| With the rapid development of modern meteorological technology(especially the weather radar), quantitative rainfall measurement technique has new development ideas. With high spatial and temporal nature, weather radar is able to meet the needs of modern rainfall observation better both in the real-time and the range of the detection of rainfall.In this paper, by use of Chongqing Meteorological Bureau CINRAD/SA weather radar rainfall data and the corresponding echo rainfall areas of the ground station data, the estab-lishments of radar quantitative precipitation estimation model are accomplished based on BP Neural Network and RBF(Radial Basis Function) Neural Network, which is used to estimate surface rainfall. As a comparison, Rainfall is estimated by the Z-R relation which is calculated by the variational method at the same time. Experimental results show that the established model of radar quantitative precipitation estimation is much better than Z-R relation in aspects of accuracy and stability, and it reflects the real situation of rainfall better.At the same time, this paper also has a groundbreaking discovery:in Chongqing Banan,such a mountainous regions, RBF neural network model effect is better than BP neural network after changing the input parameters of RBF neural network to establish estimate, this has great reference significance.The main research contents and results are as follows:(1) The basic principle of radar quantitative measurements of precipitation, summing up the discussion of the existing mature approach radar quantitative measurement of precipitation (ZI Relations Act, the average correction method, optimal interpolation method, adaptive Kalman filter algorithm, variational method), and research the advantages and limitations of each method, which you can see the change of approach in the field of technology.(2) The history, contents and main characteristics of artificial neural network; study the model structure of neurons, as well as the type and characteristics of neuronal transmission functions; describes the modeling process and the principles of neural networks, in addition to the neural network learning rules listed and described in detail.(3) Use of radar data to measure rainfall model were established by BP neural network and RBF neural network and variational method to get the Z-I relationship (Z=37I145) comparison of experimental results on the test sample, the analysis obtained using two the effect of rainfall estimation neural network model to be significantly better than the Z-I relationship, bias (bias) is reduced by about30%, the average relative root mean square error (RMSE) decreased by about1.5, the correlation coefficient (R2) increased by0.7.(4) To change parameters of RBF neural network model, I found that in this article the research of banan district of chongqing, because it has many hills, by increasing two input parameters (elevation, distance) of the RBF neural network to estimate rainfall model, I get the best results in the experimental process, so this article establish the model for the neural network application in the field of quantitative estimates radar rainfall has great reference significance.(5) Quantitative measurement of rainfall was designed and implemented in chongqing banan radar system, shows the system architecture and functions of each module, and illustrates the key technology of implementation process.The practice shows that the system can be more accurate to provide the banan township-level rainfall forecast, forecast of local grassroots forecasters work provides a good reference, at the same time because the banan hills, rivers, the monsoon rain, the system is also on the local flash floods, landslides and other meteorological disaster warning has the very high value and significance. |
PDF Full Text Request