利用卷积神经网络开展偏振雷达定量降水估测研究

蔡康龙; 胡志群; 谭浩波; 黄锦灿; 张伟强; 张晶晶; 植江玲

doi:10.16032/j.issn.1004-4965.2024.008

利用卷积神经网络开展偏振雷达定量降水估测研究

doi: 10.16032/j.issn.1004-4965.2024.008

1.
佛山市龙卷风研究中心/中国气象局龙卷风重点开放实验室，广东佛山 528000
2.
中国气象科学研究院灾害天气国家重点实验室，北京 100081
3.
广东省气象局，广东广州 510080
4.
高明区气象局，广东佛山 528000

基金项目:

广东省重点领域研发计划 2020B1111200001

广东省气象局科学技术研究重点项目 GRMC2022Z03

佛山市气象局科学技术项目 201915

详细信息

通讯作者:
胡志群，男，江西省人，研究员，主要从事雷达气象研究。E-mail: huzq@cma.gov.cn

中图分类号: P412
计量
- 文章访问数: 80
- HTML全文浏览量: 76
- PDF下载量: 53
- 被引次数: 0
出版历程
- 收稿日期: 2022-04-15
- 修回日期: 2023-11-03
- 网络出版日期: 2024-04-12
- 刊出日期: 2024-02-20

Research on Quantitative Precipitation Estimation by Polarized Radar Using CNN

1.
Foshan Tornado Research catter, China Meteorological Administration Tornado Key Laboratory, Foshan, Guangdong 528000, China
2.
Chinese Academy of Meteorological Sciences, Beijing 100081, China
3.
Guangdong Meteorological Bureau, Guangzhou, 51080, China
4.
Gaoming Meteorological Bureau, Foshan, Guangdong 528000, China

摘要

摘要: 利用偏振升级改造后的广州新一代天气雷达（CINRAD/SAD）水平反射率Z_H、差分传播相移率K_DP、差分反射率因子Z_DR和广东佛山219个地面气象自动站雨量数据，形成不同偏振量组合的8个数据集。基于卷积神经网络（CNN），建立雷达定量降水估测网络架构QPEnet，并将该架构用于雷达定量降水估测（QPE），评估结果表明：数据集通道数N的增加可降低QPEnet的定量降雨估测的均方根误差（RMSE），并提高相关系数（CORR）；对于由Z_H形成的数据集Z、Z_1~3 km和Z_6 min，随着通道数N的增加，数据集Z、Z_1~3 km和Z_6 min的性能逐步得到提高，数据集Z_1~3 km和Z_6 min的均方根误差（RMSE）分别是4.71和3.78，比数值集Z分别降低了1.3% 和18.7%；数据集Z_1~3 km和Z_6 min的CORR分别是0.82和0.88，比数据集Z分别提高了2.5%和10.0%；对于Z_H、K_DP和Z_DR偏振量组成的数据集里面，数据集Z_Z_DR_K_DP的拟合性能最好，RMSE为3.97，比数据集Z的RMSE降低了14.6%，CORR是0.86，比数据集Z提高了7.5%；分别对0.6~5mm、5~10 mm、10~20 mm、20~30 mm、30~40 mm、40~50 mm和50 mm以上的7个降水量级的均方根误差（RMSE）、平均偏差比（MBR）、平均误差（AE）和相对误差（RE）等的统计结果表明，数据集Z_6 min降雨精度最高。
- 定量降水估测 /
- 卷积神经网络 /
- S波段双偏振雷达 /
- 测雨精度
Abstract: The Z_H, Z_DR and K_DP of Guangzhou S-band dual polarization radar and rainfall data of 219 automatic meteorological stations in Foshan are used to form 8 datasets. Based on the convolutional neural network CNN, a radar quantitative precipitation estimation model is established, which will be used for ground precipitation estimation. The evaluating results of 8 datasets applied to the same precipitation estimation model are compared to each other. The results show that: The increase in the number of channels(N) of the datasets is beneficial to reduce the RMSE and improve CORR of the quantitative rainfall estimation results; For the datasets Z, Z_1~3 km and Z_6 min formed by Z_H, as the number of channels increases, the performance of the data sets Z, Z_1~3 km and Z_6 min are gradually improved, and the RMSE of Z_1~3 km and Z_6 min are 4.71 and 3.78, which are -1.3% and 18.7% lower than that of dataset Z; the CORR of Z_1~3 km and Z_6 min are 0.82 and 0.88, which are 2.5% and 10% higher than that of dataset Z; Among other datasets composed of K_DP and Z_DR, the dataset Z_Z_DR_K_DP has the best fitting performance. The RMSE is 3.97, which is 14.6% lower than that of dataset Z, and the CORR is 0.86, which is 7.5% higher than that of dataset Z; The statistical results of RMSE, MBR, AE and RE for seven precipitation levels of 0.6~5 mm, 5~10 mm, 10~20 mm, 20~30 mm, 30~40 mm, 40~50 mm and above 50 mm respectively, show that dataset Z_6 min has the highest rainfall accuracy.
- Quantitative Precipitation Estimation(QPE) /
- convolutional neural network /
- S-band dual polarization /
- measurement accuracy

HTML全文

图 1 佛山市地面气象自动站（红点）上空广州雷达（红星）某等高高度25×25个格点（黑框）数据示意图（颜色表示广州雷达的回波）

下载: 全尺寸图片幻灯片

图 2 预处理后Z_H（a）、Z_DR（b）、K_DP（c）和地面气象自动站的小时雨量（d）的箱线图

上限是表示95%位置的数值，下限则是5%位置的数值，箱体的上四分位数和下四分位数分别表示75%位置和25%位置的数值。

下载: 全尺寸图片幻灯片

图 3 定量降雨估测的深度学习框架图

输入层25×25×N特征矩阵组成的数据集；框架中间部分由两卷积层和四层全连接层组成，表示应用卷积神经网络提取特征并从数据学习映射；输出层尺寸为1，表示雨量估计值。

下载: 全尺寸图片幻灯片

图 4 8个数据集的QPEnet的损失

下载: 全尺寸图片幻灯片

图 5 不同数据集的估测降水量与实测降水量的对比

a.数据集K_DP；b.数据集Z；c.数据集Z_1~3 km；d.数据集Z_Z_DR；e.数据集Z_KDP；f.数据集Z_DR_K_DP；g.数据集Z_Z_DR_K_DP；h.数据集Z_ 6 min，色标为样本数。

下载: 全尺寸图片幻灯片

图 6 MBR(a)、KMSE(b)、ME(c)、RE(d)估测降水的可靠性对比

下载: 全尺寸图片幻灯片

表 1 各数据集的参数数据集

数据集	尺寸（25×25×N）	参数配置
Z	25×25×1	1 km等高高度的Z_H
K_DP	25×25×1	1 km等高高度的K_DP
Z_Z_DR	25×25×2	1 km等高高度的Z_H和Z_DR
Z_K_DP	25×25×2	1 km等高高度的Z_H和K_DP
Z_DR_K_DP	25×25×2	1 km等高高度的Z_DR和K_DP
Z_Z_DR_K_DP	25×25×3	1 km等高高度的Z_H、Z_DR和K_DP
Z_1~3 km	25×25×3	1 km、2 km和3 km等高高度的Z_H
Z_6 min	25×25×10	1 h逐6 min的1 km等高高度的Z_H

下载: 导出CSV

表 2 各数据集的整体评价指标

数据集	通道数	RMSE（mm）	CC	MBR	ME（mm）	RE
K_DP	1	4.88	0.78	0.996	-0.024	53%
Z	1	4.65	0.8	1.002	0.011	49%
Z_1~3 km	2	4.71	0.82	0.992	-0.047	45%
Z_Z_DR	2	4.38	0.83	1.000	0.001	46%
Z_K_DP	2	4.16	0.85	1.005	0.027	44%
Z_DR_K_DP	3	4.15	0.85	1.005	0.026	44%
Z_Z_DR_K_DP	3	3.97	0.86	0.992	-0.002	42%
Z_6 min	10	3.78	0.88	1.000	0.006	37%

下载: 导出CSV

参考文献(27)

[1]	GARCIA-PINTADO J, BARBERA G G, ERENA M, et al. Rainfall estimation by rain gauge-radar combination: A concurrent multiplicative-additive approach[J]. Water Resources Research, 45(1): W01415.
[2]	杨艳蓉, 王振会, 张沛源. 利用多普勒天气雷达资料对一次暴雨过程的同化模拟[J]. 南京气象学院学报, 2008, 31(5): 633-639.
[3]	MARSHALL J S, PALMER W Mc K. The distribution of raindrops with size[J]. J Meteor, 1948, 5(4): 165-166.
[4]	SELIGA T A, BRINGI V N. Potential use of radar differential reflectivity measurements at orthogonal polarizations for measuring precipitation[J]. J Appl Meteor, 1976, 15(1): 69-76.
[5]	BRINGI V N, CHANDRASEKAR V, BALAKRISHNAN N, et al. An examination of propagation effects in rainfall on radar measurements at microwave frequencies[J]. Journal of Atmospheric and Oceanic Technology, 1990, 7(6): 829-840.
[6]	刘黎平, 钱永甫, 王致君. 双线偏振雷达测雨效果的对比分析[J]. 大气科学, 1996, 20(5): 615-619.
[7]	殷秀良, 张培昌. 双线偏振雷达测雨公式的对比分析[J]. 南京气象学院学报, 2000, 23(3): 428-434.
[8]	赵果, 楚荣忠, 张彤, 等. 偏振多普勒雷达定量测量降雨精度的改进[J]. 高原气象, 2011, 30(2): 498-507.
[9]	汪舵, 刘黎平, 吴翀. 基于相态识别的S波段双线偏振雷达最优化定量降水估测方法研究[J]. 气象, 2017, 43(9): 1 041-1 051.
[10]	张扬. 业务双偏振雷达网与自动站联合定量降水估测方法及效果分析研究[D]. 北京: 中国气象科学研究院, 2019.
[11]	陈超, 胡志群, 胡胜, 等. CINRAD-SA双偏振雷达资料在降水估测中的应用初探[J]. 气象, 2019, 45(1): 113-125.
[12]	WEI T, LEI Y, LIU W, et al. Ground radar precipitation estimation with deep learning approaches in meteorological private cloud[J]. Journal of Cloud Computing Advances Systems and Applications, 2020, 9(1): 1-12.
[13]	YO T S, SU S H, CHU J L, et al. A deep learning approach to radar-based QPE[J]. Earth and Space Science, 2021, 8(3): 10.1029/2020EA001840.
[14]	WANG S, ZHOU A, YANG M, et al. Service composition in cyber-physical-social systems[J]. IEEE Transactions on Emerging Topics in Computing, 2017, 8(1): 82-91.
[15]	ZHOU J, SUB J, CONG P, et al. Security-critical energy-aware task scheduling for Heterogeneous real-time MPSoCs in IoT[J]. IEEE Transactions on Services Computing, 2020, 13(4): 745-758.
[16]	LIU W, WANG Z, LIU X, et al. A survey of deep neural network architectures and their applications[J]. Neurocomputing, 2017, 234: 11-26.
[17]	DARJI M P, DABHI V K, PRAJAPATI H B, Rainfall forecasting using neural network: A survey[C]//2015 International Conference on Advances in Computer Engineering and Applications. Ghaziabad: IEEE. 2015: 706-713.
[18]	NAYAK D R, MAHAPATRA A, MISHRA P A survey on rainfall prediction using artificial neural network[J]. International journal of Computer Applications, 2013, 72(16): 32-40.
[19]	张晨阳, 杨雪冰, 张文生. 气象大数据超短临精准降水机器学习与典型应用[J]. 农业大数据学报, 2019, 1(1): 78-87.
[20]	陈训来, 陈元昭, 赵春阳, 等. 梯度提升决策树在雷达定量降水估测中的应用研究[J]. 气象科技进展, 2019, 9(3): 132-137.
[21]	傅德胜, 肖晨, 谭畅, 等. RBF神经网络在雷达定量估测降水中的应用研究[J]. 气象科学, 2015, 35(2): 199-203.
[22]	CHEN H, CHANDRASEKAR V, TAN H, et al. Rainfall estimation from ground radar and TRMM precipitation radar using Hybrid Deep Neural Networks[J]. Geophysical Research Letters, 2019, 46(17-18): 10 669-10 678.
[23]	TANG Y, YANG X, ZHANG W, et al. Radar and rain gauge merging-based precipitation estimation via geographical-temporal attention continuous conditional random field[J]. IEEE Transactions on Geoscience and Remote Sensing, 2018, 56(9): 5 558-5 571.
[24]	WU Y, TANG Y, YANG X, et al. Graph convolutional regression networks for quantitative precipitation estimation[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 18(7): 1 124-1 128.
[25]	HAN L, SUN J, ZHANG W, et al. A machine learning nowcasting method based on real‐time reanalysis data[J]. Journal of Geophysical Research: Atmospheres, 2017, 122(7): 4 038-4 051.
[26]	ZHANG C, BENGIO S, HARDT M, et al. Understanding Deep Learning Requires Rethinking Generalization[C]//5th International Conference on Learning Representation. Toulon, France, 2017.
[27]	LOFFE S, SZEGEDY C. Batch Normalization: accelerating deep network training by reducing internal covariate shift[C]//Proceedings of the 32nd International Conference on International Conference on Machine Learning, Lile, France, 2015: 448-456.