【摘要】 对于火山灰云顶高度的反演问题本质上是一个不确定性反演问题,不论是一维变分的方法还是线性统计回归方法,在复杂气象条件下无法获得高精度解。为了减少对热红外辐射传输模式、物理反演方法以及辅助气象数据的依赖,提出了一种基于BP神经网络的火山灰云高度遥感反演方法。以2010年5月8日发生在冰岛的艾雅法拉火山爆发为例,利用覆盖研究区域的Meteosat-8卫星的SEVIRI仪器数据资料和美国NOAA开发的GOES-R火山灰云顶高度算法反演的火山灰云高度作为训练样本,建立了基于共轭梯度法和L-M法两种训练方法的优化神经网络反演方法体系。研究结果表明:在迭代次数相同的条件下,利用共轭梯度法的反演结果均方误差要优于L-M算法。我国新一代静止气象卫星风云四号成像仪具有与SEVIRI仪器相似的红外通道设置,本研究对建立基于我国新一代静止气象卫星FY-4数据的火山灰云顶高度定量反演模型也具有重要的借鉴意义。更多还原

【Abstract】 The inversion of the height of the volcanic ash cloud top is essentially an uncertain inversion problem. No matter the one-dimensional variation method or the linear statistical regression method, the high-precision solution can not be obtained under the complex meteorological conditions. In order to reduce the dependence on thermal infrared radiation transfer mode, physical inversion method and auxiliary meteorological data, a remote sensing inversion method of volcanic ash cloud height based on BP neural network is proposed. Taking the Eyjafjallajokull eruption in Iceland on May 8, 2010 as an example, an optimized neural network inversion method system based on conjugate gradient method and L-M method was established by using the Seviri instrument data of Meteosat-8 satellite covering the research area and the height of ash cloud retrieved by goes-r algorithm developed by NOAA as training sample. The results show that the mean square error of the inversion results of the conjugate gradient method is better than that of the L-M algorithm under the same number of iterations. China’s new generation of geostationary meteorological satellite fengyun-4 imager has similar infrared channel settings with seviri instruments. This study is also of great significance for the establishment of quantitative inversion model of volcanic ash cloud top height based on FY-4 data of China’s new generation of geostationary meteorological satellite.更多还原