Chasing After the Wind: Flow Structure Detection Strategies for Autonomous Mobile Flow Field Measurements

Modern flow measurement technology enables studies of fluid motion that, half a century ago, would have seemed unfathomable. However, despite staggering capabilities, measuring many natural flows in the field remains challenging. In particular, resolving coherent flow structures within physical scales ranging from meters to kilometers is not readily achieved. This dissertation proposes autonomous mobile flow field measurements (AMFM) as a paradigm for expanding flow field measurement capabilities into this range of scales. In the AMFM framework, a mobile platform such as a drone would identify critical flow structures and follow them autonomously as they evolve; the device would be taught, in a sense, to chase after the wind for the sake of measuring it. The greatest theoretical challenge to AMFM is that of flow structure detection: what, after all, should be identified in the flow? How is it to be measured? Answering these questions is the overarching motivation of this dissertation. In response, two principal contributions are developed. The first is a theoretical approach to gradient estimation labeled Lagrangian gradient regression (LGR), which enables instantaneous and finite-time flow gradients to be approximated from sparse flow observations. The second is a semantic approach to flow measurement, which provides the ability to discern fluid motion from complex natural images using arbitrarily defined flow tracers. Together, these tools enable a range of studies which would be difficult to conduct otherwise. To demonstrate their combined ability, two experiments are performed. The first examines the motion of imperfect surface tracers measured by the proposed methods relative to sub-surface flows measured by conventional techniques. The second experiment analyzes flow features in the Caltech turtle ponds using only tracers naturally occurring on its surface. While it is demonstrated that the methods and results obtained in this work are meritorious in their own right, they also provide a framework from which future AMFM technologies can be built.