Advances in sensing hardware and personalized computing devices (e.g. smartphones) are not only major drivers behind "big-data'', but they are also changing the face of computation. Sensors and mobile devices, such as smartphones and tablets, have made it possible for users to compute on the go. Not only generating vast amounts of data, but also appending a location to each interaction. This shift to mobile computing provides the necessary context to relate `data' to interactions and processes in the physical world like never before.

Despite becoming more cost-effective and widely integrated, data collected from many sensors remains noisy. Computing with this data, despite its uncertainty, is a complex, but necessary task in order to provide robust real-world software systems. In this thesis, I introduce novel techniques for managing the problem of data uncertainty based on the idea of exploiting spatial relationships between data objects.

I first introduce two different scenarios and show how this idea of spatial similarity can be used to improve the data management process. Both methods provide a more accurate representation of the (uncertain) data in a more efficient manner than current state-of-the-art techniques. Then, I perform a statistical study of several real-world spatial networks, showing how relationships between entities change with distance and other features of the local network structure. By studying the patterns when relationships between objects is observed, I work backward to infer the mechanisms which affect these relationships.

The ideas and techniques introduced in this thesis provide a first step toward effectively managing data uncertainty in the new era of mobile computing by exploiting spatial relationships in an efficient manner.