Streamflow estimation for underground dams using machine learning and hydrological modeling: a case study of Bartın Bahçecik underground dam

• Published in: Environmental earth sciences Vol. 84; no. 17; p. 508
• Main Authors: Ekemen Keskin, Tülay, Şander, Emrah
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.09.2025; Springer Nature B.V
• Subjects: Agricultural expansion; Agricultural production; Artificial neural networks; Artificial recharge; Biogeosciences; Case studies; Dams; Discharge measurement; Drainage; Drought; Earth and Environmental Science; Earth Sciences; Effectiveness; Environmental Science and Engineering; Flow estimation; Gaging stations; Geochemistry; Geology; Groundwater; Hydrologic data; Hydrologic models; Hydrology; Hydrology/Water Resources; Learning algorithms; Machine learning; Modelling; Neural networks; Original Article; Population growth; Precipitation; Programming languages; Python; Recharging; Regression analysis; Resource management; Site selection; Software; Stream discharge; Stream flow; Streamflow data; Streamflow estimation; Sustainability; Sustainability management; Terrestrial Pollution; Time series; Water conservation; Water levels; Water management; Water resources; Water resources management; Water shortages; Ankara Turkey; Japan; Turkey; Streamflow estimating; Artificial neural network; Multiple linear regression; Underground dam; Time series analysis; Climate changes
• ISSN: 1866-6280; 1866-6299
• DOI: 10.1007/s12665-025-12511-x

Rapid technological advances, agricultural expansion, and population growth ratio have accelerated the depletion of limited water resources, leading many countries, including Turkey, to emphasize the construction and use of underground dams as an effective strategy for sustainable water management. In order to contribute to the sustainability of underground dams, this study takes the Bahçecik (Bartın) Underground Dam as a case study, aiming to estimate the streamflow data required for the artificial recharge of underground reservoirs using surfacewater through wells. In this context, the streamflow of the main tributary recharging the dam was estimated by jointly evaluating machine learning techniques and hydrological basin modeling results. Time Series Analysis, Artificial Neural Networks (ANN), Multiple Linear Regression (MLR), and the similar basin area ratio methods used at the study. Time Series Analysis yielded Mean Absolute Percentage Error (MAPE) values ranging from 0.086 to 13.969%. The ANN method demonstrated superior performance in flow estimation at the E13A031 gauging station, achieving a coefficient of determination (𝑅²) of 0.802, while an 𝑅² value of 0.88 was obtained for the 2018 flow estimation of the Ovacuma Stream. These results underscore the effectiveness of integrating hydrological investigations with machine learning approaches in supporting sustainable water resource management.