【作者】 杨莹；

【导师】 陈圣波；

【作者基本信息】 吉林大学 ， 地球探测与信息技术， 2012， 博士

【摘要】 闪电是自然界最为壮观和重要的大气现象之一，与全球变化许多因素密切相关。利用闪电频数和降雨率的统计关系可以预报降雨；利用风暴正地闪特征可以预报强风暴中超级单体的产生、演变、龙卷风和雹暴的产生和消亡；还可以预报台风等强烈的天气现象。全球变暖模拟中，云量的分布和结构受低空间分辨率和相对简单的对流参数的限制，区域和全球的闪电观测数据可解决这种限制。现已建立了一些密集的雷电监测站网，虽然能较准确地对其附近发生的闪电进行定位和计数,甚至可测量闪电光谱和无线电信号,但由于其布点的局限性,很难给出全球范围的闪电分布图像，卫星观测则弥补了这方面的不足。美国发射光学瞬闪探测仪（Optical Transient Detector, OTD）后发射了闪电成像仪（LightningImaging Sensor, LIS），至今仍正常工作，其数据是现应用最广泛的闪电数据。但由于其搭载在低轨道卫星上，受轨道周期的限制，闪电探测能力不足。静止轨道卫星闪电成像仪可以从静止轨道平台对视场覆盖范围内的闪电进行连续不间断的观测，是闪电探测最有效的手段。美国计划2014年发射静止卫星GOSE-R上携带闪电成像仪GLM，我国即将发射的静止轨道气象卫星风云四号的有效载荷也包括闪电成像仪。在闪电仪数据预处理算法研究方面，我国从未开展过，美国同类研究也正在进行之中，属于科学技术前沿。因此，本论文研究了静止卫星闪电成像仪数据预处理中的关键技术，可对我国未来风云四号卫星的闪电成像仪地面系统建设提供一定的技术支撑。本文结合LIS数据、闪电的物理特征、中国地区闪电时空分布，通过对闪电信号的模拟、噪声的模拟以及闪电脉冲辐射传输模拟，分析了云对静止卫星闪电成像仪探测到的闪电信号的影响。通过SKT模拟开发闪电定位订正模型，得出云顶高度对闪电定位的影响。在模拟数据的基础上开发闪电数据生成算法，通过像素滤波算法、闪电聚类算法、粒子滤波等滤除闪电探测中产生的虚假信号，生成闪电数据产品。对1998年~2010年的LIS闪电数据进行再分析，生成闪电密度数据集，对闪电密度数据进行EOF分析，得到中国地区闪电的时空分布特征。并与降雨量台站数据经克里金插值后生成的格点数据进行SVD分解，分析闪电和降雨的相关关系。研究得出，我国闪电密度分布呈显著的区域性差异。高闪电密度带包括山东中西部地区、广东、广西、贵州、四川南部以及海南岛中北部；海洋的闪电密度相对较低；闪电密度分布具有一定的纬向特征，北回归线及其北部地区具有相对较高密度值。中国陆地闪电密度跟离海洋的距离也有关系，在海陆过渡带处达到峰值后呈总体下降的趋势，越远离海洋值越小。闪电和降水具有很高的相关性，当中国地区的闪电密度增高（降低）时，相应的降雨量也会增加（减少）；降雨量增加（减少）时，闪电密度也会增高（降低），降雨量受闪电密度的影响比闪电密度受降雨量的影响要大。气象场距平选择的时间尺度越大，闪电密度和降雨量的相关性越高，但在同一地区，相关系数的值一直很稳定。由于没有静止卫星闪电数据，在数据预处理技术研究中需要模拟代理数据。首先使用STK软件模拟闪电仪视场覆盖区域。在了解闪电基本电结构、放电过程以及卫星闪电探测方法的基础上，通过建立闪电云中辐射传输模型模拟静止卫星闪电成像仪探测到的云顶闪电信号特征。建立LIS数据数理统计模型，模拟闪电成像仪探测到的闪电信号数据。并通过分析LIS探测中产生的虚假闪电信号，结合蒙特卡洛方法，模拟虚假闪电信号。在以上工作的基础上结合探测区域闪电频率分布，生成代理数据。利用云中闪电传输模型和闪电定位的云高订正模型，研究云对静止卫星的闪电探测影响。云对闪电探测的影响包括云粒子散射对云顶闪电脉冲的影响，以及云顶高度对闪电定位的影响。研究得出得出云的体积、光学厚度、形状，甚至闪电在云中发生的位置，都会对云顶输出闪电能量的大小和持续时间产生影响，使有些闪电需要连续几帧图像才能被全部探测到。经纬度订正值与云高为正相关关系。云高变大，图像边缘像元位置偏差就变大。云顶高度影响的经向偏差以中心经线为对称轴，向东北和西北方向逐渐扩大；而其纬向偏差则随着纬度的增加而增加。在静止卫星闪电成像仪探测闪电定位时，必须进行闪电定位的云高订正。闪电成像仪探测到的数据中，包含着大量的虚假闪电信号，必须设计滤波算法将其滤除。在模拟数据的基础上，通过分析研究LIS闪电探测产生的虚假信号，结合静止卫星闪电探测的特点，开发基于像素级别的闪电滤波算法。像素级滤波不能完全过滤虚假闪电信号，而且，闪电产品的生成必须要对探测像素级别数据进行聚类。为了数据产品的生成和继续滤除噪声，在分析闪电时空特征的基础上开发闪电成像仪的聚类算法。使用该算法对LIS数据中Event聚类，聚类结果与数据中Group和Flash基本符合。但在进行聚类处理时，部分噪声也被聚类到2级数据中，因此需进行粒子滤波。粒子滤波的理论是保留时间和空间上聚聚集的Flash，剔除孤立的Flash，最大限度保留真实的闪电，最小限度接受噪声，使得错误发生地可能性最小。滤波处理算法保留了拒绝的Flash，以备将来将这些Flash能给重新参加到闪电数据的检验和计算中。使用模拟数据在一定程度上对于算法的精确性和准确性有影响，在进行室内外闪电实验和测试卫星发射后，可根据实验数据做适当的调整。更多还原

【Abstract】 Lighting is one of most splendid and important atmospheric phenomena, which isclosely related with numerous factors in global changes. Rainfall could be predictedthrough statistical relationship between lightning frequency and rainfall rate. Thegeneration, development and variation characteristics of strong supercell storms couldbe forecasted through characteristics of positive cloud-to-ground, so are creation anddissolution of tornado and hailstorm. In addition, the evolution process of strongweather phenomena could be also predicted. During researches on simulation ofglobal warming, cloud coverage including distribution and structure is limited by thelow spatial resolution and relatively simple convective parameters, which could besolved by regional and global lightning data observed.Lighting motoring stations have been built up rather densely around the world.Although these stations could detect and count the frequencies of lighting and evendetect lighting spectra and radio signals in their neighborhoods, it is hard for them toprovide global distribution of the lighting and satellite observation could just make upthis disadvantage. Optical Transient Detector (OTD), the first optical lighting imagemapper on the satellite, was launched by American, and after it Lighting ImagingSystem (LIS ) was set up, which is still at work and the most widely used lighting data.However it has a limited ability for it is mounted on the low-orbit satellite that leadsto limited orbit periods.It is planned in American that a stationary satellite GOSE-R will be launchedabroad GLM in 2014, and a meteorological satellite FY-4 to be launched by ourcountry will also carry the lighting image sensor. The lighting’s data productsalgorithm is a science and technology frontier internationally which has been studiedin USA but still a blank field in our country. Therefore, the key techniques of datapreprocessing on geostationary lightning mapper have been studied in this thesis,which would provide technical supports for ground system contribution ofmeteorological satellite FY-4’s lighting imaging sensor in the near future.Considering LIS data, physical features of flash and spatial variations in China,flash signals and noise as well as the radiative transfer of lighting impulse are simulated, based on which the influence of cloud on detecting ability of stationarysatellites is considered. The effect of cloud top’height on lighting position isconsidered by a flash position correction model developed by STK. The algorithm oflighting products is developed, which employs pixel filtering, clustering algorithmand particle filtering to filter the false signals and generate the lighting products.LIS lighting data from 1998 to 2010 is reanalyzed, and the Flash density datasetis generated, which is used to do the EOF analysis in order to suggest the spatial andtemporal distribution of lighting in China. Moreover, the rainfall grid data is generatedby the Kriging interpolation, which is used to do the SVD decomposition in order toanalyze the correlation between rainfall and lighting. The results show that there existsignificantly regional differences for lighting distribution around the country. Highdensity zones include middle and western Area of Shandong, Guangdong, Guangxi,Guizhou, southern part of Sichuan as well as middle and northern part of Hainan. Theocean is less dense. The lighting also presents latitudinal characteristic, and the Tropicof Cancer and its northern part is denser. In addition, the flash density is related to thedistance with the ocean, which peaks at transitional zones of land and ocean and thenshow a general downward tendency, the far the distance from the ocean the lower thefrequency will be. What’s more, the lighting is closely associated with rainfall. Whenthe Flash frequencies increase (decrease), the rainfall will increase (decrease); on thecontrary when the rainfall increases (decrease), the frequencies willincrease(decrease), and the influence of frequencies on rainfall is greater. The biggerthe time scale of meteorological field anomaly is, the higher correlation between thefrequencies and rainfall will be, but the correlation coefficient is rather steady in thesame Area.Because there is no lighting data from geostationary satellite, proxy data hasbeen simulated during the study on data preprocessing. First, the view field of lightingimage sensor simulated by STK covers the study area. Knowing basic electronicstructure, the charging procession and the method for satellites to detect the lighting, aradiation transmission model within the cloud is established and used to simulatefeatures of lighting signals captured by the sensor on top of cloud. Analyzing falselighting signals from LIS, false lighting signals are simulated by Monte Carlo Method.Based on the work mentioned above, proxy data is finally generated considering thefrequency distribution of regional lighting that has been dected.Using a lightning signal transfer model within the cloud and a cloud top heightcorrection model for lighting position, the influence of cloud on detecting ability of stationary satellites is considered, which mainly includes how cloud particle effectflash impulse of cloud top and how the height of cloud top effect the position accuracy.The volume, optical thickness, and shape of cloud, even where the lightning occurwithin the cloud will exert influence on lighting energy and its lasting time, and that’sjust why continuous images are needed to detect to all the lightings. The longitudinaland latitudinal deviations are positively related with cloud height. The higher thecloud is, the greater the positional deviations of edge pixels will be. The longitudinaldeviations effected by cloud top height show the center longitude as the axis ofsymmetry, and gradually expand to the direction of northwest and northeast; thelatitudinal deviations increase with the latitude degrees. Thus the cloud height must becorrected for lighting position when using the lighting image sensor on stationarysatellites to detect the lighting.The data collected by lighting image sensor consist of so many false signals thatthey should be removed by filtering algorithm. Based on simulated data, pixel-pixelfiltering algorithm is developed considering the characteristics of stationary satellitesdetect the lighting after analyzing the false signals from LIS data. However, it couldnot totally filter them. As to generate data products and filter, the clustering algorithmis designed, which is applied to Event of LIS data. And the clustering results coincidewith Group and Flash data. While clustering, part of noise is clustered into data ofLevel 2, and this problem could be solved by particle filtering. The theory of particlefiltering holds collective Flash of both time and space scale, and get rid of isolatedFlash which leave the actual lighting to largest extent and accept the noise to smallestextent so that the possibility of errors is lowest. The algorithm also keeps the rejectedFlash for participating testing and calculating again. To some degree, the accuracy ofalgorithm is influenced by the simulation data, which could be adjusted appropriatelyaccording to experiment data after indoor lighting test and launching of satellites.更多还原