集合同化方法在海浪同化中的试验
发布时间：2018-11-04 16:25
【摘要】：目前，集合同化方法在海浪同化中还未得到全面深入的应用。本文基于第三代海浪数值预报模式WAVEWATCH III，设计了全球区域海浪同化系统，开展了集合最优插值（Ensemble Optimal Interpolation，简称EnOI）三天海浪预报及长期的同化试验，并与最优插值（Optimal Interpolation，简称OI）同化结果进行了比较。发现，EnOI同化方法在改善海浪预报中起到了很好的作用。考虑到EnOI中的历史样本会夸大背景误差且引起虚假相关，为解决这一问题，本文又首次尝试了通过在风场叠加随机扰动的方法生成一组动态样本，评估了历史样本与动态样本的优劣，为今后继续开展集合卡尔曼滤波（Ensemble Kalman Filter，简称EnKF）同化的研究做好了准备。 主要开展了以下工作： （1）背景误差信息在资料同化中是至关重要的，在开展同化前须对模式误差有充分的认识。为此，首先对模式进行了10年的积分，通过NDBC浮标数据及Jason-1高度计资料对模式模拟结果的检验，发现模式对全球海浪模拟效果较好，并得到了有效的模式误差，为后面同化工作中背景误差协方差矩阵的构建及集合样本的选取提供了依据。 （2）由于在全球范围内同化观测点资料较多，且模式网格设置较粗，为节省计算时间，对比分析了四种同化观测点选取方案。发现观测稀疏后并不会明显削弱同化效果，采用五点平均的减薄方案滤去高频扰动，能够在不影响同化效果的前提下缩短同化计算时间。 （3）为了评估OI、EnOI同化在短期海浪预报中的改进效果，，开展了三天海浪预报同化试验。发现，数据同化能够很好地订正初始场的偏差，并对初始化过程及三天预报过程有明显改进。EnOI对模式的改进效果在时间序列上更为稳定，对于36小时以内的预报，采用EnOI同化方案能够获得比OI更理想的同化效果。 （4）为了进一步考察EnOI在长时间序列上对海浪预报的同化效果，设计了为期一年的同化试验。发现与OI相比，EnOI具有绝对的优势。采用EnOI同化方案后，全球海浪有效波高预报绝对误差小于0.5m的概率达到83.79%，小于1m的概率高达96.03%，预报精度非常可观。 （5）由于EnOI是通过预先存储好的历史样本来估计背景误差，在积分过程中始终保持不变，这往往会夸大背景误差并且引起较长时间尺度上大范围的虚假相关。为解决这一问题，通过设计多组敏感性试验继续探讨了EnKF初始样本生成的最佳方法，并将其与EnOI历史样本做了比较。发现，相对于历史样本，扰动样本能够较好的呈现出模式误差的结构和相关性。
[Abstract]:At present, the ensemble assimilation method has not been fully applied in ocean wave assimilation. In this paper, a global regional ocean wave assimilation system is designed based on the third generation wave numerical prediction model (WAVEWATCH III,). Three days of ocean wave prediction and long-term assimilation experiments are carried out with the ensemble optimal interpolation (Ensemble Optimal Interpolation, (EnOI), and the results are compared with the optimal interpolation (Optimal Interpolation,. OI) assimilation results were compared. It is found that EnOI assimilation method plays a good role in improving wave prediction. Considering that historical samples in EnOI exaggerate background errors and cause false correlation, in order to solve this problem, this paper first attempts to generate a set of dynamic samples by superimposing random disturbances in wind field. The advantages and disadvantages of historical and dynamic samples are evaluated, and the preparation for further research on EnKF assimilation based on ensemble Kalman filter (EnKF) is made. The main work is as follows: (1) background error information is very important in data assimilation. For this reason, the model is first integrated for 10 years, and the model simulation results are tested by NDBC buoy data and Jason-1 altimeter data. It is found that the model has a good effect on the global ocean wave simulation, and the effective model error is obtained. It provides a basis for the construction of background error covariance matrix and the selection of set samples in the later assimilation work. (2) in order to save calculation time, four assimilation observation point selection schemes are compared and analyzed because there are more data of assimilation observation points in the global scope and the model grid is coarse. It is found that the assimilation effect will not be significantly weakened after the observation is sparse, and the calculation time of assimilation can be shortened without affecting the assimilation effect by filtering the high-frequency disturbance by using the five-point average thinning scheme. (3) in order to evaluate the improved effect of OI,EnOI assimilation in short-term wave prediction, a three-day wave prediction assimilation experiment was carried out. It is found that the data assimilation can correct the deviation of the initial field well, and improve the initialization process and the 3-day prediction process obviously. The improved effect of EnOI on the model is more stable in time series, and is more stable for the prediction within 36 hours. The assimilation effect of EnOI assimilation scheme is better than that of OI. (4) in order to further investigate the assimilation effect of EnOI on wave prediction in long time series, a one-year assimilation experiment was designed. It is found that EnOI has an absolute advantage over OI. After adopting the EnOI assimilation scheme, the probability of global wave effective wave height prediction with absolute error less than 0.5 m is 83.79 and the probability of less than 1m is 96.03. The prediction accuracy is very considerable. (5) because the background error is estimated by pre-stored historical samples, the background error is always kept unchanged in the integration process, which often exaggerates the background error and leads to a large range of false correlation on a long time scale. In order to solve this problem, the optimal method of EnKF initial sample generation is discussed by designing multi-group sensitivity tests and compared with EnOI history sample. It is found that, compared with historical samples, disturbed samples can better present the structure and correlation of pattern errors.
【学位授予单位】：国家海洋环境预报中心
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：P731.33
本文编号：2310434
[Abstract]:At present, the ensemble assimilation method has not been fully applied in ocean wave assimilation. In this paper, a global regional ocean wave assimilation system is designed based on the third generation wave numerical prediction model (WAVEWATCH III,). Three days of ocean wave prediction and long-term assimilation experiments are carried out with the ensemble optimal interpolation (Ensemble Optimal Interpolation, (EnOI), and the results are compared with the optimal interpolation (Optimal Interpolation,. OI) assimilation results were compared. It is found that EnOI assimilation method plays a good role in improving wave prediction. Considering that historical samples in EnOI exaggerate background errors and cause false correlation, in order to solve this problem, this paper first attempts to generate a set of dynamic samples by superimposing random disturbances in wind field. The advantages and disadvantages of historical and dynamic samples are evaluated, and the preparation for further research on EnKF assimilation based on ensemble Kalman filter (EnKF) is made. The main work is as follows: (1) background error information is very important in data assimilation. For this reason, the model is first integrated for 10 years, and the model simulation results are tested by NDBC buoy data and Jason-1 altimeter data. It is found that the model has a good effect on the global ocean wave simulation, and the effective model error is obtained. It provides a basis for the construction of background error covariance matrix and the selection of set samples in the later assimilation work. (2) in order to save calculation time, four assimilation observation point selection schemes are compared and analyzed because there are more data of assimilation observation points in the global scope and the model grid is coarse. It is found that the assimilation effect will not be significantly weakened after the observation is sparse, and the calculation time of assimilation can be shortened without affecting the assimilation effect by filtering the high-frequency disturbance by using the five-point average thinning scheme. (3) in order to evaluate the improved effect of OI,EnOI assimilation in short-term wave prediction, a three-day wave prediction assimilation experiment was carried out. It is found that the data assimilation can correct the deviation of the initial field well, and improve the initialization process and the 3-day prediction process obviously. The improved effect of EnOI on the model is more stable in time series, and is more stable for the prediction within 36 hours. The assimilation effect of EnOI assimilation scheme is better than that of OI. (4) in order to further investigate the assimilation effect of EnOI on wave prediction in long time series, a one-year assimilation experiment was designed. It is found that EnOI has an absolute advantage over OI. After adopting the EnOI assimilation scheme, the probability of global wave effective wave height prediction with absolute error less than 0.5 m is 83.79 and the probability of less than 1m is 96.03. The prediction accuracy is very considerable. (5) because the background error is estimated by pre-stored historical samples, the background error is always kept unchanged in the integration process, which often exaggerates the background error and leads to a large range of false correlation on a long time scale. In order to solve this problem, the optimal method of EnKF initial sample generation is discussed by designing multi-group sensitivity tests and compared with EnOI history sample. It is found that, compared with historical samples, disturbed samples can better present the structure and correlation of pattern errors.
【学位授予单位】：国家海洋环境预报中心
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：P731.33
【参考文献】
中国期刊全文数据库 前10条
1 王关锁;乔方利;杨永增;;基于MPI的LAGFD-WAM海浪数值模式并行算法研究[J];海洋科学进展;2007年04期
2 尹宝树,王涛,范顺庭;YW-SWP海浪数值预报模式及其应用[J];海洋与湖沼;1994年03期
3 文圣常;GENERALIZED WIND WAVE SPECTRA AND THEIR APPLICATIONS[J];Science in China,Ser.A;1960年03期
4 侯一筠,楼顺里,谢强,杨联贵;浅水非线性波的演化方程[J];科学通报;1998年03期
5 齐鹏;范秀梅;;高度计波高数据同化对印度洋海域海浪模式预报影响研究[J];海洋预报;2013年04期
6 任启峰;张杰;尹训强;杨永增;;Envisat ASAR海浪谱资料的最优插值同化试验[J];热带海洋学报;2010年05期
7 袁业立,华锋,潘增弟,孙乐涛;LAGFD-WAM海浪数值模式——Ⅱ.区域性特征线嵌入格式及其应用[J];海洋学报(中文版);1992年06期
8 侯一筠;旋转流体中的非线性惯性波[J];海洋学报(中文版);1995年01期
9 ;Theoretical wind wave frequency spectra in shallow water[J];Acta Oceanologica Sinica;1988年03期
10 孙明华;;一个集合海浪预报系统及其初步试验[J];应用气象学报;2011年06期
本文编号：2310434
本文链接：https://www.wllwen.com/kejilunwen/haiyang/2310434.html
