Keywords: artificial neural network, time series decomposition, flood hazard, low pass filter, Kolmogorov-Zurbenko filter
This study introduces a hybrid model for the flood prediction in Mohawk Watershed, New York. We combine time series analysis and artificial neural networks for the explanation and prediction of the daily water discharge time series using the daily hydrological and climatic variables. Data have been used from three different stations in Mohawk Watershed, New York. One methodology applies a low pass filter (Kolmogorov-Zurbenko filter) for the decomposition of the time series into different components (long, seasonal, and short term component) of all the variables. For the prediction of the water discharge time series each component has been described by a multiple regression model. The results can be improved substantially using artificial neural networks for the explanation of each component (long, seasonal, short term component) separately. The combination of the time series decomposition and the neural networks allows us to predict the water discharge time series with high accuracy and an explanation of up to 93%.