The weather affects all of us and accurately predicting it days in advance is vital for people's lives, businesses, the transport sector and farmers.
Throughout the year, the Met Office provides forecasts for more than 10,000 locations worldwide. They do this by feeding the atmospheric conditions into a supercomputer, which uses mathematical equations to simulate how the weather will evolve over hours and days. These simulations predict temperature, atmospheric pressure, wind, humidity and cloudiness, and result in the weather forecast.
The challenge
The Met Office uses a range of advanced computer models and combines the outputs of these models with observational data in order to produce weather forecasts. The speed and accuracy of predictions is limited by the size and power of the computer itself. The bigger and more powerful the computer, the quicker it can handle a huge amount of data. Running a supercomputer, however, is costly. By making the mathematical equations behind these simulations more efficient, weather predictions can be produced much faster, and running costs can be cut.
The solution
Mathematicians at the University of Bath have been working with the Met Office to develop more efficient numerical methods that allow simulations to make better use of the available observational data. These numerical methods are called adaptive meshes, and when used in data assimilation algorithms, are better at capturing rapid changes in temperature, pressure and wind speed.
'The Met Office is always working on its models. As a result, forecasts are continuously getting better, but most improvements tend to be relatively small, maybe making the model a few percent faster. Our change will give them one of those rare, and much more significant, improvements.' — Dr Eike Mueller, reader in the Department of Maths
Over a period of ten years, the collaborative research project developed approaches that were demonstrably more accurate than existing techniques and could be readily inserted into existing software at the Met Office.
The benefits
A direct consequence of this research has been a demonstrable improvement in forecasting accuracy, as measured by comparing predictions with subsequent weather observations. This has far-reaching economic and societal impacts. For example, the Met Office is now able to better predict fog hazards at airports, and road surface temperatures in winter, ensuring that preventative measures, such as gritting roads, are used appropriately and in ways that minimise their environmental side effects.
'We're thrilled that we'll be able to reduce the overall runtime of the model and allow our modelling system to deliver accurate predictions in a shorter time. This allows us to make better use of our supercomputer resources.' — Dr Ben Shipway, Strategic head of dynamics research, Met Office
The future
Based on the success of this model, the Met Office is continuing to work with Bath mathematicians, embracing their 'multigrid solver technology' for its next-generation modelling system, expected to roll out in 2025.
To ensure weather forecasts are reliable across the world, meteorologists have divided the globe into a three-dimensional grid, with each grid cell covering an area 10km2. The supercomputer must then repeatedly solve a set of 344 million coupled equations to describe the evolution of weather patterns around the world.
In order to remain accurate, frequent global communication is needed between all grid cells - which can be time-consuming and requires a huge amount of power. The machine used by the Met Office is one of the 50 fastest supercomputers in the world, and is the most powerful dedicated to weather and climate.
The multigrid solver technology helps overcome this additional challenge. Instead of solving all equations at once, it uses a grid with much larger cells, solving an approximate version of the collection of equations. Because the mesh uses much larger cells, the calculation can be done relatively quickly and cheaply. The solution is then corrected by successively refining the grid and reducing cell size until the desired resolution is reached. This process is much more efficient overall and can handle long-range phenomena in global weather patterns.
The Met Office has been using the new multigrid solver since December 2021.
