An RBF neural network approach for retrieving atmospheric extinction coefficients based on lidar measurements
An RBF neural network approach for retrieving atmospheric extinction coefficients based on lidar measurements
复制标题
DOI：
10.1007/s00340-018-7055-1
复制
发表时间：
2018-08
期刊：
影响因子：
--
通讯作者：
Hongxu Li;Jianhua Chang;Fan Xu;Binggang Liu;Zhenxing Liu;Lingyan Zhu;Zhenbo Yang
中科院分区：
文献类型：
--
作者：
Hongxu Li;Jianhua Chang;Fan Xu;Binggang Liu;Zhenxing Liu;Lingyan Zhu;Zhenbo Yang
激光雷达是一种有效的遥感方法，可以获得气溶胶的光学特性，如气溶胶消光系数（AEC），气溶胶光学厚度（AOD），以及相关的大气能见度。然而，提高激光雷达数据检索的准确性和效率仍然具有挑战性，由于在确定AEC边界值（AEC-BV）和气溶胶散射-后向散射比（AEBR），以及所需的复杂和耗时的计算相关的不确定性。在本文中，我们提出了一种新的方法，反馈径向基函数（RBF-FB），用于检索高精度AEC轮廓的基础上径向基函数神经网络。首先，使用正割法，我们确定准确的值为AEC-BV和AEBR，并生成AEC配置文件的Fernald方法。然后，我们选择了一组激光雷达信号及其相应的AEC轮廓作为学习样本的网络训练，建立一个RBF网络模型的AEC检索。接下来，我们通过引入反馈机制来校正网络输出，该反馈机制使用由太阳光度计测量的AOD作为误差标准。实测信号的测试结果表明，所提出的RBF-FB模型的输出与Fernald方法一致，具有快速性和鲁棒性的优点。
Lidar is an effective remote sensing method for obtaining the optical properties of aerosols, such as the aerosol extinction coefficient (AEC), the aerosol optical depth (AOD), and the related atmospheric visibility. However, improving the accuracy and efficiency of lidar data retrieval remains challenging due to the uncertainties associated in determining the AEC boundary value (AEC-BV) and the aerosol extinction-to-backscatter ratio (AEBR), as well as the complex and time-consuming calculations required. In this paper, we propose a novel method, a feedback radial basis function (RBF-FB), for retrieving high-precision AEC profiles based on a radial basis function neural network. First, using the secant method, we determine accurate values for AEC-BV and AEBR, and generate the AEC profiles by the Fernald method. We then choose a set of lidar signals and their corresponding AEC profiles as learning samples for network training and establish an RBF network model for AEC retrieval. Next, we correct the network output by introducing a feedback mechanism that uses the AOD measured by a sun photometer as the error criterion. Tests on measured signals confirm that the outputs of the proposed RBF-FB model are consistent with the Fernald method and have the advantages of speed and robustness.