GFS radiation: decorrelation cloud overlap scheme #24
Conversation
|
Hi @brianhenn . This is an interesting change. What is the motivation and goal here? I don't know much about cloud overlap schemes. |
|
@lharris4 sorry I can provide a little more context. In the short term I will likely port these changes to an internal branch of SHiELD for a simulation that @brianhenn is interested in for a project. He has been thinking about this aspect of the radiation scheme for a while, so is more familiar than me, and this draft implementation will serve as a good point of reference. In the big picture though, if you are interested in potentially including these options in the main branch of SHiELD, @brianhenn can probably say more about what they offer. I believe these are options that exist in FV3GFS, but not SHiELD. |
|
Hi @spencerkclark thanks for your explanation. I think @linjiongzhou would have more to say about this than I would. |
|
I previously knew the maximum, random, and maximum/random overlapping methods. The decorrelation-length overlapping is new to me. After researching, I came to realize that this is a more advanced method than the previous ones. It is a linear combination of maximum/random overlap. There are various forms of decorrelation length. The one that is used here is a latitude-dependent form. Some researches show the cloud overlapping method could significantly impact TOA and surface radiation. It would be helpful to test this option and understand its impacts. |
|
Hi, Linjiong. There are apparently a few other new more advanced overlap
schemes out there. Here's another new one that was just published using
sophisticated statistical methods. (Looks interesting but unfortunately I
don't have the time to read it too closely :-(
https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2022MS003592
…On Mon, Jun 19, 2023 at 10:47 PM linjiongzhou ***@***.***> wrote:
I previously knew the maximum, random, and maximum/random overlapping
methods. The decorrelation-length overlapping is new to me. After
researching, I came to realize that this is a more advanced method than the
previous ones. It is a linear combination of maximum/random overlap. There
are various forms of decorrelation length. The one that is used here is a
latitude-dependent form. Some researches show the cloud overlapping method
could significantly impact TOA and surface radiation. It would be helpful
to test this option and understand its impacts.
—
Reply to this email directly, view it on GitHub
<#24 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AMUQRVCCKY3K47WA2ESYF3DXMEFNLANCNFSM6AAAAAAZGZ7RAA>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
|
Hi @lharris4 @linjiongzhou thanks for your interest. As Spencer said this is draft implementation, but the main motivation is to make this option available so that radiation outputs can be appropriately compared against FV3GFS runs that use it as implemented there. We found that when FV3GFS is run at coarse resolution, and there are many cells with fractional cloud coverage, radiative fluxes are quite sensitive to the overlap choices, and we found that we get smaller coarse-model radiation biases when running with decorrelation overlap than with maximum-random or random overlap. |
|
Hi @brianhenn, the cloud overlapping should have an impact on the radiation. It is good to know the impact is positive. What radiations bias are you referring to? Thanks! |
|
We ran an experiment where we took 3D cloud fields from C3072 X-SHiELD, coarse-grained them to C48, then prescribed those values as the cloud fields in C48 FV3GFS. The resulting radiation was compared against coarse-grained C3072 radiation, and this is where the decorrelation-length overlap option performed well as there are extensive fractional cloud fields due to coarse-graining. This is not a typical use-case, but explores how well clouds are represented on coarse model grids. We hope to have it submitted for publication soon. |
|
Hi, Brian. By any chance was the information from the native X-SHiELD grid
used to re-calculate the cloud fraction in the coarse-resolution grid cell
during the coarse graining?
Lucas
…On Fri, Jun 23, 2023 at 3:40 PM Brian Henn ***@***.***> wrote:
We ran an experiment where we took 3D cloud fields from C3072 X-SHiELD,
coarse-grained them to C48, then prescribed those values as the cloud
fields in C48 FV3GFS. The resulting radiation was compared against
coarse-grained C3072 radiation, and this is where the decorrelation-length
overlap option performed well as there are extensive fractional cloud
fields due to coarse-graining. This is not a typical use-case, but explores
how well clouds are represented on coarse model grids. We hope to have it
submitted for publication soon.
—
Reply to this email directly, view it on GitHub
<#24 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AMUQRVF77OUMC2AEFFHSG63XMXWLTANCNFSM6AAAAAAZGZ7RAA>
.
You are receiving this because you were mentioned.Message ID:
***@***.***>
|
|
Hi @brianhenn, I am not familiar with your project. Same question as Lucas. Cloud overlapping can also be computed offline. You can extract the subroutine from the radiation package and compute the cloud overlapping offline using the native X-SHiELD data and feed it into the radiation after coarse-graining. |
|
Hi, all. If this isn't intended to be merged we should consider marking this closed. Are there any objections? |
|
Thanks, yes, I'll let @brianhenn have the final word, but I think it is likely fine to close this at this point. The final version of the code that I used in the simulation for @brianhenn's paper can be found in a branch I have on GitLab (authentication required). If we ever wanted to include the decorrelation length cloud overlap scheme in |
This is a toy PR that will not be merged, but gives an idea of the code changes. Note that I implemented decorrelation length overlap via
progcld4, but it could be implemented via anyprogcld. I don't have a build system so there are probably bugs and omissions.