Seamless Evaluation of Stochastic Physics Parametrizations
Thesis or dissertation
University of Exeter
A substantial segment of the error in numerical weather prediction and climate projections comes from the intrinsic uncertainties of General Circulation Models of the atmosphere. Stochastic physics schemes are one of the preferred methods to represent the model uncertainty in Ensemble Prediction Systems, where different realizations of the same forecast are created to quantify the probabilities of different outcomes in the atmospheric flow. Stochastic physics schemes have been successfully employed in medium-range and seasonal forecasting systems, as they increase the skill of probabilistic forecasts. Similarly it has been demonstrated than these schemes can improve certain aspects of the model's climate. However, it is still not clear whether they are a truthful representation of the model uncertainties they aim to represent. In this thesis, a collection of stochastic physics schemes are evaluated using a seamless approach. It is found that they can improve the representation of the tropical climate and extra-tropical cyclones, but they degrade the individual representation of these processes deteriorating the deterministic skill of the model. Some important features of the model can be degraded by the stochastic physics schemes, like energy and moisture conservation on climate scales. Some closures to the schemes are proposed and successfully tested to remove or reduce some of the problems found. Alternative approaches in the development of stochastic parametrizations are also investigated. Stochastic physics schemes have some benefits but still require further development to produce a realistic representation of model error. It is also recommended that evaluation methodologies must be expanded to include process-based diagnostics to display the realism of its perturbations.
PhD in Mathematics