Precipitation Nowcasting Based On Weather Radar And Its Application On Hydrology Forecasting

Because of China's complex topography and climate,flood is always the main disaster that has caused massive loss.How to make accurate flood forecasting is the focus of hydrological research in China.Effective precipitation data is the key to flood forecasting.As an effective means of detecting precipitation,weather radar has been widely concerned.This paper did researches about quantitative precipitation estimation and radar echo extrapolation in Bahe River by using weather radar data and hourly rain gage data.Then based on these two,the precipitation nowcasting work was carried out.The result of the precipitation nowcasting are input to the XAJ model for flood simulation and forecasting.The specific content and main conclusion are as follows:(1)Precipitation estimation.Two precipitations on June 30 and July 4,2016 were selected for precipitation estimation research.Radar data was interpolated to heights from 1km to 5km,the climatic (5- relationships were built in each height respectively by the Probabilityfitting Technique.Three precipitations on May 3,June 10 and July 9,2017 were selected to test the estimation effect.Effects in different interpolation heights are compared to select the optimal interpolation height.The result shows that the climate Z-R relationship has a better effect on the surface rainfall estimation than on the point rainfall estimation.The optimal interpolation height of the Bahe River Basin is 2km;(2)Radar echo extrapolation.An artificial neural network combining convolutional neural network and long-short-term memory neural network was constructed.The neural network was trained by using radar echo data in Bahe River Basin in 2016 and 2017,so that it can simulate the temporal and spatial variation of radar echo.Two precipitation radar echoes on June 5 and July 9,2017 were selected to test the trained model by comparing extrapolation results and the actual observation data.The result shows that the neural network model performs better in radar echo extrapolation,and can accurately predict the position and contour of radar echoes,but the prediction effect on radar echo intensity is not good;(3)Precipitation nowcasting.Precipitation estimation using climate Z-R relationship and radar echo extrapolation using neural network are combined to obtain the results of one-hour precipitation nowcasting,the nowcasting results were compared with observation data.The comparison result shows that the forecasting effect of the areal rainfall is better,but the forecast effect on point rainfall is general.The nowcasting can basically predict the change process of areal rainfall and the time when the peak of precipitation occurs;(4)Flood forecasting.The surface rainfall obtained from the forecast of precipitation is used as the precipitation input of the XAJ Model,and the flood forecasting work is carried out.Three flood processes in June,July,2016 and April,2017 were used to determine the parameters of the XAJ Model,two flooding processes in June and July,2017 were used to test the application effect of the precipitation nowcasting results in the flood forecasting.The analysis shows that when inputting precipitation nowcasting results to XAJ Model,the change process and peak time of runoff can be basically predicted,but the forecast effect on flow rate still needs improvement.