主办单位:中国气象局沈阳大气环境研究所
国际刊号:ISSN 1673-503X
国内刊号:CN 21-1531/P

气象与环境学报 ›› 2022, Vol. 38 ›› Issue (3): 112-118.doi: 10.3969/j.issn.1673-503X.2022.03.013

• 论文 • 上一篇    下一篇

基于CART决策树的河北省低能见度分类研究

尤琦1(),曲晓黎1,2,*(),赵增保1,王洁1,张娣1,杨琳晗1   

  1. 1. 河北省气象服务中心, 河北 石家庄 050021
    2. 河北省气象与生态环境重点实验室, 河北 石家庄 050021
  • 收稿日期:2021-04-08 出版日期:2022-06-28 发布日期:2022-07-23
  • 通讯作者: 曲晓黎 E-mail:369956658@qq.com;hebqx_quxiaoli@126.com
  • 作者简介:尤琦, 男, 1994年生, 工程师, 主要从事交通气象服务研究, E-mail: 369956658@qq.com
  • 基金资助:
    河北省省级科技计划“高速公路复杂路面高分辨率恶劣天气精准预警技术研究”(19275413D);河北省自然基金面上项目“基于模式自适应源反演和TROPOMI卫星资料同化的空气质量优化预报研究”(D2020304038)

Low visibility classification based on CART decision tree in Hebei province

Qi YOU1(),Xiao-li QU1,2,*(),Zeng-bao ZHAO1,Jie WANG1,Di ZHANG1,Lin-han YANG1   

  1. 1. Hebei Meteorological Service Centre, Shijiazhuang 050021, China
    2. Key Laboratory of Meteorological and Ecological Environment of Hebei Province, Shijiazhuang 050021, China
  • Received:2021-04-08 Online:2022-06-28 Published:2022-07-23
  • Contact: Xiao-li QU E-mail:369956658@qq.com;hebqx_quxiaoli@126.com

摘要:

基于2016—2019年河北省142个国家气象站逐小时观测数据, 通过EOF时空正交分解和CART决策树分类回归等方法, 针对低能见度高发区域构建能见度预报模型, 并进行拟合检验。结果表明: 河北省雾日时空分布特征显示除张家口、承德及秦皇岛三市外, 40°N以南地区为雾日高发区域, 多年平均雾日数最高值可达50 d。相对湿度、地表温度、风速等气象要素与能见度显著相关, 将显著相关因子作为输入变量建立能见度预报模型并调参, 经检验该模型对于冬季的预报效果较好, 有较高的准确率; 夏季误报率较低; 日夜差别在夏季并不明显, 三个指数差别不大, 冬季夜晚的准确率与误报率明显优于白天, 漏报率略高。石家庄站2019年12月7—10日的三次大雾过程拟合结果较好, 有雾时次无漏报。

关键词: 能见度, CART决策树, 拟合预报

Abstract:

Based on the hourly observation data of 142 national weather stations in Hebei province from 2016 to 2019, a visibility forecast model was constructed for low visibility and high-frequency areas using Empirical Orthogonal Function and CART decision tree classification and regression methods, and a fitting test was carried out.The results show that except in Zhangjiakou, Chengde, and Qinhuangdao, fog days most appear in the areas south of 40°N, with the average peak of annual fog days more than 50 days.The relative humidity, surface temperature, wind speed, and other meteorological factors are significantly correlated with visibility.High correlation factors act as input variables to the forecast model, which is proved to have a good effect on the winter forecast with high accuracies.In summer, the false alarm rates are lower and the differences between day and night are less obvious, with little differences between the three indexes.Whereas in winter, the accuracy rates and false alarm rates are significantly superior to those in the daytime, and the missing alarm rates are slightly higher.The fitting results of three heavy fog processes on December 7-10, 2019, at Shijiazhuang station are good without any fogy time missed.

Key words: Visibility, CART decision tree, Fitting prediction

中图分类号: