Abstract. To design a monitoring network for estimating CO2 fluxes in an urban area, a high-resolution observing system simulation experiment (OSSE) is performed using the transport model Graz Mesoscale Model (GRAMMv19.1) coupled to the Graz Lagrangian Model (GRALv19.1). First, a high-resolution anthropogenic emission inventory which is considered as the truth serves as input to the model to simulate CO2 concentration in the urban atmosphere on 10 m horizontal resolution in a 12.3 km × 12.3 km domain centred in Heidelberg, Germany. By sampling the CO2 concentration at selected stations and feeding the measurements into a Bayesian inverse framework, CO2 fluxes on a neighbourhood scale are estimated. Different configurations of possible measurement networks are tested to assess the precision of posterior CO2 fluxes. We determine the trade-off between the quality and quantity of sensors by comparing the information content for different set-ups. Decisions on investing in a larger number or in more precise sensors can be based on this result. We further analyse optimal sensor locations for flux estimation using a Monte Carlo approach. We examine the benefit of additionally measuring carbon monoxide (CO). We find that including CO as tracer in the inversion enables the disaggregation of different emission sectors. Finally, we quantify the benefit of introducing a temporal correlation into the prior emissions. The results of this study have implications for an optimal measurement network design for a city like Heidelberg. The study showcases the general usefulness of the inverse framework developed using GRAMM/GRAL for planning and evaluating measurement networks in an urban area.