ICON Meeting 2024/1¶
21 March 2024, ETH Zurich (L 17.1) and via Zoom
Participants (on-site)¶
Michael Jähn (MJ), Matthieu Leclair (ML), Annika Lauber (AL), Jonas Jucker (JJ), Fulden Batibeniz (FB), Emmanuele Russo (ER), Athena Nghiem (AN), Andrea Stenke (AS), Sylvaine Ferrachat (SF)
Participants (Zoom)¶
Marco Arpagaus (MA), Will Sawyer (WS), Alina Yapparova (AY), Arash Hamzehloo (AH), Mikael Stellio (MS), David Leutwyler (DL), Dominik Brunner (DB), Corina Keller (CK), Michael Steiner (MicS), Stefan Rüdisühli (SR), Christian Steger (CS), Kutay Dönmez (KD)
Minutes by Annika Lauber
Reports¶
C2SM (Michael Jähn, Annika Lauber, Matthieu Leclair, Jonas Jucker)¶
MJ welcomes everyone and shares the latest news about ICON from C2SM .
ER asks about the bug related to volcanic transient Kinne aerosols. AL confirms that the bug also affects the CPU.
Sylvaine Ferrachat¶
SF provides an update on the progress of PhD student Andrina Caratsch, who will study tropical cyclones using ICON-HAM at high resolutions (10 km and below). They have set up the experiment successfully and ran a global 10 km simulation for a few time steps. Currently, they are using ICON version 2.6.4 but aim to transition to the open-source release.
Andrea Stenke¶
AS uses ICON-ART on Euler and now has an account at CSCS, which provides access to the data pool. With MJ's help, she has integrated her script into the Processing Chain tool on Euler. AS discovered that the latest version of ICON (2.6.6) no longer supports AMIP runs.
Athena Nghiem¶
AN is integrating arsenic into ICON-ART and will implement emissions.
Emmanuele Russo¶
ER is developing an aqua planet with a dry atmosphere. Regarding ICON-CLM, they are refining the setup for the EURO-CORDEX, which should be finalized by the end of April. They experienced problems with shortwave radiation but were able to reduce this bias by adjusting cloud cover. These adjustments require code changes in ICON, which will soon be merged into the master branch. ER anticipates that these changes may necessitate additional GPU ports.
DB asks about the project and the meaning of CLM. ER explains that CLM refers to Climate Mode, which is distinct from Climate Land Model despite the shared abbreviation.
DB asks if C2SM/EXCLAIM plans to use the same setup. MJ confirms that EXCLAIM will use a similar setup but will run it on GPU.
DB asks about the importance of transient aerosols. According to MJ, the transient aerosol dataset is crucial because the climatological aerosols showed biased and outdated, and therefore not representative of current conditions.
Fulden Batibeniz¶
FB plans to run the EURO-CORDEX with seasonal forecasts and is waiting for the final version of ICON, which includes transient aerosols.
DB inquires about FB's group at the University of Bern (Climate and Environmental Physics) and their plans for transitioning to ICON. FB confirms that she is responsible for transitioning ICON to the group.
Will Sawyer¶
WS announces the arrival of the new Alps system and the setup of a virtual cluster as a sandbox. However, technical difficulties such as frequent system downtimes are being faced, which are expected to be resolved once everyone gains access. The pre-Alps system is currently being used to test the ICON application.
WS presents slides indicating that pre-Alps is simulating 85 days per day with a 5 km resolution. Performance between Daint and Santis was compared, with an initial speedup factor of 6, slightly below expectations. However, due to the increased memory capacity of the new GPUs, a fair comparison yields a speedup factor of 9. ICON-DSL demonstrates similar performance, and although there's room for improvement, the functionality is satisfactory.
Running the same simulation on CPUs reveals the significant processing power of CPUs on pre-Alps. One Grace has 72 cores. CPUs are currently being used for ocean simulations.
DB asks about a recipe for configuring nproma efficiently on Grace-Hopper. WS explains that for the DLS version, nproma should be maximized within memory constraints. They will provide a recipe for both GPU and CPU versions, with optimal nproma appearing to be 48.
MA raises a question regarding a reported speedup factor of 1.4. WS clarifies that this speedup was gained in comparison to Balfrin.
MJ inquires about the accessibility of the reported information to others. WS confirms that they will share the slides and requests a few more weeks to provide detailed speedup factor data.
MS asks why advection is twice as slow on Alps compared to Balfrin. WS suggests that this could be due to numerical or memory issues, which require additional time to analyze and report precise numbers.
Christian Steger¶
CS has joined MCH to study the physics of ICON, focusing on the interaction between topography and surface radiation.
Alina Yapparova¶
AY, also recently joined MCH, will work on the GLORY project, specifically with 3D radiation data.
Marco Arpagaus¶
MA provides an update on ICON-22, acknowledging the ongoing challenge of the Alps system experiencing frequent reboots, which has been a drawback. Despite this issue, ICON continues to perform well overall. There have been no significant surprises since its pre-operational phase. In comparison, ICON outperforms COSMO in most aspects, although a few systematic biases persist, notably the 2m temperature. MA suggests strategies to address bias, including improving snow analysis and adjusting the minimum diffusion coefficient for better stabilization during winter. However, implementation of these changes must wait until sufficient statistical data for post-processing is collected. Operational deployment is scheduled for May.
Arash Hamzehloo¶
AH is currently working on porting ART components to the GPU. There is an open merge request to add additional functionality.
Corina Keller¶
CK is preparing to set up and run air quality simulations. Recently, a full chemistry case was incorporated into the Processing Chain.
MJ remarks that he is available to offer assistance with the Processing Chain.
Michael Steiner¶
MicS reports that ICON-ART is operating smoothly with multiple passive tracers.
Kutay Dönmez¶
KD is working on integrating something into ICON to address urban climate issues.
DB asks if this involves CORINE land cover data.
Mikael Stellio¶
MK has integrated infrastructure from EXCLAIM into HAM. MK is currently working on Tsa as it is easier to use GT4PY there.