New Thompson cloud fraction (updated subroutine cal_cldfra3) #781
Conversation
…igned with Thompson-MP for GFSv17-prototype8
|
Below are some graphics showing some quantitative differences 20190719 case using experiment 2 minus experiment 3 for outgoing longwave and shortwave radiation and total-sky albedo. The graphics clearly show the rather large increase in reflective clouds (thus less shortwave reaching the ground) and far less overall longwave IR (meaning colder clouds), particularly in the tropics Diff of outgoing longwave... Diff of outgoing shortwave...
Diff of albedo...
|
|
Below are some graphics showing some quantitative differences 20200105 case using experiment 2 minus experiment 3 for outgoing longwave and shortwave radiation and total-sky albedo. The graphics clearly show the rather large increase in reflective clouds (thus less shortwave reaching the ground) and far less overall longwave IR (meaning colder clouds), particularly in the tropics Diff of outgoing longwave...
Diff of outgoing shortwave...
Diff of albedo (total-sky)...
|
There was a problem hiding this comment.
There is some confusion around which vertical dimension to use, and I tried - within the limited abilities of GitHub comments and the limited time - to explain that. Most radiation quantities are allocated to length lmk, and if instead lm gets used then several model configurations in the UFS and other models (NEPTUNE) will stop functioning.
There is one example (cldfra) where I need to do a few more checks on what it should really be. Metadata says lmk, but the allocation in GFS_typedefs.F90 is lm.
For now, you don't need to make any changes, I will send you changes in form of a PR early tomorrow morning.
…te_from_main_20211119 Update gthompsnWRF:new_thompson_cloud_fraction with the latest code from NCAR ccpp-physics main
…mpsnWRF/ccpp-physics into new_thompson_cloud_fraction
|
There is a desire to unify the 3D instantaneous cloud fraction diagnostic. Currently, cldfra is being output to the raw history files, and a decision has been made to change this to cld_amt. Should this change be made as part of this PR or should it be a separate PR? |
Would cld_amt in this case be treated as in GFDL-MP, i.e. as a tracer in the tracer array that is not advected/diffused, or would it be treated differently? |
…S_rrtmg_post.meta and physics/GFS_rrtmg_pre.meta
…mpsnWRF/ccpp-physics into new_thompson_cloud_fraction_update_from_main_20211119
|
No, it would be a separate diagnostic output in the physics. The inline
and offline UPP for global applications would be modified such that
cld_amt is read in from the atmf* for GFDL mp and read in from the sfc*
file for all other mp schemes.
…On 11/22/2021 3:30 PM, Dom Heinzeller wrote:
There is a desire to unify the 3D instantaneous cloud fraction
diagnostic. Currently, cldfra is being output to the raw history
files, and a decision has been made to change this to cld_amt.
Should this change be made as part of this PR or should it be a
separate PR?
Would cld_amt in this case be treated as in GFDL-MP, i.e. as a tracer
in the tracer array that is not advected/diffused, or would it be
treated differently?
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#781 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ALQ75MNMBUJGJQY5INYJKCLUNKR43ANCNFSM5IFOTHXQ>.
Triage notifications on the go with GitHub Mobile for iOS
<https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675>
or Android
<https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>.
|
|
I am ok with reusing icloud = 3 for the new Thompson cloud scheme, but note the suggested changes in PR #780 - these remove icloud = 3 completely, in places that you touched ( |
|
Looks good to me.
…On Tue, Nov 23, 2021 at 11:12 AM Dom Heinzeller ***@***.***> wrote:
I am ok with reusing icloud = 3 for the new Thompson cloud scheme, but
note the suggested changes in PR #780
<#780> - these remove icloud = 3
completely, in places that you touched (physics/GFS_rrtmg_pre.F90) but
also in others (physics/radiation_clouds.f). You may want to check if the
suggested changes in PR #780
<#780> align with your changes.
I think they fit nicely, but we need to double check. Adding @mzhangw
<https://github.com/mzhangw> and @mjiacono <https://github.com/mjiacono>
for their awareness.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#781 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AG7TW2THBICOE4CHPNVBFTTUNPKSJANCNFSM5IFOTHXQ>
.
|
|
|
||
| endif ! icloud == 3 | ||
|
|
||
| if (lextop) then |
There was a problem hiding this comment.
Since both of us removed the lines of code above here related to icloud=3, then I think you should remove the next 2 chunks that are handling only the lyb layer as equal to lya layer. I no longer think those lines are relevant since there is nothing inside the cldcov array at this point. That is why I also removed it.
If I am incorrect, then my PR needs changing to put these back. Does something prior to this subroutine already pre-popluate those arrays?
|
@gthompsnWRF we didn't add the logic to use the six new diagnostic variables only if a corresponding debug switch is enabled. I think we should do that to reduce the memory footprint. If those variables are only of interest if Thompson MP and the new cloud fraction scheme is used, we could wrap all code that writes to or reads from those files in |
I agree that a debug switch could help in this regard, but I added those variables for usage by any cloud fraction or microphysics scheme in mind, so I believe they are useful regardless. The total_albedo is sorely needed as a simple way to visualize cloud reflectance (during daytime) as compared to fluxes. The explicit versus non-explicit water and ice paths for radiation are useful for anyone and 2-D arrays don't add too much memory as compared to many other 2D arrays being output. Your decision ultimately, but seems trivial to add 6 2D arrays (x,y) for a model with hundreds? of other 2D arrays (IMHO). |
The problem is that every other PR does the same and suddenly you have 5 times 6 new 2D arrays etc. I hate introducing another debug flag, though, and given that these arrays could be useful for other schemes, too, let's leave it as is. |
…actional cloudy LWP and IWP inside progcld6
…_update_from_main_20211129 greg/new_thompson_cloud_fraction update from main 2021/11/29
…educe assumed snow as ice in IWP for radiation
|
approved. I have tested the new Thompson cloud fraction many times. |
…-MP and icloud3 (cloud fraction) scheme (#1626) TYPE: enhancement KEYWORDS: microphysics, clouds, cloud fraction, Thompson microphysics SOURCE: Greg Thompson (UCAR/Joint Center for Satellite Data Assimilation) DESCRIPTION OF CHANGES: The purpose of these changes is to produce better cloud coverage and radiation amounts per tuning done with GFS global model and comparisons to observations. Problem: The Thompson-MP code (from ccpp-physics) has been used within UFS/GFS/FV-3 global model for multiple day simulations. Anning Cheng and Ruiyu Sun (NCEP/EMC) have performed most of the simulations and analysis and iterated with me to make adjustments to the Thompson-MP. At first, EMC found that high clouds were not present enough, particularly in the tropics. Once switching to a larger upper limit of cloud ice before it becomes snow, the cloud amounts improved but some clouds and radiation balances were still not matching results from FV3's GFDL microphysics (which had already been tuned to observations). Therefore an effort to account for subgrid-scale clouds using the cloud fraction scheme (icloud=3) was adopted to GFS and tested with various parameter settings. Solution: The biggest change is in how cloud ice converts to snow at a threshold size of 300 microns (up from 200 microns) and for rimed snow to convert to graupel (changed from 250 to 350 microns). A change to allowed max size of ice means that the fall velocity constant for ice was changed to keep it aligned with snow at the same cut-over size (of 300 microns). In addition, the cloud fraction scheme (icloud=3) can further improve the clouds and radiation together with Thompson-MP due to the overall under-prediction of clouds. The icloud3 option was also making far too many clouds when compared to observations so its tuning knobs were adjusted until attaining the following overall improvements compared to cloud and radiation global climatologies that EMC uses: cloud amounts of low, middle, high, and total cloud coverage, longwave radiation outgoing at top-of-atmosphere, and shortwave radiation reaching the ground. ISSUE: There are corresponding issues or pull requests in the ccpp-physics repo, [778](NCAR/ccpp-physics#778) [781](NCAR/ccpp-physics#781) [809](NCAR/ccpp-physics#809) LIST OF MODIFIED FILES: M module_mp_thompson.F M module_radiation_driver.F M module_ra_rrtmg_lw.F M module_ra_rrtmg_lwf.F M module_ra_rrtmg_sw.F M module_ra_rrtmg_swf.F TESTS CONDUCTED: 1. A series of tests in GFS including 5, 7, 16, and 30-day long simulations compared to known cloud and radiation climatologies. The plots attached here in this comment were created by Anning Cheng. The numbers at the top of each panel represent global 3-day (days 3, 4, 5) average. The first plot is the high/mid/low/total cloud amount. The tunings in this PR reduced the high cloud amount from over 55% down to the low 40s, which matches observations of global cloud coverage pretty well. And, another comparison is without the cloud fraction scheme, the mid-level clouds (and low/high) are simply less than observations, which is something well published in IPCC reports of global model clouds, especially the Southern Ocean and east sides of ocean basins (low stratus).  The next plot's panel (a) shows the outgoing longwave radiation. The target value global average is about 240 W/m2. Without the cloud fraction scheme and changes to ```D0s``` and ```D0g```, the result would be too much outgoing longwave since it will not contain enough clouds at lower temperatures. Also, the older version of the cloud fraction scheme with its excessive amount of high clouds would make the outgoing radiation closer to 225 W/m2. So the tuning of the scheme brings the results closer to observations.  Lastly, the final plot shows the downward longwave reaching the ground (panel c) and downward shortwave reaching the ground (panel d).  2. Jenkins tests are passing. RELEASE NOTE: Update of the Thompson microphysics scheme and cloud fraction scheme (icloud=3) to match the observations better. The modifications include updates to RRTMG LW and SW, and RRTMG fast LW and SW.
…-MP and icloud3 (cloud fraction) scheme (wrf-model#1626) TYPE: enhancement KEYWORDS: microphysics, clouds, cloud fraction, Thompson microphysics SOURCE: Greg Thompson (UCAR/Joint Center for Satellite Data Assimilation) DESCRIPTION OF CHANGES: The purpose of these changes is to produce better cloud coverage and radiation amounts per tuning done with GFS global model and comparisons to observations. Problem: The Thompson-MP code (from ccpp-physics) has been used within UFS/GFS/FV-3 global model for multiple day simulations. Anning Cheng and Ruiyu Sun (NCEP/EMC) have performed most of the simulations and analysis and iterated with me to make adjustments to the Thompson-MP. At first, EMC found that high clouds were not present enough, particularly in the tropics. Once switching to a larger upper limit of cloud ice before it becomes snow, the cloud amounts improved but some clouds and radiation balances were still not matching results from FV3's GFDL microphysics (which had already been tuned to observations). Therefore an effort to account for subgrid-scale clouds using the cloud fraction scheme (icloud=3) was adopted to GFS and tested with various parameter settings. Solution: The biggest change is in how cloud ice converts to snow at a threshold size of 300 microns (up from 200 microns) and for rimed snow to convert to graupel (changed from 250 to 350 microns). A change to allowed max size of ice means that the fall velocity constant for ice was changed to keep it aligned with snow at the same cut-over size (of 300 microns). In addition, the cloud fraction scheme (icloud=3) can further improve the clouds and radiation together with Thompson-MP due to the overall under-prediction of clouds. The icloud3 option was also making far too many clouds when compared to observations so its tuning knobs were adjusted until attaining the following overall improvements compared to cloud and radiation global climatologies that EMC uses: cloud amounts of low, middle, high, and total cloud coverage, longwave radiation outgoing at top-of-atmosphere, and shortwave radiation reaching the ground. ISSUE: There are corresponding issues or pull requests in the ccpp-physics repo, [778](NCAR/ccpp-physics#778) [781](NCAR/ccpp-physics#781) [809](NCAR/ccpp-physics#809) LIST OF MODIFIED FILES: M module_mp_thompson.F M module_radiation_driver.F M module_ra_rrtmg_lw.F M module_ra_rrtmg_lwf.F M module_ra_rrtmg_sw.F M module_ra_rrtmg_swf.F TESTS CONDUCTED: 1. A series of tests in GFS including 5, 7, 16, and 30-day long simulations compared to known cloud and radiation climatologies. The plots attached here in this comment were created by Anning Cheng. The numbers at the top of each panel represent global 3-day (days 3, 4, 5) average. The first plot is the high/mid/low/total cloud amount. The tunings in this PR reduced the high cloud amount from over 55% down to the low 40s, which matches observations of global cloud coverage pretty well. And, another comparison is without the cloud fraction scheme, the mid-level clouds (and low/high) are simply less than observations, which is something well published in IPCC reports of global model clouds, especially the Southern Ocean and east sides of ocean basins (low stratus).  The next plot's panel (a) shows the outgoing longwave radiation. The target value global average is about 240 W/m2. Without the cloud fraction scheme and changes to ```D0s``` and ```D0g```, the result would be too much outgoing longwave since it will not contain enough clouds at lower temperatures. Also, the older version of the cloud fraction scheme with its excessive amount of high clouds would make the outgoing radiation closer to 225 W/m2. So the tuning of the scheme brings the results closer to observations.  Lastly, the final plot shows the downward longwave reaching the ground (panel c) and downward shortwave reaching the ground (panel d).  2. Jenkins tests are passing. RELEASE NOTE: Update of the Thompson microphysics scheme and cloud fraction scheme (icloud=3) to match the observations better. The modifications include updates to RRTMG LW and SW, and RRTMG fast LW and SW.
This pull request incorporates a Thompson cloud fraction scheme (which is a heavily updated cal_cldfra3 introduced in HWRF in 2015) to improve representation of fractional cloud amounts in areas without explicit (microphysics grid-scale) clouds. Not only is cloud fraction improved to contain numerous grid cells greater than zero and less than one, but the RRTMG scheme then "sees" these 'partly cloudy' grid boxes to improve radiation. The details of the need of "boosting" clouds beyond what is captured solely in the microphysics scheme was reported by the [Developmental Testbed Center's HFIP project in 2015].(https://github.com/NCAR/ccpp-physics/files/7557574/HFIP_GT2014Dec17.pdf)
Generally no microphysics scheme is capable of handling sub-saturated grid volumes to produce explicit cloud water and ice which is why the GFDL microphysics scheme has its own "cld_amt" that produces non-zero values where the grid volumes are near yet not at saturation. So when switching to the Thompson microphysics from GFDL, one should assume that some similar treatment is logical and sensible.
Besides the changes to the direct code calculating the cloud fraction, a number of helpful 2D (x,y) diagnostics are created to assist others with the contributions by the new code additions and to utilize them for further developments regardless of cloud fraction or microphysics schemes used.
A set of 3 simulations from 20200105/00z start time is provided in graphics using the following...
subroutine progcld6and Thompson-MP (candidate version for GFSv17-prototype8)subroutine progcld_thompsonwith same Thompson-MP as abovesubroutine progcld6with changes to Thompson-MP following PR #778 using D0s=300, D0g=350, and crt_sati=0.0Another set of 3 simulations from 20190719/18z start time is provided in graphics using these powerpoint files...
subroutine progcld6subroutine progcld_thompsonsubroutine progcld6with changes proposed in #778A set of graphics from each simulation is included (in powerpoint files) to make visual comparisons of the differences - quantitative differences are being prepared. Of particular note is that more outgoing shortwave radiation (cloud reflectance) and generally lower outgoing longwave/IR (corresponding with more high clouds) occurs using either of the last 2 configurations. In other words, both code changes are capable of producing more ice clouds - primarily in the tropics and mid-latitudes. HOWEVER, the sole reason for more ice in simulation-3 is a change to initiate ice at relative humidity with respect to ice at exactly 100% whereas the Thompson microphysics was explicitly designed NOT to produce ice at RH-ice=100% because of the near total glaciation of clouds (due to the Bergeron-Wegener-Findeisen process) with this choice. As such, the impact to cloud phase from having some supercooled liquid water present to having very little is a major step backwards for forecasts of aviation icing hazards aloft and at the surface (freezing rain and freezing drizzle). Numerous papers in the literature show a distinct lack of supercooled liquid water in global weather models (e.g., Ma et al. 2014 and many papers related to the SOCRATES field experiment) in addition to the plethora of papers in the literature that show ice does not initiate at RH-ice=100%, but, rather at much higher supersaturations with respect to ice, often near water saturation (cloud chamber studies in Germany among many others).