Application of machine learning methods to study the relationships between seismicity and geodynamic features of the Dnister Hydropower Complex

• Date: 2026
• Main Authors: Brusak, Ivan, Haidus, Oleg, Kuplovskyi, Bohdan
• Format: Article
• Language: English
• Published: S. Subbotin Institute of Geophysics of the NAS of Ukraine 2026
• Subjects: Seismicity; recent geodynamics; reservoir-triggered seismicity; water level changes; Dnister Hydro Power Complex; machine learning
• Online Access: https://journals.uran.ua/geofizicheskiy/article/view/350985
• Journal Title: Geofizicheskiy Zhurnal

This study investigates the seismicity and recent geodynamic features of the Dnister Hydropower Complex in Ukraine, emphasizing the application of machine learning methods to analyze their interrelationships. The complex, situated in a seismically active transitional zone, is influenced by natural tectonic processes and anthropogenic activities, including the operations of the Dnister Hydroelectric Power Plant and active water level changes at the Dnister Reservoir. Data from digital seismic stations of the Carpathian Seismological Network, permanent Global Navigation Satellite System stations of GeoTerrace and SystemNet networks, as well as reservoir water level records of Dnister Reservoir, were collected and analyzed together. Machine learning algorithms, including Random Forest, Isolation Forest, and DBSCAN clustering, were employed to identify patterns and correlations between crustal deformations, water level fluctuations, and seismic events. Results reveal a significant association between water level changes — both short-term and long-term — and earthquake occurrences, suggesting that hydrological variations impact seismic activity. Geodynamic analysis indicates heterogeneous deformation patterns, with increased velocities in seismically active southwestern regions. Global Navigation Satellite System data shows velocities increasing by about 2 mm/year near the Dnister Hydropower Complex. Seismicity near the Dnister Hydropower Complex from 2012 to 2023 was characterized by peak earthquake years of 2014—2016 and 2022, each with over 100 events. The total seismic energy released increased from lg(ΣE)=7.5 in 2012 to 10 in 2016, then steadily declined to 7 by 2023. The findings enhance understanding of the mechanisms of induced seismicity related to reservoir operations and provide valuable insights for risk assessment and mitigation strategies in hydroelectric regions. This integrated approach demonstrates the effectiveness of machine learning in deciphering complex geodynamic and seismic interactions in tectonically sensitive environments.