Extrapolation Method Of CAPPI Radar Echo Image

With the development of various fields such as transportation,high-speed rail and flights,the impact of short-term heavy rainfall disasters caused by strong convective weather on social economy and livelihood has become increasingly prominent.Because of the convective weather will cause disasters such as extreme short-time rainfall,meteorologist has found out some traditional and effective solutions for forecasting,such as the extrapolation of CAPPI(Constant Altitude Plan Position Indicator)radar echo images.Historical meteorological data contains the regular pattern and knowledge of meteorological evolution.However,traditional forecasting methods usually consider only a small part of the massive historical data in the meteorological field and this leads to low utilization of data.Based on the full use of the CAPPI radar echo image data,this paper studies the application of the convolutional neural network model in the field of radar echo extrapolation,and realizes the extrapolation of the CAPPI radar echo image.Followings are the contents of the research:First,the method of image semantic extraction from CAPPI radar echo image is studied.Since the data source used is the CAPPI radar echo integrated image,this type of integrated image contains the radar echo block necessary for training,as well as the noise caused by the mountain building,the visualized longitude and latitude,the visualized radar radius auxiliary circle,and the provincial boundary.Sea and land and other additional information.Such additional information will have a great impact on the training of the neural network.Therefore,this paper studies and proposes an image semantic extraction algorithm for CAPPI radar echo images.The mask matrix generated by this algorithm can distinguish radar echoes from such CAPPI radar echo composite images to the pixel point.Block part,and extract the radar echo block,and finally generate the corresponding grayscale value image according to the intensity of the echo,which ensures the availability of the data set required for training from the training data source.Secondly,in view of the strong connection between the radar echo images of two adjacent frames,a radar echo extrapolation model based on Dynamic Probability Convolutional Neural Networks(DPCNN)is proposed.The difference from traditional convolutional neural networks is that the model calculates the corresponding probabilistic convolution kernel for each set of input sequences,so that when the network is in the prediction phase,some of the convolution kernel parameters are in a dynamically variable state,so that the model can The corresponding probability "adjustment" is made to different input sequences,thereby enhancing the correlation between the extrapolation result and the input sequence.Since the posterior knowledge on which this "adjustment" is based is obtained from the massive meteorological data during the training process,the advantages of massive data in the meteorological field can be brought into play.The experimental results show that the image semantic extraction algorithm of the CAPPI radar echo image used in the article can ensure the consistency of the echo intensity before and after extraction;by comparing with the traditional extrapolation methods commonly used in the meteorological field,the DPCNN mentioned in the article In the extrapolation experiment of short-term heavy precipitation data in a local area,the extrapolated image,CSI index,FAR index,and POD index all verify the effectiveness of the DPCNN model.