测绘科学技术学报2025，Vol.41Issue(4)：362-371,10.DOI:10.3969/j.issn.1673-6338.2025.04.005
基于长短期记忆神经网络改进的PWV阈值降水预报方法研究
Method Research of PWV Threshold Rainfall Forecasting Based on the Improved LSTM Neural Networks
摘要
Abstract
To address the issue of the interruption of the time series of precipitable water vapor(PWV)caused by the loss of ground-based global navigation satellite system(GNSS)observation data,which affects the continuity and accuracy of threshold precipitation forecasting,this paper constructed a PWV forecasting model using long short-term memory(LSTM)neural networks and sliding time windows.Subsequently,an improved threshold rainfall forecasting model was established as well based on the integration of actual precipitation data and LSTM-PWV.The results show that the mean bias between LSTM-PWV and GNSS-PWV is-0.1 mm,with a root mean square error(RMSE)of 1.2 mm,and the correlation coefficient is 0.99.In comparison to the GNSS-PWV,the annual average correct rate(CR)and probability of detection(POD)of the LSTM PWV threshold precipitation forecasting model have increased by 3.0％and 3.6％respectively,the false alarm rate and missed alarm rate have decreased by 4.0％and 3.6％respectively.Among them,the CR and POD at P047 station reach 95.2％and 78.2％respectively.Therefore,LSTM-PWV can not only be used to fill in the interrupted PWV time series due to missing GNSS obser-vation data,but also has more outstanding forecasting performance in threshold precipitation forecasting.关键词
全球导航卫星系统/长短期记忆神经网络/大气可降水量/阈值/降水预报Key words
GNSS/long short-term memory neural networks/PWV/threshold/precipitation forecasting分类
测绘与仪器引用本文复制引用
李锴,李黎,马乙翔,张明松,申文瑜..基于长短期记忆神经网络改进的PWV阈值降水预报方法研究[J].测绘科学技术学报,2025,41(4):362-371,10.基金项目
国家自然科学基金项目(42204014) （42204014）
江苏省研究生实践创新项目(SJCX23_1718,SJCX24_1901) （SJCX23_1718,SJCX24_1901）
江苏省科技计划项目(BK20230660). （BK20230660）
