Simultaneous retrieval of volcanic sulphur dioxide and plume height from hyperspectral data using artificial neural networks

Simultaneous retrieval of volcanic sulphur dioxide and plume height from hyperspectral data using artificial neural networks
复制标题

使用人工神经网络从高光谱数据中同时检索火山二氧化硫和羽流高度

DOI:
10.1093/gji/ggu152
复制
发表时间:
2014
影响因子:
2.8
通讯作者:
Piscini A
Piscini A
中科院分区:
地球科学2区
文献类型:
--
作者:
Piscini A

文献摘要

参考文献

被引文献

相似文献

人工神经网络是一种用于从卫星图像估计地球物理参数的有价值和行之有效的反演技术;一旦经过训练,它们有助于生成非常快的结果。此外,委员会认为,卫星遥感是一种非常有效和安全的监测火山喷发的方法,以保护环境和受这些自然灾害影响的人。本文介绍了人工神经网络作为一个反演模型的应用,同时估计柱状含量和高度的二氧化硫(SO2)。在这项研究中,两个人工神经网络的实现,以模拟一个检索模型,并估计SO2柱含量和烟羽高度。使用1000-1200和1300-1410 cm-1之间的所有红外大气探测干涉仪(IASI)通道作为输入,以及使用Carboniet等人的SO2反演方案从相同的IASI通道获得的SO2含量和烟羽高度的相应值,对ANN进行训练,作为目标输出。检索证明了埃亚菲亚德拉冰盖火山(冰岛)的喷发为2010年4月和5月,格里姆火山喷发在2011年5月。这两个神经网络的训练与时间序列包括58个高光谱喷发图像收集2010年4月14日和5月14日和16个图像从2011年5月22日至26日,并在埃亚菲亚德拉冰盖喷发的三个独立图像数据集(一个在4月,另外两个在5月)和2011年5月发生的格里姆斯火山喷发的三个独立数据集上进行了验证。神经网络输出值与目标值的均方根误差(RMSE)值分别小于20多布森单位(DU)和200毫巴(mb),均小于格里姆火山喷发目标值的标准差。当目标值在训练阶段使用的值之外时,神经网络的检索精度较低。
Artificial neural networks (ANNs) are a valuable and well-established inversion technique for the estimation of geophysical parameters from satellite images; once trained, they help generate very fast results. Furthermore, satellite remote sensing is a very effective and safe way to monitor volcanic eruptions in order to safeguard the environment and the people affected by those natural hazards.This paper describes an application of ANNs as an inverse model for the simultaneous estimation of columnar content and height of sulphur dioxide (SO2) plumes from volcanic eruptions using hyperspectral data from remote sensing.In this study two ANNs were implemented in order to emulate a retrieval model and to estimate the SO2columnar content and plume height. ANNs were trained using all infrared atmospheric sounding interferometer (IASI) channels between 1000–1200 and 1300–1410 cm−1as inputs, and the corresponding values of SO2content and height of plume, obtained from the same IASI channels using the SO2retrieval scheme by Carboniet al., as target outputs.The retrieval is demonstrated for the eruption of the Eyjafjallajökull volcano (Iceland) for the months of 2010 April and May and for the Grimsvotn eruption during 2011 May.Both neural networks were trained with a time series consisting of 58 hyperspectral eruption images collected between 2010 April 14 and May 14 and 16 images from 2011 May 22 to 26, and were validated on three independent data sets of images of the Eyjafjallajökull eruption, one in April and the other two in May, and on three independent data sets of the Grímsvötn volcanic eruption that occurred in 2011 May. The root mean square error (RMSE) values between neural network outputs and targets were lower than 20 Dobson units (DU) for SO2total column and 200 millibar (mb) for plume height.The RMSE was lower than the standard deviation of targets for the Grímsvötn eruption. The neural network had a lower retrieval accuracy when the target value was outside the values used during the training phase.
DOI: 10.5194/amt-7-4023-2014
发表时间: 2014-12
影响因子: 3.8
作者:
A. Piscini;M. Picchiani;M. Chini;S. Corradini;L. Merucci;F. Frate;S. Stramondo
通讯作者: A. Piscini;M. Picchiani;M. Chini;S. Corradini;L. Merucci;F. Frate;S. Stramondo
使用 MetOp 红外大气探测干涉仪演示了一种检测痕量物质的有效方法
DOI: --
发表时间: 2010
期刊:
影响因子: --
作者:
J. C. Walker;A. Dudhia;E. Carboni
通讯作者: E. Carboni
DOI: --
发表时间: 2001
期刊: IGARSS 2001. Scanning the Present and Resolving the Future. Proceedings. IEEE 2001 International Geoscience and Remote Sensing Symposium (Cat. No.01CH37217)
影响因子: --
作者:
F. DelFrate;A. Ortenzi;S. Casadio;Claus Zehner
通讯作者: Claus Zehner
DOI: 10.1016/j.jvolgeores.2003.12.017
发表时间: 2004
影响因子: 2.9
作者:
I. Watson;V. Realmuto;W.I Rose;A. Prata;G. Bluth;Y. Gu;C.E Bader;T. Yu
通讯作者: T. Yu
从大气红外探测器数据中反演火山 SO2 柱丰度
DOI: 10.1029/2006jd007955
发表时间: 2007
影响因子: --
作者:
A. Prata;C. Bernardo
通讯作者: C. Bernardo