Plans for the release and support of UKESM1 were presented in the previous issue of the UKESM newsletter (see https://ukesm.ac.uk/portfolio-item/release-and-support-of-ukesm1/); in this article, we describe recent work on the release, and how those plans have progressed since then.
The model was developed and tested on the internal UK Met Office HPC, with occasional test runs on MONSooN, the shared Met Office/NERC resource (including NEXCS, the NERC-only share of the machine). Following an extensive period of scientific testing, the model was made available to our collaborators for porting to other platforms. These include ARCHER, the UKRI national platform and machines run by some of the international members of the Unified Model Partnership – specifically, the Korea Meteorological Administration (KMA) and New Zealand’s National Institute of Weather and Atmospheric Research (NIWA). The port to ARCHER is being performed by our colleagues in the Computational Modelling Services (CMS) unit of the National Centre for Atmospheric Science (NCAS). As with other models, CMS will also provide front-line user support (in collaboration with the UKESM core group) for UKESM1 following its release.
Figure 1. Time evolution of global average potential temperature at model level 65, calculated from two runs of UKESM1 with the same initial conditions performed on the Met Office HPC (red points) and the NIWA HPC (blue points). Error bars represent the spread of results from an ensemble whose (fifty) members were generated by perturbations to the initial conditions; red bars show the spread in the ensemble performed on the Met Office HPC and blue bars the spread of the NIWA ensemble.
We are currently analyzing the results from the ports to other platforms. The achievement of bit-wise identical results on different machines is generally not possible for complex numerical models such as UKESM1, because of the unpredictable way in which minute differences propagate through the model simulation on different platforms, leading to differing results. Instead of asking for identical model behaviour on two different machines, we seek to verify that each model configuration is scientifically consistent with the other– that is, could each have been sampled from the same ensemble of results generated on either machine? To check this, we create an ensemble of runs on each machine by perturbing selected variables in their initial conditions, using a perturbation whose numerical value is comparable with the machine’s precision. The spread of results (at each point in time and space) on each platform can then be used to determine whether they could have come from a common ensemble. An example comparison for a single variable (drawn from a wide collection of various results and properties) for an ensemble of short runs on the machines at the Met Office and NIWA is shown in Figure 1. Comparisons such as these, and statistical analyses, are used to verify the consistency of the UKESM1 ports to other platforms.
UKESM1 will be delivered as a set of Rose suites (see Figure 2 for a snapshot from the control panel for the UKESM1 Rose suite). Rose is the Met Office framework for developing and running climate and meteorological models. We plan to deliver three configurations of the model:
- two fully coupled configurations which each make use of all model components:
- one set up to run the CMIP6 pre-industrial control experiment, and
- one to run the CMIP6 historical experiment, and
- an atmosphere-only (so-called AMIP) configuration, in which the model atmosphere is forced by observed sea surface temperature and sea ice boundary conditions.
We note that previously we were planning to offer the historical experiment along with instructions on how to change the forcing data to run the pre-industrial control experiment; we will now make these available as separate suites.
The coupled configurations of UKESM1 have already been used in the CMIP6 DECK experiments, the majority of which are complete. The AMIP configuration was finalized very recently, and the official CMIP6 AMIP experiment is now running.
Following the completion of the port testing, UKESM1 will be openly available for use by the NERC research community. We subsequently plan to offer an update to the model allowing parts of the interactive atmospheric chemistry to be turned off (specifically, chemical oxidants and ozone concentrations will be prescribed, rather than interactively calculated in the model), whilst retaining an interactive treatment of the global carbon cycle. This configuration – referred to as UKESM1-CN – runs faster than the full model, and can be used in experiments which do not require the more complete treatment of atmospheric chemistry available in the full model. Future model developments include the addition of interactive ice sheets for both Greenland and Antarctica (see the article by Smith et al. in this issue of the newsletter).
For further information, or to be kept informed about the release, please contact Jeremy Walton (firstname.lastname@example.org).
Figure 2. Part of the control panel in one of the UKESM1 Rose suites. This pane allows the user to specify the site at which the model is being run. The popup window provides further help for this option.