Autonomous underwater vehicles have been designed to make measurements, relay information, or perform tasks in a critical, inhospitable environment. Both large vehicles (transport-class platforms) and small vehicles (network-class platforms) have a role in exploring and working in the ocean. Historically, large vehicles have often been designed to address specific missions. Recently, small vehicles, enabled by advances in technology, are evolving more generically to address measurement of a variety of ocean state variables. Mapping ocean variability today and forecasting it tomorrow requires coupled modeling-sampling systems. In advanced nowcasting/forecasting systems, modeling and sampling are intimately linked by design. Sampling initializes a model which then produces three-dimensional fields used to guide further sampling. The feedback in this adaptive approach is tuned to minimize forecast error.