Study On The Convective Initiation Warning Technology Based On Numerical Model And Radar Data

Strong convective severe weather occurs frequently in China. Doppler weather radar is a major short-term severe weather monitoring tool. This paper mainly uses measurements with high spatial and temporal resolution derived from four-dimensional variational numerical model based on radar data and uses support vector machine as classifier to study the method of convective initiation warning, while the definition of CI is the first to detect the Doppler radar reflectivity of clouds produce greater than or equal to 35dBZ. The main contents of this paper are divided into the following three sections:The first part is to extract the predictors of measurements with high spatial and temporal resolution derived from four-dimensional variational numerical model. Its main idea is to use scatter plots between radar composite reflectivity of 30 to 60 minutes later and unobserved meteorological fields as the basis to chose predictors. After many experiments, the predictors have been extracted, they are wind shear, convergence lifting and vertical wind velocity and their time variables.The second part is the research about the algorithms of classifier. The predictors obtained from the first part are used as attribute characteristics of classification algorithm, radar composite reflectivity after 30 minutes is used as labels to design the classifier. By studying the classification algorithm of imbalance data set and combined with the evaluation criteria of related fields, then by means of under-sampling step by step to re-balance the training samples. After combining with the SVM, the CI warning predictor algorithm is complicated.The third part is about the classifier realization, case analysis and evaluation. The classifier designed from the second part is used as the CI predictor. By combining with the real radar composite reflectivity, then the CI predictor gives the forecast results of 30 minutes later.The CI warning method proposed of this paper is based on the measurements retrieved from four-dimensional variational numerical model. Then using the resampling techniques of imbalanced data classification to re-balance the data. Finally, combining with SVM to design the complete CI warning classifier. In the end, the experiments prove that this method is effective in CI warning, and it’s necessary to solve the problem of high false alarm rate in the future.