Global to regional km-scale hackathon 2026

Registration is now closed. The registration deadline was Friday 17th April 2026. If you would still like to join, please complete the form below and we will confirm your registration if this is possible.

Please use this Microsoft Form to register your interest and state which days you will join for. The questions will help us to make sure everything is arranged for the hackathon to run smoothly. Note that this event is free - there is no registration fee. The Met Office will provide lunch and snacks during the working day, we hope to plan a group meal on evening of Wed 20 May.

The 2026 UK K-Scale hackathon will run from lunchtime on Monday May 18 to lunchtime on Friday 22 May. Lunches will be provided at the Met Office Collaboration building each day. The draft timetable for the week is linked below.

Timetable 2026 hackathon

A K-Scale model hierarchy, encompassing a set of year-long global and regional domain simulations across a range of grid resolutions and science options. This is designed to enable analysis of a range of research questions. This year’s event builds on the 2025 WCRP Global Hackathon which aimed to: drive advances in high-resolution models; to promote and share best practices for coding and workflows; and to foster a global community with open access to datasets. The 2025 event established a number of working groups through the UK event hosted in Oxford. For this 2026 UK event, the Global to Regional km-scale hackathon, will tackle a range of questions in the context of developing evidence to address two over-arching hypotheses:

A: Better representing mesoscale phenomena, such as details of local-scale convection, feeds back on the large-scale atmospheric circulation.

B: Development of high quality, high-resolution global simulations generates new information that may be of value for enhancing development of machine learning approaches to weather and climate prediction.

A number of data sets will be made available on the JASMIN object store, in HEALPix format, and accessible remotely. Notebooks to perform the analyses above, at least at an initial stage, will be provided to all, including catalogues for data discovery. Observational products will also be provided, e.g. ERA5, IMERG, CERES. Many of the more complex data sets, e.g. TC tracks, MCS tracks, will be provided before the event. Multiple working groups will be formed around topics linked to addressing the hypotheses above, and will work together at the venue. Potential working group themes might include:

• Tropical and Extra-tropical cyclones
• Monsoon circulations and energetics
• Tropical wave activity, propagation and tele-connections
• Mesoscale convective systems
• Upscale influence of mesoscale phenomena on the large-scale circulation – utilising the local scale-to-scale energy transfer tool LoSSETT
• Land surface interactions
• Clouds and radiation
• Further topics of interest to be added ahead of and during the hackathon

Across each of these groups, it may be useful to consider:

1. How do the characteristics of the phenomena of interest differ between global and tropics-wide domain simulations with explicit representation of convection [i.e. using the RAL3.3 science configuration] relative to a benchmark high-resolution weather/climate configuration with parametrized convection [GAL9]?
2. What impact does use of a new convection parametrization scheme (CoMorphA, adapted to higher resolution) have on results?
3. How sensitive are representation of regional phenomena to the domain extent of high-resolution simulations – are there notable differences between global and tropics-wide simulations with explicit convection to benchmark regional Limited Area Models using the same science configuration?
4. How do biases evident in the hierarchy simulations compare to any known persistent biases related to phenomena of interest in coarser resolution climate models? Are there any sensitivities (e.g. magnitude, sign, extent) to the choice of science configuration, domain extent, or grid resolution of the 1-year hierarchy simulations?
5. Given the majority of current generation global machine-learning weather prediction models are trained using ERA5 reanalysis data, how does the representation of phenomena differ in the K-Scale hierarchy simulations? Are both or either representations physically realistic when compared to available observations?