Rainfall prediction of geostationary meteorological satellite images using artificial neural network

Rainfall intensity calculation is an important research theme in rainfall precipitation forecast, navigation, weather prediction, etc.. The authors propose a feature based neural network approach for rainfall prediction. A four-layer neural network is used to automatically learn the internal relationship between geostationary meteorological satellite GMS data and rainfall intensity distribution. The input data used are infrared and visible imagery of GMS image. Back propagation learning algorithm is used. The experimental area is the open sea near Shikoku, Japan. The output layer has 4 outputs, indicating rain intensity levels. The ground truth radar data is used as expectation output during the training stage. Finally, the trained network is tested on the validation data set on the same image, and is used to classify rainfall intensity of other GMS image data