The application of neural networks and partial least squares in Korea weather prediction

It is known that the rainfall in Korea is closely related to a set of weather elements measured at specific pressure levels in Monsoon area before one month. In this paper a hybrid model of stochastic method and neural network is suggested for the prediction of a weather element in Korea which has mainly influence on the degree of rainfall. After preprocessing a given training set of high dimensional data into a new set of low dimensional-but effective-data by the PLS(partial least squares) method, a feed-forward neural network is trained for forecasting the monthly change of a weather element in Korea. As the result, a neural network forecaster has many advantages such as easy design, fast learning, and easy parameter adaptation.