Coupled convolutional neural network with long short-term memory network for predicting lake water temperature

Abstract

Water temperature is critical to freshwater ecosystems and lake habitats, and it directly affects the growth, reproduction and habitat of aquatic organisms. In this study, a separable convolutional neural network (SCNN) and long short-term memory (LSTM) neural network are combined to propose a SCNN-LSTM model for lake temperature prediction with only temperature as model input. The proposed SCNN-LSTM model was constructed using 35,064 samples, including hourly measured lake water temperature at 12 different depths, air temperature, wind speed, and solar irradiance. To evaluate the performance of the hybrid SCNN-LSTM model, the model was compared with the extreme gradient boosting (XGBoost) and Air2water models, respectively. In addition, the SCNN-LSTM model that only considers temperature is compared with the SCNN-LSTM model that considers multiple factors. The results show that the SCNN-LSTM model is superior to the XGBoost and Air2water models in predicting lake water temperature, and the SCNN-LSTM model considering only temperature has certain competitiveness compared with the SCNN-LSTM model considering multiple meteorological factors such as temperature, wind speed and solar irradiance.