Executing high-level tasks in a fetch-delivery-scenario (e.g.,“Robot, bring me a muffin!”) requires a robot to reason about what the desired object is and where it is located. At best, the underlying map to plan and execute such tasks allows for a close linkage of geometric and semantic information and efficient look ups. Geographic Information Systems combine relational and geometric data in one database and can serve as powerful backends for semantic mapping, because of their in-built ability to handle spatial queries. In this paper, we present a GIS-based approach to semantic mapping that allows semantic labels to be integrated closely with geometric environment maps. We show that our semantic map is suited to store and maintain information that supports complex task planning, as well as other tasks. As an example, we show how robot navigation can be improved by including semantic information.