Forecasting Model for Danube River Water Temperature Using Artificial Neural Networks

The objective of this paper is to propose an artificial neural network (ANN) model to forecast the Danube River temperature at Chiciu–Călărași, Romania, bordered by Romanian and Bulgarian ecological sites, and situated upstream of the Cernavoda nuclear power plant. Given the temperature increase trend, the potential of thermal pollution is rising, impacting aquatic and terrestrial ecosystems. The available data covered a period of eight years, between 2008 and 2015. Using as input data actual air and water temperatures, and discharge, as well as air temperature data provided by weather forecasts, the ANN model predicts the Danube water temperature one week in advance with a root mean square deviation (RMSE) of 0.954 °C for training and 0.803 °C for testing. The ANN uses the Levenberg–Marquardt feedforward backpropagation algorithm. This feature is useful for the irrigation systems and for the power plants in the area that use river water for different purposes. The results are encouraging for developing similar studies in other locations and extending the ANN model to include more parameters that can have a significant influence on water temperature.