Improved statistical prediction of surface currents based on historic HF-radar observations

Accurate short-term prediction of surface currents can improve the efficiency of search-and-rescue operations, oil-spill response, and marine operations. We developed a linear statistical model for predicting surface currents (up to 48 h in the future) based on a short time history of past HF-radar observations (past 48 h) and an optional forecast of surface winds. Our model used empirical orthogonal functions (EOFs) to capture spatial correlations in the HF-radar data and used a linear autoregression model to predict the temporal dynamics of the EOF coefficients. We tested the developed statistical model using historical observations of surface currents in Monterey Bay, California. The predicted particle trajectories separated from particles advected with HF-radar data at a rate of 4.4 km/day. The developed model was more accurate than an existing statistical model (drifter separation of 5.5 km/day) and a circulation model (drifter separation of 8.9 km/day). When the wind forecast was not available, the accuracy of our model degraded slightly (drifter separation of 4.9 km/day), but was still better than existing models. We found that the minimal length of the HF-radar data required to train an accurate statistical model was between 1 and 2 years, depending on the accuracy desired. Our evaluation showed that the developed model is accurate, is easier to implement and maintain than existing statistical and circulation models, and can be relocated to other coastal systems of similar complexity that have a sufficient history of HF-radar observations.