基于多尺度特征融合的深度学习常德市格点气温预报 

展开
  • 1. 气象防灾减灾湖南省重点实验室,湖南 长沙 410118
    2. 湖南省常德市气象局,湖南 常德 415000
    3. 南京信息工程大学江苏省应用数学中心,江苏 南京 210044
    4. 湖南省气象台,湖南 长沙 410118
    5. 湖南省人工影响天气中心,湖南 长沙 410118

网络出版日期: 2025-05-08

基金资助

国家自然科学基金项目(42375017);中国气象局气象能力提升联合研究专项(23NLTSZ005);全国暴雨研究开放基金项目(BYKJ2024Z05);江苏省应用数学(南京信息工程大学)中心开放课题《基于多元深度学习的城市精细化气温预报研究》;湖南省气象局创新发展专项(CXFZ2024-FZZX24

Deep Learning Grid-point Temperature Forecasting in Changde City with Multi-scale Feature Integration 

Expand
  • 1. Hunan Key Laboratory of Meteorological Disaster Prevention and ReductionChangsha 410118HunanChina
    2. Changde Meteorological Bureau of Hunan ProvinceChangde 415000HunanChina
    3. Center for Applied Mathematics of Jiangsu ProvinceNanjing University of Information Science and TechnologyNanjing 210044JiangsuChina
    4. Hunan Meteorological ObservatoryChangsha 410118HunanChina
    5. Hunan Weather Modification CenterChangsha 410118HunanChina

Online published: 2025-05-08

摘要

本研究旨在通过融合高空和地面多尺度特征,提高城市格点气温预报的精度。基于2021-20244-9月常德地区的欧洲中期天气预报中心(ECMWF)模式预报数据和中国气象局高分辨率陆面数据同化系统(HRCLDAS)逐小时气温格点实况数据,采用高分辨率MU-NETMulti-Scale U-Net)模型,设计了三组试验来构建一个能够预测常德市未来 24 h每小时气温的深度学习模型。试验结果表明,融合了地面和高空特征的MU-NET模型在研究区域内展现了最佳的订正效果,其平均绝对误差(MAE)和均方根误差(RMSE)相较于 EC模式分别降低了 22%25%,并且在日变化中,MU-NET模型的 MAE值最低,尤其在日最高气温预报中表现突出,平均降低了0. 4 ℃。在空间尺度上,MU-NET模型在高海拔地区相较于 EC模式实现了 60%~80%的预报技巧评分提升,在平原和洞庭湖地区提升幅度最大。在 2024年的两次关键天气过程中,MU-NET模型由于融合了地面和高空特征,展现了稳定的预报性能,尤其在处理复杂天气现象时的适用性。本研究的发现为提高气温预报的准确性提供了新的视角,并为实际的气象预报业务提供了有价值的参考。

本文引用格式

黎 璐, 游枭雄, 陈静静, 胡振菊, 卢 姝, 李 琼 . 基于多尺度特征融合的深度学习常德市格点气温预报 [J]. 高原气象, 0 : 1 . DOI: 10.7522/j.issn.1000-0534.2025.00040

Abstract

This study aims to enhance the accuracy of urban grid temperature forecasts by integrating multi-scale features from both upper-air and surface levels. Utilizing forecast data from the European Centre for MediumRange Weather ForecastsECMWFmodel and hourly temperature grid observation data from the China Meteorological Administration's High-Resolution Land Data Assimilation SystemHRCLDASfor the Changde region from April to September 2021-2024a high-resolution Multi-Scale U-NetMU-NETmodel was employed. Three sets of experiments were designed to develop a deep learning model capable of predicting hourly temperatures in Changde City for the next 24 hours. The experimental results demonstrate that the MU-NET modelwhich integrates surface and upper-air featuresexhibits the best correction performance within the study area. The mean absolute errorMAEand root mean square errorRMSEof the MU-NET model were reduced by 22% and 25%respectivelycompared to the ECMWF model. Additionallythe MU-NET model achieved the lowest MAE values in diurnal variationsparticularly excelling in the prediction of daily maximum temperatureswith an average reduction of over 0. 4 ℃. On a spatial scalethe MU-NET model showed a 60%~80% improvement in forecast skill scores over the ECMWF model in high-altitude regionswith the most significant improvements observed in plain and Dongting Lake areas. During two critical weather events in 2024the MUNET model demonstrated stable forecast performance due to its integration of surface and upper-air featuresparticularly in handling complex weather phenomena. The findings of this study provide new insights for improving the accuracy of temperature forecasts and offer valuable references for practical meteorological forecasting.

文章导航

/