Value of information analysis for efficient sampling
This presentation describes methods for sampling (targeted data-collection) in spatial domains. By carefully designing (adaptive) sampling surveys, one can gather valuable data about the uncertain ocean environment. In the end, such information should be useful for making improved decisions related to oceanographic resources. We limit scope to sampling by autonomous underwater vehicles, and focus on temperature and salinity variability. Rather than framing a particular decision problem, we discuss different utility functions and show new approaches for improved classification of the excursion set for low salinity and low temperature (cold freshwater). A spatial Gaussian process model is used, allowing fast computation of excursion probabilities conditional on data, and on expectation - for design purposes, before the data is gathered.