基于三维卷积的双偏振雷达定量降水估测研究

张毅; 谢宸浩; 陈雨欣; 黎德波; 张永华; 熊梓立

doi:10.16032/j.issn.1004-4965.2025.018

基于三维卷积的双偏振雷达定量降水估测研究

doi: 10.16032/j.issn.1004-4965.2025.018

1.
广州市突发事件预警信息发布中心, 广东广州 510641
2.
广东省气象服务中心, 广东广州 510641

基金项目:

中国气象局智能气象观测技术重点开放实验室项目 ZNGC2024QN18

广东省气象局科学技术研究项目 GRMC2023Q36

广东省气象局科学技术研究项目 GRMC2023Z02

详细信息

通讯作者:
谢宸浩，男，广东省人，工程师，主要从事天气雷达相关研究。E-mail：884498536@qq.com

中图分类号: P412.25
计量
- 文章访问数: 10
- HTML全文浏览量: 0
- PDF下载量: 4
- 被引次数: 0
出版历程
- 收稿日期: 2024-08-15
- 修回日期: 2025-04-03
- 刊出日期: 2025-04-20

Research on Quantitative Precipitation Estimation Using Dual-Polarization Radar Based on 3D Convolution

1.
Guangzhou Emergency Warning Information Release Center, Guangzhou 510641, China
2.
Guangdong Meteorological Service Center, Guangzhou 510641, China

摘要

摘要: 强降水常常引发洪涝灾害，因此提高雷达定量降水估测（QPE）准确性对减轻灾害损失具有重要意义。利用广州双偏振雷达数据与自动站雨量数据生成四维数据集，设计了3DPoly-QPENet、3DTime-QPENet、3DEcho-QPENet三种三维卷积QPE模型进行比较试验。通过测试集的性能评估和典型暴雨个例的检验，得出以下结论：（1）四维数据集相较于传统的三维数据集，在捕捉降水分布特征和提升QPE拟合效果方面提供了更多的可能性；（2）三种三维卷积QPE模型呈现出与结构设计紧密相关的性能差异，其中3DPoly-QPENet在中等降水量区间（15~20 mm·h^-1）的平均绝对误差（MAE）较另两种模型平均降低13%；3DTime-QPENet在高降水量事件（>50 mm·h^-1）的MAE较另两种模型平均降低8.1%；3DEcho-QPENet全局误差均衡性最优，总体MAE较另两种模型平均降低20.4%；（3）三维卷积模型均系统性优于传统Z-R关系方法，平均RMSE降低46.6%，MAE下降48.6%，CC提升21.4%。
- 定量降水估测 /
- 双偏振雷达 /
- 四维数据集 /
- 三维卷积 /
- 深度学习
Abstract: Heavy precipitation often triggers flooding disasters. Therefore, enhancing radar-based quantitative precipitation estimation(QPE) accuracy is critical for disaster mitigation. This study employs Guangzhou dual-polarization radar data and automatic weather station rainfall data to construct a four-dimensional dataset. Three three-dimensional convolutional QPE models—namely, 3DPoly-QPENet, 3DTime-QPENet, and 3DEcho-QPENet—were designed and evaluated through comparative experiments. Based on test-set performance assessments and validation with typical heavy rainfall cases, we draw the following condusions: (1) Compared to traditional three-dimensional datasets, the four-dimensional dataset better captures precipitation distribution characteristics and improves QPE fitting accuracy. (2) The three three-dimensional convolutional QPE models exhibit performance differences tied to their structural designs. Specifically, 3DPoly-QPENet reduces the mean absolute error (MAE) by an average of 13% in moderate precipitation (15-20 mm · h^-1) compared to the other two models. 3DTime-QPENet achieves an average MAE reduction of 8.1% in high-intensity precipitation events (>50 mm · h^-1). 3DEcho-QPENet shows the best global error balance, with an overall MAE reduction of 20.4% on average. (3) All three three-dimensional convolutional models surpass the traditional Z-R relationship method, reducing the root mean square error (RMSE) by an average of 46.6%, lowering MAE by 48.6%, and increasing the correlation coefficient (CC) by 21.4%.
- quantitative precipitation estimation /
- dual-polarization radar /
- four-dimensional dataset /
- threedimensional convolution /
- deep learning

HTML全文

图 1 雷达站和雨量站分布图

黑色X为广州雷达位置，红色圆点为广州雷达100 km范围内的雨量站点。

下载: 全尺寸图片幻灯片

图 2 雷达参量（Z_H、Z_DR和K_DP）-自动站雨量数据集示意图

下载: 全尺寸图片幻灯片

图 3 降水强度分布统计图

下载: 全尺寸图片幻灯片

图 4 模型结构图

a. 3DPoly-QPENet；b. 3DTime-QPENet；c. 3DEcho-QPENet。

下载: 全尺寸图片幻灯片

图 5 3DPoly-QPENet （a）、3DTime-QPENet （b）、3DEcho-QPENet （c）三种模型预测结果散点密度图

下载: 全尺寸图片幻灯片

图 6 北京时间2022年6月14日09:00降水个例

图a为观测的降水场、图b为Z-R关系估测降水、图c为3DPoly-QPENet模型估测降水、图d为3DTime-QPENet模型估测降水、图e为3DEcho-QPENet模型估测降水）及北京时间2022年9月8日19:00降水个例（图f为观测的降水场、图g为Z-R关系估测降水、图h为3DPoly-QPENet模型估测降水、图i为3DTimeQPENet模型估测降水、图j为3DEcho-QPENet模型估测降水

下载: 全尺寸图片幻灯片

表 1 模型预测结果评估指标对比

模型	雨量等级/(mm·h^-1)	RE/%	MAE/mm	RMSE/mm	Bias	CC	R²
3DPoly-QPENet	Total	58.03	2.840	4.696	1.260	0.852	0.711
	0.1~5	56.24	2.110	2.911	2.051
	5~15	34.71	3.598	5.430	1.008
	15~20	38.29	5.729	7.227	-2.430
	20~50	49.89	10.604	12.923	-8.274
	>50	37.79	19.554	23.345	-18.097
3DTime-QPENet	Total	38.28	1.956	4.331	-0.089	0.869	0.748
	0.1~5	35.94	0.937	1.841	0.567
	5~15	49.38	4.004	5.516	-1.043
	15~20	49.09	6.773	8.452	-3.178
	20~50	46.42	10.778	13.525	-5.371
	>50	31.58	16.859	21.861	-6.650
3DEcho-QPENet	Total	34.91	1.847	4.018	-0.082	0.885	0.783
	0.1~5	32.99	0.924	1.729	0.048
	5~15	43.79	3.660	5.134	-0.647
	15~20	44.61	6.317	7.918	-3.085
	20~50	41.53	9.633	12.037	-5.824
	>50	32.62	17.223	21.882	-12.898

下载: 导出CSV

表 2 不同雷达定量降水估测方法对比

时间	降水估测方法	RE/%	MAE/mm	RMSE/mm	Bias	CC
2022年6月14日09:00(BJT)	Z-R关系	66.14	8.541	14.528	-1.572	0.730
	3DPoly-QPENet	37.73	3.703	6.710	-1.121	0.913
	3DTime-QPENet	51.75	5.131	8.414	-1.647	0.867
	3DEcho-QPENet	41.03	4.068	7.279	-1.462	0.908
2022年9月8日19:00(BJT)	Z-R关系	61.82	7.967	14.398	-1.654	0.728
	3DPoly-QPENet	48.23	4.185	7.374	-0.599	0.884
	3DTime-QPENet	52.03	5.383	9.243	1.122	0.870
	3DEcho-QPENet	47.80	4.148	7.308	-0.617	0.886

下载: 导出CSV

参考文献(28)

[1]	MARSHALL J S, PALMER W Mc K. The distribution of raindrops with size[J]. J Atmos Sci, 1948, 5(4): 165-166.
[2]	NAYAK D R, MAHAPATRA A, MISHRA P. A survey on rainfall prediction using artificial neural network[J]. Int J Comput Appl, 2013, 72 (16): 32-40.
[3]	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.
[4]	WEI T, LEI Y, LIU W, et al. Ground radar precipitation estimation with deep learning approaches in meteorological private cloud[J]. J Cloud Comput, 2020, 9(1): 1-12.
[5]	RUMELHART D E, HINTON G E, WILLIAMS R J. Learning representations by back-propagating errors[M]//Neurocomputing: Foundations of Research. Cambridge: MIT Press, 1988: 696-699.
[6]	皇甫江, 胡志群, 郑佳锋, 等. 利用深度学习开展偏振雷达定量降水估测研究[J]. 气象学报, 2022, 80(4): 565-577.
[7]	蔡康龙, 胡志群, 谭浩波, 等. 利用卷积神经网络开展偏振雷达定量降水估测研究[J]. 热带气象学报, 2024, 40(1): 64-74.
[8]	张永华. 基于深度学习的华南登陆台风偏振雷达定量降水估测研究[D]. 南京: 南京信息工程大学, 2022.
[9]	PENG X, LI Q, JING J. CNGAT: A graph neural network model for radar quantitative precipitation estimation[J]. IEEE Trans Geosci Remote Sens, 2021, 60(1): 1-14.
[10]	郑玉, 魏鸣, 李南. 基于循环神经网络改进雷达定量估测强降水[J]. 中国科技论文, 2020, 15(5): 585-592.
[11]	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 Trans Geosci Remote Sens, 2018, 56(9): 5135-5146.
[12]	MORAUX A, DEWITTES S, Cornelis B, et al. A deep learning multimodal method for precipitation estimation[J]. Remote Sens, 2021, 13 (16): 3278.
[13]	GAVAHI K, FOROUMAND E, MORADKHANI H. A deep learning-based framework for multi-source precipitation fusion[J]. Remote Sens Environ, 2023, 295: 113723.
[14]	WANG L. Improving radar quantitative precipitation estimation through optimizing radar scan strategy and deep learning[D]. Colorado: Colorado State University, 2024.
[15]	KIM Y, HONG S. Very short-term prediction of weather radar-based rainfall distribution and intensity over the Korean peninsula using convolutional long short-term memory network[J]. Asia-Pac J Atmos Sci, 2022, 58(4): 489-506.
[16]	MA Y, CHANDRASEKAR V, BISWASi S K. A Bayesian correction approach for improving dual-frequency precipitation radar rainfall rate estimates[J]. J Meteor Soc Japan Ser Ⅱ, 2020, 98(3): 511-525.
[17]	易雷. 基于时空预测网络和卷积神经网络的短临降水预测研究[D]. 南京: 南京信息工程大学, 2022.
[18]	TRAN D, BOURDEV L, FERGUS R, et al. Learning spatiotemporal features with 3d convolutional networks[C]//Proceedings of the IEEE international conference on computer vision. Santiago: IEEE, 2015: 4489-4497.
[19]	RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation[C]//Medical image computing and computer-assisted intervention-MICCAI 2015: 18th international conference. Munich: Springer International Publishing, 2015: 234-241.
[20]	ÇIÇEK Ö, ABDULKADIR A, LIENKAMP SS, et al. 3D U-Net: learning dense volumetric segmentation from sparse annotation[C]//Medical Image Computing and Computer-Assisted Intervention-MICCAI 2016: 19th International Conference, Athens: Springer International Publishing, 2016.
[21]	ZHANG Y, LIU L, BI S, et al. Analysis of dual-polarimetric radar variables and quantitative precipitation estimators for landfall typhoons and squall lines based on disdrometer data in southern China[J]. Atmosphere, 2019, 10(1): 30.
[22]	GABELLA M, NOTARPIETRO R. Ground clutter characterization and elimination in mountainous terrain[C]//Proceedings of ERAD. 2002, 305(311).
[23]	HELMUS J J, COLLIS S M. The Python ARM Radar Toolkit(Py-ART), a library for working with weather radar data in the Python programming language[J]. J Open Res Softw, 2016, 4.
[24]	ZHANG Y, BI S, LIU L, et al. Deep learning for polarimetric radar quantitative precipitation estimation during landfalling typhoons in South China[J]. Remote Sens, 2021, 13(16): 3157.
[25]	NEYMAN J. On the two different aspects of the representative method: the method of stratified sampling and the method of purposive selection[M]//Breakthroughs in statistics: Methodology and distribution. NY: Springer New York, 1992: 123-150.
[26]	REN J, YU C, MA K, et al. Balanced MSE for Imbalanced Visual Regression[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2022: 4175-4184.
[27]	LEGATES D R. Rea-time calibration of radar precipitation estimates[J]. Prof Geogr, 2000, 52(2): 235-246.
[28]	高烽, 周新尧, 孔凡超, 等. 滑动时间窗长度对太行山区雷达反演降水的影响[J]. 气象, 2018, 44(4): 538-547.