Abstract
Automatic observation facilities for supporting flood control activities, called the SAMTES, are located at about 100 observation sites to monitor water levels in Saitama prefecture. We developed a forecasting system of the next 3 hour-range water levels of rivers at the above sites by Neural Networks for the SAMTES. Input data for this system are observed water levels of rivers, tide levels, observed precipitations and 'Very Short-range Forcasting Precipitation' predicted by Japan Meteorological Agency. In this paper, we showed an outline of the forecasting system and the result of verification of forecasting models. Predicted values of 1-hour forecast were highly precise at most sites. The correlation coeficients of observed and predicted values were from 0.95 to 0.99. In case of 2, 3-hour forecast, the errors of predicted water levels tended to be accumulated by errors included in values of forecasting precipitation, but changing patterns of water levels of rivers were well simulated by the models.
