基于Kolmogorov数据挖掘技术和SVM模型的FY-2G卫星反演产品冰雹识别方法研究

doi:10.11755/j.issn.1006-7639-2025-04-0646

干旱气象 ›› 2025, Vol. 43 ›› Issue (4): 646-653.DOI: 10.11755/j.issn.1006-7639-2025-04-0646

• 技术报告 • 上一篇

基于Kolmogorov数据挖掘技术和SVM模型的FY-2G卫星反演产品冰雹识别方法研究

彭宇翔¹(), 唐辟如²(), 周永水¹, 李皓¹, 刘涛³, 张超⁴, 喻乙耽¹, 郭茜⁵

1.贵州省气象台，贵州贵阳 550081
2.贵州省生态与农业气象中心，贵州贵阳 550002
3.贵州省山地气象科学研究所，贵州贵阳 550002
4.贵州省榕江县气象局，贵州榕江 552707
5.贵州省气象数据中心，贵州贵阳 550002

收稿日期:2024-05-14 修回日期:2024-06-26 出版日期:2025-08-31 发布日期:2025-09-08
通讯作者: 唐辟如（1992—），女，湖南洞口人，主要从事人工影响天气科研业务工作。E-mail：350337780@qq.com。
作者简介:彭宇翔（1990—），男，贵州黄平人，高级工程师，硕士，主要从事人工影响天气科研业务工作与统计算法研究。E-mail：1070792379@qq.com。
基金资助:
贵州省科技计划项目（课题）“基于风云卫星观测资料的冰雹天气识别研究”(黔科合基础-ZK[2021]一般217);贵州省气象局省市联合科研基金项目“基于高斯分布的风云卫星短临监测定量化冰雹识别指标研究”(黔气科合SS-QN[2024]03号);风云卫星应用先行计划(2023)-天气预报应用专项(FY-APP-ZX-2023.01);贵州省科技计划项目“基于相控阵双偏振雷达的贵州冰雹云识别研究”(黔科合基础-ZK[2023]一般200);中国气象局人工影响天气中心业务项目“FY3/4卫星云特性产品在西南防雹增雨中的应用示范(二期)”共同资助

Research on hail recognition using FY-2G satellite inversion products based on Kolmogorov data mining technology and SVM model

PENG Yuxiang¹(), TANG Piru²(), ZHOU Yongshui¹, LI Hao¹, LIU Tao³, ZHANG Chao⁴, YU Yidan¹, GUO Xi⁵

1. Guizhou Meteorological Observatory， Guiyang 550081， China
2. Guizhou Center for Ecology and Agricultural Meteorology， Guiyang 550002， China
3. Guizhou Institute of Mountain Meteorological Science， Guiyang 550002， China
4. Rongjiang County Meteorological Bureau of Guizhou Province， Rongjiang 552707， Guizhou， China
5. Guizhou Meteorological Data Center， Guiyang 550002， China

Received:2024-05-14 Revised:2024-06-26 Online:2025-08-31 Published:2025-09-08

摘要/Abstract

摘要：

冰雹识别技术研究对提前防御和减轻冰雹灾害具有重要意义。本文基于支持向量机（Support Vector Machine，SVM）模型和Kolmogorov变量筛选过滤器，利用FY-2G卫星的7项反演产品开展冰雹识别方法研究。以贵州省2020—2022年30个冰雹日368组未降雹点与降雹点FY-2G卫星反演产品数据作为数据集，分别基于Linear核函数、Radial Basis Function（RBF）核函数、Sigmoid核函数建立L-SVM模型、RBF-SVM模型、S-SVM模型开展冰雹识别，通过交叉检验提升模型冰雹识别结果的可靠性，并对识别准确率分布进行分析，利用Kolmogorov变量筛选过滤器优化模型输入参数。结果表明：3种核函数SVM模型均能对降雹点和未降雹点进行有效识别，且准确率均超过70%。其中，RBF-SVM模型对总样本和未降雹点样本的识别准确率最高，分别为87.50%和91.85%；S-SVM模型对降雹点识别准确率最高（89.13%）。利用Kolmogorov变量筛选过滤器优化模型输入参数后，模型识别准确率有不同程度提升。对于未降雹点数据集，优化输入参数后RBF-SVM模型和S-SVM模型识别准确率均达92.93%；对于降雹点数据集，S-SVM模型识别效果最好；对于总数据集识别效果最好的是优化输入参数后的RBF-SVM模型。综合识别效果最好的是输入参数优化后的RBF-SVM模型，若识别降雹区域则主要关注S-SVM模型识别结果。

关键词: SVM, Kolmogorov变量筛选过滤器, 变量筛选, 冰雹识别, 核函数

Abstract:

Research on hail identification technology is of great significance for preventing and mitigating hail disasters in advance. In this paper, based on support vector machine (SVM) model and Kolmogorov variable filter, 7 inversion products of FY-2G satellite are used to study hail recognition algorithm. The inversion product data of 368 sets of non hail spots and hail spots from FY-2G satellite on 30 hail days in Guizhou Province from 2020 to 2022 were taken as the data set, and the L-SVM model, RBF-SVM model and S-SVM model were established based on Linear Kernel Function, Radial Basis Function (RBF) and Sigmoid Kernel Function, respectively, to carry out hail identification. The reliability of hail identification results of the model was improved through cross-checking, and the distribution of recognition accuracy was analyzed, and the Kolmogorov variable filter was used to optimize the input parameters of the model. The results show that three kernel function SVM models can effectively identify hail and non hail falling points, and the accuracy rate is more than 70%. Among them, the RBF-SVM model has the highest recognition accuracy for total samples and non-hail spot samples, which are 87.50% and 91.85%, respectively. The S-SVM model is the most accurate in identifying hail points, with an accuracy of 89.13%. When the Kolmogorov variable filter is used to optimize the input parameters of the models, the recognition accuracy of most models is improved to different degrees. For the non-hail point data set, when input parameters are optimized, the RBF-SVM model and S-SVM model have the best recognition effect, and the accuracy rate of both is 92.93%; for hail point data set, the recognition effect of the S-SVM model is the best; for total data set, the RBF-SVM model has the best recognition effect. Therefore, in the actual business of hail identification and early warning, the RBF-SVM model after input parameters optimization has the best comprehensive recognition effect. If the focus is to identify hail areas, the identification results of S-SVM model can be mainly concerned.

Key words: SVM, Kolmogorov variable filter, variable screening, hail identification, kernel function

中图分类号:

彭宇翔, 唐辟如, 周永水, 李皓, 刘涛, 张超, 喻乙耽, 郭茜. 基于Kolmogorov数据挖掘技术和SVM模型的FY-2G卫星反演产品冰雹识别方法研究[J]. 干旱气象, 2025, 43(4): 646-653.

PENG Yuxiang, TANG Piru, ZHOU Yongshui, LI Hao, LIU Tao, ZHANG Chao, YU Yidan, GUO Xi. Research on hail recognition using FY-2G satellite inversion products based on Kolmogorov data mining technology and SVM model[J]. Journal of Arid Meteorology, 2025, 43(4): 646-653.

图/表 7

图1 数据分类与反演产品分布散点图

Fig.1 Scatter diagrams of data classification and inversion products distribution

图2 L-SVM模型、RBF-SVM模型与S-SVM模型对未降雹点与降雹点识别结果分布（虚线为识别概率50%分界线）

Fig.2 Distribution of hail and non hail points identification results based on L-SVM model， RBF-SVM model and S-SVM model （The dotted line is the boundary line of 50% identification probability）

表1 3个模型识别结果统计

Tab.1 Statistics of identification results based on three models

模型	分类	样本数量	识别为未降雹	识别为降雹	未降雹点识别准确率/%	降雹点识别准确率/%	总识别准确率/%
L-SVM	未降雹	184	162	22	88.04		86.14
L-SVM	降雹	184	29	155		84.24	86.14
RBF-SVM	未降雹	184	169	15	91.85		87.50
RBF-SVM	降雹	184	31	153		83.15	87.50
S-SVM	未降雹	184	134	50	72.83		80.98
S-SVM	降雹	184	20	164		89.13	80.98

图3 3个模型对未降雹点的识别概率分布（红色竖线为识别概率50%分界线，下同）

Fig.3 The distribution of three models’recognition probability for non-hail spots （The red vertical line is the boundary line of 50% identification probability， the same as below）

图4 3个模型对降雹点的识别概率分布

Fig.4 The distribution of three models’ recognition probability for hail spots

图5 FY-2G卫星7项反演产品与2个随机变量（S1和S2）的K值

Fig.5 The K values of 7 inversion products of FY-2G satellite and 2 random variables （S1 and S2）

表2 Statistics of model identification results before and after variable optimization 单位：%

Tab.2

模型	未降雹点识别准确率	降雹点识别准确率	总识别准确率
L-SVM	88.04	84.24	86.14
LK-SVM	88.59	84.24	86.41
RBF-SVM	91.85	83.15	87.50
RBFK-SVM	92.93	83.70	88.04
S-SVM	72.83	89.13	80.98
SK-SVM	92.93	79.35	86.14

参考文献 30

[1]	曹晓云, 周秉荣, 周华坤, 等, 2022. 气候变化对青藏高原植被生态系统的影响研究进展[J]. 干旱气象, 40(6):1068-1 080.
[2]	冯蜀青, 王海娥, 柳艳香, 等, 2019. 西北地区未来10 a气候变化趋势模拟预测研究[J]. 干旱气象, 37(4):557-564.
[3]	冯晓莉, 马占良, 管琴, 等, 2021. 1980—2018年青海高原冰雹分布特征及其关键影响因素分析[J]. 气象, 47(6):717-726.
[4]	彭宇翔, 文继芬, 李皓, 等, 2021. 基于决策树模型的贵州降雹识别研究[J]. 中低纬山地气象, 45(6): 99-101.
[5]	彭宇翔, 吕建永, 顾赛菊, 2016. 利用支持向量机预测大磁暴期间Dst指数的变化[J]. 空间科学学报, 36(6):866-874.
[6]	王姝, 张亮, 朱红秀, 等, 2024. 2022年7—8月干旱对川西高原植被长势的影响[J]. 高原山地气象研究, 44(1):94-103.
[7]	易雪, 杨森, 刘鸣彦, 等, 2021. 辽宁省植被覆盖度时空变化特征及其对气候变化的响应[J]. 干旱气象, 39(2):252-261.
[8]	张强, 张存杰, 白虎志, 等, 2010. 西北地区气候变化新动态及对干旱环境的影响:总体暖干化,局部出现暖湿迹象[J]. 干旱气象, 28(1):1-7.
[9]	朱生翠, 周秉荣, 魏永林, 等, 2020. 气候变化对青海湖北岸天然牧草生长发育的影响[J]. 干旱气象, 38(5): 804-809.
[10]	CANDES E, TAO T, 2007. The dantzig selector: Statistical estimation when p is much larger than n[J]. The Annals of Statistics, 35(6): 2 313-2 351.
[11]	COLLIER C G, LILLEY R E, 1994. Forecasting thunderstorm initiation in north-west Europe using thermodynamic indices, satellite and radar data[J]. Meteorological Applications, 1(1): 75-84.
[12]	FAN J Q, LI R Z, 2001. Variable selection via nonconcave penalized likeli-hood and its oracle properties[J]. Journal of the American Statistical Association, 96(456): 1 348-1 360.
[13]	HUNTRIESER H, SCHIESSER H H, SCHMID W, et al, 1997. Comparison of traditional and newly developed thunderstorm indices for Switzerland[J]. Weather and Forecasting, 12(1): 108-125.
[14]	JOHNS R H, DOSWELLⅢ C A, 1992. Severe local storms forecasting[J]. Weather and Forecasting, 7(4): 588-612.
[15]	LEE R R, PASSNER J E, 1993. The development and verification of TIPS: An expert system to forecast thunderstorm occurrence[J]. Weather and Forecasting, 8(2): 271-280.
[16]	LU J Y, PENG Y X, WANG M, 2016. Support vector machine combined with distance correlation learning for Dst forecasting during intense geomagnetic storms[J]. Planetary and Space Science, 120: 48-55.
[17]	LÓPEZ L, GARCIA-ORTEGA E, SÁNCHEZ J L, 2007. A short-term forecast model for hail[J]. Atmospheric Research, 83(2/3/4): 176-184.
[18]	LÓPEZ L, MARCOS J L, SÁNCHEZ J L, et al, 2001. CAPE values and hailstorms on northwestern Spain[J]. Atmospheric Research, 56(1/2/3/4): 147-160.
[19]	MAI Q, ZOU H, 2013. The Kolmogorov filter for variable screening in high-dimensional binary classification[J]. Biometrika, 100(1): 229-234.
[20]	MCNULTY R P, 1995. Severe and convective weather: A central region forecasting challenge[J]. Weather and Forecasting, 10(2): 187-202.
[21]	NEUMANN C J, 1971. The thunderstorm forecasting system at the kennedy space center[J]. Journal of Applied Meteorology, 10(5): 921-936.
[22]	ORLANSKI I, 1975. A rational subdivision of scales for atmospheric processes[J]. Bulltin of the American Meteorological Society, 56: 527-530.
[23]	REAP R M, FOSTER D S, 1979. Automated 12-36 hour probability forecasts of thunderstorms and severe local storms[J]. Journal of Applied Meteorology, 18(10): 1 304-1 315.
[24]	REAP R M, MACGORMAN D R, 1989. Cloud-to-ground lightning: Climatological characteristics and relationships to model fields, radar observations, and severe local storms[J]. Monthly Weather Review, 117(3): 518-535.
[25]	TIBSHIRANI R, 1996. Regression shrinkage and selection via the lasso[J]. Journal of the Royal Statistical Society Series B: Statistical Methodology, 58(1): 267-288.
[26]	TUDURÍ E, RAMIS C, 1997. The environments of significant convective events in the western Mediterranean[J]. Weather and Forecasting, 12(2): 294-306.
[27]	VAPNIK V N, 1998. Statistical learning theory[M]. New York: John Wiley&SonsInc.
[28]	VAPNIK V N, 1995. The nature of statistical learning theory[M]. New York: Springer-Verlag.
[29]	YANG F, WHITE M A, MICHAELIS A R, et al, 2006. Nemani, prediction of continental-scale evapotranspiration by combining MODIS and AmeriFlux data through support vector machine[J]. IEEE Transactions on Geoscience and Remote Sensing, 44(11): 3 452-3 461.
[30]	ZOU H, 2006. The adaptive lasso and its oracle properties[J]. Journal of the American Statistical Association, 101(476): 1 418-1 429.

基于Kolmogorov数据挖掘技术和SVM模型的FY-2G卫星反演产品冰雹识别方法研究

Research on hail recognition using FY-2G satellite inversion products based on Kolmogorov data mining technology and SVM model

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 7

参考文献 30

相关文章 1

编辑推荐

Metrics