陈锦鹏1, 冯业荣2, 蒙伟光2, 文秋实2, 潘宁3, 戴光丰2
A Correction Method of Hourly Precipitation Forecast Based on Convolutional Neural Network
Chen Jinpeng1, Feng Yerong2, Meng Weiguang2, Wen Qiushi2, Pan Ning3, Dai Guangfeng2
(1.Fujian Key Laboratory of Severe Weather/Institute of Meteorological Big Data-Digital Fujian/Zhangzhou Meteorological Bureau;2.Institute of Tropical and Marine Meteorology/Key Laboratory of Regional Numerical Weather Prediction;3.Fujian Key Laboratory of Severe Weather/Fujian Meteorological Observatory)
本文已被:浏览 97次   下载 11
投稿时间:2020-02-09    修订日期:2020-10-20
中文摘要: 随着数值预报分辨率的不断提高,针对逐时降水预报的订正方法研究也日益迫切。应用2017—2018年5—9月福建省观测资料对华南区域中尺度模式(GTRAMS-3km-RUC)预报进行站点检验,建立和训练基于卷积神经网络的逐时降水分级订正模型,并与频率匹配法进行2017—2018年测试集的对比试验和2019年数据集的模拟业务检验,探讨了试验过程中遇到的样本不均衡、特征变量选取以及模型过拟合问题。结果表明:(1)模式对于15mm·h-1 以上降水的预报能力弱,各订正方法对原始预报均有不同程度的改进作用。从评估指标来看,基于卷积神经网络的订正方法相比频率匹配法的提升幅度更加突出,其中相关系数判别方案下的网络模型对强降水预报的订正效果显著优于其他方法;(2)在输入特征变量选取方面,应用主成分分析方案的模型训练收敛速度比相关系数判别方案更快,最佳训练期有所提前,但也更早进入严重的过拟合状态,而相关系数判别方案能够使网络模型的训练拥有更长的提升期以达到更具“潜力”的状态;(3)基于卷积神经网络的订正方法对减少分类降水预报的漏报率、晴雨和弱降水预报的空报率具有显著作用,其优化程度明显超过频率匹配法。
Abstract:As higher resolution NWP models become operational, more effective correction methods for model’s hourly precipitation forecast are needed. In this article, precipitation forecasts from the South China regional hourly updated cycle mesoscale model (GTRAMS-3km-RUC) are verified against observations of Fujian province during May to September in 2017 and 2018. A correction approach on categorical hourly precipitation forecast using Convolutional Neural Network (CNN) is built and trained, taking into account of issues such as sample imbalance, selection of feature and over-fitting of the system. Comparisons of test between CNN and Frequency Matching Method (FM) on 2017-2018 test set and 2019 practical data set are also made for their correction effects. Results show that all these correction methods improve the original forecast in varying degrees, which underperforms with precipitation over 15mm·h-1 . CNN performs better than FM, and CNN with Correlation Coefficient Discrimination (CCD) is the most effective for heavy rainfall category. In addition, we apply two different solutions to extract input features of the neural network system. The CNN model converges faster when principal component analysis (PCA) solution is used, however it leads to over-fitting earlier and severely which implies that the system’s generalization ability needs to be improved further. In contrast, CCD solution displays its longer promotion period and more potential state. It seems that CNN correction method improves forecast ability mainly by reducing the missing ratio for categorical precipitaition forecasts and the false alarm ratio for the rainfall and light rain forecasts.
文章编号:202002090035     中图分类号:    文献标志码:
陈锦鹏,冯业荣,蒙伟光,文秋实,潘宁,戴光丰,0.[en_title][J].Meteor Mon,():-.
Chen Jinpeng,Feng Yerong,Meng Weiguang,Wen Qiushi,Pan Ning,Dai Guangfeng,0.A Correction Method of Hourly Precipitation Forecast Based on Convolutional Neural Network[J].Meteor Mon,():-.