The objective of this project was to test the use of artificial neural networks (ANN) for improving quantitative precipitation forecasts for the Swatara Creek in eastern Pennsylvania, for which observations and numerical weather model forecasts of precipitation consistently lead to underestimation of streamflows by the hydrological models. The Swatara is representative of certain problem areas in which orographic effects on rainfall amounts are experienced, but remain generally undetected until the runoff reaches the gauging point.
A number of ANN models, operating at different time scales (daily, 6-hour and 3-hour forecasts) and based on different combinations of governing variables were used. Although some of them did not perform as well as desired for the objective, they may be useful in the future for other applications and for regions where ground observations are even more scarce or available in worse quality than in eastern Pennsylvania. In addition, a neural network scheme for replacing data was also developed and provided to the Middle-Atlantic River Forecast Center. This scheme was validated for a number of gauges and can be very useful at times of crisis, when some gauges may not be operational.