Forecasting river water temperature time series using a wavelet-neural network hybrid modelling approach

Accurate and reliable water temperature forecasting models can help in environmental impact assessment as well as in effective fisheries management in river systems. In this paper, a hybrid model that couples discrete wavelet transforms (WT) and artificial neural networks (ANN) is proposed for forecasting water temperature. Four mother wavelets, including Daubechies, Symlet, discrete Meyer and Haar, are considered to develop the WT-ANN hybrid model. The hybrid model is applied to forecast daily water temperature on the Warta River in Poland. Time series of daily water temperatures in eight river gauges as well as daily air temperatures of seven meteorological stations are used for forecasting daily water temperature. The performance of this WT-ANN hybrid model is evaluated by comparing the results with those obtained from linear and non-linear regression models as well as a traditional ANN model. The results show that the WT-ANN models perform well in simulating and forecasting river water temperature time series, and outperform the linear, non-linear and traditional ANN models. The superior performance of the WT-ANN models is particularly observed for extreme weather conditions, such as heat waves and drought. Among the four mother wavelets applied, the discrete Meyer performs the best, slightly better than the Daubechies at level 10 and Symlet, while the Haar mother wavelet has the lowest accuracy. In addition, the model performance improves with an increase in the decomposition level, indicating the importance of the choice of decomposition level. The outcomes of this study have important implications for water temperature forecasting and ecosystem management of rivers.