Related Experiment Video
Updated: Feb 26, 2026
Watershed Planning within a Quantitative Scenario Analysis Framework
Published on: July 24, 2016
Multiscale decomposition and fuzzy-rule attention: A transferable cross-basin framework for long-term water quality
Jingzhe Hu1, Ying Li2, Dawei Jiang1
1State Key Laboratory of Water Disaster Prevention, Yangtze Institute for Conservation and Development, Key Laboratory of Hydrologic-Cycle and Hydrodynamic-System of Ministry of Water Resources, Hohai University, Nanjing, 210024, Jiangsu, China; College of Hydrology and Water Resources, Hohai University, Nanjing, 210024, Jiangsu, China.
This study introduces a novel deep learning framework for accurate river water quality forecasting, outperforming existing methods. The model effectively predicts water quality indicators and transfers knowledge across different river basins.
Area of Science:
- Environmental Science
- Data Science
- Water Resource Management
Background:
- Accurate long-term river water quality forecasting is crucial for ecosystem management.
- Traditional process-based models struggle with fragmented data and complex environmental interactions.
- Existing models face challenges in balancing predictive accuracy with inter-basin generalizability.
Purpose of the Study:
- To develop a transferable deep learning framework for robust river water quality prediction.
- To address limitations of current models in handling data scarcity and non-stationary drivers.
- To improve the generalizability of water quality forecasting models across diverse river basins.
Main Methods:
- Integrated a deep learning framework combining physical signal decomposition (multichannel singular spectrum analysis) with fuzzy logic.
- Employed a Transformer architecture with adaptive fuzzy mechanisms to manage uncertainty and capture spatiotemporal dependencies.
- Validated the model using in-situ observations from 149 monitoring stations in Chinese river basins (2007-2022).
Main Results:
- Achieved an average Nash-Sutcliffe efficiency (NSE) of 0.75±0.19 for 26-step-ahead (six-month) forecasts.
- Demonstrated effective prediction for dissolved oxygen, pH, and improved stability for event-driven pollutants (CODMn, ammonia nitrogen).
- Outperformed state-of-the-art baselines with a 24.9% average NSE improvement and showed a 28.4% performance gain via cross-region modeling.
Conclusions:
- The proposed framework offers a promising solution for water quality forecasting, especially in data-scarce regions.
- Integrating structural decomposition with fuzzy logic enhances predictive accuracy and model transferability.
- The model's ability to transfer learned patterns across heterogeneous regions validates its practical applicability.
