Log In | Users | Register
COSMO Web
Edit | Attach | New | Raw | Diff | Print | | Tools
You are here: ICON » ICONmeet20200723

ICON meeting July 23, 2020

Participants:

André Walser (AW); Bernat Jiménez Esteve (BJE); Bernhard Enz (BE); Colombe Siegenthaler (CS); Dominik Brunner (DB); Doris Folini (DF); Jonas Jucker (JJ); Katherine Osterried (KO); Lukas Papritz (LK) ; Michael Steiner (MS); Pavle Arsenovic (PA); Remo Dietlicher (RD); Sascha Bellaire (SB); Sven Kotlarski (SK); William Sawyer (WS); Xavier Lapillonne (XL);

Reports:

* Remo Dietlicher (RD):
  • works with JJ to construct input files for ICON-LAM, over Switzerland
  • busy with testing the GPU ported routines against CPU version using Buildbot.
In general, daily Jenkins plans are performed for Meteoswiss-specific questions, for example a performance test is run on a daily basis by Jenkins. However, if you intend to merge back code to the official ICON repository (at DKRZ), you need to pass the Buildbot tests. So for code development, Buildbot is used.

* Colombe Siegenthaler (CS);

  • helped and supervised JJ in his ICON work.
  • set-up ICON-ART with grib2 support using eccodes library.
  • updated ICON-ART in the C2SM-ICON GitHub organization to a working version using an intermediate ICON version older than 2.6.1

* Xavier Lapillonne (XL);

  • Spack package manager is now used for compiling and installing COSMO and multiple libraries at Meteoswiss.
  • 2 new people will work with ICON at Meteoswiss/C2SM : Mikhail Zhigun will work on CLAW and Fabian Gessler will help Remo in the porting of NWP physics to GPU.
  • There are multiple people working on porting ICON routines on GPU and each is using a different testing strategy(for example WS uses a tool called fortran test generator while at Meteoswiss, XL uses serialbox). It would be good to set up a meeting for exchanging and coordinating between people interested in porting ICON components to GPU. CS will send a list to XL of people who could be interested and XL will set-up a meeting for those interested.

* Bernat Jiménez Esteve (BJE):

  • is working with ICON using the ECHAM physics, studying what ingredients are triggering the heat waves.
  • ran already the dycore-only and the amip case.
  • will run the aquaplanet and an idealized topography.

* Doris Folini (DF);

  • played with ICON on Piz Daint to understand the restart mechanism, different namelists, input, output, job chaining, etc...

* Jonas Jucker (JJ);

  • deputee for Colombe to set up ICON for C2SM users.
  • set up automated Jenkins tests for different compilers and configurations.
  • set up ICON on Euler. JJ has now a running version on Euler which is more or less stable.
  • performed scaling analysis to be able to compare performances on Daint/Euler and give a configuration recommendation for C2SM users. The results of the scaling analysis can be seen here for Piz Daint and Euler.

* Michael Steiner (MS);

  • wants to couple the online emission module (OEM) written for COSMO to ICON, and then port it to GPU.
  • played with ICON-ART
  • performed a scaling analysis of ICON-ART over Europe.
  • set-up a test case for ICON-ART.
  • More info can be found in the slides MS prepared.

* Sascha Bellaire (SB):

  • works in the ICON 2022 project, which is about getting ICON-LAM running operationally at Meteoswiss.
  • is now trying to get ICON running at ECMWF. The plan is to finish by the end of August. The reason for running at ECMWF is that there is no GPU ICON-NWP version running, and the CPU resources on our systems are quite low since the CSCS machines are highly GPU.
  • when ICON is running at ECMWF, SB wants to test different domains.

* Bernhard Enz (BE);

  • runs ICON-LAM over the North Atlantic. BE resolved his problem of unstable/messy PV by moving the north domain boundary further north.
  • compared hurricanes with 6 hours output (ERA5 data injected at domain boundaries) with observations for 2013, which was a year with very low activity. The strength of the tropical cyclones in the simulation is much too high compared to the observation. However the cyclone energy (sum of the square of cyclone energy over the season) in ICON is in the same range as observed. This might be because of the tracking algorithms not detecting the weaker tropical cyclones.
  • will do the same comparison for 2017 which was a very active year in terms of tropical cyclones over North Atlantic.

* Sven Kotlarski (SK):

  • not using ICON for the moment.
  • SK is interested in hearing what is happening in ICON community because in the Climate Evolution group at Meteoswiss, ICON-LAM might replace COSMO-CLM at some points.

* Lukas Papritz (LK):

  • just began working with ICON-ART on Euler.
  • wants to implement the tracer tracker Heini's group is using in COSMO.
  • will add a 2nd water tracer to trace the water from different sources.

* Katherine Osterried (KO):

  • worked with RD to port the NWP soil model, TERRA, to GPU in ICON. The soil model is now merged back in the NWP version of ICON.
  • KO is the source code administrator of Extpar. Concerning Extpar:
    • The latest release of Extpar includes a new option which handles the cache memory differently, allowing the processing of very high spatial resolution ICON grids.
    • DB had heard about proposals from different groups in Germany for building an online tool to map external parameters. This would bypass the use of Extpar. KO never heard of anything in this direction. KO states that historically, and also nowadays, DWD and MPI are quite engaged in developing Extpar. There are tons of new options. SK does not think they will let Extpar on side for the moment.
    • There is a project at MPI to add the JSBACH parameters to Extpar.

* André Walser (AW):

  • works with SB in the ICON 2022 project
  • did not have time to work himself with ICON, because there are too many issues with COSMO.

* Dominik Brunner (DB):

  • is interested in running ICON-ART as a transport model of passive tracer, for example:
    • for simulations of greenhouse gazes
    • to quantify emissions based on observation and inverse modeling
  • in the future DB wants to conduct air pollution simulations using ICON-ART.
  • DB is interested and will take part in the porting of ART on GPU. DB suggests to coordinate with the effort in HAM.
  • concerning MS's project, since porting ART to GPU will be a large effort, the workplan is to
    1. implement the online emissions module in ICON
    2. port the online emissions module to GPU
    3. couple the GPU ported online emissions module to ART.

* Pavle Arsenovic (PA):

  • never used ICON so far
  • submitted a SNF project for simulating ozone over Arosa and Davos using ICON-ART.

* William Sawyer (WS) :

  • PRACE QUBICC project: CSCS completed all features requested by Marco Giorgetta. Now on branch mc10_rrtmgp_remerge. Now requires PSrad to be reintegrated in order to allow incorporation into icon-2.6.x-rc. This merge back should happen this autumn.
  • CSCS, EMPA and KIT are trying to launch an effort to (ultimately) port ART to GPUS, e.g., at the CSCS GPU programming hackathon (September): port Online Emissions Module as a prototype example
  • Longer term: port individual ART components similar to the OEM (Joint effort with KIT)
  • DKRZ (Claudia Frauen, Jan-Frederick Engels, Panos Adamidis) are hoping to collect GPU programming experiences for ICON-O at the CSCS GPU programming hackathon.

Topics of common interest:

* Standard deviation of 10 years annual global means of AMIP run (exp.atm_amip settings, monthly output) has been found by JJ and CS to be very big (for example 1° for the 2m meters temperature) compared similar simulations of ECHAM-HAM. This will be investigated further by CS and JJ. The following suggestions came out:
  • maybe CDO is not doing the spatial averaging properly (maybe the spatial averaging did not use the area weights properly). cdo fldmean was used to process the data from the native ICON grid. Maybe restrict the domain to 10°S-10°N, this is the portion of Earth where the area weighting should not have a big influence.
  • try to use the ICON internal spatial remapping? BJE reports that the spatial remapping is not working so well in ICON.
  • maybe the time averaging is not done properly in ICON. To test that, we could take a daily output and average it manually. BJE has a amip simulation with daily output available.
  • ask Marco Giorgetta. At MPI they have done lots of tests.
  • JJ 29.7: The great variance of the parameters compared was caused by a "dangling" output-file only containing instantaneous data for the 1.1.2013 (end of simulation). Nervertheless differences are still substantial between Icon and Echam.
  • JJ 29.7: Welch's-Test Icon vs Echam
  • JJ 29.7: According to the Welch's-Test, Icon versions on Euler and Daint give the same results.

* BJE reports that he compared ICON output against observations (Era-interim) and ICON was doing very well. BJE compared circulation variables, for example wind and temperature in the upper and lower troposphere. The interannual standard deviation looked fine.

* Radiative transfer routines in ICON:

  • for the Echam physics, there will be two radiation schemes available: psrad and rrtmgp . Only rrtmgp is ported to GPU, the porting of psrad was unsuccessful. There are small optimization issues with rrtmgp on GPU, the responsible person is Sebastian Rast at MPI.
  • for the NWP physics, ecrad is coupled to ICON, as well as the radiative scheme used at DWD. ecrad is intended to be ported on GPU. It is not clear how.

* Infrastructure meeting :

  • last one in April 2020: nothing significant to report. CDI and parallel I/O are getting more and more attention. The new build system is quite successful.
  • WS will attend the next infrastructure meeting. If you have anything to be reported at the infrastructure meeting, send an email to WS.

* Colombe will leave C2SM at the end of September.

  • CS's successor should continue to organize the C2SM ICON meetings.
  • It is not clear when CS's successor will begin. The position will be advertised by the end of summer together with 2 others positions at C2SM.
  • C2SM will be restructured anyway. The RCM and GCM groups will be dissolved.
Attach
pdf
version 1 uploaded by JonasJucker on 24 Jul 2020 - 09:51
pdf
version 2 uploaded by JonasJucker on 29 Jul 2020 - 09:56 ...
version 1 uploaded by JonasJucker on 24 Jul 2020 - 09:53
pdf
version 1 uploaded by ColombeSiegenthaler on 24 Jul 2020 - 08:47
pdf
version 1 uploaded by JonasJucker on 29 Jul 2020 - 10:07
pdf
version 1 uploaded by JonasJucker on 29 Jul 2020 - 10:03
spacer
This site is managed by the Center for Climate Systems Modeling (C2SM).
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors. Ideas, requests, problems? Send feedback!