The Research Of GPS-IR Snow Depth Retrieval Combined With GA-BP Neural Network

The snow contains a lot of freth water,and it is an important part of the hydrologic cycle,which is one sixth of fresh water in the world.Therefor,monitoring the snow change is of great signigicance to the fresh water control.In recent years,the use of GPS-IR technique has become a hot research topic in China and abroad.In this paper,the system error and jump in snow depth retrieval,using GA-BP neural network to optimizate the results,and the feasibility and effectiveness of GA-BP neural network in the application of GPS-IR snow deep retrieval are discussed.The main research contents and results of this paper are as follows:(1)Based on the Fresnel reflection theory,the reflection region of the GPS-IR snow deep retrieval is determined.By comparing the different satellite elevation and the carrier,the smaller the satellite elevation is,the larger the reflection area is.L1 carrier's reflection area is less than L2 carrier,and L2 carrier's reflection region is not much different from L5 carrier.(2)The GPS-IR technology is realized by Matlab programming,according to the GPS observation data of AB33 station and P101 provided by PBO network,the snow depth is successfully retrievaled,and the feasibility of the GPS-IR snow deep inversion is verified.(3)Based on original research,135 days observation files of P101 station is processed,and the retrieval snow depth of 26 satellite is got,after compared with the measured snow depth,found that the retrieval precision of most of satellite is good,but the system error and jump appear in the results.The mean RMSE of 26 single satellites retrieval results is 0.319 m,the average MAE is 0.237 m,and the average correlation coefficient is 0.8259.Inversion method using will be 26 single satellites retrieval snow depth are fusion by using multiple GPS-IR retrieval method,the jump value is weakened,but the system error problems is not be solved,RMSE of retrieval results is 0.145 m,MAE is 0.132 m,the correlation coefficient is 0.9906,reaches the highest accuracy in single satellite retrieval.(4)In order to weaken the system error and jump problem in retrieval result,6 different retrieval accuracy satellites of 26 satellites are selected,using GA-BP neural network to optimize the retrieval result of 6 satellites.By training the retrieval snow depth in 1~75 days of six satellites,and output the six groups retrieval snow depth in 76-135 days of six satellites.After compared with the measured snow depth and original snow depth in 76~135 days,it is found that using GA-BP neural network can weaken the system error and jump of original inversion,the original inversion results in 76~135 days of six satellites got different degrees optimization,reduced the RMSE and MAE value,bigger correlation coefficient are obtained.After adjust the GA-BP experimental parameters,using the GA-BP to train the first half of 6 satellite retrieval snow depth,and output one set retriveal snow depth of latter half,RMSE and MAE of output reasult are respectively 0.101 m and 0.075 m,the correlation coefficient is 0.9548.It shows that appling GA-BP neural network to GPS-IR snow depth retriveal is feasible and beneficial.