Monitoring and Analysis of Surface Deformation in Mining Area Based on LuTan-1 and Sentinel-1A Data

Highlights

What are the main findings?

• Through a systematic comparative study using DInSAR and SBAS-InSAR, L-band LuTan-1 SAR demonstrated substantially superior monitoring performance in high-intensity mining areas compared with C-band Sentinel-1A.
• Accuracy validation indicates that DInSAR monitoring results derived from L-SAR data are in strong agreement with levelling measurements, enabling centimeter-level accuracy in mine subsidence monitoring.

What are the implications of the main findings?

• Compared with C-band Sentinel-1A, L-band SAR provides superior coherence preservation and deformation sensitivity, offering a practical solution for operational monitoring in active mining areas characterized by rapid subsidence, anthropogenic disturbances, and dense vegetation.
• The results confirm the advantages of L-SAR for mine subsidence monitoring and provide scientific evidence and technical guidance for applying InSAR-based approaches to mining hazard risk management.

Abstract

High-intensity mining activities in coal mining areas have produced large-gradient surface deformation, posing severe challenges to deformation monitoring using Interferometric Synthetic Aperture Radar (InSAR) techniques based on C-band Synthetic Aperture Radar (SAR) data. This study systematically evaluated the applicability of L-band LuTan-1 SAR (L-SAR) data versus C-band Sentinel-1A data for monitoring mining-induced surface deformation, using the Guqiao Coal Mine in Huainan as the study area. Based on 10 ascending-track and 13 descending-track L-SAR images and 42 Sentinel-1A images, deformation retrievals were performed using Differential InSAR (DInSAR) and the Small Baseline Subset (SBAS) InSAR approach, respectively, and the results were validated against independent levelling measurements. Results indicate that the mean coherence of descending- and ascending-track L-SAR interferometric pairs are 0.42 and 0.45, respectively, substantially higher than Sentinel-1A’s 0.25. In the DInSAR analysis along profile A–A′, the maximum line-of-sight (LOS) displacement obtained from descending- and ascending-track L-SAR are −0.40 m and −0.43 m, respectively, compared with −0.25 m from Sentinel-1A. In the SBAS-InSAR time-series analysis, descending- and ascending-track L-SAR yield 209,418 and 228,388 coherent points, respectively, clearly revealing the temporal evolution of surface deformation; their maximum LOS deformation rates are approximately −1.54 m·yr⁻¹ and −2.0 m·yr⁻¹, respectively. By contrast, Sentinel-1A selects only 81,669 coherent points, with severe loss of coherence in the subsidence center and a maximum LOS deformation rate of about −0.48 m·yr⁻¹. Accuracy validation shows that the Root Mean Square Error (RMSE) of vertical displacements obtained from DInSAR monitoring results based on descending and ascending L-SAR data is 16.1 mm, satisfying the requirement of centimeter-level accuracy for mining area surface subsidence monitoring. The study demonstrates the pronounced advantages of L-SAR for monitoring large-gradient, nonlinear deformation in mining environments. L-band data outperform C-band Sentinel-1A across coherence preservation, deformation sensitivity, and monitoring accuracy, providing a scientific basis for the broader application of domestic L-band SAR satellites in disaster risk assessment and long-term time-series monitoring of mining-induced subsidence.

1. Introduction

Underground coal extraction-induced ground deformation has become a global environmental concern. In particular, intensive and long-duration coal mining can cause extensive surface collapse, ground fissuring, and perturbations of the groundwater table, posing severe threats to regional infrastructure, land use, and ecosystems [1,2,3]. The Huainan mining district in eastern China is characterized by a high groundwater table, thick Quaternary unconsolidated deposits, and extensive coal seam extraction; these conditions have produced frequent, large-magnitude, and spatially extensive surface deformation, together with localized ponding within subsidence zones. Such complex deformation processes impose elevated requirements on precise deformation monitoring [4].

Although conventional precise levelling and GNSS provide high-accuracy point observations, their limited spatial coverage and high maintenance costs make them inadequate for large-area monitoring and risk assessment of rapidly evolving mine subsidence [5,6]. Consequently, establishing monitoring techniques that deliver high accuracy, broad spatial coverage, and temporal continuity is essential for preventing and mitigating subsidence hazards and ensuring operational safety [7]. Interferometric Synthetic Aperture Radar (InSAR) provides all-weather, day-night, high-precision, and wide-area continuous observation capabilities [8,9], and has been widely applied to monitoring mining-induced subsidence [10], landslides [11], ground fissures [12], co-seismic deformation fields [13], volcanic deformation associated with eruptions [14], and urban surface deformation [15]. Differential InSAR (DInSAR), which derives deformation directly from the phase difference between two SAR acquisitions, is well suited for rapid detection of short-term or abrupt subsidence events; in contrast, the Small Baseline Subset (SBAS) InSAR approach constructs a multi-temporal, small-baseline interferogram network and separates non-deformation signals such as atmospheric and orbital contributions to retrieve time series of surface deformation [16,17]. Through the combined application of DInSAR and SBAS-InSAR, InSAR effectively compensates for the limitations of conventional methods and can reveal the spatiotemporal evolution of mine subsidence with high precision and continuity [18].

Over the past decade, next-generation SAR missions exemplified by the European Space Agency’s freely available Sentinel-1 series have substantially advanced surface-deformation monitoring. Their wide coverage and short revisit intervals have promoted InSAR from an exploratory research tool to a mature, stable, and routinely applied monitoring technique [19,20]. However, because C-band SAR employs a shorter wavelength and has weaker penetration, it experiences pronounced coherence loss over croplands, vegetated areas, or regions with large deformation gradients, often preventing retrieval of reliable surface-deformation signals [21]. By contrast, L-band SAR inherently offers superior penetration and greater preservation of scatterer coherence, and is theoretically better able to maintain coherence and more accurately capture large-gradient deformation under complex surface conditions [22]. LuTan-1, as China’s first L-band differential interferometric SAR dual-satellite constellation, features a dual-satellite flying-around formation configuration that enables near-zero temporal baseline interferometry, effectively eliminating temporal decorrelation caused by surface changes. This represents a technical advantage unattainable by single-satellite repeat-pass systems such as ALOS-2. Furthermore, LuTan-1’s repeated observation modes and multi-polarization configurations, optimized for mining area monitoring, maintain phase continuity in high-gradient deformation zones and improve phase unwrapping reliability. Compared to ALOS-2’s 14-day revisit cycle, LuTan-1 achieves an effective revisit interval as short as 4 days through dual-satellite coordination, significantly enhancing the temporal resolution for capturing rapid, nonlinear subsidence processes in active mines [23,24].

Numerous studies have evaluated differences between C- and L-band SAR for monitoring earthquakes, landslides, volcanic activity, and glacier dynamics. For example, Taewook Kim et al. [25] analyzed the Mw 6.7 earthquake at Lake Hovsgol, Mongolia, using ascending and descending L-band ALOS-2 and descending C-band Sentinel-1B data to derive three-dimensional co-seismic surface displacements; their results showed that ALOS-2 maintained higher coherence, and the maximum LOS displacements measured by Sentinel-1B and ALOS-2 ascending/descending tracks were 22 cm, 33 cm, and 28 cm, respectively. Jin Deng et al. [26] compared Sentinel-1 and ALOS-2 using SBAS-InSAR in the steep terrain of Mao County, China, and found that L-band outperformed C-band in coherence preservation, observation coverage, and displacement retrieval, whereas C-band exhibited greater sensitivity to small-magnitude displacements. Yuji Himematsu et al. [27] conducted time-series analysis with L-band PALSAR-2 and C-band Sentinel-1 at Shinmoe-dake volcano, Japan, revealing coexisting flank inflation and crater contraction before and after the 2017 eruption; this study indicated that joint L- and C-band time-series observations provide more complete constraints on subsurface volcanic processes and improve understanding of the complex deformation dynamics associated with eruptions. Zhaohua Chen et al. [28] applied SBAS-InSAR with C-band Radarsat-2 and Sentinel-1 and L-band ALOS-2 to monitor Arctic sea-ice deformation in Nunavut, Canada, finding generally consistent spatiotemporal deformation patterns across sensors: C-band often preserves higher coherence on short timescales, while L-band shows advantages for stable observation of larger-scale and high-gradient deformation. Notably, existing comparative studies between L-band and C-band have primarily focused on single-satellite systems such as ALOS-2 and PALSAR-2, whose long revisit periods limit their applicability in rapidly subsiding mining areas. LuTan-1’s dual-satellite formation design not only inherits the inherent penetration advantages of L-band but also overcomes the temporal resolution bottleneck of conventional repeat-pass InSAR through dual-baseline interferometry, offering new technical possibilities for high-frequency monitoring in mining districts. For active mining areas, which represent complex scenarios characterized by rapid subsidence, anthropogenic disturbances, and extensive vegetation or water coverage, systematic comparative studies between LuTan-1 SAR (L-SAR) and Sentinel-1 remain lacking, particularly regarding their differences in spatiotemporal coverage, coherence preservation, and deformation detection capability.

Therefore, this study takes the Guqiao Mine, Huainan, Anhui Province as the study area to conduct a systematic comparative evaluation of L-band LuTan-1 and C-band Sentinel-1A for monitoring mine-induced surface deformation. Both datasets are processed using DInSAR and SBAS-InSAR; we quantitatively compare interferogram mean coherence, time-series deformation fields, deformation rates, and profile and point-based deformation characteristics, and perform accuracy assessment using independent levelling measurements. The results validate the advantages of L-SAR for mine subsidence monitoring and provide scientific evidence and technical guidance for employing InSAR-based methods in mining hazard risk management.

2. Study Area and Datasets

2.1. Study Area

Guqiao Mine is situated within Guqiao Town, in the northwestern part of Fengtai County, Huainan City, Anhui Province. It lies in the central-western sector of the Panxie mining subbasin of the Huainan coalfield, approximately 20 km east of Fengtai county seat. The mine’s geographic coordinates are 116°32′05″ to 116°38′49″E and 32°43′47″ to 32°51′50″N, and the mining district covers an area of roughly 92 km². The site is located on the alluvial plain of the middle Huai River, characterized by broad, flat terrain with ground elevations typically between 21 and 24 m and a gentle tilt that is higher in the northwest and lower in the southeast. The region experiences a warm-temperate, semi-humid monsoon climate and is well supplied with surface water; the mainstem of the Huai River flows south of the mine, while tributaries such as the West Fei River and the Yongxing River encircle the surrounding area.

Guqiao Mine has been designated a national key construction mine and represents the largest underground coal-mining operation in Asia by excavation scale, with an approved annual production capacity of 10 million tonnes. The mine’s geological coal reserves amount to approximately 1.82 billion tonnes, with recoverable reserves approaching 1.0 billion tonnes. The coal exhibits favorable quality, and the mine is classified as a “three-high” operation, characterized by high gas content, high ground temperature, and high geostatic pressure. As a backbone mine within the Huainan district, its development plays an important strategic role in regional energy supply for eastern China. The Huainan mining district generally suffers from a high groundwater table, thick unconsolidated overburden, and mining-induced subsidence and ponding. Against this background, the large-scale extraction, complex hydrogeological conditions, and “three-high” characteristics of Guqiao Mine make it a representative study area for investigating mining-induced surface deformation, subsidence evolution, and associated hydrological responses. The mine location is shown in Figure 1.

2.2. Datasets

To evaluate and compare the capability of LuTan-1 (L-band) and Sentinel-1A (C-band) data for monitoring mining-induced deformation, this study acquired one ascending-track and one descending-track dataset of L-SAR covering the study area, together with one ascending-track Sentinel-1A dataset. The L-SAR dataset comprises 10 ascending-track acquisitions and 13 descending-track acquisitions, while the Sentinel-1A ascending-track dataset comprises 42 acquisitions. The spatial coverage of the L-SAR and Sentinel-1A images is illustrated in Figure 1. SAR image metadata and acquisition times are listed in

Table 1. SAR data imaging information.

Data Type	Band	Wavelength (cm)	Incidence Angle (°)	Azimuth Angle (°)	Spatial Resolution (m) (Range × Azimuth)	Swath Width (km)	Revisit Cycle (Days)
L-SAR Ascending	L	23.8	32.93	349.45	1.7 × 1.7	50	28
L-SAR Descending	L	23.8	30.29	190.81	1.7 × 1.4	50	28
Sentinel-1A	C	5.6	33.74	347.14	2.3 × 14.0	250	12

and

Table 2. SAR data acquisition times.

Number	L-SAR Ascending	L-SAR Descending	Sentinel-1A
1	20231112	20240101	20231109	20240425	20241010
2	20231210	20240129	20231121	20240507	20241022
3	20240331	20240226	20231203	20240519	20241103
4	20240526	20240325	20231215	20240531	20241115
5	20240623	20240422	20231227	20240612	20241127
6	20240721	20240520	20240108	20240624	20241209
7	20240818	20240617	20240120	20240706	20241221
8	20240915	20240715	20240201	20240718	20250102
9	20241110	20240812	20240213	20240730	20250114
10	20241208	20240909	20240225	20240811	20250126
11		20241104	20240308	20240823	20250207
12		20241202	20240320	20240904	20250219
13		20241230	20240401	20240916	20250303
14			20240413	20240928	20250315

.

To support SAR-derived deformation retrieval, a 30 m resolution COP-DEM covering the study area was also acquired. This digital elevation model is produced from the same source data as TanDEM-X and is provided by the European Space Agency via the Open Topography portal [29].

3. Methods

3.1. DInSAR Method

DInSAR primarily registers two or more SAR images that cover the same target area to sub-pixel accuracy and then forms interferograms from the co-registeredd SAR images. By analyzing the differences between the echo signals corresponding to the same ground point in the two acquisitions, the elevation of the target and its temporal changes can be retrieved [30,31]. The interferometric phase can be expressed as:

$$△\varphi ={\varphi }_{flat}+{\varphi }_{topo}+{\varphi }_{def}+{\varphi }_{atm}+{\varphi }_{orb}+{\varphi }_{noise}$$

(1)

In the equation, ${\varphi }_{flat}$ denotes the flat-earth phase produced by the reference ellipsoid; ${\varphi }_{topo}$ denotes the topographic phase caused by terrain relief; ${\varphi }_{def}$ denotes the deformation phase induced by surface deformation occurring between the two observations; ${\varphi }_{atm}$ denotes the atmospheric phase produced by atmospheric disturbances; ${\varphi }_{orb}$ denotes the orbital error phase arising from errors in orbit information; and ${\varphi }_{noise}$ denotes the noise phase caused by various random noise sources. After differential interferometric processing, the phase component attributable to surface deformation can be obtained.

Ultimately, surface deformation can be retrieved by converting phase to displacement [32]:

$${\varphi }_{def}=-{\displaystyle \frac{4\pi \cdot D_{LOS}}{\lambda }}$$

(2)

In the equation, $D_{LOS}$ represents the radar LOS deformation of the ground target, and $\lambda$ denotes the radar signal wavelength.

3.2. SBAS-InSAR Method

In 2004, Hooper et al. proposed the StaMPS framework, which supports both single-master (PS) and multi-master (SBAS) InSAR analyses [33]. SBAS-InSAR first selects pixels that meet an amplitude-dispersion-index threshold within the assembled set of differential interferograms as candidate high-coherence points. Phase analysis is then employed to estimate the phase stability of each candidate; candidates are screened using a temporal-coherence coefficient threshold and iteratively refined. Finally, three-dimensional phase unwrapping is applied to retrieve the time-series deformation phase of the high-coherence points, thereby deriving surface deformation information [34,35]. The SBAS-InSAR data-processing workflow is shown in Figure 2, and the principal steps are described as follows.

(1)

Differential interferogram generation. Assuming N + 1 SAR acquisitions, one image is chosen as the common master image based on temporal and spatial baselines and Doppler centroid frequency, and the remaining N images are treated as secondary images to be co-registered to the master. According to prescribed temporal and spatial baseline thresholds, the N + 1 images are combined into M SBAS time-series interferometric pairs, where M satisfies (N + 1)/2 ≤ M ≤ N(N + 1)/2, and the interferometric pairs must form a fully connected network. The assembled SBAS time-series interferometric pairs are co-registered, interfered, corrected for flat-earth phase, and filtered to produce the corresponding sequence of differential interferograms [9].

(2)

Selection of Slowly Decorrelating Filtered Phase (SDFP) target points. Amplitude information is analyzed, and unlike the conventional amplitude-dispersion threshold, the initial candidates are selected using the amplitude-dispersion index. For Gaussian scattering pixels, the amplitude-dispersion index $D_{\mathrm{\Delta }A}$ can be expressed by Equation (3) (the threshold $D_{\mathrm{\Delta }A}$ is typically set to 0.6).

$$D_{\Delta A}={\displaystyle \frac{{\sigma }_{\Delta A}}{{\mu }_A}}={\displaystyle \frac{\sqrt{\frac{{\displaystyle \sum _{i=1}^M{\left(A_m^i-A_s^i\right)}^2}}{N}}}{\frac{{\displaystyle \sum _{i=1}^M{A}_m^i+A_s^i}}{2N}}}$$

(3)

(3)

Construction and optimization of the temporal-coherence coefficient for SDFP points, as given in Equation (4).

$${\gamma }_{\kappa }={\displaystyle \frac{1}{N}}\left|{\displaystyle {\sum }_{i=1}^{N}exp\left\{j\left({\varphi }_{diff}^i-{\varphi }^i-{\varphi }_{\theta }^{i,u}\right)\right\}}\right|$$

(4)

In the equation, ${\varphi }^i$ denotes the spatially correlated phase and ${\varphi }_{\theta }^{i,u}$ denotes the spatially uncorrelated view-angle error phase. The temporal-coherence coefficient ${\gamma }_{\kappa }$ reflects the coherence of an SDFP point; the larger the ${\gamma }_{\kappa }$ value, the higher the correlation. A coherence weighting ${\gamma }_{\kappa }$ can be used to further optimize SDFP points selection.

(2)

Estimation of deformation rate and time-series deformation. Using the selected SDFP points, a Delaunay triangulation is constructed and a three-dimensional phase-unwrapping algorithm is applied to obtain the unwrapped phase for each high-coherence point. Exploiting the differing temporal and spatial characteristics of orbital phase, deformation phase, and atmospheric noise, temporal-domain low-pass and spatial-domain high-pass filters are applied to separate and remove undesired components. The resulting deformation phase for each high-coherence point is then used to derive the deformation rate and the time-series displacement of the SDFP points.

4. Results

4.1. DInSAR Results

4.1.1. Interferogram Processing, Phase Unwrapping, and Coherence Analysis

To ensure coherence, DInSAR deformation extraction was performed using a pairwise short-baseline processing strategy for adjacent images. Based on this approach, selected summer and winter interferometric pairs from the ascending- and descending-track L-SAR datasets and from the Sentinel-1A ascending-track dataset were processed; the resulting differential interferograms, unwrapped-phase maps, and coherence maps are shown in Figure 3. Comparative analysis indicates that both sensor types can identify deformation zones, but their monitoring performance differs markedly due to wavelength disparities.

With respect to differential interferogram appearance, L-SAR, owing to its longer wavelength, exhibits stronger penetration and superior coherence; deformation fringes are distinct and can be used to coarsely estimate deformation magnitude (one 2π fringe cycle corresponds to a displacement of λ/2). By contrast, Sentinel-1A, with its shorter wavelength, is limited in resolving large deformation gradients: only partial boundary fringes are discernible in the differential interferograms, fringes become increasingly blurred toward the deformation center, and severe phase ambiguity occurs, which impedes direct interpretation of deformation magnitude. Regarding unwrapped-phase characteristics, L-SAR data show good phase continuity, with deformation phase increasing continuously from the subsidence margin toward the center and only localized phase loss in waterlogged areas; Sentinel-1A unwrapped-phase continuity is poorer, with pronounced phase jumps and reduced reliability of the unwrapping results. In terms of coherence, L-SAR maintains generally higher coherence and remains relatively stable even in active deformation zones, with reduced coherence mainly confined to ponding areas; Sentinel-1A exhibits moderate overall coherence and experiences pronounced coherence loss in ponding and deformation zones, detracting from interferogram quality and monitoring accuracy. Furthermore, in areas with large subsidence, the reliability of phase unwrapping depends not only on wavelength but is also significantly affected by spatial sampling density. L-SAR data, with its high resolution and higher spatial sampling rate, can capture spatial variations in subsidence gradients more finely. In regions with large deformation gradients, such as the edges of subsidence funnels, dense sampling points help maintain phase continuity and reduce phase unwrapping errors. In contrast, Sentinel-1A has relatively lower resolution; its coarser resolution results in each pixel covering a larger ground area. When non-uniform deformation exists within a pixel, volume decorrelation and mixed-pixel effects occur, increasing the uncertainty of phase unwrapping. Overall, L-SAR demonstrates clear advantages in coherence preservation and phase quality, ensuring more reliable DInSAR monitoring results, whereas Sentinel-1A has limited applicability when comparable accuracy is required.

The foregoing qualitative analysis of differential interferograms, unwrapped phase, and coherence highlights the advantages of L-SAR data. To quantitatively validate this conclusion, Figure 4 presents statistics of the mean coherence for interferometric pairs from descending- and ascending-track L-SAR and Sentinel-1A. Due to the extensive availability of Sentinel-1A interferometric pairs, only those employed in the DInSAR experiments are presented herein. The results indicate that the mean coherence of descending- and ascending-track L-SAR interferometric pairs is markedly higher than that of Sentinel-1A. For all three datasets, interferometric pairs acquired in winter generally exhibit higher coherence than those acquired in summer; however, owing to the longer wavelength and stronger penetration of L-SAR, the seasonal coherence difference for L-SAR is small and the overall coherence level remains high. By contrast, C-band Sentinel-1A is highly susceptible to agricultural activities, vegetation cover, and large deformation gradients, resulting in substantially lower overall coherence. Among the eight Sentinel-1A interferometric pairs selected from September to December 2024, the four interferometric pairs acquired before November all have mean coherence below 0.20, rendering DInSAR-based monitoring results unreliable; although coherence improves after November, it remains clearly lower than that of contemporaneous L-SAR interferograms. Overall, the mean coherence for descending- and ascending-track L-SAR is approximately 0.42 and 0.45, respectively, while Sentinel-1A exhibits a substantially lower mean coherence of about 0.25.

4.1.2. DInSAR Monitoring Results and Comparison

To ensure temporal consistency with the Sentinel-1A interferometric pairs used in the previous section, interferometric pairs from descending- and ascending-track L-SAR acquired during the same periods were selected. After removal of the flat-earth and topographic phases and application of adaptive filtering, phase unwrapping was performed using the Minimum Cost Flow (MCF) algorithm. Following phase-to-deformation conversion and geocoding, the LOS DInSAR deformation maps for Guqiao Mine were obtained and are shown in Figure 5 and Figure 6.

For the interferograms from September to early November 2024 (Figure 5a,c), the mean coherence of both descending- and ascending-track L-SAR data is 0.27. Although the results are affected by noise, DInSAR is still capable of detecting surface deformation signals. For the interferograms from November to December 2024 (Figure 5b,d), the mean coherence of the descending- and ascending-track L-SAR interferograms increases to 0.49 and 0.57, respectively, yielding improved monitoring performance and clearer delineation of the subsidence area. By contrast, for Sentinel-1A during September to early November 2024 (Figure 6a–e), low coherence and large deformation gradients largely preclude effective detection of surface deformation by DInSAR. After November 2024 (Figure 6f–h), valid deformation signals are identifiable, but phase jumps persist in some regions, owing to the large deformation magnitudes. Compared with the contemporaneous descending- and ascending-track L-SAR DInSAR results, Sentinel-1A yields qualitatively similar spatial patterns but systematically smaller displacement magnitudes, and its effectiveness for retrieving surface deformation is markedly inferior to that of L-SAR.

To further compare the DInSAR monitoring results from descending- and ascending-track L-SAR and Sentinel-1A, cumulative deformation along profile A–A′ for November–December 2024 was selected for comparative analysis. Because Sentinel-1A has a short revisit interval, the DInSAR results of three Sentinel-1A interferometric pairs acquired between 20241103 and 20241209 were stacked to obtain cumulative deformation. The Sentinel-1A cumulative DInSAR result and the location of profile A–A′ are shown in Figure 7. These results are compared with the descending-track L-SAR DInSAR result for 20241104 to 20241202 (Figure 5b) and the ascending-track L-SAR DInSAR result for 20241110 to 20241208 (Figure 5d).

The cumulative DInSAR comparison for descending- and ascending-track L-SAR and Sentinel-1A is presented in Figure 8. The monitoring periods for descending- and ascending-track L-SAR are both 28 days, with an overall temporal offset of 6 days; the difference in single-day displacement magnitude is minor and is therefore neglected. Although the Sentinel-1A monitoring interval spans 36 days, i.e., 8 days longer than the L-SAR period, C-band coherence in vegetated areas and regions with large deformation gradients decays rapidly, and the deformation signals captured by Sentinel-1A remain weaker than those from L-SAR. All three datasets identify broadly consistent locations and extents of surface deformation, but the displacement magnitudes retrieved from L-SAR substantially exceed those from Sentinel-1A. Along profile A–A′, descending-track L-SAR records a maximum subsidence of approximately −0.40 m, ascending-track L-SAR about −0.43 m, and Sentinel-1A approximately −0.25 m.

Notably, the subsidence basins derived from descending- and ascending-track L-SAR data exhibit a clear apparent offset. Two primary factors explain this phenomenon. First, horizontal ground movement induced by mining face advance produces differing LOS contributions for ascending and descending geometries. Because the incidence directions of ascending and descending L-SAR are opposite while the horizontal movement at the mining face is generally directed toward the subsidence basin center, the horizontal displacement contributes to LOS deformation with opposite signs in the two geometries, as illustrated in Figure 9. Consequently, the LOS deformation at the same ground location differs between ascending- and descending-track observations, producing an apparent spatial shift in the derived subsidence basin. Second, geocoding offsets caused by orbit inaccuracies can produce positional discrepancies between ascending and descending-track products; differences in orbit precision between the two pass directions may therefore yield additional apparent offsets during geocoding.

4.2. SBAS-InSAR Results

4.2.1. SBAS-InSAR Parameter Settings and Coherence Analysis

Because DInSAR only monitors surface deformation over a specific observation interval, its results are constrained by the temporal separation of the two SAR acquisitions and by various decorrelation factors, and therefore it is difficult to effectively reveal long-term time-series deformation characteristics. To further compare the capability of descending- and ascending-track L-SAR and Sentinel-1A for time-series monitoring of mine-induced surface deformation, SBAS-InSAR processing was applied to each dataset to retrieve temporal deformation fields for the study area.

During SBAS-InSAR processing, and to ensure overall interferogram coherence, interferometric pairs for both descending- and ascending-track L-SAR were generated with a temporal baseline of 56 days, while Sentinel-1A interferometric pairs were configured with a temporal baseline of 24 days. Due to the limited availability of descending- and ascending-track L-SAR images, no spatial baseline constraints were imposed on the L-SAR and Sentinel-1A data to ensure the formation of a complete interferometric network. To improve SBAS-InSAR monitoring performance, interferometric pairs exhibiting severe decorrelation or deformation-phase aliasing were removed; ultimately, 15 and 11 interferometric pairs were selected for the descending and ascending-track L-SAR datasets, respectively, and 81 interferometric pairs were selected for the Sentinel-1A dataset. The spatiotemporal baselines of the selected interferometric pairs for the three datasets are shown in Figure 10.

To evaluate the coherence of the selected interferometric pairs for the three datasets, this study compared the mean coherence of SBAS-InSAR interferometric pairs derived from descending- and ascending-track L-SAR and from Sentinel-1A; the results are presented in Figure 11. The mean coherence of the L-SAR descending- and ascending-track interferometric pairs is 0.41 and 0.43, respectively, which is substantially higher than the mean coherence of 0.19 for Sentinel-1A. Although Sentinel-1A interferometric pairs have shorter temporal baselines (12 days), whereas the shortest temporal baseline among L-SAR interferometric pairs is 28 days and some L-SAR pairs exceed 28 days, Sentinel-1A’s shorter C-band wavelength and weaker penetration make it difficult to compensate for coherence loss caused by agricultural fields and vegetation in the mining area, resulting in low coherence across most regions except built-up areas. By contrast, L-SAR interferometric pairs maintain high coherence over most land-cover types except water bodies. These results indicate that L-band L-SAR data exhibit a markedly superior ability to preserve interferogram coherence for deformation monitoring in mining areas compared with C-band Sentinel-1A data.

To retrieve surface subsidence information in the study area, this study employed the StaMPS (v3.3) software to conduct SBAS-InSAR time-series deformation inversion using L-SAR and Sentinel-1A imagery data. The key parameters adopted for SBAS-InSAR time-series deformation analysis of both datasets are listed in

Table 3. Parameters used for SBAS-InSAR deformation analysis of the two datasets.

Parameter	Parameter Function	L-SAR	Sentinel-1A
max_topo_err	Estimating phase-unwrapped DEM phase, filtering, and eliminating coherent points with large noise	5	5
filter_grid_size		50	32
weed_standard_dev		5	1
weed_max_noise		10	5
unwrap_prefilter_flag	Phase unwrapping	y	y
unwrap_grid_size		8	8
unwrap_time_win		12	50
scla_deramp	Removal of orbital, DEM, and atmospheric phases	y	y

. Given the generally low coherence of Sentinel-1A data, several parameter thresholds for noise suppression were moderately relaxed during the inversion process to ensure sufficient coherent point density for time-series inversion, thereby improving the spatial coverage and stability of the deformation time series. The SBAS-InSAR time-series deformation results for Sentinel-1A are detailed in Section 4.2.2.

4.2.2. SBAS-InSAR Monitoring Results and Comparison

Using descending- and ascending-track L-SAR and Sentinel-1A datasets, SBAS-InSAR time-series analysis was conducted with the StaMPS software to derive LOS deformation rates and their standard deviations for the three datasets; the results are presented in Figure 12 and Figure 13.

In terms of coherent-point density, the descending- and ascending-track L-SAR datasets yielded 209,418 and 228,388 coherent points, respectively, with comparable counts and dense, uniform spatial distributions; by contrast, the Sentinel-1A dataset produced only 81,669 coherent points, demonstrating substantially lower density. The primary reason for this difference lies in the fact that L-SAR has longer wavelength, stronger penetration capability, and higher spatial resolution, exhibiting good coherence under temporal baseline control. Moreover, L-SAR’s high resolution effectively reduces intra-pixel averaging effects, enabling more accurate localization of subsidence extrema and boundaries, thus maintaining a higher density of coherent points even in areas with intense deformation. In comparison, Sentinel-1A’s C-band has shorter wavelength, weaker penetration capability, and lower spatial resolution. For large-gradient deformation areas exceeding 1 m, Sentinel-1A’s low resolution may result in each pixel containing multiple subsidence centers or complex deformation patterns, producing an averaging effect that potentially underestimates maximum subsidence and blurs subsidence boundaries. Regarding deformation retrieval capability, descending-track L-SAR clearly captures deformation features at the edges of the subsidence funnel, whereas ascending-track L-SAR, having fewer and temporally less continuous images, exhibits localized decorrelation that hinders full delineation of the subsidence zone. Sentinel-1A, owing to insufficient coherence, fails to resolve the overall deformation pattern and only detects subsidence at the periphery. The monitoring results indicate maximum LOS deformation rates of approximately −1.54 m·yr⁻¹ and −2.0 m·yr⁻¹ for descending- and ascending-track L-SAR, respectively, which are substantially larger than the −0.48 m·yr⁻¹ observed for Sentinel-1A. These findings indicate that L-band is more suitable for monitoring large-gradient deformation and more faithfully reflects actual mining-induced ground movement.

Furthermore, because mining-induced deformation often exhibits pronounced nonlinearity, SBAS-InSAR results in deforming areas display relatively large standard deviations, while non-deforming areas show smaller standard deviations. Notably, the standard deviations derived from L-SAR data are generally larger than those from Sentinel-1A, which is mainly attributable to L-band’s greater sensitivity to nonlinear deformation. Standard deviation here is computed from the residuals of the linear fit to the deformation rate; therefore, the more significant the nonlinear behavior within the monitored area, the larger the standard deviation. Owing to its longer wavelength and capability to capture broader deformation gradients, L-SAR produces monitoring results that more closely match the actual deformation process, and consequently exhibits larger standard deviations in regions of intense deformation.

Figure 14 presents the time-series deformation of the study area derived from descending-track L-SAR data using the SBAS-InSAR method. The selected coherent points are numerous and densely distributed, providing relatively complete temporal deformation information. Three deformation zones are identified, with clearly delineated locations and extents and evident temporal evolution. The lower deformation zone was first detected and exhibited increasing spatial extent and magnitude from 20240101 to 20240422; thereafter the lower zone stabilized with minor changes, indicating that the underlying coal face may have ceased extraction and subsequent surface movement is dominated by residual deformation. The central deformation zone showed no obvious activity before 20240422, but from 20240520 its spatial extent and magnitude increased progressively and continued to develop through 20241230. The evolution direction of surface deformation suggests that the coal-face advance is from west to east and that extraction is ongoing. The upper deformation zone began to register deformation signals from 20241202, and both its spatial extent and magnitude increased markedly with time, indicating that the associated coal face is in an early stage of extraction. The maximum cumulative surface deformation measured from descending-track L-SAR with SBAS-InSAR is approximately −1.51 m, located in the central deformation zone.

Figure 15 shows the time-series deformation derived from ascending-track L-SAR data with SBAS-InSAR. The number of selected coherent points is sufficient, and the results overall reflect the temporal evolution of surface deformation; the general temporal trend is consistent with the descending-track result. However, the upper deformation zone is not conspicuous in the ascending-track result, primarily because only an ascending-track image acquired on 20241208 was available; at that time, the upper zone had just begun to develop and the displacement magnitude was small, rendering it less apparent in the ascending-track product. The maximum cumulative surface deformation derived from ascending-track L-SAR and SBAS-InSAR is approximately −2.15 m, located in the central deformation zone.

Figure 16 presents the time-series deformation retrieved from Sentinel-1A using SBAS-InSAR. Owing to the short C-band wavelength and coherence decay associated with increasing deformation gradient, coherent points are severely lacking in the subsidence center, resulting in incomplete delineation of deformation zones in the Sentinel-1A product. Nevertheless, the detected deformation locations and extents are generally consistent with those from L-SAR. Sentinel-1A resolves temporal changes for the middle and lower deformation zones only, and their boundaries are blurred; the upper deformation zone is not clearly represented. Moreover, the temporal pattern of the lower deformation zone inferred from Sentinel-1A differs from that of L-SAR: the lower zone shows a continuous increase throughout the monitoring period. This behavior is attributable to the limited capability of short-wavelength C-band data to resolve large-gradient deformation in time-series inversion, which effectively smooths large-gradient deformation across the entire temporal window, causing underestimation of instantaneous deformation rates while producing a continuously increasing cumulative displacement. The maximum cumulative displacement derived from Sentinel-1A is only approximately −0.64 m, substantially lower than the SBAS-InSAR results from L-SAR.

In summary, L-SAR and Sentinel-1A show good agreement in identifying the locations and extents of deformation, but significant differences exist in displacement magnitude, temporal evolution detail, and spatial completeness due to disparities in wavelength, resolution, and coherence preservation. The long-wavelength characteristic of L-SAR renders it more effective for monitoring large-gradient, rapidly evolving surface deformation in mining areas.

To further compare differences between L-SAR and Sentinel-1A in SBAS-InSAR monitoring of mining-induced deformation, profile A–A′ and four characteristic points (P1, P2, P3, and P4) were selected; the locations of the profile and points are shown in Figure 12. The cumulative time-series surface deformation derived from the two datasets was compared, and the results are presented in Figure 16 and Figure 17.

Figure 17 displays the cumulative time-series deformation along profile A–A′ derived from SBAS-InSAR using descending- and ascending-track L-SAR and Sentinel-1A data. The figure indicates that the three datasets exhibit broadly similar trends and spatial extents of cumulative deformation along profile A–A′, whereas the displacement magnitudes differ substantially. Owing to the longer wavelength of L-band, the maximum deformations retrieved from descending- and ascending-track L-SAR are approximately −1.5 m and −2.1 m, respectively, while the maximum deformation from Sentinel-1A is only about −0.6 m. In addition, the locations of the maximum deformations differ: L-SAR places the maximum near 1100 m along profile A–A′, whereas Sentinel-1A locates it near 1500 m.

Figure 18 presents the time-series deformation comparison for the four selected points (P1–P4). In regions with small-magnitude deformation, the temporal trends and magnitudes from Sentinel-1A and L-SAR are generally consistent; however, where deformation gradients are large, the short-wavelength C-band signal suffers from phase aliasing and decorrelation, which together limit Sentinel-1A’s accuracy for rapid deformation signals. Sentinel-1A therefore tends to indicate only the deformation trend and underestimates cumulative displacement. By contrast, L-SAR’s long wavelength preserves high coherence and robustly captures nonlinear deformation throughout the monitoring period, yielding cumulative displacements that better reflect the temporal evolution of deformation. Consequently, L-SAR provides markedly higher accuracy and reliability than Sentinel-1A for monitoring large-gradient, nonlinear deformation in mining areas, and is thus better suited for long-term SBAS-InSAR analysis in zones of intensive extraction.

5. Discussion

To establish a reliable ground elevation benchmark for InSAR result validation, this study employed fourth-order levelling measurement methods for field observations. Observations were conducted using calibrated digital levels with calibrated staves, with survey lines arranged in forward and backward runs to improve accuracy and facilitate closure error checking. The benchmarks for each survey line utilized established stable permanent benchmarks, and levelling points were established based on stable wooden stakes with point intervals of 100–150 m. The two levelling lines were 1.7 km and 1.4 km in length, respectively.

To validate the accuracy of L-SAR for monitoring surface deformation in the mining area, six L-SAR images were acquired: ascending-track 20250330, 20250427 and 20250525, and descending-track 20250324, 20250421 and 20250519. To approximate the L-SAR revisit cadence, three fourth-order precise levelling surveys were conducted on 29 March 2025, 26 April 2025, and 26 May 2025. Due to the limited availability of additional L-SAR data, only DInSAR results are presented for comparison. The DInSAR monitoring results for the corresponding periods are shown in Figure 19.

Based on the D-InSAR monitoring results derived from ascending and descending L-SAR data, two-dimensional deformation decomposition was performed to obtain vertical displacements, which were then compared with levelling measurements. Due to water accumulation in Area B leading to poor decomposition results, the second epoch data from Area A was selected for accuracy validation, with results shown in Figure 20 and Figure 21 and

Table 4. Accuracy assessment results.

Maximum Error (mm)	Minimum Error (mm)	Standard Deviation (mm)	RMSE (mm)
26.4	1.6	17.0	16.1

. Overall, the vertical displacements obtained from two-dimensional decomposition are slightly larger than the levelling results. This is primarily because the actual observation period of ascending and descending L-SAR data is approximately 4 days longer than the levelling survey period, and the study area is in an active mining stage with rapid daily subsidence rates, resulting in higher cumulative subsidence. Nevertheless, the Root Mean Square Error (RMSE) between levelling data and two-dimensional decomposition vertical displacements is 16.1 mm, validating that DInSAR technology based on L-SAR data can achieve centimeter-level accuracy for mining area surface deformation monitoring, and the obtained results possess high precision and reliability.

6. Conclusions

This study uses descending- and ascending-track L-band L-SAR and Sentinel-1A imagery, and applies DInSAR and SBAS-InSAR techniques to perform an integrated monitoring and comparative analysis of surface deformation at the Guqiao mine in Huainan. The main conclusions are as follows:

(1)

L-band L-SAR demonstrates clear superiority over C-band Sentinel-1A in terms of interferometric coherence within the mining area. In the interferometric pairs analyzed, descending and ascending L-SAR data achieve mean coherence values of approximately 0.42 and 0.45, respectively, substantially exceeding the 0.25 observed for Sentinel-1A. Under conditions of extended temporal baselines and surface coverage by cropland and vegetation, L-SAR maintains high coherence across most land cover types except water bodies, benefiting from its longer wavelength and finer spatial sampling rate. In contrast, the relatively coarse resolution of Sentinel-1A results in larger ground areas covered by each pixel, which leads to volume decorrelation and mixed pixel effects when non-uniform deformation occurs within a pixel. These effects increase the uncertainty in phase unwrapping and cause rapid coherence loss in areas of intense deformation and vegetation cover, rendering it inadequate for high-precision deformation retrieval.

(2)

In DInSAR deformation monitoring, L-SAR more faithfully captures the magnitude of large-gradient mining subsidence. While all three datasets demonstrate good consistency in identifying the subsidence location and overall extent, the maximum LOS displacements along profile A–A′ differ substantially between sensors. Descending and ascending L-SAR data yield approximately −0.40 m and −0.43 m, respectively, whereas Sentinel-1A measures only about −0.25 m. This pronounced discrepancy in magnitude indicates that L-SAR, benefiting from the longer wavelength and higher spatial resolution of L-band, exhibits stronger deformation detection capability in large-gradient deformation zones.

(3)

In SBAS-InSAR time-series inversion, L-SAR data are able to fully reveal the nonlinear evolution of the subsidence areas. The descending- and ascending-track L-SAR datasets yield 209,418 and 228,388 coherent points, respectively. In the time-series deformation fields, three subsidence zones and their evolution are clearly identifiable: the lower subsidence zone appeared first and gradually stabilized, the middle subsidence zone continued to develop, and the upper subsidence zone was in the initial stage of mining. The maximum LOS deformation rates for descending- and ascending-track L-SAR data are approximately −1.54 m·yr⁻¹ and −2.0 m·yr⁻¹, respectively. By contrast, Sentinel-1A selects only 81,669 coherent points; severe loss of coherent points in the central subsidence area leads to blurred boundaries, smoothing and underestimation of deformation magnitudes, and a maximum deformation rate of only about −0.48 m·yr⁻¹, which prevents accurate characterization of rapid, nonlinear subsidence processes in the mine.

(4)

The vertical displacements obtained from D-InSAR monitoring results based on L-SAR data show high consistency with levelling measurements, enabling centimeter-level accuracy for mining subsidence monitoring. Accuracy assessment against fourth-order levelling results indicates that D-InSAR results from descending and ascending L-SAR data can reproduce the subsidence zones spatially, with subsidence extent and magnitude generally agreeing with levelling observations, achieving a RMSE of 16.1 mm. These results validate that DInSAR technology based on L-SAR data can achieve centimeter-level accuracy for mining area surface deformation monitoring, and the obtained results possess high precision and reliability.