Optimizing sensor placement for environmental monitoring

Abstract

Nowadays, environmental monitoring is playing an increasingly important role in providing not only timely information to the public for decision making, but also accurate data for health studies. However, very few monitoring stations have been installed even for big urban cities due to its high cost. In addition, different goals will result in different sensor placements. Therefore, it is important to optimize the sensor placement to achieve specific goals. In this thesis, we propose new sensor placement strategies for environmental monitoring with different objectives, including the following topics:•When the complete field information of interest can be obtained, we propose greedy approaches to find: (1) the number and locations of sensors required for achieving a certain accuracy goal; (2) the achievable accuracy given a fixed budget of sensors available. We also propose a similarity based clustering approach when the obtained similarity matrix satisfies a certain structure, and provide the conditions for guaranteed recovery under a Gaussian model. When some data from a pilot deployment is available and a probabilistic Gaussian Process model can be built with the data, we propose a multi-type sensor placement scheme which extends the commonly used single-type sensor placement scheme that aims to maximize information gain. The proposed approach can better handle situations for complex phenomena such as air pollution that involve multiple variables to be monitored at the same time.