Nothing Special   »   [go: up one dir, main page]
Next Article in Journal
Assessment of Vegetation Drought Loss and Recovery in Central Asia Considering a Comprehensive Vegetation Index
Previous Article in Journal
Global Semantic Localization from Abstract Ellipse-Ellipsoid Model and Object-Level Instance Topology
You seem to have javascript disabled. Please note that many of the page functionalities won't work as expected without javascript enabled.
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Remote Sensing
Volume 16
Issue 22
10.3390/rs16224188
remotesensing-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Long-Term Ground Deformation Monitoring and Quantitative Interpretation in Shanghai Using Multi-Platform TS-InSAR, PCA, and K-Means Clustering

by
Yahui Chong
1,2 and
Qiming Zeng
1,2,*
1
Institute of Remote Sensing and Geographic Information System, School of Earth and Space Sciences, Peking University, Beijing 100871, China
2
Spatial Information Integration and 3S Engineering Application Beijing Key Laboratory, Beijing 100871, China
*
Author to whom correspondence should be addressed.
Remote Sens. 2024, 16(22), 4188; https://doi.org/10.3390/rs16224188
Submission received: 13 October 2024 / Revised: 6 November 2024 / Accepted: 8 November 2024 / Published: 10 November 2024
Browse Figures
Figure 1
<p>The geographic location of Shanghai and spatial coverage of SAR datasets.</p> "> Figure 2
<p>SAR image acquisition dates and the spatial baselines of three SAR sensors with respect to the reference image for each sensor. The star represents the reference image of each sensor.</p> "> Figure 3
<p>The data processing flowchart of this study.</p> "> Figure 4
<p>Spatial–temporal baseline configurations of SAR datasets: (<b>a</b>) ALOS-1 PALSAR, (<b>b</b>) ENVISAT ASAR, and (<b>c</b>) Sentinel-1A.</p> "> Figure 5
<p>Annual LOS deformation rates obtained from three SAR datasets: (<b>a</b>) ALOS-1 PALSAR, (<b>b</b>) ENVISAT ASAR, and (<b>c</b>) Sentinel-1A.</p> "> Figure 6
<p>(<b>a</b>) Average vertical deformation rate and (<b>b</b>) time series cumulative surface deformation obtained from ALOS-ENVISAT-S1A fusion results for Shanghai from 2007 to 2018.</p> "> Figure 7
<p>Comparisons of long-term time series cumulative deformation obtained by ALOS-ENVISAT-S1A and self-weight consolidation settlement model: (<b>a</b>) P1, (<b>b</b>) P2, (<b>c</b>) P3, and (<b>d</b>) P4.</p> "> Figure 8
<p>(<b>a</b>) Scatterplot and (<b>b</b>) correlation coefficient graph of TS-InSAR deformation rate and field measurements.</p> "> Figure 9
<p>Landsat TM/ETM optical images of Pudong International Airport for the following years: (<b>a</b>) 2007, (<b>b</b>) 2010, (<b>c</b>) 2015, and (<b>d</b>) 2018.</p> "> Figure 10
<p>Vertical annual deformation rates of Pudong International Airport during (<b>a</b>) 2007–2010 and (<b>b</b>) 2015–2018 (base image is from Google Map).</p> "> Figure 11
<p>PCA result derived from ALOS-ENVISAT: (<b>a</b>) variance explained by the PC 1–4 and (<b>b</b>) eigenvectors obtained from PC 1–4.</p> "> Figure 12
<p>PCA result derived from Sentinel-1A: (<b>a</b>) variance explained by the PC 1–4 and (<b>b</b>) eigenvectors obtained from PC 1–4.</p> "> Figure 13
<p>Correlation map between eigenvectors of PC 1–3 obtained from ALOS-ENVISAT and temperature, groundwater level, precipitation, groundwater extraction volume, and impervious surface area: (<b>a</b>) PC1, (<b>b</b>) PC2, and (<b>c</b>) PC3.</p> "> Figure 14
<p>Correlation map between eigenvectors of PC 1–2 obtained from Sentinel-1A and temperature, groundwater level, precipitation, groundwater extraction volume, and impervious surface area: (<b>a</b>) PC and (<b>b</b>) PC2.</p> "> Figure 15
<p>The deformation rate of Pudong International Airport on the east–west profile line: (<b>a</b>) AB profile and (<b>b</b>) CD profile.</p> "> Figure 16
<p>P1–P7 time series cumulative deformation acquired by ALOS-ENVISAT in the runway and terminal areas.</p> "> Figure 17
<p>Q1–Q7 time series cumulative deformation acquired by Sentinel-1A in the runway and terminal areas.</p> "> Figure 18
<p>K-means clustering results of the long-term time series deformation obtained from ALOS-ENVISAT-S1A: (<b>a</b>) spatial distribution of each cluster, (<b>b</b>) percentage of each cluster, (<b>c</b>) time series of cumulative deformation of the cluster center, and (<b>d</b>) violin map of the annual deformation velocity for each cluster.</p> ">
Versions Notes

Abstract

:
Ground subsidence in urban areas is mainly due to natural or anthropogenic activities, and it seriously threatens the healthy and sustainable development of the city and the security of individuals’ lives and assets. Shanghai is a megacity of China, and it has a long history of ground subsidence due to the overexploitation of groundwater and urban expansion. Time Series Synthetic Aperture Radar Interferometry (TS-InSAR) is a highly effective and widely used approach for monitoring urban ground deformation. However, it is difficult to obtain long-term (such as over 10 years) deformation results using single-platform SAR satellite in general. To acquire long-term surface deformation monitoring results, it is necessary to integrate data from multi-platform SAR satellites. Furthermore, the deformations are the result of multiple factors that are superimposed, and relevant studies that quantitatively separate the contributions from different driving factors to subsidence are rare. Moreover, the time series cumulative deformation results of massive measurement points also bring difficulties to the deformation interpretation. In this study, we have proposed a long-term surface deformation monitoring and quantitative interpretation method that integrates multi-platform TS-InSAR, PCA, and K-means clustering. SAR images from three SAR datasets, i.e., 19 L-band ALOS-1 PALSAR, 22 C-band ENVISAT ASAR, and 20 C-band Sentinel-1A, were used to retrieve annual deformation rates and time series deformations in Shanghai from 2007 to 2018. The monitoring results indicate that there is serious uneven settlement in Shanghai, with a spatial pattern of stability in the northwest and settlement in the southeast of the study area. Then, we selected Pudong International Airport as the area of interest and quantitatively analyzed the driving factors of land subsidence in this area by using PCA results, combining groundwater exploitation and groundwater level change, precipitation, temperature, and engineering geological and human activities. Finally, the study area was divided into four sub-regions with similar time series deformation patterns using the K-means clustering. This study helps to understand the spatiotemporal evolution of surface deformation and its driving factors in Shanghai, and provides a scientific basis for the formulation and implementation of precise prevention and control strategies for land subsidence disasters, and it can also provide reference for monitoring in other urban areas.

1. Introduction

As cities expand rapidly and urban population continuously increases, various natural or man-made activities, such as urban expansion and infrastructure construction, groundwater exploitation, static load of high-rise buildings and dynamic load generated by traffic facilities, and compression and consolidation of soft soil layer, could lead to ground subsidence, which will seriously threaten the healthy development of cities and the security of individuals’ lives and assets [1,2,3,4]. Therefore, it is vital to conduct long-term and continuous monitoring of ground subsidence in cities. Compared with traditional ground subsidence monitoring techniques like leveling and Global Navigation Satellite System (GNSS), Time Series Synthetic Aperture Radar Interferometry (TS-InSAR) has unparalleled advantages such as high precision (i.e., centimeter to millimeter level), high spatial and temporal resolution, large coverage area, all-weather, all-sky, and long-distance non-destructive detection, etc. and is widely used in global and regional surface deformation monitoring [5,6,7]. In particular, Persistent Scatterer for Synthetic Aperture Radar Interferometry (PS-InSAR) technique can achieve millimeter-level monitoring accuracy by detecting persistent scatterers that retain high coherence over extended durations, which are widely distributed in cities. This makes it ideal for the fine monitoring of cities and their infrastructure [8,9,10,11].
Shanghai stands as a major national hub and a metropolis of significant scale, serving as China’s economic, financial, trade, and shipping center. Due to excessive exploitation of groundwater, Shanghai has been experiencing ground subsidence since the 1920s, and the cumulative subsidence was up to 2.63 m by 1965 [12]. After 1966, the government departments have taken measures such as restricting groundwater pumping and groundwater recharge, and land subsidence has been alleviated and controlled to a certain extent [13]. However, land subsidence in Shanghai has been exacerbated in recent years by urban expansion and large-scale engineering construction, raising concerns about the safety of the city and its infrastructure. Some scholars have monitored the highways, subways, and Pudong International Airport in Shanghai using TS-InSAR technology and found that uneven settlement has occurred in these infrastructures [14,15,16]. In addition, to alleviate the land shortage caused by population growth, Shanghai has implemented China’s largest land reclamation project, Nanhui New City. Due to the unconsolidated soil, land subsidence is an inherent problem in the land reclamation area. In Shanghai’s coastal area, the coupling of land subsidence and sea level rise is prone to cause seawater intrusion and flood disasters [17]. Therefore, it is necessary to conduct long-term and continuous land subsidence monitoring in Shanghai to comprehensively grasp and understand the distribution characteristics and temporal evolution of surface deformation in this region, and to identify areas with severe land subsidence. In order to obtain long-term continuous deformation results, this paper uses a geotechnical model to unify the deformation results obtained from single-platform SAR satellite with time gap to the same time reference. By selecting the area with serious land subsidence as the region of interest, we achieved the quantitative analysis of the driving factors of land subsidence for the first time in this region. This could provide scientific basis and decision support for formulating effective land subsidence prevention and control measures, thereby reducing the hazards caused by land subsidence and contributing to the sustainable development of the city.
Although TS-InSAR technology has been widely used in urban land subsidence monitoring, the most of developed TS-InSAR algorithms are basically aimed at data acquired by a single SAR satellite, and only a handful of studies have explored the combined monitoring of surface deformation employing multi-platform SAR satellites. However, a single SAR satellite has a relatively short orbital life, making it difficult to perform long-term uninterrupted imaging observations of a specific area. Therefore, to capture surface deformation over a long time span (e.g., more than 10 years), it is essential to integrate the deformation results obtained from multi-platform SAR satellites. For example, Haghighi and Motagh assumed that the subsidence rate remained constant during the observation period and integrated the deformation results obtained from multi-sensor to obtain long-term deformation results in Tehran, Iran, spanning from 2003 to 2017 [18]. Yastika et al. applied the Hyperbolic Method, which is commonly employed as a geotechnical engineering tool for fitting monitored subsidence data, to integrate SBAS DInSAR results across various time periods [19]. Zhang et al. proposed a model-backfeed deformation estimation method to reveal the 20-year surface dynamics of the Groningen gas field by utilizing multi-platform SAR datasets [20]. Zhao et al. used the modified Quantile–Quantile Adjustment (MQQA) algorithm, along with a geotechnical model, to study the long-term evolution of ground subsidence in the coastal area of Shanghai [21]. Wang et al. proposed an approach for continuously updating long-term multi-sensor InSAR deformation time series using robust sequential least squares [22]. At present, the fusion of deformation results from multi-platform SAR satellites is mostly based on certain model constraints or assumptions. In applications, the appropriate fusion method should be selected based on the particular circumstances of the research region.
The purpose of conducting long-term InSAR monitoring in urban areas is to identify the regions of land subsidence, thereby formulating prevention and control measures to avoid casualties and reduce economic losses. However, current analyses of the driving factors of land subsidence are generally qualitative, lacking insight into the quantitative relationship between different driving factors and land subsidence. Furthermore, correctly interpreting the time series deformation results obtained from millions of measurement points remains a challenge. The commonly used approach is to select a few feature points in areas with severe land subsidence and plot their cumulative deformation results over time to observe their temporal evolution characteristics. This method lacks comprehensive analysis and understanding of the temporal evolution characteristics across the entire monitoring area, obstructing the efforts to prevent and manage land subsidence.
In this paper, we collected 19 L-band ALOS-1 PALSAR, 22 C-band ENVISAT ASAR, and 20 C-band Sentinel-1A SAR datasets to retrieve the annual deformation rate and time series of cumulative displacement in Shanghai from 2007 to 2018. In order to obtain continuous deformation results in time, a multi-platform SAR deformation fusion method is proposed. Then, we selected Pudong International Airport as the area of interest, and principal component analysis (PCA) was performed on the time series cumulative deformation results. The driving factors of land subsidence were quantitatively analyzed by combining PCA results with groundwater exploitation and groundwater level change, precipitation, temperature, and engineering geological and human activities. Finally, the K-means clustering was used to obtain the classification results of surface deformation patterns in the whole study area during the monitoring period. This study can provide scientific basis for government departments to formulate effective land subsidence control measures and can provide reference for other cities to analyze the driving factors of land subsidence.
The structure of the remaining paper is as follows. Section 2 introduces the study area and datasets, and Section 3 outlines the research route of this paper and elaborates the principles of TS-InSAR, principal component analysis, and K-means clustering. Section 4 presents the deformation monitoring results of TS-InSAR, the PCA results and the subsidence driving factors analysis for Pudong International Airport, and the classification results of temporal deformation patterns obtained from K-means clustering, followed by conclusions in Section 5.

2. Study Area and Datasets

2.1. Study Area

Shanghai is a megacity in China with a permanent resident population of more than 24 million, covering an area of 6340.5 km2. It is situated in the Yangtze River Delta region, along the eastern coastline of China. Its geographic extent ranges from 120°52′ to 122°12′ longitude and 30°40′ to 31°53′ latitude. To the north, it abuts the Yangtze River. To the east, it borders the East China Sea. To the south, it adjoins Hangzhou Bay, and to the west, it neighbors Jiangsu and Zhejiang provinces. The majority area of Shanghai is flat and is characterized by elevations that span from 2 to 6 m above sea level [23]. Shanghai is predominantly characterized by soft soil layers that possess high water content, low strength properties, and exhibit high compressibility. The engineering geological conditions are poor, making it susceptible to subsidence when subjected to additional loads [15]. The study area includes the Downtown, Pudong New District, Baoshan District, Minhang District, and Fengxian District. The specific geographical location is shown in Figure 1.

2.2. Datasets

In this study area, we collected 19 L-band ALOS-1 PALSAR, 22 C-band ENVISAT ASAR, and 20 C-band Sentinel-1A SAR datasets. Table 1 presents the data parameter information. The spatial and temporal distributions of these SAR datasets are displayed in Figure 1 and Figure 2, respectively. It should be pointed out that we cropped the Sentinel-1A image according to the geographical scope of the study area, and Figure 1 displays the coverage of the cropped Sentinel-1A image. To eliminate the topographic phase from the interferograms, we utilized the 30 m resolution Digital Elevation Model (DEM) derived from the Shuttle Radar Topographic Mission (SRTM). Orbit corrections on ENVISAT ASAR datasets were carried out using the precise orbital data furnished by the Technical University of Delft. In addition, orbital errors in Sentinel-1A datasets were removed using the precise orbital data supplied by the European Space Agency (ESA).
To verify the precision of the TS-InSAR deformation outcomes, the field measurement data of five monitoring stations were collected from Shanghai Public Data Open platform. Five stations are located in Pudong New district. The monitoring period of S1–S4 is from September 2007 to May 2018, and the monitoring period of S5 is from September 2015 to May 2018. The measurement value of each station represents the elevation difference between the monitoring facility and the datum plane with a stable foundation. Figure 1 illustrates the spatial location of the monitoring stations within the study area.
To conduct a quantitative assessment of the factors contributing to land subsidence, the annual groundwater pumping volume of Shanghai from 2007 to 2018 was obtained from the Shanghai Water Resources Bulletin. To examine the relationship between groundwater level changes and land subsidence, groundwater level monitoring data, attributed to the Shanghai Geological Survey Research Institute, were obtained from the Shanghai Geological Data Information Sharing Platform. Additionally, the 0.1° by 0.1°precipitation data and daily mean temperature data were collected to assess the link between seasonal variations and ground subsidence. Furthermore, Landsat TM/ETM optical imagery, engineering geological data, and 30 m resolution impervious surface area change data published by Gong et al. were also collected to analyze the relationship between urban expansion and ground subsidence [24].

3. Methods

The workflow of the SAR dataset processing, driving factor analysis of subsidence, and time series deformation pattern classification are presented in Figure 3. The workflow mainly consists of three parts. At first, Time Series Synthetic Aperture Radar Interferometry was used to retrieve the annual deformation rate and time series of cumulative deformation of the study area. It includes two steps. The first step is to obtain the LOS deformation by using the single satellite platform, and the second step is to integrate the deformation results obtained by multiple satellite platforms with the self-weight consolidation settlement model [17,23] to obtain the long-term continuous vertical surface deformation results. Then, PCA was applied to the time series cumulative deformation of Pudong International Airport. The PCA results, groundwater exploitation and groundwater level change, precipitation, temperature, and engineering geological and human activities were combined to quantitatively analyze the main driving factors of settlement. Finally, the spatiotemporal unsupervised K-means clustering was performed to partition large-scale deformation fields into sub-regions with the same deformation pattern. The SAR datasets were processed by open-source software ISCE (InSAR Scientific Computing Environment) version 2 and StaMPS (Stanford Method for Persistent Scatterers) version 4. PCA and K-means clustering were performed by MATLAB R2021a. The detailed principles of MT-InSAR, PCA, and K-means clustering are as follows.

3.1. Long Time Series InSAR for Joint Multi-Platform Satellites

3.1.1. TS-InSAR for Single Satellite Platforms

In this study, we adopt the StaMPS-PSInSAR technology put forward by Hooper to process the SAR datasets obtained from a single satellite platform. The selection of common reference images is dependent on the contributions of the two factors, temporal baseline and spatial baseline, to the coherence of the interferogram. A comprehensive correlation measure is adopted as the evaluation index for the coherence of the interferogram. The model of the comprehensive correlation function is as follows [25,26]:
γ m = 1 N − 1 ∑ k = 1 N − 1 [ c ( T k , m , T c ) ] α * [ c ( B ⊥ k , m , B c ) ] β  
c ( x , a ) = { 1 − | x | a ,       x < a           0 ,                     x ≥ a
where γ m represents the comprehensive correlation coefficient, and T k , m and B ⊥ k , m are the temporal baseline and perpendicular spatial baseline of the interferogram formed by images k and m, respectively. T c and B c are their respective critical conditions. To improve the accuracy of the coherence measure’s numerical expression, the maximum value of the temporal baselines among all the interferograms is taken as the critical value for the temporal baseline, and the maximum value of the spatial baselines is taken as the critical value for the spatial baseline. α and β are the corresponding exponential factors representing the contribution degree of each individual factor. Considering formula (1) as an optimization function, when the values of the independent variables T k , m and B ⊥ k , m maximize the objective function, this set of parameters is considered as the optimal solution of the model, and the corresponding image m is the optimal common reference image with the best comprehensive baseline distribution.
The remaining images were registered with the optimal common reference image and interferometric processing was carried out. After removing terrain phase by using SRTM DEM with 30 m resolution, the differential interferograms were obtained. Then, the methodology introduced by Hooper was employed to recognize PS [9]. The first step uses the amplitude deviation index to select PS candidate points, and the second step uses the phase stability to select PS. The phase of any pixel x in the jth differential interferogram is mainly composed of the following parts:
ϕ   i n t , x , j = W { ϕ d e f , x , j + ϕ a t m , x , j + ϕ o r b , x , j + Δ ϕ θ , x , j + ϕ n , x , j }
where ϕ d e f , x , j is the deformation phase that originates from the displacement of the pixel within the line-of-sight (LOS) direction, ϕ a t m , x , j is the phase that represents the variation in atmospheric retardation when compared between the reference and secondary images, Δ ϕ θ , x , j is the residual topographic phase that arises due to inaccuracies in the DEM, ϕ o r b , x , j is the residual phase that originates from errors in the orbit, ϕ n , x , j is the noise component that arises due to thermal noise and coregistration discrepancies, and W{·} denotes the wrapping phase.
The amplitude dispersion index can be formulated as follows [8]:
D A = σ A m A
where D A is the amplitude dispersion index. σ A and m A represent the standard deviation and mean of the amplitude, respectively. Subsequently, PS is chosen based on phase stability and assessed through the calculation of the pixel’s temporal coherence:
  γ x = 1 N | ∑ i = 1 N exp { j ( ϕ i n t ,   x , i − ϕ ¯ i n t ,   x , i − Δ ϕ ^ θ , x , i ) } |
where γ x denotes the temporal coherence associated with pixel x, N represents the count of interferograms, and j equals − 1 . The phase average of all pixels located in a circle of radius L, centered at pixel x, is ϕ ¯ i n t ,   x , i . Δ ϕ ^ θ , x , i serves as the approximation for Δ ϕ θ , x , i .
After selecting the PS points, the DEM error is removed from their phases. The phase error from uncertainty in the DEM is proportional to the perpendicular component of the baseline. Therefore, the DEM error can be estimated by the least square method. Then, we applied a three-dimensional phase unwrapping method (comprising one-dimensional time and two-dimensional space domains) to the PS. Variation in atmospheric delay between passes is mainly due to the variation in the water vapor content of the troposphere, which is spatially correlated. Orbital errors are also correlated spatially in azimuth, and interferometric processing leads to the spatial correlation of the residual orbit error term, which is also in range. Thus, through high-pass filtering in time and low-pass filtering in space, the atmospheric delay phase and orbital error phase can be estimated and removed, and the deformation phase can be separated to obtain the deformation result [9,27].

3.1.2. Fusion the Deformation Results Obtained from Multi-Platform Satellites

To achieve long-term, continuous surface deformation measurements, the data fusion in this study includes two steps. The first step is fusion of ALOS-1 and ENVISAT deformation results with time overlap, and the second step is fusion of ALOS-ENVISAT and Sentinel-1A deformation results with time interval. Since ALOS-1, ENVISAT, and Sentinel-1A satellites have different radar coordinate systems, it is necessary to convert them into the same geographical coordinate system, namely, the World Geodetic System 1984 (WGS-84) coordinate system, prior to fusion. The measurement results of TS-InSAR are relative to a certain temporal benchmark point and a certain spatial benchmark point; therefore, the temporal and spatial benchmarks must be unified before fusion. Taking the moment of the first SAR image acquired by ALOS-1 as the temporal benchmark, which meant that the surface deformation at that time was assumed to be zero. To ensure the consistency of spatial benchmarks, ALOS-1, ENVISAT, and Sentinel-1A are set to the same spatial reference point before phase unwrapping. Furthermore, the deformation outcomes derived through TS-InSAR signify the actual deformation’s projection along the LOS direction, yet the incidence angles of different SAR satellites are often varied. According to the published literature [12,28], the surface deformation in the Shanghai region is primarily vertical. Hence, the deformation results obtained by ALOS-1, ENVISAT, and Sentinel-1A can be convert to the vertical direction using the following equation.
V v e r t i c a l = V L O S / c o s θ
where V L O S indicates the deformation rate along the line-of-sight direction, V v e r t i c a l signifies the vertical deformation rate, and θ denotes the incidence angle of the radar satellite.
Next, the integration of ALOS-1 and ENVISAT deformation results is introduced with the same time range, which primarily includes the selection of common persistent scatterer pairs (CPSPs), the fusion of deformation rates and time series cumulative deformations for these CPSPs, and the screening and supplementation of non-common PS pairs. In order to obtain the CPSPs of ALOS-1 and ENVISAT, we search for the nearest PS points from ENVISAT within a certain distance range centered on the PS points acquired by ALOS-1. Considering the spatial resolutions of ALOS-1 and ENVISAT, the spatial distance threshold is set to 30 m. In urban areas, if PS pairs are distributed on adjacent buildings and grounds, they do not belong to the scatterer of the same ground target even if they are very close in spatial distance. To avoid this situation, PS pairs possessing an elevation difference greater than 0.5 m are discarded. In addition, to avoid some measurement points that have large errors, the mean value of the deformation rate difference in all CPSPs plus or minus half of the mean square error is used as the upper and lower thresholds for judgment, and the CPSPs that do not fall within this threshold range are discarded.
Coherence serves as a criterion for assessing the quality of PS, so coherence is taken as the weight when fusing the deformation rates of CPSPs. The deformation rates of CPSPs after fusion can be obtained by the following equations:
W A L O S = C o h A L O S
W A S A R = C o h A S A R  
V C P S P = W A L O S W A L O S + W A S A R * V A L O S + W A S A R W A L O S + W A S A R * V A S A R  
where C o h A L O S represents the coherence of PS obtained from ALOS-1 in CPSPs, and C o h A S A R represents the coherence of PS obtained from ENVISAT. W A L O S and W A S A R represent the weights of PS obtained from ALOS-1 and ENVISAT, respectively. V A L O S represents the deformation velocity of PS derived from ALOS-1, V A S A R represents the deformation rate of PS obtained from ENVISAT, and V C P S P represents the deformation rate of CPSP after fusion.
Then, the time series cumulative deformation of CPSP will be fused. Assuming that the imaging times of ALOS-1 SAR images are t 1 A L O S , t 2 A L O S , … , t m A L O S , the acquisition times of the ENVISAT SAR images are t 1 E N V , t 2 E N V , … , t n E N V , where t 1 A L O S < t 1 E N V . The time series cumulative deformation derived from ALOS-1 and ENVISAT by using TS-InSAR technology are d 1 A L O S , d 2 A L O S , … , d m A L O S and d 1 E N V , d 2 E N V , … , d n E N V . Since the cumulative deformation obtained by ALOS-1 and ENVISAT takes the first SAR image date as the time starting point, it is necessary to correct the cumulative deformation obtained by ENVISAT to the same time starting point as the ALOS-1, i.e., to add systematic offset. Firstly, the cumulative deformation at the acquisition time of ENVISAT SAR images can obtained by linearly interpolating the time series cumulative deformation of ALOS-1:
d i n t e r p E N V ( t i E N V ) =   d A L O S (   t N A L O S ) − d A L O S ( t P A L O S ) t i E N V − t P A L O S + d A L O S ( t P A L O S ) ( i = 1 , 2 , … n )
where, t i E N V represents the acquisition time of the ENVISAT SAR images, t P A L O S represents the previous moment closest to t i E N V in ALOS-1, t N A L O S represents the next moment closest to t i E N V in ALOS-1, d A L O S ( t P A L O S ) and d A L O S (   t N A L O S ) represent the cumulative deformation at the moment t P A L O S and t N A L O S , respectively. The systematic offset of ENVISAT deformation is obtained by minimizing the sum of squares of the differences between the corrected ENVISAT deformation and the linearly interpolated ENVISAT deformation, i.e.,
d o f f s e t = arg m i n ∑ i = 1 n [ ( d t i E N V + d o f f s e t ) − d t i i n t e r p E N V ] 2
The deformation result of ENVISAT after correction is d 1 E N V + d o f f s e t , d 2 E N V + d o f f s e t , … , d n E N V + d o f f s e t . Finally, according to the chronological order of SAR image acquisition time, the time series cumulative deformation of ALOS-1 datasets and the corrected ENVISAT datasets are sorted to obtain the fused cumulative deformations for CPSPs, i.e., d 1 A L O S − E N V , d 2 A L O S − E N V , … d m + n A L O S − E N V . There are also some non-common PS pairs in ALOS-1 and ENVISAT. The PS points exhibiting a coherence value above 0.65 are chosen and combined with the common PS pairs to obtain the final fusion result.
To acquire the consecutive surface deformation in Shanghai from 2007 to 2018, it is necessary to link the TS-InSAR measurement results obtained from ALOS-ENVISAT and Sentinel-1A, which have an interval of nearly five years. In this study, we adopted the self-weight consolidation settlement model to achieve this. This time-dependent geotechnical model was developed by Yang et al. to simulate the self-weight consolidation and settlement process of the alluvial soil in Shanghai by centrifugal model experiment [29]. The expression of the model is defined by the following [17,23]:
S ( t ) = S m ( t − δ ) λ k λ + ( t − δ ) λ
where t represents the acquisition time of SAR image; S(t) signifies the cumulative deformation at time t; S m denotes the total cumulative deformation when t approaches infinity (i.e., the overall deformation); k and λ are the curvature parameters of the model, primarily determining the degree of curvature; and δ represents the time delay coefficient, which considers the differing start times for the consolidation phase.
Taking each PS acquired after the fusion of ALOS-ENVISAT as the center, the spatial distance threshold was set to 30 m, and the common PS pairs of ALOS-ENVISAT and Sentinel-1A were obtained by searching for the nearest PS points from Sentinel-1A within this distance. Then, the time series cumulative deformations obtained from ALOS-ENVISAT at common PS points were used as the observations, and the parameters of the optimal fitting model are calculated by applying the Levenberg–Marquardt least squares (LS) minimization method to Formula (12). The evaluation criterion of the optimal fitting model is to minimize the Root Mean Square Error (RMSE) between the cumulative displacement of ALOS-ENVISAT derived from TS-InSAR and the fitting cumulative deformation of the model. Thus, a lower RMSE indicates a better fit of the model.
Then, the self-weight consolidation settlement optimal fitting model is used to predict the time series cumulative deformation of Sentinel-1A SAR image acquisition time. By reducing the squared discrepancies between the Sentinel-1A model predictions and the TS-InSAR measured values to a minimum, the systematic deformation offset of the Sentinel-1A due to the different time reference points with ALOS-ENVISAT is obtained. After correcting the TS-InSAR deformation measurements of Sentinel-1A using the systematic deformation offset, the cumulative deformation that is temporally continuous with ALOS-ENVISAT is obtained. The long-term continuous deformation results of Shanghai obtained from 2007 to 2018 using the method described are presented in Section 4.1.2.

3.2. Principal Component Analysis

Principal component analysis (PCA) is a prevalent multivariate statistical method that facilitates dimensionality reduction of extensive datasets, enhancing interpretability while minimizing the loss of information [30,31]. It converts a collection of correlated variables into a new set of uncorrelated ones, allowing for the depiction of temporally and spatially evolving deformation patterns without prior constraints [32,33]. The cumulative deformation in the time series arises from the overlay of multiple signals stemming from various driving factors. The PCA method, which is utilized for isolating and analyzing these deformation signals, essentially performs a blind signal separation [1,34].
There are two ways for applying PCA to the InSAR time series dataset: the T-mode and S-mode [1,35,36,37]. In this study, we chose to use T-PCA because it focuses on the time series displacements. The dimension of InSAR observation matrix D for PCA is m rows and n columns, where m is the count of PS points and n is the count of the observation epoch. The fundamental procedures involved in the PCA process are outlined below [37,38]:
  • D is centralized and standardized to obtain D c ;
  • The covariance matrix C of matrix D c is calculated by the following equation:
      C n × n = 1 m − 1 D c T D c
  • By applying eigen-decomposition to C, it can obtain the eigenvalue matrix denoted as ∧ and the corresponding orthogonal eigenvector matrix P, which satisfied P T C P = ∧ . The eigenvectors, also known as coefficients or loadings, represent the contribution of each original variable to each principal component (PC). The eigenvalue quantifies the proportion of variance in the original data that is captured by each PC. The eigenvalues are usually arranged in a descending order, and this implies that the first principal component (PC), which has the highest variance contribution, serves as the primary explainer of the dataset’s variability. The variance contribution (VC) of each PC can be calculated by the following equation:
    V C i = λ i ∑ j = 1 n λ j ,   i = 1 , 2 , … , n
    where λ i is the eigenvalue of the ith PC, and n is the total number of PC.
  • Automated determination of the best number of PCs is carried out [32]. This process involves utilizing a Scree plot, which graphs the eigenvalues against the component count, identifying the optimal number of PCs by locating the “elbow” point on the curve.

3.3. K-Means Clustering

The K-means algorithm according to MacQueen is a classical unsupervised learning clustering method [39]. It has been widely used in time series clustering, and it can partition the massive time series of InSAR deformations into K homogenous clusters with similar temporal patterns in terms of the Euclidean distance [31,40,41]. Considering a data matrix Ω of dimensions N by M, N rows represent individual measurement points and the M columns the satellite acquisition time. We aim to assign the partitioning of Ω into K clusters with similar deformation characteristics. The essential procedures for K-means clustering are outlined below [42,43]:
  • The centroids u k (k = 1, 2, …, K) of the K desired clusters were initialized randomly. The quantity of clusters, K, is autonomously set based on the ideal count of PCs. Performing PCA prior to K-means clustering can mitigate the issue of Euclidean space inflation and enhance computational efficiency [31].
  • Each data sample x i was allocated to the nearest cluster centroid u k , i.e., with the the smallest Euclidean distance which is defined by the following equation:
    min { ∑ k = 1 K ∑ x i ∈ C k | | x i − u k | | 2 }
  • The cluster centroids were adjusted to the average values of their respective associated data samples:
    u k ( t + 1 ) = 1 N k ∑ x i ∈ C k x i
    where u k ( t + 1 ) is the new cluster center of moment t + 1, and N k is the number of measure points in the subcluster C k .
  • Iterative steps 2 and 3 until the centroids and the assignment of measurement points remain unchanged.

4. Results and Discussion

4.1. Deformation Monitoring Results of TS-InSAR

4.1.1. LOS Deformation Result from Single Satellite Platform

When selecting the common reference image for StaMPS-PSInSAR, the optimal spatial–temporal baseline configuration is found by utilizing a comprehensive correlation function model. For ALOS-1 PALSAR, ENVISAT ASAR, and Sentinel-1A datasets, SAR images with acquisition dates of 14 April 2009, 4 August 2008, and 11 March 2017 were selected as the reference images, and the remaining images were used as secondary images. A total of 18, 21, and 19 interferograms were generated. Figure 4 displays the resultant spatial–temporal baseline diagram.
During the selection of PS, considering the lower spatial resolution of ALOS-1 PALSAR, ENVISAT ASAR, and Sentinel-1A, the threshold for the amplitude dispersion index is set to 0.35. The average annual line-of-sight (LOS) deformation rates acquired from ALOS-1 PALSAR, ENVISAT ASAR, and Sentinel-1A are presented in Figure 5. A positive value depicts that the ground target is moving towards the radar sensor, while a negative value depicts that the ground target is moving away from the radar sensor. According to the deformation results of ALOS-1 and ENVISAT, the average annual LOS deformation rate of Shanghai from 2007 to 2010 is distributed in [−30, 20] mm/annual. It should be noted that mm/annual is the unit of the annual deformation rate. In order to simplify the expression and conform to common expression habits, mm/annual is expressed as mm/a in the following articles. The primary regions experiencing subsidence are concentrated in Downtown, Minhang District, and Pudong New District. Notably, the eastern coastal region of Pudong New District has experienced significant subsidence, recording a rate surpassing −20 mm/a. This area belongs to the land reclamation zone, and the compression and consolidation of the filled soil are prone to causing ground subsidence.
Based on the deformation results from Sentinel-1A, the LOS deformation rate of Shanghai between 2015 and 2018 is mainly distributed within [−10, 10] mm/a. Subsidence has been observed in Minhang District, Fengxian District, and the central portion of Pudong New District, with a subsidence rate of approximately −10 mm/a. In addition, Sentinel-1A data detected severe uneven subsidence along the eastern coast of Pudong New District, with a subsidence rate falling within a range of −30 mm/a to −20 mm/a.
The density of PS points detected in the study area is 154/km2 for ALOS-1 PALSAR, 255/km2 for ENVISAT ASAR, and 318/km2 for Sentinel-1A. Studies have shown that the decorrelation time depends on the wavelength of the radar, and it increases as the wavelength becomes longer [25,44,45]. As a result, ALOS-1 at L-band (23.6 cm) can maintain coherence better than ENVISAT and Sentinel-1 at C-band (5.6 cm) in vegetated areas. The PS density in TS-InSAR measurement is mainly influenced by two categories of factors. One is the land cover type. The other influencing factor is SAR satellite parameters, for instance, wavelength, resolution, etc. [46,47]. SAR image resolution from high to low order is as follows: Sentinel-1A > ENVISAT > ALOS-1. Therefore, Sentinel-1A achieves the highest PS density, ENVISAT follows in second, and ALOS-1 has the least. According to the land cover map of Shanghai (Figure 1), the land cover type along the coastal areas of Pudong New District is vegetation, which is easily affected by temporal decorrelation. Both ENVISAT and Sentinel-1 barely detect deformation in this region, whereas ALOS-1 is capable of obtaining deformation monitoring results in this area.

4.1.2. Long-Term Vertical Deformation Result from Multi-Platform Satellite

In order to obtain long-term and continuous results of surface deformation in Shanghai from 2007 to 2018, we fused the deformation results of ALOS-1, ENVISAT, and Sentinel-1A using the method introduced in Section 3.1.2. Specifically, we first performed spatial–temporal fusion of the deformation results obtained from ALOS-1 and ENVISAT with overlapping time ranges and then used the self-weight consolidation settlement model to link the deformation results of ALOS-ENVISAT and Sentinel-1A with time intervals, generating long-term and continuous deformation results. In the SAR image acquired by ALOS-1, we selected the first image as the time benchmark, assuming that the deformation at this moment (i.e., 7 January 2007) was zero, and the deformation measured by subsequent SAR images was relative to this time reference point. Based on the spatial coverage of ALOS-1 datasets, there were 413,250 and 514,975 PS points involved in the fusion of ALOS-1 and ENVISAT, respectively. After screening common PS point pairs and filling in non-common PS points, a total of 744,051 fused PS points were obtained. By merging the deformation results of ALOS-1 and ENVISAT, the temporal resolution of deformation observation is improved, and the spatial density of measurement points is also increased. Through the selection of the common PS points of ALOS-ENVISAT and Sentinel-1A, a total of 455,332 PS points were finally obtained. The average vertical deformation rate and time series cumulative deformation of Shanghai from 2007 to 2018 obtained after the fusion of ALOS-1, ENVISAT and Sentinel-1A are displayed in Figure 6.
Figure 6 illustrates that the vertical deformation rate of Shanghai from 2007 to 2018 is distributed in [−30, 20] mm/a. The deformation rate of most regions is between −10 mm/a and 10 mm/a, indicating that most of the surface in the study area is basically in a stable state during the monitoring period. Severe land subsidence was detected along the coast and in the central northern parts of Pudong New District, with subsidence rates exceeding −20 mm/a and cumulative surface subsidence exceeding −150 mm. Due to the five-year gap between ALOS-1, ENVISAT, and Sentinel-1A, we obtained continuous surface deformation results in the study area from 2007 to 2018 by using the self-weight consolidation settlement model. In order to check the difference between the deformation results predicted by the self-weight consolidation settlement model and the time series InSAR measurements, we selected four measurement points in the study area (Figure 6b) and calculated the RMSE of the difference between the model predictions and the TS-InSAR measurements of Sentinel-1A. As depicted in Figure 7, the RMSE between the model’s predicted value and the TS-InSAR’s measured value is relatively low, suggesting a strong alignment between the model’s predictions and the TS-InSAR’s observations.

4.1.3. Verification of Deformation Results from TS-InSAR

In order to verify the accuracy of the deformation results obtained by a single satellite platform and the fusion results obtained from ALOS-ENVISAT-S1A, PS points within 100 m around the monitoring stations were selected, and the mean value of their deformation rates was calculated. After the deformation rate in LOS direction V L O S is converted into the vertical direction V v e r t i c a l by formula (6), the RMSE of the difference between the deformation rate of ALOS-1 PALSAR, ENVISAT ASAR, Sentinel-1A, and ALOS-ENVISAT-S1A obtained by TS-InSAR processing, and the deformation rate of the field measurement point is calculated, which is 0.41 mm/a, 0.16 mm/a, 0.92 mm/a, and 0.16 mm/a, respectively. The scatterplot of deformation rates calculated from TS-InSAR and field measurement is displayed in Figure 8, and the Pearson’s correlation coefficient r is 0.935. From the above comparison results, we can conclude that the TS-InSAR monitoring results obtained from a single satellite platform and the fusion results of multi-platform satellites are in good agreement with the field measurement results. It should be pointed out that the vertical deformation rates obtained from monitoring stations during the period from 2007 to 2010 were employed to verify the accuracy of the measurement results from ALOS-1 and ENVISAT. Compared to the Sentinel-1A measurement results, the vertical deformation rates from 2015 to 2018 were used. When comparing with the ALOS-ENV-S1A fusion results, the vertical deformation rates from 2007 to 2018 were used. Since the monitoring station S5 obtained the measurement results from 2015 to 2018, the field measurement data of S5 were only used to verify the accuracy of Sentinel-1A deformation results.

4.2. Quantitative Analysis of Driving Factors for Subsidence at Pudong International Airport Based on PCA

As seen from the deformation results in Section 4.1, there is obvious uneven land settlement in Shanghai, exhibiting a spatial pattern of stability in the northwest and subsidence in the southeast. In particular, severe settlement exhibiting a subsidence rate ranging from −30 mm/a to −20 mm/a has been monitored in the land reclamation area of Pudong New District. Severe land subsidence poses a threat to the safe functioning of important urban infrastructure such as airports. In view of this, Pudong International Airport is selected as the region of interest in this section, and principal component analysis is performed on the time series cumulative deformation obtained in this region. Subsequently, leveraging the PCA results, combined with groundwater exploitation and groundwater level change, precipitation, temperature, engineering geological and human activities, quantitative analysis is performed on the driving factors of regional ground subsidence.
Although the mean rate of deformation and maximum accumulated deformation in Shanghai from 2007 to 2018 can be obtained by combining the deformation results of multi-platform satellites and the self-weight consolidation subsidence model, the lack of SAR data between 2010 and 2015 makes it impossible to obtain the time series accumulated deformation during 2010–2015. Even if the time series accumulated deformation during 2010–2015 can be obtained through linear interpolation, the interpolated results cannot reflect the true temporal evolution behavior of surface deformation. Therefore, the vertical time series of cumulative deformation derived from the fusion of ALOS and ENVISAT during 2007–2010 and the vertical time series of cumulative deformation obtained from Sentinel-1A during 2015–2018 were used for PCA.
Pudong International Airport is situated in Pudong New District, approximately 30 km from downtown. It is a 4F-class civil airport, representing one of China’s key transportation hubs in the eastern region, specifically the largest and most significant gateway airport hub. The construction of Pudong International Airport has been a complex and ambitious project involving years of planning, design, construction, and expansion. The first phase of Pudong Airport began in 1997, with the completion and opening of the first terminal and first runway in 1999. The second runway was completed and officially opened in 2005. The second phase expansion project was fully launched in 2005, and the second terminal and third runway were officially opened for traffic in 2008. The fourth and fifth runways were completed and put into use in 2015 and 2017, respectively. The construction changes in Pudong International Airport during 2007–2018 can be observed from the Landsat TM/ETM optical images in Figure 9. By the end of 2016, Pudong International Airport had 220 routes (106 domestic and 114 international). As of 2023, the passenger throughput reached 54,476,397.
Pudong International Airport is constructed on a foundation of soft soil layers, with the foundation soil within a depth of 65 m from the surface primarily consisting of fill soil, clayey soil, silty soil, and sandy soil. The names and burial depths of the foundation soil layers are shown in Table 2 [48]. Figure 9d illustrates that the airport is divided into eastern and western parts by a dashed black line located between Runway 1 and Runway 2. The western section of the airport consists of naturally occurring sediments with silt-bearing mud and muddy clay. The eastern part of the airport is a newly formed land area created through sea reclamation and landfilling, filled with loose sandy hydraulic fill. The hydraulic fill can be classified into recently filled soil and old filled soil based on the length of time since filling. The consolidation time of old filled soil is generally more than 10 years, while the consolidation time of recently filled soil is less than 5 years. Due to the under-consolidated state of the hydraulic fill, it inevitably induces land subsidence, necessitating long-term and continuous deformation monitoring of the area.
Based on the monitoring results from Section 4.1, the vertical annual deformation rate of Pudong International Airport during 2007–2010 and 2015–2018 can be obtained as shown in Figure 10. The black rectangular box represents Pudong International Airport, which is the analysis area for PCA. It can be observed that Pudong International Airport experienced significant uneven land subsidence during both 2007–2010 and 2015–2018, with the western part of the surface remaining relatively stable while the newly constructed eastern area formed through land reclamation experienced severe subsidence. To recognize the driving factors of ground subsidence in this area, PCA was conducted on PS points exhibiting negative deformation rates. Among the deformation results obtained from the fusion of ALOS-1 and ENVISAT, 2959 PS points were selected in total, characterized by an average deformation rate of −7.49 mm/a. In the deformation results obtained from Sentinel-1A, 3738 PS points were chosen in total, with an average deformation rate of −5.76 mm/a.
Subsequently, PCA was employed separately on the PS subsidence points obtained from the fusion of ALOS and ENVISAT during 2007–2010 and Sentinel-1A during 2015–2018. Figure 11 and Figure 12 display the explained variance of the first four PCs and their eigenvectors, respectively. It is evident that the variance contributions of PC 1–4 derived from the ALOS-ENVISAT time series cumulative deformation are 70.66%, 7.98%, 6.92%, and 3.49%, respectively, while those derived from the Sentinel-1A time series cumulative deformation are 71.59%, 13.11%, 3.87%, and 2.21%, respectively. The first principal component (PC1) accounts for over 70% of the variance in both cases, indicating that PC1 can explain more than 70% of the variability in the dataset. Since the variance contributions of PC 4 from ALOS-ENVISAT and PC 3–4 from Sentinel-1A are not significant and lack statistical significance, their corresponding eigenvectors are not analyzed in the subsequent studies.
The eigenvectors corresponding to the principal components (PCs) can be used to describe the variation in deformation over time. The eigenvectors of PC1 obtained from both ALOS-ENVISAT and Sentinel-1A are relatively stable over time, suggesting stability in the driving factors linked to PC1 over an extended period. Apart from PC1, almost all other principal components exhibit significant fluctuations, suggesting that seasonal variations have a greater impact on these PCs compared to PC1. The driving factors of urban land subsidence can be classified into two categories: human activities and natural factors. Human activities encompass construction projects, groundwater extraction, etc., while natural factors encompass precipitation, temperature, etc. Therefore, to assess the primary causes of land subsidence in this region quantitatively, daily mean temperature data, monthly precipitation data, groundwater level variation data (see Figure 10 for the locations of groundwater level monitoring wells), annual groundwater pumping data, and impervious surface area change data were collected. Overlay correlation analysis was performed between the eigenvectors of PC 1–3 obtained from ALOS-ENVISAT, and PC 1–2 obtained from Sentinel-1A with these datasets. The results are depicted in Figure 13 and Figure 14, respectively.
Figure 13a illustrates that the eigenvector of PC1 exhibits a linearly increasing trend over time, with a deceleration in growth after 2008 and subsequently maintaining a stable state, suggesting that the driving factor corresponding to PC1 is consistently stable. In comparison, the area of impervious surfaces exhibits a similar temporal pattern as the eigenvector of PC1. The change in the area of impervious surfaces is closely related to human construction activities. This figure indicates a linear increase in impervious surface area in this region from 2007 to 2010, indicating continuous construction activities in the area. The optical image in Figure 9 reveals that the land reclamation projects were underway in this region during 2007–2010. The engineering geological conditions of the strata formed in land reclamation areas are poor, making them prone to ground subsidence. It can be observed from Figure 15 that the two profile lines AB and CD crossing the east–west dividing line of Pudong International Airport are basically stable in the west, and the deformation rate is distributed between −5 mm/a and 5 mm/a. In the east, the settlement is serious, and the subsidence rate increases steadily with the extension of the section line to the east, and the highest subsidence rate can reach −30 mm/a.
To further investigate the relationship between the construction time of airport runways, terminals, and the occurrence of land subsidence, the time series cumulative deformation at seven measurement points P1–P7 obtained from ALOS-ENVISAT was plotted (Figure 16). Points P1–P5 are sequentially located along Runways 1 to 5, while P6 and P7 are positioned at Terminals 1 and 2, respectively. Figure 16 reveals that the time series cumulative deformation at P1 and P6 ranges from −10 mm to 10 mm, indicating a relatively stable state of the surface at Runway 1 and Terminal 1 after more than a decade of consolidation and settlement since their completion in 1999. Terminal 2 and Runway 3 started construction in December 2005 and were completed in 2008, with the construction period overlapping with the monitoring time of TS-InSAR. The plot shows that the cumulative deformation at P7 reaches −40 mm, while that at P3 is around −25 mm. This is attributed to Terminal 2 being built on the land reclaimed from the sea, whereas Runway 3 is constructed on older, more stable land. Runway 2, located in a land-reclaimed area and officially opened in 2005, still exhibited settlement between 2007 and 2010, with a cumulative deformation at P2 of approximately −25 mm, implying incomplete self-weight consolidation of the surface. Between 2007 and 2010, Runways 4 and 5 were not yet constructed, but they are located in land-reclaimed areas near the coastline. The cumulative deformations at P4 and P5, which are in these areas, exceeded −60 mm. Based on the above analysis, it can be concluded that land subsidence corresponding to PC1 during the period from 2007 to 2010 was caused by engineering construction activities related to land reclamation, contributing 70.66% to the total subsidence.
Figure 13b indicates that the eigenvector corresponding to PC2 exhibits a downward trend over time, consistent with the trend in groundwater extraction volume. Although there is an upward trend in the eigenvector of PC2 from February to November 2008, similar to the trends in precipitation and groundwater level, an overall downward trend is observed during this monitoring period. Hence, the ground subsidence linked to PC2 between 2007 and 2010 was attributed to groundwater extraction, contributing 7.98%.
As depicted in Figure 13c, the eigenvector corresponding to PC3 exhibits a fluctuating upward trend over time, similar to the variations in precipitation and groundwater level. Additionally, the trends in precipitation and groundwater level are also similar, with the latter exhibiting a certain degree of lag. This indicates that precipitation can result in alterations in groundwater levels. When precipitation increases, groundwater levels tend to rise. When precipitation decreases, groundwater levels tend to fall. As groundwater levels decline, the moisture content in soil and aquifers decreases, increasing the effective stress between soil particles, and leading to progressive soil layer compaction. This compaction reduces the soil volume, thereby triggering land subsidence. Therefore, it can be inferred that the ground subsidence associated with PC3 during 2007–2010 was jointly caused by variations in precipitation and groundwater levels, with a contribution rate of 6.92%.
Next, we will analyze the eigenvectors corresponding to PC1-2 obtained from Sentinel-1A. As seen in Figure 14a, the eigenvector of PC1 displays a linearly increasing trend over time, with a deceleration in growth after 2016 and subsequently maintaining a stable state, suggesting that the driving factor corresponding to PC1 is consistently stable. By comparison, it can be found that the area of impervious surfaces exhibits a similar temporal evolution with the eigenvector of PC1. Furthermore, Figure 15 illustrates that the two profile lines AB and CD crossing the east–west division line of Pudong International Airport are basically stable in the west, and the deformation rate spans a range of −5 mm/a to 5 mm/a. In the east, the subsidence is serious, and the peak subsidence rate reaches up to −30 mm/a.
To further explore the relationship between the construction time of airport runways, terminals, and the occurrence of land subsidence, the time series cumulative deformation at seven measurement points Q1–Q7 obtained from Sentinel-1A was plotted (Figure 17). Points Q1–Q5 are sequentially located along Runways 1 to 5, while Q6 and Q7 are positioned at Terminals 1 and 2, respectively. Figure 17 shows that the time series cumulative deformation at Q1 and Q6 varies between −10 mm and 10 mm, consistent with the monitoring results from 2007 to 2010, indicating a relatively stable state of the surface at Runway 1 and Terminal 1 after more than a decade of consolidation and settlement since their completion in 1999. The construction of Terminal 2 and Runway 3 was started in December 2005 and were completed in 2008. The plot reveals that the cumulative deformation at Q7 is around −30 mm, indicating a mitigation of subsidence compared to the period between 2007 and 2010. The cumulative deformation at Q3 is also around −30 mm, suggesting mild subsidence at both Terminal 2 and Runway 3. Runway 2 is located on a land-reclaimed area and was officially opened in 2005, exhibiting significant subsidence between 2015 and 2018, with a cumulative deformation at Q2 of approximately −50 mm. This may be related to Runway 2 being located on a land-reclaimed area and the influence of construction activities next to it, specifically the construction of Runway 4. Runways 4 and 5 were completed in 2015 and 2017, respectively, overlapping with the monitoring period of TS-InSAR. The cumulative deformations at Q4 and Q5 both exceeded −80 mm, indicating severe land subsidence in these areas during the construction period. Following the preceding analysis, the conclusion can be drawn that land subsidence corresponding to PC1 during the period from 2015 to 2018 was caused by engineering construction activities related to land reclamation, contributing 71.59% to the total subsidence.
Figure 14b indicated that the eigenvector of PC2 exhibits a downward trend over time, consistent with the trend in groundwater extraction volume. Therefore, it can be deduced that the land subsidence associated with PC2 during 2015–2018 was caused by groundwater extraction, with a contribution rate of 13.11%.

4.3. Classification Results of Time Series Deformation Patterns Based on K-Means Clustering

The K-means clustering results of the long-term time series deformation obtained from ALOS-ENVISAT-S1A in Shanghai ranging from 2007 to 2018 are presented in Figure 18. The number of clusters can be decided by the optimal number of PCs. For this research, we established the cluster count as 4. We can see from Figure 18 that the K-means clustering technique has categorized the time series deformation into four types: uplift, stable, slight subsidence, and evident subsidence. Figure 18a, Figure 18b, and Figure 18c, respectively, show the spatial distribution of each cluster, the percentage of each cluster, and the time series of cumulative deformation in each cluster center. Figure 18d is the violin map of the average annual deformation velocity for each cluster.
In Figure 18, around 53% of the PS points belong to Cluster 3 (Stable). It is distributed throughout the whole study area. Cluster 1 (uplift) accounts for 10% and is aggregated on the southern study area, that is, southern Pudong New District and eastern Fengxian District. It should be noted that, after 2010, Cluster 1 showed a tendency of slow subsidence, but its time series cumulative deformation was still greater than zero. Cluster 4 (slight subsidence) is mainly distributed in the Minhang District, northern Fengxian District, and central Pudong New District. It accounts for 34%. Although Cluster 2 accounts for the smallest proportion (3%), it represents evident subsidence, and a large number of PS points have sedimentation rates exceeding −10 mm/a. It is mainly distributed in the central part of Minhang District and the northern part and the coastal areas of Pudong New District. The evident subsidence regions which are marked as Cluster 2 are surrounded by the slight subsidence labeled Cluster 4. As shown in Figure 18c, they all show a tendency of slowing down in subsidence rate after 2010.
Figure 18d illustrates that the deformation rates of Clusters 1, 3, and 4 are all distributed around zero, making it difficult to distinguish them in the deformation rate map (Figure 5). However, in fact, these PS points possess different temporal evolution characteristics. Especially, Cluster 4, which accounts for 34%, has the temporal evolution characteristics of slight subsidence, which may develop into severe subsidence over time. However, this potential hazard is easily overlooked when only examining the deformation rate map.

5. Conclusions

This paper jointly utilized multi-platform SAR satellites, including ALOS-1 PALSAR, ENVISAT-ASAR, and Sentinel-1A, to obtain the long-term surface deformation in Shanghai during the period between 2007 and 2018. By leveraging the PCA results, combined with groundwater exploitation and groundwater level change, precipitation, temperature, and engineering geological and human activities, we quantitatively analyzed the driving factors of land subsidence at Pudong International Airport during 2007–2010 and 2015–2018. Finally, combined with the K-means unsupervised learning clustering methodology, the classification results of temporal deformation characteristics in Shanghai during 2007–2018 were obtained. The conclusions are as follows:
(1) The LOS deformation rate in Shanghai during 2007–2010 spanned from −30 mm/a up to 20 mm/a. The subsidence rate has slowed down over time. By 2015–2018, the LOS deformation rate in Shanghai ranged from −20 mm/a to 10 mm/a. Shanghai exhibited vertical deformation rates varying between −30 mm/a and 20 mm/a during the period spanning from 2007 to 2018, with significant uneven subsidence observed during the monitoring period, characterized by a spatial pattern of stability in the northwest and subsidence in the southeast. The RMSEs of the deformation rates between ALOS-1 PALSAR, ENVISAT ASAR, Sentinel-1A, and the field measurements are 0.41 mm/a, 0.16 mm/a, and 0.92 mm/a, respectively. The RMSE of the fusion result of ALOS-ENVISAT-S1A is 0.16 mm/a, which is a reduction of 61%–83% compared to single-platform SAR satellites, indicating that the integration of multi-platform SAR satellites can not only obtain long-term deformation results but also improve the precision of monitoring results.
(2) Throughout the observation period, Pudong International Airport suffered from serious uneven subsidence. Specifically, the consolidated western area was relatively stable, while the newly reclaimed eastern area suffered from severe subsidence with a subsidence rate exceeding −20 mm/a. For the first time in this region, PCA was employed to quantitatively assess the contributing factors of land subsidence, offering significant practical implications for targeted prevention and mitigation of land subsidence. From 2007 to 2010, 70.66% of the land subsidence at Pudong International Airport was caused by reclamation engineering activities, 7.98% by groundwater exploitation, and 6.92% jointly by precipitation and groundwater level changes. Between 2015 and 2018, 71.59% of the land subsidence was attributed to reclamation engineering activities, and 13.11% to groundwater exploitation. These results can provide decision-making basis for the effective control of land subsidence of important infrastructure and offer reference for similar studies in other urban areas.
(3) The time series deformation characteristics of Shanghai during 2007–2018 can be mainly divided into four categories. Among them, Cluster 1 is the uplift type, accounting for 10%, mainly distributed in the southern portion of Pudong New District and the eastern portion of Fengxian District. Cluster 2 is the evident subsidence type, accounting for 3%, primarily located within the central portion of Minhang District, the northern portion of Pudong New District, and the coastal areas. Cluster 3 is the stable type, accounting for 53%, distributed throughout the study area. Cluster 4 is the slight subsidence type, accounting for 34%, mainly distributed in Minhang District, the northern portion of Fengxian District, and the central portion of Pudong New District.

Author Contributions

Conceptualization, Y.C. and Q.Z.; methodology, Y.C. and Q.Z.; software, Y.C.; validation, Y.C. and Q.Z.; formal analysis, Y.C.; investigation, Y.C.; resources, Y.C.; data curation, Y.C.; writing—original draft preparation, Y.C.; writing—review and editing, Y.C.; visualization, Y.C.; supervision, Q.Z.; project administration, Q.Z. All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

Data are contained within the article.

Acknowledgments

We sincerely thank European Space Agency (ESA) for providing ENVISAT ASAR and Sentinel-1 data, and the ALOS-1 PALSAR data were freely provided by Japan Aerospace Exploration Agency (JAXA). We also thank the Shanghai Public Data Open Platform for providing land subsidence monitoring data, and the groundwater level monitoring data were provided by the Shanghai Geological Data Information Sharing Platform. The temperature and precipitation datasets were provided by National Tibetan Plateau/Third Pole Environment Data Center (http://data.tpdc.ac.cn, accessed on 1 June 2024). InSAR processing was supported by the High-performance Computing Platform of Peking University. We thank JPL for providing the ISCE software package. We thank Andrew Hooper for providing the StaMPS software package.

Conflicts of Interest

The authors declare no conflicts of interest.

References

  1. Rigamonti, S.G.; Frattini, D.P.; Crosta, G.B. A Multivariate Time Series Analysis of Ground Deformation Using Persistent Scatterer Interferometry. Remote Sens. 2023, 15, 3082. [Google Scholar] [CrossRef]
  2. Wang, R.; Yang, T.; Yang, M.; Liao, M.; Lin, J. A safety analysis of elevated highways in Shanghai linked to dynamic load using long-term time-series of InSAR stacks. Remote Sens. Lett. 2019, 10, 1133–1142. [Google Scholar] [CrossRef]
  3. Wu, S.; Zhang, B.; Liang, H.; Wang, C.; Ding, X.; Zhang, L. Detecting the Deformation Anomalies Induced by Underground Construction Using Multiplatform MT-InSAR: A Case Study in To Kwa Wan Station, Hong Kong. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 9803–9814. [Google Scholar] [CrossRef]
  4. Zheng, Y.; Peng, J.; Chen, X.; Huang, C.; Chen, P.; Li, S.; Su, Y. Spatial and Temporal Evolution of Ground Subsidence in the Beijing Plain Area Using Long Time Series Interferometry. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2023, 16, 153–165. [Google Scholar] [CrossRef]
  5. Berardino, P.G.; Lanari, F.R.; Sansosti, E. A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms. IEEE Trans. Geosci. Remote Sens. 2002, 40, 2375–2383. [Google Scholar] [CrossRef]
  6. Raucoules, D.B.; Michele, B.M.; Cozannet, G.L.; Closset, L.; Veldkamp, B.H.; Bateson, T.L.; Crosetto, M.; Agudo, M.; Engdahl, M. Validation and intercomparison of persistent scatterers interferometry: PSIC4 project results. J. Appl. Geophys. 2009, 68, 335–347. [Google Scholar] [CrossRef]
  7. Ferretti, A.; Fumagalli, A.; Novali, F.; Prati, C.; Rocca, F.; Rucci, A. A New Algorithm for Processing Interferometric Data-Stacks: SqueeSAR. IEEE Trans. Geosci. Remote Sens. 2011, 49, 3460–3470. [Google Scholar] [CrossRef]
  8. Ferretti, A.; Prati, C.; Rocca, F. Permanent scatterers in SAR interferometry. IEEE Trans. Geosci. Remote Sens. 2001, 39, 8–20. [Google Scholar] [CrossRef]
  9. Hooper, A.; Zebker, H.; Segall, P.; Kampes, B. A new method for measuring deformation on volcanoes and other natural terrains using InSAR persistent scatterers. Geophys. Res. Lett. 2004, 31, L23611. [Google Scholar] [CrossRef]
  10. Yang, M.; Li, M.; Huang, C.; Zhang, R.; Liu, R. Exploring the InSAR Deformation Series Using Unsupervised Learning in a Built Environment. Remote Sens. 2024, 16, 1375. [Google Scholar] [CrossRef]
  11. Zhang, J.; Ke, C.; Shen, X.; Lin, J.; Wang, R. Monitoring Land Subsidence along the Subways in Shanghai on the Basis of Time-Series InSAR. Remote Sens. 2023, 15, 908. [Google Scholar] [CrossRef]
  12. Dai, K.; Liu, G.; Li, Z.; Li, T.; Yu, B.; Wang, X. Extracting Vertical Displacement Rates in Shanghai (China) with Multi-Platform SAR Images. Remote Sens. 2015, 7, 9542–9562. [Google Scholar] [CrossRef]
  13. Dong, S.; Samsonov, S.; Yin, H.; Ye, S.; Cao, Y. Time-series analysis of subsidence associated with rapid urbanization in Shanghai, China measured with SBAS InSAR method. Environ. Earth Sci. 2014, 72, 677–691. [Google Scholar] [CrossRef]
  14. Perissin, D.; Wang, Z.; Lin, H. Shanghai subway tunnels and highways monitoring through Cosmo-SkyMed persistent scatterers. ISPRS J. Photogramm. Remote Sens. 2012, 73, 58–67. [Google Scholar] [CrossRef]
  15. Qin, X.; Yang, T.; Yang, M.; Zhang, L.; Liao, M. Health Diagnosis of Major Transportation Infrastructures in Shanghai Metropolis Using High-Resolution Persistent Scatterer Interferometry. Sensors 2017, 17, 2270. [Google Scholar] [CrossRef]
  16. An, B.; Jiang, Y.; Wang, C.; Shen, P.; Song, T.; Hu, C.; Liu, K. Ground Infrastructure Monitoring in Coastal Areas Using Time-Series inSAR Technology: The Case Study of Pudong International Airport, Shanghai. Int. J. Digit. Earth. 2023, 16, 355–374. [Google Scholar] [CrossRef]
  17. Zhao, Q.; Pepe, A.; Gao, Z.; Lu, M.; Bonano, M.; Wang, H.; Tang, X. A DInSAR Investigation of the Ground Settlement Time Evolution of Ocean-Reclaimed Lands in Shanghai. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2015, 8, 1763–1781. [Google Scholar] [CrossRef]
  18. Haghighi, M.H.; Motagh, M. Ground surface response to continuous compaction of aquifer system in Tehran, Iran: Results from a long-term multi-sensor InSAR analysis. Remote Sens. Environ. 2019, 221, 534–550. [Google Scholar] [CrossRef]
  19. Yastika, P.E.; Shimizu, N.; Abidin, H.Z. Monitoring of long-term land subsidence from 2003 to 2017 in coastal area of Semarang, Indonesia by SBAS DInSAR analyses using Envisat-ASAR, ALOS-PALSAR, and Sentinel-1A SAR data. Adv Space Res. 2019, 63, 1719–1736. [Google Scholar] [CrossRef]
  20. Zhang, B.; Chang, L.; Stein, A. A model-backfeed deformation estimation method for revealing 20-year surface dynamics of the Groningen gas field using multi-platform SAR imagery. Int. J. Appl. Earth Obs. Geoinf. 2022, 111, 102847. [Google Scholar] [CrossRef]
  21. Zhao, Q.; Ma, G.; Wang, Q.; Yang, T.; Liu, M.; Gao, W.; Falabella, F.; Mastro, P.; Pepe, A. Generation of long-term InSAR ground displacement time-series through a novel multi-sensor data merging technique: The case study of the Shanghai coastal area. ISPRS J. Photogramm. Remote Sens. 2019, 154, 10–27. [Google Scholar] [CrossRef]
  22. Wang, B.; Zhao, C.; Zhang, Q.; Lu, Z.; Pepe, A. Long-term continuously updated deformation time series from multisensor InSAR in Xi’an, China from 2007 to 2021. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2021, 14, 7297–7309. [Google Scholar] [CrossRef]
  23. Pepe, A.; Bonano, M.; Zhao, Q.; Yang, T.; Wang, H. The use of C-/X-band time-gapped SAR data and geotechnical models for the study of Shanghai’s ocean-reclaimed lands through the SBAS-DInSAR technique. Remote Sens. 2016, 8, 911. [Google Scholar] [CrossRef]
  24. Gong, P.; Li, X.; Zhang, W. 40-Year (1978–2017) human settlement changes in China reflected by impervious surfaces from satellite remote sensing. Sci. Bull. 2019, 64, 756–763. [Google Scholar] [CrossRef]
  25. Zebker, H.; Villasenor, J. Decorrelation in interferometric radar echoes. IEEE Trans. Geosci. Remote Sens. 1992, 30, 950–959. [Google Scholar] [CrossRef]
  26. Zhang, H.; Zeng, Q.; Liu, Y.; Li, X.; Gao, L. The Optimum Selection of Common Master Image for Series of Differential SAR Processing to Estimate Long and Slow Ground Deformation. In Proceedings of the IGARSS 2005—2005 IEEE International Geoscience and Remote Sensing Symposium, Seoul, Republic of Korea, 25–29 July 2005; pp. 4586–4589. [Google Scholar]
  27. Hooper, A.; Segall, P.; Zebker, H. Persistent scatterer interferometric synthetic aperture radar for crustal deformation analysis, with application to Volca’n Alcedo, Gala´pagos. J. Geophys. Res. Solid Earth 2007, 112, B07407. [Google Scholar] [CrossRef]
  28. Yu, L.; Yang, T.L.; Zhao, Q.; Liu, M.; Pepe, A. The 2015–2016 Ground Displacements of the Shanghai Coastal Area Inferred from a Combined COSMO-SkyMed/Sentinel-1 DInSAR Analysis. Remote Sens. 2017, 9, 1194. [Google Scholar] [CrossRef]
  29. Yang, P.; Tang, Y.; Zhou, N.; Wang, J. Consolidation settlement of Shanghai dredger fill under self-weight using centrifuge modeling test. J. Cent. S. Univ. Technol. 2008, 39, 862–866. (In Chinese) [Google Scholar]
  30. Jolliffe, I.T.; Cadima, J. Principal component analysis: A review and recent developments. Philos. Trans. R. Soc. Math. Phys. Eng. Sci. 2016, 374, 20150202. [Google Scholar] [CrossRef]
  31. Festa, D.; Novellino, A.; Hussain, E.; Bateson, L.; Casagli, N.; Confuorto, P.; Soldato, M.D.; Raspini, F. Unsupervised detection of InSAR time series patterns based on PCA and K-means clustering. Int. J. Appl. Earth Obs. Geoinf. 2023, 118, 103276. [Google Scholar] [CrossRef]
  32. Abdi, H.; Williams, L.J. Principal component analysis. Wiley Interdiscip. Rev. Comput. Stat. 2010, 2, 433–459. [Google Scholar] [CrossRef]
  33. Chaussard, E.; Bürgmann, R.; Shirzaei, M.; Fielding, E.J.; Baker, B. Predictability of hydraulic head changes and characterization of aquifer-system and fault properties from InSAR-derived ground deformation. J. Geophys. Res. Solid Earth. 2014, 119, 6572–6590. [Google Scholar] [CrossRef]
  34. Gaddes, M.E.; Hooper, A.; Bagnardi, M.; Inman, H.; Albino, F. Blind Signal Separation Methods for InSAR: The Potential to Automatically Detect and Monitor Signals of Volcanic Deformation. J. Geophys. Res. Solid Earth. 2018, 123, 10226–10251. [Google Scholar] [CrossRef]
  35. Ebmeier, S.K. Application of independent component analysis to multitemporal InSAR data with volcanic case studies. J. Geophys. Res. Solid Earth. 2016, 121, 8970–8986. [Google Scholar] [CrossRef]
  36. Chen, Y.; Tan, K.; Yan, S.; Zhang, K.; Zhang, H.; Liu, X.; Li, H.; Sun, Y. Monitoring land surface displacement over Xuzhou (China) in 2015-2018 through PCA-based correction Applied to SAR interferometry. Remote Sens. 2019, 11, 1494. [Google Scholar] [CrossRef]
  37. Richman, M.B. Rotation of principal components. J. Climatol. 1986, 6, 293–335. [Google Scholar] [CrossRef]
  38. Wang, G.; Li, P.; Li, Z.; Liang, C.; Wang, H. Coastal subsidence detection and characterization caused by brine mining over the Yellow River Delta using time series InSAR and PCA. Int. J. Appl. Earth Obs. Geoinf. 2022, 114, 103077. [Google Scholar] [CrossRef]
  39. MacQueen, J. Some methods for classification and analysis of multivariate observations. In Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, Berkeley, CA, USA, 21–18 June 1965 and 27 December 1965–7 January 1966; pp. 281–297. [Google Scholar]
  40. Taravatrooy, N.; Nikoo, M.R.; Sadegh, M.; Parvinnia, M. A hybrid clustering-fusion methodology for land subsidence estimation. Nat. Hazards 2018, 94, 905–926. [Google Scholar] [CrossRef]
  41. Izumi, Y.; Nico, G.; Sato, M. Time-Series Clustering Methodology for Estimating Atmospheric Phase Screen in Ground-Based InSAR Data. IEEE Trans. Geosci. Remote Sens. 2022, 60, 5206309. [Google Scholar] [CrossRef]
  42. Abdullahi, S.; Schardt, M.; Pretzsch, H. An unsupervised two-stage clustering approach for forest structure classification based on X-band InSAR data—A case study in complex temperate forest stands. Int. J. Appl. Earth Obs. Geoinf. 2017, 57, 36–48. [Google Scholar] [CrossRef]
  43. Peng, M.; Motagh, M.; Lu, Z.; Xia, Z.; Guo, Z.; Zhao, C.; Liu, Q. Characterization and Prediction of InSAR-Derived Ground Motion with ICA-Assisted LSTM Model. Remote Sens. Environ. 2024, 301, 113923. [Google Scholar] [CrossRef]
  44. Wei, M.; Sandwell, D. Decorrelation of L-Band and C-Band Interferometry Over Vegetated Areas in California. IEEE Trans. Geosci. Remote Sens. 2010, 48, 2942–2952. [Google Scholar] [CrossRef]
  45. Morishita, Y.; Hanssen, R. Temporal Decorrelation in L-, C-, and X-band Satellite Radar Interferometry for Pasture on Drained Peat Soils. IEEE Trans. Geosci. Remote Sens. 2015, 53, 1096–1104. [Google Scholar] [CrossRef]
  46. Chong, Y.; Zeng, Q.; Long, J. The Influence of SAR Image Resolution, Wavelength and Land Cover Type on Characteristics of Persistent Scatterer. PFG–J. Photogramm. Remote Sens. Geoinf. Sci. 2024, 92, 271–290. [Google Scholar] [CrossRef]
  47. Huang, S.; Zebker, H. Persistent Scatterer Density by Image Resolution and Terrain Type. IEEE J. Sel. Top. Appl. Earth Obs.Remote Sens. 2019, 12, 2069–2079. [Google Scholar] [CrossRef]
  48. Shi, Y.; Yan, X.; Chen, D. Construction of engineering geological structure and geological condition evaluation of Shanghai sea-land body. Hydrogeol. Eng. Geol. 2017, 44, 96–101. (In Chinese) [Google Scholar]
Remotesensing 16 04188 g001
Figure 1. The geographic location of Shanghai and spatial coverage of SAR datasets.
Figure 1. The geographic location of Shanghai and spatial coverage of SAR datasets.
Remotesensing 16 04188 g001
Remotesensing 16 04188 g002
Figure 2. SAR image acquisition dates and the spatial baselines of three SAR sensors with respect to the reference image for each sensor. The star represents the reference image of each sensor.
Figure 2. SAR image acquisition dates and the spatial baselines of three SAR sensors with respect to the reference image for each sensor. The star represents the reference image of each sensor.
Remotesensing 16 04188 g002
Remotesensing 16 04188 g003
Figure 3. The data processing flowchart of this study.
Figure 3. The data processing flowchart of this study.
Remotesensing 16 04188 g003
Remotesensing 16 04188 g004
Figure 4. Spatial–temporal baseline configurations of SAR datasets: (a) ALOS-1 PALSAR, (b) ENVISAT ASAR, and (c) Sentinel-1A.
Figure 4. Spatial–temporal baseline configurations of SAR datasets: (a) ALOS-1 PALSAR, (b) ENVISAT ASAR, and (c) Sentinel-1A.
Remotesensing 16 04188 g004
Remotesensing 16 04188 g005
Figure 5. Annual LOS deformation rates obtained from three SAR datasets: (a) ALOS-1 PALSAR, (b) ENVISAT ASAR, and (c) Sentinel-1A.
Figure 5. Annual LOS deformation rates obtained from three SAR datasets: (a) ALOS-1 PALSAR, (b) ENVISAT ASAR, and (c) Sentinel-1A.
Remotesensing 16 04188 g005
Remotesensing 16 04188 g006
Figure 6. (a) Average vertical deformation rate and (b) time series cumulative surface deformation obtained from ALOS-ENVISAT-S1A fusion results for Shanghai from 2007 to 2018.
Figure 6. (a) Average vertical deformation rate and (b) time series cumulative surface deformation obtained from ALOS-ENVISAT-S1A fusion results for Shanghai from 2007 to 2018.
Remotesensing 16 04188 g006
Remotesensing 16 04188 g007
Figure 7. Comparisons of long-term time series cumulative deformation obtained by ALOS-ENVISAT-S1A and self-weight consolidation settlement model: (a) P1, (b) P2, (c) P3, and (d) P4.
Figure 7. Comparisons of long-term time series cumulative deformation obtained by ALOS-ENVISAT-S1A and self-weight consolidation settlement model: (a) P1, (b) P2, (c) P3, and (d) P4.
Remotesensing 16 04188 g007
Remotesensing 16 04188 g008
Figure 8. (a) Scatterplot and (b) correlation coefficient graph of TS-InSAR deformation rate and field measurements.
Figure 8. (a) Scatterplot and (b) correlation coefficient graph of TS-InSAR deformation rate and field measurements.
Remotesensing 16 04188 g008
Remotesensing 16 04188 g009
Figure 9. Landsat TM/ETM optical images of Pudong International Airport for the following years: (a) 2007, (b) 2010, (c) 2015, and (d) 2018.
Figure 9. Landsat TM/ETM optical images of Pudong International Airport for the following years: (a) 2007, (b) 2010, (c) 2015, and (d) 2018.
Remotesensing 16 04188 g009
Remotesensing 16 04188 g010
Figure 10. Vertical annual deformation rates of Pudong International Airport during (a) 2007–2010 and (b) 2015–2018 (base image is from Google Map).
Figure 10. Vertical annual deformation rates of Pudong International Airport during (a) 2007–2010 and (b) 2015–2018 (base image is from Google Map).
Remotesensing 16 04188 g010
Remotesensing 16 04188 g011
Figure 11. PCA result derived from ALOS-ENVISAT: (a) variance explained by the PC 1–4 and (b) eigenvectors obtained from PC 1–4.
Figure 11. PCA result derived from ALOS-ENVISAT: (a) variance explained by the PC 1–4 and (b) eigenvectors obtained from PC 1–4.
Remotesensing 16 04188 g011
Remotesensing 16 04188 g012
Figure 12. PCA result derived from Sentinel-1A: (a) variance explained by the PC 1–4 and (b) eigenvectors obtained from PC 1–4.
Figure 12. PCA result derived from Sentinel-1A: (a) variance explained by the PC 1–4 and (b) eigenvectors obtained from PC 1–4.
Remotesensing 16 04188 g012
Remotesensing 16 04188 g013
Figure 13. Correlation map between eigenvectors of PC 1–3 obtained from ALOS-ENVISAT and temperature, groundwater level, precipitation, groundwater extraction volume, and impervious surface area: (a) PC1, (b) PC2, and (c) PC3.
Figure 13. Correlation map between eigenvectors of PC 1–3 obtained from ALOS-ENVISAT and temperature, groundwater level, precipitation, groundwater extraction volume, and impervious surface area: (a) PC1, (b) PC2, and (c) PC3.
Remotesensing 16 04188 g013
Remotesensing 16 04188 g014
Figure 14. Correlation map between eigenvectors of PC 1–2 obtained from Sentinel-1A and temperature, groundwater level, precipitation, groundwater extraction volume, and impervious surface area: (a) PC and (b) PC2.
Figure 14. Correlation map between eigenvectors of PC 1–2 obtained from Sentinel-1A and temperature, groundwater level, precipitation, groundwater extraction volume, and impervious surface area: (a) PC and (b) PC2.
Remotesensing 16 04188 g014
Remotesensing 16 04188 g015
Figure 15. The deformation rate of Pudong International Airport on the east–west profile line: (a) AB profile and (b) CD profile.
Figure 15. The deformation rate of Pudong International Airport on the east–west profile line: (a) AB profile and (b) CD profile.
Remotesensing 16 04188 g015
Remotesensing 16 04188 g016
Figure 16. P1–P7 time series cumulative deformation acquired by ALOS-ENVISAT in the runway and terminal areas.
Figure 16. P1–P7 time series cumulative deformation acquired by ALOS-ENVISAT in the runway and terminal areas.
Remotesensing 16 04188 g016
Remotesensing 16 04188 g017
Figure 17. Q1–Q7 time series cumulative deformation acquired by Sentinel-1A in the runway and terminal areas.
Figure 17. Q1–Q7 time series cumulative deformation acquired by Sentinel-1A in the runway and terminal areas.
Remotesensing 16 04188 g017
Remotesensing 16 04188 g018
Figure 18. K-means clustering results of the long-term time series deformation obtained from ALOS-ENVISAT-S1A: (a) spatial distribution of each cluster, (b) percentage of each cluster, (c) time series of cumulative deformation of the cluster center, and (d) violin map of the annual deformation velocity for each cluster.
Figure 18. K-means clustering results of the long-term time series deformation obtained from ALOS-ENVISAT-S1A: (a) spatial distribution of each cluster, (b) percentage of each cluster, (c) time series of cumulative deformation of the cluster center, and (d) violin map of the annual deformation velocity for each cluster.
Remotesensing 16 04188 g018
Table 1. Parameter information of SAR datasets.
Table 1. Parameter information of SAR datasets.
SatellitePolarizationModeOrbit DirectionNumber of ImageAcquisition Span
ALOS-1 PALSARHHFBDAscending19January
2007 to September 2010
ENVISAT ASARVVIMSAscending22February 2007 to September 2010
Sentinel-1AVVIWAscending 20April 2015 to May 2018
Table 2. Name of the foundation soil level of Pudong International Airport.
Table 2. Name of the foundation soil level of Pudong International Airport.
Geologic TimeSoil Layer NameBurial Depth (m)Genetic TypeCompactness
Holocene Q4-3Dredger fill0LaborLoose
Holocene Q4-3Brown-yellow clay0.5–2Littoral estuaryPlasticity
Holocene Q4-2Gray silty clay3–7Coastal–shallow seaRheoplastic
Pleistocene Q3-2Dark green clay15–32Estuary–lakePlastic–hard plastic
Pleistocene Q3-2Grass yellow-gray silty sand20–35Estuarine–coastalMedium–dense
Pleistocene Q3-2Gray fine sand35–40Littoral estuaryDense
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Chong, Y.; Zeng, Q. Long-Term Ground Deformation Monitoring and Quantitative Interpretation in Shanghai Using Multi-Platform TS-InSAR, PCA, and K-Means Clustering. Remote Sens. 2024, 16, 4188. https://doi.org/10.3390/rs16224188

AMA Style

Chong Y, Zeng Q. Long-Term Ground Deformation Monitoring and Quantitative Interpretation in Shanghai Using Multi-Platform TS-InSAR, PCA, and K-Means Clustering. Remote Sensing. 2024; 16(22):4188. https://doi.org/10.3390/rs16224188

Chicago/Turabian Style

Chong, Yahui, and Qiming Zeng. 2024. "Long-Term Ground Deformation Monitoring and Quantitative Interpretation in Shanghai Using Multi-Platform TS-InSAR, PCA, and K-Means Clustering" Remote Sensing 16, no. 22: 4188. https://doi.org/10.3390/rs16224188

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop