Artificial Neural Networks and Long-Range Precipitation Prediction in California

Abstract

Artificial neural networks (ANNs), which are modeled on the operating behavior of the brain, are tolerant of some imprecision and are especially useful for classification and function approximation/mapping problems, to which hard and fast rules cannot be applied easily. Using ANNs, this study maps a 1-yr monthly (January?December) time series of the 700-hPa teleconnection indices and ENSO indicators onto the water year (October?September) total precipitation of California?s seven climatic zones, with different lag times between the inputs and outputs. It was found that the pattern of rainfall predicted by the ANN model matched closely the observed rainfall with a 1-yr time lag for most California climate zones and for most years. This research shows the possibility of making long-range predictions using ANNs and large-scale climatological parameters. This research also extends the use of neural networks to determine important parameters in long-range precipitation prediction by comparing results gained using all the inputs with results from leaving an individual index out of the network training. This comparison gives insight into the physical meteorological factors that influence California?s rainfall.