Prediction of extreme precipitation using a neural network:: application to summer flood occurrence in Moravia

Dramatic floods occurred in Central Europe in summer 1997, and Czech Republic has been seriously affected in its eastern part-Moravia. A predictive approach based on modelling flood recurrence may be helpful in flood management. Summer floods are typically characterized by saturated catchment due to long-lasting heavy precipitation followed by a sudden extreme rainfall. In present work an artificial neural network (ANN) model was evaluated for precipitation forecasting. Back propagation neural networks were trained with actual monthly precipitation data from two Moravian meteorological stations for a time period of 38 years. Predicted amounts are of next-month-precipitation and summer precipitation in the next year. The ANN models provided a good fit with the actual data, and have shown a high feasibility in prediction of extreme precipitation. (C) 2000 Elsevier Science Ltd. All rights reserved.