Dynamic ship routing through stochastic, spatially dependent ocean currents

Although potential benefits of ocean current routing could be substantial, available current information is inadequate to provide for optimal routing. Therefore, we develop a prototype system which utilizes satellite altimetry to detect ocean currents in near real-time for improved routing of commercial vessels. Potential problems in this system include measurement errors, inaccuracies of geoid models, temporal and spatial sampling characteristics, and time lags between information collection and utilization. Although the measurement errors and spatial sampling seem insignificant, the limitations of the gravimetric geoid models and the combined spatial and temporal sampling seem more troubling. We tested two other geoid approaches, the hydrographic and the synthetic geoid approaches. Their performances seem promising, though each has its own limitations. The results of other tests suggest the need of a spatial-temporal interpolation scheme to deduce synoptic information.; Finally, results indicate that the effects of time lags are significant. The dynamics of current patterns makes 7-day old information virtually worthless. This fact suggests the need to provide timely information or to develop a routing algorithm that improves the performances of old information. To utilize current information, we developed a minimum fuel dynamic programming procedure which allows for heading and power controls. Due to the computational burden of this formulation, we developed two simpler heuristics, which decompose the combinatorial optimization into separate optimizations. Tests of these heuristics through current patterns in the "Harvard Gulf Stream area" showed that the simpler heuristic performed comparably with the more complex formulation.; To handle the uncertainty of current information introduced by the time lag, we developed a stochastic dynamic programming algorithm which explicitly incorporates such uncertainty. We modeled this uncertainty with a one-stage spatial dependency property, and estimated the associated conditional transition probabilities by using historical data. Under different time lag situations, the stochastic algorithm performed consistently better than its deterministic counterpart for "with-current" voyages. For "against-current" voyages, the stochastic algorithm introduced little marginal benefit over the deterministic one.