Identification of Severe Precipitation Radar Echo Reflectivity with Back-Propagation ANN

Abstract

In this thesis, the radar echo reflectivity of severe precipitation in the flood season of Changjiang-Huaihe area was identified by a Back-Propagation (BP) Model of Artificial Neural Network (ANN). The trained network was applied in a precipitation progress in the same area in 2001. The results illustrate that: the single hide-layer BP ANN can be used to identify the target radar echo at a high succeed rate. It is also validated that the performance of the network is influenced by following factors: the quality and input sequence of the training sample, the framework of hide layer and the learning rate.