Work

Optimizing dissolved oxygen monitoring in Eutrophic Lakes: A Kalman Filter-LSTM approach for distinct water quality zones

Year: 2026

Type: article

Abstract: Accurate dissolved oxygen (DO) forecasting is crucial for managing eutrophic ecosystems. However, determining optimal monitoring intervals for different water quality zones remains challenging. This study focuses on integrating a Kalman Filter (KF) with a Long Short-Term Memory (LSTM) model to reduce noise and uncertainty arising during the sampling process. The improved KF-LSTM model was then compared with the traditional LSTM model to determine the most effective approach and to identify the optimal monitoring frequency for various water quality zones. Results revealed that, compared with the stable and good water quality S2 zone, exhibiting a coefficient of variation ( CV ) from 0.13 to 0.14, the DO concentration variability in the poor water quality S1 zone rises significantly, with its CV ranging from 0.21 to 0.22. KF-LSTM outperformed LSTM, especially in long monitoring intervals, improving Nash-Sutcliffe Efficiency ( NSE ) in S2 zone from 0.68 to 0.74 and in S3 zone with average water quality from 0.18 to 0.41. In S1 zone, the R 2 value for KF-LSTM's 2-week predictions dropped from 0.93 with 4-h monitoring to 0.52 with 7-day monitoring, emphasizing the need for high-frequency monitoring. The optimal temporal resolution for accurately capturing the dynamic high-frequency variability characteristics of DO at S1 zone sites is a 4-h interval, achieving an R 2 value of 0.92. In contrast, for S2 zone sites with better and relatively stable water quality, the model maintains an R 2 value above 0.85 even when the interval monitoring is extended to 24hours or up to 7 days. These findings validate KF-LSTM's effectiveness and demonstrate the value of tiered monitoring strategies for improved DO concentration prediction and cost-effective management, offering a practical framework for optimizing monitoring efforts in eutrophic lakes. • Dissolved oxygen (DO) variability differs significantly across trophic lakes in different water quality zones. • The novel Kalman Filter-Long Short-Term Memory (KF-LSTM) model demonstrates superior accuracy in predicting DO concentrations, especially under low-frequency monitoring conditions. • A monitoring frequency of 4-h is recommended for highly variable regions, while 24-h or 7-day monitoring intervals are sufficient for stable water quality zones. • A hierarchical monitoring strategy is proposed to enhance DO prediction capabilities while promoting cost-effective watershed management practices.

Source: Ecological Indicators

DOI: 10.1016/j.ecolind.2026.114668

Authors: Jingjing Ji, WenJun Yang, Jialong Chen, Chang Han, Yujie Zhang, Junteng Huang

Institutions: Changjiang River Scientific Research Institute, Hohai University

Language: English

Topic: Hydrological Forecasting Using AI

Subfield: Environmental Engineering

Field: Environmental Science

Domain: Physical Sciences

SDG: Clean water and sanitation

Open Access status: gold

Estimated APC paid: $3,650