【作者】 王春晖；

【导师】 吕咸青；

【作者基本信息】 中国海洋大学 ， 物理海洋学， 2013， 博士

【摘要】 海洋生态系统具有典型的非线性特征，微小的扰动通过非线性作用得以放大，因此在海洋生态系统动力学数值模拟中，参数的取值可以显著影响模型的模拟结果。然而模型中的参数很难精确确定，不仅因为生态模型中某些参数之间具有很高的相关性，还因为模型中很多参数不是常数，在空间大尺度上，由于温度、光照等环境因素的不同，生态模型中的参数在不同海区的取值也不尽相同；即使是同一海区，温度、光照等环境因素也会随时间发生变化，从而导致参数的取值发生变化。在以往的研究中，参数通常取常数，不随空间、时间发生变化，导致模型的模拟结果与观测结果存在较大差异，且该问题不能通过增加模型复杂性得到解决。在本文中，首先在渤黄海建立了一个典型的三维营养盐-浮游植物-浮游动物-碎屑（NPZD）生态系统动力学模型及伴随模型，模型的背景流场由POM(Princeton Ocean Model)模式提供，只考虑背景流场对生态变量的作用，而未考虑生态变量对背景流场的反作用，根据已有的SeaWiFS叶绿素资料，利用伴随同化方法对模型中的全部12个参数进行优化，研究发现模型中的某些参数之间具有很高的相关性，且优化后的参数都具有明显的季节变化，大部分参数的季节变化可以在生物学上得到很好的解释。与参数取常数相比，随时间变化的参数可以显著提高模型的模拟能力。对生态模型中随空间变化的参数进行反演时，首先通过敏感性分析，找出模型中对模拟结果影响最大的5个参数作为研究对象。为了保证参数空间分布的连续性，使得模拟结果更加合理，选取一些网格点作为独立网格点，只需对独立网格点的参数值进行调整，其它网格点的参数值通过Cressman插值得到，利用此方法通过孪生数值实验得到了最优的影响半径。给定两种形式的空间分布，孪生数值实验表明只对模型中的一个参数进行反演时，对于给定的两种空间分布，每个参数都可以得到很好的反演，且参数的空间分布越符合实际情况越容易反演；同时对5个参数进行反演时，只有当参数空间分布的搭配与它们在生态系统中引起浮游植物生物量变化的作用相一致时，5个参数才能得到较准确的反演。实验表明海洋生态系统动力学模型中参数空间变化是合理可行的，可将此方法应用到实际问题，从而更好地模拟叶绿素等生态变量的空间分布特征。渤海是我国唯一的内海，陆源排污量很大，但水交换能力低下，一旦遭到污染将很难得到改善，因此准确模拟渤海污染物（总氮、总磷、COD等）的时空变化特征，对实现经济的可持续发展具有重要意义。对污染物的时空分布进行数值模拟时，初始场对模拟结果的影响很大，本文中将污染物当作保守物质，只考虑污染物一种状态变量的输运扩散过程，同时借鉴参数空间分布中“独立网格点”的思想，利用伴随同化方法对渤海区污染物的初始场进行反演。孪生数值实验表明：给定旋转抛物面和圆锥面两种形式的污染物初始分布，不论污染物浓度中间高、四周低还是四周高、中间低，均可得到较好的反演结果。与传统的插值方法相比，伴随同化方法可有效减少模拟结果与观测值的误差，模拟结果能更好的反映污染物的全场分布特征，验证了模型的稳定性与可靠性。最后把该模型应用到实际实验中，利用已有常规监测数据，对渤海污染物的初始分布及时间变化情况进行了准确模拟，从而求得渤海污染物的月平均分布。该方法可用于海洋环境质量的监测与评价，具有重要的现实意义。更多还原

【Abstract】 One typical feature of ecosystems is nonlinearity. Even small perturbations can beamplified through nonlinear interactions. Therefore, parameter values cansignificantly affect the model results in the simulation of marine ecosystem dynamics.However, it is very difficult to estimate parameter values accurately. Not only becauseare there very high correlations between some parameters, but also that someparameters are not constant. Parameter values in different sea areas are not same dueto different environmental factors, such as temperature, light intensity and so on. Evenwithin the same sea area, environmental factors such as temperature, light intensitymay change over time, indicating that the values of the parameter changes over time,too. In previous studies, there always has been a large discrepancy between simulationresults and observations because most researchers treated ecological parameters asconstant, and this difficult problem can not be solved by increasing the complexity ofthe model.In this paper, a typical3-dimension nutrient-phytoplankton-zooplankton-detritus(NPZD) marine ecosystem dynamical model and the corresponding adjoint model areconstructed in the Bohai Sea and the Yellow Sea. The three-dimensional PrincetonOcean Model (POM) is used to calculate the ambient physical velocities, thetemperature, and the eddy diffusivities. Only the impact of background field onecological state variables is considered while the impact of ecological state variableon background field is neglected. The variational adjoint method is applied to estimatevalues of all the12parameters by utilizing real chlorophyll data from theSea-Viewing Wide Field-of-view Sensor (SeaWiFS). The seasonal variabilities ofthese parameters are reconstructed. It looks that that some parameters are highly correlated and the time variances of most parameters seem reasonable. The simulatedresults show that using time varing parameters values accord much better with theobservations than by using constant parameters.In the study of estimating spatially varying parameters, five parameters whichhave greatest impact on simulation results are selected by a conventional sensitivityanalysis. In order to guarantee the continuity of parameter values and make thecalculated results more reasonable, several grids are selected as independent grids,and only the parameter value of these independent grids need to be optimized whilethe other grids are calculated by interpolation method. Based on this method, weconfirmed the optimal influence radius by a twin experiment. In the followingexperiments, when the five parameters are inversed respectively, spatial variations ofthe two given types can be reproduced very well. But when all the five parameters areestimated simultaneously, the collocation of the changing trend of each parameterinfluences the estimation results remarkably. Only when this collocation coincideswith the ecological mechanisms which influence the growth of the phytoplankton inthe model, can the five parameters be estimated accurately. The result demonstratesthat it is reasonable to take the spatial variation of parameters into account in theecosystem dynamical model. This method can be applied to real experiments toestimate the distributions of ecological control variables such as chlorophyll.The Bohai Sea is the only inland sea of China，and its organic pollutants aretremendous. Unfortunately, water exchange of the Bohai Sea is very weak and itsphysical self-clean capacity is poor due to its special geographical position, so it ishard to recover if the Bohai Sea is polluted. Therefore, an accurate simulation of thetime varying pollutant (e.g. total Nitrate, total Phosphate, COD) distribution is neededif we want to achieve sustainable development of economy. The initial condition hasdramatic influence on the results when simulating the time varying pollutantdistribution. In this paper, the pollution is treated as conservative substance, and onlythe transporting diffusion process of pollution is considered. The adjoint assimilationmethod is firstly applied to estimate the pollutant initial field as far as I know, and some independent grids are also selected as before to guarantee the continuity ofpollutant distribution. No matter a parabolic or conical surface the initial distributionof pollutant concentration shows, and regardless the pollutant concentration is higherin the center or not, the initial distribution can be inversed successfully. Comparedwith the traditional interpolation methods, the adjoint assimilation method caneffectively reduce the errors between simulation results and observations, which provethe stability and reliability of this model. Therefore, this model could be applied inreal experiment to simulate the initial field and the distribution of pollution in anytime step by using the regular monitoring observations, and then the monthly meandistribution of pollution can be calculated by the statistical method. The results showthat this method has important practical significance because it can be used for themonitoring and evaluation of the marine environmental quality.更多还原