蒋薇,刘芸芸,陈鹏,张志薇. 2021. 利用深度神经网络和先兆信号的江苏夏季降水客观预测方法[J]. 气象学报, 79(6):1035-1048, doi:10.11676/qxxb2021.057
利用深度神经网络和先兆信号的江苏夏季降水客观预测方法
Prediction of summer precipitation in Jiangsu province based on precursory factors:A deep neural network approach
投稿时间:2021-01-15  修订日期:2021-06-28
DOI:10.11676/qxxb2021.057
中文关键词:  夏季降水  季节预测  先兆信号  深度神经网络  动态权重集合方案
英文关键词:Summer precipitation  Seasonal prediction  Precursory signals  Deep Neural Network (DNN)  Dynamic weight set scheme
基金项目:江苏省气象局科研项目重点项目(KZ202004)、江苏省气象局科研项目面上项目(KM202009)和中国气象局预报员专项(CMAYBY2020-164)
作者单位E-mail
蒋薇 江苏省气候中心南京210041  
刘芸芸 国家气候中心中国气象局气候研究开放实验室北京100081 liuyuny@cma.gov.cn 
陈鹏 江苏省气象信息中心南京210041  
张志薇 江苏省气象科学研究所南京210041  
摘要点击次数: 168
全文下载次数: 134
中文摘要:
      利用1961—2019年江苏省67个站降水量和气候指数数据集等资料,选取大气环流、海温和积雪等先兆信号的不同组合作为预测因子方案,通过对比不同机器学习方法对江苏省夏季降水开展预测试验。结果表明,深度神经网络(Deep Neural Network,DNN)较传统统计方法和其他机器学习方法有一定优势,深度神经网络结合动态权重集合因子方案对江苏省夏季降水的预测技巧最高,其独立样本检验结果稳定,2015—2019年的平均PS评分为76.0,距平符号一致率为0.62,距平相关系数达0.35,尤其对江苏省中南部的预测技巧更高,具有业务应用价值。不同预测因子方案对比分析表明,大气环流因子在江苏省夏季降水预测中做主要贡献,而海温因子和积雪等其他因子也有正贡献,说明使用综合性预测因子以及集合方案有助于提升季节预测准确率。
英文摘要:
      Based on precipitation data collected at 67 national stations in Jiangsu province and a series of climatic indices from 1961 to 2019, the prediction experiment on summer precipitation in Jiangsu province is carried out using different machine learning methods accompanied by five prediction schemes with different combinations of precursor signals, including atmospheric circulation, sea surface temperature and snow cover, etc. It is shown that the deep neural network (DNN) method has advantages over traditional statistical methods and other machine learning methods on the prediction of summer precipitation in Jiangsu province. The comparison of the prediction results of five different prediction schemes with the DNN method further indicates that the model of DNN mixed dynamic weight set scheme (DMDW) has the highest prediction skill for summer precipitation in Jiangsu province. The results of the independent sample test by DMDW are stable with the five-year average PS score of 76.0, the symbol consistency rate of 0.62, and the abnormality correlation coefficient (ACC) of 0.35. In the operational application, the model shows higher accuracy of precipitation forecast over central and southern Jiangsu province. Furthermore, the potential impacts of the precursor signals in the prediction factor schemes on the prediction accuracy of the summer precipitation in Jiangsu province are also investigated in this work. The atmospheric circulation factors play a major role in the summer precipitation prediction in Jiangsu province, while other factors such as SST and snow cover have positive contributions. Therefore, the DMDW model with the comprehensive precursory factors has the best prediction effect, which can effectively improve the accuracy of seasonal prediction of summer precipitation in Jiangsu province.
HTML   查看全文   查看/发表评论  下载PDF阅读器
分享按钮