设为首页
加入收藏
Hail Identification Technology in Eastern Hubei Based on Decision Tree Algorithm
-
摘要: 冰雹是对流性天气常见的灾害之一，雷达是识别冰雹强有利的工具，为克服现有方法主观性强、特征量阈值不明确以及虚警率高的不足，探究机器学习算法用于冰雹识别的可行性，基于决策树算法利用2015年1月1日—2021年12月31日鄂东地区冰雹灾情资料、武汉多普勒天气雷达以及探空资料，将湿球温度高度引入冰雹识别因子中，并根据命中率、虚警率和临界成功指数定量评估其识别能力。结果表明：仅包含回波强度的决策树（强度决策树）和包含回波强度和湿球温度高度的决策树（强度-高度决策树）均能有效识别冰雹，强度-高度决策树较强度决策树的命中率和临界成功指数均小幅提高，且虚警率明显降低；强度决策树识别冰雹的关键因子为组合反射率因子，底层多为0.5°和1.5°仰角反射率因子，强度-高度决策树的关键因子为0.5°仰角反射率因子，底层多为风暴的整体强度属性；个例分析显示强度-高度决策树减少了湿球0℃层高度较高时的虚警次数，展现出良好的应用前景。Abstract: Hail refers to the solid precipitation with a diameter greater than or equal to 5 mm caused by convection. Hail is also one main disastrous weather phenomenon in eastern Hubei, while Doppler weather radar is the most favorable tool for hail identification. At present, there are two hail identification methods used in the actual operation in eastern Hubei, one is artificial conceptual model, the other is the self-contained identification technology in short-time and proximity prediction system. The conceptual model needs to be judged by human, which is too subjective and threshold values of radar echo characteristic are not clear, while the false alarm rate(FAR) of existing automatic technologies in prediction system are too high. To overcome the shortcoming of the above methods, feasibilities of machine learning algorithms for hail identification are explored and a decision tree algorithm is established. Based on hail disaster data of Wuhan, Huanggang, Huangshi, Ezhou, Xianning and Xiaogan, Doppler weather radar data and convention high altitude sounding data of Wuhan from 2015 to 2021, the height of wet bulb 0℃(HWB0) and the height of wet bulb -20℃(HWB-20) are introduced into the hail identification factors, and artificial intelligence technology is applied in hail recognition. The performance is evaluated according to probability of detection(POD), FAR and critical success index(CSI). The result shows that both the decision tree algorithm with radar echo intensity (intensity decision tree) and the decision tree algorithm with radar echo intensity and wet bulb temperature height (intensity-height decision tree) can identify hail effectively. The POD results of the two decision tree algorithms are higher than 0.88, while the FAR are lower than 0.12, and the CSI are higher than 0.8, but the intensity-height decision tree performs better, with the POD and CSI increased by 5.68% and 7.5% than intensity decision tree respectively, while the FAR decrease 41.67%. The key factor of hail recognition by intensity decision tree is the combined reflectivity factor, and the bottom layer is the reflectivity factor of 0.5° and 1.5° elevation. The key factor of intensity-height decision tree is the reflectivity factor of 0.5° elevation and the judgment modules of radar echo extension height with the height of wet bulb temperature, especially with HWB0 included in the middle, and the bottom layer is the strength attributes of storm (vertically integrated liquid water and combined reflectivity). The analysis results of three cases with different occurrence time, location and hail size show that, due to the introduction of height of wet bulb temperature, the intensity-height decision tree reduces the number of empty alarm when the height of HWB0 and HWB-20 are high, especially when the HWB0 is high, thus reduces its FAR and improves its CSI, which indicate its potential wide prospect for operational application.
-
图 1 鄂东地区判别降雹的强度决策树
(a)Rc≤56.7 dBZ, (b)Rc>56.7 dBZ
Fig. 1 Decision tree of intensity for hail in eastern Hubei
(a)Rc≤56.7 dBZ, (b)Rc>56.7 dBZ
图 2 鄂东地区判别降雹的强度-高度决策树
(a)R0.5≤51.8 dBZ, (b)R0.5>51.8 dBZ
Fig. 2 Decision tree of intensity-height for hail in eastern Hubei
(a)R0.5≤51.8 dBZ, (b)R0.5>51.8 dBZ
图 3 2020年3月26日16：18武汉雷达组合反射率因子(相邻距离圈相距50 km, 下同)(a) 和过降雹点的反射率因子剖面(b)
Fig. 3 Combined reflectivity(the distance between adjacent rings is 50 km, hereinafter)(a) and reflectivity factor profile(b) of hail point of Wuhan radar at 1618 BT 26 Mar 2020
图 4 2020年3月26日16：18决策树的识别结果
(红色圆圈表示以降雹点为中心，半径为5 km的区域)
Fig. 4 Identification of decision tree at 1618 BT 26 Mar 2020
(red circle area denotes the hail location with a radius of 5 km)
图 5 2021年5月14日16：12武汉雷达组合反射率因子(a)和过降雹点的反射率因子剖面(b)
Fig. 5 Combined reflectivity(a) and reflectivity factor profile(b) of hail point of Wuhan radar at 1612 BT 14 May 2021
图 6 2021年5月14日16：12决策树识别结果
(红色圆圈表示以降雹点为中心半径为5 km的区域)
Fig. 6 Identification of decision tree at 1612 BT 14 May 2021
(red circle area denotes the hail location with a radius of 5 km)
图 7 2021年9月28日19：48武汉雷达组合反射率因子(a)和过汉川降雹点反射率因子剖面(b)
Fig. 7 Combined reflectivity(a) and reflectivity factor profile(b) of Hanchuan hail point of Wuhan radar at 1948 BT 28 Sep 2021
图 8 2021年9月28日19：48决策树识别结果
(红色圆圈表示以降雹点为中心半径为5 km的区域)
Fig. 8 Identification of decision tree at 1948 BT 28 Sep 2021
(red circle area denotes the hail location with a radius of 5 km)
表 1 非冰雹样本选取标准及对应的样本量
Table 1 Selection standard and corresponding quantity of non-hail samples
组合反射率因子/dBZ 样本量 40.0~44.9 63 45.0~49.9 158 50.0~54.9 188 55.0~59.9 158 ≥60.0 63 
下载: 导出CSV
表 2 不同决策树算法的评分
Table 2 Scores of different decision trees
决策树算法 命中率 虚警率 临界成功指数 强度决策树 0.88 0.12 0.80 强度-高度决策树 0.93 0.07 0.86
下载: 导出CSV
-
[1] 郑永光, 周康辉, 盛杰, 等.强对流天气监测预报预警技术进展.应用气象学报, 2015, 26(6):641-657. doi: 10.11898/1001-7313.20150601Zheng Y G, Zhou K H, Sheng J, et al. Advance in techniques of monitoring, forecasting and warning of severe convective weather. J Appl Meteor Sci, 2015, 26(6): 641-657. doi: 10.11898/1001-7313.20150601 [2] 徐双柱, 韦慧红. 关于强对流天气的几点思考. 暴雨灾害, 2016, 35(3): 197-202. doi: 10.3969/j.issn.1004-9045.2016.03.001Xu S Z, Wei H H. Some thoughts on the weather forecast of severe convective storms. Torrential Rain Disaste, 2016, 35(3): 197-202. doi: 10.3969/j.issn.1004-9045.2016.03.001 [3] 施望芝, 金琪, 郭施, 等. 湖北一次冰雹天气过程的落区诊断分析和预报. 热带气象学报, 2004, 20(2): 212-217. doi: 10.3969/j.issn.1004-4965.2004.02.014Shi W Z, Jin Q, Guo S, et al. An analysis and forecast for the area of an hail weather in Hubei Province. J Trop Meteor, 2004, 20(2): 212-217. doi: 10.3969/j.issn.1004-4965.2004.02.014 [4] 俞小鼎, 王迎春, 陈明轩, 等. 新一代天气雷达与强对流天气预警. 高原气象, 2005, 24(3): 456-464. doi: 10.3321/j.issn:1000-0534.2005.03.025Yu X D, Wang Y C, Chen M X, et al. Severe convective weather warnings and its improvement with the introduction of the NEXRAD. Plateau Meteor, 2005, 24(3): 456-464. doi: 10.3321/j.issn:1000-0534.2005.03.025 [5] 刁秀广, 朱君鉴, 黄秀韶, 等. VIL和VIL密度在冰雹云判据中的应用. 高原气象, 2008, 27(5): 1131-1139. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX200805023.htmDiao X G, Zhu J J, Huang X S, et al. Application of VIL and VIL density in warning criteria for hailstorm. Plateau Meteor, 2008, 27(5): 1131-1139. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX200805023.htm [6] 胡胜, 罗聪, 张羽, 等. 广东大冰雹风暴单体的多普勒天气雷达特征. 应用气象学报, 2015, 26(1): 57-65. doi: 10.11898/1001-7313.20150106Hu S, Luo C, Zhang Y, et al. Dopper radar features of severe hailstorm in Guangdong Province. J Appl Meteor Sci, 2015, 26(1): 57-65. doi: 10.11898/1001-7313.20150106 [7] 王萍, 潘跃. 基于显著性特征的大冰雹识别模型. 物理学报, 2013, 62(6): 515-524. https://www.cnki.com.cn/Article/CJFDTOTAL-WLXB201306077.htmWang P, Pan Y. Severe hail identification model based on saliency characteristics. Acta Physica Sinica, 2013, 62(6): 515-524. https://www.cnki.com.cn/Article/CJFDTOTAL-WLXB201306077.htm [8] 王莎, 沙勇, 宋金妹, 等. 冀东地区冰雹云多普勒雷达参数特征分析. 气象, 2019, 45(5): 713-722.Wang S, Sha Y, Song J M, et al. Characteristic analysis of hail cloud Doppler radar parameters in the eastern Hebei Province. Meteor Mon, 2019, 45(5): 713-722. [9] 吴剑坤, 陈明轩, 秦睿, 等. 变分回波跟踪算法及其在对流临近预报中的应用研究. 气象学报, 2019, 77(6): 999-1014. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201906003.htmWu J K, Chen M X, Qin R, et al. The veriatiobal echo tracking method and its application in convective storm nowcasting. Acta Meteor Sinica, 2019, 77(6): 999-1014. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201906003.htm [10] 王洪, 吴乃庚, 万齐林, 等. 变分回波跟踪算法及其在对流临近预报中的应用研究. 气象学报, 2018, 76(1): 92-103. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201906003.htmWang H, Wu N G, Wan Q L, et al. Analysis of S-band polarimetric radar observations of a hail-producing supercell. Acta Meteor Sinica, 2018, 76(1): 92-103. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201906003.htm [11] 冯晋勤, 张深寿, 吴陈锋, 等. 双偏振雷达产品在福建强对流天气过程中的应用分析. 气象, 2018, 44(12): 1565-1574. doi: 10.7519/j.issn.10000526.2018.12.006Feng J Q, Zhang S S, Wu C F, et al. Appliation of dual polarization weather radar products to severe convective weather in Fujian. Meteor Mon, 2018, 44(12): 1565-1574. doi: 10.7519/j.issn.10000526.2018.12.006 [12] 刁秀广, 李芳, 万夫敬. 两次强冰雹超级单体风暴双偏振特征对比. 应用气象学报, 2022, 33(4): 414-428. doi: 10.11898/1001-7313.20220403Diao X G, Li F, Wan F J. Comparative analysis on dual polarization features of two severe hail supercells. J Appl Meteor Sci, 2022, 33(4): 414-428. doi: 10.11898/1001-7313.20220403 [13] 徐舒扬, 吴翀, 刘黎平. 双偏振雷达水凝物相态识别算法的参数改进. 应用气象学报, 2020, 31(3): 350-360. doi: 10.11898/1001-7313.20200309Xu S Y, Wu C, Liu L P. Parameter improvements of hydrometeor classification algorithm for the dual-polarimetric radar. J Appl Meteor Sci, 2020, 31(3): 350-360. doi: 10.11898/1001-7313.20200309 [14] 俞小鼎, 郑永光. 中国当代强对流天气研究与业务进展. 气象学报, 2020, 78(3): 391-418. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB202003006.htmYu X D, Zheng Y G. Advances in severe convective weather research and operational service in China. Acta Meteor Sinica, 2020, 78(3): 391-418. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB202003006.htm [15] 王婷波, 周康辉, 郑永光. 我国中东部雷暴活动特征分析. 气象, 2020, 46(2): 189-199.Wang T B, Zhou K H, Zheng Y G. Statistic analysis of thunderstorm characteristics in central and eastern China. Meteor Mon, 2020, 46(2): 189-199. [16] 周康辉, 郑永光, 韩雷, 等. 机器学习在强对流监测预报中的应用进展. 气象, 2021, 47(3): 274-289. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX202103002.htmZhou K H, Zheng Y G, Han L, et al. Advances in application of machine learning to severe weather monitoring and forecasting. Meteor Mon, 2021, 47(3): 274-289. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX202103002.htm [17] 王瑾, 刘黎平. WSR-88D冰雹探测算法在贵州地区的评估检验. 应用气象学报, 2011, 22(1): 96-106. http://qikan.camscma.cn/article/id/20110110Wang J, Liu L P. The evaluation of WRS-88D hail detection algorithm over Guizhou Region. J Appl Meteor Sci, 2011, 22(1): 96-106. http://qikan.camscma.cn/article/id/20110110 [18] 周康辉. 基于卷积神经网络的强对流天气预报方法研究. 北京: 中国科学院大学, 2021.Zhou K H. Convective Weather Forecasting with Convolutional Neural Networks. Beijing: University of Chinese Academy of Sciences, 2021. [19] 刘新伟, 蒋盈沙, 黄武斌, 等. 基于雷达产品和随机森林算法的冰雹天气分类识别及预报. 高原气象, 2021, 40(4): 898-908. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX202104016.htmLiu X W, Jiang Y S, Huang W B, et al. Classified identification and nowcast of hail weather based on radar products and random forest algorithm. Plateau Meteor, 2021, 40(4): 898-908. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX202104016.htm [20] 王萍, 高毅, 李聪. 50 km以内雷暴系统的分类识别方法研究. 气象, 2016, 42(2): 230-237. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX201602011.htmWang P, Gao Y, Li C. Method study of classification and recognition of thunderstorm system less than 50 km. Meteor Mon, 2016, 42(2): 230-237. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX201602011.htm [21] 方德贤, 李红斌, 董新宁, 等. 风暴分类识别技术在人工防雹中的应用. 气象, 2016, 42(9): 1124-1134. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX201609010.htmFang D X, Li H B, Dong X N, et al. Application of storm auto-classification technology in artificial hail prevention. Meteor Mon, 2016, 42(9): 1124-1134. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX201609010.htm [22] 郑建琴, 路明月, 王曙东, 等. 基于决策树的天津地区冰雹天气雷达因子分析. 气象科技, 2017, 45(2): 349-354. https://www.cnki.com.cn/Article/CJFDTOTAL-QXKJ201702020.htmZheng J Q, Lu M Y, Wang S D, et al. Analysis of hail weather based on decision-making tree using radar data in Tianjin. Meteor Sci Technol, 2017, 45(2): 349-354. https://www.cnki.com.cn/Article/CJFDTOTAL-QXKJ201702020.htm [23] 濮文耀, 李红斌, 宋煜, 等. 0℃层高度的变化对冰雹融化影响的分析与应用. 气象, 2015, 41(8): 980-985. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX201508007.htmPu W Y, Li H B, Song Y, et al. Analysis and application of the effect of 0℃ layer height on melting hail. Meteor Mon, 2015, 41(8): 980-985. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX201508007.htm [24] 俞小鼎. 关于冰雹的融化层高度. 气象, 2014, 40(6): 649-654. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX201406001.htmYu X D. A note on the melting level of hail. Meteor Mon, 2014, 40(6): 649-654. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXX201406001.htm [25] Steiner M, Smith J. Use of three-dimensional reflectivity structure for automated detection and removal of non-precipitating echoes in radar data. J Atmos Ocean Technol, 2014, 40(6): 649-654. [26] 刘伯骏, 张亚萍, 黎中菊, 等. 一种基于地面实况的降雹风暴体客观标识方法. 应用气象学报, 2021, 32(1): 78-90. doi: 10.11898/1001-7313.20210107Liu B J, Zhang Y P, Li Z J, et al. An objective hailstorm labeling algorithm based on ground observation. J Appl Meteor Sci, 2021, 32(1): 78-90. doi: 10.11898/1001-7313.20210107 [27] 王研峰, 黄武斌, 王聚杰, 等. 一次甘肃天水强冰雹的雷达回波特征及成因分析. 高原气象, 2019, 38(2): 368-376. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201902013.htmWang Y F, Huang W B, Wang J J, et al. Analysis on the characteristic of radar echo and the causes of a strong hail in Tianshui City of Gansu Province. Plateau Meteor, 2019, 38(2): 368-376. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX201902013.htm [28] 郭欣, 郭学良, 陈宝君, 等. 一次大冰雹形成机制的数值模拟. 应用气象学报, 2019, 30(6): 651-664. doi: 10.11898/1001-7313.20190602Guo X, Guo X L, Chen B J, et al. Numerical simulation on the formation of large-size hailstone. J Appl Meteor Sci, 2019, 30(6): 651-664. doi: 10.11898/1001-7313.20190602 [29] 张曦, 黄兴友, 刘新安, 等. 北京大兴国际机场相控阵雷达强对流天气监测. 应用气象学报, 2022, 33(2): 192-204. doi: 10.11898/1001-7313.20220206Zhang X, Huang X Y, Liu X A, et al. The hazardous convective storm monitoring of phased-array antenna radar at Daxing International Airport of Beijing. J Appl Meteor Sci, 2022, 33(2): 192-204. doi: 10.11898/1001-7313.20220206 [30] Knox J A, Nevius D S, Knox P N. Two simple and accurate approximations for wet-bulb temperature in moist conditions with forecasting applications. Bull Amer Meteor Soc, 2017, 98(9): 1897-1906. [31] 修媛媛, 韩雷, 冯海磊. 基于机器学习方法的强对流天气识别研究. 电子设计工程, 2016, 24(9): 4-7. https://www.cnki.com.cn/Article/CJFDTOTAL-GWDZ201609002.htmXiu Y Y, Han L, Feng H L. The identification of strong convective weather based on machine learning methods. Electronic Design Engineering, 2016, 24(9): 4-7. https://www.cnki.com.cn/Article/CJFDTOTAL-GWDZ201609002.htm [32] 陈训来, 刘军, 郑群峰, 等. 基于卷积门控循环单元神经网络的临近预报方法研究. 高原气象, 2020, 40(2): 411-423. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX202102018.htmChen X L, Liu J, Zheng Q F, et al. A study on radar echo nowcasting based on convolutional gated recurrent unit neural network. Plateau Meteor, 2020, 40(2): 411-423. https://www.cnki.com.cn/Article/CJFDTOTAL-GYQX202102018.htm [33] 章国材. 强对流天气分析与预报. 北京: 气象出版社, 2011.Zhang G C. Analysis and Forecast for Severe Convective Weather. Beijing: China Meteorological Press, 2011. [34] 李颖, 陈怀亮. 机器学习技术在现代农业气象中的应用. 应用气象学报, 2020, 31(3): 257-266. doi: 10.11898/1001-7313.20200301Li Y, Chen H L. Review of machine learning approaches for modern agrometeoroly. J Appl Meteor Sci, 2020, 31(3): 257-266. doi: 10.11898/1001-7313.20200301 [35] 史达伟, 张静, 曹庆, 等. 基于决策树算法的海州湾地区海雾预测. 气象科学, 2022, 42(1): 136-142. https://www.cnki.com.cn/Article/CJFDTOTAL-QXKX202201015.htmShi D W, Zhang J, Cao Q, et al. Research on sea fog diagnosis in Haizhou Bay based on decision tree algorithm. J Meteor Sci, 2022, 42(1): 136-142. https://www.cnki.com.cn/Article/CJFDTOTAL-QXKX202201015.htm [36] 俞小鼎, 周小刚, 王秀明, 等. 雷暴与强对流临近天气预报技术进展. 气象学报, 2012, 70(3): 311-337. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201203001.htmYu X D, Zhou X G, Wang X M, et al. The advances in the nowcasting techniques on thunderstorms and severe convection. Acta Meteor Sinica, 2012, 70(3): 311-337. https://www.cnki.com.cn/Article/CJFDTOTAL-QXXB201203001.htm [37] 俞小鼎, 王秀明, 李万莉, 等. 雷暴与强对流临近预报. 北京: 气象出版社, 2020.Yu X D, Wang X M, Li W L, et al. Thunderstorm and Strong Convection Nowcasting. Beijing: China Meteorological Press, 2020. [38] 李博勇, 胡志群, 郑佳锋, 等. 利用贝叶斯方法改进华南地区冰雹识别效果. 热带气象学报, 2021, 37(1): 112-125. https://www.cnki.com.cn/Article/CJFDTOTAL-RDQX202101011.htmLi B Y, Hu Z Q, Zheng J F, et al. Using Bayesian method to improve hail identification in South China. J Trop Meteor, 2021, 37(1): 112-125. https://www.cnki.com.cn/Article/CJFDTOTAL-RDQX202101011.htm [39] 刘海知, 徐辉, 包红军, 等. 机器学习分类算法在降雨型滑坡预报中的应用. 应用气象学报, 2022, 33(3): 282-292. doi: 10.11898/1001-7313.20220303Liu H Z, Xu H, Bao H J, et al. Application of machine learning classification algorithm to precipitation landslides forecasting. J Appl Meteor Sci, 2022, 33(3): 282-292. doi: 10.11898/1001-7313.20220303 [40] 刘娜, 熊安元, 张强, 等. 强对流天气人工智能应用训练基础数据集构造. 应用气象学报, 2021, 32(5): 530-541. doi: 10.11898/1001-7313.20210502Liu N, Xiong A Y, Zhang Q, et al. Development of basic dataset of severe convective weather for artificial intelligence training. J Appl Meteor Sci, 2021, 32(5): 530-541. doi: 10.11898/1001-7313.20210502 -
图(8) / 表(2)
计量
- 摘要浏览量: 1310
- HTML全文浏览量: 159
- PDF下载量: 153
- 被引次数: 0
