An exploratory statistical analysis of the covariance structure of joint atmosphere-ocean variables is performed on global scales in preparation for a coupled atmosphere-ocean model assimilation scheme for long-range forecasting. We focus here on redundancy analysis which can be a powerful tool for determining directionality of causal influence in the natural system. This statistical technique allows one to identify regions of high variability associated with the propagation of information and errors from the one medium to the other, which can be of importance to data assimilation. State space models are introduced to aid in the interpretation of the redundancy analysis eigenvectors. The data sets used in this study are the global fields of Sea Surface Temperature (SST) and Sea Level Pressure (SLP) from (i) an NCEP reanalysis and (ii) output from the Canadian Centre for Climate Modelling and Analysis coupled model. Time-lagged Redundancy analyses are carried out on global scales for: 1) SST forcing SLP and 2) SLP forcing SST. In all case studies, SLP is found to be the principal forcing agent leading the SST by about 1 month. The two dominant global patterns identified are 1) ENSO SST patterns and low-high pressure bands in the Pacific Ocean and 2) equatorial heating in the Atlantic with mid-Pacific pressure anomaly. In a regional study of the North Atlantic, the Atlantic Multidecadal Oscillation (AMO) and Atlantic Meridional Mode (AMM) are observed after a low pass filtering of the data and the North Atlantic SST Tripole anomaly after a high pass filtering of the data. Possible applications of this technique to downscaling studies of the natural system are reviewed. The implications of these statistical relationships for modeling the background error covariance, and ultimately prediction of the coupled atmosphere-ocean system using a data assimilative model, are discussed.