Short-term Rainfall Prediction Method Based On Radar Echo

Nowcasting technology can predict the weather conditions in the next 2 hours by monitoring the current weather conditions.Affected by monsoons in our country,frequent torrential rains have seriously affected social stability and people’s lives.Providing reliable and accurate short-term rainfall forecasts has become a widely concerned issue.At present,shortimminent rainfall prediction mainly relies on radar echo extrapolation technology,that is,the shape and intensity of future echo images are estimated from historical radar echo images,which has the problems of insufficient accuracy and short extrapolation timeliness.Based on this,this thesis aims to carry out theoretical research on short-term rainfall prediction based on historical radar data from the perspective of machine learning and deep learning.The main research of this thesis includes the establishment of short-imminent rainfall prediction model and the practice of short-imminent rainfall prediction in Liaoning Province.During the research process,historical radar data is fully utilized in order to achieve a certain improvement in the precision of the forecast and minimize the loss of meteorological disasters.First,decode and preprocess the acquired raw radar data,draw radar echo images,and extract radar image features.Then from the original precipitation data,the data corresponding to the radar is screened out,and after further cleaning,it is matched with the radar data to obtain a data set corresponding to the radar echo and the precipitation.Then,on the basis of the definition and modeling method of the existing short-imminent precipitation forecast,the theoretical framework of the short-imminent rainfall prediction model based on deep learning is constructed.The RNN model has the problem of gradient disappearance and explosion,and the LSTM model has the problem that it cannot describe features well due to spatial redundancy.To address the above problems,this thesis gradually adds the CNN layer and the Attention mechanism.A shortimminent rainfall prediction method based on the CNN-LSTM-Attention model is given.Finally,for the radar data at different altitudes,select the altitude data closely related to precipitation,divide the constructed 73,739 pieces of data into a training set and a test set in a ratio of 8:2 for experiments,and compare the data based on RNN,LSTM,and CNN-The precipitation prediction results of the LSTM model are compared,and the mean square error,the average absolute error ratio,and the average absolute error are used as indicators to compare the predicted rainfall and the measured rainfall,and verify the effectiveness of the given method.On this basis,the direction of follow-up research is given.