Information integration implies access to distributed information sources without interfering with the autonomy of the underlying data sources. Integration of distributed geospatial data requires a mechanism for selecting the data sources and performing data processing operations on the selected sources efficiently. We describe a semantics-based information integration approach that uses a spatio-temporal semantic model to define the geospatial information content of the sources, employs a conflict resolution ontology to resolve semantic heterogeneity, and uses geospatial metadata to help the users evaluate usefulness of the available data sources. We show how the captured metadata can be used for efficient query planning. Based on our proposed approach, we are developing GeoCosm, a web-based prototype that would help integrate autonomous distributed heterogeneous geospatial data.