Topics in non-linear data assimilation and structure alignment and their applications in ensemble forecasting.

摘要

This thesis is a collection of work detailing techniques of analyzing and applying nonstandard methods of sequential data assimilation. First is a study on the effects of coupling between non-linear dynamics, instability and model error in a quasi-geostrophic model. A method is employed to combine information from simulated observations, the model, and prior knowledge to assign probabilities to the model's parameters. This process results in the convergence to high probabilities for parameter values that produce the simulated observed field. The second body of work focuses on identifying and correcting location errors in coherent structures. A method is detailed that identifies jet structures in a quasi-geostrophic channel model and then aligns them with observed jet structures prior to each analysis cycle. The application of this method is shown to be both effective in reducing the error in the simulations as well as computationally inexpensive. Finally the incorporation of non-standard data with high temporal and spatial resolution into a complex prognostic model is investigated. A method to weight an ensemble forecast produced by the WRF model by comparing simulated radar reflectivity with observed radar reflectivity is presented and the implementation of this post-process is shown to improve the skill of the forecasted precipitation fields.