Spatial-temporal Analysis and Prediction of Precipitation Extremes:A Case Study in the Weihe River Basin,China

• 作者: QIU Dexun WU Changxue MU Xingmin ZHAO Guangju GAO Peng
• 来源期刊: 中国地理科学（英文版）
• 年，卷（期）: 2022,（2）
• 页码范围: 358-372
• 页数: 15页
• 语种: 英文

摘要

Extreme precipitation events bring considerable risks to the natural ecosystem and human life.Investigating the spatial-tem-poral characteristics of extreme precipitation and predicting it quantitatively are critical for the flood prevention and water resources planning and management.In this study,daily precipitation data(1957-2019)were collected from 24 meteorological stations in the Weihe River Basin(WRB),Northwest China and its surrounding areas.We first analyzed the spatial-temporal change of precipitation extremes in the WRB based on space-time cube(STC),and then predicted precipitation extremes using long short-term memory(LSTM)network,auto-regressive integrated moving average(ARIMA),and hybrid ensemble empirical mode decomposition(EEMD)-LSTM-ARIMA models.The precipitation extremes increased as the spatial variation from northwest to southeast of the WRB.There were two clusters for each extreme precipitation index,which were distributed in the northwestern and southeastern or northern and southern of the WRB.The precipitation extremes in the WRB present a strong clustering pattern.Spatially,the pattern of only high-high cluster and only low-low cluster were primarily located in lower reaches and upper reaches of the WRB,respectively.Hot spots(25.00％-50.00％)were more than cold spots(4.17％-25.00％)in the WRB.Cold spots were mainly concentrated in the northwestern part,while hot spots were mostly located in the eastern and southern parts.For different extreme precipitation indices,the performances of the different models were different.The accuracy ranking was EEMD-LSTM-ARIMA＞LSTM＞ARIMA in predicting simple daily intensity index(SDII)and consecutive wet days(CWD),while the accuracy ranking was LSTM＞EEMD-LSTM-ARIMA＞ARIMA in predicting very wet days(R95P).The hybrid EEMD-LSTM-ARIMA model proposed was generally superior to single models in the pre-diction of precipitation extremes.