Information-theoretic framework for uncertainty quantification and improving Lagrangian predictions based on imperfect Eulerian models
Abstract
The research proposed here aims to answer fundamental questions regarding the transition from Eulerian (velocity field based) ensemble forecasts to subsequent skilful Lagrangian (trajectory based) forecasts. Given the operational need for deriving reliable Lagrangian ensemble forecasts from Eulerian ensembles of imperfect velocity fields, a development of a systematic mathematical framework for tackling this fundamental problem is long overdue. Operational ocean models perform well when judged by Eulerian metrics (?72h lead time), but produce only marginally useful trajectory information (less than 24h lead time). Consequently, there is an urgent need for a new framework that unifies notions of Eulerian predictability and uncertainty interacting with Lagrangian predictability and uncertainty. The scope of this proposal aims beyond Eulerian ensemble forecasting for oceanic flows; it focuses on a probabilistic framework for (i) quantifying uncertainty in Lagrangian predictions, and (ii) improving these predictions through Lagrangian ensemble forecasting. Both these issues can be addressed by a synergistic approach combining information-theoretic ideas with dynamical systems theory and stochastics, and with statistical inference techniques on the path space of Lagrangian trajectories. The ultimate goal is to use this framework for tuning imperfect Eulerian models based on the available information so that the loss of relevant information in the subsequent Lagrangian predictions is minimised. In particular, this framework will enable identification of those flow characteristics which are most important to reproduce in imperfect Eulerian models in order to maximise the Lagrangian prediction skill. It is expected that the relevant flow features will be associated with certain coherent structures in the flow and that the Eulerian models which are optimal for skilful Lagrangian predictions do not coincide with optimal models for Eulerian predictions. The necessary groundwork has already been laid by the PI and collaborators in the Eulerian framework and it should lead to accelerated progress in this area.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 12, 2016
- Source ID
- N000141512351
Entities
People
- Michał Branicki
Organizations
- Office of Naval Research
- United States Navy
- University of Edinburgh