26 November 2018 San Rafael HailPixel Survey Data and Analysis

The following collection is used to demonstrate the HailPixel survey technique as part of an AGU GRL publication. The following is an abstract for this paper: A new technique, named "HailPixel," is introduced for measuring the maximum dimension and intermediate dimension of hailstones from aerial imagery. The photogrammetry procedure applies a convolutional neural network for robust detection of hailstones against complex backgrounds and an edge detection method for measuring the shape of identified hailstones. This semi-automated technique is capable of measuring many thousands of hailstones within a single survey, which is several orders of magnitude larger than population sizes from existing sensors (e.g., a hail pad). Comparison with a co-located hail pad for an Argentinan hailstorm event demonstrates the larger population size of the HailPixel survey significantly improves the shape and tails of the observed hail size distribution. When hailfall is sparse, such as during large and giant hail events, the large survey area of this technique is especially advantageous for resolving the hail size distribution. The dataset contains the DEM and orthomosaic imagery, processing reports, final location of hail centroids, final measurements of hail major and minor axis, subset offsets and hail pad data. For more information applying the subset offsets to calculate the true position of the hail centroids please see the paper.

本数据集用于演示作为美国地球物理联合会（American Geophysical Union, AGU）《地球物理研究快报》（Geophysical Research Letters, GRL）论文组成部分的HailPixel探测技术。以下为该论文的摘要：本研究提出一种名为‘HailPixel’的新技术，可通过航空影像测量冰雹的最大直径与中间直径。该摄影测量流程采用卷积神经网络（Convolutional Neural Network）在复杂背景下稳健检测冰雹，并结合边缘检测方法对已识别冰雹的形状进行测量。该半自动化技术单次探测即可完成数千颗冰雹的尺寸测量，其样本种群规模较现有传感器（如冰雹垫）大数个数量级。通过与阿根廷一次冰雹事件中同步布设的冰雹垫开展对比，结果表明HailPixel探测所获得的更大样本量，显著优化了观测到的冰雹尺寸分布的形态与尾部特征。当冰雹降落较为稀疏时，例如在大冰雹和巨型冰雹事件中，该技术的大面积探测范围对解析冰雹尺寸分布尤为有利。本数据集包含数字高程模型（Digital Elevation Model, DEM）与正射镶嵌影像（Orthomosaic Imagery）、处理报告、冰雹质心的最终定位结果、冰雹长轴与短轴的最终测量值、子集偏移量数据以及冰雹垫观测数据。如需了解如何应用子集偏移量计算冰雹质心的真实位置，请参阅该论文。