ODA product intercomparisons are intended to improve our modeling of forecast error covariances. We plan to identify strengths of particular methods and areas that need development, and to quantify improvements produced by particular developments. Since all ocean models have biases and coupled models undergo shocks and systematic drifts after coupling of the ocean and atmosphere, one avenue of investigation, tied to theme 4, is to develop effective assimilation strategies to initialize ocean models for forecasts. This strategy may not be the same as that for ocean state estimation.
Intercomparisons of products based on comprehensive, global GCMs will be conducted with MOM and Poseidon. A quasi-linear model for the tropical Pacific Ocean will be used to provide insight into data impacts and issues relevant to ensemble and reduced state space implementations. Different assimilation procedures (OI, 3DVAR, Kalman filter) will be developed and tested as will ensemble and reduced state space implementations of these methods. The issue of the representation of systematic model biases during the assimilation process will also be addressed. We will compare different strategies for correcting model salinity during assimilation. The focus will be on the tropical oceans. Under this theme, we will generate routine (weekly) real-time products, retrospective products from the 1980’s onwards. To reduce the sources of product differences, we will undertake the intercomparisons within controlled experiments, i.e., we will adopt a consensus analysis period, set of forcings, observations, validation suite, and evaluation metrics. The evaluation metrics will include a quantification of how different the products are in regions of low data volume.
Initial discussions indicate the likelihood that we will adopt the set of surface forcings assembled by GFDL for our retrospective analyses. However, one of the first tasks we will undertake is to document our current and planned model configurations, forcing, observation suite, etc., so as to form the consensus for controlled experiments.
NCEP, GFDL, IRI, LDEO and GMAO will generate assimilation products. GFDL, LDEO and IRI will collaborate on reduced state space assimilation implementations, especially those for MOM. NCEP and GMAO will collaborate on methodologies for salinity corrections. The IRI (with GFDL) will generate retrospective (1980's-present) and routine products tailored to the resolution and physics of the COLA and IRI coupled models. GMAO, GFDL and the IRI will collaborate on bias assimilation. COLA will collaborate with NCEP, GFDL and IRI on utilization of MOM-based products for coupled forecasts. All will collaborate on the development of intercomparisons metrics, and on developing the most effective strategy for initializing the ocean for coupled predictions.