【作者】 丁又专；

【导师】 潘德炉； 韦志辉；

【作者基本信息】 南京理工大学 ， 模式识别与智能系统， 2009， 博士

【摘要】 目前,卫星遥感和数值模拟已经成为我们理解海洋过程的两大主要手段。卫星遥感具有周期性、宏观性、实时性和费用低等特点,被广泛应用于海洋的水体监测；数值模拟能够从整体上把握海洋现象的时空变化规律,在海洋预报中发挥着重要作用。由于海洋上空覆盖的云层、传感器扫描轨道变化等原因,使用可见光和红外波段反演的遥感数据往往存在较大比例的数据缺失区域；其次,难以准确检测的薄云会造成反演数据异常.数值模拟中控制方程是对现实世界的简化,模型、初始条件、边界条件的误差会导致预报时效的降低。结合卫星遥感和数值模拟两者的优势,利用数据同化方法,融合遥感观测和数值模拟数据,构建海洋数据同化系统,可以有效地提高数值预报的精度.针对以上问题,本文首先提出了结合经验模态分解(EMD)与经验正交函数(EOF)的自适应EMD-EOF资料重构方法,并应用该方法对2003年长江口海域5天平均的海表温度(SST)与表层悬浮泥沙浓度(SSC)遥感产品进行了资料重构。结果表明：(1)SST重构的均方根误差为0.9℃、SSC重构的对数均方根误差为0.137(log10mg/L); (2)相对于Alvera提出的DINEOF方法,EMD-EOF方法的计算时间不到DINEOF方法的50%,同时重构精度提高10%左右；(3) EMD-EOF方法可以有效的剔除遥感反演中薄云未准确检测导致的噪声点,提高原始遥感图像的准确度；(4) EMD-EOF方法可以有效的重构数据量极少的遥感图像,得到高空间分辨率、全覆盖的遥感再分析产品。海温与悬浮泥沙是影响中国近海浮游植物生长的主要因素之一,也是进行海洋生态模拟与预报的基础。本文使用减秩卡尔曼滤波(SEEK)方法,结合COHERENS数值模型与遥感观测数据,初步建立了杭州湾三维海温与悬浮泥沙的数据同化系统,利用2003年春季的遥感SST与SSC数据对同化系统进行了后报同化实验。结果表明：(1)相对于遥感SST,模拟数据、预报数据、分析数据的均方根误差分别为2.13、1.65和0.75℃,而相对于遥感SSC,三者的对数均方根误差分别为0.62、0.53和0.26(log10mg/L);(2)对分析数据与遥感数据、分析数据与预报数据的差异进行分析表明,分析数据在分布趋势上接近预报数据,在数值上接近观测数据,观测对同化的影响效果显著；(3)数据同化方法可以有效的结合遥感观测与数值模拟两者的优势,改进数值预报的精度。为了更好的利用遥感数据,提高海洋数值预报的精度,还需要在以下两个方面开展工作：(1)使用EMD-EOF方法对其他遥感数据产品(如CHL-a,透明度等)进行资料重构,同时通过对EOF分解后的时间模态系数进行预测,构建一个基于统计方法的短期海洋遥感预测系统。(2)利用数据同化方法,同化CHL-a、颗粒有机碳等遥感数据,提高海洋生态模拟与预报的精度。更多还原

【Abstract】 At present, satellite remote sensing and numerical simulation are the two major means by which we learn more about ocean processes. Satellite remote sensing is characterized by periodicity, macroscopy, real-time and low cost, which is the reason why it is widely used in ocean monitoring. Numerical simulation can grasp the rules of ocean spatial-temporal variations as a whole, playing an important role in ocean forecasting. Because of the clouds coverage over the ocean and changes in scanning orbit of sensors, the satellite remote sensing data obtained by the visible and infrared bands often show missing data in a large proportion. Besides, thin clouds which are difficult to precisely detect could result in abnormal data retrieval。The control functions in numerical simulation predigest the real world. And errors of model, initial conditions and boundary conditions will reduce the forecast abilities. Combining the advantages of satellite remote sensing and numerical simulation, we can make use of the data assimilation method, merge the remote data and simulated data, construct the ocean data assimilation system and improve the accuracy of ocean forecast.In response to the above problems, we advance an EMD-EOF data reconstruction method, which combines empirical mode decomposition (EMD) and Empirical Orthogonal Function (EOF). By applying the new method, we reconstruct the five-day-average sea surface temperature (SST) and suspended sediment concentration (SSC) data of Changjiang estuary sea area in 2003.The conclusions are as follows. Firstly, the root mean squared error(RMSE) of SST reconstruction is 0.9℃and log RMSE of SSC reconstruction is 0.137(log1O mg/L). Secondly, the calculating time of EMD-EOF method is less than half of that of the DINEOF method raised by Alvera, and the reconstruction precision is comparatively improved. Thirdly, the EMD-EOF method can effectively eliminate the abnormal data which result from undetected thin clouds in remote sensing retrieve, improving the precision of original remote sensing images. Lastly, the EMD-EOF method can effectively reconstruct remote sensing images of little data, which leads to reanalysis remote sensing products of high spatial-resolution and full coverage.Sea temperature and suspended sediment affect the growth of phytoplankton in China Adjacent Seas and they are also the basis of ocean ecological simulation and forecast. Using singular evolutive extended kalman filter(SEEK), combined with the simulation result of COHERENS model and remote sensing observation data, we initially build the three-dimensional data assimilation system of sea surface temperature and suspended sediment in Hangzhou Bay. This system is further tested via hindcast validation experiment by using the remote sensing data of SST and SSC of Spring in 2003.Our research results are as follows. Firstly, compared with the remote sensing SST, the RMSEs of simulated data, forecast data and analyzed data are 2.13,1.65 and 0.75℃respectively, and compared with the remote sensing SSC, the log RMSEs of simulated data, forecast data and analyzed data are 0.62,0.53 and 0.26 (log10 mg/L) respectively. Secondly, as the difference between the analyzed data and remote sensing data and the difference between the analyzed data and forecast data show, the analyzed data are identical to the forecast data in terms of distributing trend and the analyzed data are close to the observed data in terms of numerical value. Therefore, observation has obvious effect on assimilation. Lastly, the data assimilation method can effectively combine the advantages of both remote sensing observation and numerical simulation, improving the precision of numerical forecast.In order to better utilize the remote sensing data and improve the precision of ocean numerical forecasting, further research work is to be complemented from two perspectives. On the one hand, other remote sensing data (CHL-a, SDD eg.) are to be reconstructed by using the EMD-EOF method. Meanwhile, by forecasting the time-coefficients of EOF decomposition, we can build a short ocean remote sensing forecasting system. On the other hand, to enhance the precision of ocean ecological simulation and forecast, the data assimilation method is to be used to assimilate such remote sensing data as CHL-a and Particulate Organic Carbon (POC).更多还原