杭州湾三维悬浮泥沙输运模型初始场的伴随法反演研究

In the numerical simulation of suspended sediment transport, accurate setting of initial condition is important. Available methods of initial condition determination have some defects, and thus the accuracy of initial condition needs to be improved. Based on a three-dimensional suspended sediment transport model with the adjoint data assimilation, adjoint method was used to optimize initial condition of the model by running twin experiments and practical experiments. In the twin experiments, the influence of errors of initial condition on the simulation results was studied. Then, the sensitivity of inversion results of initial condition to different influencing factors such as optimization algorithms, initial guess values, number of satellite remote sensing data, window width of assimilation time, and errors of background flow field were discussed. At last, the ability to reconstruct initial condition with adjoint method and interpolation method was compared. The results of the twin experiments show that, the steepest descent algorithm was better than three conjugate-gradient algorithms and limited-memory Broyden-Fletcher-Goldfarb-Shanno algorithm in optimizing initial condition. The inversion results are less sensitive to initial guess values, number of satellite remote sensing data, and errors of background flow field, but sensitive to the width of assimilation window. Compared with interpolation methods, the adjoint method was a more effective method of reconstructing initial condition of a model. In the practical experiments, the surface suspended sediment concentration data obtained from the GOCI satellite remote sensing data in typical neap and spring tide periods in Hangzhou Bay were used to optimize the initial condition. Results of the practical experiments show that after data assimilation, a more realistic optimal initial condition could be obtained, indicating that the adjoint method was an effective tool for initial condition optimization. This study provides references for improving initialization schemes of suspended sediment transport model, and for the initialization schemes of other numerical models.