As remotely sensed data become an increasingly dominant source of the information provided, through assimilation procedures, to operational forecast models, the related problems of correlated or biased measurement error and the poor numerical conditioning of the formal inversion by the assimilation become increasingly severe. This paper will address a unified approach to the treatment of measurement bias and correlation through the use of ancillary variables. The proposed treatment further inflates the already large condition number intrinsic to the analysis inversion problem but, by employing an observation-space form of the analysis and adopting an extension to the averaged block-matrix preconditioners recently advocated by R. Daley and E. Barker based on grouping the data into overlapping small clusters, we expect to be able to achieve a dramatic reduction in the condition number. An outline of the proposed techniques and preliminary results will be presented.