Deep learning-based water quality prediction model combined with meteorological data for rivers around Taihu Lake

Abstract

[Objective] To optimize water quality monitoring and early warning systems and promote the sustainable development of river ecosystems. [Methods] Using daily water quality monitoring data from 19 monitoring stations around Taihu Lake from 2021 to 2023 and daily meteorological data from 7 meteorological stations, the water quality conditions were quantitatively evaluated using the comprehensive water quality index(WQI) method. Meteorological data spatially consistent with water quality monitoring stations were obtained through Kriging interpolation. Meteorological input variables were selected by comprehensively considering the influencing mechanisms of meteorological factors on water quality parameters and the result of Spearman correlation analysis. Predictions of the comprehensive WQI were conducted using Long Short-Term Memory(LSTM), Gated Recurrent Units(GRU), Backpropagation(BP) neural network, and a parallel GRU-LSTM model. [Results] The result showed that in the water quality prediction model, model accuracy was affected by the input step length. The parallel GRU-LSTM model using a 14-day input step length achieved the best performance, with a prediction accuracy of R²=0.98. [Conclusion] The deep learning-based prediction model provides a new technical approach for long-term monitoring and prediction of river water quality. Water quality prediction combined with meteorological data can help relevant authorities to warn water quality changes in advance in practical applications, optimize water resources scheduling and management strategies, and improve the sustainable management of water environment.

Keywords

water quality prediction model combined with meteorological data / deep learning / comprehensive water quality index / river water quality / influencing factors