Bayesian Hierarchical Model Characterization of Model Error in Ocean Data Assimilation and Forecasts
Abstract
The Bayesian Hierarchical Model (BHM) methodology is exploited to identify, characterize, and model the irreducible model error in ocean data assimilation and forecast systems. We describe 4 objectives addressed in the fiscal year October 2012-September 2013. First, we seek to extend the proof-of-concept results comparing a BHM surface wind ensemble with the increments in the surface momentum flux control vector in a four-dimensional variational (4dvar) assimilation system. The current objective is to convert BHM surface wind realizations to create an ensemble of surface stress vectors. Second, continuing the effort to understand irreducible model error induced by representing the ocean state vector on a discrete grid, the current objective is to estimate the Hellinger distance between posterior distributions described in the next section. Third, we have extended the hierarchical models for stochastic time-varying error-covariance matrices associated with data assimilation to include the case where both the observation and background error covariances are updated, yet dependent upon each other. Fourth, we have extended the emulator-assisted data assimilation methodology by extending the parameterization of the spectral quadratic nonlinear spatio-temporal models to accommodate the inclusion of nonlinear interactions from small scales to inform the evolution of large scale modes.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 30, 2013
- Accession Number
- ADA601834
Entities
People
- Christopher K. Wikle
- L. M. Berliner
- Radu Herbei
- Ralph F. Milliff
Organizations
- University of Colorado Boulder