Assimilation of High Frequency Radar Data into a Shelf Sea Circulation Model

Abstract

With the development of High-Frequency (HF) radar application in mapping coastal currents, more and more oceanic data assimilation efforts have been focused on coastal regions. The assimilation methods have been developed in different ways, including inverse method, optimal interpolation, variational method and Kalman filter technique. Meanwhile, the assimilation objects also have been taken into account from univariable to multivariable. Hence, it is getting increasingly important to find a feasible assimilation method to combine radar current data and sophisticate ocean models. In this thesis, the main work has been restricted to develop a feasible and operational assimilation method that can be implemented in a shelf ocean model with realistic topography and boundary conditions. The assimilation method used here is a combination of the Ensemble Kalman Filter (EnKF) and the Canadian Quick Covariance (CQC) method. As we know, the main objective of all the assimilation methods is to combine the model results and observations with a reasonable treatment of model errors. The EnKF method uses an ensemble of shortrange forecast model runs to describe the model errors. However, as we proved in the thesis, such a method can not realize real-time assimilations using high frequency observations. On the other hand, the generation of this ensemble is arduous, since this ensemble members should represent all the possibilities of model results using the assumption that the estimation of background errors is correct. A quick and easy way to get the background error covariances is the CQC method, which involves obtaining the model states from one single forecast at a fixed interval, calculating the covariances by using differences between successive output fields thereafter. Such covariances are assumed as a proxy for the model errors. Thus, we use the CQC method to substitute the background error covariances calculated from the ensemble of forecast model runs. The assumption behind this implementation is that forecast errors can be resembled by forecast tendencies. The time interval in this study is chosen according to the character of observed data, which typically is 20 minutes. As the tidal forcing is included in the model, this time interval is reasonable for resolving the tidal phase error. Our numerical experiments show that this newly developed method is appropriate for real-time assimilation of HF observations for shelf ocean model. With implementation of this assimilation scheme the model also provides more realistic results for shelf currents, as well as for temperature and salinity distributions. With the traditional EnKF method, the posterior variance keeps decreasing with time, consequently a divergence will happen and the analysis will ignore the observations at the end. Compared to the traditional EnKF method, the newly developed method proves that the ensemble is really flow-dependent. No matter whether with a linear model or a complex nonlinear model, the newly developed method is consistent in its performance. To summarize, we developed a real-time and effective assimilating scheme, which is suitable for operational ocean modeling using HAMSOM for instance, as a fully non-linear model system including tide and baroclinic effect as well as the dynamics at shelf slope.