Arrange and average algorithm for the retrieval of aerosol parameters from multiwavelength high-spectral-resolution lidar/Raman lidar data.
Arrange and average algorithm for the retrieval of aerosol parameters from multiwavelength high-spectral-resolution lidar/Raman lidar data.
复制标题
DOI：
10.1364/ao.53.007252
复制
发表时间：
2014-11
期刊：
影响因子：
1.9
通讯作者：
E. Chemyakin;D. Müller;S. Burton;A. Kolgotin;C. Hostetler;R. Ferrare
中科院分区：
文献类型：
--
作者：
E. Chemyakin;D. Müller;S. Burton;A. Kolgotin;C. Hostetler;R. Ferrare
我们展示了一项可行性研究的结果，其中使用一种简单、自动化和无监督的算法（我们称之为排列和平均算法）来推断大气气溶胶颗粒的微物理参数（复折射率、有效半径、总数、表面积和体积浓度）。该算法使用 355、532 和 1064 nm 处的反向散射系数以及 355 和 532 nm 处的消光系数作为输入信息。该算法的测试基于合成光学数据，这些数据是根据规定的单峰粒径分布和描述球形（主要是精细模式污染颗粒）的复折射率计算得出的。我们测试了多波长气溶胶高光谱分辨率激光雷达 (HSRL) 或拉曼激光雷达的“3 反向散射 (β)+2 消光 (α)”配置的算法性能。我们研究了该算法检索的微物理结果在多大程度上取决于输入反向散射和消光系数的数量。例如，我们测试了“3β+1α”、“2β+1α”和“3β”激光雷达配置。这种排列和平均算法可以以两种方式使用。首先，它可以应用于激光雷达获取的实验数据的快速数据处理。鉴于 NASA 兰利研究中心的机载“3β+2α”高光谱分辨率激光雷达 (HSRL-2) 可以获取大量数据，因此需要快速自动检索微物理粒子属性。事实证明，它对于越来越多的地面多波长激光雷达网络很有用，并且将为分析未来星载多波长激光雷达获取的大量光学数据提供一种选择。第二个潜在应用是通过我们现有的反演算法来改进微物理粒子表征，该算法使用吉洪诺夫正则化反演。这种先进的算法最近得到了发展，可以实现自动化和无监督的处理；排列平均算法可以作为预分类器，进一步提高其速度和精度。对排列和平均算法性能的首次测试令人鼓舞。我们使用了一组 48 种不同的单峰粒径分布、复折射率的 4 个实部和 15 个虚部。总而言之，我们针对 0%、10% 和 20% 高斯测量噪声（一标准差）测试了 2880 个不同的光学数据集。在具有 10% 测量噪声的“3β+2α”配置的情况下，我们将 1964 年（68.2%）测试光学数据集的粒子有效半径恢复到 27% 以内。获得的数量浓度达到 76%，表面积浓度达到 16%，体积浓度达到 30% 精度。 “3β”配置的性能明显较差。 “3β+1α”和“2β+1α”配置的性能介于“3β+2α”和“3β”之间。
We present the results of a feasibility study in which a simple, automated, and unsupervised algorithm, which we call the arrange and average algorithm, is used to infer microphysical parameters (complex refractive index, effective radius, total number, surface area, and volume concentrations) of atmospheric aerosol particles. The algorithm uses backscatter coefficients at 355, 532, and 1064 nm and extinction coefficients at 355 and 532 nm as input information. Testing of the algorithm is based on synthetic optical data that are computed from prescribed monomodal particle size distributions and complex refractive indices that describe spherical, primarily fine mode pollution particles. We tested the performance of the algorithm for the "3 backscatter (β)+2 extinction (α)" configuration of a multiwavelength aerosol high-spectral-resolution lidar (HSRL) or Raman lidar. We investigated the degree to which the microphysical results retrieved by this algorithm depends on the number of input backscatter and extinction coefficients. For example, we tested "3β+1α," "2β+1α," and "3β" lidar configurations. This arrange and average algorithm can be used in two ways. First, it can be applied for quick data processing of experimental data acquired with lidar. Fast automated retrievals of microphysical particle properties are needed in view of the enormous amount of data that can be acquired by the NASA Langley Research Center's airborne "3β+2α" High-Spectral-Resolution Lidar (HSRL-2). It would prove useful for the growing number of ground-based multiwavelength lidar networks, and it would provide an option for analyzing the vast amount of optical data acquired with a future spaceborne multiwavelength lidar. The second potential application is to improve the microphysical particle characterization with our existing inversion algorithm that uses Tikhonov's inversion with regularization. This advanced algorithm has recently undergone development to allow automated and unsupervised processing; the arrange and average algorithm can be used as a preclassifier to further improve its speed and precision. First tests of the performance of arrange and average algorithm are encouraging. We used a set of 48 different monomodal particle size distributions, 4 real parts and 15 imaginary parts of the complex refractive index. All in all we tested 2880 different optical data sets for 0%, 10%, and 20% Gaussian measurement noise (one-standard deviation). In the case of the "3β+2α" configuration with 10% measurement noise, we retrieve the particle effective radius to within 27% for 1964 (68.2%) of the test optical data sets. The number concentration is obtained to 76%, the surface area concentration to 16%, and the volume concentration to 30% precision. The "3β" configuration performs significantly poorer. The performance of the "3β+1α" and "2β+1α" configurations is intermediate between the "3β+2α" and the "3β."