Meeting minutes (22/07/2025)
Participants:
@anton
@lachlanswhyborn
@inh599
@MartinDix
@Jhan
@pearseb
@Aidan
@adfraser
@dhb599
@harshula
@jemmajeffree
@manodeep
@paulleopardi
@RachelLaw
@rbeucher
@tiloz
@cbull
@clairecarouge
Chair: @clairecarouge
Minutes: @cbull
payu dev bug
Spencer added things to calendar handling for the UM in esm1.6 late last week in Payu. This unfortunately broke running new experiments with Payu (existing submitted runs should have been okay), but was fixed yesterday and new payu dev should be working again.
water conservation in the um
Ian, Spencer and Martin have been looking into water conservation issue in the UM. Unravelling compensating errors. Could be a substantive bit of work to fix all the bugs. Considering rolling back to an earlier version. Jhan asked for further detail. CICE4 is doing a better job (Martin it’s not actually a CICE issue but the issue was found through the model change). Martin: Atmosphere is creating water, which could be balanced with scaling the river outflow (by a few %). Ian had a look at this and thinks it goes through about 14 layers of code. Martin: Could a module be used? Spencer has been looking into it. Martin is keen to keep looking into how feasible is it.
Lachlan doing some tests with some compiler flags off
Lachlan: ESM1.6 with CABLE included as a library within UM7, rather than as source code, bitwise reproduces the release build when optimisations are turned off i.e. release compiled with -O0 is identical to library build with -O0. Doing a 10 year run with optimisations turned back on with -O2, and comparing results to determine effect of optimisations on behaviour. First few years suggest runtime/year is about the same between builds, perhaps even slightly faster with the library, but most likely just natural variance.
Australian PFTs
Ian: Martin and Jhan are still looking into Australian PFTs.
Radiation frequency
Martin: Using hourly radiation (rather than 3 hourly) gives a better land surface energy balance but also has a larger effect on net surface SW than expected (even seems to affect something as basic as amount of Rayleigh scattering). Likely to have an unacceptably large effect on the overall surface energy budget. Ian and Martin think it needs to be parked.
Forcing update
Martin: no update on the forcing but still waiting on the ozone. Paul hasn’t seen it yet (July is a possibility).
Optimisation
Manodeep: 20 model years per day using 4 nodes per day (1000 Su). 5 nodes can get us up to 22 model years per day (1200 Su). Anton: is it harder to justify on a NCIMASS type application, Manodeep highlighted that he has tested 1-4 nodes as well and they all are more mpi efficient so there’s a balance. Tilo: 1kSu 20 years/d is similar to esm1.5 and would be happy with that (for production runs). Aidan: are you resource or time constrained? Tilo: both and storage. Martin would like to see both configs available. Aidan: highlighted current tests are using cice4 would be good to confirm that cice5 has similar performance. Rachel: we are generally limited by our ability to post-process, quality control and assess simulations.
General updates?
None.