There are two basic approaches for inferring surface-atmosphere exchange for trace gases on regional scales: a bottom-up approach, in which local process knowledge is scaled up, and a top-down approach, in which the larger-scale constraint from atmospheric concentration measurements is applied in combination with transport models. Here we combine the two approaches, and assess the information content added by boundary layer concentration data. More specifically, we analyze the potential for inferring spatially resolved surface fluxes from atmospheric tracer observations within the mixed layer, such as from monitoring towers, using a receptor oriented transport model (Stochastic Time-Inverted Lagrangian Transport [STILT] model, [Lin et al., 2003]) coupled to a simple biosphere in which CO2 fluxes are represented as functional responses to environmental drivers (radiation and temperature, [Gerbig et al., 2003]). Transport and fluxes are coupled on a dynamic grid using a polar projection with high horizontal resolution (~20 km) in near field, and low resolution far away (as coarse as 2000 km), reducing the number of surface pixels without significant loss of information. To test the system, and to evaluate the errors associated with the retrieval of fluxes from atmospheric observations, a pseudo data experiment was performed. A large number of realizations of measurements (pseudo data) and a priori fluxes was generated, and for each case spatially resolved fluxes were retrieved. Results indicate strong potential for high resolution retrievals based on a network of tall towers, subject to the requirement of correctly specifying the a priori uncertainty covariance, especially the off diagonal elements that control spatial correlations.