UFS Weather Model project board. Removing

[yangfanglin]

When we started developing GFSv16 back in the mid-2010s, I used variable standard-atmosphere lapse rates to construct the GFSv16 vertical coordinate and initial conditions. These can be implemented in core_init_atmosphere, if desired, to apply different lapse rates in different atmospheric layers for interpolation and extrapolation.