Tsuyoshi Wakamatsu
Institute of Ocean Sciences, Department of Fisheries and Oceans
Sidney, BC, Canada
A data assimilation system is under development at the Institute of Ocean
Sciences based on IPEZ (Inverse Primitive Equation model in Z coordinate)
package based in IOM (Inverse Ocean Model) system developed by Chua and
Bennett (1991) at the Oregon State University. The goal of the system is
to create the best estimate of the Pacific Ocean state during the last
decade (1992-2002). We solve the weak-constraint 4DVAR problem using the
representer method in this system. Assuming that the dynamics are weakly
nonlinear for the basin scale ocean circulation, each representer
describes spatial and temporal structure of the impact from a
corresponding datum to the optimal solution. Since the representer
structure is determined by the model error covariance matrix and model's
dynamical operator, we need to search for the best balance between them to
determine suitable a interpolation kernel for the reanalysis purpose. In
the current system, the error covariance matrix is designed to be
univariate while multivariate structure of the representers solely depends
on the dynamical operator, which changes its dominant balance according to
a data assimilation period. In this presentation we discuss the
sensitivity of the analysis to the decorrelation length-scale in the
dynamical error covariance matrix and the length of the assimilation
period.