In this thesis, I explore integration strategies in the visualization of multifaceted spatial data. First, I chart the design space of visual and functional integration in the scope of a taxonomy. Visual integration describes how representations of the different data facets can be visually composed. Functional integration, describes the various ways in which events can be coordinated across the visually integrated representations. In the second part of this thesis, I present contributions to the field of visualization in the context of concrete integration applications for exploration and presentation scenarios. Thereby I first address a set of challenges in the scope of a decision making scenario in lighting design. In this context, I propose LiteVis, a system that integrates representations of the simulation parameter space with representations of all relevant aspects of the simulation output. ^The integration of these heterogeneous aspects together with a novel ranking visualization are the key to enabling an efficient exploration and comparison of lighting parametrizations. In presentation scenarios, on the other side, the generation of insights often cannot rely on user interaction and therefore needs a different approach. The challenge is to generate visually appealing, yet information-rich representations for mainly passive observation. Integration can thereby be applied as a substitute for interaction. In this context, I address two different challenges in the domain of molecular visualization. First, I investigate how to convey relations between different representations of a molecular data set, such as a virus. I thereby propose a novel technique for creating transitions between representations that are re-usable for different data sets, and can be combined in a modular fashion. ^The second challenge concerns the presentation of transitions between development states of molecular models, where the actual biochemical process of the transition is not exactly known or it is too complex to represent. In order to overcome a potential indication of false relationship information, a technique is introduced that proposes the continuous abstraction to the lowest common denominator in terms of available visual detail, at which the relationship between two models can be accurately conveyed. Finally, I reflect on the taxonomy in respect to the presented techniques. The results demonstrate that integration is a versatile tool in overcoming key challenges in the visualization of multifaceted spatial data.