Effective Benthic Surveying with Autonomous Underwater Vehicles

The finite velocity and battery life of an AUV impose constraints on the extents of surveys and the spatial density of data recorded. Using a Gaussian process approach, we develop a method for quantifying the survey error resulting from spatial undersampling of the sample field. We also show how the Gaussian process model can be used to predict the information gain from a proposed AUV action. These techniques are demonstrated using a real world data-set collected during deployments at Ningaloo Marine Park, Western Australia.