Pre-collapse spaceborne deformation monitoring of the Kakhovka dam, Ukraine, from 2017 to 2023

The Kakhovka Dam, completed in 1956 as the final dam in the Dnieper cascade, is critical for hydroelectric generation, irrigation, navigation, and water supply for the Zaporizhzhia Nuclear Power Plant. During the 2022–2023 conflict, satellite imagery revealed abnormal operations and damage, culminating in the dam’s collapse on June 6, 2023. The cause has been debated, with reports suggesting an explosion; other factors such as earlier damage, extreme water levels, and operational anomalies have been discussed. This study aims to determine whether precursory structural deformations were detectable prior to failure using spaceborne MT-InSAR, quantify their timing and magnitude, and relate them to environmental drivers (e.g., water level) after removing seasonal thermal effects.

Remote sensing, and specifically MT-InSAR, has been increasingly used in conflict zones and for monitoring critical infrastructure, including bridges, levees, and dams. Prior work in Ukraine showed seasonal thermal deformation patterns on dam walls (horizontal motion toward the reservoir and vertical uplift in warm months, reversing in cold months). Studies have demonstrated MT-InSAR’s effectiveness in detecting precursory deformations before catastrophic dam failures (e.g., Brumadinho) and monitoring large dams (e.g., Three Gorges, Pertusillo). These findings motivate its application to the Kakhovka Dam to identify anomalous trends beyond thermal dilation.

Datasets: Two Sentinel-1 stacks were analyzed: (1) ascending track (IW3, orbit 14) with 180 images (Sentinel-1A/B) from July 2017 to June 2023; look angle 33.79°, heading −170.82°. (2) descending track (IW1, orbit 65) with 226 images (Sentinel-1A/B) from November 2015 to June 2023; look angle 43.66°, heading −10.82°. A SRTM 30 m DEM was used for topographic phase removal. A stable reference point (lat 46.7668, lon 33.3721) was used for both tracks. Processing: MT-InSAR was performed in SARPROZ using a persistent scatterer approach. A star-graph connection was adopted; PS candidates were selected via the amplitude stability index (ASI) with a threshold of 0.7. Interferometric phases were unwrapped and modeled using a five-degree polynomial for deformation while estimating residual height via periodogram optimization. Atmospheric phase screen effects were neglected due to the small spatial extent. Thermal detrending and hydrologic correlation: To isolate non-thermal deformation, LOS time series were regressed against weather temperature (Visual Crossing). If the correlation exceeded 0.7, the estimated thermal component was subtracted. Hydrologic influence was assessed by correlating PDW (perpendicular-to-dam-walls) deformation with Hydroweb reservoir water levels. Decomposition: Ascending and descending LOS measurements were inverted to retrieve vertical (up–down) and east–west (approximated here as PDW relative to dam orientation) components using track geometry (look and heading angles). This enabled quantification of vertical subsidence and upstream-directed PDW motion over defined polygons on the dam (I–V).

• Spatial patterns: Polygons I and III remained largely stable over the observation period. Polygons II, IV, and V showed pronounced deformation, with activation beginning June 2021.
• Rates (LOS, June 23, 2021–May 28, 2023): Polygon II: −21.2±1.3 mm/yr (ASC), −15.0±1.3 mm/yr (DSC). Polygon IV: −22.0±0.8 mm/yr (ASC), −8.9±1.3 mm/yr (DSC). Polygon V: −8.3±0.7 mm/yr (ASC), −5.5±1.3 mm/yr (DSC). Polygons I and III showed near-zero trends over the full period (e.g., Polygon III: −0.7±0.2 mm/yr ASC; −0.4±0.1 mm/yr DSC).
• Decomposition (June 1, 2021–May 28, 2023): Polygon II: vertical −22.7±0.3 mm/yr (down), PDW +28.3±0.3 mm/yr (upstream). Polygon IV: vertical −20.3±0.3 mm/yr, PDW +21.3±0.2 mm/yr. Polygon V: vertical −8.8±0.1 mm/yr, PDW +5.3±0.1 mm/yr. Polygon III: vertical −1.2±0.1 mm/yr, PDW ~0.0±0.2 mm/yr.
• Temporal behavior: Polygon V shows a pre-2021 steady subsidence of ~−2.4±0.2 mm/yr, accelerating ~4× to ~−8.8 mm/yr after June 2021. Polygons II and IV exhibit accelerating subsidence beginning June 2021 and upstream PDW motion beginning June 2022.
• Hydrologic coupling: Negative correlation between water level and PDW deformation for polygons II and IV (r ≈ −0.57 and −0.62) indicates increased upstream motion with decreasing water level, with the correlation persisting up to five days before collapse. Other dam segments near II and IV show similar correlations, while the road segment between II and III does not, likely due to measurement spatial distribution.
• Overall, deformation hotspots are concentrated on the southern side of dam segments that later breached (polygons II and IV), with detectable precursory motions up to two years before collapse.

After removing seasonal thermal dilation, clear non-thermal deformation signals emerge. From June 2021, subsidence accelerates over key dam segments (II and IV) that later failed, and from June 2022 upstream PDW motion intensifies in correlation with reservoir level changes, indicating hydrologically driven lateral responses. A nearby facility (polygon V) experienced a fourfold acceleration in subsidence beginning June 2021, while the hydroelectric plant area (polygon III) stayed largely steady in 2017–2021 despite historical subsidence in earlier decades. The combined vertical and PDW displacements are compatible with several damage mechanisms, including effects of overtopping, sediment and debris accumulation, faulty gate operations, and potential lack of maintenance during the war, which could have degraded foundations and structural integrity. While the immediate cause may involve an explosion on June 6, 2023, the InSAR evidence indicates that the structure had been undergoing significant deformations well before failure, suggesting pre-existing distress that could have reduced resilience. The findings highlight InSAR’s value for forensic interpretation and proactive monitoring of large hydraulic structures.

Multi-temporal SAR interferometry detected and characterized precursory deformation of the Kakhovka Dam prior to its June 2023 collapse. Key signals include subsidence starting in June 2021 and upstream (PDW) lateral motion from June 2022, with the latter correlated to reservoir water level changes. Deformation hotspots coincide with subsequently breached segments. These observations are consistent with potential mechanisms such as overtopping, sediment/debris buildup, and faulty gate operation, possibly exacerbated by insufficient maintenance during conflict. The study demonstrates the effectiveness of InSAR for long-term, high-temporal-resolution monitoring of critical infrastructure, supporting risk assessment and management. Future work could integrate higher-resolution SAR, in-situ instruments (e.g., GNSS), and structural modeling to refine mechanism attribution and thresholds for early warning.

• InSAR geometry limitations: Reduced sensitivity to motion along the satellite flight (north–south) direction and to certain vertical/horizontal components depending on look angle; decomposition relies on two geometries and assumes negligible along-track motion.
• Atmospheric and model assumptions: APS was considered negligible over the small area; a 5th-degree polynomial deformation model and thermal detrending thresholds may introduce modeling biases.
• Measurement coverage: Persistent scatterers are unevenly distributed (e.g., road segment between polygons II and III), potentially limiting local correlation analyses.
• Causality: InSAR cannot directly determine the collapse trigger (e.g., explosion) but identifies pre-existing deformation patterns and correlations.
• Data timeframe: Sentinel-1 acquisitions end in late May/early June 2023; no ultra-high-cadence data immediately preceding the collapse are available beyond those dates.