1. Introduction
Located at the northern end of the North China Plain, Beijing is ranked as the 5th most water-stressed city in the world [
1], and groundwater is the main water source for industrial, agricultural and household activities. With its rapid urban growth, there has been increasing water demand in Beijing. Previous studies [
2,
3] reveal that the Beijing region has been suffering from land subsidence due to over-exploitation of groundwater since 1935, and more seriously, the rate and extent of land subsidence shows an increasing trend. Land subsidence is a severe geohazard threating the safety of the public and urban infrastructure. Hence, continuous monitoring of land subsidence is critical for detecting potential hazards and designing compensation strategies.
Interferometric Synthetic Aperture Radar (InSAR) is a powerful tool widely used to map land subsidence over wide regions with high spatial-temporal resolutions. InSAR time series techniques such as Persistent Scatterer Interferometry (PSI) [
4,
5] and the Small Baseline Subset (SBAS) [
6,
7,
8] can minimise the limitations of traditional InSAR (e.g., spatial and temporal decorrelation and atmospheric effects) and have been demonstrated to be able to map ground displacements with precision comparable to traditional geodetic techniques such as leveling [
9,
10,
11,
12]. In this work, the Small Baseline InSAR technique has been employed to investigate land subsidence in the Beijing region and its relationship with different triggering and conditioning factors.
Gong
et al. [
13] explored the application of PSI method to monitor land subsidence in Beijing with C-band Envisat ASAR data acquired from 2003 to 2006. Their results showed that the spatial pattern of the subsidence is influenced by Quaternary faults, and land subsidence often develops in areas with clay layer thicker than 50 m. Li
et al. [
14] employed SBAS InSAR to investigate city subsidence in Beijing with two adjacent tracks of C-band Envisat ASAR images, reporting a maximum subsidence rate reaching 100 mm/year in Chaoyang district. Chen
et al. [
15] studied the spatial and temporal evolution of the groundwater depression funnels in the Beijing plain and the 2D and 3D spatial evolution of land subsidence from a model based on the combination of piezometric, GPS, and radar data. Liang
et al. [
16] monitored the accumulated crustal deformation and its characteristics in Beijing and its surrounding areas by traditional D-InSAR method with L-band SAR data. They reported that the maximum deformation rate along Line of Sight (LOS) was about 125 mm/year between 2007 and 2010. Ng
et al. [
17] applied PS InSAR to 41 Envisat ASAR images (acquired from 2003 to 2009) and 24 ALOS PALSAR images (acquired from 2007 to 2009) to investigate the land deformation rate maps over Beijing showing that the vertical displacement rates were in the range of −115 to 6 mm/year. Zhu
et al. [
18] analyzed the spatial relationship between land subsidence and three factors (groundwater level, compressible soil thickness and building areas) by means of remote sensing and Geographical Information System (GIS) tools and used the Back Propagation Neural Network (BPN) model combined with Genetic Algorithm (GA) to simulate regional distribution of land subsidence. Hu
et al. [
19] used 52 Envisat ASAR images acquired from 2003 to 2010 to derive land deformation information in Beijing, and found a maximum subsidence velocity of up to 110 mm/year. Zhang
et al. [
20] utilized stress-strain analysis and oedometer tests to characterize the hydraulic and mechanical properties of the five hydrogeologic units at different depths in the Beijing plain. The results revealed that the second (64.5–82.3 m) and third (102–117 m) aquitards contributed 39% of the total compression deformation, and the second (82.3–102 m) and third (117–148 m) confined aquifers exhibited elasto-plastic mechanical behaviour. Zhu
et al. [
21] used SAR images, optical images and hydrogeological data to study the land subsidence in the northern Beijing plain. PS InSAR was applied to C-band Envisat ASAR images acquired from 2003 to 2010 to estimate land subsidence on this area. The results indicated that the largest subsidence rates reached 52 mm/year and the silty clay layers contribute to the larger land subsidence.
Compared with medium resolution SAR imagery such as C-band Envisat ASAR and L-band ALOS PALSAR data, recent advanced imaging radar sensors such as the TerraSAR-X SAR system (X-band radar wavelength and a shorter revisiting interval) can provide higher spatial resolution and geometric accuracy imagery, which makes it possible to monitor land subsidence at a finer scale [
22,
23]. Chen
et al. [
24] utilized C-band Envisat ASAR data and X-band COSMO-SkyMED images to detect land subsidence rates in Shanghai and validated the results by spirit leveling measurements. The results indicated the location of a significant subsidence funnel and the root mean square error of differences of the point targets between two band SAR images was 3 mm/year. Luo
et al. [
25] extracted land deformation information from L-band ALOS PALSAR data between 2007 and 2010 and X-band TerraSAR-X data from 2009 to 2010. The obtained results indicated that the subsidence rate was in the range of −90 mm/year to −10 mm/year from TerraSAR-X data and −190 to −10 mm/year from PALSAR data, which was in good agreement with leveling and GPS data.
Although existing investigations have tried different ways of monitoring land subsidence in the Beijing region [
13,
14,
15,
16,
17,
18,
19,
20,
21], the consistency of the land subsidence monitoring results in this densely populated area in the world acquired by multi-wavelength InSAR time series analysis and validation by GPS data is rarely covered. Furthermore, the comprehensive and quantitative spatio-temporal analysis to identify the main conditioning and triggering factors of land subsidence still need to be systematically researched further. This paper aims to use TerraSAR-X (X-band) stripmap images collected from 2010 to 2011 and Envisat ASAR (C-band) images acquired between 2003 and 2010 to provide a new insight into the spatial and temporal distribution characteristics and main conditioning and triggering factors of land subsidence in the Beijing plain. The joint cross analysis of radar-derived deformation data and existing geo-information will help to improve the knowledge of the mechanisms that govern land subsidence to be used for the development of suitable groundwater management policies in this area. The paper is organized as follows:
Section 2 describes the geographical and geological setting of the study area. SAR images and InSAR processing used in this work is described in
Section 3.
Section 4 shows the main InSAR-derived products, validation of the results and the spatial-temporal characteristics of land subsidence over the study area.
Section 5 discusses and analyses the effect of different conditioning and triggering factors on land subsidence. The main conclusions are summarized in
Section 6.
2. Description of the Beijing Basin
Beijing is the political, cultural and economic centre of China. As one of the most populated cities in the world, its population has reached 20 million [
26]. As one of the most water-scarce cities in the world, the major water source for the Beijing municipality is groundwater, accounting for about two thirds of water use [
27].
Beijing is located in northern China (
Figure 1a), at 39°28′–41°05′ north latitude and 115°25′–117°30′ east longitude, with an area of 16,807 square kilometers, of which 6390 square kilometers are the plain. The northern and north-eastern parts of Beijing are dominated by (
Figure 1b) the Jundu Mountains, while the western part of Beijing is framed by the Taihang Mountains. The southeastern part of Beijing is alluvial-pluvial plains crossed by five rivers (
Figure 1b): Chaobai, Yongding, Wenyu, Ju and Juma. The terrain of Beijing is high in the west-north area but low in south-east part, sloping from the front of the mountain to the south-east. The elevations range from around 60–80 m.a.s.l. in the front of mountains to around 20–60 m.a.s.l. in the plains.
The study area is characterized by a monsoon-influenced semi-arid and semi-humid continental climate [
20]. The average annual temperature is around 11.7 °C. Annual precipitation, mainly concentrated in summer months (from June to September), is around 570 millimetres and constitutes the main source of water resources in Beijing [
27].
The thickness of quaternary deposits, whose lithologies change from single gravel to multilayer structures of clay with sand interbeds and whose grain-size turns from coarse to fine, varies from a few meters over the mountain-front area to hundreds of meters in the plain zone [
20,
21]. Groundwater levels change from deep to shallow and the groundwater varies greatly from phreatic to multilayer confined water throughout the study area, increasing from tens of meters in the mountain-front zone to several hundred meters in the central or southeast plain area [
28].
The sediment, forming the principal aquifer system units, predominantly consists of gravel, fine-to-coarse sand, silt and clays (
Figure 1c).
Figure 1b,c illustrate the gradual transition from fine-to-coarse sand and gravel in the proximal fan to fine sand, silt or clay in the mid and distal-fan areas.
The Quaternary aquifers of Beijing plain can be divided into three aquifer groups [
2,
20,
29]: (a) The first aquifer group (late Pleistocene-Holocene) is widespread over the Beijing plain. The depths of the bottom of this group are less than 100 m. In this aquifer group, the phreatic aquifer with single gravel structure located in the top area of the alluvial-diluvial fan and over the middle and lower parts of the alluvial-proluvial fan, distributes phreatic aquifer and the shallow confined water with a multilayer structure; (b) The second aquifer group (middle Pleistocene) is mainly located over the middle and lower segments of the alluvial-diluvial fan, which is the middle and deep confined aquifers with depths of the roof from 80 to 100 m, and depths of the bottom plate around 300 m. This aquifer group presents a multilayer structure, which is mainly made up of medium-coarse sand and small amounts of gravel; (c) The third aquifer group (early Pleistocene) consists on deep confined aquifers with the depth of the top over 300 m. This aquifer group, mainly distributed over the north-east part and the south-east part of the centre of the depression land in Beijing plain, is mainly composed of medium-coarse sand and gravel, with clay aquiclude thicker than 30 m covering the top.
The unconfined aquifer and the first confined aquifer are intensely exploited for irrigation providing about 68% of the total groundwater pumpage of the aquifer system. However, the deeper aquifer units are not significantly exploited [
20].
Corresponding to the classification of the aquifers, three compressible layers responsible of land subsidence can be recognized in Beijing [
2,
29]. The first compressible layer is widely distributed over the Beijing plain. The depth of the bottom of this layer, which is composed of normally consolidated silt, silty clay and clay with plastic or hard plastic states, is less than 100 m. The second compressible layer is widely distributed over the middle and lower parts of the alluvial-proluvial fan in Beijing. The main lithologics of this layer are silt, silty clay and clay. The depth of the bottom over the south-west part of Beijing is less than 150 m, while the depth of the bottom over the east and north parts of Beijing can reach a depth of 280 m. The compressible over-consolidated layers with plastic or hard plastic clays account for 60~80 percent of the overall thickness. The third compressible layer is distributed mainly over the centre of the depression of the Beijing plain. The depth of the roof of this layer is larger than 270 m. This layer is mainly composed of over-consolidated clays with a solid state [
28].
3. InSAR Processing
The SAR dataset used in this study consists of 41 images from Envisat-ASAR acquired from 2003 to 2010 and 14 images from TerraSAR-X (TSX) acquired from 2010 to 2011. The Envisat ASAR images were collected in Stripmap mode from descending track with VV polarisation, and the TSX images were obtained in Stripmap mode from ascending track with HH polarisation (
Figure 1). The interferograms were generated from single-look complex (SLC) images using DORIS (Delft object oriented radar interferometric software) [
30]. Among all the total amount of possible interferograms formed by pairs of images, only 91 interferograms from Envisat ASAR images and 36 interferograms from TerraSAR-X images were selected for further processing. The selection of interferograms was restricted to those interferograms with spatial and temporal baselines smaller than 1070 m and 1500 days for Envisat ASAR images, and temporal baselines smaller than 200 m and 200 days for TerraSAR-X images. The external Shuttle Radar Topography Mission (SRTM) DEM with 90 m resolution was used to remove the topographic component of the interferometric phase and geocode the interferograms.
In this study, land subsidence information was obtained by using a small baseline subset of interferograms with the StaMPS (Stanford Method for Persistent Scatterer/Multi-Temporal InSAR) technique [
31]. The processing includes three main steps:
Firstly, a network of multi-master differential interferograms was created by taking into account small spatial and temporal baselines and reduced Doppler centroid frequency differences to minimize the spatial and temporal decorrelation and topographic errors for the group of small baseline interferograms.
Figure 2 shows the network of the small baseline interferograms used for the SBAS analysis of the land subsidence in Beijing.
Secondly, after filtering the azimuth and range spectra to reduce decorrelation effects due to geometry and non-overlapping Doppler frequencies, the slowly decorrelating filtered phase (SDFP) pixels whose phase shows slow decorrelation over short time intervals were selected. For computational efficiency, the SDFP candidates were initially selected by setting a threshold for the amplitude difference dispersion, which is the standard deviation of the amplitude difference between the master and slave divided by the mean amplitude. Afterwards, an iteratively conducted phase analysis yielded the phase stability estimation of each candidate to generate the final set of SDFP pixels [
31]. The amplitude dispersion index threshold was set at 0.6 to generate the largest set of candidate pixels.
Thirdly, after the iteratively robust process of SDFP selection and correcting for spatially uncorrelated look angle errors, a three-dimensional phase unwrapping was applied on the final sets of SDFP pixels by using a statistical cost flow algorithm [
32,
33]. The time series of displacement for each SDFP pixel was derived through a least-squares inversion method [
34]. This technique estimates displacement time series without prior consideration of a temporal model for the deformation and thus allows the derivation of temporally varying deformation processes. From these values, long-wavelength atmospheric effects and orbit error are estimated from the SDFP pixels using a spatial-temporal filter method.
6. Conclusions
In this study, land subsidence due to over-extraction of groundwater in the Beijing region was investigated with SBAS InSAR using 41 Envisat ASAR and 14 TSX images. The results reveal that the Beijing region has experienced significant ground subsidence from 2003 to 2010 with a maximum accumulative displacement of 790 mm. The SBAS results have been validated by GPS measurements with a mean difference of 2.41 ± 1.84 mm/year and a RMS of 2.94 mm/year, demonstrating that SBAS InSAR can effectively monitor and detect complicated settlements. A high correlation coefficient of 0.92 and a standard deviation of 7.48 mm/year between both TerraSAR-X and Envisat ASAR InSAR-derived subsidence rates was observed, indicating the reliability of InSAR derived land subsidence rate maps.
The spatiotemporal analysis of land subsidence indicates an increasing trend in the rate and extent of land subsidence. The join spatial and/or temporal analysis of InSAR data and conditioning and triggering factors shows that land subsidence is correlated with groundwater levels, active faults, different soft soil ranges and aquifer types. Furthermore, a relationship between land subsidence and the distance to the pumping wells has been found.
The analysis of different stress-strain relationships shows that the aquifer-system exhibits different behaviour, such as quite elastic behaviour and predominantly inelastic behaviour. Additionally, the analysis of the variation of the inelastic storage coefficient (Skv) of the aquifer system computed from the stress-strain curves shows that the deformability of the aquifer system increases from the north towards the south.
The distribution and development trends of the land subsidence over the study area are obviously controlled by its geological structures. Among the biggest contributors to land subsidence are the clayey layers. The subsidence bowls located in the east, north-east and north are mostly placed over a zone with a clayey layer underneath, with a thickness around 50 to 70 m, and those subsidence bowls are located over predominantly fine-grain sediments. The InSAR data statistics and the distance to the pumping wells show that land subsidence rates are higher near the pumping wells, as expected according to the shape of a well’s cone of depression.
To summarize, accurate InSAR-derived data from two different sensors have allowed us to perform a comprehensive spatial and temporal study of the land subsidence in Beijing basin from 2003 to 2011. Complementarily, these data and existing geo-information have been used to perform a study of the role of the main conditioning and triggering factors controlling land subsidence mechanisms in this area, providing useful information for improved development of future models for the prediction of land subsidence affecting this area.