Water quality parameter prediction using a temporal fusion transformer

Abstract Effective monitoring and prediction of river water quality parameters are vital for sustainable water resource management and ecosystem preservation. This study introduces an innovative Temporal Fusion Transformer (TFT) model, leveraging emerging technologies such as the Internet of Things (IoT) and artificial intelligence, to enhance the accuracy of water quality parameter forecasting, with a focus on dissolved oxygen, a critical indicator of aquatic ecosystem health due to its direct impact on biological activity, pollutant degradation, and habitat sustainability. The model addresses the challenges of noisy, incomplete, and irregular IoT-generated data by integrating advanced preprocessing techniques, including seasonal-trend decomposition based on loess and the Savitzky-Golay filter, to decompose time-series data into trend, seasonal, and residual components. By incorporating supplementary meteorological data, such as temperature, the TFT model captures environmental influences, improving robustness and predictive performance. Empirical evaluations using real-world datasets demonstrate that the proposed model outperforms traditional methods, including support vector machine, long short-term memory networks, and temporal convolutional networks, across metrics such as RMSE, MAE, and MAPE. This approach not only advances real-time water quality monitoring but also supports sustainable water management by enabling proactive decision-making and resource allocation in dynamic environmental conditions. The model’s scalability and adaptability make it a promising tool for broader applications in environmental monitoring and sustainable ecosystem management.