Need for extra radiation layers?

[dustinswales]

Description

In the GFS, RRTMG radiation uses a different vertical grid than the rest of the physics. This vertical dimension has the CCPP standard name adjusted_vertical_layer_dimension_for_radiation. This requires new variables for the radiation, and all of these fields need to be copied in/out before/after radiation. Lot's of overhead.

As far as I can tell, adjusted_vertical_layer_dimension_for_radiation = vertical_layer_dimension + LTP, where LTP=0.

Solution

If there is no need for LTP, use vertical_layer_dimension for RRTMG interstitials.

For RRTMGP, there is a more elegant way of computing the boundary conditions for low top model configurations.

@climbfuji

[climbfuji]

Thanks @dustinswales. I am tagging @matusmartini because I think he experimented with LTP in NEPTUNE in the past. I am somewhat positive that we don't need it, though. Especially since the code probably crashes at the moment if LTP > 0.

[matusmartini]

Thanks @dustinswales. I am tagging @matusmartini because I think he experimented with LTP in NEPTUNE in the past. I am somewhat positive that we don't need it, though. Especially since the code probably crashes at the moment if LTP > 0.

Our current default with NEPTUNE is still to run with LTP = 1. The reason we needed an extra layer in the past was not because of low model top but because of a large cooling bias from longwave at the top resulting from a thinner top-most layer due to spectral element discretization. However, we may not need the extra top layer with the new vertical discretization anymore. This would have to be tested.

[climbfuji]

Thanks @dustinswales. I am tagging @matusmartini because I think he experimented with LTP in NEPTUNE in the past. I am somewhat positive that we don't need it, though. Especially since the code probably crashes at the moment if LTP > 0.

Our current default with NEPTUNE is still to run with LTP = 1. The reason we needed an extra layer in the past was not because of low model top but because of a large cooling bias from longwave at the top resulting from a thinner top-most layer due to spectral element discretization. However, we may not need the extra top layer with the new vertical discretization anymore. This would have to be tested.

Thanks @matusmartini. We'll need to look into this on the NEPTUNE side.

[climbfuji]

We are tracking this in a separate issue for NEPTUNE and will report back.

[dustinswales]

@grantfirl It looks like RRTMG coupled to MPAS-A internally adds extra-layers to the radiation, so we will probably need to implement this as an option in RRTMG/RRTMGP in the CCPP.

In WRF/MPAS, they do this internally at the beginning of the radiation scheme, just before computing the cloud-optical properties.
In the CCPP, this is done externally and requires new separate variables passed from the interstitial schemes to the radiation scheme.

For MPAS, I propose that we make this an optional internal feature in RRTMG and RRTMGP (i.e., like WRF/MPAS), instead of the existing LTP method which relies on additional interstitials, and probably doesn't work anymore...

In RRTMG, we would need to compute the extra layers here
In RRTMGP, we wound need to compute the extra layers here.
In both cases, the radiation outputs (flux profiles from RRTMGP and heating rates from RRTMG) will need to be mapped back at the end of the scheme.

We don't need to do this as part of our initial implementation of MPAS in the UFS/CCPP, but we will want to explore the forecast sensitivity w/ and w/o these extra layers at some point.