Monitoring the Transformation of an Aging Plain Reservoir: InSAR-Derived Deformation Associated with Dredging and Capacity Expansion at Suyahu Reservoir, China

Highlights

What are the main findings?

• Optical imagery and satellite altimetry documented the expanded inundation extent and stage-dependent water level variations during the dredging and capacity expansion project.
• InSAR revealed localized line-of-sight velocities reaching approximately −40 mm/a along the embankment and −50 mm/a along monitored artificial island margins. No statistically significant lagged correlations were found between embankment displacement and water level.

What are the implications of the main findings?

• Local excavation, fill placement, and construction histories provide a plausible first-order interpretation of the observed deformation.
• Time-series InSAR enables system-scale monitoring and supports targeted field inspection of aging plain reservoirs undergoing major engineering transformation.

Abstract

Reservoir sedimentation reduces operational performance, while large-scale dredging and capacity expansion redistribute earth materials and alter engineering conditions at aging plain reservoirs. This study integrates Sentinel-1 time-series Interferometric Synthetic Aperture Radar (InSAR), multitemporal optical imagery, and ICESat-2 and Sentinel-3A altimetry to characterize the engineering transformation and associated deformation at Suyahu Reservoir, China. Optical imagery documented land-based excavation, cofferdam construction, and subsequent reinundation. Under comparable instantaneous water levels, the mapped water surface area increased from approximately 72.06 km² in April 2020 to 87.45 km² in April 2025, an increase consistent with changes in reservoir-bed topography and shoreline morphology. Satellite altimetry showed that the mean water level decreased from about 52.62 m before construction to 52.15 m during the main construction period, while the mean annual observed range narrowed from 1.33 to 0.68 m before increasing again from mid-2023 onward. InSAR revealed spatially heterogeneous negative line-of-sight velocities reaching approximately −40 mm/a along localized embankment sections and −50 mm/a along the monitored margins of the artificial islands. Lagged cross-correlation analysis identified no statistically significant correlations between water level variations and the monitored embankment displacement. The spatial correspondence between deformation and project-modified areas supports a plausible first-order interpretation based on local excavation, fill placement, and construction histories.

1. Introduction

Reservoir sedimentation progressively reduces the storage capacity and operational performance of aging reservoirs, compromising their flood-control, water-supply, irrigation, and ecological functions [1,2,3]. Dredging can restore storage capacity and extend infrastructure service life, but large-scale projects also reshape reservoir-bed topography, redistribute substantial volumes of earth material, and create new engineered landforms. Such interventions may therefore transform the physical and engineering setting of an aging reservoir rather than merely restoring its lost storage capacity.

Plain reservoirs are broad, shallow storage systems constructed in low-relief terrain and commonly enclosed by long earthfill embankments [4]. Compared with valley reservoirs, they are typically characterized by large water surface areas, shallow water depths, low dams, and long embankments [5]. Their gentle bed gradients favor sediment deposition, while their extensive embankments and frequently alluvial foundations may exhibit spatially variable seepage and deformation. During rehabilitation, excavation may alter reservoir bathymetry and local hydraulic boundary conditions, whereas the placement of dredged or excavated material introduces new loads and fills that may undergo compaction and consolidation. The resulting reservoir system comprises mature hydraulic structures, a reshaped basin, and recently modified ground whose deformation may vary spatially.

Conventional dam safety monitoring relies primarily on leveling, settlement gauges, Global Navigation Satellite System (GNSS) observations, piezometers, seepage measurements, and visual inspections. These observations are essential for safety assessment but generally concentrate on selected structures or cross-sections. Interferometric Synthetic Aperture Radar (InSAR) provides repeated, spatially distributed measurements of line-of-sight (LOS) displacement [6,7,8]. Its sensitivity to small displacements, wide-area coverage, and regular repeat acquisitions make InSAR well suited to monitoring ground and infrastructure deformation [9,10,11].

Over the past two decades, InSAR has been increasingly applied to deformation monitoring across different types of dams and reservoir-related infrastructure. For concrete gravity and arch dams, Wang et al. [12], Liao et al. [13], Tavakkoliestahbanati et al. [14] and Ziemer et al. [15] used InSAR to characterize millimeter- to centimeter-scale deformation associated with thermal effects, reservoir-level variations, long-term structural responses, and, in some cases, pre-failure deformation. Jänichen et al. [16] further demonstrated the integration of InSAR with in situ measurements for operational dam monitoring.

Earth and rockfill dams generally exhibit larger and more persistent deformation associated with fill consolidation, foundation compression, seepage, and reservoir loading. Tomás et al. [17] reconstructed approximately 15 years of deformation at the La Pedrera earthfill dam and attributed the observed displacement primarily to long-term consolidation. Zhou et al. [18], Li et al. [19] and Xie et al. [20] further demonstrated the application of InSAR to post-construction settlement monitoring and highlighted the effects of radar geometry, surface conditions, and seasonal inundation on deformation measurement and spatial coverage. Milillo et al. [21] combined multi-sensor SAR observations to identify persistent subsidence at the Mosul Dam associated with dissolution of underlying evaporite layers and changes in maintenance grouting.

Long embankments surrounding plain reservoirs share several InSAR monitoring characteristics with levees and other elongated flood-defense structures, including extended linear geometry, spatially variable settlement, and heterogeneous surface conditions. Hanssen and van Leijen [22] demonstrated systematic PSI monitoring of Dutch flood defenses and found that hard surfaces provided reliable scatterers, whereas vegetated slopes were frequently affected by temporal decorrelation. Özer et al. [23] detected sub-seasonal swelling and shrinkage of canal levees associated with meteorological and hydrogeological variations, while Nguyen et al. [24] obtained InSAR deformation patterns consistent with in situ measurements along levees in New Orleans. Xiao et al. [25] and Xie et al. [26] independently investigated the Sardoba reservoir failure using InSAR and identified localized differential settlement near the breached section. Xiao et al. [27] reconstructed seven years of deformation along the 9.6 km long Shuangwangcheng reservoir embankment and identified long-term consolidation as the dominant process.

Previous studies have focused on dams as the primary monitoring objects. Comparatively little attention has been paid to system-scale deformation when an aging reservoir undergoes major sediment removal, basin reshaping, and excavated material redistribution. In particular, the spatial and temporal deformation of mature hydraulic structures relative to recently modified ground remains insufficiently documented. Suyahu Reservoir in central China provides a suitable case for investigating this issue [28]. The reservoir is enclosed by an exceptionally long earthfill embankment and has operated for more than six decades. Following substantial loss of effective storage through sediment accumulation, a large-scale dredging and capacity expansion project was initiated in 2019. The project reshaped parts of the reservoir basin and placed excavated materials in disposal and newly formed engineered areas, producing a spatially heterogeneous reservoir setting.

This study uses 128 ascending Sentinel-1 acquisitions, multitemporal optical imagery, ICESat-2 and Sentinel-3A satellite altimetry observations, and available engineering records to characterize LOS deformation at Suyahu Reservoir. It addresses three questions: (1) what LOS deformation patterns are observable along the long earthfill dam and adjacent project-modified ground? (2) Do representative dam sections and modified or filled areas exhibit different time-series behavior? (3) What temporal correspondence, if any, exists between the observed deformation and reservoir water level variations? By addressing these questions, this study extends structure-specific dam monitoring to a system-scale characterization of an aging plain reservoir undergoing and adjusting to large-scale engineering rehabilitation. The results provide remote sensing evidence for identifying areas that warrant further field inspection and illustrate both the potential and limitations of time-series InSAR for characterizing the deformation state of aging reservoirs following large-scale dredging and capacity expansion projects.

2. Study Area and Engineering Background

2.1. Reservoir Setting and Dam Characteristics

Located in Henan Province, central China, Suyahu Reservoir is reported to be the largest artificial plain reservoir in Asia. Flood control is its primary function, complemented by irrigation and other water-resource uses. The reservoir controls a drainage area of approximately 4498 km² and has a designed total storage capacity of about 1.64 billion m³ [28], making it an important flood-regulation reservoir on the main stem of the Huaihe River.

Construction of the reservoir began in 1958, and initial impoundment commenced in 1959. The principal water-retaining structure is a homogeneous earthfill dam approximately 35 km in length (Figure 1a), with multiple flood discharge and irrigation structures distributed along the dam alignment [28]. The exceptionally long earthfill dam and broad, shallow reservoir basin are characteristic of large reservoirs constructed in low-relief plain terrain.

The latest hazard-elimination and reinforcement project for Suyahu Reservoir was approved and implemented in 2010, involving dam strengthening, localized foundation seepage control, rehabilitation of hydraulic structures, and installation of safety monitoring facilities [29]. In addition to the original dam and associated hydraulic structures, this study also considers areas subsequently modified by large-scale dredging and capacity expansion works.

2.2. Long-Term Reservoir Sedimentation

Since operation began more than 60 years ago, Suyahu Reservoir has experienced more than 20 major flood events, resulting in nearly 100 million cubic meters of sediment accumulation, with a maximum deposit depth of 3 m. The average water depth at normal storage levels is less than 2 m [28]. The primary causes of this sedimentation are as follows: (1) loose soils, sparse vegetation, and pronounced soil erosion in parts of the upstream catchment generate high sediment loads. During major floods, especially the catastrophic August 1975 (“75.8”) flood, channel erosion, bank collapse, and farmland soil loss transported large quantities of sediment and debris into the reservoir. (2) Human activities further enhanced sediment delivery. Conversion of natural channels and adjacent land to farmland reduced vegetation cover and disturbed protective surface layers, increasing the exposure of soil and weathered materials to erosion. (3) The reservoir was not originally equipped with dedicated sediment-flushing or sediment-discharge facilities. Its low-relief plain setting, gentle bed slope, and reduced inflow velocity further favor sediment deposition. (4) Seasonal inflow variability and water scarcity limit discharge for sediment management, constraining operations that retain clear water while releasing sediment-laden flows.

These conditions have progressively reduced the effective storage capacity of the reservoir and created a strong engineering need for sediment removal and basin expansion. The subsequent intervention was therefore intended not only to remove accumulated sediment, but also to reshape parts of the reservoir basin and restore storage capacity lost during long-term operation.

2.3. Dredging and Capacity Expansion Project

The dredging and capacity expansion project at Suyahu Reservoir was initiated in 2019 and is reported as the first large-scale reservoir dredging project in China, with an approved total investment of approximately CNY 3.17 billion [28]. The project involved more than 95 million m³ of excavation within the existing reservoir basin and the western capacity expansion area. After cut-and-fill balancing, the project restored approximately 33.38 million m³ of dead storage and increased the operational storage capacity by approximately 54 million m³ [30].

Construction combined land-based excavation and underwater dredging, with cofferdams used to isolate selected working areas. These operations deepened parts of the reservoir basin and substantially reshaped its topography. The excavated materials were managed through three principal pathways (Figure 1a): (1) placement along the wetland patrol and flood-control road on the upstream side of the reservoir, (2) reuse within the reservoir to construct artificial islands, and (3) hydraulic transport to designated disposal areas. Specifically, excavated material placed along the patrol road formed a deposit approximately 2 m high and 120 m wide. Three artificial islands exceeding 40 m in height were constructed within the reservoir, and dredged sediment from the central reservoir area was conveyed through pressurized pipelines to a disposal area downstream of the dam. This disposal area extended approximately 300 m outward from the dam and approximately 14.5 km along the dam.

Consequently, the project produced an environment of pronounced spatial heterogeneity, in which the long-standing earthfill embankment and hydraulic structures coexist with recent excavation faces, material disposal zones, cofferdams, and newly placed fill bodies. Within the engineering context, this study analyzes the deformation phenomena and their spatial distribution characteristics revealed by InSAR time series observations from 2021 to 2025.

3. Materials and Methods

3.1. Datasets

3.1.1. SAR Data and Auxiliary Data

The SAR dataset used in this study comprised 128 C-band Sentinel-1 images acquired from an ascending orbit between January 2021 and May 2025. All images were acquired in the Interferometric Wide Swath (IW) mode using the Terrain Observation with Progressive Scans (TOPS) technique, with VV polarization, and were provided as single-look complex (SLC) products. The nominal revisit interval was 12 days, and the approximate ground resolutions were 5 m and 20 m in the range and azimuth directions, respectively. The spatial coverage of the Sentinel-1 observations is shown in Figure 1b.

Precise orbit ephemerides provided by the European Space Agency (ESA) were used to refine the satellite orbital information and reduce orbital errors during interferometric processing. In addition, a Copernicus Digital Elevation Model (DEM) with a spatial resolution of 30 m was employed to simulate and remove the topographic phase contribution and to support the geocoding of the InSAR results.

3.1.2. Satellite Altimetry Water Level Data

Reservoir water levels between April 2016 and June 2026 were jointly derived from ICESat-2 and Sentinel-3 satellite-altimetry observations. ICESat-2 carries the Advanced Topographic Laser Altimeter System (ATLAS), a photon-counting lidar that provides dense along-track surface-elevation measurements on a 91-day repeat orbit. ICESat-2 ground tracks intersecting the open water of Suyahu Reservoir were selected, and water surface elevation measurements were extracted from L3A Along Track Inland Surface Water Data (ATL13) [31].

Sentinel-3A carries a dual-frequency Ku- and C-band Synthetic Aperture Radar Altimeter (SRAL) operating on a 27-day repeat orbit [32]. SRAL determines the satellite-to-surface range from the two-way travel time of radar pulses, and the water surface elevation is obtained by subtracting this range and the relevant atmospheric and geophysical corrections from the satellite altitude. Water surface heights over the reservoir were extracted from the Level-2 SRAL land products, whose 20 Hz measurements are spaced approximately 300 m apart along track.

For both datasets, measurements that did not satisfy the product quality-control criteria or were located over the exposed reservoir bed, dam, islands, or other non-water surfaces were removed. After statistical outliers were excluded, the remaining measurements from each overpass were averaged to obtain a representative reservoir water level for that date. The selected ICESat-2 and Sentinel-3 ground tracks and representative along-track elevation profiles are shown in Figure 2.

Before merging, a common vertical reference was adopted for the two datasets. ICESat-2 elevations used in this study were already referenced to the EGM2008 geoid, while the Sentinel-3A elevations were transformed to the same reference using the provided geoid correction parameters. The inter-sensor offset was estimated from three pairs of ICESat-2 and Sentinel-3A observations acquired on the same day or within two days of each other. The mean difference, calculated as Sentinel-3A minus ICESat-2, was approximately −0.16 m. Using ICESat-2 as the reference, we therefore added 0.16 m to the Sentinel-3A elevations to correct this estimated mean offset. The two records were then merged into a combined altimetry time series. The combined record was used to characterize the reservoir water level regime and to evaluate its temporal correspondence with the InSAR displacement series as described in Section 3.3.

3.1.3. Optical Imagery for Construction Progress Mapping

Multitemporal Sentinel-2 optical imagery, supplemented by high-resolution imagery available through Google Earth, was used to reconstruct the spatial evolution of the dredging and capacity expansion project. Sentinel-2 true-color composites were generated from the 10 m resolution B4 (red), B3 (green), and B2 (blue) bands in an RGB configuration. The images were subjected to atmospheric correction, band compositing, study-area clipping, and contrast enhancement. Water boundaries were derived from the 10 m Dynamic World V1 classifications corresponding to the selected Sentinel-2 images, with additional visual checks around the artificial islands.

Visual interpretation of images acquired at different construction stages revealed changes in reservoir inundation, exposure of the reservoir bed, excavation in the western capacity expansion area, cofferdam construction, material placement, artificial island formation, and development of the downstream sediment-disposal area. The derived construction-progress information was compared with the InSAR results to assess whether deformation zones coincided with excavation areas, newly placed fills, material disposal areas, or existing hydraulic structures.

3.2. Time-Series InSAR Processing

The Sentinel-1 SLC images were preprocessed using GAMMA software (https://www.gamma-rs.ch/, accessed on 18 August 2026) including cropping to the area of interest and image co-registration. The acquisition obtained on 24 November 2021 was selected as the common co-registration reference, and all remaining acquisitions were co-registered to this image to ensure spatial consistency throughout the time series. A small-baseline strategy was adopted to construct the interferometric network, with pairs selected based on a maximum temporal baseline of 90 days. In total, the network comprised 625 interferograms. The distributions of the temporal and perpendicular baselines are shown in Figure 3.

Time-series processing was performed using the small baseline implementation of the StaMPS/MTI framework [33,34]. Candidate pixels were initially selected using an amplitude difference dispersion threshold of 0.6 and were subsequently refined through phase stability analysis. Pixels with an estimated phase standard deviation greater than 0.8 rad were excluded from further processing. These thresholds were selected to balance phase quality and measurement point density, ensuring sufficient spatial coverage of the monitoring targets. The wrapped phase was filtered using a Goldstein adaptive filter and then unwrapped in three dimensions. The unwrapped interferograms were checked for spatial discontinuities and temporal inconsistencies. Loop closure checks and visual inspection guided the identification of regions with clearly identifiable boundaries and unambiguous integer-cycle phase offsets. Manual correction was restricted to these regions and performed by adding the appropriate integer multiples of 2π. Interferograms containing complex or unresolved unwrapping errors were excluded from subsequent analysis.

Mean LOS velocities and displacement time series were estimated through least-squares inversion. The displacement estimates were referenced to the area considered stable during the monitoring period. Negative LOS displacement denotes motion away from the satellite, whereas a positive value denotes motion toward the satellite. To characterize the temporal uncertainty of the fitted linear velocities, an ordinary least squares regression was fitted to the cumulative displacement time series at each valid measurement point, and the standard error of the fitted slope was used to describe the scatter of the observations around the linear trend.

3.3. Correspondence Analysis Between Water Level and Dam Deformation

Lagged cross-correlation analysis was used to evaluate the relationship between satellite-derived reservoir water levels and InSAR-derived embankment deformation. Water level and InSAR displacement observations were aggregated into monthly time series by calculating the arithmetic mean of all available observations. Long-term trends were removed from both series, and residual autocorrelation was reduced through autoregressive pre-whitening.

Cross-correlation coefficients [35] were calculated for lags of 0-3 months, with water level variations occurring contemporaneously with or preceding displacement. This lag range was selected based on the previous experience in embankment dam deformation monitoring to examine potential responses within the same month and over the following three months. Statistical significance was assessed using circular shifts in the residual displacement series. The maximum absolute correlation coefficient within the tested lag range was used as the test statistic, thereby accounting for the selection of the strongest lag. The resulting point-level p values were subsequently adjusted for multiple comparisons using the Benjamini–Hochberg false-discovery-rate procedure [36].

4. Results

4.1. Engineering Transformation Revealed by Remote Sensing Imagery

Multi-temporal Sentinel-2 imagery reveals substantial changes in the surface water extent and shoreline configuration of Suyahu Reservoir associated with the dredging and capacity expansion project. Three representative Sentinel-2 images were selected to delineate the reservoir water boundaries, as shown in Figure 4.

During the initial phase of the project, on 3 April 2019, the mapped water surface area was approximately 77.14 km². By 12 April 2020, it had decreased to 72.06 km², representing the smallest extent among the selected images. Construction in the western expansion area involved cofferdams and land-based excavation and coincided with a reduction in the water surface area and extensive exposure of the reservoir bed.

The image acquired on 6 April 2025 shows the late-stage reinundation configuration of the reservoir. The mapped water surface area had expanded to approximately 87.45 km², representing increases of 15.39 km² relative to April 2020 and 10.31 km² relative to April 2019. Areas exposed or excavated during construction in the western expansion area had subsequently been inundated. The comparative imagery illustrates the overall evolution of the site, ranging from shoreline retreat and exposure of the reservoir bed during construction to the expansion of the inundated area in the project’s later stages. This transformation can be broadly divided into three stages: an initial construction stage represented by the 2019 image, a low-inundation and land-based excavation stage represented by the 2020 image, and a late-stage reinundation phase represented by the 2025 image.

An enlarged view of the western expansion area further illustrates the construction sequence within the project zone (Figure 5). In April 2020, most of the expansion area remained exposed and was divided by cofferdams, temporary access roads, and excavation surfaces, indicating that land-based excavation was still underway. By December 2021, excavation and terrain reshaping had progressed substantially, and reservoir water had begun to reoccupy parts of the previously enclosed construction areas. Although their degrees of separation from the western shoreline varied, the three artificial islands were clearly distinguishable from the surrounding construction areas.

Comparison of the connections between the islands and the shoreline suggests that cofferdam removal progressed from south to north. Artificial Island 1, the southernmost island, appears to have been separated from the surrounding construction area first and was already largely surrounded by water in the December 2021 image. Artificial Island 2 was separated later, whereas the cofferdams or temporary connections associated with Artificial Island 3, the northernmost island, remained in place for the longest period. By November 2023, a greater proportion of the former excavation area had been inundated, and the three islands had approached their final configurations. This asynchronous construction sequence implies that the islands experienced different durations of filling, compaction, exposure, and contact with reservoir water, providing important temporal context for the differences observed in their deformation time series.

4.2. Reservoir-Level Variations from Satellite Altimetry

Observations from Sentinel-3A and ICESat-2 reveal distinct changes in the water level regime of Suyahu Reservoir before, during, and after the main construction phase (Figure 6). Although ICESat-2 provided fewer measurements because of its lower sampling frequency over the reservoir, the two datasets show generally consistent variations when their observation times are close.

Before project implementation, Sentinel-3A observations from 2016 to 2018 indicated a mean water level of approximately 52.62 m and a mean annual observed range of approximately 1.33 m. During the main construction phase from 2019 to 2022, the mean water level decreased to approximately 52.15 m, approximately 0.47 m below the pre-project mean, while the mean annual observed range narrowed to about 0.68 m. Water levels during this period remained largely concentrated around 52 m, despite occasional short-term increases in 2020 and 2021. These lower and less variable water levels coincided with cofferdam construction in the western expansion area and the maintenance of exposed working areas required for land-based excavation.

From 2023 onward, both the water level and the magnitude of its fluctuations began to increase. By 2025, the intra-annual observed range had returned to the typical pre-project range. This change indicates a transition toward the more variable water level regime associated with later-stage reservoir operation.

The altimetry observations also provide useful context for the changes in surface water extent identified from the Sentinel-2 imagery (Figure 4). The images shown in Figure 4a–c were acquired on 3 April 2019, 12 April 2020, and 6 April 2025, respectively. The corresponding nearest altimetry observations were acquired on 31 March 2019, 12 April 2020, and 26 March 2025, recording water levels of 51.95 m, 51.87 m, and 51.99 m, respectively. Consequently, the larger inundated area mapped in April 2025 is unlikely to be explained solely by a higher instantaneous water level but is consistent with changes in reservoir bed and shoreline geometry resulting from excavation, capacity expansion, and subsequent reinundation. Quantifying the contribution of reservoir-bed topographic changes to the observed area increase would require bathymetric or elevation data.

4.3. Spatial Distribution and Uncertainty of LOS Deformation

The spatial distributions of the mean LOS velocity derived from InSAR and the formal standard error of the fitted velocity within the study area are shown in Figure 7a,b, respectively. Valid InSAR measurement points are unevenly distributed across the study area. Open water yields no coherent measurement points, whereas relatively dense measurements are obtained from surfaces with stable radar backscatter, including the dam crest, paved roads, artificial island margins, and residential areas. Although extensive ground surfaces were exposed in the western expansion area during construction, measurement points remain relatively sparse. This limited coverage is consistent with temporal decorrelation caused by frequent excavation, material stockpiling, and surface modification. Pronounced negative LOS velocities are concentrated primarily along sections of the eastern embankment, parts of the artificial island margins, and the built-up area southeast of the reservoir.

The reference point was located in a relatively stable area, with a mean LOS deformation rate of approximately −1 mm/a over the monitoring period. The eastern embankment exhibits relatively continuous linear measurement coverage. LOS velocities are predominantly negative, indicating motion away from the satellite. The velocity distribution along the dam axis is spatially heterogeneous, with sections exhibiting near-zero velocities alternating with localized areas of pronounced negative motion. In some sections, negative LOS velocities reach approximately −40 mm/a, indicating clear along-dam segmentation and localized deformation concentrations.

On the three artificial islands, valid measurement points are concentrated mainly along perimeter roads and marginal structures, whereas coverage within the island interiors is limited. The monitored island margins are characterized primarily by negative LOS velocities, with the most negative values reaching approximately −50 mm/a. Velocity differences are observed both among the three islands and between different margins of the same island.

The negative LOS velocities in the built-up area southeast of the reservoir may be associated with local building construction, as suggested by the temporal correspondence between redevelopment and the observed deformation. A direct link between this urban deformation and the reservoir dredging and capacity expansion project has not been established. Therefore, a detailed investigation of this urban deformation is beyond the scope of the present study.

Based on these spatial patterns, representative points along the embankment and artificial islands were selected for the displacement time-series analysis presented in Section 4.4. The standard error distribution shown in Figure 7b indicates low fitted-rate uncertainty for most coherent targets.

4.4. Displacement Time Series of the Dam and Artificial Islands

Representative points were selected from valid InSAR measurements based on their deformation behavior and engineering locations. The selection included both pronounced deformation and relatively stable behavior and covered key engineering features to illustrate spatial differences across the monitored areas. These points are distributed along the embankment, at the dock and in the adjacent sediment disposal zone, and along the margins of the three artificial islands.

4.4.1. Dam and Associated Engineering Features

Eight representative monitoring points (D01-D08) were selected along the eastern embankment to characterize distinct temporal deformation patterns (Figure 8). Among them, D01, D02, D04, and D08 exhibited the largest cumulative negative LOS displacements. D01 and D04 showed nearly continuous negative displacement between 2021 and the second half of 2023, followed by a marked reduction in deformation rate. By the end of the monitoring period, both points had accumulated approximately −110 mm of LOS displacement. D02 and D08 exhibited larger and more persistent negative displacements, reaching approximately −150 and −140 mm, respectively. Despite short-term fluctuations, their time series were dominated by sustained motion away from the satellite. All four points are located along embankment sections adjacent to areas affected by material placement or surface modification.

In contrast, D03 and D07 exhibited the smallest net displacements among the selected embankment points. The D07 time series fluctuated within a narrow range and returned to approximately its initial displacement level by the end of the monitoring period. D03 showed a small long-term negative displacement and several short-term fluctuations, with a final cumulative displacement of approximately −10 mm. The contrasting behavior of these two points relative to the more strongly deforming locations indicates that the observed deformation was localized rather than uniformly distributed along the entire embankment.

D05 and D06 exhibited a distinct three-stage pattern. Both points remained relatively stable during the initial monitoring period but entered a phase of accelerated negative displacement beginning in the second half of 2022. The cumulative displacement reached approximately −60 mm at D05 and −40 mm at D06. From late 2023 onward, the deformation rates at both points decreased markedly, and their cumulative displacement curves remained relatively stable. These time series therefore show a transition from an initially stable stage to a period of pronounced negative displacement and finally to a stage of relative stability.

A more detailed analysis was conducted in the dock area, where points P01-P06 are distributed across the dock structure, the embankment, and the disposal area behind the dam (Figure 9). All six points exhibited cumulative negative LOS displacement, although the magnitudes varied markedly over short distances. P02, located on the embankment near the dock, recorded the largest displacement, reaching approximately −180 mm. P03, situated along the margin of the disposal zone behind the dam, accumulated approximately −130 mm. P04 and P06 exhibited intermediate cumulative displacements of approximately −80 and −90 mm, respectively, whereas P01 on the dock structure and P05 on the embankment showed smaller displacements of approximately −65 and −35 mm, respectively. The pronounced differences among the points on the dock, embankment, and disposal area further indicate that deformation in this section of the reservoir infrastructure was localized and spatially heterogeneous.

4.4.2. Newly Filled Artificial Islands

Excavated soil and rock materials from the western capacity expansion area of Suyahu Reservoir were air-dried, placed as fill, and compacted on site to construct three artificial islands. Multitemporal optical imagery showed that the islands underwent repeated fill placement and surface modification as the capacity expansion project progressed. These disturbances resulted in pronounced temporal decorrelation within the island interiors. Consequently, valid InSAR measurement points were concentrated primarily along the island margins, particularly on the perimeter roads. Two representative InSAR measurement points were selected on each artificial island, and their cumulative LOS displacement time series are shown in Figure 10.

Among the six selected points, AI11 on Artificial Island 1 exhibited the largest cumulative LOS displacement. Its displacement became progressively more negative in a nearly linear manner throughout the monitoring period, ultimately reaching approximately −230 mm. In contrast, AI12 on the same island exhibited relatively minor displacement and remained comparatively stable after 2023, with a final cumulative displacement of approximately −15 mm.

AI21 on Artificial Island 2 showed limited displacement during the initial monitoring period but developed a more pronounced negative displacement trend beginning in the second half of 2022. By the end of the monitoring period, its cumulative displacement had reached approximately −130 mm. AI22 exhibited substantially smaller displacement, with a final cumulative negative LOS displacement of approximately −20 mm.

On Artificial Island 3, AI31 exhibited continuous, nearly linear negative displacement, reaching approximately −170 mm by the end of the monitoring period. AI32 displayed a more moderate, stage-dependent pattern, with its cumulative displacement reaching approximately −70 mm. Although its deformation rate decreased during the later monitoring period, the displacement had not fully stabilized.

The paired time series reveal marked differences not only among the three islands but also between different margins of the same island. Construction-related temporal decorrelation limited InSAR coverage within the island interiors. Consequently, the available measurements mainly characterize selected sections of the perimeter roads and marginal structures and do not fully represent the deformation behavior of the entire island fill bodies. As surface conditions stabilize following project completion, the stability of the filled slopes will require particular attention.

5. Discussion

5.1. Engineering Interpretation of Contrasting Deformation Patterns

The displacement time series presented in Section 4 reveal contrasting deformation patterns along the embankment and across the three newly constructed artificial islands. Along the embankment, D01, D02, D04, and D08 exhibited pronounced and persistent LOS displacement, whereas D03 and D07 remained relatively stable and D05 and D06 showed stage-dependent deformation. Marked spatial differences were also observed over short distances in the dock area, where the cumulative displacements at P02 and P03 were substantially larger than those at P01 and P05.

The dredging and capacity expansion project involved the large-scale excavation and redistribution of soil and sediment. Part of the excavated material was placed in disposal areas behind the dam, which may have introduced additional loading conditions to the existing ground. Such loading may have contributed to the compression of the newly placed fill and consolidation of the underlying soil layers, particularly where the material has a high-water content, variable compaction, or substantial fill thickness. Dredging and reservoir-bed excavation may also have modified local seepage paths, hydraulic gradients, and soil stress conditions. Moreover, over the 35 km long embankment, spatial variations in embankment materials, foundation conditions, and construction history may have contributed to a heterogeneous deformation response.

Although the Suyahu Reservoir embankment has been in service for approximately 60 years, the major earthworks conducted nearby may have introduced new loading histories and boundary conditions. Consequently, even if long-term consolidation had previously approached stability, sections adjacent to disposal areas may have undergone renewed localized adjustment. The spatial correspondence between the observed deformation and local construction or material-placement areas is consistent with this interpretation. In particular, the pronounced LOS displacements at P02 and P03, in contrast to the smaller displacements at nearby points, are more consistent with a localized response to heterogeneous loading than with uniform movement of the entire embankment. Similarly, the stage-dependent deformation at D05 and D06 may be associated with an interval of construction-related disturbance followed by post-construction adjustment as the loading and surface conditions became more stable.

The artificial islands represent a different engineering setting because they were constructed from excavated soil and rock and underwent repeated fill placement and compaction during the capacity expansion project. Sustained loading may have induced self-weight consolidation of the fill and compression of the underlying reservoir-bed deposits. Variations in fill thickness, material composition, water content, placement history, compaction quality, and interaction with reservoir water may have resulted in strongly heterogeneous deformation. This interpretation is consistent with the marked differences between paired points on the same island. AI11, AI21, and AI31 exhibited substantial and persistent negative LOS displacement, whereas AI12 and AI22 remained comparatively stable and AI32 showed a more moderate, stage-dependent response.

Overall, the observations indicate two contrasting deformation settings: localized adjustment of the mature embankment and adjacent ground potentially associated with construction-related disturbance, and continued deformation along selected margins of the newly constructed island fills. During the later monitoring period, deformation rates at several embankment points decreased markedly, whereas the more strongly deforming island-margin points had not reached a comparable level of stability. Local construction and material-placement histories therefore provide a more plausible first-order explanation for the observed spatial variability than a spatially uniform response to reservoir water-level loading, although secondary hydrological influences cannot be excluded.

5.2. Limited Correspondence Between Dam Deformation and Reservoir Water Level

Using points D01-D08, which represent the main deformation patterns along the embankment, the relationship between reservoir water level variations and deformation was analyzed. These points exhibited different deformation histories but were subject to the same reservoir water level variations. If reservoir level loading were the dominant control, the response amplitudes could differ among embankment sections because of variations in local geometry, material properties, and foundation conditions, but broadly consistent temporal responses and lag relationships would be expected. However, the selected points exhibited distinctly different long-term and stage-dependent displacement patterns.

The maximum absolute cross-correlation coefficients between the detrended and pre-whitened monthly series ranged from 0.19 to 0.32. None of the point-level correlations was statistically significant in the global circular-shift tests, with all global p values exceeding 0.05 even before correction for multiple comparisons across the eight monitoring points. The Benjamini–Hochberg-adjusted q values ranged from 0.41 to 0.69, providing no evidence of statistically robust lagged correlations. The heterogeneous deformation histories and absence of statistically robust lagged correlations provide no strong evidence that reservoir water level fluctuations were the dominant control on the monitored embankment deformation during the study period. An additional analysis using a 0–6-month lag window likewise identified no statistically significant lagged correlations. However, more subtle or complex water level influences cannot be excluded, given the temporal sampling and number of available altimetry and SAR observations.

5.3. Implications for Monitoring Aging Plain Reservoirs

For aging reservoirs with low dams and limited water level fluctuations, long-term consolidation of the original dam body may be assumed to be largely complete after several decades of operation, and the remaining deformation is therefore often expected to be small and relatively stable. However, the InSAR observations at Suyahu Reservoir revealed localized negative LOS displacement during the construction and subsequent adjustment periods of the dredging and capacity expansion project. Negative LOS velocities reached approximately −40 mm/a along some embankment sections and approximately −50 mm/a along monitored margins of the newly constructed artificial islands. The dam, previously considered to be stable, thus entered a new phase of active deformation potentially associated with engineering activities external to the dam itself.

The observed deformation patterns show spatial correspondence with the excavation and material placement configuration of the project. Reservoir bed excavation may have modified local hydraulic boundary conditions, depending on the subsurface materials and hydraulic connectivity. Meanwhile, the placement of large volumes of dredged sediment and excavated earth introduced new loads on disposal areas and the underlying ground. Persistent negative LOS displacement along the monitored margins of the artificial islands is also consistent with continued post-placement adjustment and consolidation of recently deposited fill. Although these observations do not indicate a risk of instability for the entire dam structure, they confirm that even when operational conditions appear stable, localized deformation may still occur following major engineering interventions.

This finding offers practical insights for numerous aging reservoirs that may face similar engineering interventions in the future. First, large-scale engineering works inside or around a reservoir may alter the loading and drainage conditions surrounding existing structures and disturb their previous deformation regime. Engineering activities should be regarded as potential contributors to deformation, and particular attention should be paid to dam sections adjacent to spoil areas, material disposal areas, or excavation zones. Second, time-series InSAR provides an effective system-scale tool for this type of monitoring. With its dense spatial sampling, wide area coverage, and non-contact nature, it is well suited to the routine monitoring of long dams in plain reservoirs; moreover, the availability of archived data supports retrospective analysis of conditions before and during project implementation.

For reservoirs where similar dredging projects are planned, it is recommended that an InSAR baseline monitoring system be established prior to construction, and that newly constructed fill structures (such as artificial islands and spoil areas) be incorporated into the monitoring plan. Particular attention should be directed toward the more strongly deforming island margins, which had not reached a clear state of stabilization by the end of the observation period. InSAR observations should ultimately be integrated with leveling, GNSS, piezometric measurements, and geotechnical inspections to evaluate the engineering significance of the detected deformation.

5.4. Limitations and Future Work

This study used a single ascending-track Sentinel-1 observation geometry. Consequently, the derived deformation reflects only the projection of the three-dimensional displacement onto the line of sight and cannot be uniquely decomposed into vertical and horizontal components. Combined ascending- and descending-track observations could be used to decompose the LOS measurements into vertical and east–west components, although the north–south component would remain poorly constrained [11,37]. Given that the main embankment of the Suyahu Reservoir runs roughly north–south, horizontal displacement perpendicular to the dam axis is oriented east–west. The well-known insensitivity of InSAR to north–south displacements therefore poses a limited constraint for this particular structure. Temporal decorrelation restricted the available measurements on the artificial islands mainly to their margins, limiting the extent to which these observations can reflect deformation within the island interiors.

Upcoming SAR missions and the continued expansion of satellite constellations are expected to provide shorter revisit intervals, higher spatial resolution, and more diverse observation geometries. With its broad spatial coverage, dense measurement sampling, and repeat-pass observation capability, time-series InSAR is particularly well suited to the long-term monitoring of long earth and rockfill dams and other aging structures common to plain reservoirs.

6. Conclusions

This study integrates Sentinel-1 time-series InSAR, multitemporal optical imagery, and ICESat-2 and Sentinel-3A altimetry data to characterize the engineering transformation and associated deformation patterns of Suyahu Reservoir. The findings demonstrate the value of time-series InSAR for the system-scale monitoring of aging plain reservoirs undergoing large-scale engineering transformation. The main conclusions are as follows:

• Multitemporal optical imagery documented the evolution of the western capacity expansion area from land-based excavation and cofferdam construction to reinundation. With similar water levels recorded by the nearest available altimetry observations, the mapped water-surface area increased from 72.06 km² in April 2020 to 87.45 km² in April 2025. This increase is consistent with changes in reservoir-bed topography and shoreline morphology and is unlikely to be explained solely by differences in water level.
• Satellite altimetry observations revealed distinct changes in the reservoir water level regime across different engineering stages. The mean water level decreased from approximately 52.62 m before project implementation (2016–2018) to 52.15 m during the main construction period (2019–2022), while the mean annual observed range narrowed from approximately 1.33 m to 0.68 m. From 2023 onward, both the water level and the magnitude of its fluctuations increased, and by 2025 the intra-annual observed range had returned to a level comparable to that observed before construction.
• The InSAR results revealed pronounced spatial heterogeneity in the observed deformation. Negative LOS velocities reached approximately −40 mm/a along localized sections of the existing embankment and approximately −50 mm/a along the monitored margins of the newly constructed artificial islands. Representative embankment points exhibited persistent, stage-dependent, and near-stable displacement patterns, indicating that the deformation was localized rather than uniformly distributed. Marked differences were also observed between different margins of the same artificial island, and some of the more strongly deforming island-margin points showed no clear tendency toward stabilization by the end of the monitoring period.
• Lagged cross-correlation analysis provided no statistically robust evidence that reservoir water level fluctuations were the dominant control on the monitored embankment deformation. Instead, the spatial correspondence between the observed deformation and project-modified areas suggests that local excavation, fill placement, and construction histories provide a more plausible first-order interpretation, although secondary hydrological influences and other mechanisms cannot be excluded.