Improving probabilistic weather forecasting
Scientists at the University of Southampton are developing a new framework which will facilitate more accurate probabilistic weather forecasting.
Over the years, many different techniques have been used in weather forecasting, from relatively simple observations of the sky to highly complex mathematical models run on the world's largest computers. Despite significant advances in this field, due to the unpredictability of the weather, forecasting remains a complex business.
Scientists are working towards a solution. Professor Manfred Opper from the University's School of Electronics & Computer Science (ECS) has started work with Aston University, the University of Surrey and the Met Office to develop new methods to improve probabilistic weather prediction.
The researchers are starting their work with the assumption that all models have errors and that they will therefore need to adopt a probabilistic framework which will allow them to characterise not just the typical behaviour but also the uncertainty that results from model error and other sources. This approach will set them apart from most other environmental models which are essentially based on a deterministic view of the world.
Their first task will be to produce a computationally efficient algorithm that can propagate the uncertainty in the model state through space and time. Unlike many other approaches to quantifying uncertainty they will exploit the known physics when this is available, and be able to estimate unknown or imperfectly known model parameters from observations to augment this. Their methods will be tested on a range of simplified models which exhibit the same behaviour as weather forecast models, but have a controlled number of degrees of freedom.
Professor Opper commented: 'The need for probabilistic models is becoming increasingly recognised in the academic and research community across environmental science, but it is yet to make a strong showing in the more operational setting of commercial weather forecasting. Our work will enable a more principled and accurate approach to probabilistic forecasting to be considered.'
Dr Dan Cornford from Aston University added: 'We expect that the result of this research project will be a new framework for conducting probabilistic modelling for environmental systems, which will allow us to make more accurate probabilistic forecasts. We hope that this will bring improved weather and climate forecasts in the future, but it clearly also has applications beyond these areas.'