Highway Slope Displacement Monitoring and Stability Analysis Based on D-InSAR and GNSS

Highway slopes are important components of highway infrastructure in mountainous areas, and their deformation directly affects operational safety and traffic function. Engineering excavation, geological heterogeneity, rainfall infiltration, groundwater variations, and long-term environmental effects may cause slow, progressive, and spatially variable deformation. Subtle early changes are often difficult to identify through manual inspection. Their accumulation may eventually lead to slope cracking, damage to protective structures, or local sliding, thereby disrupting traffic. Operational highway slopes are numerous and widely distributed, and their responses may vary among slope stages and elevations. A small number of monitoring stations or periodic inspections therefore cannot fully characterize slope conditions. Deformation monitoring that combines temporal continuity with broad spatial coverage can help identify anomalous areas, characterize overall deformation, and define priorities for field verification.

Researchers have investigated highway-slope deformation monitoring, hazard mitigation, and operational safety using field observations, numerical simulation, remote sensing, and transportation resilience assessment. Li et al. [1] combined displacement, rainfall, and anti-slide pile observations from a slope along the Yunmao Expressway with finite-element modeling to evaluate the effects of rainfall and pile arrangement on slope deformation. Zhang et al. [2] used the finite-element and material-point methods, surrogate models, and Monte Carlo simulation to assess road functionality and recovery under rainfall-induced slope failure. They found that uncertainty in slope strength strongly affected highway resilience. Liu et al. [3] developed an automated monitoring and emergency-response system for a high, steep sandstone slope along a highway under construction. Deep-borehole displacement data were used to identify deformation stages and slip surfaces, and the effectiveness of warning and mitigation measures was demonstrated. Wu et al. [4] converted MEMS acceleration measurements into subsurface displacement and compared the results with particle image velocimetry. Their method identified displacement within the sliding zone of a gravelly-soil slope. Liu et al. [5] monitored a small highway slope using Sentinel-1A and ALOS-2 data. C-band observations were constrained by slope dimensions and coherence, whereas L-band TCP-InSAR identified local active areas. Zhang et al. [6] deployed a general-purpose BeiDou monitoring system on an expansive-soil slope protected by soilbags and analyzed its deformation using soilbag, rainfall, and crack observations. However, many existing studies rely on a limited number of stations or focus on individual rainfall events, construction periods, and local deformation processes. Long-term, continuous, and spatially distributed deformation of operational highway slopes therefore requires further investigation.

The Global Navigation Satellite System (GNSS) continuously records coordinate changes at monitoring stations. Berru Garcia et al. [7] tested a low-cost differential GNSS system at the Chin Coulee landslide. Six-month displacement trends agreed closely with those from a commercial system and previous results, although the power supply and thermal insulation required improvement. Zhang et al. [8] integrated an adaptive receiver, communications, a deployment unmanned aerial vehicle, and a cloud platform. Tests at the Dangchuan landslide demonstrated unmanned deployment and real-time monitoring in high-risk areas. Li et al. [9] compared the data quality, positioning performance, and deformation-detection capability of miniaturized GNSS antennas and conducted six months of monitoring at the Baige landslide. Although signal-to-noise ratio and multipath affected the compact antennas, they still yielded useful displacement observations. Wu et al. [10] incorporated multipath correction and stochastic-model optimization into carrier-phase receiver autonomous integrity monitoring, reducing gross errors and improving RTK reliability. Lu et al. [11] classified mountainous-area GNSS signals using K-means++ and developed adaptive elevation-angle and class-specific stochastic models, which improved ambiguity resolution and positioning under obstruction and multipath conditions. These studies expanded GNSS applications through low-cost devices, unmanned aerial vehicle deployment, and error control. Nevertheless, GNSS measurements remain limited to discrete stations, and their spatial coverage depends on the number and placement of instruments, making it difficult to describe continuous deformation across an entire slope.

Interferometric Synthetic Aperture Radar (InSAR) can identify spatially distributed slope deformation. Cheng et al. [12] reviewed spaceborne InSAR applications in landslide monitoring and susceptibility assessment. They summarized advances in three-dimensional monitoring, failure prediction, and multisource integration, while noting the limited use of dynamic information. Novellino et al. [13] reviewed satellite-based landslide mapping and concluded that Sentinel data and artificial intelligence have accelerated inventory updating, although operational applications remain constrained by data availability and quality. Zheng et al. [14] fused asynchronous observations using a spatiotemporally constrained Kalman filter, improving the characterization of three-dimensional movement and block interactions. Xu et al. [15] linked satellite-observed precursory creep with granular-flow simulations and demonstrated the feasibility of estimating the inundation extent of the Chunchi landslide. Sreejith et al. [16] combined InSAR, GPS, rainfall data, and singular spectrum analysis to identify seasonal movement and cascading acceleration at the Joshimath landslide. Wang et al. [17] integrated SBAS-InSAR, field monitoring, machine learning, and a limit-equilibrium model to establish a data-physics framework for landslide prediction and warning. Wen et al. [18] used ascending- and descending-track InSAR together with optical imagery to investigate the Wudongde reservoir area. They found that reservoir-bank landslides were more likely to change their deformation state after impoundment.

Figure 1. Research workflow

3.1 Basic principles and processing of D-InSAR

Interferometric Synthetic Aperture Radar (InSAR) is a surface-deformation measurement technique developed from synthetic aperture radar imaging. It processes two complex SAR images acquired over the same area at different times or from different orbital positions. Differences in phase information between the images are then used to derive surface deformation or topographic information. The interferometric geometry is shown in Figure 3.

Figure 3. InSAR interferometric geometry

Two accurately co-registered complex SAR images form an interferogram. The interferometric phase can be decomposed as follows:

$\begin{aligned} \Delta \varphi_{\text {int}}=\Delta \varphi_{\text {flat}} & +\Delta \varphi_{\text {topo}}+\Delta \varphi_{\text {def}}+\Delta \varphi_{\text {atm}} +\Delta \varphi_{\text {noise}}\end{aligned}$                  (1)

where, Δφ_flat is the flat-earth phase, Δφ_topo is the topographic phase, Δφ_def is the deformation phase, Δφ_atm is the atmospheric-delay phase, and Δφ_noise represents noise and other residual phase components. After the flat-earth and topographic phases are removed using precise orbit data and an external DEM, the differential interferometric phase can be expressed as:

$\Delta \varphi_D=\Delta \varphi_{\text {def }}+\Delta \varphi_{\text {atm }}+\Delta \varphi_{\text {noise }}$                  (2)

The relationship between the deformation phase and the slant-range change between the ground target and the satellite is:

$\Delta \varphi_{\text {def }}=-4 \pi \Delta R / \lambda$                   (3)

where, λ is the radar wavelength and ΔR is the slant-range change between the ground target and the satellite. D-InSAR measures the projection of the true three-dimensional surface displacement onto the LOS direction rather than an independent vertical or horizontal displacement component. Accordingly, all D-InSAR results in this study are referred to as LOS displacement. GNSS vertical displacement was used to evaluate the D-InSAR results in terms of temporal trend, fluctuation magnitude, and characteristic periods. The sign of LOS displacement indicates the slant-range change of a target relative to the reference image along the radar line of sight (LOS). Its spatial meaning must be interpreted together with the satellite imaging geometry and slope aspect.

D-InSAR processing was performed with the SARscape module in ENVI. The workflow included precise orbit correction, subsetting, master and slave image selection, co-registration, interferogram generation, removal of flat-earth and topographic phases, Goldstein filtering, phase unwrapping, orbit refinement and re-flattening, phase-to-LOS displacement conversion, and geocoding. The earlier image was used as the master and the later image as the slave, and the multilook ratio was set to 1:3. A DEM with a resolution of approximately 30 m was used to simulate and remove the topographic phase. Goldstein filtering was applied to suppress phase noise, followed by phase unwrapping with the minimum-cost-flow method at a coherence threshold of 0.15. The orbit-refinement radius was set to 30 m. After re-flattening, the differential phase was converted to LOS displacement. The 10 interferometric pairs were used to derive LOS displacement in the study area from December 2021 to August 2022.

3.2 Comparison of D-InSAR LOS displacement with GNSS vertical displacement

Vertical displacement measured by the field GNSS stations was used in this study. For each GNSS station, the corresponding spatial grid cell was located in ArcGIS Pro using the station coordinates, and the multi-temporal D-InSAR LOS displacement series was extracted from that cell. No neighborhood average was calculated from surrounding grid cells. The extracted series therefore represents LOS displacement in the grid cell corresponding to the GNSS coordinates. Although GNSS vertical displacement and D-InSAR LOS displacement are measured along different directions, both record deformation changes at the monitoring location. GNSS field observations were used as a reference to evaluate the consistency of D-InSAR results in terms of temporal trend, fluctuation magnitude, and correspondence during characteristic periods. The grid cells corresponding to the GNSS coordinates are shown in Figure 4.

Figure 4. D-InSAR grid cells corresponding to the GNSS monitoring stations