Happy Hour Seminar - Polarimetric radar-based precipitation classification, estimation and nowcasting using deep learning

It is long believed that radar observations are rich in information and the traditional ways of evaluating them have only been able to extract part of the information due to the nature of the analytical tools. This talk presents various machine learning techniques for precipitation classification, estimation, and nowcasting using polarimetric radar measurements. A hybrid random forest approach is developed to identify different hydrometeor types and quantify the hail size if hails are detected. A neural network model is designed to quantify surface rain rates with multidimensional polarimetric radar measurements aloft. This deep neural network can extract the complex relation from high dimensional input space (i.e., radar data) to the target space (i.e., surface rain rate). This talk also discusses the applicability of deep learning for radar-based precipitation nowcasting. A convolutional neural network is utilized as benchmark, and a transfer learning framework is developed to incorporate the learned knowledge at one location to other locations characterized by different precipitation features to produce enhanced prediction. The experimental results demonstrate that deep learning techniques can improve radar hydrometeorology applications and facilitate data fusion in weather radar observations.