Deep Learning-Based Multi-Step Ahead Streamflow Forecasting under RCP 4.5 and 8.5 Scenarios
WATER RESOURCES MANAGEMENT, cilt.40, sa.11, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 40 Sayı: 11
- Basım Tarihi: 2026
- Doi Numarası: 10.1007/s11269-026-04875-x
- Dergi Adı: WATER RESOURCES MANAGEMENT
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, ABI/INFORM, CAB Abstracts, Compendex, Environment Index, Geobase, INSPEC, Natural Science Collection (ProQuest), Biological Science Database (ProQuest), Earth, Atmospheric, & Aquatic Science Collection (ProQuest), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Ankara Üniversitesi Adresli: Evet
Özet
Effective water resource management during climate change is increasingly dependent on accurate streamflow forecasting. This study explores the potential of advanced deep learning methods, specifically Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) models, for long-term, multistep ahead, streamflow forecasting in two sub-basins of the Sakarya Basin in T & uuml;rkiye. By integrating high-resolution CMIP6-based climate projections under RCP 4.5 and RCP 8.5 scenarios, spatial interpolation techniques, and diverse meteorological and land use datasets, the research investigates streamflow dynamics until 2099. Approximately 40,000 alternative configurations, including hyperparameter adjustments, model architecture modifications, and training strategies, were tested to optimise model performance. The results demonstrate that deep learning models, particularly when enriched with spatially interpolated inputs and land cover variability, significantly enhance forecasting accuracy. This approach not only provides valuable projections for future water availability but also informs sustainable water management strategies in climate-sensitive regions. In terms of model performance, CNN generally outperformed LSTM in both sub-basins, achieving Kling-Gupta Efficiency (KGE) values ranging between 0.80 and 0.85 during the testing phase, indicating strong predictive capability. Regarding future projections, streamflow in the E12A033 sub-basin is expected to decline by approximately 37% under RCP 4.5 and 42% under RCP 8.5 compared to the historical period. In the E12A053 sub-basin, the reductions are even more severe, with projected decreases of 59% under RCP 4.5 and 62% under RCP 8.5. These substantial declines underscore the escalating impact of climate change on water availability and highlight the need for adaptive, data-driven strategies for regional water resource planning.