Monitoring lakes, rivers, and oceans is critical to improving our understanding of complex large-scale ecosystems. In this work, we develop and analyze three path planning algorithms for underwater robots to optimize sensing in conjunction with networks of underwater sensors. The algorithms require different levels of knowledge about the environment: global, local, and decentralized control of the robot by the sensor network. We find our global Voronoi approach produces paths that are typically best for sensing, but are longer, which can be problematic if the robot has limited endurance. The local algorithm, inspired by Tangent Bug, produces paths that are usually shorter while still having good sensing. The decentralized controller also has good sensing and short paths and has the advantage that it can also adapt the depths of the underwater sensors to jointly optimize the sensor network and robot sensing and the robot path length. The drawback is the somewhat higher communication and processing requirements. For each of these algorithms we perform a detailed analysis and comparison in simulation. We identify limitations of each and provide framework for future improvements.