基于深度学习的中国地面气温的多模式集成预报研究
投稿时间:2020-02-19  修订日期:2020-02-19  点此下载全文
引用本文:
摘要点击次数: 20
全文下载次数: 0
作者单位E-mail
智协飞 南京信息工程大学 zhi@nuist.edu.cn 
王田 南京信息工程大学  
季焱 南京信息工程大学  
基金项目:面向一体化保障的天气现实及发展态势产品生成技术,国家自然科学基金面上项目 (41575104)。
中文摘要:基于TIGGE资料中的欧洲中期天气预报中心(ECMWF)、英国气象局(UKMO)、美国国家环境预报中心(NCEP)、韩国气象厅(KMA)、和日本气象厅(JMA)5个中心2015年1月1日—9月30日中国及周边地区地面2m气温24~168 h集合预报资料,利用长短期记忆神经网络(LSTM)、浅层神经网络(NN)、滑动训练期消除偏差集合平均(BREM)和滑动训练期多模式超级集合(SUP)方法对2015年9月5日—30日26 d预报期进行集成预报试验。结果表明,BREM的预报结果易受预报效果较差的模式的不利影响,虽然预报效果整体优于大部分单个模式,但不如单个模式中预报技巧最高的ECMWF预报技巧高,特别是在新疆南部,误差比较大。超级集合的预报结果比所有单个模式的更为准确。在144 h之前,SUP的误差明显小于ECMWF,但随预报时效的增加,误差增长幅度增大。NN对地面气温的预报效果与SUP的预报效果相当。LSTM整体预报效果最好,特别是在预报时效较长(超过72 h)时,相比其他方法预报准确率明显提升。LSTM神经网络方法明显改进我国西北、华北、东北、西南和华南大部分地区的气温预报,但在南疆部分地区误差较大。
中文关键词:深度学习  人工智能;多模式集成预报;LSTM神经网络;地面气温
 
Multimodel Ensemble Forecasts of Surface Air Temperature over China Based on Deep Learning Approach
Abstract:Based on the 1-7 days ensemble forecasts of the European Centre for Medium-Range Weather Forecasts (ECMWF), US National Centers for Environmental Prediction (NCEP), and the Japan Meteorological Agency (JMA), the UK Met Office (UKMO) as well as the Korea Meteorological Administration (KMA) in the TIGGE datasets, the multimodel ensemble forecasts of the surface air temperature in China and its adjacent area during the period from 1 January to 30 September 2015 have been conducted by using long-term memory (LSTM) neural networks, neural networks (NN), bias-removed ensemble mean (BREM) and the superensemble (SUP) with sliding training period for the forecast period from September 5 to 30, 2015. The results showed that the BREM forecast is not better than the ECMWF forecast due to the impact of low skill model forecasts among five different models. The forecast skill of SUP is better than that of all single models. For 24h-144h forecasts, the root mean square error (RMSE) of SUP is significantly smaller than that of ECMWF forecast. As the forecast lead time increases, the RMSE increases as well. The forecast skill of NN is roughly equivalent to that of SUP. Overall, the LSTM approach has the best forecast performance, especially when the forecast lead time is longer than 72h, the RMSE of the LSTM forecast is considerably smaller than that of ECMWF, BREM, NN, and SUP forecast. The LSTM neural networks approach significantly reduced the forecast RMSE of the surface air temperature in the northwestern, northern, northeastern, southwestern, and southern China. However, the RMSE of the LSTM forecast is relatively large in part of southern Xinjiang compared with ECMWF forecast.
keywords:Deep learning  Artificial intelligence  Multimodel ensemble  LSTM  Surface air temperature.
  查看/发表评论  下载PDF阅读器