Statistical approaches to cloud classification

For the determination of clouds from satellite data there exist in general more unknown parameters than independent observations. If the bispectral observations are used from the geostationary satellites in the solar (VIS-channel) and in the infrared (IR-channel) range to derive cloud parameters, information is needed whether a pixel radiance is from a cloud free or a cloudy scene. Statistical methods are applied to derive those informations. Various proposed statistical methods are discussed.The histogram analysis developed at the University of Cologne is described in detail: bispectral (two-dimensional) histograms are partitioned into clusters. Cloud cover results are shown. A comparison is given between the results of the histogram analysis, threshold methods (VIS- and IR threshold separately and both combined) and the spatial coherence method developed by Coakly and Bretherton. The cloud cover varies in this example by a factor of two depending on the definition of the threshold between cloud free and cloudy pixels. It is further shown that after a cluster analysis of a two-dimensional histogram the derived cloud cover is not as sensitive to the threshold as for a threshold method. The methods which are discussed here are those proposed for the International Satellite Cloud Climatology Project (ISCCP). The results are from the pilot study of the ISCCP.     

基于卫星云图的大区域云层预测方法

云层覆盖是影响对地观测卫星成像的一个重要问题,如果遥感图像中云层比例太高,或者特定目标不可见,则遥感图像就会失效。对地观测卫星能够根据云层预测信息,在多个观测目标之间进行选择。面向对地观测卫星任务规划的应用,设计了大区域范围的短期云层预测方法,首先通过光流法获取云运动矢量,然后依据云运动矢量外推获得预测的云层图像,同时引入拉普拉斯算子刻画云层运动过程中的扩散现象,利用风云二号卫星的真实云图序列数据,通过神经网络的反向传播算法优化扩散因子,以提升云层预测的效果。通过对结果进行分析,引入的拉普拉斯算子方法能够提高云层预测的精度,80%分位数的云层覆盖率误差约为11.7%,该精度的云层预测可以用于指导对地观测卫星任务规划。     

基于对抗和迁移学习的灾害天气卫星云图分类

针对卫星云图中的灾害天气数据存在严重不平衡问题,提出一个结合生成对抗学习（GAN）和迁移学习（TL）的卷积神经网络（CNN）框架以解决上述问题进而提高基于卫星云图的灾害天气分类精度。该框架主要包含基于GAN的数据均衡化模块和基于迁移学习的CNN分类模块。上述2个模块分别从数据和算法层面解决数据的类间不平衡问题,分别得到一个相对均衡的数据集和一个可在不同类别数据上提取相对均衡特征的分类模型,最终实现对卫星云图的分类,提高其中灾害天气的卫星云图类别分类准确率。与此同时所提方法在自建的大规模卫星云图数据上进行了测试,消融性和综合实验结果证明了所提数据均衡方法和迁移学习方法是有效的,且所提框架模型对各个灾害天气类别的分类精度都有显著提升。      

Radiation budget and cloud properties from METEOSAT observations

Studies of the Earth's radiation budget from polar orbiting satellite systems, such as the forthcoming NASA Earth Radiation Budget Experiment, suffer from errors due to a poor temporal sampling of the diurnal variations in the radiation field. A knowledge of the causes and magnitudes of such variations is of importance in minimising these errors. This paper presents data on daily mean radiation budget parameters, together with their variation over the daylight hours, relating them to physical processes within the earth/atmosphere system. The most significant cause of variability is shown to be persistent high level cloud. The relative magnitude of cloud induced variability in the visible and infrared spectral regions is derived.     

Nighttime cloud properties retrieval using MODIS and artificial neural networks

In this work a methodology for inferring water cloud macro and microphysical properties from nighttime MODIS imagery is developed. This method is based on the inversion of a theoretical radiative transfer model that simulates the radiances detected in each of the sensor infrared bands. To accomplish this inversion, an operational technique based on Artificial Neural Networks (ANNs) is proposed, whose main characteristic is the ability to retrieve cloud properties much faster than conventional methods. Furthermore, a detailed study of input data is performed to avoid different sources of errors that appear in several MODIS infrared channels. Finally, results of applying the proposed method are compared with in-situ measurements carried out during the DYCOMS-II field experiment.     

ISCCP cloud analysis algorithm intercomparison

The International Satellite Cloud Climatology Project (ISCCP) will provide a uniform global climatology of satellite-measured radiances and derive a climatology of cloud radiative properties from these radiances. For this purpose, a pilot study of cloud analysis algorithms was initiated to define a state-of-the-art algorithm for ISCCP. This study compared the results of applying the nine different algorithms to the same satellite radiance data. The comparison allowed for a sharper understanding of the process of detecting clouds and shows that all algorithms can be improved by better information about clear sky radiance values (essentially equivalent to surface property information) and by better understanding of cloud size distribution variations. The dependence of all methods on cloud size distribution led to selection of an advanced bispectral threshold technique for ISCCP because this method is currently better understood and more developed. Further research on cloud algorithms is clearly suggested by these results.     

Statistical studies of cloud cover based on pictures from meteorological satellites

Cloud amounts data of a five year period manually digitized from meteorological satellite imagery are used for cloud statistics over the European region. Spatial and temporal structure functions of the cloud field are computed and analysed for the seasons. The results give a numerical representation of the effects of the general circulation and the geographical-physical conditions on cloudiness.     

Effects of concurrent snow and cloud cover on planetary albedo

The effects of snow and cloud cover on planetary albedo are examined using observations from NOAA polar orbiting satellites. Reflected radiation was measured in the visible range (0.5 – 0.7 μm). Planetary albedos resulting from different cloud/snow cover conditions are compared using Northern Hemisphere snow cover maps, surface weather charts, satellite photos and data on land surface types. None of the cases studied show that concurrent cloud and snow cover produces significantly different planetary albedos than cloud cover alone. Cloud cover alone is found to yield higher planetary albedos than snow cover alone; the difference being greatest over forested areas. With and without snow cover present, clear-sky planetary albedos over farming and grazing lands (snow(0.45), no snow(0.15)) are found to be significantly higher than those over forested regions (snow(0.33), no snow(0.11)). Variations in satellite zenith angle are not found to produce significant effects in most cases studied.     

气象卫星遥感海面温度的资料处理方法

气象卫星遥感海面温度的资料处理关键环节是云检测(筛选晴空区)和大气削弱订正。目前卫星气象中心的卫星海面温度处理业务系统中,针对美国的NOAA 卫星是利用国际上普遍使用的红外窗区多通道法。对于中国发射的风云一号(FY-1)气象卫星,由于星载辐射计只有一个红外窗区通道能用于大气削弱订正,为此专门开发了单通道海面温度处理方法。文章介绍了两种方法的基本原理、算法和处理结果举例,并对两种方法的特点作了简要评价。     

基于高分四号数据的多时相多通道云检测算法

针对高分四号（GF-4）卫星影像波段较少导致传统云检测算法难以区分云与冰雪像元的问题,提出一种多时相多通道云检测算法。该算法首先对GF-4卫星影像进行辐射定标和配准,然后利用云与典型地表的光谱差异得到潜在云像元,之后利用序列GF-4卫星影像之间的差异识别出移动的云像元,最后利用中红外波段反演地表亮度温度来去除冰雪像元。该算法在海南、辽宁和安徽3个研究区域进行验证,并将检测结果与传统单时相云检测算法、支持向量机（SVM）云检测算法和实时差分（RTD）云检测算法的检测结果进行对比。结果表明,该算法优于其他3种云检测算法,准确识别率均达到90%以上,误检率均低于5%,有利于GF-4卫星影像的进一步利用。     

Discriminating satellite IR anomalies associated with the MS 7.1 Yushu earthquake in China

In the process of exploring pre-earthquake thermal anomalies using satellite infrared data, Blackett et al. (2011) found that the previously reported anomalies before the 2001 Mw 7.7 Gujarat earthquake, in India, were related to positive biases caused by data gaps due to cloud cover and mosaicing of neighboring orbits of MODIS satellite data. They supposed that such effects could also be responsible for other cases. We noted a strip-shaped TIR anomaly on March 17th, 2010, 28?days before the Ms. 7.1 Yushu earthquake (Qin et al., 2011). Here we again investigate multi-year infrared satellite data in different bands to discriminate whether the anomaly is associated with the earthquake, or is only bias caused by the data gaps. From the water vapor images, we find lots of clouds that have TIR anomalies. However, on the cloudiness background, there is an obvious strip-shaped gap matching the tectonic faults almost perfectly. In particular, the animation loops of hourly water vapor images show that the cloud kept moving from west to east, while they never covered the strip-shaped gap. We consider that the cloud with this special spatial pattern should have implied the abnormal signals associated with the seismogenic process. Based on current physical models, the satellite IR anomalies both on TIR and water vapor bands can qualitatively be explained using synthetic mechanisms.     

Assessment of thin cirrus and low cloud over snow by means of the maximum likelihood method

Cirrus clouds and low clouds over snow are sometimes difficult to assess by common retrieval methods. In the case of cirrus the reason is the highly variable optical depth while low clouds have approximately the same temperature and reflection properties as snow covered mountains (or plains). An empirical interactive method is described, which allows to classify with great detail clouds of the described types and to determine the fractional coverage of each cloud type as seen from the satellite. The statistical properties of the cloud classes are determined by analyzing small areas of uniform cloudiness. The algorithms applied to pairs of spectral images is the standard maximum likelihood method.     

Real-time stereoscopic applications using the goes operational satellites

We have investigated the use of real and synthetic stereo satellite images and stereo graphics in applications such as cloud-tracked winds, severe storm cloud analysis, and general meteorological interpretation. We have concluded that a stereo meteorological presentation is possible and desirable in an operational environment. Synthetic stereo could be used immediately in cloud-tracked wind operations. The presentation allows one to appreciate the interrelations between cloud motions and cloud structures, especially in multi-layered situations. Reprocessing of FGGE tropical wind sets with a synthetic stereo presentation showed some improved yields of low-level vectors, a significant increase in mid-level vectors, and very little change in the high-level vectors. Severe local storm real-stereo presentations are possible operationally because the 15 minute RISOP operations of GOES-East allow simultaneous scanning of both geosynchronous satellites twice per hour. The real-stereo height measurements of overshooting turrets are an improvement over infrared heights and can be used to monitor the strength of the thunderstorm updraft. Synthetic stereo presentations of thunderstorm tops can be presented in a non-linear fashion which stretches out the cloud top features. The synthetic stereo presentation is easier for most people to see. We recommend the use of a hybrid system where the viewing is done on the synthetic stereo image and the quantitative measurements are done on the real-stereo pairs.     

Stereoscopic imaging from polar orbit and synthetic stereo imaging

The photogrammetric determination of cloud top heights from stereoscopic satellite images seems to be a very good solution to this hitherto unresolved problem. Whereas in the U.S.A., stereoscopic imaging is done by use of geosynchronous weather satellites, in Europe such a system cannot be used because there is only one geosynchronous satellite (METEOSAT). An alternative could be a Stero Line Scanner (SLS) operating on a polar orbiter.SLS would scan twice, forward and backward, producing in this way two image strips for steroscopic viewing and photogrammetric measurements from pole to pole. Because of the cloud motion between the two scans, a SLS needs additional independent height information for reference points, e.g. from a Laser Ranger (LAR). The advantage of this method is that cloud motion, and therefore wind, can also be determined for these reference points. Another solution is a system of two SLS satellites flying one after the other and scanning the same area simultaneously. This allows cloud motion determination across the whole image. The use of infrared channels also allows night operation and provides additional data such as improved seas surface temperatures.The DFVLR is currently studying these problems. Synthetic stereoscopic imaging is being used in a forerunner programm to the SLS project and also for simulation in SLS studies.     

Generalised variational assimilation of cloud-affected brightness temperature using simulated hyper-spectral atmospheric infrared sounder data

Assimilated channel brightness temperature data from infrared sounders accounting for cloud effects have a positive effect on weather forecasting, especially in weather-sensitive areas. When cloud effects are included, the channel brightness temperature deviations follow a non-Gaussian distribution. However, classical variational data assimilation follows a Gaussian distribution. When processing the cloud-affected brightness temperature, useful data are lost through the cloud detection process, thus assimilating some channel brightness temperatures with weight function peaks above the cloud top. Furthermore, strict quality control of brightness temperature removes outliers. By adopting the generalised variational assimilation method, which assumes that errors follow a non-Gaussian distribution, this paper assimilates the cloud-affected brightness temperature using simulated data for the hyper-spectral atmospheric infrared sounder (AIRS). A channel set is formed by dynamically selecting AIRS channels. The experiments for retrieving temperature and humidity data demonstrate that the generalised variational assimilated cloud-affected brightness temperature method performs better than the classical method.     

Identification of a debris cloud from the nuclear powered SNAPSHOT satellite with haystack radar measurements

Data from the Massachusetts Institute of Technology Lincoln Laboratory Long Range Imaging Radar (known as the Haystack radar) have been used in the past to examine families of objects from individual satellite breakups or families of orbiting objects that can be isolated in altitude and inclination. This is possible because, for some time after a breakup, the debris cloud of particles can remain grouped together in similar orbit planes. This cloud will be visible to the radar, in fixed staring mode, for a short time twice each day, as the orbit plane moves through the field of view. There should be a unique three-dimensional pattern in observation time, range, and range rate which can identify the cloud. Eventually, through slightly differing precession rates of the right ascension of ascending node of the debris cloud, the observation time becomes distributed so that event identification becomes much more difficult.     

Climatological characteristics of cloud cover over Europe from meteorological satellites

Climatological characteristics of cloud cover - means, variances, structure functions - previously computed for five years are compared with the results of a new test cloud cover data set of one year period. The effects on the cloud field of the Alps, the Carpathian basin and the land-ocean interface, further the seasonal changes of the general circulation are well reflected. The new test data set is based upon digitized data of one morning as well as one afternoon picture per day. Therefore some results are also obtained of the cloud cover variations during the day.     

面向战斗机云作战的构造型仿真平台架构

随着大数据、云计算、物联网、移动互联网等信息技术的迅猛发展与广泛应用,新的作战模式不断涌现,以任务分布式指控流程为核心的云作战成为一种全新的跨域全维作战样式。在分析作战云与云作战特征的基础上,结合传统作战仿真流程提出了云作战体系仿真流程,并提出了云作战构造型仿真平台框架的总体方案设计与系统功能设计。通过云作战构造型仿真示例,对比了传统作战样式与云作战样式的观察-调整-决策-行动（OODA）循环,结果表明,云作战样式能够有效缩短OODA循环时间。      

自研激光雷达三维点云配准技术

为了实现激光雷达点云与图像重建点云的三维空间配准,基于自研三维扫描激光雷达系统,提出了新型的快速多尺度因子（FMSR）点云配准算法,研究了空间点云配准技术。该算法主要包括初始配准和精确配准2个步骤:初始配准使用基于尺度自适应关键点质量（ASKQ）的点云特征提取算法,提取关键点的特征匹配对,求解点云配准初始参数;精确配准利用K-邻近（KNN）算法全局搜索,提升计算效率,多次迭代得到2组点云之间的最优旋转矩阵、最优平移向量和最优尺度因子。仿真和实验结果表明,所提出的算法对空间目标（尺寸为20.30 m×7.85 m×26.56 m）实现空间点云配准,配准精度达到0.194 m,运行时间为16.207 s;与多尺度迭代最近点（S-ICP）算法相比,配准精度提高了0.131 m,运行时间提高了30%。所提出的空间点云配准技术可为场景重建和纹理匹配提供算法基础。      

Comparison of cloud top heights measured by airborne Lidar and TIROS-N image data

The determination of the cloud top height by means of satelliteborne IR-radiometers requires the conversion of the measured radiance to an equivalent blackbody temperature and the assignment of this temperature to a geometrical height. The latter is associated with errors which add up easily to several kilometers. DFVLR did a case study to compare satellite derived cloud top heights with those from airborne Lidar measurements. The difference of the radiosonde temperature from the standard temperature profile results in a 1.8 km difference in cloud top height. The achievable accuracy using actual radiosonde temperatures is ± 0.4 km for optical thick clouds and much less for optical thin clouds.