In the evolving field of water resource management, underground dams have garnered significant attention for their ability to enhance groundwater storage and secure water supply in regions vulnerable to drought and climate variability. A groundbreaking study conducted by researchers Ekemen Keskin and Eren Şander, recently published in Environmental Earth Sciences, delves into the innovative synergy of machine learning methodologies and hydrological modeling to estimate streamflow for an underground dam located in Bartın Bahçecik, Turkey. This pioneering research not only pushes the boundaries of traditional hydrological studies but also offers a scalable approach that could revolutionize water management practices globally.
Underground dams are critical infrastructures constructed beneath riverbeds or other permeable sediments to intercept and store subsurface flows. Unlike conventional surface dams, these subterranean barriers minimize evaporation losses and reduce ecological disruption, making them ideal for semi-arid and arid climates. However, assessing their effectiveness requires precise estimation of streamflow and groundwater recharge rates, which traditionally depends on extensive field measurements and complex hydrological modeling techniques. Keskin and Şander’s study ingeniously addresses these challenges by integrating machine learning algorithms with classical hydrological models to improve the accuracy of streamflow predictions while optimizing data utilization.
The Bartın Bahçecik underground dam offers a compelling case study due to its unique hydrogeological settings and the increasing water stress in the Black Sea region of Turkey. The researchers collected an extensive dataset encompassing precipitation, temperature, land use, soil characteristics, and streamflow records. They employed a suite of supervised machine learning models, including Random Forests, Support Vector Machines, and Gradient Boosting, to identify nonlinear relationships within the hydrological data that are often overlooked by traditional methods. This approach harnessed the power of pattern recognition and data-driven insights to supplement physical process-based models.
One of the key technical achievements of this work is the hybrid modeling framework proposed by Keskin and Şander. They used a conventional hydrological model, SWAT (Soil and Water Assessment Tool), to capture the basin-scale hydrological processes such as surface runoff, infiltration, and evapotranspiration. The residual errors and prediction uncertainties from the SWAT simulations were then addressed by the machine learning models, which learned from observational data to adjust the output streamflow estimates dynamically. This cascading model architecture significantly reduced bias and enhanced the predictive performance over the entire simulation period.
Moreover, the study presents a detailed sensitivity analysis, revealing which climatic and watershed parameters most influence streamflow variability and recharge potential in the underground dam’s catchment. Precipitation intensity and soil transmissivity emerged as dominant factors, underscoring the importance of local meteorological patterns and subsurface conditions. The authors also highlight the temporal resolution’s effect on model accuracy, demonstrating that daily data offers better granularity for streamflow estimation than monthly averages, a nuance critical for operational water resource planning.
Importantly, Keskin and Şander’s methodology underscores the value of machine learning not as a standalone tool but as a complementary enhancement to physically based hydrological models. In regions where ground truth data are sparse or expensive to obtain, this synergistic approach enables more robust estimates without sacrificing interpretability. The hybrid model’s adaptability and scalability mean it can be deployed in similar underground dam projects worldwide, particularly in developing countries facing water scarcity challenges.
The implications of accurate streamflow estimation extend beyond water storage. They influence ecosystem sustainability, agricultural planning, and disaster mitigation strategies. By improving the predictability of how underground dams modulate subsurface flows, this research paves the way for integrated water resource management frameworks that balance human use with environmental conservation. Furthermore, such predictive capabilities allow for real-time operational adjustments in dam management during extreme weather events, enhancing resilience in the face of climate change.
Another notable contribution of this work lies in its methodological transparency and replicability. The authors provide detailed model parameterizations, validation metrics, and the statistical techniques used to optimize machine learning hyperparameters. Their rigorous cross-validation and uncertainty quantification protocols set a high standard for future studies merging machine learning with traditional hydrological sciences. This rigor ensures that the reported improvements in streamflow estimation are both statistically significant and practically meaningful.
The Bartın Bahçecik case study also reveals practical insights into underground dam performance evaluation. The study indicates that while underground dams can substantially augment groundwater storage, their benefits are maximized when integrated with upstream watershed management practices. Maintaining vegetation cover and reducing land degradation in the catchment area substantially enhance recharge efficiency, as confirmed by the hybrid model’s simulation scenarios. These findings empower policymakers to adopt holistic watershed management strategies that synergize engineering solutions with ecological stewardship.
From a technological standpoint, the use of ensemble learning methods, which integrate predictions from multiple machine learning models, contributed to the robustness of the new framework. Ensemble approaches inherently reduce overfitting and handle noisy environmental data more effectively than individual algorithms. This advance is critical given the inherent variability and uncertainty in hydrological processes, particularly in regions with complex topography and heterogeneous soil conditions such as Bartın Bahçecik.
The research also acknowledges the limitations inherent in both modeling approaches. While the hybrid model substantially improved streamflow estimation accuracy, uncertainties remain due to unmeasured subsurface heterogeneities and data gaps in climatic records. The authors advocate for continued investment in sensor networks and remote sensing technologies to provide higher resolution data streams. They envision that coupling these real-time data with adaptive machine learning models will further elevate underground dam management capabilities.
This study is situated within a broader scientific discourse emphasizing the transformative potential of artificial intelligence in environmental modeling. By concretely demonstrating successful integration with hydrological simulation, Keskin and Şander contribute to a paradigm shift where data-driven and mechanistic models coalesce for better environmental decision-making. Such interdisciplinary innovations hold promise not only for water resource engineering but also for addressing global challenges like ecosystem degradation and sustainable agriculture.
In conclusion, the research led by Ekemen Keskin and Eren Şander represents a milestone in the application of AI-enhanced hydrology to subterranean water infrastructure. Their hybrid modeling framework delivers a powerful, scalable tool for accurately estimating streamflow, improving underground dam performance assessment, and informing water resource management under climate uncertainty. As groundwater depletion continues to threaten socio-economic stability worldwide, such innovative approaches could become indispensable in securing water sustainability for future generations.
Subject of Research: Streamflow estimation for underground dams using machine learning and hydrological modeling
Article Title: Streamflow estimation for underground dams using machine learning and hydrological modeling: a case study of Bartın Bahçecik underground dam
Article References:
Ekemen Keskin, T., Şander, E. Streamflow estimation for underground dams using machine learning and hydrological modeling: a case study of Bartın Bahçecik underground dam. Environ Earth Sci 84, 508 (2025). https://doi.org/10.1007/s12665-025-12511-x
Image Credits: AI Generated
