In this paper, we describe a new two-dimensional and multi-channel feature detection algorithm (2D-McDA) and demonstrate its application to lidar backscatter measurements from the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) mission. Unlike previous layer detection schemes, this context-sensitive feature finder algorithm is applied to a 2-D lidar “scene”, i.e., to the image formed by many successive lidar profiles. Features are identified when an extended and contiguous 2-D region of enhanced backscatter signal rises significantly above the expected “clear air” value. Using an iterated 2-D feature detection algorithm dramatically improves the fine details of feature shapes and can accurately identify previously undetected layers (e.g., subvisible cirrus) that are very thin vertically but horizontally persistent. Because the algorithm looks for contiguous 2-D patterns using successively lower detection thresholds, it reports strongly scattering features separately from weakly scattering features, thus potentially offering improved discrimination of juxtaposed cloud and aerosol layers. Moreover, the 2-D detection algorithm uses the backscatter signals from all available channels: 532 nm parallel, 532 nm perpendicular and 1064 nm total. Since the backscatter from some aerosol or cloud particle types can be more pronounced in one channel than another, simultaneously assessing the signals from all channels greatly improves the layer detection. For example, ice particles in subvisible cirrus strongly depolarize the lidar signal and, consequently, are easier to detect in the 532 nm perpendicular channel. Use of the 1064 nm channel greatly improves the detection of dense smoke layers, because smoke extinction at 532 nm is much larger than at 1064 nm, and hence the range-dependent reduction in lidar signals due to attenuation occurs much faster at 532 nm than at 1064 nm. Moreover, the photomultiplier tubes used at 532 nm are known to generate artifacts in an extended area below highly reflective liquid clouds, introducing false detections that artificially lower the apparent cloud base altitude, i.e., the cloud base when the cloud is transparent or the level of complete attenuation of the lidar signal when it is opaque. By adding the information available in the 1064 nm channel, this new algorithm can better identify the true apparent cloud base altitudes of such clouds.

中文翻译：

---

CALIPSO激光雷达测量的二维和多通道特征检测算法

在本文中，我们描述了一种新的二维和多通道特征检测算法（2D-McDA），并展示了其在云气溶胶激光雷达和红外探路者卫星观测（CALIPSO）任务进行的激光雷达后向散射测量中的应用。与以前的图层检测方案不同，此上下文相关的特征查找器算法应用于二维激光雷达“场景”，即应用于许多连续的激光雷达轮廓所形成的图像。当增强的反向散射信号的扩展且连续的二维区域明显高于预期的“晴朗空气”值时，将识别特征。使用迭代的二维特征检测算法可以显着改善特征形状的精细细节，并且可以准确地识别以前未检测到的垂直（水平）持久性很薄的层（例如，亚可见卷云）。因为该算法使用相继较低的检测阈值来寻找连续的2D模式，所以它报告的强散射特征与弱散射特征分开，从而有可能提供更好的分辨并列的云层和气溶胶层的能力。此外，二维检测算法使用来自所有可用通道的反向散射信号：平行532 nm，垂直532 nm和总计1064 nm。由于某些气溶胶或云颗粒类型的反向散射在一个通道中比在另一个通道中更明显，因此同时评估来自所有通道的信号会大大改善层检测。例如，亚可见卷云中的冰粒使激光雷达信号强烈去极化，因此，在532 nm垂直通道中更容易检测到。1064 nm通道的使用大大改善了浓烟层的检测，因为在532 nm处的烟消光远大于在1064 nm处的烟消光，因此，由于衰减引起的激光雷达信号在距离范围内的减少在532 nm处的发生比在15 nm处的发生要快得多。 1064 nm。此外，已知在532 nm处使用的光电倍增管在高反射性液态云下方的扩展区域中会产生伪影，从而引入错误的检测，从而人为地降低了明显的云底高度，即当云透明或云层水平时的云底高度。不透明的激光雷达信号完全衰减。通过添加1064 nm通道中可用的信息，此新算法可以更好地识别此类云的真实表观云基准高度。因为在532 nm处的烟消光远大于在1064 nm处的烟消光，因此在532 nm处由于衰减引起的激光雷达信号在距离范围内的减小比在1064 nm处快得多。此外，已知在532 nm处使用的光电倍增管在高反射性液态云下方的扩展区域中会产生伪影，从而引入错误的检测，从而人为地降低了明显的云底高度，即当云透明或云层水平时的云底高度。不透明的激光雷达信号完全衰减。通过添加1064 nm通道中可用的信息，此新算法可以更好地识别此类云的真实表观云基准高度。因为在532 nm处的烟消光远大于在1064 nm处的烟消光，因此在532 nm处由于衰减引起的激光雷达信号在距离范围内的减小比在1064 nm处快得多。此外，已知在532 nm处使用的光电倍增管会在高反射性液态云下方的扩展区域中产生伪影，从而引入错误检测，从而人为地降低了表观云底高度，即当云层透明或云层水平降低时的云底高度。不透明的激光雷达信号完全衰减。通过添加1064 nm通道中可用的信息，此新算法可以更好地识别此类云的真实表观云基准高度。众所周知，在532 nm处使用的光电倍增管会在高反射性液态云下方的扩展区域中产生伪影，从而引入错误的检测，从而人为地降低了明显的云底高度，即云透明或完全衰减时的云底高度。不透明时的激光雷达信号的变化。通过添加1064 nm通道中可用的信息，此新算法可以更好地识别此类云的真实表观云基准高度。众所周知，在532 nm处使用的光电倍增管会在高反射性液态云下方的扩展区域中产生伪影，从而引入错误的检测，从而人为地降低了明显的云底高度，即云透明或完全衰减时的云底高度。不透明时的激光雷达信号的变化。通过添加1064 nm通道中可用的信息，此新算法可以更好地识别此类云的真实表观云基准高度。