Answers are grounded in this paper and cite their sources.
Demonstrates improved prediction of water quality indicators in aquatic ecosystems, indicating effective ecological monitoring tools.
Accurate assessment of aquatic ecological health relies on the effective characterization of the temporal dynamics of key water quality and algal indicators. To address challenges such as strong non-stationarity, significant noise interference, and complex multivariate coupling in water environment monitoring time series, this paper introduces a differential attention mechanism (DIFF Transformer) and constructs a time series prediction model for key indicators of aquatic ecological health. Taking the Huangtaiqiao monitoring section of the Xiaoqing River in Jinan, China, as the study area, multivariate time series prediction research was conducted for key proxy indicators of aquatic ecological health, including blue-green algae, chlorophyll, dissolved oxygen, turbidity, total phosphorus, ammonia nitrogen, permanganate index, and TOC, based on hourly measured data from 2017 to 2024. The results show that the differential attention mechanism improves balanced forecasting across multiple water quality and algal indicators by reducing isolated and unstable high-frequency residual fluctuations and enhancing the reconstruction of dominant temporal patterns. Compared with LSTM, GRU, and the standard Transformer, DIFF Transformer achieved the highest macro-averaged Nash-Sutcliffe efficiency coefficient among the learning-based models, with a macro-averaged NSE of 0.93. Per-indicator evaluation further showed that DIFF Transformer achieved NSE values above 0.95 for 11 of the 12 water quality indicators. In addition, DIFF Transformer obtained lower MSE, MAE, and RMSE than LSTM, GRU, and the standard Transformer. These results indicate that DIFF Transformer can provide continuous, stable, and high-temporal-resolution predictions of key aquatic ecological proxy indicators, supporting routine water quality trend tracking and dynamic aquatic ecological assessment.
No takes yet. Share an insight, caveat, or question.
Li et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: