Collaborators:
| Dr. Ralph F. Milliff Colorado Research Associates Division (CoRA), NorthWest Research Associates, Inc. milliff@cora.nwra.com |
Professor L. Mark Berliner Department of Statistics Ohio State University mb@stat.ohio-state.edu |
| Professor Nadia Pinardi Corso di Scienze Ambientali University of Bologna n.pinardi@sincem.unibo.it |
Professor Christopher K. Wikle Department of Statistics University of Missouri wikle@stat.missouri.edu |
| Professor Emanuele Di Lorenzo School of Earth and Atmospheric Sciences Georgia Institute of Technology 311 Ferst Drive, Atlanta, GA 30332 edl@gatech.edu |
ABSTRACT:
Bayesian Hierarchical Models (BHM) are implemented to establish ensemble ocean forecasting tools for the Mediterranean Forecast System (MFS). Progress is reported for MFS-Wind-BHM and MFS-Error-BHM from Phase I of the research. Ocean ensemble initial conditions and ensemble forecasts driven by MFS-Wind-BHM exhibit distributions of ocean circulation uncertainty concentrated in mesoscale features. Plans for MFS-Error-BHM are described to refine vertical error covariance estimates, and build daily time-dependence in background error covariance for MFS reduced-order optimal interpolation. Research plans for Phase II are proposed to focus on extending skill in targeted ocean forecasts using BHM methodology for superensemble forecast implementations. Med-MultiModel-BHM is a superensemble forecast system that will include contributions from a proposed Mediterranean implementation of the Regional Ocean Modelling System (ROMS) to combine with MFS forecasts. Two variants of an MFS-MultiParam-BHM methodology are described. In one, a superensemble is constructed of MFS models employing different vertical mixing parameterizations. In a second variant, a superensemble using different model resolutions will be constructed. Large ensemble sizes from coarse resolution MFS models can combine with high-resolution MFS implementations to yield an optimal superensemble forecast. We propose to explore the accumulating impacts of BHM on MFS using stochastic optimal/non-normal operator diagnostics.
