Dynamic Inversion Algorithm for Rainfall Intensity Based on Dual-Mode Microwave Radar Combined Rain Gauge
-
摘要: 微波气象雷达探测降雨特征的应用前景较为广泛,但其数据维度单一,测量精度受限于传统反演算法局限性。该文提出基于双模(FMCW-CW)微波雷达联合雨量计的数据进行融合反演降雨强度。针对FMCW模式的雨滴谱数据特征,该文提出基于注意力机制(Attention)连接双层长短期记忆网络(LSTM)的融合算法(LSTM-Attention-LSTM),通过Attention和 LSTM提取雨滴谱数据与实际降雨强度之间的依赖关系,聚焦重要特征并进行解码预测。同时,针对CW模式反演算法难以获取雨滴谱数据,仅能依赖反射率因子和降雨率(Z-R)关系的问题,提出基于扩展卡尔曼算法(EKF)优化Z-R关系,通过动态建模Z-R参数、融合多源数据、施加物理约束,以便准确拟合Z-R关系。实验结果表明:(1) LSTM-Attention-LSTM显著提升降雨率反演精度,相较实际测量降雨强度相关系数(R2)达到0.95,均方根误差(RMSE)为
0.1623 mm/h。(2) 基于EKF优化动态Z-R关系法能够更加精准地确定Z-R关系的参数,拟合结果与实际数据分布情况相关度最高R2为0.972,RMSE为0.1076 mm/h。-
关键词:
- 微波雷达 /
- 降雨强度 /
- 长短期记忆神经网络 /
- 反射率因子和降雨率关系 /
- 扩展卡尔曼滤波
Abstract:Objective Microwave meteorological radar has broad application potential in rainfall detection due to its non-contact measurement, high spatiotemporal resolution, and multi-parameter retrieval capability. However, in the context of climate change, increasingly complex rainfall events require monitoring systems to deliver high-precision, multi-dimensional, real-time data to support disaster warning and climate research. Conventional single-mode radars, constrained by fixed functionalities, cannot fully meet these requirements, which has led to the development of multi-mode radar technology. The dual-mode radar examined in this study employs Frequency Modulated Continuous Wave (FMCW) and Continuous Wave (CW) modes. These modes adopt different algorithmic principles for raindrop velocity measurement: FMCW enables spatially stratified detection and strong anti-interference performance, whereas CW provides more accurate measurements of raindrop fall speed, yielding integral rainfall information in the vertical column. Despite these advantages, retrieval accuracy remains limited by the reliance of traditional algorithms on fixed empirical parameters, which restrict adaptability to regional climate variations and dynamic microphysical precipitation processes, and hinder real-time response to variations in rain Drop Size Distribution (DSD). Ground rain gauges, by contrast, provide near-true reference data through direct measurement of rainfall intensity. To address the above challenges, this paper proposes a dynamic inversion algorithm that integrates dual-mode (FMCW–CW) radar with rain gauge data, enhancing adaptability and retrieval accuracy for rainfall monitoring. Methods Two models are developed for the two radar modes. For the FMCW mode, which can retrieve DSD parameters, a fusion algorithm based on Attention integrated with a double-layer Long Short-Term Memory (LSTM) network (LSTM–Attention–LSTM) is proposed. The first LSTM extracts features from DSD data and rain gauge–measured rainfall intensity through its hidden state output, with a dropout layer applied to randomly discard neurons and reduce overfitting. The Attention mechanism calculates feature similarity using dot products and converts it into attention weights. The second LSTM then processes the time series and integrates the hidden-layer features, which are passed through a fully connected layer to generate the retrieval results. For the CW mode, which cannot directly retrieve DSD parameters and is constrained to the reflectivity factor–Rainfall rate (Z–R) relationship (Z=aRb), an algorithm based on the Extended Kalman Filter (EKF) is proposed to optimize this relationship. The method dynamically models the Z–R parameters, computes the residual between predicted rainfall intensity and rain gauge observations, and updates the prior estimates accordingly. Physical constraints are applied to parameters a and b during state updates to ensure consistency with physical laws, thereby enabling accurate fitting of the Z–R relationship. Results and Discussions Experimental results show that both models enhance the accuracy of rainfall intensity retrieval. For the FMCW mode, the LSTM–Attention–LSTM model applied to the test dataset outperforms traditional physical models, single-layer LSTM, and double-layer LSTM. It effectively captures the temporal variation of rainfall intensity, with the absolute error relative to observed values remaining below 0.25 mm/h ( Fig. 5 ). Compared with the traditional physical model, the LSTM–Attention–LSTM reduces RMSE and MAE by 46% and 38%, achieving values of0.1623 mm/h and 0.147 mm/h, respectively, and increases R2 by 14.5% to 0.95 (Table 2 ). For the CW mode, the Z–R relationship optimized by the EKF model provides the best fit for the Z and R distribution in the validation dataset (Fig. 6 ). Rainfall intensity retrieved with this algorithm on the test set exhibits the smallest deviation from actual observations compared with convective cloud empirical formulas, Beijing plain area empirical formulas, and the dynamic Z–R method. The corresponding RMSE, MAE, and R2 reach0.1076 mm/h, 0.094 mm/h, and 0.972, respectively (Fig. 7 ;Table 4 ).Conclusions This study proposes two multi-source data fusion schemes that integrate dual-mode radar with rain gauges for short-term rainfall monitoring. Experimental results confirm that both methods significantly improve the accuracy of rainfall intensity retrieval and demonstrate strong dynamic adaptability and robustness. -
表 1 LSTM网络参数设置
模型参数 第1层LSTM 第2层LSTM 学习率 0.009 0.001 遗忘率 0.200 — 隐藏层神经单元数量 256 128 最大迭代次数 300 200 表 2 FMCW模式各模型反演结果指标
模型 RMSE MAE R2 雨滴谱算法
LSTM
双层LSTM
LSTM-attention-LSTM0.3009 0.2773 0.2260 0.1623 0.2371 0.2156 0.1982 0.1470 0.8285 0.8544 0.9032 0.9501 表 3 CW模式各Z-R模型最优参数
模型 a b 对流型经验公式
北京平原经验公式
动态Z-R
EKF优化动态Z-R300
386
298
3741.40
1.32
1.50
1.24表 4 CW模式各Z-R模型反演结果指标
模型 RMSE MAE R2 对流型经验公式
北京平原经验公式
动态Z-R
EKF优化Z-R0.2059 0.1770 0.1466 0.1076 0.1799 0.1479 0.1186 0.0940 0.9197 0.9207 0.9593 0.9720 -
[1] SEGOVIA-CARDOZO D A, BERNAL-BASURCO C, and RODRÍGUEZ-SINOBAS L. Tipping bucket rain gauges in hydrological research: Summary on measurement uncertainties, calibration, and error reduction strategies[J]. Sensors, 2023, 23(12): 5385. doi: 10.3390/s23125385. [2] 戴强, 刘超楠, 张亚茹, 等. 基于多模式雷达遥感的陆表降雨反演研究进展[J]. 遥感学报, 2023, 27(7): 1574–1589. doi: 10.11834/jrs.20231768.DAI Qiang, LIU Chaonan, ZHANG Yaru, et al. Development of precipitation retrieval based on multimode radar remote sensing[J]. National Remote Sensing Bulletin, 2023, 27(7): 1574–1589. doi: 10.11834/jrs.20231768. [3] ZHOU Ruiyang, GONG Aofan, and NI Guangheng. A machine learning model for radar quantitative precipitation estimation in Beijing, China[C]. IGARSS 2023 - 2023 IEEE International Geoscience and Remote Sensing Symposium, Pasadena, US, 2023: 3823–3826. doi: 10.1109/IGARSS52108.2023.10282719. [4] FARIZHANDI A A K and MAMIVAND M. Spatiotemporal prediction of microstructure evolution with predictive recurrent neural network[J]. Computational Materials Science, 2023, 223: 112110. doi: 10.1016/j.commatsci.2023.112110. [5] POLZ J, GLAWION L, GEBISSO H, et al. Temporal super-resolution, ground adjustment, and advection correction of radar rainfall using 3-D-convolutional neural networks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2024, 62: 5103710. doi: 10.1109/TGRS.2024.3371577. [6] LI Wenyuan, CHEN Haonan, and HAN Lei. Polarimetric radar quantitative precipitation estimation using deep convolutional neural networks[J]. IEEE Transactions on Geoscience and Remote Sensing, 2023, 61: 4102911. doi: 10.1109/TGRS.2023.3280799. [7] WANG Fuzeng, CAO Yaxi, WANG Qiusong, et al. Estimating precipitation using LSTM-based raindrop spectrum in Guizhou[J]. Atmosphere, 2023, 14(6): 1031. doi: 10.3390/atmos14061031. [8] HARILAL G T, DIXIT A, and QUATTRONE G. Establishing hybrid deep learning models for regional daily rainfall time series forecasting in the United Kingdom[J]. Engineering Applications of Artificial Intelligence, 2024, 133: 108581. doi: 10.1016/j.engappai.2024.108581. [9] 卓健, 廖胜石, 苏传程, 等. MLP在雷达定量降水估测中的应用[J]. 热带气象学报, 2023, 39(3): 289–299. doi: 10.16032/j.issn.1004-4965.2023.027.ZHUO Jian, LIAO Shengshi, SHU Chuancheng, et al. Application of MLP in radar quantitative precipitation estimation[J]. Journal of Tropical Meteorology, 2023, 39(3): 289–299. doi: 10.16032/j.issn.1004-4965.2023.027. [10] CUI Liman, LI Haoren, SU Aifang, et al. Raindrop size distributions in the Zhengzhou extreme rainfall event on 20 July 2021: Temporal-spatial variability and implications for radar QPE[J]. Journal of Meteorological Research, 2024, 38(3): 489–503. doi: 10.1007/s13351-024-3119-9. [11] ZHANG Peng, LIU Xichuan, and PU Kang. Improving quantitative precipitation estimation using adaptive Z-R relationship adjustment: Merging weather radar with commercial microwave links[J]. IEEE Transactions on Geoscience and Remote Sensing, 2025, 63: 2001310. doi: 10.1109/TGRS.2025.3552794. [12] MAITRA A, RAKSHIT G, and JANA S. Three-parameter rain drop size distributions from GPM dual-frequency precipitation radar measurements: Techniques and validation with ground-based observations[J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 4110711. doi: 10.1109/TGRS.2022.3227622. [13] LEIGHTON H, BLACK R, ZHANG Xuejin, et al. The relationship between reflectivity and rainfall rate from rain size distributions observed in hurricanes[J]. Geophysical Research Letters, 2022, 49(23): e2022GL099332. doi: 10.1029/2022GL099332. [14] SHA Lina, LÜ Jingjing, ZHU Bin, et al. Investigation of summer raindrop size distributions and associated relations in the semi-arid region over Inner Mongolian Plateau, China[J]. Advances in Atmospheric Sciences, 2025, 42(5): 1026–1042. doi: 10.1007/s00376-024-4059-0. [15] TAO Shuting, PENG Peng, LI Yunfei, et al. Supervised contrastive representation learning with tree-structured parzen estimator Bayesian optimization for imbalanced tabular data[J]. Expert Systems with Applications, 2024, 237: 121294. doi: 10.1016/j.eswa.2023.121294. [16] 覃睿, 张小琴. 不同输入设置对LSTM洪水预报模型应用效果的影响[J]. 湖泊科学, 2025, 37(4): 1470–1480. doi: 10.18307/2025.0444.QIN Rui and ZHANG Xiaoqin. Comparison on the application performance of LSTM flood forecasting model under different input methods[J]. Journal of Lake Sciences, 2025, 37(4): 1470–1480. doi: 10.18307/2025.0444. [17] LIU Miaomiao, ZUO Juncheng, TAN Jianguo, et al. Comparing and optimizing four machine learning approaches to radar-based quantitative precipitation estimation[J]. Remote Sensing, 2024, 16(24): 4713. doi: 10.3390/rs16244713. [18] LI Jianzhu, SHI Yi, ZHANG Ting, et al. Radar precipitation nowcasting based on ConvLSTM model in a small watershed in North China[J]. Natural Hazards, 2024, 120(1): 63–85. doi: 10.1007/s11069-023-06193-6. [19] KARBASI M, JAMEI M, MALIK A, et al. Multi-steps drought forecasting in arid and humid climate environments: Development of integrative machine learning model[J]. Agricultural Water Management, 2023, 281: 108210. doi: 10.1016/j.agwat.2023.108210. [20] 胡欢, 贾田鹏, 张英. 基于特征点云统计的多传感器融合定位方法[J]. 传感器与微系统, 2025, 44(3): 125–129. doi: 10.13873/J.1000-9787(2025)03-0125-05.HU Huan, JIA Tianpeng, and ZHANG Ying. Multi-sensor fusion localization method based on feature point cloud statistics[J]. Transducer and Microsystem Technologies, 2025, 44(3): 125–129. doi: 10.13873/J.1000-9787(2025)03-0125-05. [21] SITU Zuxiang, WANG Qi, TENG Shuai, et al. Improving urban flood prediction using LSTM-DeepLabv3+ and Bayesian optimization with spatiotemporal feature fusion[J]. Journal of Hydrology, 2024, 630: 130743. doi: 10.1016/j.jhydrol.2024.130743. [22] ELABD E, HAMOUDA H M, ALI M A M, et al. Climate change prediction in Saudi Arabia using a CNN GRU LSTM hybrid deep learning model in al Qassim region[J]. Scientific Reports, 2025, 15(1): 16275. doi: 10.1038/s41598-025-00607-0. [23] 赵城城, 张乐坚, 梁海河, 等. 北京山区和平原地区夏季雨滴谱特征分析[J]. 气象, 2021, 47(7): 830–842. doi: 10.7519/j.issn.1000-0526.2021.07.006.ZHAO Chengcheng, ZHANG Lejian, LIANG Haihe, et al. Microphypical characteristics of the raindrop size distribution between mountain and plain areas over Beijing in summer[J]. Meteorological Monthly, 2021, 47(7): 830–842. doi: 10.7519/j.issn.1000-0526.2021.07.006. -