Quantifying Prediction Fidelity in Ocean Circulation Models

Abstract

Abstract : The long-term goal is the development, validation and application of robust uncertainty quantification methods to ocean modeling, forecasting, and parameter estimation. This project explores the use of Polynomial Chaos (PC) expansions for improving our understanding of uncertainties in Ocean General Circulation Models (OGCM). Given adequate initial and boundary conditions, most OGCMs can be used to forecast the evolution of the oceanic state consistent with known physical laws. Reliable ocean forecasts, however, require an objective, practical and accurate methodology to assess the inherent uncertainties associated with the model and data used to produce these forecasts. OGCMs uncertainties stem from several sources that include: physical approximation of the equations of oceanic motion; discretization and modeling errors; an incomplete set of sparse (and often noisy) observations to constrain the initial and boundary conditions of the model; and uncertainties in surface momentum and buoyancy fluxes. Our objective is the development of an uncertainty quantification methodology that is efficient in representing the solution s dependence on the stochastic data, that is robust even when the solution depends discontinuously on the stochastic inputs, that can handle non-linear processes, that propagates the full probability density functions without apriori assumption of Gaussianity, and that can be applied adaptively to probe regions of steep variations and/or bifurcation in a high-dimensional parametric space. In addition we are interested in developing utilities for decision support analysis; specifically, we plan to demonstrate how PC representations can be used effectively to determine the non-linear sensitivity of the solution to particular components of the random data, identify dominant contributors to solution uncertainty, as well as guide and prioritize the gathering of additional data through experiments or field observations.