Dam Deformation Monitoring at Jatiluhur Dam, Indonesia, Using Multi-Temporal Synthetic Aperture Radar Interferometry and Integrated Field Observations

Highlights

What are the main findings?

• TW-PSI mitigates tropical decorrelation, increasing on-structure measurement density by 40% compared with conventional PSI for dense dam crest monitoring.
• The integrated InSAR–leveling comparison reveals a persistent central settlement bowl around C7–C12, with strong agreement between TW-PSI-derived quasi-vertical deformation and long-term leveling.

What are the implications of the main findings?

• The combined use of TW-PSI, SBAS, leveling, GNSS, and reservoir water-level data provides operational support for dam deformation monitoring, demonstrating how satellite geodesy complements, rather than replaces, in situ observations.
• ASDTR effectively prioritizes critical monitoring zones (C7, C7B, C8), distinguishing long-term hydro-mechanical deformation from co-seismic impacts.

Abstract

Monitoring dam deformation is critical for ensuring structural integrity and identifying long-term settlement trends. However, traditional InSAR techniques often face limitations in tropical environments due to severe temporal decorrelation. This study addresses these challenges at Jatiluhur Dam, Indonesia, by implementing an integrated framework using Sentinel-1 InSAR, in situ leveling, GNSS, and reservoir water-level data from 2019 to 2024. To overcome the observation bottlenecks, Tracy–Widom-guided PSI (TW-PSI) was employed and compared against SBAS and conventional PSI. The TW-PSI approach successfully increased on-structure measurement point density by approximately 40%, supporting a first-order ascending–descending decomposition into east–west and quasi-vertical components. The analysis reveals a persistent settlement bowl at the central crest (C7–C12), consistent with long-term leveling observations and supported by regional GNSS trend checking. While the 2022 Mw 5.6 Cianjur earthquake showed no statistically significant co-seismic crest deformation, a strong correlation (r = −0.709) was identified between crest deformation and reservoir water-level variations, suggesting an observational association between reservoir level and crest settlement tendency. Furthermore, the application of the Annual Structural Deformation Tolerance Ratio (ASDTR) identified specific priority monitoring zones. These findings demonstrate that the proposed integrated framework can support operational dam deformation monitoring by linking satellite-derived measurements with in situ observations and engineering-oriented interpretation.

1. Introduction

Dams serve as critical infrastructure for water resource management, flood mitigation, irrigation, and clean energy generation [1,2,3,4,5]. Despite their benefits, dams remain exposed to deformation and instability driven by geological conditions and natural hazards, including earthquakes [6,7], landslides [8], and soil liquefaction [9], as well as construction defects, material aging, operational changes, and inadequate maintenance practices [10,11,12,13,14]. Recent failures, including the 2009 Situ Gintung Dam failure in Indonesia [15], the 2023 Derna dam failures in Libya [16,17], and the 2024 Arbaat Dam failure in Sudan [18], highlight the need for continuous structural health monitoring to support early detection of abnormal deformation and improve dam safety management [19,20,21,22].

Interferometric Synthetic Aperture Radar (InSAR) has become a key technique for monitoring infrastructure deformation [23], including dams [24,25], bridges [26,27], railways [28,29], mining sites [30,31], and airports [32,33]. InSAR provides dense spatial sampling and millimeter-scale sensitivity over wide areas, making it suitable for observing subtle deformation of large or difficult-to-access hydraulic structures [34]. For dam applications, time-series analysis is essential to distinguish long-term settlement, seasonal or reservoir-related fluctuations, and possible event-related perturbations, rather than relying solely on single-epoch displacement [35]. Foundational processing algorithms such as Small Baseline Subset (SBAS) and Persistent Scatterer Interferometry (PSI) have been widely applied [36,37,38]; SBAS is commonly used to retrieve spatially continuous deformation over distributed targets, whereas PSI is better suited to stable scatterers on engineered structures [39]. Recent studies further integrate ascending and descending orbit geometries to derive first-order quasi-vertical and east–west deformation components, thereby improving the physical interpretability of InSAR time series while acknowledging the limited sensitivity to north–south motion [40].

Jatiluhur Dam, the oldest and largest concrete-faced rockfill dam in Indonesia, faces ongoing challenges related to sedimentation and land use changes around its reservoir [41]. The dam is located in a tectonically active region of West Java, where seismic loading represents an additional external factor for structural monitoring [42]. A magnitude 5.6 earthquake occurred near Cianjur in November 2022 [43,44], approximately forty kilometers from the dam, providing an event reference for evaluating whether measurable deformation perturbations occurred during the monitoring period [45]. Although terrestrial laser scanning surveys have been conducted, they have not provided periodic deformation records. Routine deformation monitoring primarily relies on leveling techniques, while many Indonesian dams still lack continuous on-structure GNSS instrumentation, leaving gaps in spatially continuous and temporally repeated observation. These limitations motivate the use of multi-temporal InSAR as a complementary monitoring tool for measuring spatially distributed dam deformation.

This study applies a dual-orbit Sentinel-1 time-series framework to measure and interpret deformation at Jatiluhur Dam from 2019 to 2024, with the 2022 Cianjur earthquake treated as an event reference within a longer-term deformation monitoring context. Because conventional PSI may lose reliable measurement points in coherence-limited dam environments affected by vegetation, moisture variability, mixed scattering, and residual noise, the workflow includes an enhanced Persistent Scatterer InSAR approach [46] that uses Tracy–Widom-guided eigenvalue screening to improve signal isolation and point selection in low-coherence dam regions [47]. Compared with SBAS, which is more suitable for broad spatial deformation patterns, the enhanced PS approach increases on-structure sampling over stable engineered targets and improves the observability of localized crest deformation [48]. The dual-orbit configuration enables first-order decomposition of LOS displacement into quasi-vertical and east–west components, supporting the interpretation of settlement-dominated deformation along critical dam segments [49,50]. To translate deformation rates into an engineering-oriented monitoring quantity, the Annual Structural Deformation Tolerance Ratio (ASDTR) is used to compare measured quasi-vertical deformation with an adopted screening-level tolerance. Reservoir water-level observations are analyzed independently to evaluate hydro-mechanical controls on the observed deformation, while ASDTR is used as the primary tolerance-normalized monitoring-priority indicator [51]. Earlier InSAR analyses at Jatiluhur suggest that deformation gradients are concentrated near the crest [52], while basin-wide responses are influenced by hydrological loading [53], reinforcing the need to integrate InSAR time series with reservoir operation data for dam deformation monitoring [54,55].

2. Materials and Methods

2.1. Study Area and Structural Context

Jatiluhur Dam is located on the Citarum River in Jatiluhur District, Purwakarta Regency, West Java, Indonesia, about 84 km southeast of Jakarta and roughly 60 km northwest of Bandung, as shown in Figure 1. It is also known as Juanda Dam, with construction beginning in 1957 and inauguration in 1967 [56]. The dam is a concrete-faced rockfill structure with an inclined clay core. Based on technical specification documents from Perum Jasa Tirta II [57], the crest elevation is 114.50 m, the maximum operating reservoir level is 111.60 m, the normal reservoir level is 107.00 m, and the minimum operating reservoir level is 87.50 m. The resulting design freeboard is therefore 2.90 m, defined as the difference between the crest elevation and the maximum operating reservoir level. The main spillway is a tower-type morning glory structure, and its lower section also serves as the powerhouse, with hydropower facilities arranged around the base of the morning glory structure [58].

The dam regulates the Citarum River and its tributaries and supports potable water supply, irrigation, hydropower generation, and flood control for West Java and Jakarta. Jatiluhur is the downstream component of the Citarum cascade, with Saguling upstream and Cirata midstream [45]. The catchment lies in a humid tropical climate consistent with Indonesia’s monsoonal seasonality, and the reservoir supports multiple and sometimes competing demands across water, energy, food, and land systems [41]. The reservoir also supports extensive floating net cage aquaculture. Water quality assessments classify the reservoir as moderately polluted and estimate substantial nitrogen and phosphorus loads attributable to aquaculture at prevailing cage counts. Management recommendations include limiting total cage area to about one percent of the reservoir surface to protect its primary functions [56,59].

2.2. Datasets

We used Sentinel-1 synthetic aperture radar imagery from the European Space Agency to build InSAR-based time series over Jatiluhur Dam. From January 2019 to December 2024, we collected 341 scenes consisting of 185 ascending and 156 descending acquisitions. All images were acquired in Interferometric Wide mode with VV polarization and a consistent 250 × 250 km frame. Precise orbit files from the ESA auxiliary repository were applied to improve interferogram geometry.

Daily reservoir water-level observations for 2019–2024 were provided by Perum Jasa Tirta II, which operates Jatiluhur Dam and maintains the official reservoir operation records. These data were used to evaluate the relationship between reservoir loading and InSAR-derived deformation. Regular in situ leveling measurements at the dam crest were also obtained from Perum Jasa Tirta II and used for structural comparison with the InSAR-derived crest deformation pattern.

For independent regional trend checking, we used GNSS coordinates from the CPWK station of InaCORS operated by Badan Informasi Geospasial. The CPWK station is located west of the dam in Purwakarta and spans January 2021 to December 2023. Because CPWK is not installed on the dam body, it is used only as a regional vertical-trend benchmark rather than as direct structural validation of crest deformation. Direct structural comparison is instead provided by the crest leveling benchmarks. We also referenced the official technical specification of Jatiluhur Dam (Figure 2) for the crest elevation, operating reservoir levels, and derived freeboard used later in the monitoring indicators. A summary of all datasets used in this study is presented in

Table 1. Datasets used in this study.

Dataset	Details	Period	Source
Sentinel-1 SAR	Interferometric wide mode, VV polarization, 250 × 250 km frame; 341 scenes total, consisting of 185 ascending and 156 descending acquisitions.	January 2019–December 2024	European Space Agency
Water level (reservoir)	Daily in situ measurements at Jatiluhur Dam used to evaluate reservoir loading and deformation response.	January 2019–December 2024	Perum Jasa Tirta II
In situ leveling (crest)	Regular in situ crest measurements used for structural comparison with InSAR-derived deformation patterns.	October 1965–December 2024; recent survey subset December 2023–December 2024	Perum Jasa Tirta II
GNSS (CPWK)	Continuous regional GNSS station west of the dam, used for independent vertical-trend checking; not installed on the dam crest.	January 2021–December 2023	InaCORS, Badan Informasi Geospasial
Dam specification	Crest elevation 114.50 m; maximum operating reservoir level 111.60 m; normal reservoir level 107.00 m; minimum operating reservoir level 87.50 m; freeboard 2.90 m.	–	Perum Jasa Tirta II, Spesifikasi Teknis Bendungan Jatiluhur

.

2.3. Multi-Temporal InSAR Processing: SBAS, PSI, and TW-PSI

The fundamental principle of Interferometric Synthetic Aperture Radar (InSAR) is the observation of phase differences between repeated SAR acquisitions. In this study, Sentinel-1 interferogram generation and preprocessing were performed using ISCE2 version 2.6.3. The interferometric phase can be expressed as

$$\Delta \varphi =\Delta {\varphi }_{\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{p}}+\Delta {\varphi }_{\mathrm{h}\mathrm{e}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}}+\Delta {\varphi }_{\mathrm{a}\mathrm{t}\mathrm{m}}+\Delta {\varphi }_{\mathrm{n}\mathrm{o}\mathrm{i}\mathrm{s}\mathrm{e}}$$

(1)

where $\Delta {\varphi }_{\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{p}}$ represents the deformation phase, $\Delta {\varphi }_{\mathrm{h}\mathrm{e}\mathrm{i}\mathrm{g}\mathrm{h}\mathrm{t}}$ accounts for residual topographic or DEM-related errors, $\Delta {\varphi }_{\mathrm{a}\mathrm{t}\mathrm{m}}$ represents atmospheric propagation delay, and $\Delta {\varphi }_{\mathrm{n}\mathrm{o}\mathrm{i}\mathrm{s}\mathrm{e}}$ captures decorrelation and system noise [60]. Time-series InSAR processing aims to isolate $\Delta {\varphi }_{\mathrm{d}\mathrm{i}\mathrm{s}\mathrm{p}}$ from these other phase contributions using multi-temporal observations and appropriate spatial and temporal filtering.

The Small Baseline Subset (SBAS) approach was implemented using the MintPy version 1.6.1.post7 [61], which extends the foundational SBAS formulation in [62]. SBAS constructs an interferometric network using image pairs with relatively short temporal and perpendicular baselines to reduce decorrelation and improve temporal phase continuity. The deformation time series is estimated through a linear inversion:

$$\Phi =Gm+\epsilon$$

(2)

where $\Phi$ is the observed phase vector, $G$ is the design matrix, $m$ contains the deformation parameters and residual topographic terms, and $\epsilon$ represents residual noise. In this study, SBAS was used primarily to characterize broader spatial deformation patterns around the dam and surrounding terrain.

The Persistent Scatterer Interferometry (PSI) technique introduced in [46] identifies pixels that maintain high phase stability over a long acquisition period. The StaMPS/MTI framework version 4.0b6 [63] further refines PS selection through iterative phase analysis and spatio-temporal filtering. The temporal coherence of a candidate pixel can be expressed as

$${\gamma }_x={\displaystyle \frac{1}{N}}\left|{\displaystyle \sum _{i=1}^N}\exp \left[j\left({\varphi }_i-\overline{\varphi }-\Delta {\varphi }_{\theta ,i}\right)\right]\right|$$

(3)

where ${\varphi }_i$ is the interferometric phase at epoch $i$, $\overline{\varphi }$ is the mean phase over time, $\Delta {\varphi }_{\theta ,i}$ represents the look-angle error term, and $N$ is the number of interferograms. PSI is particularly suitable for stable engineered targets such as asphalt, concrete, parapet walls, exposed crest surfaces, and spillway structures. However, conventional PSI can produce sparse point distributions in vegetated or moisture-affected dam environments.

SBAS and PSI were therefore used as complementary time-series InSAR approaches because they are optimized for different scattering conditions. SBAS is generally effective for retrieving spatially distributed deformation over broader areas, but its spatial averaging, multi-looking, and filtering steps may smooth localized deformation signals on narrow engineered structures. In contrast, PSI can preserve localized deformation signals at coherent point targets, but its spatial coverage may be sparse where stable scatterers are limited. Previous integrated PSInSAR–SBAS studies have shown that PSInSAR may provide higher coherence but lower point density than SBAS, supporting the complementary use of both approaches for deformation monitoring [64,65]. This distinction is important for Jatiluhur Dam, where the crest and spillway contain stable engineered scatterers, while the downstream and surrounding slopes include vegetated or mixed-surface areas.

Figure 3 summarizes the temporal–perpendicular baseline configuration used in this study. The SBAS processing was based on small-baseline interferogram networks to reduce temporal and geometric decorrelation, whereas PSI and TW-PSI were implemented using a single-master configuration. The selected PSI master acquisitions were 20211107 for the descending geometry and 20220115 for the ascending geometry.

To improve PS selection in decorrelated dam regions, we implemented an adapted TW-PSI refinement strategy based on Random Matrix Theory (RMT) and Tracy–Widom eigenvalue screening, following the top-eigenvalue concept for persistent scatterer selection [47]. For each candidate pixel, the largest eigenvalue ${\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}$ of the sample coherence matrix was evaluated to distinguish signal-dominated scatterers from noise-dominated candidates. Under the null hypothesis of random phase noise, the distribution of the largest eigenvalue can be standardized using the Tracy–Widom distribution:

$$z={\displaystyle \frac{{\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}-{\mu }_{np}}{{\sigma }_{np}}}$$

(4)

where ${\mu }_{np}$ and ${\sigma }_{np}$ are the centering and scaling terms determined by the temporal and dimensional parameters of the coherence matrix. A candidate pixel is retained when

$$z>F_{\mathrm{T}\mathrm{W}}^{-1}\left(1-\alpha \right)$$

(5)

where ${\mathrm{F}}_{\mathrm{T}\mathrm{W}}^{-1}$ (·) is the Tracy–Widom quantile function and α controls the false-alarm probability [47]. The centering and scaling terms are defined as

$${\mu }_{np}={\displaystyle \frac{{\left(\sqrt{n-1}+\sqrt{p}\right)}^2}{n}}$$

(6)

$${\sigma }_{np}={\displaystyle \frac{\sqrt{n-1}+\sqrt{p}}{n}}{\left({\displaystyle \frac{1}{\sqrt{n-1}}}+{\displaystyle \frac{1}{\sqrt{p}}}\right)}^{1/3}$$

(7)

where $n$ and $p$ denote the temporal and dimensional parameters of the coherence matrix, respectively. Because tropical dam environments may exhibit spatially variable and non-Gaussian decorrelation due to vegetation, moisture, mixed pixels, and residual atmospheric effects, the Tracy–Widom screening was combined with a robust local adaptive threshold. A candidate was also required to satisfy

$${\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}>\mathrm{M}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\left({\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}\right)+\kappa \cdot \mathrm{M}\mathrm{A}\mathrm{D}\left({\lambda }_{\mathrm{m}\mathrm{a}\mathrm{x}}\right)$$

(8)

where $\kappa$ controls the strictness of the local threshold, and the median and MAD are computed from the local ensemble of candidate pixels. In the implementation used here, the local MAD coefficient was set to $\kappa$ = 3.5, the local ensemble was evaluated using a 200-pixel grid, and bins with fewer than 10 valid candidates were excluded from local threshold estimation. A coherence lower bound of 0.1 was also applied to avoid retaining extremely unreliable candidates. Thus, the Tracy–Widom formulation provides the statistical basis for identifying signal-dominated candidates, while the final practical selection was constrained by the local TW–MAD threshold and minimum coherence requirement. This TW–MAD refinement is intended to improve reliable scatterer retention over low-coherence dam surfaces, particularly along the crest, spillway, and engineered slopes, where conventional fixed coherence thresholds may remove useful monitoring points. The resulting enhanced PS dataset was used for crest-scale deformation interpretation and for the subsequent ascending–descending decomposition. Accordingly, the enhanced TW-PSI product was treated as the primary crest-scale monitoring result, while SBAS was retained as an independent broader-scale consistency check.

Table 2. Key processing configurations used for SBAS, PSI, and TW-PSI analyses.

Parameter	SBAS	PSI	TW-PSI
Network type	Small-baseline	Single-master	Single-master
Baseline network	Figure 3	Figure 3	Figure 3
Master acquisition	–	ASC: 20220115; DSC: 20211107	ASC: 20220115; DSC: 20211107
Minimum temporal coherence	0.4	–	–
Amplitude dispersion threshold	–	D_A = 0.4	D_A = 0.4
Refinement criterion	–	StaMPS phase-stability refinement	TW eigenvalue and local MAD refinement
TW-MAD coefficient	–	–	κ = 3.5
Coherence lower bound	–	StaMPS gamma-based refinement	0.1

summarizes the key processing configurations used for SBAS, PSI, and TW-PSI analyses.

Atmospheric and topographic residuals were mitigated through the standard correction steps implemented in MintPy and StaMPS, including interferogram network optimization, DEM-error correction, spatio-temporal filtering, and atmospheric phase-screen estimation and, where available, ERA5-based atmospheric correction through the MintPy/PyAPS workflow. These corrections reduce long-wavelength atmospheric artifacts and temporally uncorrelated noise; however, residual atmospheric effects cannot be completely excluded. Therefore, the interpretation of periodic deformation in this study is supported by spatial consistency along the crest, comparison with in situ leveling, and correlation with reservoir water-level variations rather than by single-epoch displacement alone. Spatial visualization and map preparation were conducted using QGIS version 3.44.11.

2.4. 2.5D Decomposition, GNSS, and Leveling Comparison

To obtain a first-order geometry-aware interpretation of the deformation field, ascending and descending LOS measurements were decomposed into east–west (E) and quasi-vertical (U) components. Single-track InSAR measures only the projection of ground displacement along the satellite line of sight. Because Sentinel-1 ascending–descending geometries have limited sensitivity to north–south motion, the north–south component cannot be robustly resolved using this two-orbit configuration alone. Therefore, the retrieved E and U components are interpreted as a 2.5D approximation rather than a full three-dimensional displacement solution.

The decomposition follows the ASC–DSC LOS geometry used in MintPy, where LOS observations are projected using the incidence angle and radar look azimuth. In this study, the radar look azimuth γ was computed from the Sentinel-1 satellite heading angle as $\gamma ={\alpha }_{\mathrm{h}\mathrm{e}\mathrm{a}\mathrm{d}}+{90}^{\circ }$ for right-looking acquisitions. The LOS projection for each orbit is expressed as

$$d_{\mathrm{L}\mathrm{O}\mathrm{S}}=-\sin \theta \sin \gamma \hspace{0.17em}E+\cos \theta \hspace{0.17em}U$$

(9)

where $d_{\mathrm{L}\mathrm{O}\mathrm{S}}$ is the LOS displacement, $\theta$ is the incidence angle, $\gamma$ is the radar look azimuth, E is the east–west displacement component, and U is the vertical displacement component. For paired ascending and descending observations, the system becomes

$$\left[\begin{array}{c}{d}_{\mathrm{a}\mathrm{s}\mathrm{c}}\\ d_{\mathrm{d}\mathrm{e}\mathrm{s}\mathrm{c}}\end{array}\right]=\left[\begin{array}{cc}-{\sin \theta }_{\mathrm{a}\mathrm{s}\mathrm{c}}{\sin \gamma }_{\mathrm{a}\mathrm{s}\mathrm{c}}& {\cos \theta }_{\mathrm{a}\mathrm{s}\mathrm{c}}\\ -{\sin \theta }_{\mathrm{d}\mathrm{e}\mathrm{s}\mathrm{c}}{\sin \gamma }_{\mathrm{d}\mathrm{e}\mathrm{s}\mathrm{c}}& {\cos \theta }_{\mathrm{d}\mathrm{e}\mathrm{s}\mathrm{c}}\end{array}\right]\left[\begin{array}{c}E\\ U\end{array}\right]+\eta$$

(10)

where $d_{\mathrm{a}\mathrm{s}\mathrm{c}}$ and $d_{\mathrm{d}\mathrm{e}\mathrm{s}\mathrm{c}}$ are the ascending and descending LOS displacements, and η is the residual term. The system was solved using least squares for pixels with valid measurements from both orbital geometries. The resulting U component is therefore referred to as quasi-vertical displacement throughout this study. This approximation is appropriate for screening settlement-dominated deformation along the dam crest, but it does not exclude possible unresolved north–south motion.

GNSS data from the CPWK station spanning January 2021 to December 2023 were obtained from the Indonesia Continuously Operating Reference Stations (InaCORS) managed by Badan Informasi Geospasial. The station is located in Purwakarta to the west of Jatiluhur Dam and is marked in Figure 1. The data were processed using the Bernese GNSS Software version 5.2 with a double-difference strategy to estimate daily coordinate solutions in the ITRF2014 reference frame [66]. This approach minimizes common satellite and atmospheric errors through differencing, and the use of IGS stations ALIC, IISC, KARR, NTUS, and PIMO provides robust reference points for consistent positioning. Following prior work [67], a Heaviside step function [68] was included to account for coseismic offsets, antenna changes, or other abrupt shifts detected at CPWK.

A weighted least squares model with constraint-based adjustment was applied to estimate linear velocities and step offsets, improving ambiguity resolution and solution stability in GNSS processing. The model is

$$x\left(t\right)=a+bt+\sum _{j}H\left(t-t_j\right){\Delta }_j$$

(11)

where $a$ is the intercept representing the initial coordinate, $b$ is the linear velocity, ${\Delta }_j$ is the amplitude of each offset, and $H\left(t-t_j\right)$ is the Heaviside step function representing abrupt positional changes [68]. A linear model was considered sufficient when no nonlinear trends were evident in the residuals. Outliers were defined as coordinates exceeding the 95% confidence level of the time series, and observations were weighted by the inverse of the squared coordinate uncertainty. Final outputs include GNSS velocity trends, permanent offsets, and their standard deviations derived by classical least squares applied to the daily time series. Because CPWK is not located on the dam body, the GNSS result was used to assess regional vertical-trend consistency only. It was not used as a direct validation of crest-scale InSAR deformation.

Crest leveling data were used as the primary in situ structural reference for evaluating the spatial deformation pattern detected by InSAR. The leveling benchmarks are distributed along the dam crest and provide direct measurements of vertical deformation at engineered monitoring points. Because leveling and InSAR may differ in spatial sampling, temporal reference, and measurement geometry, the comparison focuses on the consistency of deformation patterns along the crest, particularly the presence and position of the central settlement bowl. Because the leveling profile and Sentinel-1 InSAR velocities represent different temporal references, the comparison is interpreted mainly as spatial-pattern agreement along the crest rather than direct temporal equivalence.

2.5. ASDTR-Based Monitoring Priority and Reservoir-Level Comparison

In this study, the Annual Structural Deformation Tolerance Ratio (ASDTR) is used as a screening-level indicator by extending the tolerance-based deformation assessment framework proposed by [69] and adapting it to the operational context of Jatiluhur Dam. The purpose of ASDTR is not to define a deterministic failure threshold, but to normalize the InSAR-derived quasi-vertical deformation rate against an engineering reference value so that crest sections with relatively higher deformation demand can be identified consistently. The index is expressed as

$$ASDTR\left(t\right)={\displaystyle \frac{\left|v_U\left(t\right)\right|}{v_{\mathrm{t}\mathrm{o}\mathrm{l}}}}\times 100\%$$

(12)

where $v_U\left(t\right)$ is the InSAR-derived quasi-vertical deformation rate in year t expressed in mm/year, and v_tol is the adopted annual deformation screening tolerance. The absolute value is used because the index evaluates the magnitude of vertical deformation demand relative to the adopted tolerance, regardless of whether the displacement is expressed as settlement or uplift in the sign convention of the InSAR product.

The adopted annual tolerance was derived from the available freeboard of Jatiluhur Dam and used only as a screening-level reference. Based on the PJT II technical specification, the crest elevation is 114.50 m and the maximum operating reservoir level is 111.60 m, giving an available freeboard of 2.90 m [57,70]. The cross-section in Figure 2 provides the structural and operational context for this definition by showing the dam crest elevation, reservoir operating levels, internal zoning, and crest monitoring condition. Long-term dam deformation can be influenced by cyclic reservoir loading, rainfall, self-consolidation, and aging processes [35]. Non-uniform crest settlement may also affect crest alignment and contribute to cracking or differential deformation that requires monitoring [71]. Because progressive crest settlement reduces the available freeboard, and insufficient freeboard may increase vulnerability to overtopping-related failure mechanisms [72], only part of the available freeboard is treated as a deformation reserve for screening purposes. This interpretation is further supported by recent embankment-dam failure studies showing that overtopping and breach processes can become serious failure mechanisms under extreme hydrological loading [17,73].

As a conservative operational assumption in this study, half of the available freeboard was treated as the deformation reserve. This corresponds to

$$0.5\times 2.90 \mathrm{m}=1.45 \mathrm{m}.$$

(13)

This conservative reserve was annualized over a long service-life normalization horizon of 150 years:

$${\displaystyle \frac{1.45 \mathrm{m}}{150 \mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}\mathrm{s}}}=0.0097 \mathrm{m}/\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}\approx 9.7 \mathrm{m}\mathrm{m}/\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}.$$

(14)

The resulting value was rounded to 10 mm/year and used as the annual screening tolerance for ASDTR normalization. Therefore, in this study

$$v_{\mathrm{t}\mathrm{o}\mathrm{l}}=10 \mathrm{m}\mathrm{m}/\mathrm{y}\mathrm{e}\mathrm{a}\mathrm{r}.$$

(15)

This value is not interpreted as a deterministic failure threshold or a universal allowable settlement criterion for all dams. Instead, it is used as an operational reference for identifying crest sections with relatively higher deformation demand relative to the available freeboard and long-term monitoring context. The use of 50% of the available freeboard over a 150-year normalization period is therefore not intended to represent a formal design-code requirement, but is adopted as a transparent conservative screening assumption to preserve part of the available freeboard for other safety-relevant components. This assumption is aligned with the operational safety-monitoring philosophy reflected in Indonesian dam-safety standards and dam-operation monitoring modules, where the remaining freeboard and post-construction settlement are treated as important elements of dam-safety surveillance. It is also consistent with international freeboard guidelines that consider wind setup, wave run-up, reservoir operation, settlement, and other uncertainty-related components in freeboard assessment [74,75].

Reservoir water-level observations were analyzed independently from ASDTR to evaluate whether hydrological loading influences the observed deformation pattern. Instead of converting water level into a composite risk index, the reservoir record was compared directly with the mean quasi-vertical crest deformation time series. This approach was adopted to avoid over-interpreting a non-standardized load-adjusted risk formulation while still addressing the hydro-mechanical control of reservoir operation on dam deformation. Reservoir-level variation is considered because dam deformation may be influenced by hydrostatic loading, seasonal operation, consolidation, and time-dependent material response [51,54,55].

The correlation between reservoir water level and mean crest deformation was therefore evaluated using Pearson’s correlation coefficient. This comparison is used only to assess whether the temporal deformation fluctuations are associated with reservoir operation. It is not used to define a deterministic failure criterion or a formal dam-safety index. In the subsequent analysis, ASDTR identifies where deformation demand is concentrated along the crest, while the reservoir-level comparison evaluates whether this deformation is temporally associated with hydrological loading.

3. Results

3.1. Multi-Technique InSAR-Derived LOS Deformation and Measurement Density

The multi-technique InSAR results show that LOS displacement velocities over the broader Jatiluhur area generally range from approximately −25 to +25 mm/year (Figure 4). The SBAS, PSI, and TW-PSI products exhibit different spatial sampling characteristics because they are optimized for different scattering conditions, interferometric networks, and filtering strategies. These differences should not be interpreted simply as processing inconsistency, but as the expected consequence of different target types and measurement assumptions in multi-temporal InSAR analysis [39,64,65]. This distinction is important because dam environments commonly contain contrasting surface types within a short distance, including engineered crest surfaces, concrete or asphalt elements, spillway structures, vegetated slopes, and mixed natural surfaces [76,77,78].

The regional maps in Figure 4 provide the broader deformation context around Jatiluhur Dam, while the enlarged dam-scale comparison in Figure 5 clarifies the role of each method for the structure itself. The SBAS results (Figure 5a,d) provide useful spatial information over the wider dam body, downstream embankment, and surrounding terrain. In particular, the negative LOS velocities distributed across parts of the downstream embankment indicate a broader deformation pattern that is not limited only to the crest.

In contrast, PSI and TW-PSI provide more localized measurements over stable point scatterers. The crest of Jatiluhur Dam includes asphalt pavement, concrete elements, parapet structures, spillway components, and benchmark installations, which are more likely to behave as temporally stable radar scatterers than vegetated or mixed-surface slopes. Conventional PSI captures coherent targets along the crest and morning-glory spillway, but its spatial coverage remains limited in areas affected by vegetation, moisture, mixed pixels, or reduced coherence. Under the same initial amplitude-dispersion candidate threshold, the TW–MAD refinement increased the number of measurement points over the dam-structure domain from 333 to 464 in the ascending geometry and from 495 to 721 in the descending geometry, corresponding to increases of approximately 39.3% and 45.7%, respectively (

Table 3. Comparison of conventional PSI and TW-PSI measurement point counts over the Jatiluhur Dam structure.

Orbit Geometry	Conventional PSI Points	TW-PSI Points
Ascending	333	464
Descending	495	721

).

The comparison in Figure 5 shows that SBAS provides broader distributed coverage over the dam body and surrounding slopes, whereas TW-PSI provides denser on-structure sampling over the crest and spillway. Because the subsequent analysis focuses on crest settlement, benchmark-level comparison with leveling data, and tolerance-normalized monitoring indicators, the TW-PSI product is used as the primary dataset for the following 2.5D decomposition and crest-scale deformation interpretation, while SBAS is retained as a complementary product for regional and dam-body deformation context.

The ascending and descending LOS results also differ in local detail because each orbit observes the dam from a different viewing geometry. Therefore, identical LOS patterns are not expected between the two acquisition directions. The combined ascending and descending TW-PSI measurements are used in the next subsection to derive first-order east–west and quasi-vertical deformation components.

3.2. Regional Vertical-Trend Check Using GNSS

The CPWK GNSS station was used as a regional comparison reference to assess the vertical deformation tendency around the Jatiluhur area. CPWK is not installed on the dam body or along the dam crest. Therefore, the comparison in Figure 6 is interpreted only as a regional vertical-trend check, not as direct structural validation or magnitude validation of the crest-scale InSAR deformation.

At the CPWK comparison point, all three datasets indicate a subsidence tendency. The estimated vertical velocity from GNSS is −4.9106 mm/year, while the corresponding InSAR-derived quasi-vertical velocities are −2.8083 mm/year for TW-PSI and −2.342 mm/year for SBAS. Although the InSAR-derived rates are smaller than the GNSS velocity, the sign of deformation is consistent across the three datasets.

The TW-PSI and SBAS quasi-vertical velocities at CPWK are also close to each other, indicating that both InSAR products retrieve a comparable long-term vertical tendency at the regional comparison point. The difference in magnitude between GNSS and InSAR is expected because the datasets differ in measurement principle, spatial sampling, and reference characteristics. GNSS measures displacement at a single geodetic monument, whereas InSAR represents radar-derived displacement over coherent pixels or distributed scatterers.

Consequently, GNSS is used in this study only to provide an independent regional reference for the vertical deformation trend. Direct structural comparison of the dam crest is instead provided by the in situ leveling benchmarks discussed in the following subsection.

3.3. Quasi-Vertical Crest Deformation and Comparison with Leveling

The ascending–descending decomposition provides quasi-vertical deformation estimates from both SBAS and TW-PSI over the dam body and crest (Figure 7a,b). Here, U-SBAS refers to the SBAS-derived quasi-vertical velocity, whereas U-PSI refers to the TW-PSI-derived quasi-vertical velocity. Because the two-orbit decomposition does not resolve the north–south component, these estimates are interpreted as quasi-vertical rather than absolute vertical deformation. The U-SBAS result provides broader spatial coverage across the dam body, abutment areas, and surrounding terrain (Figure 7a), whereas the U-PSI result provides more localized measurements on coherent crest and structural scatterers (Figure 7b).

The decomposed quasi-vertical velocities generally show negative deformation over the crest, indicating settlement. Both U-SBAS and U-PSI identify the central portion of the crest as the main deformation zone, although their spatial expressions differ. U-SBAS shows a smoother and broader deformation pattern, while U-PSI preserves sharper local variations along the crest.

The in situ leveling records provide the main structural reference for evaluating the spatial consistency of the crest-scale deformation pattern. The leveling benchmarks are distributed along the crest from C2 to C19 and are compared with the selected U-PSI and U-SBAS velocities along the same crest alignment (Figure 8). Because the leveling profile is referenced to the initial survey on 6 October 1965, whereas the Sentinel-1 InSAR velocities represent the recent observation period, the comparison is interpreted primarily as a spatial-pattern consistency assessment rather than a direct one-to-one comparison of cumulative displacement magnitude.

The leveling-derived average settlement rate was computed using the cumulative displacement between the initial survey on 6 October 1965 and the final survey on 31 December 2024. The elapsed period is approximately 59.24 years, and the average leveling rate was calculated as

$$v_{\mathrm{l}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{l}\mathrm{i}\mathrm{n}\mathrm{g}}={\displaystyle \frac{D{Z}_{\mathrm{c}\mathrm{u}\mathrm{m},31\mathrm{D}\mathrm{e}\mathrm{c}2024}}{T}}$$

(16)

where $D{Z}_{\mathrm{c}\mathrm{u}\mathrm{m},31\mathrm{D}\mathrm{e}\mathrm{c}2024}$ is the cumulative leveling displacement at the final survey epoch and T is the elapsed time in years.

Table 4. Per-benchmark comparison between long-term leveling-derived average settlement rates and decomposed InSAR quasi-vertical velocities along the Jatiluhur Dam crest.

BM	Lat.	Lon.	$\mathit{D}{\mathit{Z}}_{\mathbf{c}\mathbf{u}\mathbf{m},31\mathbf{D}\mathbf{e}\mathbf{c}2024}$ (mm)	Leveling avg. (mm/yr)	U-PSI (mm/yr)	U-SBAS (mm/yr)
C2	−6.522224	107.384857	−82	−1.38	−0.37	−2.24
C3	−6.522474	107.385483	−182	−3.07	−1.57	−3.02
C4	−6.522631	107.385948	−275	−4.64	−1.63	−4.29
C5	−6.522978	107.386719	−364	−6.14	−2.92	−5.39
C6	−6.523104	107.387314	−398	−6.72	−3.63	−5.09
C7	−6.523372	107.387856	−429	−7.24	−5.51	−5.96
C7B	−6.523506	107.38842	−468	−7.90	−6.29	−8.13
C8	−6.523575	107.388695	−468	−7.90	−5.16	−10.23
C9	−6.523658	107.388901	−422	−7.12	−4.47	−5.66
C10	−6.523829	107.389297	−373	−6.30	−3.86	−6.6
C11	−6.523939	107.389969	−333	−5.62	−3.52	−6.54
C12	−6.524090	107.39045	−286	−4.83	−3.81	−4.91
C13	−6.524141	107.390808	−252	−4.25	−2.85	−4.09
C14	−6.524295	107.391281	−221	−3.73	−2.58	−4.63
C15	−6.524360	107.391571	−180	−3.04	−2.77	−2.38
C16	−6.524580	107.392326	−131	−2.21	−1.48	−3.46
C17	−6.524692	107.392909	−90	−1.52	−0.67	−2.78
C18	−6.524761	107.393837	−47	−0.79	−0.63	−3.28
C19	−6.524889	107.393852	−21	−0.35	−0.47	−3.45

summarizes the per-benchmark comparison between the leveling-derived average settlement rate, U-PSI velocity, and U-SBAS velocity.

The values in Table 4 show a clear bowl-shaped settlement pattern along the crest. The largest long-term leveling-derived settlement rates occur around C7B–C8, where the average settlement rate reaches approximately −7.90 mm/year. The western crest near C2–C5 shows smaller settlement rates, the central section around C7–C12 forms the main settlement bowl, and the eastern crest toward C16–C19 shows a reduction in settlement magnitude. This pattern is also visible in the U-PSI and U-SBAS crest profiles shown in Figure 8.

To quantify the spatial agreement among the three profiles, Pearson and Spearman correlation coefficients were computed using the 19 paired benchmark values listed in Table 4. The results are summarized in

Table 5. Spatial agreement between leveling-derived average settlement rates and decomposed InSAR quasi-vertical velocities along the Jatiluhur Dam crest.

Comparison	N	Pearson R	Spearman $\mathit{\rho }$
Leveling avg. 1965–2024 vs. U-PSI	19	0.932	0.958
Leveling avg. 1965–2024 vs. U-SBAS	19	0.843	0.898
U-PSI vs. U-SBAS	19	0.828	0.875

.

The correlation results in Table 5 indicate that U-PSI shows the strongest agreement with the leveling-derived average settlement profile (R = 0.932, $\mathit{\rho }$ = 0.958). U-SBAS also shows strong agreement with the leveling profile (R = 0.843, $\mathit{\rho }$ = 0.898), and the agreement between U-PSI and U-SBAS is also high (R = 0.828, $\mathit{\rho }$ = 0.875). These results show that the dominant mid-crest settlement morphology is consistently detected by the leveling, TW-PSI, and SBAS profiles.

Because U-PSI shows the strongest agreement with the leveling-derived crest profile, the following time-series analyses use the TW-PSI product as the primary deformation record for the selected crest benchmarks.

3.4. Event-Based Assessment of the 2022 Cianjur Earthquake

The Mw 5.6 Cianjur earthquake on 21 November 2022 is treated in this study as an event reference for evaluating whether a measurable deformation perturbation occurred at Jatiluhur Dam during the monitoring period. To evaluate possible event-related deformation, quasi-vertical InSAR displacement time series were extracted at the crest benchmarks C2–C19 using the nearest TW-PSI measurement points. Each time series was referenced to the first available epoch in January 2019.

Linear regression was applied separately to the pre-event and post-event periods to estimate the pre-event velocity $v_{\mathrm{p}\mathrm{r}\mathrm{e}}$, post-event velocity $v_{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}$, and velocity change $\Delta v=v_{\mathrm{p}\mathrm{o}\mathrm{s}\mathrm{t}}-v_{\mathrm{p}\mathrm{r}\mathrm{e}}$. A displacement step, referred to here as the event jump, was estimated at the earthquake epoch by comparing the modeled displacement immediately before and after the event. The velocity change Δv is used as a diagnostic indicator of pre- and post-event trend modification, whereas the jump term is used to evaluate possible abrupt displacement around the earthquake epoch. To avoid relying only on visual interpretation, the event jump was further normalized by the pre-event displacement variability. The normalized jump statistic was computed as

$$Z_{\mathrm{j}\mathrm{u}\mathrm{m}\mathrm{p}}={\displaystyle \frac{\mathrm{J}\mathrm{u}\mathrm{m}\mathrm{p}}{{\sigma }_{\mathrm{p}\mathrm{r}\mathrm{e}}}}$$

(17)

where ${\sigma }_{\mathrm{p}\mathrm{r}\mathrm{e}}$ is the standard deviation of the pre-event displacement time series. A jump was considered statistically significant when |$Z_{\mathrm{j}\mathrm{u}\mathrm{m}\mathrm{p}}$| > 1.96, corresponding to a two-sided 95% screening criterion. These metrics are summarized in

Table 6. Pre- and post-event quasi-vertical deformation characteristics and event-jump significance screening at Jatiluhur Dam crest benchmarks.

Point	${\mathit{v}}_{\mathbf{p}\mathbf{r}\mathbf{e}}$ (mm/yr)	${\mathit{v}}_{\mathbf{p}\mathbf{o}\mathbf{s}\mathbf{t}}$ (mm/yr)	Δv (mm/yr)	Jump (mm)	${\mathit{Z}}_{\mathbf{j}\mathbf{u}\mathbf{m}\mathbf{p}}$	Significant at 95%
C2	−0.85	+0.66	+1.51	1.36	0.23	No
C3	−1.08	−2.57	−1.49	3.34	0.50	No
C4	−1.89	−0.49	+1.40	7.02	1.01	No
C5	−3.90	−1.60	+2.30	4.01	0.50	No
C6	−5.28	−1.58	+3.70	3.44	0.41	No
C7	−7.14	−4.24	+2.90	4.48	0.44	No
C7B	−6.70	−6.59	+0.11	2.65	0.28	No
C8	−7.22	−3.97	+3.25	4.12	0.42	No
C9	−5.23	−2.66	+2.57	−0.99	−0.12	No
C10	−4.95	−3.14	+1.80	7.46	0.88	No
C11	−4.57	−2.59	+1.98	2.70	0.35	No
C12	−4.45	−0.47	+3.99	3.46	0.47	No
C13	−3.40	+1.44	+4.83	1.06	0.16	No
C14	−3.60	+0.46	+4.06	4.15	0.61	No
C15	−2.91	−0.42	+2.49	3.40	0.53	No
C16	−1.86	−0.37	+1.49	4.49	0.72	No
C17	−1.89	+3.17	+5.06	5.94	0.97	No
C18	−0.08	−3.42	−3.34	3.83	0.64	No
C19	−1.33	+1.93	+3.26	3.45	0.61	No

.

The pre-event velocities are dominated by subsidence, particularly in the central crest region. Points C6–C12 and C7B generally show negative pre-event velocities, with the strongest settlement tendency around C7–C8, where $v_{\mathrm{p}\mathrm{r}\mathrm{e}}$ reaches approximately −7 mm/year. This pattern is consistent with the settlement bowl identified in the quasi-vertical deformation map and the leveling profile.

After the earthquake, most benchmarks show less negative velocities or weak positive trends, producing mostly positive $\Delta v$ values. Exceptions occur at C3 and C18, where the post-event velocities become more negative. This indicates that the post-event period was not characterized by a systematic acceleration of settlement along the entire crest.

The estimated event jumps range from approximately −1 to 7.5 mm. The largest positive jump is observed at C10, while C9 shows a small negative jump. However, the normalized jump statistics remain below the two-sided 95% screening threshold at all benchmarks, with the largest absolute value of $Z_{\mathrm{j}\mathrm{u}\mathrm{m}\mathrm{p}}$ being approximately 1.01. The jump values also do not show a uniform crest-wide sign or spatial pattern at the earthquake epoch. Therefore, based on the benchmark-level TW-PSI time series and event-jump significance screening, no statistically significant co-seismic displacement was detected along the Jatiluhur Dam crest at the 95% confidence level.

3.5. Reservoir-Level Influence and ASDTR-Based Monitoring Priority

The time-series comparison in Figure 9 indicates that the crest deformation of Jatiluhur Dam is associated with reservoir-level variation during the 2019–2024 observation period. The mean quasi-vertical crest displacement shows a negative correlation with reservoir water level, with r = −0.709. This indicates that higher reservoir levels tend to coincide with more negative vertical displacement, or increased settlement tendency at the crest.

To translate the observed crest-scale quasi-vertical deformation into an engineering-oriented monitoring quantity, the Annual Structural Deformation Tolerance Ratio (ASDTR) was computed using the full-period U-PSI velocities because U-PSI showed the strongest spatial agreement with the leveling-derived crest settlement profile. This follows the concept of converting geodetic deformation into an engineering-scale tolerance ratio, as proposed by [69], but is adapted here to the operational and structural context of Jatiluhur Dam. In this study, the adopted annual screening tolerance is 10 mm/year, as described in the methodology.

The ASDTR results in

Table 7. ASDTR-based monitoring priority at Jatiluhur Dam crest benchmarks.

Point	Velocity (mm/yr)	ASDTR (%)	Monitoring Priority
C2	−0.373	3.73	Low
C3	−1.567	15.67	Low
C4	−1.625	16.25	Low
C5	−2.924	29.24	Low
C6	−3.627	36.27	Moderate
C7	−5.507	55.07	High
C7B	−6.294	62.94	High
C8	−5.162	51.62	High
C9	−4.474	44.74	Moderate
C10	−3.857	38.57	Moderate
C11	−3.521	35.21	Moderate
C12	−3.810	38.10	Moderate
C13	−2.852	28.52	Low
C14	−2.578	25.78	Low
C15	−2.772	27.72	Low
C16	−1.480	14.80	Low
C17	−0.666	6.66	Low
C18	−0.629	6.29	Low
C19	−0.471	4.71	Low

show that the highest monitoring priorities are concentrated at C7, C7B, and C8, where the ASDTR values exceed 50%. These benchmarks coincide with the central settlement bowl identified from the TW-PSI quasi-vertical deformation map and the benchmark-level leveling comparison discussed above. C7B shows the highest ASDTR value, reaching 62.94%, followed by C7 and C8 with values of 55.07% and 51.62%, respectively.

Moderate monitoring priority is observed around C6 and C9–C12, where ASDTR values range from approximately 35% to 45%. These points form a transitional zone around the central settlement bowl, suggesting that the deformation demand gradually decreases away from the highest-settlement area. In contrast, the western crest near C2–C5 and the eastern crest toward C13–C19 generally show lower ASDTR values, indicating weaker recent settlement rates relative to the adopted screening tolerance.

All ASDTR values remain below 100%, meaning that the annual quasi-vertical deformation rates during 2019–2024 remain below the adopted screening tolerance of 10 mm/year. Therefore, the classification in Table 7 is used to identify relative monitoring priority along the crest, with the highest-priority zone concentrated around C7–C8 and C7B.

4. Discussion

4.1. Methodological Value of TW-PSI and Multi-Sensor Consistency

The results show that the three InSAR products provide different but complementary information for Jatiluhur Dam. SBAS provides broader spatial coverage over the dam body, downstream embankment, and surrounding slopes, whereas PSI-based products provide more localized measurements over stable engineered scatterers. This distinction is important for dam monitoring because dam environments commonly include a mixture of asphalt pavement, concrete elements, spillway structures, vegetated slopes, and mixed natural surfaces within a short distance [76,78,79]. Therefore, differences among SBAS, PSI, and TW-PSI should not be interpreted simply as processing inconsistency, but as the expected consequence of different scattering assumptions, spatial filtering strategies, and measurement targets in multi-temporal InSAR analysis [39,64,65].

To clarify these complementary roles,

Table 8. Operational distinction among SBAS, PSI, and TW-PSI in this study.

Method	Measurement Target	Main Advantage	Intended Use in This Study
SBAS	Distributed scatterers and broader deformation fields.	Provides wider spatial coverage over the dam body, downstream embankment, and surrounding slopes.	Used to describe the broader deformation context around Jatiluhur Dam.
PSI	Stable point scatterers on coherent engineered surfaces.	Captures localized deformation over stable crest, spillway, asphalt, and concrete elements.	Used as the conventional point-scatterer reference for comparison with TW-PSI.
TW-PSI	Refined signal-dominated point scatterers.	Improves on-structure measurement density over the crest and spillway.	Used as the primary crest-scale product for decomposition, leveling comparison, and ASDTR-based monitoring prioritization.

summarizes the practical distinction among SBAS, PSI, and TW-PSI in this study.

The main methodological value of TW-PSI in this study is the improvement of on-structure measurement density over the dam body and crest relative to conventional PSI. Because conventional PSI and TW-PSI were initialized using the same amplitude-dispersion candidate threshold (D_A = 0.4), the point-count comparison in Table 3 provides a direct assessment of the added effect of the Tracy–Widom and local TW–MAD refinement under the adopted processing configuration. This comparison indicates that the improvement is not simply a result of relaxing the initial candidate selection, but is associated with the subsequent signal-dominated scatterer refinement. Nevertheless, the result should be interpreted as an improvement in observability and measurement density for this tropical dam setting, rather than as a universal point-density gain applicable to all sites.

This improvement is particularly relevant because conventional InSAR and PSI applications are affected by temporal decorrelation, geometrical decorrelation, atmospheric disturbance, and the stability of persistent scatterers [46,63]. In a tropical dam environment such as Jatiluhur, where engineered surfaces are located close to vegetated slopes, reservoir margins, and mixed natural surfaces, improving the density and continuity of reliable on-structure points is essential for crest-scale interpretation. In this context, TW-PSI does not replace SBAS, but strengthens the measurement basis for localized structural monitoring by retaining denser coherent points over engineered crest and spillway elements.

The complementary role of SBAS- and PSI-based products is also consistent with recent operational dam-monitoring studies, which emphasize that InSAR applicability depends strongly on persistent-scatterer availability, viewing geometry, dam orientation, surface condition, and the deformation component being interpreted [76,80,81]. Similar studies further show that PSI-based monitoring becomes more useful for dam operators when it is integrated with in situ observations and interpreted as a complementary spatial monitoring layer rather than as a full replacement for conventional measurements [76,79,81]. This interpretation is consistent with the present study, where TW-PSI is used for crest-scale deformation mapping, SBAS is retained for broader spatial context, and leveling remains the main structural reference.

The comparison with in situ leveling further supports the use of TW-PSI as the primary crest-scale product. The strongest spatial agreement with the leveling-derived average settlement profile is obtained by U-PSI, with R = 0.932 and $\mathit{\rho }$ = 0.958, followed by U-SBAS with R = 0.843 and $\mathit{\rho }$ = 0.898. These results indicate that the decomposed InSAR products reproduce the same dominant mid-crest settlement morphology observed in the long-term leveling profile. However, this agreement should be interpreted as spatial-pattern consistency rather than exact temporal validation, because the leveling profile represents a long-term average from 1965 to 2024, whereas Sentinel-1 InSAR represents the recent 2019–2024 observation period.

The CPWK GNSS comparison provides an additional regional vertical-trend check, but it should not be overinterpreted as direct validation of crest deformation. CPWK is not installed on the dam body and represents a regional geodetic monument outside the structure. The consistent subsidence tendency from GNSS, TW-PSI, and SBAS supports the regional downward trend around the comparison location, while the in situ leveling benchmarks remain the main structural reference for the dam crest. Thus, the integrated use of TW-PSI, SBAS, leveling, and GNSS strengthens the interpretation while preserving the different roles and limitations of each dataset.

4.2. Hydro-Mechanical Interpretation and Earthquake-Related Deformation Assessment

The deformation pattern observed at Jatiluhur Dam is dominated by a persistent central settlement bowl, with the strongest settlement concentrated around C7–C12 and especially near C7B–C8. This morphology is consistently expressed in the long-term leveling profile, the TW-PSI quasi-vertical velocity, and the SBAS quasi-vertical velocity. The persistence of this pattern across independent measurement concepts suggests that the observed deformation is not merely an LOS-geometry artifact or isolated scatterer noise, but reflects a stable structural deformation feature of the dam crest. Similar InSAR-based dam studies have shown that long-term and spatially heterogeneous deformation can be associated with consolidation settlement, material heterogeneity, and differences in structural or embankment conditions [35,80].

The negative correlation between reservoir water level and mean quasi-vertical crest displacement (r = −0.709) indicates that higher reservoir levels are associated with stronger settlement tendency at the crest. This relationship is physically plausible for a rockfill dam with an inclined impervious clay core, where increasing reservoir level raises hydrostatic loading on the upstream shell, impervious core, filter zones, and foundation. Such loading can contribute to compression, self-consolidation, and time-dependent adjustment of the dam body. Recent dam-monitoring studies have also shown that reservoir level, temperature, and other external variables can act as deformation drivers in dam segments, and that water-level changes may explain short-term variations superimposed on longer-term deformation behavior [77]. In long embankment–dam systems, water-level changes have also been reported to influence deformation, although consolidation settlement may remain the dominant control over the monitoring period [80].

The event-based assessment indicates that the 2022 Mw 5.6 Cianjur earthquake did not produce a statistically significant crest-wide deformation response at Jatiluhur Dam. Although several benchmarks show changes in pre- and post-event velocity, the direction and magnitude of these changes are not spatially uniform. Most post-event velocities become less negative or weakly positive, whereas C3 and C18 become more negative. The estimated event jumps range from approximately −1 to 7.5 mm; however, the normalized jump statistics remain below the two-sided 95% screening threshold at all benchmarks, with the largest absolute Zjump value being approximately 1.01. The Cianjur earthquake remains important as a regional seismic perturbation because West Java is affected by active shallow crustal faults, and recent studies have shown that the 2022 Cianjur event involved a destructive shallow earthquake and a complex faulting process [42,43,44]. However, within the Jatiluhur crest time series, the event is better interpreted as an external reference for testing the monitoring framework rather than as the primary driver of the observed deformation.

Therefore, the combined evidence points more strongly toward long-term settlement and reservoir-related hydro-mechanical adjustment than toward earthquake-induced structural damage during the 2019–2024 observation period. This interpretation is supported by the spatial consistency among leveling, TW-PSI, and SBAS; the concentration of deformation in the central settlement bowl; the negative reservoir-level correlation; and the absence of statistically significant co-seismic displacement at the crest. A lagged-correlation sensitivity check was additionally performed between the mean crest TW-PSI quasi-vertical displacement and reservoir water level for lag intervals from 0 to 120 days. The strongest association was obtained at zero-day lag, with Pearson r = −0.709 and Spearman $\mathit{\rho }$ = −0.670 over 174 Sentinel-1 epochs, indicating that higher reservoir levels are observationally associated with greater downward crest displacement. The correlation weakened with increasing lag, suggesting that the reservoir-level association is mainly contemporaneous within the Sentinel-1 sampling interval rather than showing a stronger delayed response. Nevertheless, atmospheric residuals, seasonal moisture variation, reservoir-operation variability, and unresolved short-term effects cannot be fully excluded in Sentinel-1 time-series InSAR. Therefore, the reservoir-level correlation reported here should be interpreted as an observational association, while detailed seasonal decomposition remains a topic for future work.

For this reason, the interpretation is based on multiple lines of evidence rather than on a single epoch, a single sensor, or a single statistical indicator. Overall, the deformation behavior of Jatiluhur Dam is more consistent with gradual hydro-mechanical and consolidation-related response than with event-driven structural damage.

4.3. Operational Implications, Limitations, and Future Monitoring

The ASDTR-based assessment provides an engineering-oriented way to translate InSAR-derived quasi-vertical deformation rates into monitoring priorities. The highest ASDTR values occur at C7B, C7, and C8, which also coincide with the central settlement bowl identified from leveling and InSAR decomposition. This convergence indicates that the mid-crest zone should receive closer monitoring attention in future inspection and surveillance programs. However, the ASDTR classification should not be interpreted as a deterministic failure-risk category. Instead, it is a screening-level monitoring indicator that normalizes the observed deformation rate relative to an adopted annual tolerance, consistent with the broader concept of translating geodetic deformation into an engineering-oriented monitoring quantity [35,69].

The adopted annual screening tolerance is derived from the available freeboard of Jatiluhur Dam based on the PJT II technical specification, where the crest elevation is 114.50 m and the maximum operating reservoir level is 111.60 m, giving an available freeboard of 2.90 m [57,70]. The tolerance is used only as a conservative operational reference for identifying crest sections with relatively higher deformation demand. This framing is important because progressive crest settlement can reduce freeboard reserve, while insufficient freeboard may increase vulnerability to overtopping-related failure mechanisms [17,72,73]. Non-uniform settlement is also relevant because differential deformation along the crest may affect alignment and contribute to cracking or localized structural distress [71].

Several limitations should be considered when interpreting the results. First, the ascending–descending decomposition provides a 2.5D approximation into east–west and quasi-vertical components, but it cannot robustly resolve north–south motion using only two Sentinel-1 viewing geometries. This limitation is inherent to LOS-based InSAR observation and multi-geometry decomposition, because InSAR measures only the projection of three-dimensional displacement onto the radar line of sight, and two conventional ascending–descending observations remain insufficient for a fully constrained three-dimensional solution [82,83]. Therefore, the retrieved U component should be interpreted as quasi-vertical and may contain bias if an unmodelled north–south component is present. Because no on-structure north–south GNSS observation or third independent SAR viewing geometry was available, a reliable quantitative correction of this unresolved component was not attempted in this study. Second, CPWK GNSS is useful for regional trend checking, but it cannot replace on-structure GNSS instrumentation. Third, leveling and InSAR represent different temporal references, so their comparison is most appropriate for spatial-pattern consistency rather than exact temporal validation or magnitude equivalence. Fourth, residual atmospheric effects, geocoding uncertainty, outliers, and seasonal surface-condition changes may still influence Sentinel-1 time-series measurements, particularly in infrastructure monitoring applications [48].

Despite these limitations, the integrated framework provides practical value for operational dam monitoring. TW-PSI improves on-structure measurement density, ASC–DSC decomposition improves the physical interpretability of LOS deformation, leveling provides direct structural reference, GNSS supports regional trend checking, reservoir data help interpret hydro-mechanical forcing, and ASDTR translates deformation rates into monitoring-priority zones. This type of integrated interpretation is consistent with recent operational dam-monitoring studies, which emphasize that InSAR is most valuable when combined with in situ observations, dam-specific geometry assessment, and operator-oriented interpretation [76,79,81]. For Jatiluhur Dam, the integrated evidence indicates that the observed 2019–2024 deformation is more consistent with long-term settlement and reservoir-related hydro-mechanical response than with earthquake-induced structural damage. Future monitoring would benefit from permanent on-crest GNSS or artificial radar reflectors, continued leveling surveys, updated reservoir-operation records, and periodic reprocessing of Sentinel-1 time series to evaluate whether the central settlement bowl remains stable or evolves toward higher deformation demand.

5. Conclusions

This study measured and interpreted the 2019–2024 deformation of Jatiluhur Dam using multi-temporal Sentinel-1 InSAR, in situ leveling, GNSS, and reservoir water-level observations. The main conclusions are as follows:

• SBAS, conventional PSI, and TW-PSI provide complementary deformation information for Jatiluhur Dam. SBAS captures broader deformation patterns over the dam body, downstream embankment, and surrounding slopes, whereas PSI-based products provide more localized measurements over stable engineered scatterers along the crest and spillway. Under the same initial amplitude-dispersion candidate threshold of D_A = 0.4, TW-PSI increased the number of on-structure measurement points relative to conventional PSI, supporting more detailed crest-scale deformation assessment in a tropical dam environment.
• The ascending–descending decomposition revealed a settlement-dominated quasi-vertical deformation pattern along the crest, with the strongest settlement concentrated in the central crest zone around C7–C12 and particularly near C7B–C8. The decomposed U-PSI product showed stronger spatial agreement with the long-term leveling-derived settlement profile than U-SBAS. However, because two-orbit Sentinel-1 decomposition cannot robustly resolve north–south motion, the retrieved U component should be interpreted as quasi-vertical rather than as a fully constrained vertical displacement.
• The CPWK GNSS station provides only a regional vertical-trend check, not direct structural validation of crest deformation, because it is not installed on the dam body. Similarly, the leveling and Sentinel-1 InSAR datasets represent different observation periods; therefore, their comparison should be interpreted as spatial-pattern consistency rather than exact temporal validation.
• The 2022 Mw 5.6 Cianjur earthquake did not produce a statistically significant co-seismic crest deformation signal in the available Sentinel-1 observations. The normalized jump statistics remained below the two-sided 95% screening threshold at all benchmarks, and the pre- and post-event velocity changes did not show systematic settlement acceleration along the crest. Thus, the earthquake is interpreted as an external event reference for testing the monitoring framework rather than as the dominant deformation driver.
• The observed deformation is more consistent with long-term settlement and reservoir-related hydro-mechanical response than with earthquake-induced structural damage. The lagged-correlation sensitivity check showed that the strongest association between mean crest deformation and reservoir water level occurred at zero-day lag, with Pearson r = −0.709 and Spearman $\mathit{\rho }$ = −0.670, suggesting a mainly contemporaneous observational association between higher reservoir levels and increased settlement tendency. The ASDTR-based screening identifies C7B, C7, and C8 as monitoring-priority zones, while all values remain below the adopted annual screening tolerance. Future monitoring should prioritize the central crest zone and would benefit from continued leveling, permanent on-crest GNSS or artificial radar reflectors, updated reservoir-operation records, and periodic reprocessing of Sentinel-1 time series.