A neural network regression model for tropical cyclone forecast

In recent years a large amount of literature has evolved on the usage of artificial neural network (ANNs) for weather forecasting, particularly because of ANNs´ ability to model an unspecified nonlinear relationship of various meteorological variables. In this paper we proposed a dynamic competitive neural network classifier to predict the maximum potential intensity (MPI) of a given tropical cyclone, based on a 10-year period of Western North Pacific tropical cyclones and monthly mean sea surface temperature (MSST). A procedure to select most significant-correlated attributes of tropical cyclones is designed to for fast and accurate neural network training. A binary trigger is adopted to adjust the structure of the network layers. To justify the performance, we carry out a set of experiments to prove that our proposed model is promising.