On water level forecasting using artificial neural networks: the case of the Río de la Plata Estuary, Argentina

The Río de la Plata Estuary (RdP) is frequently affected by large storm surges that have historically caused social and economic losses. According to recent research, the number and strength of surge events have been increasing over time as a result of climate change. Although process-based models have been widely used for the storm surge prediction, their high computational demand may be a significant disadvantage in some applications, such as rapid or neartime forecasting. Artificial neural network (ANN) becomes an alternative tool to forecast the water level, taking into account meteorological and astronomical forcing as numerical models also do. In this work, an ANN model performance was evaluated to hindcast and forecast water levels in the RdP. Several combinations of lead times and inputs were assessed in order to find the best configuration. The resulting model provides 4-day forecasts for Buenos Aires and Torre Oyarvide stations (located at the upper and intermediate estuary, respectively), using observed water levels, meteorological inputs and predicted astronomical tides. Results also support the ANN model’s ability to simulate even extreme events. For instance, for a 12 h-forecast, the RMSE is about 20 cm. Finally, we conclude that the model developed here can effectively complement the empirical and numerical forecasts executed by Naval Hydrographic Service, reducing computational costs and leveraging available datasets.