A study on radar echo extrapolation nowcasting method combined with physical constraints of mesoscale model
-
摘要: 研究设计了一种结合中尺度模式物理约束的雷达回波临近智能外推预报方法，该方法在外推预报时效（0—2 h）内即利用中尺度高分辨率模式信息对外推进行约束。首先将模式风场和雷达回波轨迹风场融合成融合风场，然后利用融合风场光流外推形成动力约束外推；并在此基础上利用模式诊断产品和雷达历史资料通过投票回归器集成多种深度学习算法构建回波强度频率分布的预测模型，最终基于预测模型结果利用降水频率匹配订正技术对外推预测的原始回波强度进行订正形成物理约束外推方法。通过2个典型个例，以及2年主汛期的长期检验对原始光流法、动力约束外推方法和物理约束外推方法进行综合评估，结果表明：动力约束外推通过改善光流法回波在边缘的堆积扭曲从而改进了预报性能，物理约束外推通过基于模式信息预测的回波频率分布调整回波强度实现回波的增强和减弱来改善预报性能，随着时效延长改善越来越明显，整体而言物理约束外推是其中最优的方案。Abstract: This study develops an intelligent extrapolation nowcasting method combined with physical constraints of mesoscale numerical model. The method uses the high-resolution mesoscale model information to constrain the extrapolation within 2 h. Based on optical flow (OF) method extrapolation, the dynamic constraint extrapolation optical flow (DCOF) method is applied to reconstruct the motion vector; the mesoscale numerical model diagnostic products and radar historical data are used to train the echo intensity frequency prediction model, which integrates multiple deep learning algorithms through the voting regressor. Finally, a physical constraint optical flow (PCOF) method is used to correct the original echo intensity of the OF method by using the frequency-matching method. The performance of the OF method, DCOF method and PCOF method in 2 typical cases and cases over a long-term period are evaluated. The results show that the DCOF method improves the prediction by improving the unrealistic accumulation and distortion produced by the OF method at the edge of the echoes. The PCOF method improves the prediction by adjusting the echo intensity frequency distribution prediction based on the model information. With increasing forecast lead time, the improvements caused by the two methods are becoming more obvious. Overall, the PCOF method is the best among the three methods.
-
Key words:
- Optical flow /
- Mesoscale model /
- Deep learning /
- Nowcasting
-
图 1 选取的试验范围 （蓝色方框为选取的试验区域 （29°—37°N，114°—122°E）；红色圆点为雷达站点位置，数字为雷达站号，红色曲线范围为雷达站点探测覆盖范围）
Figure 1. Distribution of radar stations （red points） and radar detection range （red line） （The blue rectangle denotes the selected area （29°—37°N，114°—122°E））
图 2 动力约束风场融合效果 （a. 光流风场（大于 5 m/s 区域用紫色实线圈出），b. 模式无辐散风场，c. 融合风场）
Figure 2. Comparison between radar the optical flow model produced wind field （a） （the purple line denotes the area where the wind speed is greater than 5 m/s），PWAFS harmonic wind field （b） and the combined wind field （c）
图 3 回波强度和其最优相关的强对流产品归一化时间序列 （a. 回波强度大于10 dBz，b. 回波强度大于20 dBz，c. 回波强度大于30 dBz， d. 回波强度大于40 dBz）
Figure 3. Normalized time series of precipitation echo and the most relevant convective products （a. precipitation echo above 10 dBz，b. precipitation echo above 20 dBz，c. precipitation echo above 30 dBz，d. precipitation echo above 40 dBz）
图 4 回波强度频率智能预测模型流程
Figure 4. Flow chart of intelligent echo intensity frequency prediction model
图 5 2019年7月6日09时雷达外推预报120 min效果对比 （a. 实况，b. 光流外推，c. 动力约束外推，d. 物理约束外推）
Figure 5. Comparison between observations and forecasts on 6 July 2019 （a. observations of composite reflectivity at 11：00 BT，b. 120 min forecast at 09：00 BT using OF，c. 120 min forecast at 09：00 BT using DCOF，d. 120 min forecast at 09：00 BT using PCOF）
图 6 2019年7月6日09时雷达外推预报和实况在研究范围内20—30 dBz （a）、超过30 dBz （b） 的格点数时间序列
Figure 6. Time series of number of grid points with radar echoes within 20—30 dBz （a） and above 30 dBz （b） from radar extrapolation forecast and observations in the study area at 09：00 BT 6 July 2019
图 7 2019年7月6日09时3种外推预报方案 （光流外推：黑线，动力约束外推：蓝线，物理约束外推：红线） 20 dBz的CSI评分 （a） 和BIAS评分 （b），30 dBz的CSI评分 （c） 和BIAS评分 （d），以及均方根误差 （e） 和相关系数 （f）
Figure 7. Experimental results in terms of CSI （a. 20 dBz，c. 30 dBz），BIAS （b. 20 dBz，d. 30 dBz），RMSE （e） and CC （f） by OF （black line），DCOF （blue line） and PCOF （red line） at 09：00 BT 6 July 2019
图 8 2020年5月18日02时雷达外推预报60 min效果对比 （a. 实况，b. 光流外推，c. 动力约束外推，d. 物理约束外推）
Figure 8. Comparison between observations and forecasts on 18 May 2020 （a. observations of composite reflectivity at 03：00 BT，b. 60 min forecast at 02：00 BT using OF，c. 60 min forecast at 02：00 BT using DCOF，d. 60 min forecast at 02：00 BT using PCOF）
图 9 2020年5月18日02时雷达外推预报和实况在研究范围内20—30 dBz （a）、超过30 dBz （b） 的格点数时间序列
Figure 9. Time series of number of grid points with radar echoes within 20—30 dBz （a） and above 30 dBz （b） from radar extrapolation forecast and observations in the study area at 02：00 BT 18 May 2020
图 10 2020年5月18日02时3种外推预报方案 （光流外推：黑线，动力约束外推：蓝线，物理约束外推：红线） 20 dBz的CSI评分 （a） 和BIAS评分 （b），30 dBz的CSI评分 （c） 和BIAS评分 （d），以及圴方根误差 （e） 和相关系数 （f）
Figure 10. Experimental results in terms of CSI （a. 20 dBz，c. 30 dBz），BIAS （b. 20 dBz，d. 30 dBz），RMSE （e） and CC （f） by OF （black line），DCOF （blue line） and PCOF （red line） at 02：00 BT 18 May 2020
图 11 三种外推预报方案 （光流外推：黑线，动力约束外推：蓝线，物理约束外推：红线） 20 dBz的CSI评分 （a） 和BIAS评分 （b），30 dBz的CSI评分 （c） 和BIAS评分 （d），以及均方根误差 （e） 和相关系数 （f）
Figure 11. Experimental results in terms of CSI （a. 20 dBz，c. 30 dBz），BIAS （b. 20 dBz，d. 30 dBz），RMSE （e） and CC （f） by OF （black line），DCOF （blue line） and PCOF （red line）
表 1 训练集构建方式
Table 1. Construction of training set
预报因子 预报量 起报时刻某强度区间回波格点比例 Rt0 某预报时 较前1 h（9次）某强度区间回波格点比例变化Rt0−Rt0-n 效某强度 第2时次模式物理因子P2 区间回波 较第1时次模式物理因子变化量P2−P1 格点比例 
下载: 导出CSV
-
[1] 曹春燕,陈元昭,刘东华等. 2015. 光流法及其在临近预报中的应用. 气象学报,73(3):471-480 doi: 10.11676/qxxb2015.034Cao C Y,Chen Y Z,Liu D H,et al. 2015. The optical flow method and its application to nowcasting. Acta Meteor Sinica,73(3):471-480 (in Chinese) doi: 10.11676/qxxb2015.034 [2] 陈明轩,王迎春,俞小鼎. 2007. 交叉相关外推算法的改进及其在对流临近预报中的应用. 应用气象学报,18(5):690-701 doi: 10.3969/j.issn.1001-7313.2007.05.014Chen M X,Wang Y C,Yu X D. 2007. Improvement and application test of TREC algorithm for convective storm nowcast. J Appl Meteor Sci,18(5):690-701 (in Chinese) doi: 10.3969/j.issn.1001-7313.2007.05.014 [3] 陈元昭,林良勋,王蕊等. 2019. 基于生成对抗网络GAN的人工智能临近预报方法研究. 大气科学学报,42(2):311-320Chen Y Z,Lin L X,Wang R,et al. 2019. A study on the artificial intelligence nowcasting based on generative adversarial networks. Trans Atmos Sci,42(2):311-320 (in Chinese) [4] 程丛兰,陈明轩,王建捷等. 2013. 基于雷达外推临近预报和中尺度数值预报融合技术的短时定量降水预报试验. 气象学报,71(3):397-415 doi: 10.11676/qxxb2013.041Cheng C L,Chen M X,Wang J J,et al. 2013. Short-term quantitative precipitation forecast experiments based on blending of nowcasting with numerical weather prediction. Acta Meteor Sinica,71(3):397-415 (in Chinese) doi: 10.11676/qxxb2013.041 [5] 程丛兰,陈敏,陈明轩等. 2019. 临近预报的两种高时空分辨率定量降水预报融合算法的对比试验. 气象学报,77(4):701-714 doi: 10.11676/qxxb2019.017Cheng C L,Chen M,Chen M X,et al. 2019. Comparative experi-ments on two high spatiotemporal resolution blending algorithms for quantitative precipitation nowcasting. Acta Meteor Sinica,77(4):701-714 (in Chinese) doi: 10.11676/qxxb2019.017 [6] 邓涤菲,周玉淑. 2011. 无旋转风分量在台风“桑美”急剧增强和急剧减弱过程中的分析和应用. 高原气象,30(2):406-415Deng D F,Zhou Y S. 2011. Analysis and application of irrotational wind component in the drastically increasing and decreasing processes of "Saomai" Typhoon. Plateau Meteor,30(2):406-415 (in Chinese) [7] 郭瀚阳,陈明轩,韩雷等. 2019. 基于深度学习的强对流高分辨率临近预报试验. 气象学报,77(4):715-727 doi: 10.11676/qxxb2019.036Guo H Y,Chen M X,Han L,et al. 2019. High resolution nowcasting experiment of severe convections based on deep learning. Acta Meteor Sinica,77(4):715-727 (in Chinese) doi: 10.11676/qxxb2019.036 [8] 韩丰,龙明盛,李月安等. 2019. 循环神经网络在雷达临近预报中的应用. 应用气象学报,30(1):61-69 doi: 10.11898/1001-7313.20190106Han F,Long M S,Li Y A,et al. 2019. The application of recurrent neural network to nowcasting. J Appl Meteor Sci,30(1):61-69 (in Chinese) doi: 10.11898/1001-7313.20190106 [9] 黄旋旋,朱科锋,赵坤. 2017. 改进后TREC外推方法在台风临近降雨预报中的应用. 气象科学,37(5):610-618Huang X X,Zhu K F,Zhao K. 2017. The improved TREC algorithm and its application to the precipitation now-casting for landfall typhoon. J Meteor Sci,37(5):610-618 (in Chinese) [10] 李俊,杜钧,陈超君. 2014. 降水偏差订正的频率(或面积)匹配方法介绍和分析. 气象,40(5):580-588 doi: 10.7519/j.issn.1000-0526.2014.05.008Li J,Du J,Chen C J. 2014. Introduction and analysis to frequency or area matching method applied to precipitation forecast bias correction. Meteor Mon,40(5):580-588 (in Chinese) doi: 10.7519/j.issn.1000-0526.2014.05.008 [11] 苏翔,袁慧玲,朱跃建. 2021. 四种定量降水预报客观订正方法对比研究. 气象学报,79(1):132-149Su X,Yuan H L,Zhu Y J. 2021. A comparative study of four objective quantitative precipitation forecast calibration methods. Acta Meteor Sinica,79(1):132-149 (in Chinese) [12] 吴剑坤,陈明轩,秦睿等. 2019. 变分回波跟踪算法及其在对流临近预报中的应用试验. 气象学报,77(6):999-1014 doi: 10.11676/qxxb2019.062Wu J K,Chen M X,Qin R,et al. 2019. The variational echo tracking method and its application in convective storm nowcasting. Acta Meteor Sinica,77(6):999-1014 (in Chinese) doi: 10.11676/qxxb2019.062 [13] 曾明剑,王桂臣,吴海英等. 2015a. 基于中尺度数值模式的分类强对流天气预报方法研究. 气象学报,73(5):868-882Zeng M J,Wang G C,Wu H Y,et al. 2015a. Study of the forecasting method for the classified severe convection weather based on a meso-scale numerical model. Acta Meteor Sinica,73(5):868-882 (in Chinese) [14] 曾明剑,张备,吴海英等. 2015b. 基于接近度概念的强对流天气预报方法研究. 高原气象,34(5):1357-1368Zeng M J,Zhang B,Wu H Y,et al. 2015b. Study on severe convective weather forecast method based on approach concept. Plateau Meteor,34(5):1357-1368 (in Chinese) [15] Atencia A,Rigo T,Sairouni A,et al. 2010. Improving QPF by blending techniques at the meteorological service of catalonia. Nat Hazards Earth Syst Sci,10(7):1443-1455 doi: 10.5194/nhess-10-1443-2010 [16] Ayzel G,Heistermann M,Winterrath T. 2019. Optical flow models as an open benchmark for radar-based precipitation nowcasting (rainymotion v0.1). Geosci Model Dev,12(4):1387-1402 doi: 10.5194/gmd-12-1387-2019 [17] Breiman L. 2001. Random forests. Mach Learn,45(1):5-32 doi: 10.1023/A:1010933404324 [18] Chen Q S,Kuo Y H. 1992. A consistency condition for wind-field reconstruction in a limited area and a harmonic-cosine series expansion. Mon Wea Rev,120(11):2653-2670 doi: 10.1175/1520-0493(1992)120<2653:ACCFWF>2.0.CO;2 [19] Chen Y Z,Lan H P,Chen X L,et al. 2017. A nowcasting technique based on application of the particle filter blending algorithm. J Meteor Res,31(5):931-945 doi: 10.1007/s13351-017-6557-9 [20] Haiden T,Kann A,Wittmann C,et al. 2011. The integrated nowcasting through comprehensive analysis (INCA) system and its validation over the eastern alpine region. Wea Forecasting,26(2):166-183 doi: 10.1175/2010WAF2222451.1 [21] Li X,Zeng M J,Wang Y,et al. 2016. Evaluation of two momentum control variable schemes and their impact on the variational assimilation of radarwind data:Case study of a squall line. Adv Atmos Sci,33(10):1143-1157 doi: 10.1007/s00376-016-5255-3 [22] Liang Q Q,Feng Y R,Deng W J,et al. 2010. A composite approach of radar echo extrapolation based on TREC vectors in combination with model-predicted winds. Adv Atmos Sci,27(5):1119-1130 doi: 10.1007/s00376-009-9093-4 [23] Liu Y,Xi D G,Li Z L,et al. 2015. A new methodology for pixel-quantitative precipitation nowcasting using a pyramid Lucas Kanade optical flow approach. J Hydrol,529:354-364 doi: 10.1016/j.jhydrol.2015.07.042 [24] Pedregosa F,Varoquaux G,Gramfort A,et al. 2011. Scikit-learn:Machine learning in Python. J Machine Learn Res,12:2825-2830 [25] Wong W K, Yeung L H Y, Wang Y C, et al. 2009. Towards the blending of NWP with nowcast operation experience in BO8FDP∥WMO. Symposium on Nowcasting. 30 Aug - 4 Sep 2009, Whistler B. C. Canada [26] Zhu Y J,Luo Y. 2015. Precipitation calibration based on the frequency-matching method. Wea Forecasting,30(5):1109-1124 doi: 10.1175/WAF-D-13-00049.1 -
下载:
点击查看大图
计量
- 文章访问数: 433
- HTML全文浏览量: 142
- PDF下载量: 158
- 被引次数: 0
