A two-layer pollutant advection-diffusion model is built to investigate the pollutant transport in the sea area around Maidao off Qingdao coast. An adjoint data assimilation technique is applied to estimate the optima...A two-layer pollutant advection-diffusion model is built to investigate the pollutant transport in the sea area around Maidao off Qingdao coast. An adjoint data assimilation technique is applied to estimate the optimal values of the model parameters. The experimental results on the initial field of pollutant indicate that the distribution of Chemical Oxygen Demand (COD) concentra- tion is sensitive to the horizontal eddy diffusivity. An appropriate value of horizontal eddy diffusivity is necessary in order to depict the influence of the initial field precisely, and it is also essential to the simulation of the advection-diffusion process of the pollutant. By inversion of the model parameters and optimization of the initial COD concentrations, the simulation results are improved sig- nificantly. The cost function is reduced to 40% of its first step value. The average misfit between the model outputs and the observa- tions in the upper layer decreases from 0.46 to 0.25 mg L-1, and that in the lower layer decreases from 0.22 to 0.14 mg L-1.展开更多
文摘A two-layer pollutant advection-diffusion model is built to investigate the pollutant transport in the sea area around Maidao off Qingdao coast. An adjoint data assimilation technique is applied to estimate the optimal values of the model parameters. The experimental results on the initial field of pollutant indicate that the distribution of Chemical Oxygen Demand (COD) concentra- tion is sensitive to the horizontal eddy diffusivity. An appropriate value of horizontal eddy diffusivity is necessary in order to depict the influence of the initial field precisely, and it is also essential to the simulation of the advection-diffusion process of the pollutant. By inversion of the model parameters and optimization of the initial COD concentrations, the simulation results are improved sig- nificantly. The cost function is reduced to 40% of its first step value. The average misfit between the model outputs and the observa- tions in the upper layer decreases from 0.46 to 0.25 mg L-1, and that in the lower layer decreases from 0.22 to 0.14 mg L-1.
