Assimilation Of Water Level Data Into A Coastal Hydrodynamic Model By An Adjoint Variational Optimal Technique

A two-dimensional barotropic version of the Princeton Ocean Model (POM) is employed to simulate wind-driven low-frequency subtidal water levels along the East Coast of the United States. In this model, an orthogonal curvilinear grid system has been used, and 48-km 3 hourly ETA Data Assimilation System (EDAS) analyzed wind fields are used to force the model. The model is evaluated by comparing the simulated water levels with the observed water levels at 15 tide gauge stations, part of the National Ocean Service抯 (NOS抯) National Water Level Observation Network (NWLON). The surface wind forcing is the most important dynamic mechanism to the subtidal water level generation and variations along the East Coast of the United States except around Cape Hatteras. The response of water level to surface wind forcing is about 9 hours later. From the subtidal water level simulations, it is found that the Root Mean Square (RMS) errors vary from Scm to 15cm, and correlation coefficient between the simulated and observed water levels ranges from maximum 0.91 at Willets Point to minimum 0.54 at Cape Hatteras. Altgough this is reasonably good agreement on an overall statistical basis, it is not good enough for the model to be used in a nowcastlforecast system. A water level data assimilation system is needed to improve model nowcasts which provide the initial conditions of model forecasts. An adjoint data assimilation system is developed and used to assimilate observed coastal water level data. In this system, linear two-dimensional POM is used as forward model. Since surface wind forcing is the most important dynamic mechanism affecting subtidal water level variations along the East Coast of the United States (except around Cape Hatteras), and since errors in the wind field may be responsible for errors in the predicted water levels, the adjoint model uses wind drag coefficient as the control variable. It is used to determine gradient of cost function. Limited memory Broyden- Fletcher-Goldfarb-Shanno (BFGS) quasi-Newton method for large scale optimization is implemented to minimize the cost function. Identical twin experiments with model generated pseudo-observations are performed to verify the data assimilation system. The results of identical twin experiments show that the true solution of control variable can be recovered by assimilating pseudo-observations into model. The data assimilation system is then applied to actual subtidal water level data assimilation. The results show that the simulated subtidal water level with data assimilation is better than that without data assimilation even if only one control variable is used in data assimilation process. The results from 16 control variable experiment demonstrate that the correlation coefficients at 18 tide gauge stations are all greater than 0.93, and RMS errors are all less than 5.3cm. Thus, the simulated water levels are much improved by water level data assimilation technique. The nowcastlforecast experiments demonstrate that the average RMS of forecasted subtidal water levels over 18 tide gauge stations within 24-hour forecast circle for without data assimilation varies in the range of 8.8cm to 12cm. The improvement of subtidal water level forecasts by assimilating observed subtidal water levels at tide gauges along the East Coast of United States into the nowcastlforecast system mostly takes place within the first 6-hour water level forecasts. The impact of initial conditions created from...