Quantitative models of deep convection play a central role to improve understanding of weather, trace gas distributions, and radiative regime of the upper troposphere. Cloud-resolving models of deep convection are useful tools to simulate relevant processes. Observations of tracers such as CO2 can provide critical constraints on mass transport within these models. However, such measurements do not span the entire four-dimensional domain in space and time. We introduce a new method to improve tracer constraints on such models, combining a Receptor-Oriented Atmospheric Modeling (ROAM) framework with airborne and ground-based CO2 data. We illustrate the application of ROAM in generating initial and boundary conditions of CO2 for cloud-resolving model simulations, for a case study in the CRYSTAL-FACE campaign. Observations and model results were compared for CO2 profiles from the surface up to 16 km, inside and outside of a deep convective cloud. ROAM generated concentration fields that agreed within 0.5 ppm (1-sigma) of observations outside the cloud. When ROAM-derived initial and boundary CO2 concentrations were fed to a state-of-the-art cloud-resolving model (DHARMA), the combined modeling system successfully reproduced observed concentration differences, 0.2–0.8 ppm, between in-cloud and out-of-cloud air at 9~14 km. Results suggest that 25% of air at 14 km was lifted through the convective system from the PBL. This study demonstrates the potential of the receptor-oriented framework to constrain redistribution of air within convective systems using CO2, and it points to the need for better coordinated tracer measurements in future field missions.