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 ›› 2020, Vol. 36 ›› Issue (3): 106-112.doi: 10.3969/j.issn.1673-503X.2020.03.015

• 快报 • 上一篇    

基于多种模型的云南元江哈尼云海景观预报研究

丁圣1(),段玮2,*(),朱勇3,李刚1   

  1. 1. 元江县气象局, 云南 玉溪 653300
    2. 云南省气象科学研究所, 云南 昆明 650034
    3. 云南省气候中心, 云南 昆明 650034
  • 收稿日期:2019-10-10 出版日期:2020-06-30 发布日期:2020-07-09
  • 通讯作者: 段玮 E-mail:3499428@qq.com;duanwain@hotmail.com
  • 作者简介:丁圣,男, 1981年生,高级工程师,主要从事应用气象研究, E-mail:3499428@qq.com
  • 基金资助:
    国家自然科学基金项目(91537212);国家自然科学基金项目(41665004);国家自然科学基金项目(41565002);云南省气象局推广应用项目(YZ201907);省部合作重点工程建设项目“云南省高原特色农业气象服务体系建设”(云财农[2016]18号);云南省中青年学术和技术带头人后备人才项目(2017HB040);云南省科技计划项目(2017FB076)

Forecasting of Ha'ni cloud sea landscape in Yuanjiang county of Yunnan province based on multiple models

Sheng DING1(),Wei DUAN2,*(),Yong ZHU3,Gang LI1   

  1. 1. Meteorological Service of Yuanjiang County, Yuanjiang 653300, China
    2. Yunnan Institute of Meteorological Science, Kunming 650034, China
    3. Yunnan Climate Center, Kunming 650034, China
  • Received:2019-10-10 Online:2020-06-30 Published:2020-07-09
  • Contact: Wei DUAN E-mail:3499428@qq.com;duanwain@hotmail.com

摘要:

为推动区域旅游事业发展,满足旅游气象服务需求,研究针对云南元江哈尼云海景观进行了立体气象观测和业务预报实验。本研究基于云南元江云海气候站2016—2019年观测数据,利用经验预报法、Logistics回归、支持向量机、决策树分析等方法,进行了云海景观出现与否的二分类预报实验。结果表明:各预报方法间训练样本总体准确率在74.3%—82.2%之间差别不大,但传统经验预报基于云海机理研究背景,预报指标物理意义明确,随着预报经验的积累经验预报2019年TS评分为54.8,优于2016—2018年TS评分46.0,也优于仅使用局地数据的统计学习算法的预报评分,且其他几种统计学习预报方法的检验样本TS评分均不如训练样本评分高。云海景观出现需要水汽条件和大气静稳条件的配合,局地云海气象观测站建设收集的立体气候数据有利于预报人员改进预报指标体系,提高预报准确率,有利于提升区域旅游气象服务能力发展。

关键词: 云海, 预报, 逻辑回归, 决策树, 支持向量机

Abstract:

For promoting the development of local tourism industry and satisfying the demand for tourism meteorological services, the three-dimensional (3D) meteorological observation and the forecast experiments for the Ha'ni cloud sea landscape in Yuanjiang county were conducted.Based on the meteorological observation data from the weather station in the studied area during 2016-2019, the dichotomous prediction experiments for the cloud sea landscape appearing or not were carried out by using the empirical forecasting, the logistics regression, the support vector machine (SVM), the decision tree analysis and other analysis methods.The results show that the overall forecast accuracies of the training samples based on the different prediction methods are between 74.3% and 82.2% with a small difference, but the prediction indicator of the traditional empirical prediction based on the research background of the cloud sea mechanism is of a definite physical meaning.With the accumulation of the prediction experience, the TS score of empirical prediction i.e.54.8 in 2019 is better than that i.e.46.0 in 2016-2018, and also better than that based on the statistical learning algorithm using only local data.In addition, the TS scores for the test sample employing the other several learning forecast methods are lower than that for the training sample.The occurrence of cloud sea landscape requires the coordination of both water vapor and atmospheric static stability.Besides, the collected 3D climate data from the local cloud sea observatory is beneficial for improving the forecast index system, increasing forecast accuracy and promoting the development of local tourism industry.

Key words: Cloud sea, Forecast, Logistics regression, Decision tree, Supporot vector machine (SVM)

中图分类号: