集合调整Kalman滤波同化模块的建立及其在海洋和气候系统模式中的应用

发布时间：2018-08-27 12:27

【摘要】：海洋模式和气候系统模式在实际应用过程中往往存在较大的偏差,亟需利用较为成熟的数据同化方法在数值模拟过程中有效结合观测信息,得到更合理的模拟结果或通过改进初值场提高预测的精度。在目前海洋与气候研究的主要的数据同化方法中,集合调整Kalman滤波(EAKF)同化方法不需要扰动观测,可以充分保留数值模式的先验信息,其计算和存储方面的需求也相对较少,适合用于开展海洋和气候系统模式的数据同化。本文从方法实现的角度详细阐述了EAKF方法的基本理论和相关假定,讨论了EAKF同化方法的串行实现、并行实现、观测数据处理和集合样本处理等过程,建立了EAKF同化模块,随后将其应用于区域海洋模式、全球海洋模式和海气耦合模式中。本文首先在基于POM建立的西北太平洋环流模式中,通过EAKF同化模块开展了2005年到2009年的Argo资料的集合滤波同化实验。为对比分析区域海洋模式数据同化的效果,本研究设计了3组数值实验：控制实验(单模式运行,无数据同化)、集合自由发散实验(集合运行,无数据同化)和EAKF同化实验(集合运行,Argo数据同化)。该区域海洋模式将不同年份的初始场作为2005年不同模式样本的初始场,实现集合模式运行,开展集合自由发散实验和EAKF同化实验。自由发散实验的集合样本分布情况的分析表明：由于模式对初始场的适应过程,集合模拟结果在开始几个月内出现集合样本分布有所减少,但随后稳定在一定范围。这种构造集合初始场的方法应用在区域模式中,所有集合样本具有一定的发散性,可用来开展准确的集合滤波同化。经过EAKF同化后的集合样本分布相比无同化的自由发散实验略小,但仍保持了一定的量值,对后续的滤波同化过程不会造成不良影响。通过分析SST的集合模拟结果相对特定参考点的相关系数,模式背景误差协方差表现出较强的各向异性特征。为考察同化的效果,所有实验的结果与卫星观测SST、GTSPP温盐剖面数据和卫星高度计观测数据等进行了细致的对比。误差统计结果显示：同化结果相对卫星SST的误差比同化前在整体上减少,平均减少量为10%左右：相比独立于Argo数据的GTSPP温盐剖面观测,同化后的温度和盐度误差比同化前均有大幅减小,相对控制实验和自由发散实验的误差减少最大百分比分别达到85%和80%。同化前后的模拟结果与卫星高度计观测海面高度数据的对比显示：同化过程还增强了模式对海洋中尺度活动能力模拟能力,这一改进在黑潮及其延伸体附近,以及10°N断面上尤为突出。在基于MOM4建立的全球大洋环流模式中,开展了2008年的Argo浮标数据的EAKF同化,对比分析了4组同化实验与控制实验(未同化)的实验结果。初步同化实验(实验1)中初始温度场的扰动采用了1.0℃的振幅对上层海洋进行扰动,且无集合样本扩展,其实验结果表明：通过Argo数据同化后的温度(盐度)在上400m(500m)水层偏差显著减小,然而这些偏差在更深水层增大;SST的误差在上半年的减小值明显高于其余时段。为了考察同化的改善作用在不同深度和不同时段的差异,本文设计了3个敏感性实验。其中2个实验用于分析不同垂向扰动的敏感性：扰动深度(实验2)和扰动振幅(实验3)。实验2采用了整层水柱的扰动,扰动振幅仍为1.0℃,实验结果表明：模拟温度、盐度的偏差在整个水体中均得到减小。实验3采用较小扰动振幅(0.1℃),相比实验2说明合适的扰动振幅也非常重要。实验4采用了集合样本扩展,其扩展系数则是通过一系列的数值实验的敏感性分析所得到的5%。与其它3个实验相比,实验4的同化性能有了较大的提高。综合上述实验结果,我们认为：对于初始场的扰动应考虑模式的所有分层：合适的扰动振幅对EAKF同化具有重要作用;最优集合样本扩展系数的选择有助于提高EAKF同化的效果。基于国家海洋局第一海洋研究所地球系统模式(FIO-ESM),采用微扰动法构建了集合初始场,开展了卫星SST和SLA等数据的EAKF同化实验。本研究利用气候系统模式数据同化后的海洋模式、大气模式、海冰模式、陆面模式和海浪模式等分量模式的同化结果重构了1992-2013年的气候再分析数据,并从整体上对重构的再分析数据进行了评估。本研究采用了ERA-Interim再分析数据集、EN4温盐再分析数据集、GPCP降水数据集、AVISO沿轨道观测海浪有效波高等多种数据,对FIO-ESM同化再分析数据行了对比分析,结果显示：重构的再分析数据可以成功再现1992-2013年间上层海洋、大气运动和水汽分布、海冰变化、海浪气候态分布等方面的气候特征。在进一步研究中,该数据将用于开展气候分析和未来气候预测,提高我们对气候变化的认知水平
[Abstract]:There are often large deviations between ocean model and climate system model in the actual application process. It is urgent to use more mature data assimilation method to effectively combine observation information in the numerical simulation process to obtain more reasonable simulation results or to improve the prediction accuracy by improving the initial value field. In data assimilation method, ensemble-adjusted Kalman filter (EAKF) assimilation method does not need perturbation observation, can fully retain the prior information of numerical model, and its calculation and storage requirements are relatively small. It is suitable for data assimilation of ocean and climate system models. The basic theory and related hypotheses are discussed. The serial implementation, parallel implementation, observation data processing and set sample processing of EAKF assimilation method are discussed. The EAKF assimilation module is established and then applied to the regional ocean model, the global ocean model and the air-sea coupling model. In order to compare and analyze the effect of regional ocean model data assimilation, three groups of numerical experiments were designed: control experiment (single mode operation, no data assimilation), aggregate free divergence experiment (aggregate operation, no data assimilation) and EAKF experiment (aggregate operation, no data assimilation). Assimilation experiment (collective operation, Argo data assimilation). The regional ocean model takes the initial field of different years as the initial field of different model samples in 2005, realizes the collective mode operation, carries out the collective free divergence experiment and EAKF assimilation experiment. The results of ensemble simulation show that the distribution of ensemble samples decreases in the first few months, but then stabilizes in a certain range. This method of constructing the initial field of ensemble is applied to the regional model. All the ensemble samples have certain divergence and can be used to carry out accurate ensemble filtering assimilation. The sample distribution is slightly smaller than that of the non-assimilated free-divergence experiment, but it still keeps a certain amount of value, which will have no adverse effect on the subsequent filtering assimilation process. As a result, all the experimental results were compared with SST, GTSPP temperature and salinity profiles and satellite altimeter data in detail. The error statistics showed that the error of the assimilation results was less than that of the satellite SST before assimilation, and the average reduction was about 10%. Compared with the GTSPP temperature and salinity profiles independent of Argo data, the error of the assimilation results was less than that of the satellite SST before assimilation. The results show that the temperature and salinity errors of the assimilated model are greatly reduced compared with those of the pre-assimilated model, and the maximum errors of relative control experiment and free divergence experiment are 85% and 80% respectively. In the global oceanic circulation model based on MOM4, EAKF assimilation of Argo buoy data in 2008 was carried out, and the results of four groups of Assimilation Experiments and control experiments (not assimilated) were compared and analyzed. The results show that the deviations of temperature (salinity) in the upper 400 m (500 m) water layer after assimilation of Argo data are significantly reduced, but these deviations increase in the deeper water layer; the deviations of SST in the first half of the year are significantly higher than those in the deeper water layer. In order to investigate the difference of the improvement effect of assimilation at different depths and different periods, three sensitivity experiments were designed. Two of them were used to analyze the sensitivity of different vertical disturbances: disturbance depth (experiment 2) and disturbance amplitude (experiment 3). In experiment 2, the disturbance of the whole water column was used, and the disturbance amplitude was still 1.0 C. The results show that the deviation of salinity decreases with the simulated temperature in the whole water body. In experiment 3, a small disturbance amplitude (0.1 C) is used, and the appropriate disturbance amplitude is also very important compared with experiment 2. In experiment 4, the set sample expansion is used, and the expansion coefficient is 5% obtained by a series of numerical experiments. Comparing with the three experiments, the assimilation performance of Experiment 4 has been greatly improved. Based on the above experimental results, we consider that all the layers of the model should be considered for the initial field perturbations: the appropriate amplitude of the perturbation plays an important role in EAKF assimilation; the selection of the optimal set of sample expansion coefficients helps to improve the effect of EAKF assimilation. The Earth System Model (FIO-ESM) of the First Institute of Oceanography, Jiahai Oceanic Administration, was used to construct a set of initial fields and to carry out EAKF Assimilation Experiments of satellite SST and SLA data. In this study, the assimilation of ocean model, atmospheric model, sea ice model, land surface model and ocean wave model was carried out using the assimilated data of climate system model. Results The climate reanalysis data from 1992 to 2013 were reconstructed and the reconstructed reanalysis data were evaluated as a whole. The data of ERA-Interim reanalysis data set, EN4 thermohaline reanalysis data set, GPCP precipitation data set, AVISO along-track observation of wave effective wave height were used in this study. The results show that the reconstructed reanalysis data can successfully reproduce the climatic characteristics of the upper ocean, atmospheric movement and water vapor distribution, sea ice changes, and wave climatic distribution during 1992-2013. Knowledge level
【学位授予单位】：中国海洋大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：P73;P435

【相似文献】