MUMIP Meeting 5

Monday 29 April 2024
2-3:30pm (UK)

Present:

Jeff Beck (NOAA / DTC)
Hannah Christensen (University of Oxford)
Richard Forbes (ECMWF)
Edward Groot (University of Oxford)
Hugo Lambert (University of Exeter)
Wahiba Lfarh (Météo-France)
Sarah-Jane Lock (ECMWF)
Will Mayfield (NCAR)
John Methven (University of Reading)
Kathryn Newmann (NCAR / DTC)
Romain Roehrig (Météo-France)
Kasturi Singh (University of Exeter)
Xia Sun (NOAA / DTC)
Keith Williams (UK Met Office)
Nils Wedi (ECMWF)

Apologies

Martin Leutbecher

NOAA/NCAR DTC: XS, KN, WM update

• Analysing the data.

Produced PDFs of ICON temperature and wind conditioned on SCM predicted tendencies (following Christensen 2020).

For both, see that standard deviation increases with magnitude of mean tendency for negative tendencies (some evidence for SPPT), but constant sigma for positive tendencies.

Uncertainty in T and wind seems to be poorly correlated.

Also see some systematic differences between the two physics suites

Further have computed snapshot of ‘optimal perturbation’ fields following Christensen 2020

RR: do you remove spin-up? XS: no

HC: suggestion to remove the spin up and recompute conditioned PDFs

• New high-resolution UFS runs

Have produced limited area high-resolution UFS runs. 10 days simulation. 30 days sims to be done next year. Used GFS for initial conditions. Hourly output after 12-hour spin-up. 64 vertical levels. Consider same IO region.

Coarse grained these fields using HC’s approach. One issue was lack of grid corners for conservative interpolation. Instead using ncl’s regridding tool for curvilinear grids to compute grid corners. Then used cdo for the regridding.

Coarse graining finished, and format consistent with ICON data. Located on NCAR’s HPC. Available for use.

Have published SCM code to github repo. Public release of code. Compatible with Dephy format data.

• Next steps

Wil provide extra fields including cloud cover and precipitation. Continue to test multiplicative noise hypothesis. E.g. conditioned PDFs at different levels. Explore how can high-resolution tendencies be used

• Questions/discussion

HC: discussion about possible use of high-res tendencies, and how the partitioning between physics and dynamics will be different between the two models
JM: coarse graining in time?
XS: No, but analysis thus far on 6-hr accumulated tendency. May move to 3-hr windows
HC: plans with new high-res dataset?
XS: how does reference dataset impact SPPT tests? But also will archive high-res tendencies from these runs which could be useful.
NW: spectrally analyse the tendencies, in space, from high-resolution runs? At which scales do the physics and dynamics force - before and after coarse-graining

Météo France – WL, RR update

• Progress with simulations

Using ICON coarse grained inputs to derive fields needed by Météo France ARPEGE-clima SCM

Use vertical velocity and geostrophic wind forcing together with surface temperature as lower boundary condition

Initial tests carried out on 10x10 grid domain. By end of last week, managed to run model in parallel over the whole of the Indian Ocean domain – 200x220 points. Have collected these together into a single file. Have started comparing SCM with ICON data. Significant difference in local values, though larger scales well reproduced.

Next steps – run over more initial times, and start quantifying errors associated with convection in different meteorological situations.

• Questions/discussion

HC: How long does it take to run the whole domain? WL: 1 hour wall-clock time per initial condition using 10 nodes.
EG: scales of variability seem very different – can you comment on the grid spacing? Is the effective resolution different?
RR: grid spacing all the same.
XS: we saw this behaviour too
RR: Note that there is no feedback from the dynamics in the SCM so it is noisier – these interactions would smooth the fields
RR: see big difference in calculated and ICON surface fluxes - SCM underestimates fluxes compared to ICON – back to our old discussion. Is this part of the error we want to assess or not? Different energetics in the column
HC: Oxford will test both
RR: Météo France also
JM: grey zone – issues to do with unresolved fluxes
RR: agree – would get systematic enhancement of surface fluxes by mesoscale circulations
RR: some of this may be specific to ICON and its turbulence scheme. Will be good to have an alternative high-resolution simulation available from the DTC

Oxford – EG, HC update

• New Group workspace on UK JASMIN machine

Common ICON inputs available. Subfolders set up for IFS data. All groups encouraged to request access to Jasmin and push their data there

• IFS simulations update

Working with new OpenIFS release CY48. Software being updated to parallelise the SCM runs

Have spent time checking the consistency of budgets in IFS SCM – e.g. changes between tendencies and change in state in IFS. Care is needed in interpreting the timestamps of the files.

Have chosen various other settings to generate initial condition files, including sourcing missing variables and padding the data above ICON data top.

Currently have MPI issues in parallelising the SCM

RF: ECMWF will provide help on this issue
XS: how many cores are you using?
EG: just one core, 16 tasks. Testing this to begin with

• First analysis of NCAR data

Computing correlation between temperature and humidity tendencies as a function of Temperature and simulation day, for both convection and microphysics parametrisation for two physics suites. Reveals differences between different convection schemes between two models, especially at low levels and near the freezing level. Illustrates differences between physics packages.

A recent paper looked at systematic changes in ERA5 rainfall over time – shift from convective to large scale precipitation. Is this due to the spin-up? -> Use MUMIP data to study spin-up of models, since we are using non-native ICs for the SCMs. Have therefore been analysing DTC data looking at CAPE and CIN to predict vertical velocity, conditional on existence of buoyancy. Stratify data according to when convection was active in each scheme. Use to assess differences between physics packages.

NW: vertical velocity derived from CAPE? How does it compare to high resolution fields?
EG: Yes. DTC physics packages on same vertical levels which is convenient. Would be very helpful to have ICON interpolated to DTC model levels, plus the tendency data at 3 hrs
XS: Please can we have a copy of the slides – will forward to Lisa Bengtsson. Will provide cloud and precipitation data, and +3hr forecast.

Exeter – KS, HL update

• Update

Have been working on speeding up simulations. Currently 4,000 columns take 45 minutes.

Will share preliminary results soon.

HC: suggest to share technical experience between MF – DTC – Exeter – Oxford given the issues with parallelisation at Oxford and Exeter.
XS: Happy to provide tips of how DTC has parallelised the problem.
KS: Using Met Office SILK software

AOB

HC: We are organising a Workshop on Model Uncertainty to be held at University of Oxford in w/c 23 September. We also plan to hold a MUMIP meeting at the end of the workshop – fully hybrid – an opportunity to look at first results and plan out an implementation paper.

Next meeting: Early summer