Reservoir eutrophication prediction with explainable framework integrating temporal convolutional networks and bidirectional sequence modeling
Abstract
Accurate prediction of total nitrogen (TN) concentration in dam reservoirs is recognized as one of the significant challenges in water resources management. In this regard, the present study, to develop an intelligent and robust modeling framework, introduced the integrated model, which consists of the combination of the improved variational mode decomposition algorithm (IVMD), the attention mechanism-based temporal convolutional neural network (AttTCN), the bidirectional gated recurrent unit (BiGRU), and the grasshopper optimization algorithm (GOA), which is used to predict the total nitrogen of Mahabad Dam. The results of this study suggest that the proposed framework has a significant advantage over conventional models, as the analysis of the indices indicated that, with each layer of complexity added to the model, prediction accuracy improved significantly. The 80% reduction in the mean absolute error value from the basic BiGRU model to the final IVMD-GOA-AttTCN-BiGRU model, along with the improvement of Willmott’s index of agreement from 0.75 to 0.99, confirms the superiority of the proposed integrated approach. This significant improvement in accuracy demonstrated that the intelligent combination of data preprocessing, an attention mechanism, and hyperparameter optimization can yield models with greater reliability in water quality prediction. The use of interaction terms such as turbidity × pH corrected the systematic bias of the model, which is hydrologically justified. In summary, the proposed framework shows promise as an accurate prediction tool and provides a useful analytical basis for better understanding water quality dynamics in the Mahabad reservoir, subject to further validation with larger datasets.
Related
- Hybrid stacked bidirectional deep learning with attention based air pollution prediction for environmental sustainability
- Application of long short-term memory and convolutional neural networks for real-time bridge pier scour prediction
- Integrating Molecular Docking, Dynamics, and AI-Based ADMET Prediction to Decipher the Toxicity Profile of Isometamidium Chloride Against Animal Trypanosomiasis
- Machine learning models with explainable AI for PM2.5 prediction for Urban air quality assessment
- An explainable deep learning framework for sub-seasonal to seasonal agricultural drought prediction over India
BibTeX
@article{sharafi2026reservoir,
title = {{Reservoir eutrophication prediction with explainable framework integrating temporal convolutional networks and bidirectional sequence modeling}},
author = {Milad Sharafi and Javad Behmanesh and Vahid Rezaverdinejad and Saeed Samadianfard},
journal = {Scientific Reports},
year = {2026},
month = oct,
doi = {10.1038/s41598-026-74162-7},
url = {https://www.nature.com/articles/s41598-026-74162-7},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Reservoir eutrophication prediction with explainable framework integrating temporal convolutional networks and bidirectional sequence modeling
AU - Milad Sharafi
AU - Javad Behmanesh
AU - Vahid Rezaverdinejad
AU - Saeed Samadianfard
PY - 2026
DA - 2026/10/12
JO - Scientific Reports
DO - 10.1038/s41598-026-74162-7
UR - https://www.nature.com/articles/s41598-026-74162-7
AB - Accurate prediction of total nitrogen (TN) concentration in dam reservoirs is recognized as one of the significant challenges in water resources management. In this regard, the present study, to develop an intelligent and robust modeling framework, introduced the integrated model, which consists of the combination of the improved variational mode decomposition algorithm (IVMD), the attention mechanism-based temporal convolutional neural network (AttTCN), the bidirectional gated recurrent unit (BiGRU), and the grasshopper optimization algorithm (GOA), which is used to predict the total nitrogen of Mahabad Dam. The results of this study suggest that the proposed framework has a significant advantage over conventional models, as the analysis of the indices indicated that, with each layer of complexity added to the model, prediction accuracy improved significantly. The 80% reduction in the mean absolute error value from the basic BiGRU model to the final IVMD-GOA-AttTCN-BiGRU model, along with the improvement of Willmott’s index of agreement from 0.75 to 0.99, confirms the superiority of the proposed integrated approach. This significant improvement in accuracy demonstrated that the intelligent combination of data preprocessing, an attention mechanism, and hyperparameter optimization can yield models with greater reliability in water quality prediction. The use of interaction terms such as turbidity × pH corrected the systematic bias of the model, which is hydrologically justified. In summary, the proposed framework shows promise as an accurate prediction tool and provides a useful analytical basis for better understanding water quality dynamics in the Mahabad reservoir, subject to further validation with larger datasets.
SN - 2045-2322
ER -