基于机器学习的华北气温多模式集合预报的订正方法

门晓磊; 焦瑞莉; 王鼎; 赵晨光; 刘亚昆; 夏江江; 李昊辰; 严中伟; 孙建华; 王立志

doi:10.3878/j.issn.1006-9585.2018.18049

基于机器学习的华北气温多模式集合预报的订正方法

doi: 10.3878/j.issn.1006-9585.2018.18049

1.
北京信息科技大学信息与通信工程学院, 北京 100101
2.
北京信息科技大学信息管理学院, 北京 100192
3.
中国科学院大气物理研究所东亚区域气候-环境重点实验室, 北京 100029
4.
北京大学数学科学学院, 北京 100871
5.
中国科学院大学, 北京 100049

基金项目:

中国科学院战略性先导科技专项 A-XDA19030403

中国科学院战略性先导科技专项 XDA19040202

详细信息

作者简介:
门晓磊, 男, 1993年出生, 硕士研究生, 主要从事机器学习和深度学习方向的研究。E-mail:1160727887@qq.com

通讯作者:
夏江江, E-mail:xiajj@tea.ac.cn

中图分类号: P457.3
计量
- 文章访问数: 1609
- HTML全文浏览量: 79
- PDF下载量: 1455
- 被引次数: 0
出版历程
- 收稿日期: 2018-04-02
- 网络出版日期: 2018-07-04
- 刊出日期: 2019-01-20

A Temperature Correction Method for Multi-model Ensemble Forecast in North China Based on Machine Learning

1.
School of Information and Communication Engineering, Beijing Information Science and Technology University, Beijing 100101
2.
School of Information Management, Beijing Information Science and Technology University, Beijing 100192
3.
Key Laboratory of Regional Climate-Environment for Temperate East Asia, Institute of Atmospheric Physics, Chinese Academy of Sciences, Beijing 100029
4.
School of Mathematical Sciences, Peking University, Beijing 100871
5.
University of Chinese Academy of Sciences, Beijing 100049

Funds:

Strategic Priority Research Program of Chinese Academy of Sciences A-XDA19030403

Strategic Priority Research Program of Chinese Academy of Sciences XDA19040202

摘要

摘要: 模式预报的订正是决定局地天气预报结果的一个重要步骤，基于机器学习的后处理模型近年来开始崭露头角。本文发展了基于岭回归（Ridge）、随机森林（Random Forest，RF）和深度学习（Deep Learning，DL）的3种后处理模型，基于中国气象局（CMA）的BABJ模式、欧洲中期天气预报中心（ECMWF）的ECMF模式、日本气象厅（JMA）的RJTD模式和NCEP的KWBC模式这4个数值天气预报模式2014年2月至2016年9月（训练期）近地面2 m气温预报和实况资料确定各模型参数，进而对2016年10月至2017年9月（预报期）华北地区（38°N~43°N，113°E~119°E）的逐日地面2 m气温预报进行了多模式集合预报分析。采用均方根误差对预报效果进行评估，这3种后处理模型的预报效果和4个数值天气预报模式以及通常的多模式集合平均（Ensemble Mean，EMN）的预报效果的对比表明：1）随着预报时长增加，4个数值预报模式及各种后处理模型的均方根误差均呈上升趋势；但区域平均而言，Ridge、RF和DL的预报效果在任何预报时长上都明显优于EMN和单个天气预报模式；特别是前几天的短期预报DL的预报效果更好，中后期预报Ridge的预报效果略好。2）华北地区的东南部均方根误差较小，其余格点上均方根误差较高，从空间分布而言，DL的订正预报效果最好，3种机器学习模型的误差在1.24~1.26℃之间，而EMN的误差达1.69℃。3）夏季各种方法的预报效果都较好，冬季预报效果都较差；但是Ridge、RF和DL的预报效果明显优于EMN，这3种模型预报的平均均方根误差在2.15~2.18℃之间，而EMN的平均均方根误差达2.45℃。
- 地面2 m气温 /
- 多模式集合平均 /
- 岭回归 /
- 随机森林 /
- 深度学习
Abstract: Post-forecast data processing is critical for obtaining reliable local weather forecast. In this study, the authors developed three post-processing models based on ridge regression (Ridge), random forest (RF), and deep learning (DL) methods. The post-processing models were trained by observational and forecast data of daily 2-m above surface air temperature in North China (38°N-43°N, 113°E-119°E) from four numerical weather forecast (NWF) models (BABJ model from China Meteorological Administration, ECMF model from ECMWF, RJTD model from Japan Meteorological Agency, and KWBC model from NCEP, respectively), for the training period from February 2014 to September 2016, and then applied to the forecast period from October 2016 to September 2017. The forecast results of the post-processing models together with those of commonly-used multi-model ensemble mean (EMN) and individual NWF models were evaluated according to the root-mean-square error (RMSE). The main results are as follows:1) For the region as a whole, with the increase in the forecast lead time, all the NWF models, EMN and the post-processing models exhibit increasing RMSEs, but the RMSEs of the three post-processing models are all significantly smaller than those of EMN and individual NWF models; especially, DL is slightly better for the short-term (the first few days) forecast and RF is slightly better for the longer-term prediction. 2) The RMSEs are relatively smaller in the southeastern part of North China, approximately in the range of (38°N-41°N, 115.5°E-119°E) than else where; on average, DL is slightly better, and the RMSEs of the three machine learning models are between 1.24℃ and 1.26℃, while the EMN error is 1.69℃. 3) There are seasonal differences:The results of all the models are relatively good for the summer, but poor in general for the winter. All the three post-processing models perform better than EMN and individual NWF models, with a smallest average RMSE of 2.15℃ for Ridge compared with 2.45℃ for EMN.
- 2-m above surface air temperature /
- Multi-model ensemble mean forecast /
- Ridge regression /
- Random forest /
- Deep learning

HTML全文

图 1 深层神经网络构建的多模式集合温度预报模型

Figure 1. Multi-model ensemble temperature prediction model constructed by deep neural network

下载: 全尺寸图片幻灯片

图 2 4个模式以及4种后处理模型的均方根误差（RMSE）随预报时长的变化

Figure 2. Root-mean-square errors (RMSEs) in four models and four post-processing models as a function of forecast lead time

下载: 全尺寸图片幻灯片

图 3 4种预报订正处理模型的均方根误差空间分布

Figure 3. RMSEs spatial distributions in four post-predictive processing models

下载: 全尺寸图片幻灯片

图 4 4种预报订正处理模型的均方根误差的时间序列（2016年10月至2017年9月）

Figure 4. Time series of RMSEs for four forecast correction processing models (Oct 2016−Sep 2017)

下载: 全尺寸图片幻灯片

图 5 4个数值预报模式的均方根误差的时间序列（2016年10月至2017年9月）

Figure 5. Time series of RMSEs for four numerical weather prediction models (Oct 2016−Sep 2017)

下载: 全尺寸图片幻灯片

表 1 4个模式以及4种后处理模型的均方根误差随预报时长的线性趋势

Table 1. Trends of RMSEs in four models and four postrocessing models with the forecast lead time

℃ d⁻¹
模式或模型名称	线性趋势
BABJ	0.22
ECMF	0.29
RJTD	0.22
KWBC	0.27
EMN	0.20
Ridge	0.23
RF	0.23
DL	0.23

下载: 导出CSV

参考文献(21)

[1]	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. doi: 10.3969/j.issn.1004-9045.2015.01.009 Chen Boyu, Dai Kan, Guo Yunqian. 2015. Precipitation verification and analysis of ECMWF ensemble statistic products in 2013 flooding season[J]. Torrential Rain and Disasters (in Chinese), 34 (1):64-73, doi:10.3969/j.issn.1004-9045. 2015.01.009.
[2]	范苏丹, 盛春岩, 肖明静, 等. 2015.多模式集合对山东省气象要素预报效果检验[J].气象与环境学报, 31 (6):68-77. doi: 10.3969/j.issn.1673-503X.2015.06.009 Fan Sudan, Sheng Chunyan, Xiao Mingjing, et al. 2015. Forecast effect verification of multi-model ensemble for meteorological elements in Shandong Province[J]. Journal of Meteorology and Environment (in Chinese), 31 (6):68-77, doi: 10.3969/j.issn.1673-503X.2015.06.009.
[3]	冯慧敏, 智协飞, 崔慧慧, 等. 2016.基于多模式集成技术的地面气温精细化预报[J].气象与环境科学, 39 (4):73-79. doi: 10.16765/j.cnki.1673-7148.2016.04.012 Feng Huimin, Zhi Xiefei, Cui Huihui, et al. 2016. Refined forecasting of surface temperature based on multi-model ensemble technology[J]. Meteorological and Environmental Sciences (in Chinese), 39 (4):73-79, doi: 10.16765/j.cnki.1673-7148.2016.04.012.
[4]	Hernández E, Sanchez-Anguix V, Julian V, et al. 2016. Rainfall prediction: A deep learning approach[C]//Proceedings of the 11th International Conference on Hybrid Artificial Intelligence Systems. Seville: Springer, 151-162, doi: 10.1007/978-3-319-32034-2_13.
[5]	Hinton G E, Osindero S, Teh Y W. 2006. A fast learning algorithm for deep belief nets[J]. Neural Computation, 18 (7):1527-1554, doi: 10.1162/neco.2006.18.7.1527.
[6]	黄威, 牛若芸. 2017.基于集合预报和支持向量机的中期强降雨集成预报试验[J].气象, 43 (9):1110-1116. doi: 10.7519/j.issn.1000-0526.2017.09.008 Huang Wei, Niu Ruoyun. 2017. The medium-term multi-model integration forecast experimentation for heavy rain based on support vector machine[J]. Meteorological Monthly (in Chinese), 43 (9):1110-1116, doi:10.7519/j.issn.1000-0526.2017. 09.008.
[7]	焦李成, 杨淑媛, 刘芳, 等. 2016.神经网络七十年:回顾与展望[J].计算机学报, 39 (8):1697-1716. http://d.old.wanfangdata.com.cn/Periodical/jsjxb201608015 Jiao Licheng, Yang Shuyuan, Liu Fang, et al. 2016. Seventy years beyond neural networks:Retrospect and prospect[J]. Chinese Journal of Computers (in Chinese), 39 (8):1697-1716. http://d.old.wanfangdata.com.cn/Periodical/jsjxb201608015
[8]	Krishnamurti T N, Kishtawal C M, LaRow T E, et al. 1999. Improved weather and seasonal climate forecasts from multimodel super ensemble[J]. Science, 285 (5433):1548-1550, doi: 10.1126/science.285.5433.1548.
[9]	李刚, 谢清霞, 魏涛. 2016.集合预报在贵州最低气温中的应用[J].安徽农业科学, 44 (14):229-231. doi: 10.3969/j.issn.0517-6611.2016.14.078 Li Gang, Xie Qingxia, Wei Tao. 2016. Application of multi-model ensemble method for minimum temperature in Guizhou Province[J]. Journal of Anhui Agricultural Sciences (in Chinese), 44 (14):229-231, doi: 10.3969/j.issn.0517-6611.2016.14.078.
[10]	李丽辉, 朱建生, 强丽霞, 等. 2017.基于随机森林回归算法的高速铁路短期客流预测研究[J].铁道运输与经济, 39(9):12-16. doi: 10.16668/j.cnki.issn.1003-1421.2017.09.03 Li Lihui, Zhu Jiansheng, Qiang Lixia, et al. 2017. Study on forecast of high-speed railway short-term passenger flow based on random forest regression[J]. Railway Transport and Economy (in Chinese), 39 (9):12-16, doi:10. 16668/j.cnki.issn.1003-1421.2017.09.03.
[11]	马清. 2008.中尺度集合预报的偏差订正与多模式集成研究[D].南京信息工程大学硕士学位论文, 78pp. http://cdmd.cnki.com.cn/Article/CDMD-10300-2008092054.htm Ma Qing. 2008. Study of the bias-correction and multi-model combine of mesoscale ensemble forecast[D]. M. S. thesis (in Chinese), Nanjing University of Information Science and Technology, 78pp. http://cdmd.cnki.com.cn/Article/CDMD-10300-2008092054.htm
[12]	牛金龙, 张东方, 姚鹏, 等. 2016.多模式资料在成都地区的温度预报研究应用[J].高原山地气象研究, 36 (3):66-70, 75. http://d.old.wanfangdata.com.cn/Periodical/scqx201603011 Niu Jinlong, Zhang Dongfang, Yao Peng, et al. 2016. Application study of multi-mode data in the forecast of temperaturein Chengdu[J]. Plateau and Mountain Meteorology Research (in Chinese), 36 (3):66-70, 75. http://d.old.wanfangdata.com.cn/Periodical/scqx201603011
[13]	潘留杰, 张宏芳, 朱伟军, 等. 2013. ECMWF模式对东北半球气象要素场预报能力的检验[J].气候与环境研究, 18 (1):111-123. doi: 10.3878/j.issn.1006-9585.2012.11097 Pan Liujie, Zhang Hongfang, Zhu Weijun, et al. 2013. Forecast performance verification of the ECMWF model over the Northeast Hemisphere[J]. Climatic andEnvironmental Research (in Chinese), 18 (1):111-123, doi: 10.3878/j.issn.1006-9585.2012.11097.
[14]	Wang H Z, Li G Q, Wang G B, et al. 2017. Deep learning based ensemble approach for probabilistic wind power forecasting[J]. Applied Energy, 188:56-70, doi: 10.1016/j.apenergy.2016.11.111.
[15]	王奕森, 夏树涛. 2018.集成学习之随机森林算法综述[J].信息通信技术, 12 (1):49-55. doi: 10.3969/j.issn.1674-1285.2018.01.009 Wang Yisen, Xia Shutao. 2018. A survey of random forests algorithms[J]. Information and Communications Technologies (in Chinese), 12 (1):49-55, doi: 10.3969/j.issn.1674-1285.2018.01.009.
[16]	邢彩盈, 张京红, 黄海静. 2016.基于BP神经网络的海口住宅室内气温预测[J].贵州气象, 40 (5):38-42. doi: 10.3969/j.issn.1003-6598.2016.05.007 Xing Caiying, Zhang Jinghong, Huang Haijing. 2016. Forecast of residential indoor temperature based on BP neural network in Haikou[J]. Journal of Guizhou Meteorology (in Chinese), 40 (5):38-42, doi: 10.3969/j.issn.1003-6598.2016.05.007.
[17]	熊国经, 董玉竹, 宗瑾. 2017.基于岭回归法对"三废"排放影响因素的研究——以江西省为例[J].生态经济, 33 (2):103-107. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=671176711 Xiong Guojing, Dong Yuzhu, Zong Jin. 2017. Research on the discharge of "Three Wastes" factors based on ridge regression:Taking Jiangxi Province as an example[J]. Ecological Economy (in Chinese), 33 (2):103-107. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=671176711
[18]	叶笃正, 严中伟, 戴新刚, 等. 2006.未来的天气气候预测体系[J].气象, 32 (4):3-8. doi: 10.3969/j.issn.1000-0526.2006.04.001 Ye Duzheng, Yan Zhongwei, Dai Xingang, et al. 2006. A discussion of future system of weather and climate prediction[J]. Meteorological Monthly (in Chinese), 32 (4):3-8, doi:10.3969/j.issn. 1000-0526.2006.04.001.
[19]	张恒德, 张庭玉, 李涛, 等. 2018.基于BP神经网络的污染物浓度多模式集成预报[J].中国环境科学, 38 (4):1243-1256. doi: 10.19674/j.cnki.issn1000-6923.2018.0147 Zhang Hengde, Zhang Tingyu, Li Tao, et al. 2018. Forecast of air quality pollutants' concentrations based on BP neural network multi-model ensemble method[J]. China Environmental Science (in Chinese), 38 (4):1243-1256, doi: 10.19674/j.cnki.issn1000-6923.2018.0147.
[20]	张禄, 杨志军. 2016.基于神经网络和主分量的日极值气温预测方法[C]//第33届中国气象学会年会S20气象信息化——业务实践与技术应用.西安: 中国气象学会, 5pp. Zhang Lu, Yang Zhijun. 2016. BP neural network prediction model of the extreme temperature based on principal component analysis[C]//33rd Annual Meeting of China Meteorological Society. Xi'an: Chinese Meteorological Society, 5pp.
[21]	张伟, 王自发, 安俊岭, 等. 2010.利用BP神经网络提高奥运会空气质量实时预报系统预报效果[J].气候与环境研究, 15 (5):595-601. doi: 10.3878/j.issn.1006-9585.2010.05.08 Zhang Wei, Wang Zifa, An Junling, et al. 2010. Update the ensemble air quality modeling system with BP model during Beijing Olympics[J]. Climatic and Environmental Research (in Chinese), 15 (5):595-601, doi: 10.3878/j.issn.1006-9585.2010.05.08.