[Objective]Mastering the refined spatiotemporal distribution characteristics of rainfall is of great significance for im-proving the level of urban flood risk management.The quality of rainfall monitoring data in China in the past decade is good,with a dense network of stations and high precision of rainfall data,but the time series is relatively short.[Methods]In order to make more effective use of historical rainfall data,this study introduces manifold learning algorithm into the reconstruction of historical rainfall data.From high-resolution rainfall data in the past decade,spatiotemporal features of rainfall are extracted.Based on this feature,6 h interval rainfall spatial data is reconstructed into 1 h interval spatial data to meet the requirements of urban flood risk analysis.[Results]The result show that:The average error between the high value area of the reconstructed data and the meas-ured value is within 15％,and the low value area is within 20％,the error in the high value area of the data is reduced by 45％～85％,while the error in the low value area is reduced by about 10％～40％,compared to traditional method.[Conclusion]The reconstructed historical spatial rainfall data using manifold learning algorithms conforms to the spatiotemporal distribution charac-teristics of rainfall in various regions,which can improve the granularity of rainfall spatial data and achieve effective and reasona-ble extraction and summary of refined features of rainfall spatiotemporal distribution.
关键词
流形学习/机器学习/暴雨时空分布/特征提取/低分辨率重构/泸州/降水
Key words
manifold learning/machine learning/spatial-temporal distribution of rainstorm/feature extraction/low resolution da-ta reconstruction/Luzhou/precipitation
维普
万方数据