Wave Data Assimilation Using Support Vector Regression (SVR) Model and Ensemble Kalman Filter (EnKF)

The prediction of wind driven ocean waves is of primary importance for the safety of shipping and off-shore operations, as well as for scientific studies of, e.g., sediment transport and ocean-atmosphere interaction. Traditionally, wave models do not explicitly use the wave observations to estimate the present sea state: the only input to the models is a sequence of wind fields from a meteorological model. However, it is obvious that the model estimate of both the present and the future sea state can be improved if all available knowledge is combined, which can be done by assimilation of the observations into the model. In the present study a new approach for assimilating wave measurements is presented. A soft computing method namely Support Vector Regression (SVR) is used as a surrogate model to simulate wave heights. This model is trained with a data set of wind and wave measurements and is capable of predicting wave characteristics with relatively good quality. In the other hand, swells and short duration storms cannot be well modeled by the SVR model. Therefore assimilating the wave height measurements into the SVR model using an Ensemble Kalman Filter (EnKF) was utilized to ensure better efficiency of the model. A pretty complicated Matlab and Shell scripts were modified in order to establish appropriate data assimilation (DA) system. Using two data sets from NDBC buoys in different geographical zones and performing statistical comparisons showed that the assimilation scheme can reduce the errors in predicting swells, storm peaks and also the storm duration and time. Also it was seen that the combination of SVR model and the EnKF method can be easily used for producing high quality wave forecasts.