Fitting precipitation particle size-velocity data to mixed joint probability density function with an expectation maximization algorithm
Fitting precipitation particle size-velocity data to mixed joint probability density function with an expectation maximization algorithm
复制标题
使用期望最大化算法将降水粒径-速度数据拟合到混合联合概率密度函数
DOI：
10.1175/jtech-d-19-0150.1
复制
发表时间：
2020
影响因子：
2.2
通讯作者：
Inatsu Masaru
中科院分区：
文献类型：
--
作者：
Katsuyama Yuta;Inatsu Masaru
本文提出了一种基于多类型降水粒子采样资料的节理尺寸和终端速度分布的估算方法。假设速度服从正态分布，尺寸服从伽玛分布，该方法使用期望最大化算法在实际参数范围内搜索局部最大对数似然。用实际数量的元素制备了几个测试群体，然后通过从其样品中检索群体来评价该方法。结果表明，在大多数情况下，初始参数的估计是成功的测试人口包含一些液体，谷粒，和镶边和未镶边团聚体。元素的原始数量也是通过调整元素的数量来估计的，调整的方式是使得元素的每个少数分数都超过阈值。该方法应用于二维降速仪观测数据，有助于剔除掉降速过大的常观测错误数据。
This paper proposes an estimation method of joint size and terminal velocity distribution on the basis of sampling data of precipitation particles containing multiple types. Assuming that the velocity follows the normal distribution and the size follows the gamma distribution, the method searches a locally maximum logarithmic likelihood within a realistic parameter range using the expectation–maximization algorithm. Several test populations were prepared with a realistic number of elements, and then the method was evaluated by retrieving the populations from their sample. The results showed that the original parameters were successfully estimated in most cases of the test population containing some of liquids, graupels, and rimed and unrimed aggregates. The original number of elements was also estimated with an adjustment of the number of elements in a manner such that each of their minority fractions exceeded a threshold. Applied to the two-dimensional disdrometer observation data, the method was helpful to discard frequently observed erroneous data with unrealistically large fall velocity.