The ability to monitor and understand natural and anthropogenic variability in atmospheric carbon dioxide (CO.sub.2) is a growing need of many stakeholders across the world. Systems that assimilate satellite observations, given their short latency and dense spatial coverage, into high-resolution global models are valuable, if not essential, tools for addressing this need. A notable drawback of modern assimilation systems is the long latency of many vital input datasets; for example, inventories, in situ measurements, and reprocessed remote-sensing data can trail the current date by months to years. This paper describes techniques for bias-correcting surface fluxes derived from satellite observations of the Earth's surface to be consistent with constraints from inventories and in situ CO.sub.2 datasets. The techniques are applicable in both short-term forecasts and retrospective simulations, thus taking advantage of the coverage and short latency of satellite data while reproducing the major features of long-term inventory and in situ records. Our approach begins with a standard collection of diagnostic fluxes which incorporate a variety of remote-sensing driver data, viz. vegetation indices, fire radiative power, and nighttime lights. We then apply an empirical sink so that global budgets of the diagnostic fluxes match given atmospheric and oceanic growth rates for each year. This step removes coherent, systematic flux errors that produce biases in CO.sub.2 which mask the signals an assimilation system hopes to capture. Depending on the simulation mode, the empirical sink uses different choices of atmospheric growth rates: estimates based on observations in retrospective mode and projections based on seasonal forecasts of sea surface temperature in forecasting mode. The retrospective fluxes, when used in simulations with NASA's Goddard Earth Observing System (GEOS), reproduce marine boundary layer measurements with comparable skill to those using fluxes from a modern inversion system. The forecasted fluxes show promising accuracy in their application to the analysis of changes in the carbon cycle as they occur.