«remote sensing ISSN 2072-4292 Article L- and X-Band Multi-Temporal InSAR Analysis of Tianjin Subsidence Qingli Luo ...»
Remote Sens. 2014, 6, 7933-7951; doi:10.3390/rs6097933
L- and X-Band Multi-Temporal InSAR Analysis of
Qingli Luo 1,*, Daniele Perissin 2, Yuanzhi Zhang 3 and Youliang Jia 4
Center of Remote Sensing, Tianjin University, No. 92, Weijin Road, Nankai District,
Tianjin 300072, China 2 School of Civil Engineering, Purdue University, 550 Stadium Mall Drive, West Lafayette, IN 47907–2051, USA; E-Mail: firstname.lastname@example.org 3 National Astronomical Observatories, Chinese Academy of Sciences, Beijing 100012, China;
E-Mail: email@example.com 4 Tianjin Institute of Surveying and Mapping, No. 9, Lingkou Road, Xiqing District, Tianjin 3000381, China; E-Mail: firstname.lastname@example.org * Author to whom correspondence should be addressed; E-Mail: email@example.com;
Received: 9 April 2014; in revised form: 28 July 2014 / Accepted: 29 July 2014 / Published: 26 August 2014 Abstract: When synthetic aperture radar interferometry (InSAR) technology is applied in the monitoring of land subsidence, the sensor band plays an important role. An X-band SAR system as TerraSAR-X (TSX) provides high resolution and short revisit time, but it has no capability of global coverage. On the other side, an L-band sensor as Advanced Land Observing Satellite-Phased Array L-band Synthetic Aperture Radar (ALOS-PALSAR) has global coverage and it produces highly coherent interferograms, but it provides much less details in time and space. The characteristics of these two satellites from different bands can be regarded as complementary. In this paper, we firstly present a possible strategy for X-band optimized acquisition planning combining with L-band. More importantly, we also present the multi-temporal InSAR (MT-InSAR) analysis results from 23 ALOS-PALSAR images and 37 TSX data, which show the complementarity of L- and X-band allows measuring deformations both in urban and non-urban areas. Furthermore, the validation between MT-INSAR and leveling/GPS has been carried out. The combination analysis of L- and X-band MT-InSAR results effectively avoids the limitation of X-band, providing a way to define the shape and the borderline of subsiding center and helps us to understand the Remote Sens. 2014, 6 7934 subsidence mechanism. Finally, the geological interpretation of the detected subsidence center is given.
Keywords: subsidence monitoring; L- and X-band; multi-temporal InSAR (MT-InSAR);
Permanent Scatterers (PS); Quasi-Permanent Scatterers (QPS)
1. Introduction Ground subsidence has always been one of the most severe and widespread geological hazards in Chinese cities . Compared with the single-point-measurement methods of ground leveling and GPS techniques, an advanced remote sensing technique, referred to as differential synthetic aperture radar interferometry (D-InSAR), was introduced as a feasible way to monitor deformation over wide areas with centimeter-to-millimeter accuracy. Nevertheless, the technique was mainly limited by the spatiotemporal decorrelation and atmospheric disturbance [2,3].
To compensate for these shortages, the permanent scatterers (PS) technique [4,5] was proposed as the most powerful multi-temporal tool for wide area deformation monitoring with millimetric accuracy and high spatial sampling density , well addressing the above well-known limitations of conventional D-InSAR by examining point-like radar targets (buildings, structures, etc.). In the following years, different implementations of PS technique have been realized by many research groups [7,8], commonly referred as persistent scatterers interferometry (PSI). In addition, another approach has been proposed to extract information from distributed targets by generating many interferograms formed in the conventional way [9,10], known as “small baseline” approach. These two broad categories of methods are exploiting two different kinds of radar scattering targets within resolution cells. PSI is looking for a point-like scatterer while small baseline methods are looking for a distributed scatterer within resolution cells. More recently, methods have been proposed by exploring both types of scatterer [11–13]. To supplement it, the detection of partially coherent targets has been carried out by Quasi-PS (QPS) technique . These different techniques can be optional methods for multi-temporal InSAR (MT-InSAR) analysis when applied to monitor deformation in diverse applications according to real conditions and restrictions. In recent years, significant attention has been focused on investigating subsidence over urban areas [15–17] with MT-InSAR analysis.
The main drawback of SAR images is the low resolutions. With the launch of new generation high resolution SAR satellites, the level of details visible in SAR images increased dramatically. For example, TerraSAR-X (TSX) can provide high resolution data to the scale of 1 m [18,19]. Moreover, compared with the 35 days of ERS and ENVISAT, the short revisit time of 11 days favors a fast build-up of interferometric data stacks, for which it is possible to develop thermal effects and a multi-scatters dispersion model [20,21]. Due to the incredible spatial density of identified PS points and high temporal coherence, X-band PSI results has been proved to have the potential of monitoring simultaneously of multiple towns with relatively high accuracy, as well as large-scale man-made linear features (LMLFs) such as highways, railways and powerlines [22,23].
TSX has the advantages of high precision and short revisit period. However, almost no information can be detected by applying PS analysis with X-band over extra-urban areas . In the contrast, Remote Sens. 2014, 6 7935 ALOS-PALSAR satellite can provide SAR data with global coverage , which has relatively high temporal and spatial coherence even in vegetated and forested areas . Besides, the coverage of TSX in strip mode is not globally continuous. Then, how to make an optimized acquisition planning becomes a challenging problem for data provider. The characteristics of these two satellites can be regarded as complementary. Combining L- and X-band can enhance the ability of subsidence monitoring and provide more reliable results.
In this paper, we firstly present a strategy for X-band optimized acquisition planning combining with L-band. Moreover, with multi-band InSAR data, we present MT-InSAR analysis results by applying PS/QPS technique for subsidence monitoring, taking Tianjin suburbs as the study area, which is well validated with leveling and GPS data from two continuously operating reference stations (CORS). PS, QPS techniques were applied in our processing with the software SARPROZ, developed by Daniele Perissin. The results comparison show the complementarity of L- and X-band allows measuring deformations both in urban and non-urban areas. The combination analysis of X-band with L-band data offers a chance for us to describe the shape of subsiding centers and to further understand the causes of different subsidence rates. Besides, integration analysis of geological materials facilitates our understanding of subsidence mechanisms in Tianjin suburbs.
This article is organized as follows. The strategy for X-band optimized acquisition planning combining with L-band is described in Section 2. Then, MT-InSAR analysis with L- and X-band are presented using the datasets from TSX and ALOS-PALSAR and the MT-InSAR results are validated with leveling data and GPS data from two CORS in Section 3. Section 4 shows the combination analysis with L- and X-band MT-InSAR analysis and geological interpretation. The last section is the conclusions and perspectives.
2. X-Band Optimized Acquisition Planning Combining with L-Band
The strategy we adopted can be divided into the following steps as illustrated in Figure 1:
(1) D-InSAR analysis is applied for processing L-band data over the whole region. In this way we can exploit the high coherence and the wide coverage of L-band to find hot spot areas (HSAs) affected by ground surface subsidence. (2) Once the HSAs are identified, we can focus the attention on smaller areas and collect the corresponding TSX images, acquired with shorter revisit time. We can then carry out MT-InSAR analysis and study the displacement time series.
L-band (23 cm wavelength) data supports the relatively high temporal and spatial coherence data with wide-area coverage. In order to make full use of temporal and spatial coherence, we constructed interferometric pairs among these time series SAR images based on minimum spanning tree (MST) algorithm . It provides a different choice for the selection of interferometric pairs. Then, we can construct several interferometric pairs from available time series SAR images. An external digital elevation model (DEM) was applied to remove topographic phase, and thus the differential interferograms were generated. From the analysis of these differential interferograms, the HSAs were highlighted.
After the HSAs were identified by L-band D-InSAR analysis, we try to apply X-band PS analysis for one of the potential deformation areas. Due to better resolution, X-band is expected to have the potential ability to monitor subsidence of individual infrastructure including high speed railway as well as urban buildings. The applied PS processing steps are as follows: master image selection, SAR data Remote Sens. 2014, 6 7936 focusing and registration, baseline construction, DEM simulation, differential interferogram generation, persistent scatterers candidate (PSC) selection, multi-image sparse grid phase unwrapping, atmospheric phase screen (APS) estimation and removal, PS point selection, PS point displacement history analysis and average deformation estimation.
Figure 1. The framework of X-band optimized acquisition planning in combination with L-band, including multi-baseline construction based on MST, L-band D-InSAR analysis, potential deformation area detected and X-band PS analysis.
3. Multi-Band MT-InSAR Analysis
3.1. MT-InSAR Analysis with L- and X-Band With the rapid development of industry and urbanization, excessive groundwater withdrawal in Tianjin leads itself to one of the most serious ground subsidence regions in China. The study area is located in the west of Tianjin, including Wuqing District, Jinghai County and several other towns.
Figure 2 shows the coverage of L-Band (red rectangle) data and X-Band (blue one) data which are superimposed on Google Earth.
The available SAR datasets are composed of 23 ALOS-PALSAR L-band images acquired from 17 January 2007 to 28 October 2010 and 37 TSX imageries acquired from 29 April 2009 to 11 November 2010. ALOS-PALSAR images are acquired by single polarization model with the spatial resolution of about 7 m on the ground. The incidence angles of TSX and ALOS-PALSAR are
41.08 and 38.70 degrees, respectively. With the aforementioned multi-band InSAR methodology, MTInSAR analysis can be applied to process the available SAR datasets. The basic principles and algorithms of the PS and QPS technique can be referred to [4,5,14]. The PS and QPS analysis are conducted by the processing software SARPROZ , which provides data processing, data analysis, Remote Sens. 2014, 6 7937 data visualization and data exportation in different formats. DEM from shuttle radar topography mission (SRTM) with 90 m resolution was applied for topographic phase removal. There are two major outputs for each PS analysis: average displacement velocity map derived for the whole time period and displacement history , displayed exploiting Google Earth imagery.
Figure 2. Spatial coverage of L- (red) and X-band (blue).
The geological profile is marked with white line and three drilling locations are marked as W1, W2 and W3. Color scale represents SAR backscattering.
As using the same TSX dataset and PS processing with single master image (Figure 3a), the X-band MT-InSAR results are the same as published in , which were rectified with one reference leveling data and displayed as Figure 4a. The results can then be considered as the absolute deformation rates to some extent.
For processing ALOS-PALSAR dataset, multiple reference images were selected and 98 interferograms were generated (Figure 3b). L-band QPS analysis results can be represented as Figure 4b. More than 3,867,649 QPS points (approximately 915 QPS/km2 with the temporal coherence threshold of 0.7) were identified and the subsidence velocity ranges from −190 to −10 mm/year. The results also show that Wangqingtuo, Shengfang and Nanhe Town suffer from the serious subsidence, which matches well with X-band PS results .
By comparison, the spatial distribution of the targets detected from X-band PS analysis is non-homogeneous. Most targets are identified as man-made objects and they are aggregated in residential areas, where the density of targets from X-band (1500–2000 PS/km2) is far more than that from L-band (915 QPS/km2). All the advantages of X-band allow its wide applications in monitoring LMLFs and detecting thermal expansion, which requires high precision and detail information capture.
However, almost no targets (0–10 PS/km2) can be detected in non-urban area from X-band MT-InSAR analysis. In the contrast, the spatial distribution of the targets identified from L-band MT-InSAR Remote Sens. 2014, 6 7938 analysis seems more homogeneous (915 QPS/km2) both for the agricultural parcels and residential areas. The reasonable explanation should be the longer wavelength of L-band supports good coherence even over the vegetation areas, where more distributed targets exist instead of point-like targets and they can be detected by QPS technique.
Figure 4. Linear deformation trends along line of sight (LOS) direction.
Three major subsiding centers are located in A: Wangqingtuo Town, B: Dongguguang Town, C: Shengfang Town, D: Nanhe Town. (a) Linear deformation trend of TSX using PS analysis. The colorbar ranges from −90 to −10 mm/year ; (b) Linear deformation trend of ALOS-PALSAR using the QPS technique. The colorbar ranges from −190 to −10 mm/year.