Short-term water quality variable prediction using a hybrid CNN–LSTM deep learning model
Rahim Barzegar,Mohammad Taghi Aalami,Jan Adamowski
DOI: https://doi.org/10.1007/s00477-020-01776-2
IF: 3.821
2020-02-01
Stochastic Environmental Research and Risk Assessment
Abstract:Water quality monitoring is an important component of water resources management. In order to predict two water quality variables, namely dissolved oxygen (DO; mg/L) and chlorophyll-a (Chl-a; µg/L) in the Small Prespa Lake in Greece, two standalone deep learning (DL) models, the long short-term memory (LSTM) and convolutional neural network (CNN) models, along with their hybrid, the CNN–LSTM model, were developed. The main novelty of this study was to build a coupled CNN–LSTM model to predict water quality variables. Two traditional machine learning models, support-vector regression (SVR) and decision tree (DT), were also developed to compare with the DL models. Time series of the physicochemical water quality variables, specifically pH, oxidation–reduction potential (ORP; mV), water temperature (°C), electrical conductivity (EC; µS/cm), DO and Chl-a, were obtained using a sensor at 15-min intervals from June 1, 2012 to May 31, 2013 for model development. Lag times of up to one (t − 1) and two (t − 2) for input variables pH, ORP, water temperature, and EC were used to predict DO and Chl-a concentrations, respectively. Each model's performance in both training and testing phases was assessed using statistical metrics including the correlation coefficient (r), root mean square error (RMSE), mean absolute error (MAE), their normalized equivalents (RRMSE, RMAE; %), percentage of bias (PBIAS), Nash–Sutcliffe coefficient (\(E_{NS}\)), Willmott's Index, and graphical plots (Taylor diagram, box plot and spider diagram). Results showed that LSTM outperformed the CNN model for DO prediction, but the standalone DL models yielded similar performances for Chl-a prediction. Generally, the hybrid CNN–LSTM models outperformed the standalone models (LSTM, CNN, SVR and DT models) in predicting both DO and Chl-a. By integrating the LSTM and CNN models, the hybrid model successfully captured both the low and high levels of the water quality variables, particularly for the DO concentrations.
environmental sciences,engineering, environmental,water resources, civil,statistics & probability