Study On Four - Dimensional Variational Assimilation Of Three - Dimensional Storm Surge Model In Bohai - East China Sea

Based on Regional Ocean Modelling System(ROMS) with its four-dimensionalvariational (4D-Var) data assimilation modules, a model is developed for storm surgesimulation in the East China Sea. The4D-Var assimilation technique is able to compensatefor the errors between forecasts and observations by integrating the tangent linear model andthe adjoint model, as well as using the conjugate gradient algorithm. During the assimilationwindow, the dynamics is consistent and persistent.Tides are simulated by using terrain following bottom friction coefficients, and thecotidal charts of four main constituent(M2, S2, K1, O1) are analyzed. Results show that themodel has fair ability to tide simulation, as both crossover points’ location and cotidal linesare consistent with the previous study. Also, most of the tidal station simulation results are ingood agreement with observations.Two extratropical storm surges are selected as the study cases. By assimilating theobservation from tidal stations into the model, the simulation results are improved to a largeextent. Five groups of twin sensitivity tests are designed on the factors of the simulationresults. Results indicate that wind stress plays a decisive role in the storm surge simulation,while the initial field only affects the simulation in a short period, the greater the wind blows,the shorter time the initial field affects. So taking the initial field and the wind stress as thecontrol variables, assimilation can improve the simulation accuracy of storm surge, not onlywhere the stations are located, but also the place around them. In the assimilating process,iteration is a very important parameter, which determines the decline degree of the costfunction. The cost function describes the discrepancy between model results and observations.As the iteration increases, the assimilation results will be better coincide with theobservations, at the same time, more computational cost is needed.After a period of assimilating, an optimal initial field consistent with the observationsand the inverted optimal drag coefficients are obtained. Then, storm surges are forecasted byusing the optimal drag coefficients, and original ones respectively. It turns out that whetherthe optimal initial field improves the accuracy of the forecasting results is case-dependent. The falling surge case does not see clear improvement, likely because the physicaldescription of falling surges is deficient, and facing the strong wind, the error is significant.While in the rising surge case, the forecasting results have a good improvement from theinitial time to the observed peak time. It means the optimal initial field do good to thenowcasting.Whether inverted optimal drag coefficients can further improve the forecasting quality,depends on two aspects. One is the wind error between the assimilation window andforecasting window. If the error varies greatly, the suitability of optimal drag coefficientsmust become bad. The other aspect is the range of drag coefficient variation under differentwind speed condition. That is to say, if the drag coefficients vary widely, the suitability of theoptimal drag coefficients will also be challenged.