AH-GCAN-LSTM: Adaptive Hybrid-Graph Convolution Attention Network with LSTM and Genetic Optimization for Hydrological time Series Forecasting

Huseyin Cagan Kilinc (), Sina Apak, Furkan Ozkan and Adem Yurtsever
Additional contact information
Huseyin Cagan Kilinc: Istanbul Aydın University, Department of Civil Engineering
Sina Apak: Istanbul Aydın University, Department of Management Information Technology
Furkan Ozkan: Çukurova University, Department of Computer Engineering
Adem Yurtsever: Istanbul University-Cerrahpaşa, Department of Environmental Engineering

Water Resources Management: An International Journal, Published for the European Water Resources Association (EWRA), 2025, vol. 39, issue 15, No 21, 8293-8312

Abstract: Abstract Monitoring the flow rates of rivers and appropriately managing their basins can be possible by developing effective forecasting methods. Therefore, streamflow prediction can be a crucial indicator for determining the future conditions of water resources. This study presents a novel approach for accurate hydrological time series forecasting. The proposed Adaptive Hybrid-Graph Convolution Attention Network, referred to as AH-GCAN-LSTM, integrates Long Short-Term Memory (LSTM) to capture temporal dependencies and Graph Convolutional Networks (GCN) to model spatial relationships between hydrological stations. To assess the robustness of the proposed method, we compared it to other algorithms such as FFO (Fire Fly Algorithm), PSO (Particle Swarm Optimization) and GA (Genetic Algorithm). A multi-head attention mechanism dynamically enhances the model’s ability to focus on crucial spatial and temporal features. Additionally, genetic algorithms are used to optimize the model’s parameters, further improving its predictive accuracy. In the study, statistical measurement metrics such as RMSE, MAE, R2 and KGE were used. Results demonstrate significant improvements in forecasting accuracy compared to traditional and state-of-the-art models, with reduced error rates and enhanced predictive performance across the Ulucami, Kaptanpaşa, and Cumhuriyet datasets. Specifically, the proposed AH-GCAN-LSTM model achieved RMSE values of 0.0295, 0.0894, and 0.0769; MAE values of 0.0258, 0.0308, and 0.0333; and R2 scores of 0.9874, 0.9863, and 0.9866, respectively. Additionally, the model attained high NSE values—0.9905, 0.9965, and 0.9964—demonstrating consistent robustness. Compared to GRU, Multi GRU-LSTM Attention, and A-LSTM, AH-GCAN-LSTM outperformed all baselines across all metrics. Genetic Algorithm-based optimization further improved performance, reducing RMSE to 0.02 at Ulucami and 0.06 at Cumhuriyet stations. These results confirm the model’s strong ability to capture complex spatial and temporal dependencies, establishing it as a highly effective tool for hydrological forecasting and water resource management.

Keywords: Graph convolution attention network; Deep learning; LSTM; streamflow; Forecasting (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://link.springer.com/10.1007/s11269-025-04343-y Abstract (text/html)
Access to the full text of the articles in this series is restricted.

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:spr:waterr:v:39:y:2025:i:15:d:10.1007_s11269-025-04343-y

Ordering information: This journal article can be ordered from
http://www.springer.com/economics/journal/11269

DOI: 10.1007/s11269-025-04343-y

Access Statistics for this article

Water Resources Management: An International Journal, Published for the European Water Resources Association (EWRA) is currently edited by G. Tsakiris

More articles in Water Resources Management: An International Journal, Published for the European Water Resources Association (EWRA) from Springer, European Water Resources Association (EWRA)
Bibliographic data for series maintained by Sonal Shukla () and Springer Nature Abstracting and Indexing ().