Application and analysis of multi-model synamic fusion correction method in heavy to rainstorm forecast in Qinghai Province

doi:10.11755/j.issn.1006-7639-2026-01-0149

Abstract

Abstract:

To improve the accuracy of 24-hour heavy to rainstorm forecasts in Qinghai Province, particularly the capability to capture sudden rainstorm events, this study first evaluates and selects the optimal models using precipitation observations and multiple model forecast datasets from July to September 2023 in Qinghai Province. After applying quantile mapping correction to the selected model forecast products, a combination of multi-model dynamic fusion and intensity-based calibration is employed for objective heavy to rainstorm forecasting over the Tibetan Plateau. The results show that CMA-SH9 (China Meteorological Administration Shanghai 9-km model), CMA-BJ (China Meteorological Administration Beijing Rapid-Update Cycling Forecast System), and SCMOC (Central Meteorological Observatory Guidance Forecast Products) demonstrate relatively superior performance. Therefore, these three models are selected as the preferred models for fusion. After quantile mapping correction and subsequent dynamic fusion of the three preferred models, verification metrics including the threat score (TS) and accuracy of categorical precipitation forecasts improved to varying degrees across all intensity thresholds, although the forecasts exhibited a systematic positive bias with overestimated intensity. The intensity-based calibration applied after dynamic fusion of the three preferred models significantly reduced forecast bias, increasing the categorical forecast accuracy by 7.8%-27.7% and the TS score by 9.3%-22.8% compared to single models. Case analysis of a regional heavy to rainstorm event indicates that the bias of the dynamically fused and intensity-calibrated forecast is closer to 1, with the predicted rainband position, particularly the heavy to rainstorm area, more consistent with observations. Compared with the best single model, its categorical forecast accuracy, TS score for heavy to rainstorm, and probability of detection (POD) for heavy to rainstorm increased by 2.2%, 19.4%, and 27.4%, respectively.

Key words: heavy to rainstorm forecast, multi-model dynamic fusion correction, quantile mapping method, Qinghai Province

摘要：

为提升青海省未来24 h大到暴雨预报准确率，尤其是对突发性暴雨事件的捕捉能力，本文利用青海省2023年7—9月降水观测和多个模式预报资料进行检验优选，再对优选模式预报产品进行分位数映射订正，将多模式动态融合与分级强度订正方法相结合，应用于青藏高原地区的大到暴雨客观预报。结果表明：中国气象局（China Meteorological Administration，CMA）华东区域数值预报系统（CMA-SH9）、中国气象局北京快速更新循环数值预报系统（CMA-BJ）及中央台指导预报产品（SCMOC）在青海省大到暴雨预报中效果相对较优，因此最终考虑选取这3种模式作为优选模式进行融合；对3类优选模式分位数映射订正再动态融合后，各量级晴雨预报准确率和TS（Threat Score）评分等检验指标均有不同程度提升，但强度偏大，存在系统性正偏差；3类优选模式动态融合分级订正明显改善了预报偏差，较单模式预报晴雨预报准确率提升7.8%~27.7%，TS评分提升9.3%~22.8%；区域性大到暴雨个例分析表明：动态融合分级订正预报偏差更接近1，降水预报的雨带位置特别是大到暴雨落区也更接近降水实况；其晴雨预报准确率、大到暴雨TS评分、大到暴雨命中率分别较最优单模式提升2.2%、19.4%、27.4%。

关键词: 大到暴雨预报, 多模式动态融合订正, 分位数映射方法, 青海省

CLC Number:

P457.6

WANG Qian, LI Jing, ZHANG Qingmei, LI Jinhai, ZHAXI Cairang, QI Wanpeng. Application and analysis of multi-model synamic fusion correction method in heavy to rainstorm forecast in Qinghai Province[J]. Journal of Arid Meteorology, 2026, 44(1): 149-158.

王倩, 李静, 张青梅, 李金海, 扎西才让, 祁万鹏. 多模式动态融合订正方法在青海省大到暴雨预报中的应用与分析[J]. 干旱气象, 2026, 44(1): 149-158.

Figures/Tables 9

Fig.1 Spatial distribution of precipitation observation stations in Qinghai Province

Fig.2 Comprehensive verification results of 3-hourly and 24-hour accumulated precipitation forecasts at 24-hour lead time from multiple products in Qinghai Province from July to September 2023

Tab.1 The percentiles of daily precipitation from observations and multiple forecast products before and after correction in Qinghai Province from July to September 2023

百分位数	观测降水量/mm	不同预报产品/mm
百分位数	观测降水量/mm	CMA-BJ	SCMOC	CMA-SH9	QM-BJ	QM-SCMOC	QM-SH9
20	0	0	0	0	0	0	0
40	0	0.1	0	0	0	0	0
50	0	0.3	0.5	0.4	0	0	0
55	0.1	0.6	0.9	0.8	0	0	0
60	0.3	1.0	1.3	1.5	0.1	0	0.1
65	0.6	1.5	1.8	2.6	0.4	0.2	0.4
70	1.2	2.1	2.3	4.0	1.0	0.7	1.0
75	2.1	3.0	3.0	5.7	1.8	1.4	2.0
80	3.3	4.1	3.8	8.0	3.2	2.7	3.4
85	5.0	5.7	4.0	10.8	5.4	4.7	5.5
90	7.2	8.0	6.7	14.8	8.7	8.0	8.8
95	11.5	12.4	11.6	21.0	15.3	14.6	15.1
99	22.7	23.6	20.8	34.6	32.9	33.1	32.6
99.5	29.1	29.3	26.6	40.0	43.2	43.4	41.2
99.9	44.0	42.8	35.7	51.7	77.6	78.5	64.8
99.99	69.2	65.8	48.0	67.5	142.1	139.6	102.9
99.999	89.7	90.9	54.2	81.4	162.2	151.8	135.2
100	96.2	114.9	57.1	103.3	200.1	171.5	149.0

Fig.3 The cumulative probability distribution curves of daily precipitation from observations and multiple forecast products before and after correction in Qinghai Province from July to September 2023

Fig.4 Verification of 3-hourly and 24-hour cumulative precipitation forecasts at 24-hour lead time before and after correction for multiple forecast products in Qinghai Province from July to September 2023

Fig.5 Trends of standardized weight coefficients for 3-hourly forecasts from three preferred models with dynamic fusion at 08:00 (a) and 20:00 (b) in Qinghai Province from July to September 2023

Fig.6 Comprehensive verification of 3-hourly and 24-hour cumulative precipitation forecasts at 24-hour lead time from multiple forecast products in Qinghai Province from July to September 2023

Tab.2 Comprehensive verification of 3-hourly and 24-hour cumulative precipitation forecasts at 24-hour lead time from multiple forecast products from 08：00 on 3 to 08：00 on 5 September 2024 in Qinghai Province

预报间隔/h	检验指标	CMA-BJ	CMA-SH9	SCMOC	QH-M1	QH-M2	QH-SPCC
3	PC	0.63	0.66	0.57	0.71	0.80	0.58
	TS	0.56	0.58	0.53	0.62	0.69	0.53
	PO	0.03	0.02	0.02	0.02	0.08	0.03
	FAR	0.43	0.41	0.47	0.37	0.27	0.46
	BIAS	1.71	1.67	1.83	1.55	1.25	1.79
	TS（≥10 mm）	0.13	0.14	0.15	0.30	0.17	0.20
	BIAS（≥10 mm）	0.47	0.81	0.45	2.93	0.38	2.50
	POD（≥10 mm）	0.17	0.22	0.19	0.90	0.22	0.59
24	PC	0.89	0.89	0.90	0.90	0.92	0.89
	TS	0.89	0.89	0.90	0.90	0.91	0.89
	PO	0.001	0.001	0.003	0.001	0.01	0.002
	FAR	0.11	0.11	0.10	0.10	0.08	0.11
	BIAS	1.12	1.12	1.11	1.11	1.07	1.12
	TS（≥25 mm）	0.26	0.31	0.31	0.28	0.37	0.23
	BIAS（≥25 mm）	0.81	1.52	1.61	3.56	2.36	3.94
	POD（≥25 mm）	0.38	0.63	0.73	0.99	0.93	0.93

Fig.7 Distribution of accumulated precipitation from observations and multiple forecast products over eastern Qinghai Province from 08：00 on 3 September to 08：00 on 5 September 2024 （Unit： mm）

References 32

[1]	曹萍萍, 肖递祥, 龙柯吉, 等, 2023a. 基于分位数映射法的四川省ECMWF模式降水预报误差订正分析[J]. 干旱气象, 41(4): 666-675
[2]	曹萍萍, 杨康权, 肖递祥, 等, 2023b. 2022年汛期四川多模式降水预报检验[J]. 高原山地气象研究, 43(2):28-35.
[3]	曹勇, 张恒德, 郭云谦, 等, 2022. 基于多模式预报优选融合技术的短时定量降水预报[J]. 河海大学学报:自然科学版, 50(5):10-15.
[4]	陈锦鹏, 冯业荣, 蒙伟光, 等, 2021. 基于卷积神经网络的逐时降水预报订正方法研究[J]. 气象, 47(1):60-70.
[5]	陈晓燕, 孔祥伟, 彭筱, 等, 2022. 全球和区域数值模式在甘肃2020年汛期降水预报中的检验评估[J]. 干旱气象, 40(3):524-535. DOI
[6]	狄靖月, 赵琳娜, 张国平, 等, 2013. 降水集合预报集成方法研究[J]. 气象, 39(6): 691-698.
[7]	吉璐莹, 智协飞, 季焱, 等, 2025. 基于多模式集成的降水空间结构预报改进研究[J]. 大气科学, 49(1): 257-278.
[8]	焦洋, 郑丽娜, 张永婧, 等, 2024. 两种降水客观统计方法对ECMWF集合平均降水预报的订正研究[J]. 干旱气象, 42(2): 293-304 DOI
[9]	雷华锦, 马佳培, 李弘毅, 等, 2020. 基于分位数映射法的黑河上游气候模式降水误差订正[J]. 高原气象, 39(2):266-279. DOI
[10]	林春泽, 祁海霞, 智协飞, 等, 2013. 中国夏季降水多模式集成概率预报研究[J]. 暴雨灾害, 32(4):354-359.
[11]	刘娜, 黄武斌, 杨建才, 等, 2022. 基于SCTP-RF算法的甘肃省短期定量降水客观预报方法研究[J]. 干旱气象, 40(1): 146-155. DOI
[12]	龙柯吉, 杨康权, 康岚, 2024. 多模式对四川盆地强降水过程的预报性能检验[J]. 干旱气象, 42(3): 473-483. DOI
[13]	罗聪, 张华龙, 曾沁, 等, 2021. 多模式融合的广东网格定量降水预报方法的研发与评估[J]. 气象, 47(5):539-549.
[14]	青海省气象局, 2022. 短时临近降雨强度等级划分:DB63/T 2048—2022[S].
[15]	沈洁, 王镜, 李忠亮, 等, 2023. 2016—2020年青海省区域性大到暴雨分布特征及预报指标检验[J]. 农业灾害研究, 13(12): 188-191.
[16]	唐健, 代刊, 宗志平, 等, 2018. 主客观融合定量降水预报方法及平台实现[J]. 气象, 44(8):1 020-1 032.
[17]	田笑, 余文韬, 从靖, 等, 2022. 统计降尺度方法在天津小时降水和气温精细化预报中的应用[J]. 干旱气象, 40(1): 135-145. DOI
[18]	王丽芳, 漆梁波, 周伟, 等, 2021. 频率匹配技术在暴雨预报中的应用及改进分析[J]. 暴雨灾害, 40(4):352-361.
[19]	王叶红, 赵玉春, 2023. GRAPES-REPS对我国南方2017年初夏持续性降水预报的检验评估[J]. 干旱气象, 41(2): 328-340. DOI
[20]	危国飞, 刘会军, 吴启树, 等, 2020. 多模式降水分级最优化权重集成预报技术[J]. 应用气象学报, 31(6): 668-680.
[21]	吴启树, 韩美, 刘铭, 等, 2017. 基于评分最优化的模式降水预报订正算法对比[J]. 应用气象学报, 28(3):306-317.
[22]	肖丹, 胡超, 陈宁, 等, 2023. 基于频率匹配法和最优TS评分法的降水预报订正效果对比[J]. 气象与减灾研究, 46(4):287-295.
[23]	许晶, 程丛兰, 宋林烨, 等, 2025. 鄂尔多斯高分辨率降水融合预报模型及检验分析[J]. 沙漠与绿洲气象, 19(6):149-157.
[24]	闫慧, 赵桂香, 董春卿, 等, 2017. 集合预报方法在山西暴雨预报中的应用试验[J]. 干旱气象, 35(2):321-329. DOI
[25]	张娇, 王东勇, 郑淋淋, 等, 2021. ECMWF模式强降水预报偏差订正方法研究及应用[J]. 暴雨灾害, 40(4): 430-436.
[26]	智协飞, 赵忱, 2020. 基于集合成员订正的强降水多模式集成预报[J]. 应用气象学报, 31(3):303-314.
[27]	朱科旭, 管琴, 白爱娟, 2024. 青海省洪涝灾害时空分布和致灾雨量特征[J]. 沙漠与绿洲气象, 18(1):81-88.
[28]	朱文刚, 盛春岩, 范苏丹, 等, 2024. 基于前馈神经网络的多模式集成降水预报研究[J]. 干旱气象, 42(1): 117-128. DOI
[29]	GUDMUNDSSON L, BREMNES J B, HAUGEN J E, et al, 2012. Technical note: Downscaling RCM precipitation to the station scale using statistical transformations-a comparison of methods[J]. Hydrology and Earth System Sciences, 16(9):3 383-3 390. DOI URL
[30]	LIU Q, LOU X, YAN Z W, et al, 2023. Deep-learning post-processing of short-term station precipitation based on NWP forecasts[J]. Atmospheric Research, 295, 107032. DOI:10.1016/j.atmosres.2023.107032.
[31]	WANG Y, DAI K, ZONG Z, et al, 2021. Quantitative precipitation forecasting using multi-model blending with supplemental grid points: Experiments and prospects in China[J]. Journal of Meteorological Research, 35(3):521-536. DOI
[32]	ZHAO J, LIU Y, XU J C, et al, 2023. Multi-source precipitation data fusion analysis and application based on Bayesian-Three Cornered Hat method[J]. Advances in Water Science, 34(5): 685-696.