基于分级跨级权重评分的快速更新数值模式集合降水预报订正技术研究

Research on rapid refresh numerical model ensemble precipitation forecasting correction technology based on hierarchical cross-level weighted scoring

摘要

基于中国气象局广东快速更新同化数值系统 CMA-GD-R3 和中国气象局上海快速更新同化数值预报系统 CMA-SH3 和中国气象局中尺度天气数值预报系统 CMA-MESO (以下简称原模式) 的 24 h 内逐 1 h、3 h 起报降水产品以及湖南区域气象观测站降水实况资料，在传统降水检验方法的基础上创建了分级跨级权重降水评分法 (Cross-magnitude Weight，CMW)，并设计了基于 CMW 优选的时间滞后和空间平移快速更新数值模式集合降水订正预报模型 (STDA-CMW)，实现了效果更优的逐时更新的未来 12 h 内逐时定量降水预报。为验证 STDA-CMW 各模块 (分别为缺失空间平移模块、缺失强降水融合模块、将 CMW 优选替换为 TS 评分优选方案，以下简称对比模型) 的有效性，将该模型、三个原模式和对比模型进行对照试验，以检验各模块的独立贡献。结果表明：(1) 对于一般性降水[0.1，20) mm·h⁻¹，STDA-CMW 模型 TS 评分 (0.325) 优于所有原模式及对比模型。空间平移、CMW 优选、强降水融合模块均有正贡献，其中 CMW 优选模块贡献最大，明显优于传统 TS 评分优选模块，空间平移与强降水融合模块贡献接近。(2) 对于强降水 (≥20 mm·h⁻¹)，STDA-CMW 模型在 TS 评分 (0.011)、空报率 (0.971)、平均绝对误差 (25.141 mm) 和 CMW (−0.004) 均优于原模式及对比模型。三个模块均有正贡献，但只有三个模块同时集成时，预报效果才能优于所有原模式预报。(3) 三个模块贡献各有侧重，空间平移模块主要提升强降水命中率，强降水订正模块同时提升命中率并降低空报率，CMW 优选模块主要降低强降水空报率，且对强降水量级的贡献明显大于另外两个模块。三个增强模块对短临降水预报的贡献各有侧重，且仅在三者协同集成时才能达到最优的强降水预报效果。上述研究成果为基于快速更新中尺度数值模式的短临降水订正提供了一种可行路径，在降低空报率和减少平均绝对误差等业务关键指标上具有相对优势。

Abstract

Based on the 1-hourly and 3-hourly updated precipitation products within 24 hours from three models (China Meteorological Administration Guangdong Rapid Refresh and Assimilation Numerical System CMA-GD-R3, China Meteorological Administration Shanghai Rapid Refresh and Assimilation Numerical Prediction System CMA-SH3, and China Meteorological Administration Mesoscale Weather Numerical Prediction System CMA-MESO, hereinafter referred to as the original models) and the precipitation observations from regional automatic weather stations in Hunan Province, this study develops a novel verification metric named Cross-magnitude Weight (CMW) on the basis of traditional precipitation verification methods. A short-term ensemble correction and forecasting model (STDA-CMW) is designed, which incorporates time-lagging and spatial shifting techniques optimized by CMW, to achieve more effective hourly-updated quantitative precipitation forecasts for the next 12 hours. The proposed model is compared with the three original models and three sets of comparative experiments designed to validate the effectiveness of each module of STDA-CMW (namely, the scheme without the spatial shifting module, the scheme without the heavy precipitation fusion module, and the scheme replacing CMW optimization with the traditional TS score optimization), in order to examine the independent contributions of each module. The results are as follows. (1) For general precipitation [0.1, 20) mm·h⁻¹, the STDA-CMW model achieves a TS score (0.325) superior to all original models and comparative models. All three enhancement modules (spatial shifting module, CMW optimization module, and heavy precipitation fusion module) make positive contributions, among which the CMW optimization module contributes the most, significantly outperforming the traditional TS score optimization, while the contributions of the spatial shifting module and the heavy precipitation fusion module are comparable. (2) For heavy precipitation (≥20 mm·h⁻¹), the STDA-CMW model outperforms the original models and comparative models in terms of TS score (0.011), false alarm rate (0.971), mean absolute error (25.141 mm), and CMW (-0.004). All three modules make positive contributions, but the forecast performance can surpass that of all original models only when the three modules are integrated simultaneously. (3) The contributions of the three modules have distinct focuses. The spatial shifting module primarily improves the hit rate of heavy precipitation; the heavy precipitation correction module enhances both the hit rate and reduces the false alarm rate; the CMW optimization module mainly reduces the false alarm rate of heavy precipitation and makes a significantly greater contribution to the magnitude of heavy precipitation than the other two modules. The three enhancement modules contribute differently to short-term precipitation forecasting, and optimal heavy precipitation forecasts can only be achieved when the three modules are integrated synergistically. The above findings provide a feasible pathway for short-term precipitation correction based on rapidly updated mesoscale numerical models, demonstrating relative advantages in reducing false alarm rates and mean absolute error—key operational early warning metrics.

关键词

时间滞后 / 空间平移 / 分级跨级权重评分 / 快速更新数值模式 / 逐小时降水预报 / 短时强降水

Key words

time-Lag / space translation / Cross-Magnitude Weigh / rapid refresh numerical forecasting model / hourly precipitation forecast / short term heavy precipitation

参考文献

[1] 曹伟华,南刚强,陈明轩,等. 2022.基于深度学习的京津冀地区精细尺度降水临近预报研究[J].气象学报,80(4):546-564.

[2] Cao W H, Nan G Q, Chen M X, et al. 2022. A study on fine scale precipitation nowcasting in Beijing—Tianjin—Hebei region based on deep learning [J].Acta Meteorologica Sinica,80(4):546-564 (in Chinese). doi:10.11676/qxxb2022.027

[3] 陈博宇,代刊,郭云谦. 2015. 2013 年汛期 ECMWF 集合统计量产品的降水预报检验与分析[J].暴雨灾害,34(1):64-73.

[4] Chen B Y, Dai K, Guo Y Q. 2015. Verification and analysis of precipitation forecasts of ECMWF ensemble statistic products in the rainy season of 2013 [J].Torrential Rain and Disasters,34(1):64-73 (in Chinese). doi:10.3969/j.issn.1004−9045.2015.01.009

[5] 陈敏,范水勇,仲跻芹,等. 2010.全球定位系统的可降水量资料在北京地区快速更新循环系统中的同化试验[J].气象学报,68(4):450-463.

[6] Chen M, Fan S Y, Zhong J Q, et al. 2010. An experimental study of assimilating the global positioning system—precipitable water vapor observations into the rapid updated cycle system for the Beijing area [J].Acta Meteorologica Sinica,68(4):450-463 (in Chinese). doi:10.11676/qxxb2010.044

[7] 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[J].气象学报,71(3):397-415.

[8] Cheng C L, Chen M X, Wang J J, et al. 2013. Short—term quantitative precipitation forecasting experiment based on fusion of radar extrapolation nowcasting and mesoscale numerical forecast [J].Acta Meteorologica Sinica,71(3):397-415 (in Chinese). doi:10.11676/qxxb2013.041

[9] 崔粉娥,畅巨峥,燕成玉,等. 2019.一次燕山南麓局地大暴雨的成因分析及数值预报检验[J].气象科技进展,9(6):53-60.

[10] Cui F E, Chang J Z, Yan C Y, et al. 2019. Cause analysis and numerical prediction test of a local heavy rain at the southern foot of Yanshan Mountain [J].Advances in Meteorological Science and Technology,9(6):53-60 (in Chinese). doi:10.3969/j.issn.2095−1973.2019.06.009

[11] 高梦竹,刘松涛,陈雪,等. 2025.基于多量级 TS 评分动态权重的多模式融合降水预报方法[J].暴雨灾害,44(3):371-379.

[12] Gao M Z, Liu S T, Chen X, et al. 2025. A multi—model blending precipitation prediction method based on the dynamic weight of multi—level TS score [J].Torrential Rain and Disasters,44(3):371-379 (in Chinese). doi:10.12406/byzh.2024−072

[13] 胡嘉缨,董春卿,操俊伟. 2023.山西复杂地形下 CMA—MESO 3 km 系统降水预报检验及订正[J].暴雨灾害,42(4):384-394.

[14] Hu J Y, Dong C Q, Cao J W. 2023. Assessment and correction of precipitation forecast with the CMA—MESO 3 km model under the complex terrain in Shanxi Province [J].Torrential Rain and Disasters,42(4):384-394 (in Chinese). doi:10.12406/byzh.2022−218

[15] 廖慕科,王俊,唐熠,等. 2026.短时强降水和持续强降水过程雨滴谱特征对比分析[J].暴雨灾害,45(2):229-238.

[16] Liao M K, Wang J, Tang Y, et al. 2026. Comparative analysis of raindrop spectrum characteristics between short—term heavy precipitation and continuous heavy precipitation processes [J].Torrential Rain and Disasters,45(2):229-238 (in Chinese). doi:10.12406/byzh.2024−149

[17] 毛冬艳,周雨华,张芳华. 2006. 2005 年初夏湖南致洪大暴雨中尺度分析[J].气象,32(3):63-70.

[18] Mao D Y, Zhou Y H, Zhang F H. 2006. Mesoscale analysis of torrential rain causing flood in Hunan in early summer of 2005 [J].Meteorological Monthly,32(3):63-70 (in Chinese). doi:10.7519/j.issn.1000−0526.2006.3.011

[19] 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[J].气象学报,79(1):132-149.

[20] Su X, Yuan H L, Zhu Y J. 2021. Comparative study of four objective correction methods for quantitative precipitation forecast [J].Acta Meteorologica Sinica,79(1):132-149 (in Chinese). doi:10.11676/qxxb2020.071

[21] 唐文苑,郑永光. 2019.基于快速更新同化数值预报的小时降水量时间滞后集合订正技术[J].气象,45(3):11-23.

[22] Tang W Y, Zheng Y G. 2019. Ensemble correction technique of hourly precipitation time lag based on fast updating assimilation numerical forecast [J].Meteorological Monthly,45(3):11-23 (in Chinese). doi:10.7519/j.issn.1000−0526.2019.03.002

[23] 王玉虹,孔德璇,朱寿鹏,等. 2025.基于深度学习的京津冀地区短时降水预报订正[J].应用气象学报,36(3):257-267.

[24] Wang Y H, Kong D X, Zhu S P, et al. 2025. Correction of short—term precipitation forecast in the Beijing—Tianjin—Hebei region based on deep learning [J].Journal of Applied Meteorological Science,36(3):257-267 (in Chinese). doi:10.11898/1001−7313.20250301

[25] 徐成鹏,曹勇,张恒德,等. 2022. U—Net 模型在京津冀临近降水预报中的应用和检验评估[J].气象科学,42(6):781-792.

[26] Xu C P, Cao Y, Zhang H D, et al. 2022. Application and test evaluation of U—Net model in Beijing—Tianjin—Hebei precipitation nowcasting [J].Journal of the Meteorological Sciences,42(6):781-792 (in Chinese). doi:10.12306/2022jms.0078

[27] 徐月飞,顾松山,黄兴友,等. 2011.计算几何法在风暴识别中的应用[J].热带气象学报,27(2):173-178.

[28] Xu Y F, Gu S S, Huang X Y, et al. 2011. Application of computational geometry method in storm identification [J].Journal of Tropical Meteorology,27(2):173-178 (in Chinese). doi:10.3969/j.issn.1004−4965.2011.02.005

[29] 闫之辉,王雨,朱国富. 2010.国家气象中心业务数值预报发展的回顾与展望[J].气象,36(7):26-32.

[30] Yan Z H, Wang Y, Zhu G F. 2010. Review and prospect of operational numerical forecast development at National Meteorological Center [J].Meteorological Monthly,36(7):26-32 (in Chinese). doi:10.7519/j.issn.1000−0526.2010.7.007

[31] 杨丽敏,格央,次旦巴桑,等. 2019. SWC—WARMS 模式产品对西藏灾害性天气过程的预报能力检验[J].西藏科技,(5):54-56.

[32] Yang L M, Ge Y, Ci D B S, et al. 2019. Prediction ability test of SWC—WARMS model products for disastrous weather processes in Tibet [J].Tibet Science and Technology,(5):54-56 (in Chinese). doi:10.3969/j.issn.1004−3403.2019.05.018

[33] 杨楠,邓荣耀,张兰英,等. 2025.基于风云卫星产品的一次宜宾强降水过程分析[J].高原山地气象研究,45(S1):13-18.

[34] Yang N, Deng R Y, Zhang L Y, et al. 2025. Analysis of a heavy precipitation process in Yibin based on Fengyun satellite products [J].Plateau and Mountain Meteorology Research,45(S1):13-18 (in Chinese). doi:10.3969/j.issn.1674−2184.2025.Z1.003

[35] 俞小鼎,郑永光. 2020.中国当代强对流天气研究与业务进展[J].气象学报,78(3):391-418.

[36] Yu X D, Zheng Y G. 2020. Research and operational progress of severe convective weather in China [J].Acta Meteorologica Sinica,78(3):391-418 (in Chinese). doi:10.11676/qxxb2020.035

[37] 袁凯,庞晶,李武阶,等. 2023.深度学习模型在 2021 年汛期武汉市雷达回波临近预报中的应用评估[J].干旱气象,41(1):173-185.

[38] Yuan K, Pang J, Li W J, et al. 2023. Application evaluation of deep learning models in radar echo nowcasting in Wuhan during flood season of 2021 [J].Journal of Arid Meteorology,41(1):173-185 (in Chinese). doi:10.11755/j.issn.1006−7639(2023)−01−0173

[39] 袁良,谭江红,闫彩霞,等. 2024.概率匹配订正法在湖北襄阳地区降水预报中的应用[J].沙漠与绿洲气象,18(2):92-98.

[40] Yuan L, Tan J H, Yan C X, et al. 2024. Application of probability matching correction method in precipitation forecast in Xiangyang area of Hubei Province [J].Desert and Oasis Meteorology,18(2):92-98 (in Chinese). doi:10.12057/j.issn.1002−0799.2024.02.012

[41] 曾小团,邹晨曦,范娇,等. 2024.基于 GRU 深度学习的短时临近降水预报订正方法[J].应用气象学报,35(5):513-525.

[42] Zeng X T, Zou C X, Fan J, et al. 2024. Correction method of short—term and nowcasting precipitation forecast based on GRU deep learning [J].Journal of Applied Meteorological Science,35(5):513-525 (in Chinese). doi:10.11898/1001−7313.20240501

[43] 张涵斌,李玉焕,陈敏,等. 2020.集合变分混合同化方案在快速循环同化系统中的应用研究[J].大气科学,44(6):1349-1363.

[44] Zhang H B, Li Y H, Chen M, et al. 2020. Study on the application of ensemble variational hybrid assimilation scheme in fast cycling assimilation system [J].Chinese Journal of Atmospheric Sciences,44(6):1349-1363 (in Chinese). doi:10.3878/j.issn.1006−9895.2008.20118

[45] 张亚萍,张焱,翟丹华,等. 2022.深度学习方法在强降水临近预报技术中的应用及其改进思考[J].暴雨灾害,41(5):506-514.

[46] Zhang Y P, Zhang Y, Zhai D H, et al. 2022. Thoughts on application and improvement of deep learning in severe precipitation nowcasting technology [J].Torrential Rain and Disasters,41(5):506-514 (in Chinese). doi:10.12406/byzh.2022−079

[47] Benjamin S G, Weygandt S S, Brown J M, et al. 2016. A North American hourly assimilation and model forecast cycle: The Rapid Refresh [J].Monthly Weather Review,144(4):1669-1694. doi:10.1175/MWR−D−15−0242.1

[48] Dixon M, Wiener G. 1993. TITAN: Thunderstorm identification, tracking, analysis, and nowcasting a radar—based methodology [J].Journal of Atmospheric and Oceanic Technology,10(6):785-797. doi:10.1175/1520−0426(1993)0102.0.CO;2

[49] Han L, Chen M X, Chen K K, et al. 2021. A deep learning method for bias correction of ECMWF 24—240 h forecasts [J].Advances in Atmospheric Sciences,38(9):1444-1459. doi:10.1007/s00376−021−0215−y

[50] Lu C, Yuan H L, Schwartz B E, et al. 2007. Short—range numerical weather prediction using time—lagged ensembles [J].Weather and Forecasting,22(3):580-595. doi:10.1175/WAF999.1

[51] Mittermaier M P. 2007. Improving short—range high—resolution model precipitation forecast skill using time—lagged ensembles [J].Quarterly Journal of the Royal Meteorological Society,133(627):1487-1500. doi:10.1002/qj.135

[52] Shi X J, Chen Z, Wang H, et al. 2015. Convolutional LSTM network: A machine learning approach for precipitation nowcasting [C]//Advances in Neural Information Processing Systems 28 (NIPS 2015): 802-810

[53] Shi X J, Gao Z, Lausen L, et al. 2017. Deep learning for precipitation nowcasting: A benchmark and a new model [C]//Advances in Neural Information Processing Systems 30 (NIPS 2017): 5617-5627. doi:10.48550/arXiv.1706.03458

[54] Wilson J W, Feng Y, Chen M, et al. 2010. Nowcasting challenges during the Beijing Olympics: Successes, failures, and implications for future nowcasting systems [J].Weather and Forecasting,25(6):1691-1714. doi:10.1175/2010WAF2222417.1

基金资助

全国暴雨研究开放基金项目(BYKJ2024M08)

中国气象局复盘总结专项(FPZJ2026-086)

湖南省重点研发计划项目(2025AQ2012)

中国气象局创新发展专项(CXFZ2023J025，CXFZ2026QA15)

湖南省自然科学基金项目(2026JJ80889，2026JJ80880)