Research on weather forecast based on neural networks

The traditional method based on aerodynamics and expertise for weather forecast focuses on present data, but the historical database is not utilized. The paper presents research on weather forecasting through the historical database. Because the atmosphere pattern is a complex, nonlinear system, the traditional method is not an effective method. The multi-layer neural network is a potent method for resolving this problem. With the backpropagation neural networks the atmosphere pattern is divided into five patterns based on the rainfall. Comments on this method and a special atmosphere pattern are given at the end of the paper.