EPSRC DTP PhD studentship: Advanced statistical post-processing of ensemble weather predictions
University of Exeter - College of Engineering, Mathematics and Physical Sciences
|Funding for:||UK Students, EU Students|
|Funding amount:||£14,296 per annum|
|Placed on:||26th October 2016|
|Closes:||11th January 2017|
|★ View Employer Profile|
Probabilistic weather forecasts present users with likelihoods for the occurrence of different weather events. Demand for such forecasts is increasing as they provide users with a basis for risk-based decisions. For example, a council may decide to deploy a road gritting service if the probability of widespread ice formation exceeds 50%. It is crucial that probabilistic forecasts are well calibrated. For example, events predicted to occur with probability 70% should subsequently occur 70% of the time.
Decisions based on poorly calibrated forecasts, forecasts in which the probability of an event is systematically underestimated for example, could lead to inappropriate actions and significant losses. This is particularly true for extreme weather events, such as heavy precipitation or severe winds, which impact most heavily on society. Calibrated forecasts that accurately quantify the likelihoods of such events are therefore highly desirable to a wide range of users. While an extreme event at a single location can be damaging to the local area, the consequences may be even more serious if the event affects a wider area. Therefore, calibrated forecasts that also accurately reflect the spatial dependence of extreme events are highly desirable.
Probabilistic weather forecasts are typically derived from ensemble forecasts. An ensemble is a collection of deterministic forecasts, where the forecasts differ either in the numerical weather prediction model used or the initial conditions supplied to the model. Forecast ensembles are still biased both in location and dispersion. They tend to be underdispersive, leading to overconfident uncertainty estimates and an underestimation of extreme weather events.
The project will develop novel multivariate statistical techniques for recalibrating forecast ensembles that capture spatial, temporal and cross-variable dependence. These will improve probabilistic prediction of compound weather risk, both multi-site and multi-variable. A particular emphasis will lie on high-impact extreme weather events. Forecast performance will be assessed with scoring rules and graphical tools. The project will compare the skill of the novel methods with standard recalibration techniques.
The research will be conducted in close collaboration with the Met Office. We will use historical data from the Met Office's ensemble prediction system MOGREPS together with the corresponding verifications.Meteorological variables of interest are temperature, surface pressure, wind speed and precipitation. We are interested in short- to medium-range weather forecasting where there is considerable variability but still some skill in the ensemble.
The project is interdisciplinary in nature, drawing on applied statistics, applied mathematics as well as weather and climate science.
Share this PhD
Type / Role:
South West England