RAINFALL FORECASTING IN SPACE AND TIME USING A NEURAL NETWORK

A neural network is developed to forecast rainfall intensity fields in space and time; it is a three-layer learning network with input, hidden, and output layers. Training is conducted using back propagation where the input and output rainfall fields are presented to the neural network as a series of learning sets. After training is complete, the neural network is used to forecast rainfall intensity fields with a lead time of 1 h using only the current field as input. Rainfall fields are generated using a space-time mathematical rainfall simulation model, and forecasted fields are compared with the perfectly known model-produced fields. Results indicate that a neural network is capable of learning the complex relationship describing the space-time evolution of rainfall such as that inherent in a complex rainfall simulation model. One hour ahead forecasts are produced, and comparisons with true mean areal intensities and percent areal coverage indicate that in most cases the method performs well when applied to the events used in training. The neural network is used to forecast a series of events not included in the training data and is shown to perform well when a relatively large number of hidden nodes are utilized. Performance of the neural network is compared with two other methods of short-term forecasting, persistence and nowcasting.