基于FFT与ConvLSTM的ERA5水汽预测

ERA5 PWV Prediction Based on FFT and ConvLSTM

ES评分 0

DOI 10.14075/j.jgg.2024.11.504
刊名
Journal of Geodesy and Geodynamics
年,卷(期) 2025, 45(10)
作者
作者单位

1 天津城建大学地质与测绘学院,天津市津静路26号, 300384;
2 天津城建大学经济与管理学院,天津市津静路26号, 300384

摘要
利用ERA5多要素数据集,综合FFT与ConvLSTM方法开展ERA5水汽预测。首先采用FFT提取各气象要素的公共周期;然后分别选择不同公共周期的多气象要素数据、单一水汽数据作为模型输入,使用ConvLSTM方法进行训练并构建ERA5水汽预测模型;最后以GNSS水汽、ERA5实测水汽分别开展ERA5点、面水汽的模型预测效果评价。结果表明,基于FFT与ConvLSTM的ERA5水汽预测方法可有效预测未来120 h的水汽,在单点和面空间范围内均有较高精度,可为短临降水预警提供数据基础。
Abstract
This paper utilizes the ERA5 multi-element dataset to predict ERA5 water vapor by integrating the fast Fourier transform(FFT) and convolutional long short-term memory(ConvLSTM) methods. Firstly, the common periods of various meteorological elements are extracted using FFT. Subsequently, multiple meteorological elements with different common periods and individual water vapor data are selected as model inputs, and the ConvLSTM method is employed to train the ERA5 water vapor and construct the prediction model. Finally, the prediction performance of the model for point and spatial water vapor is evaluated using GNSS water vapor and actual ERA5 water vapor. The results indicate that the ERA5 water vapor prediction method based on FFT and ConvLSTM can effectively predict water vapor for the next 120 hours, with high accuracy in both single-point and spatial areas, providing a data foundation for short-term precipitation warnings.
关键词
水汽; 快速傅里叶变换; 卷积长短时记忆网络; ERA5; GNSS
KeyWord
PWV; FFT; ConvLSTM; ERA5; GNSS
基金项目
页码 1033-1036
  • 参考文献
  • 相关文献
  • 引用本文

王勇, 孟祥顺, 杜玮, 刘严萍, 刘宇博. 基于FFT与ConvLSTM的ERA5水汽预测 [J]. 大地测量与地球动力学. 2025; 45; (10). 1033 - 1036.

  • 文献评论

相关学者

相关机构