The ability to make predictions is fundamental to almost all human endeavours. Almost every decision we make is coloured by our own expectations of the outcomes of future events. As a general rule, the more confidently we believe the predictions that inform our decision the easier the decision is to make. On the other hand, greater confidence married to inaccurate predictions can lead to very poor decisions. Often our current knowledge is too imprecise, or simply insufficient, to know with certainty what will happen in the future. In such cases it is vital to have some quantification of the impact of the uncertainty of our present knowledge on future predictions. The study of predictability aims to provide this quantification.
In weather and climate forecasting, predictions are made by evolving the best estimate of the current state of the atmosphere and ocean forward in time using very large computer simulations of the physical laws which govern the flow. There are two main sources of uncertainty which impact on the uncertainty of predictions. The first is uncertainty due to the fact that the current state of the system and the relevant external forcings are not known precisely. This first source of uncertainty is usually referred to as initial and boundary condition uncertainty. The main strategy which has been developed to quantify this is ensemble forecasting, in which multiple forecasts with small differences in initial and/or boundary conditions are created. The second source of uncertainty is the equations used to evolve the atmospheric state forward in time. This second source of uncertainty stems from terms in the equations that are inaccurate or missing, either due to lack of knowledge or simply because computer resources are limited and not all terms can be included. To account for this type of uncertainty two main strategies have emerged. One is to create ensembles using different models. These are known as multi-model ensembles. The second strategy is to explicitly incorporate the uncertainty into the equations as a random process. This is referred to as stochastic parameterisation and essentially means that two forecasts produced with the same model and initial conditions will not be identical.
Predictability Research within NCAS Weather
NCAS weather conducts research into the predictability of weather and climate in both the Department of Meteorology at the University of Reading and in Oxford University.
Recent work on predictability and ensemble forecasting in the University of Reading has focused on two main topics:
- The predictability of the North Atlantic jet regimes.
- The predictability of frontal waves and cyclones in the Met Office MOGREPS ensemble forecasting system. This work forms part of the DIAMET project and is being performed in collaboration with project partners in the Met Office.
Recent work on predictability and ensemble forecasting at Oxford University has focussed on these topics:
- Predictability on monthly, seasonal and decadal time scales
- Assessing model uncertainty in weather and climate models
- Seamless prediction of weather and climate
- Development of stochastic parameterisations
Staff involved in this research are Dr Tom Frame and Dr John Methven at Reading, and Dr Antje Weisheimer and Prof Tim Palmer at Oxford