引入预报时效的中国近海海面风速预报深度学习订正方法-海洋预报

引入预报时效的中国近海海面风速预报深度学习订正方法

单位：1. 河海大学自然资源部海洋灾害预报技术重点实验室, 江苏南京 210024;
2. 河海大学海洋学院, 江苏南京 210024;
3. 北部湾大学海洋学院广西海洋环境灾害过程与生态保护技术重点实验室, 广西钦州 535011;
4. 国家海洋环境预报中心, 北京 100081

关键词：海面风风速预报时效深度学习数值模式订正

分类号：P732.1

出版年·卷·期（页码）：2025·42·第五期（1-10）

摘要：

评估了美国国家环境预报中心全球预报系统（GFS）模式和欧洲中期天气预报中心综合预报系统（IFS）模式在中国近海海面风速的预报表现，并基于深度神经网络（DNN）模型对GFS模式的海面风速预报进行订正。结果表明：订正前，GFS模式在中国近海预报时效120 h的海面风速预报均方根误差为3.3 m/s,IFS模式为2.5 m/s,IFS模式的预报性能优于GFS模式；DNN模型可有效降低GFS模式的负系统偏差，多要素联合订正方法具有较好的订正效果。为减小预报时效延长导致的误差累积，提出一种引入预报时效的订正方法。将预报时效作为输入要素的订正模型，相较其他订正模型预报误差进一步减小。与原始GFS模式相比，该订正模型系统偏差减少1.3 m/s，均方根误差减少1.1 m/s，相对误差减少7个百分点，离散指数减少3个百分点。在最高风速区间，引入预报时效的订正模型，其预报时效96 h和120 h的预报负偏差，较IFS模式分别减少1.0 m/s和0.7 m/s，性能优于IFS模式。

This study evaluates the performances of the Global Forecast System（GFS） of the National Centers for Environmental Prediction and the Integrated Forecasting System（IFS） of the European Centre for MediumRange Weather Forecasts in forecasting sea surface wind speed along the nearshore areas of China. Additionally,a deep neural network（DNN）-based correction method is applied to improve the GFS wind speed forecasts along the nearshore areas of China. Results show that, prior to correction, the root mean square error（RMSE） of the 120-hour sea surface wind speed forecasts is 3.3 m/s for the GFS model and 2.5 m/s for the IFS model, with the IFS model demonstrating better performance. The DNN model effectively reduces the negative systematic bias of the GFS forecasts, and the multi-variable correction approach achieves significant improvements. To mitigate error accumulation caused by increasing forecast lead time, this study proposes a correction method incorporating forecast lead time as an input variable. Compared to other correction models, the lead-time-aware correction model further reduces forecast errors. Compared to the original GFS model, this approach reduces systematic bias by 1.3 m/s, RMSE by 1.1 m/s, relative error by 7%, and scatter index by 3%. In the highest wind speed range, the lead-time-aware correction model reduces the negative bias of the 96-hour and 120-hour forecasts by 1.0 m/s and 0.7 m/s, respectively, outperforming the IFS model.

参考文献：

[1] BAUER P, THORPE A, BRUNET G. The quiet revolution of numerical weather prediction[J]. Nature, 2015, 525(7567):47-55.
[2] DEE D P, UPPALA S. Variational bias correction of satellite radiance data in the ERA-Interim reanalysis[J]. Quarterly Journal of the Royal Meteorological Society, 2009, 135(644):1830-1841.
[3] BI K F, XIE L X, ZHANG H H, et al. Accurate medium-range global weather forecasting with 3D neural networks[J]. Nature,2023, 619(7970):533-538.
[4] KURTH T, SUBRAMANIAN S, HARRINGTON P, et al. Four CastNet:accelerating global high-resolution weather forecasting using adaptive Fourier neural operators[C]//Proceedings of the Platform for Advanced Scientific Computing Conference. Davos:Association for Computing Machinery, 2023:13.
[5] CHEN L, ZHONG X H, LI H, et al. A machine learning model that outperforms conventional global subseasonal forecast models[J].Nature Communications, 2024, 15(1):6425.
[6] KALNAY E. Atmospheric modeling, data assimilation, and predictability[M]. Cambridge:Cambridge University Press, 2003.
[7] JOSE D M, VINCENT A M, DWARAKISH G S. Improving multiple model ensemble predictions of daily precipitation and temperature through machine learning techniques[J]. Scientific Reports, 2022, 12(1):4678.
[8] MCGOVERN A, ELMORE K L, GAGNE D J, et al. Using artificial intelligence to improve real-time decision-making for high-impact weather[J]. Bulletin of the American Meteorological Society, 2017, 98(10):2073-2090.
[9] 陈德花,刘铭,苏卫东,等. BP人工神经网络在MM5预报福建沿海大风中的释用[J].暴雨灾害, 2010, 29(3):263-267.CHEN D H, LIU M, SU W D, et al. Interpretation and application of BP artificial neural network in MM5 model forecasting gale for coastal regions of Fujian province[J]. Torrential Rain and Disasters, 2010, 29(3):263-267.
[10] 张颖超,肖寅,邓华.基于ELM的风电场短期风速订正技术研究[J].气象, 2016, 42(4):466-471.ZHANG Y C, XIAO Y, DENG H. Modification technology research of short-term wind speed in wind farm based on ELM method[J]. Meteorological Monthly, 2016, 42(4):466-471.
[11] 郝雨辰,杨勤胜,霍雪松,等.基于Resnet50的江苏近海海面风场预报订正方法研究[J].海洋预报, 2024, 41(4):57-65.HAO Y C, YANG Q S, HUO X S, et al. Correction method for surface wind forecasts in Jiangsu offshore area based on Resnet50[J]. Marine Forecasts, 2024, 41(4):57-65.
[12] 罗江珊,杨凡,毕玮,等. EC细网格10 m风场产品在青岛港区的预报检验和随机森林订正[J].海洋预报, 2024, 41(3):110-119.LUO J S, YANG F, BI W, et al. Verification and random forest correction of the EC fine-grid model 10 m wind in Qingdao port[J]. Marine Forecasts, 2024, 41(3):110-119.
[13] FU S, HUANG W Y, LUO J J, et al. Deep learning improves GFS sea surface wind field forecast accuracy in the northwest pacific ocean[J]. Journal of Geophysical Research:Atmospheres, 2024,129(13):e2024JD041188.
[14] 王国松,王喜冬,侯敏,等.基于观测和再分析数据的LSTM深度神经网络沿海风速预报应用研究[J].海洋学报, 2020, 42(1):67-77.WANG G S, WANG X D, HOU M, et al. Research on application of LSTM deep neural network on historical observation data and reanalysis data for sea surface wind speed forecasting[J]. Haiyang Xuebao, 2020, 42(1):67-77.
[15] 谢凯文,杨昊,邹茂扬,等.一种融合多要素时空特征的数值预报风场订正模型[J].软件导刊, 2024, 23(11):84-92.XIE K W, YANG H, ZOU M Y, et al. A model for correcting numerical wind field forecast with multifactor spatiotemporal features[J]. Software Guide, 2024, 23(11):84-92.
[16] KÖHL A. Evaluating the GECCO3 1948-2018 ocean synthesisa configuration for initializing the MPI-ESM climate model[J].Quarterly Journal of the Royal Meteorological Society, 2020, 146(730):2250-2273.
[17] LORENZ E N. Deterministic nonperiodic flow[J]. Journal of the Atmospheric Sciences, 1963, 20(2):130-141.
[18] DAVIES L A, BROWN A R. Assessment of which scales of orography can be credibly resolved in a numerical model[J].Quarterly Journal of the Royal Meteorological Society, 2001, 127(574):1225-1237.
[19] LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature,2015, 521(7553):436-444.
[20] HINTON G, DENG L, YU D, et al. Deep neural networks for acoustic modeling in speech recognition:the shared views of four research groups[J]. IEEE Signal Processing Magazine, 2012, 29(6):82-97.
[21] WANG J J, LI B X, GAO Z Y, et al. Comparison of ECMWF significant wave height forecasts in the China sea with buoy data[J]. Weather and Forecasting, 2019, 34(6):1693-1704.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】