LAUSR

The prediction of water current (WC) is key in understanding oceanographic phenomena, and serves as a corner stone for maritime applications. While numerical models are capable to characterize the main phenomenon effecting the current, a good estimate of the current’s velocity requires accurate environmental information such as topographic maps, temperature, and wind velocity. Without such information, common practice involves deploying a large number of drifting devices to measure the WC in good resolution. In this paper, we purpose a statistical approach to predict the WC’s velocity field, namely, the relation between the drifters’ locations and their drifting velocity to estimate the velocity of the WC across a given environment. We offer two statistical approaches. Assuming the above relation is linear, the first uses a weighted least squares to estimate the parameters of the velocity field. The second uses a support vector regression to predict the WC by a classifier. The first approach is more traceable, while the second can manage a non-linear space-speed relation of the WC. Results are studied numerically using an WC model, and experimentally from a sea trial performed across the shores of San Diego including 13 acoustically localized submerged drifters. In both cases, performances show a good agreement between the ground truth of the WC and the predicted one.