Research On Precipitation Retrieval Model Based On C-band Weather Radar

Aiming at the core problem of improving the effect of radar quantitative estimation of precipitation,this study selected the observation data of C-band weather radar of Yinchuan Station and the ground rainfall gauge network set up in the surrounding area,and selected the observation data of 10 precipitation processes as the basic data to establish the radar-rainfall gauge data pair.The radar precipitation inversion model was constructed by using two classification methods of automatic recognition and classification of precipitation types and dynamic adjustment combined with optimization processing.The effect of precipitation inversion using different combination models was analyzed in detail,and the average deviation method and Kalman filter calibration method were used to further calibrate and optimize radar precipitation.The optimization effects of two calibration methods for radar precipitation under different rainfall gauge densities were studied.On the other hand,combined with Support Vector Regression(SVR)and Random Forest(RF),the radar precipitation inversion model based on machine learning is constructed with longitude and latitude,altitude and other terrain factors as the input of the model,and the parameters are optimized by grid search and cross verification.To evaluate the application potential of two machine learning methods in radar quantitative precipitation estimation.The main conclusions are as follows:(1)The classification dynamic optimization model using the two classification methods had the best inversion effect(CC=0.72,BIAS=0.75),MAE and RMSE increased by 35.4% and 39.8%,respectively.The model underestimated precipitation in most of the time,but accumulated precipitation was very close to the real precipitation,which had good consistency with the real precipitation situation.Using the fixed Z-R relationship will obviously overestimate the precipitation at low rain intensity and underestimate the precipitation at high rain intensity.Using the optimization processing method alone to estimate precipitation has certain optimization effect at low rain intensity,but it will incorrectly underestimate precipitation at high rain intensity,which has certain limitations.(2)The calibration optimization effect of Kalman calibration method is better than that of MFB calibration method under different precipitation intensity,which can improve the deviation between radar precipitation and real precipitation,and better describe the falling area and magnitude of heavy precipitation distribution;On the whole,the effect of MFB belongs to negative optimization.The optimization effect of MFB calibration method is largely affected by the number of rain gauges.The more rain gauges,the better the calibration optimization effect.The MFB calibration method has obvious limitations in calibrating and optimizing radar precipitation.(3)The machine learning method shows good application potential in radar quantitative precipitation inversion,and SVR model has better performance,CC=0.87,MAE=0.5,RMSE=1.25.The BIAS of both models is above 0.9,but the slope k of fitting curve of scatter distribution image is low.RF has a relatively obvious underestimation of precipitation.When the precipitation intensity R is low,it will slightly overestimate precipitation,while when R is high,it will significantly underestimate precipitation.