1.珠海市公共气象服务中心,广东 珠海 519082
2.珠海市气象局,广东 珠海 519082
3.珠海市国家气候观象台,广东 珠海 519082
蓝玉峰(1990年生),女;研究方向:大气科学;E-mail: lyfnuist@163.com
孙磊(1992年生),男;研究方向:大气科学;E-mail: sunleimodel@gmail.com
收稿:2026-04-03,
修回:2026-09-01,
录用:2026-09-08,
网络首发:2026-11-02,
移动端阅览
蓝玉峰, 郑浩阳, 张生奇, 等. 粤港澳大湾区短时降水CNN概率降尺度[J/OL]. 中山大学学报(自然科学版)(中英文), 2026,1-11.
Lan Yufeng, Zheng Haoyang, Zhang Shengqi, et al. Probabilistic downscaling of short-range precipitation forecasting using CNN in the Guangdong-Hong Kong-Macao Greater Bay Area[J/OL]. Acta Scientiarum Naturalium Universitatis Sunyatseni, 2026, 1-11.
蓝玉峰, 郑浩阳, 张生奇, 等. 粤港澳大湾区短时降水CNN概率降尺度[J/OL]. 中山大学学报(自然科学版)(中英文), 2026,1-11. DOI: 10.11714/acta.snus.ZR20260082.
Lan Yufeng, Zheng Haoyang, Zhang Shengqi, et al. Probabilistic downscaling of short-range precipitation forecasting using CNN in the Guangdong-Hong Kong-Macao Greater Bay Area[J/OL]. Acta Scientiarum Naturalium Universitatis Sunyatseni, 2026, 1-11. DOI: 10.11714/acta.snus.ZR20260082.
精细尺度短时降水预报是粤港澳地区防灾减灾的重要支撑。本研究利用卷积神经网络(CNN,convolutional neural network)构建短时降水概率降尺度预报模型,该模型以GFS(global forecast system)模式气象背景场预报资料作为输入信息,获得区域内各站点3 h降水预报的概率分布参数,并评价不同试验的预报性能。结果表明:1)确定性预报结果中,各试验均不同程度地降低了GFS原始预报误差,CNN-PR试验进一步强化了空间特征,预报效果最好(RMSE=3.31 mm);SINGLE试验(未引入全局信息)和CNN0试验(未进行空间特征提取)表现较差;2)概率预报结果中,各试验总体校准性能较好,CONTROL试验的综合评分最优(CRPS=0.69 mm);除SINGLE和CNN0试验外,其余试验对降水发生事件均有一定的区分能力,其中纳入时序演变信息的TREND试验对强降水的区分能力最优;CNN-PR试验的区间条件校准性最优,但以牺牲预报锐度为代价;3)各试验对强降水的预报结果偏小,概率预报水平受“整体-区间校准”“校准-锐度”2组内在矛盾的限制。本研究可为粤港澳大湾区短时降水精细化概率预报的业务化应用提供方法参考。
High-resolution short-range precipitation forecasting is critical for regional disaster prevention and mitigation over the Guangdong-Hong Kong-Macao Greater Bay Area (GBA). This study developed a probabilistic downscaling model for short-range precipitation using a Convolutional Neural Network (CNN), which takes large-scale meteorological forecast information from GFS (global forecast system)as input and outputs the parameters of the probabilistic distribution for 3 h precipitation at multiple stations. The model skill was evaluated through a set of carefully designed experiments. The results show that: 1) All experiments effectively reduce forecasting errors compared with the raw GFS outputs. Among them, the CNN-PR experiment, which incorporates enhanced spatial feature extraction, achieves the best deterministic performance (RMSE=3.31 mm). In contrast, the SINGLE experiment, which relies solely on local information, and the CNN0 experiment, which omits spatial feature extraction, show inferior performance. 2) In terms of probabilistic skill, all models are well calibrated overall, with the CONTROL experiment achieving the optimal skill (CRPS=0.69 mm). All experiments except SINGLE and CNN0 show improved discrimination ability for precipitation occurrence, and the TREND experiment incorporating temporal evolution information is optimal for discriminating heavy rainfall events. The CNN-PR experiment is superior in conditional calibration, albeit at the cost of reduced sharpness. 3) The study also reveals that all experiments systematically underestimate heavy rainfall events. Furthermore, the probabilistic forecasting skill is constrained by two key trade-offs: one between calibration and sharpness, and the other between overall and conditional calibration. This study provides a methodological reference for operational high-resolution short-range probabilistic precipitation forecasting in the GBA.
董晓云 , 余锦华 , 梁信忠 , 等 , 2020 . CWRF模式极端降水模拟误差订正 [J]. 应用气象学报 , 31 ( 4 ): 504 - 512 .
范丽军 , 符淙斌 , 陈德亮 , 2005 . 统计降尺度法对未来区域气候变化情景预估的研究进展 [J]. 地球科学进展 , 3 : 320 - 329 .
李广宗 , 孙赫敏 , 苏布达 , 等 , 2025 . 珠江流域不同历时极端降水雨型时空分布特征研究 [J]. 水利水电技术(中英文) , 56 ( 10 ): 72 - 83 .
李丽平 , 章开美 , 王超 , 等 , 2010 . 近40年华南前汛期极端降水时空演变特征 [J]. 气候与环境研究 , 15 ( 4 ): 443 - 450 .
李卓凡 , 王东海 , 曾智琳 , 2023 . 粤港澳大湾区一次局地极端强降水过程的多尺度观测特征与演变分析 [J]. 热带气象学报 , 39 ( 6 ): 915 - 928 .
孙健 , 曹卓 , 李恒 , 等 , 2021 . 人工智能技术在数值天气预报中的应用 [J]. 应用气象学报 , 32 ( 1 ): 1 - 11 .
孙磊 , 蓝玉峰 , 梁秀姬 , 等 , 2024 . 基于图卷积记忆网络对珠海臭氧时空预测 [J]. 中山大学学报(自然科学版)(中英文) , 63 ( 3 ): 48 - 59 .
王蕾 , 陈起英 , 徐国强 , 等 , 2024 . CMA-GFS对一次强降水过程预报评估及诊断改进 [J]. 中山大学学报(自然科学版)(中英文) , 63 ( 2 ): 46 - 58 .
王垚 , 史海匀 , 2024 . 粤港澳大湾区极端降雨的时空变化及其驱动因素探究 [J]. 水资源保护 , 40 ( 6 ): 28 - 37,77 .
杨绚 , 代刊 , 朱跃建 , 2022 . 深度学习技术在智能网格天气预报中的应用进展与挑战 [J]. 气象学报 , ( 5 ): 649 - 667 .
张博 , 张芳华 , 李晓兰 , 等 , 2024 . “ 23·7”华北特大暴雨数值预报检验评估 [J]. 应用气象学报 , 35 ( 1 ): 17 - 32 .
张进 , 孙健 , 沈学顺 , 等 , 2023 . CMA-GFS V 4.0 模式关键技术研发和业务化 [J]. 应用气象学报 , 34 ( 5 ): 513 - 526 .
张欣 , 刘凑华 , 戴建华 , 等 , 2026 . 长三角地区不同环流型下强降水模式预报性能评估 [J]. 气象 , 52 ( 5 ): 580 - 594 .
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 .
Baño-Medina J , Manzanas R , Gutiérrez J M , 2020 . Configuration and intercomparison of deep learning neural models for statistical downscaling [J]. Geosci Model Dev , 13 ( 4 ): 2109 - 2124 .
Cannon A J , 2008 . Probabilistic multisite precipitation downscaling by an expanded bernoulli-gamma density network [J]. J Hydrometeorol , 9 ( 6 ): 1284 - 1300 .
Deng Z , Wang Z , Wu X , et al , 2023 . Effect difference of climate change and urbanization on extreme precipitation over the Guangdong-Hong Kong-Macao Greater Bay Area [J]. Atmos Res , 282 : 106514 .
Dong F , Zhi X , Zhu S , et al , 2023 . Principal modes of diurnal cycle of rainfall over South China during the presummer rainy season [J]. J Clim , 36 ( 8 ): 2457 - 2470 .
Gneiting T , Balabdaoui F , Raftery A E , 2007 . Probabilistic forecasts, calibration and sharpness [J]. J R Stat Soc Ser B Stat Methodol , 69 ( 2 ): 243 - 268 .
Li Z , Chan J C L , Zhao K , et al , 2021 . Impacts of urban expansion on the diurnal variations of summer monsoon precipitation over the South China coast [J]. J Geophys Res Atmos , 126 ( 22 ): e2021JD035318 .
Liang X Z , Sun C , Zheng X , et al , 2019 . CWRF performance at downscaling China climate characteristics [J]. Clim Dyn , 52 ( 3 ): 2159 - 2184 .
Liang X Z , Xu M , Yuan X , et al , 2012 . Regional climate-weather research and forecasting model [J]. Bull Am Meteor Soc , 93 ( 9 ): 1363 - 1387 .
Reichstein M , Camps-Valls G , Stevens B , et al , 2019 . Deep learning and process understanding for data-driven Earth system science [J]. Nature , 566 ( 7743 ): 195 - 204 .
Sloughter J M L , Raftery A E , Gneiting T , et al , 2007 . Probabilistic quantitative precipitation forecasting using Bayesian model averaging [J]. Mon Weather Rev , 135 ( 9 ): 3209 - 3220 .
Sun L , Lan Y , 2021 . Statistical downscaling of daily temperature and precipitation over China using deep learning neural models: Localization and comparison with other methods [J]. Int J Climatol , 41 ( 2 ): 1128 - 1147 .
Sun L , Lan Y , Jiang R , 2023 . Using CNN framework to improve multi-GCM ensemble predictions of monthly precipitation at local areas: An application over China and comparison with other methods [J]. J Hydrol , 623 : 129866 .
Sun L , Lan Y , Sun X , et al , 2024 . Deterministic forecasting and probabilistic post-processing of short-term wind speed using statistical methods [J]. J Geophys Res Atmos , 129 ( 7 ): e2023JD040134 .
Vrac M , Naveau P , 2007 . Stochastic downscaling of precipitation: From dry events to heavy rainfalls [J]. Water Resour Res , 43 ( 7 ): 2006WR005308 .
Wang Y , Han Z , Gao R , 2021 . Changes of extreme high temperature and heavy precipitation in the Guangdong-Hong Kong-Macao Greater Bay Area [J]. Geomatics, Natural Hazards and Risk , 12 ( 1 ): 1101 - 1126 .
Wilks D S , 2011 . Statistical methods in the atmospheric sciences [M]. 3rd ed . Oxford : Academic press:257- 267 .
Xue Y , Janjic Z , Dudhia J , et al , 2014 . A review on regional dynamical downscaling in intraseasonal to seasonal simulation/prediction and major factors that affect downscaling ability [J]. Atmos Res , 147 / 148 : 68 - 85 .
Yu R , Zhou T , Xiong A , et al , 2007 . Diurnal variations of summer precipitation over contiguous China [J]. Geophys Res Lett , 34 ( 1 ): L01704 .
0
浏览量
0
下载量
0
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621
