ICON Meeting 2023/4 (19 December 2023)

Venue

Hybrid: ETH Zurich (L 17.1) and via Zoom

Participants (on-site)

Michael Jähn (MJ), Matthieu Leclair (ML), Annika Lauber (AL), Tina Schnadt (TS), Andrea Stenke (AS), Brigitta Goger (BG), Guillaume Bertolli (GB), Stefan Rüdisühli (SR), Jan Zibell (JZ), Mikael Stellio (MS)

Participants (Zoom)

Clarissa Kroll (CK), Emmanuele Russo (ER), Fulden Batibeniz (FB), Doris Folini (DF), Will Sawyer (WS)

Minutes by Annika Lauber

Reports

C2SM (Michael Jähn, Annika Lauber, Matthieu Leclair)

MJ welcomes everyone and shares the latest news about ICON from C2SM.

Aerosol Dataset Tests

ER asks about awareness of the aerosol dataset tests.

MJ is unsure of the location of the tests.

ER asks about the nesting and setup of their domain.

MJ explains that they are using the EURO-CORDEX setup, currently 12 km resolution (aiming for higher). They are using SPICE, which has a default configuration.

ER warns of ongoing problems with the default configuration, despite higher radiation it leads to a too low 2m-temperature. Work is ongoing in the CLM community, with progress expected next month.

MJ acknowledges time constraints and suggests discussing offline.

Progress on Two-Moment Microphysics Scheme on GPU

BG asks since when the explicit scheme is on the master.

AL answers that it was merged a few weeks ago (information on that can be found in the GitHub issue Speed up two-moment microphysics scheme on GPU).

WS asks why the implicit scheme does not run on Daint.

AL replies that this is due to a compiler bug, which will not be fixed before Daint’s decommissioning.

Probtest (Tolerance Tests) Development

JZ asks if the GPU run is wrong, it means that it is physically wrong.

AL answers that if there is a bug in the GPU port it can be physically wrong, but small bugs may not be obviously physically wrong.

Andrea Stenke

AS uses ICON-ART to simulate atmospheric trace element cycles. She is currently involved in model development, setting up a global AMIP-type simulation running on Euler.

AS asks if there are additional input files available on CSCS.

MJ replies that availability depends on the specific requirements.

AS plans to request specific resolution data for Euler.

WS comments that CSCS has data for test cases covering a wide range of scenarios. In addition, data can be transferred from Levante.

Brigitta Goger

BG uses ICON on Daint with a limited area setup at high resolution down to 100 m. The latest version of ICON is stable below 100 m and works down to 65 m.

Guillaume Bertolli

GB is focusing on the radiation component of the ICON model and trying to emulate it using machine learning (ML). A super-simple model could solve the problems of the upper atmosphere. The GPU version with ML is three times faster than the ECRAD version.

TS asks about the frequency of radiation calls, and GB mentions that they occur every 3 minutes.

Stefan Rüdisühli

SR is working on a high resolution setup over the Atlantic. He also reports news from Sebastian Schemm that the aquaplanet simulation paper has been accepted (Link to preprint).

Jan Zibell

JZ analyses ICON simulations and plans aquaplanet ensemble runs focusing on storm tracks, using a 1 degree simulation.

ML asks about the duration and number of ensembles.

JZ clarifies that it is five times in 10 years, and suggests that specific questions about high resolution should be directed to Sebastian Schemm or Anurag Dipankar.

Mikael Stellio

MS is working on ICON-HAM with Sylvaine, focusing on GPU ports. They are currently running at 160 km resolution. They have been successfully running at 40 km resolution, but are having problems trying to reduce it further due to problems with the input data. MS is also doing scaling analysis, but it doesn’t scale as well as ICON-NWP, mainly due to the increased number of tracers in ICON-HAM. He also helped with pollen simulation issues.

MS also reports experimenting with the new version of probtest and managing to reduce the ensemble members to 5 for their test case. This reduction seems to be stable for now.

Will Sawyer

WS reports several delays at Alps. Daint is due to be decommissioned in the summer of next year. In January, a pre-Alps system will be launched on a virtual platform resembling several machines. This pre-Alps system will be dedicated to EXCLAIM. They will start compiling the DSL branch in January and try to get benchmarks. Full implementation of Alps is expected around April/May and once operational, Daint will be rapidly decommissioned as it is not feasible to run two systems simultaneously.

Doris Folini

DF, who hasn’t used ICON recently, provides an update on behalf of Boriana, who is currently using an older version of ICON. They express the hope of receiving revisions for Boriana’s paper before Daint is shut down.

Emmanuele Russo

ES uses the same version as DF with ECHAM physics. Significant developments have taken place, including the introduction of a version of ICON known as ICON-DRY. This version provides the ability to add weights within the model, allowing the adjustment of Sea Surface Temperature (SST) values.

Fulden Batibeniz

FB runs ICON for seasonal forecasting studies, with a particular focus on drought. However, she notices that there is no moisture available on the layers. She asks if anyone has relevant data.

WS replies and mentions the possibility that they might have the required data. Although he does not know the specific name for layer moisture, WS suggests checking Levante as it may contain the relevant information.

Clarissa Kroll

CK is working with ICON-DSL and aims to perform kilometre-scale simulations on Alps. The focus is on investigating the atmospheric response to different SST patches in the Pacific. CK is currently running sensitivity tests with ICON.

She comments on FB’s request and reports difficulties in accessing soil moisture data, mentioning that access can be obtained through MCH, but the process takes months. It would be much easier if the data were available on Levante.

WS asks if there have been any recent technical problems. CK replies that the model is currently running. They plan to run tests on the pre-Alps system in April.