Inverse modelling for trace gas surface flux estimation, impact of a non-diagonal B-matrix

One of the most appealing uses of data assimilation is to infer useful information about a dynamical system that is not observed directly. This is the case for the estimation of surface fluxes of trace gases (like methane). Such fluxes are not easy to measure directly on a global scale, but it is possible to measure the trace gas itself as it is transported around the globe. This is the purpose of INVICAT (the inverse modelling system of the chemical transport model TOMCAT), which has been developed here. INVICAT interprets observations of (e.g.) methane over a time window to estimate the initial conditions (ICs) and surface fluxes (SFs) of the TOMCAT model.

This talk will show how INVICAT has been expanded from a diagonal background error covariance matrix (B-matrix, DB) to allow an efficient representation of a non-diagonal B-matrix (NDB). The results of this process are mixed. A NDB-matrix for the SF field improves the analysis against independent data, but a NDB-matrix for the IC field appears to degrade the analysis. This paper presents these results and suggests that a possible reason for the degraded analyses is the presence of a possible bias in the system.