New science upgrades to our global wave model
Significant science upgrades were introduced to the global wave model run at the Met Office in November 2016. The testing of the latest upgrades took full advantage of the new supercomputer, currently the fastest environmental supercomputer in the world, capable of up to 14 trillion calculations per second.
Parallel testing of new science changes
Upgrades to configurations of the Unified Model (UM) run at the Met Office are tested in parallel to the operational version of our weather forecasting models in what are known as parallel suites. The latest upgrades, which were tested in parallel suite 38 (PS38), are now fully operational.
PS38 has seen a series of improvements to the global wave model. This model provides operational forecasts out to 5 days ahead. The same grid scheme and physics configuration have also been applied to a North Atlantic-UK ensemble wave model which provides probabilistic forecasts out to 7 days ahead.
- A version update to WAVEWATCH III® – taking us up to version 4.18.
- A change in the source term physics used in the model – these represent wave growth and dissipation processes.
- The configuration uses a Spherical Multiple-Cell (SMC) grid with refined cell resolution in coastal waters.
This is an unstructured grid, which uses refined cell scales reducing from 25km in the open ocean up to 3km at the coastline around the UK. The choice of a refined grid method has been made in order to improve overall model accuracy through providing a better description of coastline and islands, but has the important additional benefit of enabling good quality forecasts to be generated in coastal waters without the requirement to set up and run nested regional models. Consequently we have increased our ability to provide forecasts in shelf seas waters worldwide without having to run and maintain a large number of dedicated regional models.
Because the model uses varying time-steps for different resolution cells and retains conventional latitude-longitude grid features, such as rectangular cells and finite difference schemes, it is almost as efficient to run on the Met Office HPC as a conventional latitude-longitude grid model. The use of rectangular cells enables further efficiencies when post-processing the data from the model for downstream use.
Comparisons, between the new configuration and the global model run operationally during the trials period, show a major overall improvement. The changes result from a combination of both the grid and source term updates in the wave model, plus a small improvement in the skill of our global marine wind forecasts.
Considerable reduction in significant wave height errors
Significant wave height errors are reduced from approximately 18% of the background signal in the old operational model to 14% in the new configuration, representing relative change in the errors of over 20%. A similar relative improvement is seen in errors for prediction of peak wave period.
Improvement in forecast lead times
In forecast lead time terms, our improved model achieves approximately a 2 day improvement in errors (i.e. our day 3 forecast now has same level of error as our old day 1 forecast).
Reduction in large over-prediction errors during storms
These changes in model quality have mostly resulted from a reduction in large over-prediction errors during storms. Improving prediction of high energy sea-states is particularly important for maritime safety, since our forecasts need to be of a dependable quality when making decisions that may affect Safety of Life at Sea (SOLAS).
Read the science report
Results of the parallel suite trials and further details of our new global wave model are available in a Met Office science techical report titled "Application of a refined grid global model for operational wave forecasting".