杨璐. 2021. 基于三种机器学习方法的降水相态高分辨率格点预报模型的构建及对比分析[J]. 气象学报, (0):-, doi:10.11676/qxxb2021.059
基于三种机器学习方法的降水相态高分辨率格点预报模型的构建及对比分析
The Construction and Comparison of High Resolution Prediction type Prediction Models based on three machine learning methods
投稿时间:2020-11-24  修订日期:2021-05-07
DOI:10.11676/qxxb2021.059
中文关键词:  降水相态  客观预报 数值模式  气候统计
英文关键词:Precipitation  phase, objective  forecast, numerical  model, climate  statistics
基金项目:国家重点研发计划课题;国家重点研发计划课题
作者单位邮编
杨璐 北京城市气象研究院 100089
摘要点击次数: 87
全文下载次数: 127
中文摘要:
      [目的]冬季降水相态及其转变时间的精细化客观预报,对提高气象预报和服务质量具有重要现实意义。[资料和方法]本文基于京津冀地区1955-2019年国家级气象站的温度和天气现象资料,以及快速更新多尺度分析和预报系统(睿图/RMAPS)集成子系统(RMAPS-IN)高分辨率集成预报产品,分别利用提升树XGBoost、支持向量机SVM和深度神经网络DNN三种机器学习方法,建立了京津冀复杂地形下降水相态高分辨率网格预报模型。首先统计分析了京津冀地区174个国家级观测站点各类降水相态及其对应的气温、湿球温度平均气候概率的空间分布差异,以及不同降水相态时RMAPS-IN提供的网格化快速更新精细集成产品中7个可能影响降水相态判断的因子信息,包括地面2-m气温、露点温度、相对湿度、雪线高度、雪混合比占雨和雪混合比的比例,以及气温和湿球温度三维气象要素客观分析等;然后将地面观测天气现象资料、复杂地形下降水相态气候特征及高分辨率模式输出产品作为特征向量,[结果]分别基于XGBoost、SVM、DNN建立了降水相态分类模型,并对同样条件下三种机器学习方法对降雨、雨夹雪和雪3种京津冀主要降水相态的预报效果进行了对比检验。[结论]研究结果表明:1)构建的特征参数中增加复杂地形下降水相态气候特征,可以明显提升三种机器学习方法对于雨、雨夹雪和雪的预测准确率;2)三种机器学习方法对雨和雪的预报能力都明显优于雨夹雪;3)XGBoost和DNN的预报能力相当,都明显优于SVM。研究结果为基于机器学习方法开展降水相态的高精度格点预报提供了借鉴。
英文摘要:
      Refined and objective prediction of precipitation type and its transition time in winter is of great practical significance for improving the quality of forecast service. This paper establishes the high-resolution precipitation type prediction models based on the temperature and weather phenomena data at 174 national automatic weather stations in 1955-2019 in Beijing-Tianjin-Hebei and the high-resolution integrated forecast product of rapid update multi-scale analysis and forecast system integrated subsystem (RMAPS-IN) by using three machine learning methods, i.e. XGBboost, support vector machine and depth neural network prediction model. Firstly, the spatial distribution differences of various precipitation types and their corresponding average climate probabilities of air temperature and wet bulb temperature at 174 national stations in Beijing-Tianjin-Hebei region are statistically analyzed. The fine integrated products provided by RMAPS-IN, i.e. 2m air temperature, dew point temperature, relative humidity, snowline height, and the mixture ratio of frozen part precipitation in the near surface atmosphere under different precipitation types and the analysis field of three-dimensional meteorological elements such as temperature and wet bulb temperature are analyzed. Then, the observational weather phenomenon, precipitation type climate characteristics under complex terrain and high-resolution model output products are taken as feature vectors, the classification model of precipitation type is established respectively based on XGBboost, SVM and DNN, and the prediction effect of three machine learning algorithms on rain, sleet and snow is compared and tested. The results show that: 1) the accuracy of three machine learning methods for rain, sleet and snow prediction can be significantly improved by adding climate features of precipitation type under complex terrain to the feature parameters. 2) The prediction ability of three machine learning methods for rain and snow is better than that of sleet. 3) XGBboost and DNN have the same prediction ability, which are obviously better than SVM.
查看全文   查看/发表评论  下载PDF阅读器
分享按钮