Development of water quality models for supporting NH3-N control in a dam regulated river

A seasonal occurrence of high ammonia nitrogen (NH3-N) concentrations has hampered chemical treatment processes of a water plant in Geum river of Korea. Monthly flow allocation from upstream dam is important for downstream NH3-N control. In this study, water quality models based on multiple regression (MR) and artificial neural network (ANN) methods were developed to support dam operations through providing forecasted NH3-N concentrations. The models were calibrated with 7 years of monthly data, and verified with another 3 years of independent data. In the models, the NH3-N concentration for next time step is dependent on dam outflow, river water quality such as alkalinity, temperature, and NH3-N of previous time step. During the calibration phase, the ANN models compared to MR models showed a better agreement with observed data as indicated with small RMSE and high R2 values. However, in the verification phase, the performance of ANN models was decreased and showed a little difference to MR models. From the model comparisons, it is recommended for both ANN and MR models to include the autocorrelation of NH3-N concentrations up to 3-lag months to avoid overestimation during low flow season.