Improving weather forecasts through code improvements
Researchers from the Bureau of Meteorology (BoM) are using NCI’s high performance computing facilities and expertise to research and develop the ACCESS (Australian Community Climate and Earth System Simulator) model used in daily weather forecasts all over the country. While the BoM uses its own supercomputer to run the daily forecasts that go out to the public, some advanced work in the performance analysis of future candidate systems is conducted at NCI.
Improvements in computing power make it possible to simulate the atmosphere, water and land much more reliably and accurately than ever before, but to make this a reality, the model’s code needs to be improved as well. NCI leads a local team that optimises the next generation of ACCESS code across the different timescales – from numerical weather prediction, seasonal forecasts and climate – that are then used for the Bureau’s operational needs. Tim Pugh, Supercomputer Programme Director at the BoM, says: “The ACCESS optimisation team has a mix of staff from NCI, BoM, and CSIRO, who understand both the models and computational science.”
One major improvement by the team reduced the model run time for the Bureau’s operational forecast by 30%. This means that the Bureau can now fit more model runs inside their set forecasting schedule, or make additional improvements to the model. That produces better outcomes for citizens, and a huge time and cost benefit to the BoM, says Tim Pugh. “Our weather applications share common code with the climate applications, so benefits in weather modelling translate into benefits in climate modelling. If CSIRO researchers are doing a 200 year simulation, saving 30% of the elapsed time is enormous.”
Any performance and scalability improvements to the ACCESS model code are also regularly incorporated into the Unified Model (UM) of the UK Meteorological Office, which ACCESS is based on. Other weather and climate centres and research teams using the UM are then able to apply the improvements to their own modelling. For the operational ACCESS model which runs at the Bureau, the improvements are due to appear in the newest modelling systems to be released next year.
The UK Met Office UM collaboration partners are currently working to improve the convection scale modelling in ACCESS in higher resolution grids. Convection models, which describe the movements of air in the atmosphere, work best when the resolution gets to the order of hundreds of metres. Going from the 1.5 kilometre resolution that is currently in the next operational model down to hundreds of metres in key forecast areas of interest, such as populated areas, will involve a huge increase in computing power and more effective computing solutions built into the code.
Pugh says “Whenever we have more computational power available, we can always improve the modelling through improved data assimilation, physics and dynamics. The NCI computational specialists have been doing a great job. We’ve seen a lot of very significant computational advancements in the software from their work, and each of those has resulted in a more efficient modelling system in our operations.”