TORNADO DETECTION ALGORITHM BASED ON RANDOM FOREST IN RADAR NETWORK
-
摘要: 近年来我国极端灾害性天气频发,造成了重大人员伤亡和财产损失,随着防灾减灾工作的推进,龙卷等中小尺度强对流灾害性天气的预警预报工作的关注度正逐步提升。现有龙卷检测算法基于对新一代天气雷达基数据在多个仰角和体积扫描中进行阈值判断得到龙卷涡旋特征TVS,在自适应协同观测背景下表现为自适应策略同步较慢,预警预报准确率不高,提前预警时间短。使用机器学习算法结合龙卷在雷达反射率、径向速度和速度谱宽的多重特征能有效提高龙卷识别的准确率和预警时间,能提高组网雷达的协同观测能力。基于随机森林的龙卷检测算法(TDA-RF),使用CINRAD雷达历史龙卷数据作为训练集,通过随机森林算法对训练集进行分类学习得到龙卷预测模型,使用预测模型对实时雷达数据进行龙卷检测。试验结果表明,TDA-RF算法能有效识别不同强度的龙卷,较TVS龙卷检测算法能给出龙卷区域的分类概率值,无需对龙卷特征时空连续性进行判断;TDA-RF算法对多个特征进行综合判断具有较好的抗干扰能力,使基于组网雷达的龙卷预警时间最高可达18分钟。Abstract: In recent years, extreme severe weather has frequently occurred in China, causing heavy casualties and property losses. As a result, a growing focus has been on the early warning and forecasting of tornadoes and other small to medium-scale severe convective weather events as part of national disaster prevention and mitigation efforts. However, the current tornado detection algorithm exhibits limitations in terms of accuracy and warning time. This algorithm relies on the threshold judgment of the new generation weather radar base data in multiple elevation angles and volume scans to obtain the tornado vortex signature (TVS). The TVS algorithm's tornado warning forecasts are characterized by low accuracy, and the early warning time is short due to the sudden onset of tornadoes and their short generation and lysis duration. The integration of a machine learning algorithm with the multiple features of the tornado, including radar reflectivity, radial velocity, and velocity spectrum width, can effectively improve the accuracy and warning time of tornado recognition. In the present study, the tornado detection algorithm based on random forest (TDA-RF) utilizes CINRAD radar historical tornado data as the training set, classifies the training set through the random forest algorithm to generate a tornado prediction model, and employs the prediction model to detect tornadoes in real-time radar data. The test results show that the TDA-RF algorithm can effectively identify tornadoes of varying intensities. Compared with the TVS tornado detection algorithm, it can calculate the classification probability value of the tornado area. TDA-RF does not require the judgment of the spatiotemporal continuity of tornado features. It is less affected by false and invalid echoes, and the early warning time can reach up to 18 minutes.
-
Key words:
- weather radar /
- tornado /
- random forest /
- machine learning
-
表 1 基于随机森林的组网雷达龙卷检测算法性能测试结果
年份 击中 漏报 误报 2011 3 5 4 2013 19 5 5 2015 0 2 0 2016 33 5 4 2017 9 3 1 2018 5 2 2 表 2 样本集特征(注:c4开头的特征表示从4×4网格正中间的2×2的网格计算得到的特征)
特征 含义 单位 r_average 4×4网格内反射率平均值 dBZ r_max 4×4网格内反射率最大值 dBZ r_min 4×4网格内反射率最小值 dBZ v_average 4×4网格内径向速度平均值 m/s v_max 4×4网格内径向速度最大值 m/s v_min 4×4网格内径向速度最小值 m/s w_average 4×4网格内谱宽平均值 m/s w_max 4×4网格内谱宽最大值 m/s w_min 4×4网格内谱宽最小值 m/s s_average 4×4网格内切变平均值 s-1 s_max 4×4网格内切变最大值 s-1 s_min 4×4网格内切最小变值 s-1 l_average 4×4网格内角动量平均值 m2/s l_max 4×4网格内角动量最大值 m2/s l_min 4×4网格内角动量最小值 m2/s vt_average 4×4网格内旋转速度平均值 m/s vt_max 4×4网格内旋转速度最大值 m/s vt_min 4×4网格内旋转速度最小值 m/s c4_d_v_max 2×2网格内最大的速度差值 m/s c4_s_average 2×2网格内切变的平均值 s-1 c4_s_max 2×2网格内切变的最大值 s-1 c4_s_min 2×2网格内切变的最小值 s-1 c4_l_average 2×2网格内角动量的平均值 m2/s c4_l_max 2×2网格内角动量的最大值 m2/s c4_l_min 2×2网格内角动量的最小值 m2/s c4_vt_average 2×2网格内旋转速度的平均值 m/s c4_vt_max 2×2网格内旋转速度的最大值 m/s c4_vt_min 2×2网格内旋转速度的最小值 m/s w_range 4×4网格内谱宽的范围 m/s w_40 4×4网格内大于谱宽网格内40%谱宽值的阈值 m/s w_60 4×4网格内大于谱宽网格内60%谱宽值的阈值 m/s w_80 4×4网格内大于谱宽网格内80%谱宽值的阈值 m/s -
[1] 俞小鼎, 赵娟, 范雯杰. 中国龙卷的时空分布与关键环境参数特征[J]. 热带气象学报, 2021, 37(Z1): 681-692. [2] 何建新, 曾强宇, 王皓, 等. 龙卷的雷达探测研究进展[J]. 成都信息工程大学学报, 2018, 33(5): 477-489. [3] LEMON L R, DONALDSON Jr R J, BURGESS D W, et al. Doppler radar application to severe thunderstorm study and potential real-time warning[J]. Bull Amer Meteor Soc, 1977, 58(11): 1 187-1 193. [4] DOSWELL III C A. Small sample size and data quality issues illustrated using tornado occurrence data[J]. E-Journal of Severe Storms Meteorology, 2007, 2(5): 1-16. [5] 孟智勇. 张福青. 罗德海, 等. 新中国成立70年来的中国大气科学研究: 天气篇[J]. 中国科学: 地球科学, 2019, 49(12): 1 875-1 918. [6] BLUESTEIN H B, LADUE J G, STEIN H, et al. Doppler radar wind spectra of supercell tornadoes[J]. Mon Wea Rev, 1993, 121(8): 2 200-2 222. [7] HEINSELMAN P, LADUE D, KINGFIELD D M, et al. Tornado warning decisions using phased-array radar data[J]. Wea Forecasting, 2015, 30(1): 57-78. [8] 赵鑫樾, 陈石, 张晓杰, 等. 中国龙卷风等级分布研究[J]. 科技与创新, 2019(21): 136-137. [9] 郑永光, 陶祖钰, 俞小鼎. 强对流天气预报的一些基本问题[J]. 气象, 2017, 43(6): 641-652. [10] 李兆慧, 王东海, 麦雪湖, 等. 2015年10月4日佛山龙卷过程的观测分析[J]. 气象学报, 2017, 75(2): 288-313. [11] 王秀明, 俞小鼎. 热带一次致灾龙卷形成物理过程研究[J]. 气象学报, 2019, 77(3): 387-404 [12] 唐嘉蕙, 冉令坤, 沈新勇, 等. 广东佛山EF3级龙卷超级单体风暴高分辨率数值模拟[J]. 地球物理学报, 2019, 62(11): 4 082-4 097. [13] 植江玲, 张伟强, 黄先香. "山竹"(1822)台风外围佛山强龙卷天气过程观测分析[J]. 气象与减灾研究, 2019, 42(4): 251-260. [14] ZHANG X, HE J, ZENG Q, et al. Weather radar echo super-resolution reconstruction based on nonlocal self-similarity sparse representation [J]. Atmos, 2019, 10(5): 254. [15] 顾瑜, 孙即霖, 楚合涛. 2016年6月江苏阜宁一次超强龙卷的特征分析[J]. 中国海洋大学学报(自然科学版), 2018, 48(2): 11-21. [16] 朱江山, 刘娟, 边智, 等. 一次龙卷生成中风暴单体合并和涡旋特征的雷达观测研究[J]. 气象, 2015, 41(2): 182-191. [17] 曾明剑, 吴海英, 王晓峰, 等. 梅雨期龙卷环境条件与典型龙卷对流风暴结构特征分析[J]. 气象, 2016, 42(3): 280-293. [18] 植江玲, 李彩玲, 蔡康龙, 等. 雷州半岛一次台风外围龙卷天气过程分析[J]. 气象科技进展, 2020, 10(6): 28-33. [19] 张一平, 梁俊平, 牛淑贞, 等. 豫东龙卷的环境条件和雷达监测预警分析[J]. 高原气象, 2020, 39(5): 1 089-1 101. [20] BROWN R A, LEMON L R, BURGESS D. Tornado detection by pulsed Doppler radar[J]. Mon Wea Rev, 1978, 106(1): 29-38. [21] ZRNIC D S, BURGESS D W, HENNINGTON L D. Automatic detection of mesocyclonic shear with Doppler radar[J]. J Atmos Ocean Technol, 1985, 2(4): 425-438. [22] WAKIMOTO R M, WILSON J W. Non-supercell tornadoes[J]. Mon Wea Rev, 1989, 117(6): 1 113-1 140. [23] PALMER R D, BODINE D, KUMJIAN M, et al. Observations of the 10 May 2010 tornado outbreak using OU-PRIME: Potential for new science with high-resolution polarimetric radar[J]. Bull Amer Meteor Soc, 2011, 92(7): 871-891. [24] 程元慧, 傅佩玲, 胡东明, 等. 广州相控阵天气雷达组网方案设计及其观测试验[J]. 气象, 2020, 46(6): 823-836. [25] 张羽, 田聪聪, 傅佩玲, 等. 广州X波段双偏振相控阵天气雷达观测试验进展[J]. 气象科技进展, 2020, 10(6): 80-85. [26] WANG Y, YU T Y. Novel tornado detection using an adaptive neuro-fuzzy system with S-band polarimetric weather radar[J]. J Atmos Ocean Technol, 2015, 32(2): 195-208. [27] 谭仲辉, 马烁, 韩丁, 等. 基于随机森林算法的FY-4A云底高度估计方法[J]. 红外与毫米波学报, 2019, 38(3): 381-388. [28] HILL A J, HENMAN G R, SCHUMACHER R S. Forecasting severe weather with random forests[J]. Mon Wea Rev, 2020, 148(5): 2 135-2 161. [29] PEERBHAY K Y, MUTANGA O, ISMAIL R. Random forests unsupervised classification: The detection and map of solanum mauritianum infestations in plantation forestry using hyperspectral data[J]. IEEE J Sel Top Appl Earth Obs Remote Sens, 2015, 8(6): 3 107-3 122 [30] BREIMAN L. Random forests[J]. Mach Learn, 2001, 45: 5-32. [31] 方匡南, 吴见彬, 朱建平, 等. 随机森林方法研究综述[J]. 统计与信息论坛, 2011, 26(3): 32-38. [32] HO T K. The random subspace method for constructing decision forests[J]. IEEE Trans Pattern Anal Mach Intell, 1998, 20(8): 832-844. [33] AHIJEVYCH D, PINTO J O, WILLIAMS J K, et al. Probabilistic forecasts of mesoscale convective system initiation using the random forest data mining technique[J]. Wea Forecasting, 2016, 31(2): 581-599. [34] 郑永光, 朱文剑, 姚聃, 等. 风速等级标准与2016年6月23日阜宁龙卷强度估计[J]. 气象, 2016, 42(11): 1 289-1 303. [35] 张小玲, 杨波, 朱文剑, 等. 2016年6月23日江苏阜宁EF4级龙卷天气分析[J]. 气象, 2016, 42(11): 1 304-1 314. [36] 张建云, 张持岸, 葛元, 等. 1522号台风外围佛山强龙卷X波段双偏振多普勒雷达反射率因子特征[J]. 气象科技, 2018, 46(1): 163-169. [37] 王炳赟, 魏鸣, 范广洲, 等. 1522强台风"彩虹"螺旋雨带中衍生龙卷的超级单体演变与机理研究Ⅰ: 谱宽和速度[J]. 热带气象学报, 2018, 34(4): 472-480. [38] 马骊. 随机森林算法的优化改进研究[D]. 广州: 暨南大学, 2016.