Fusion Calibration and Comprehensive Evaluation Precipitation Estimation by Multi-source Precipitation Products Based on Deep Learning in Ungauged Region
The study of satellite ground station precipitation fusion correction is of great significance for the hydrolog-ical application of remote sensing precipitation data in ungauged region.This article constructs a deep neural network(CNN-LSTM)fusion model that considers spatiotemporal factors.Combining terrain factors and meteorological station measurement data,the TRMM 3B42 remote sensing data in the Xieshui River Basin is fused and corrected,and the cor-rected daily precipitation error is quantitatively evaluated.The results show that the accuracy of precipitation data correc-ted by the CNN-LSTM model has been improved,with a correlation coefficient above 0.65 with station precipitation da-ta.The accuracy of RRMSE,MMAE,and PPOD has been improved by 4.01％,8.09％,and 16.61％respectively compared to the original TRMM data,and the corrected data significantly underestimates precipitation.In addition,the accuracy of remote sensing fusion correction precipitation is good under different rainfall intensity conditions,but the accuracy of rain-storm event detection still needs to be improved.
deep learningprecipitation fusionCNN-LSTM modelungauged regionXieshui River Basin
NETL
NSTL
万方数据
维普
