Long term observations of atmospheric greenhouse gas measuring stations improve our understanding of greenhouse gas sources and sinks. These dry mole fraction measurements can be linked to surface fluxes by atmospheric transport inversions. In the framework of ICOS (Integrated Carbon Observation System), more observing stations are to be deployed within the European domain. A quantitative network design study is required to perform this optimization and to assess potential observing networks. A regional inverse modeling framework was set up that derives biosphere-atmosphere exchange fluxes at regional scales using CO2 measurements from tall towers, and ground stations. The modeling framework consists of the following components: the global transport model TM3, the regional Stochastic Time-Inverted Lagrangian Transport model (STILT), the Vegetation Photosynthesis and Respiration Model (VPRM), gridded emissions from fossil fuel burning, ocean-atmosphere exchange fluxes, and a Bayesian inversion scheme. This thesis first studies the flux error structure, and explains how these uncertainties are distributed spatially and temporally. Fluxes from three biosphere models were used and compared against flux observations from 53 Eddy covariance flux towers and from an aircraft campaign. Spatial and temporal autocorrelations of the daily model-data flux residuals were approximated by an exponentially decay error model. This flux error information is implemented in the inversion system. A synthetic experiment was performed using two different biosphere models, one to produce the a-priori flux field, and the other to provide fluxes that served as a “known truth”. This experiment allows to quantitatively assess the system’s ability to correct fluxes at different spatial and temporal scales. A network design study is conducted, using different network configurations for ICOS current and future stations, evaluating the uncertainty reduction on the terrestrial fluxes.