This paper describes a geospatial data system that produces KML representations of complex spatio-temporal datasets related to modeling the atmospheric carbon cycle. KML is an open standard language for transferring annotated geospatial data that can be used by many modern geospatial software packages, particularly virtual globe applications. The server component of the data system is built using a variety of open source software packages, which provide flexibility for creating custom geospatial representations of the datasets. The paper shows examples of how KML representations of atmospheric CO2 datasets and model outputs can be visualized with virtual globe client applications, allowing a diverse group of users to explore the complex scientific datasets that are central to the discussion of climate change and global warming.