| People has paid more attention to the global climate change in recently years, thus to improve numerical modeling ability on climate change and to evaluate its influence on human living become significantly important. As a crucial role in the issue, more researchers are focusing on marine ecosystem modeling and more robust models are urgent in demand. Under such circumstances, the number of model dimensions, application area and model complexity are undergoing quickly development. However, in most previous studies, the spatial variations of key parameters are always ignored or not taken into account at all, and researchers always assumes that a set of optimized parameters is suitable for the whole study area. It results in big spatial differences between modeled values and the observations, and these differences can not be effectively reduced by improving physical background or adding new ecological mechanisms, especially when modeling on basin or global scales. To solve the problem, spatially variable parameters are needed. Traditional ways are optimizing local parameters or estimating parameters in each sub-area independently, and then smoothing the parameter values. In such a way, spatially variable parameters are obtained, but the model and the observations are not taken as a whole and the parameters are not optimized globally, thus these spatially variable parameters can not represent the reality accurately. In this study, a new method (spatial parameterization) is utilized to realize spatial variations of the parameters. The article tested if the new method can help us improve marine ecosystem modeling.In this effort, a simple 3-D marine ecosystem model and the corresponding adjoint model are constructed on a global scale under climatological physical environment provided by FOAM. The numerical study and assimilation experiments are implemented within a depth of 500m. A relaxed term of nutrient in the bottom layer is designed to prevent nutrient from dissipation, which insures the positive simulation.Five uncorrelated and sensitive parameters are selected as control variables by a conventional sensitivity analysis and investigating the gradients of the cost function with respect to each parameter, respectively. In twin experiments, model-generated data are served as fictitious observations, and they are assimilated into the model to estimate the control variables. Firstly, the spatially-constant control variables are estimated, which validates the adjoint model. Furthermore, based on spatial parameterization, the spatially varying control variables are estimated by the adjoint method, which indicates the validity of spatial parameterization and the feasibility of estimating spatially varying parameters by the adjoint method.In real experiments, SeaWiFS chlorophyll-a data are assimilated into the model based on the adjoint method on a global scale. After spatial parameterization is utilized, the mean error of phytoplankton in the surface layer (nitrogen) reduced to 0.0584 mmolN·m-3, which is a reduction of 62.4% compared with the assimilation experiment without spatial parameterization. The assimilation results show that spatially varying parameters are more reasonable than a set of constants, since the selected parameters exhibit great spatial variations. By utilizing spatial parameterization, most regional features in SeaWiFS imagery can be reproduced. The assimilation results are sensitive to the amount of independent grids and the selected influencing radius: the more independent grids we use, the better the assimilation results are; the influencing radius should be suitable for the configuration of the independent grids, since either big or small influencing radius is unfavorable for the assimilations. Because of the deficiency of the model (iron-limitation is not considered), the model fails to simulate high nitrate zones in southern ocean, which indicates that when modeling on a global scale, some important mechanisms such as iron-limitation should be included in.
|
PDF Full Text Request