Near-real-time volcanic ash cloud detection: Experiences from the Alaska Volcano Observatory

Volcanic eruptions produce ash clouds, which are a major hazard to population centers and the aviation community. Within the North Pacific (NOPAC) region, there have been numerous volcanic ash clouds that have reached aviation routes. Others have closed airports and traveled for thousands of kilometers. Being able to detect these ash clouds and then provide an assessment of their potential movement is essential for hazard assessment and mitigation. Remote sensing satellite data, through the reverse absorption or split window method, is used to detect these volcanic ash clouds, with a negative signal produced from spectrally semi-transparent ash clouds. Single channel satellite is used to detect the early eruption spectrally opaque ash clouds. Volcanic Ash Transport and Dispersion (VATD) models are used to provide a forecast of the ash clouds' future location. The Alaska Volcano Observatory (AVO) remote sensing ash detection system automatically analyzes satellite data of volcanic ash clouds, detecting new ash clouds and also providing alerts, both email and text, to those with AVO. However, there are also non-volcanic related features across the NOPAC region that can produce a negative signal. These can complicate alerts and warning of impending ash clouds. Discussions and examples are shown of these non-volcanic features and some analysis is provided on how these features can be discriminated from volcanic ash clouds. Finally, there is discussion on how information of the ash cloud such as location, particle size and concentrations, could be used as VATD model initialization. These model forecasts could then provide an improved assessment of the clouds' future movement.     

Volcanic ash hazard in the Central Mediterranean assessed from geological data

Volcanic ash produced during explosive eruptions can have very severe impacts on modern technological societies. Here, we use reconstructed patterns of fine ash dispersal recorded in terrestrial and marine geological archives to assess volcanic ash hazards. The ash-dispersal maps from nine Holocene explosive eruptions of Italian volcanoes have been used to construct frequency maps of distal ash deposition over a wide area, which encompasses central and southern Italy, the Adriatic and Tyrrhenian seas and the Balkans. The maps are presented as two cumulative-thickness isopach maps, one for nine eruptions from different volcanoes and one for six eruptions from Somma-Vesuvius. These maps represent the first use of distal ash layers to construct volcanic hazard maps, and the proposed methodology is easily applicable to other volcanic areas worldwide.     

Volcanic and atmospheric controls on ash iron solubility: A review

The ash material produced by volcanic eruptions carries important information about the underground magma eruptive conditions and subsequent modifications in the volcanic plume and during atmospheric transport. Volcanic ash is also studied because of its impacts on the environment and human health. In particular, there is a growing interest from a multidisciplinary scientific community to understand the role that ash deposition over open ocean regions may play as a source of bioavailable Fe for phytoplankton production. Similar to aeolian mineral dust, the processes that affect the mineralogy and speciation of Fe in ash may promote solubilisation of Fe in ash, and thus may increase the amount of volcanic Fe supplied to ocean surface waters. Our knowledge of these controls is still very limited, a situation which has hindered quantitative interpretation of experimental Fe release measurements. In this review, we identify the key volcanic and atmospheric controls that are likely to modulate ash Fe solubility. We also briefly discuss existing data on Fe release from ash and make some recommendations for future studies in this area.     

模糊聚类与遗传算法相结合的卫星云图云分类

针对模糊C均值聚类（FCM）方法存在的缺陷,提出运用遗传算法（GA）全局寻优与FCM局部寻优以及模糊减法聚类客观估算聚类数等优势互补的思想和途径进行卫星云图云分类判别.试验结果表明,综合方法（GA\|FCM）的云分类效果明显优于单一的FCM和GA算法,可有效弥补FCM和GA算法在云分类中存在的不足,并可运用于实况云图中云类的客观、自动判别.     

基于图像语义分割的层位自动追踪方法

层位追踪是地震资料解释中基础且重要的工作.传统的层位追踪主要依靠人的主观性判断,是一项既要求专业知识经验,同时又费时费力的环节,为此本文提出了一种基于图像语义分割技术的层位自动追踪方法.图像语义分割是像素级别的分类技术,我们将地震剖面中的层位看作图像中的一个语义类别,借助分割网络进行层位的分类与追踪.本文选择了改进的Dense-UNet神经网络作为分割算法,该网络强调对特征的传递和复用,同时避免了梯度消失的问题,能够很好适应地震数据的分割任务.首先以实际地震数据制作实验所用数据集,应用数据扩充方法增大训练样本数量,再借助改进的Dense-UNet神经网络训练出最优的追踪模型,通过测试集在最优模型上的表现来评价追踪的效果.实验结果表明,改进的Dense-UNet网络能够准确且快速的自动完成层位追踪任务,验证了该方法的可行性和有效性.     

基于改进云图法的结构概率地震需求分析

概率地震需求分析是美国太平洋地震工程研究中心(Pacific Earthquake Engineering ResearchCenter,PEER)提出的新一代"性能化地震工程(Performance-Based Earthquake Engineering,PBEE)"理论框架的重要一环。传统的概率地震需求分析方法称为"云图法",这种方法针对确定性结构进行一系列地震动作用下的非线性动力分析,从而得到地震动强度参数与结构地震需求的"云图"。然而,传统的云图法只能考虑地震动的不确定性,而无法考虑结构的不确定性。为此,结合拉丁超立方体抽样技术,提出一种能综合考虑地震动不确定性和结构不确定性的改进云图法,并将传统的概率地震需求分析内容拓展为概率地震需求模型、概率地震需求易损性分析、概率地震需求危险性分析三个层次。以一榀五层三跨钢筋混凝土框架结构为例,分别采用传统云图法和改进云图法对其进行概率地震需求分析,得到了该结构的概率地震需求模型、地震需求易损性曲线和地震需求危险性曲线。分析结果表明:提出的方法可以有效地考虑地震动与结构的不确定性,避免不考虑结构的不确定性而低估结构的地震风险性。     

基于像素差分神经网络的断层识别方法

断层识别是地震解释的一项重要任务.相对传统的基于人工或半人工的断层检测方法, 基于深度学习的断层检测方法具有自动化程度高等特点, 近年来已经吸引了业界广泛兴趣.目前, 大多数利用深度学习的断层识别方法都是基于语义分割或图像分类, 但基于语义分割或图像分类的方法预测出的断层噪声较多.而边缘检测网络可以通过学习断层在地震剖面中不连续的特征, 从而提高网络的抗噪能力.为了利用边缘检测网络的抗噪优点并提升其性能, 本文采用基于像素差分的卷积运算构建出像素差分网络模型(Pixel Difference Networks, Pidinet).相较于传统的边缘检测网络, Pidinet将传统的边缘检测算子与深度学习网络结合, 有效地提升了边缘检测的效果.为了使Pidinet网络更好地预测断层, 本文对原始Pidinet网络进行了优化, 去掉了相对断层识别任务而言多余的分支结构和卷积层.相对于传统的卷积运算, 基于像素差分的卷积运算可以让神经网络更好地学习断层信息.为了让网络充分学习数据中的断层特征, 本文将真实地震样本与合成地震样本混合得到训练所用数据集.实验证明Pidinet在交并比(Intersection over Union, IoU)方面较Holistically Nested Edge Detection (HED)边缘检测网络提升了10%左右.为了测试网络的迁移能力, 本文只使用少量的数据样本对网络进行微调, 迁移的结果较分类网络在F1分数(F1 Score)、灵敏度(sensitivity)等指标均提升了10%以上.最后, 本文使用公开的实际地震数据进行测试, 实验结果显示Pidinet识别出的断层连续且清晰, 从而证明了基于边缘检测的深度学习算法在断层识别问题中的有效性.

     

基于EMD方法与RBF神经网络的结构损伤检测

针对结构损伤检测中损伤的识别、定位以及程度的标定这三个独立并按一定先后顺序进行的检测过程,提出了一种能将以上三者同时进行的联合检测方法。该方法首先利用经验模态分解(EMD)方法将三层钢筋混凝土剪切型结构在各种损伤工况下的顶层地震作用加速度响应分解为若干固有模态函数(IMF)分量,然后以此IMF分量和未经EMD分解的原始加速度响应数据来构造损伤标识量,作为特征参数依次输入到径向基函数神经网络(RBFNN)中进行损伤检测。给出了应用此方法的具体步骤,通过仿真实验证明了利用该方法进行结构损伤一次检测的可行性和有效性,结果表明,由加速度响应经EMD分解而得到的IMF分量输入到RBFNN中能够更为精确地一次检测出结构所有损伤信息,并且RBFNN在结构损伤损度大时具有更好的检测效果。     

基于局部二值法的断层检测方法应用研究

利用地震资料检测断层对油气勘探具有重要的意义,也是地震资料解释的重点.文中发展了局部二值模式(Local Binary Pattern,LBP)方法来检测断层,并结合地震剖面的旋转不变局部方差(Rotation Invariant Variance,VAR)信息,提出一种基于LBP/VAR特征的断层检测方法.首先利用相干属性、曲率属性和边缘检测等常规的断层检测方法和LBP/VAR算子对原始地震数据进行断层刻画,说明LBP/VAR算子的有效性,再对LBP、LBP/VAR进行阈值优化处理,使其检测断层的效果更佳.其次用LBP/VAR算子实现对含噪声的理论断层模型以及实际断层图片检测并对比分析.试验结果表明LBP/VAR算子对含噪声的断层理论模型仍然有效,具有一定的鲁棒性.实际断层图片检测证明该方法对断层检测具有更好的效果.研究结果表明LBP/VAR算子具有连续性好、一定的抗干扰能力,能较好的刻画断层,为地震资料断层检测提供了一种有效的检测工具.     

基于广义RI指数的微波观测资料干扰识别方法

基于AMSR2(the Advanced Microwave Scanning Radiometer 2)一级亮温数据,提出了广义RI指数(Radio Frequency Interference Index)识别方法来检测无积雪覆盖陆地和洋面区域AMSR2 7.3 GHz通道受RFI(Radio Frequency Interference)影响的观测数据,进而用主成分分析法以及谱差法的检测结果进行验证,并分析了三种方法检测结果间的差异.结果表明:广义RI指数识别方法的检测出的RFI和其他两种方法的检测结果基本一致,这表明新算法对识别陆地和洋面上的RFI是有效的.但这三种检测方法的区别是,主成分分析法识别的RFI信号比其他两种方法确定的范围小、数量少,谱差法次之,广义RI指数检测方法检测出的RFI信号最多.本研究为评估不同识别算法提供了参考,还为进行卫星遥感技术微波频段保护提供了理论依据.     

应用人工神经元网络方法自动检测地震事件

利用人工神经元网络方法,提出了一种从连续的地震数据中检测出地震事件的方法。该方法分两步,首先,低阈值的STA／LTA算法从连续的波形中检测出类似地震事件;其次利用神经元网络方法,区分事件是地震事件还是噪声事件。通过对数据检测结果比较,找出了适合地震检测的神经元网络训练方法和神经元传递函数。在对天山流动台阵其中两个台的检测结果表明,在连续约两个月数据中,39RLS台检测出地震75个,30RNA台检测出地震95个,证明该方法对地震事件检测来说是一种有效的方法。     

基于卷积神经网络的电成像图像空白条带填充方法

全井眼微电阻率成像(Fullbore Formation Micro Imager,FMI)技术广泛应用于强非均质性的碳酸盐岩储层测并精细解释.然而,由于各极板间间隙的影响,FMI无法测量360°全井壁地层电阻率信息,这使得FMI图像出现空白条带,严重影响了图像视觉和测井解释进程.为获取全井壁的电阻率信息,必须有效的填充空白条带.本文在探讨现有的电成像图像空白条带填充方法基础上,提出了一种基于卷积神经网络的电成像图像空白条带填充方法.与常用的Criminisi插值法对比填充效果显示:该方法完全适用于各种特征的电成像图像空白条带填充,延续了所有待填充边缘的暗色条带和亮色背景,填充结果与原始图像的像素值频率占比相关性高达0.9581.Criminisi插值法在复杂区域电成像图像的填充结果连通性较差且存在干扰区域,填充结果与原始图像的像素值频率占比相关性仅为0.5924.本文提出的方法具有处理速度快、干扰区域少、连通性强、相关性高的特点,能够满足电成像空白条带成规模填充的实际应用需求,还可为测量极板间的测量缺失带信息处理提供参考手段.     

Forest NPP estimation based on MODIS data under cloudless condition

Based on light-use efficiency model, an MODIS-derived daily net primary production (NPP) model was developed. In this model, a new model for the fraction of photosynthetically active radiation absorbed by vegetation (FPAR) is developed based on leaf area index (LAI) and albedo parameters, and a pho- tosynthetically active radiation (PAR) is calculated from the combination of Bird's model with aerosol optical thickness and water vapor derived from cloud free MODIS images. These two models are inte- grated into our predicted NPP model, whose most parameters are retrieved from MODIS data. In order to validate our NPP model, the observed NPP in the Qianyanzhou station and the Changbai Mountains station are used to compare with our predicted NPP, showing that they are in good agreement. The NASA NPP products also have been downloaded and compared with the measurements, which shows that the NASA NPP products underestimated NPP in the Qianyanzhou station but overestimated in the Changbai Mountains station in 2004.     

太湖蓝藻水华的MODIS卫星监测

本文基于MODIS卫星影像探讨太湖蓝藻水华的识别、监测问题.通过比较MODIS影像上不同蓝藻浓度的波段反射率值的差别,得到对蓝藻信息响应的敏感波段,利用特定波段的合成,可基本识别蓝藻的分布信息,并基于蓝藻的光谱响应特征,采用比值模型方法、进一步确定蓝藻分布的相对浓度信息.最后通过不同时相蓝藻浓度的叠加分析,得出了蓝藻水华的动态变化信息.     

基于图像边缘检测方法提取地震剖面同相轴

地球物理勘探中,地震剖面中的同相轴对地质分层、断层解释、有无形成空洞以及其它异常体的检测中作用重要.根据地震剖面上同相轴具有的性质,采用图像边缘检测方法提取同相轴简单且有效.由于边缘检测中常用的Sobel算子和Canny算子检测同相轴时对噪声免疫差,常因噪声因素引起边缘信息缺失导致同相轴提取不连续,本文采用中值滤波并对Sobel算子进行改进,对地震剖面图进行预处理,结合均方差自动确定阈值并进行异常检测,提取完整同相轴.实验结果表明应用该方法可以提取连续、准确的同相轴.     

基于地震图像校准的共成像点道集增强技术

由共成像点道集抽取的共偏移距道集可以当成相同地下成像的多次观测.由于偏移误差的影响,在不同共偏移距道集上,同一采样点存在水平和垂直方向上的错位.本文提出一种基于地震图像校准的共成像点道集增强技术,实现了不同偏移距道集在时间和空间上的逐点匹配对齐.在本文中以2D局部归一化互相关来表征共偏移距道集和叠加道集在相同时空位置的相似度,假设在不同水平和垂直移动量时的互相关满足连续凸函数,利用求导方法估计共偏移距道集在该位置处的非整数校正量,最后采用双线性内插方法估计成像振幅.传统道集拉平技术在垂直方向进行校正量消除,本文方法能有效估计共偏移距道集中的水平和垂直校正量,并在亚像素域估计正确的成像振幅.模型数据和实际数据的处理结果表明,本文方法能有效增强共成像点道集中同相轴的一致性,提高叠加结果的分辨率.     

A robust method for microseismic event detection based on automatic phase pickers

We present a robust method for the automatic detection and picking of microseismic events that consists of two steps. The first step provides accurate single-trace picks using three automatic phase pickers adapted from earthquake seismology. In the second step, a multi-channel strategy is implemented to associate (or not) the previous picks with actual microseismic signals by taking into account their expected alignment in all the available channels, thus reducing the false positive rate. As a result, the method provides the number of declared microseismic events, a confidence indicator associated with each of them, and the corresponding traveltime picks. Results using two field noisy data records demonstrate that the automatic detection and picking of microseismic events can be carried out with a relatively high confidence level and accuracy.     

区域地震信号自动识别方法及应用(英文)

地震信号的实时、自动、准确识别对于地震自动速报和地震预警十分重要。仿真信号试验分析表明,观测数据的四阶统计量函数(BKCF)对信号与噪声在能量和(或)频率方面的微弱差异变化具有较高的分辨能力。以此为基础,本文提出了一种新的自动探测区域地震事件的方法和测定直达波震相到时的BKCF-AIC方法。为了进一步提高波震相到时测定的精度,本文首先对指定时段的P-波记录进行偏振特性分析,其次对含有P波的S波记录进行偏振滤波处理,再次应用上述方法测定震相到时。与传统算法相比,基于山东测震台网记录的区域地震震例分析结果表明,使用本文提出的方法能够大幅度降低地震事件误检、漏检率,进一步提高了震相识别精度。