Prediction of water temperature in a subtropical subalpine lake using an artificial neural network and three-dimensional circulation models

Abstract

Water temperature is considered as a dominant factor in controlling the stratification, water quality, and ecological environment of lakes because many biological processes are temperature-dependent. Therefore accurate prediction of water temperature is crucially important for lake management. Traditional three-dimensional circulation models have been widely adopted to predict lake water temperatures spatially and temporally. An artificial neural network (ANN) technique is used as an alternative in water temperature simulation studies. The present study compares the performance of the ANN technique with a physically based three-dimensional circulation model. Four ANN models were established to simulate the time-series water temperature at a buoy station of the Yuan-Yang Lake (YYL) in north-central Taiwan at various measured depths. To evaluate the performance of the ANN and the three-dimensional circulation model, three different statistical indicators were used, including the root mean square error, the mean absolute error, and the coefficient of correlation. The simulated results reveal that the three-dimensional circulation model provides a better prediction of water temperature at different layers, except at the 3m below water surface, during the calibration phase. For the validation phase, the three-dimensional circulation model can predict water temperature satisfactorily at different layers than ANN model. Overall, the performance of water temperature prediction with three-dimensional circulation model is better than that with the ANN model. However ANN is a black box model and fails to simulate the internal physical processes in the lake, while three-dimensional circulation model is physical model which can be used to predict water temperature in spatial and temporal variations simultaneously.