Conference Agenda

Topographic mapping and surface motion estimation with spaceborne SAR sensors are the main topics of the Dragon-4 project "Multi-baseline SAR processing for 3D/4D reconstruction (id 32278-2)” under the framework of THREE- AND FOUR- DIMENSIONAL TOPOGRAPHIC MEASUREMENT AND VALIDATION (id 32278). In Dragon-4, we work on different test sites investigating the following topics:

1. Topographic mapping with SAR. Interferometric SAR (InSAR) is the main method for the generation of digital elevation models (DEM) with SAR observations. The StereoSAR-assisted InSAR topographic mapping strategy is presented in the following three steps. Firstly, the StereoSAR DEM can be employed as a reference to remove the main topographic phase from the InSAR interferogram, which can reduce the fringe frequency and facilitate the phase unwrapping of the InSAR interferogram. Then, the StereoSAR DEM can be used to calibrate the unwrapped phase of InSAR to determine the absolute phase deviation. Finally, the StereoSAR DEM and InSAR DEM are fused by weighted averaging, in which the determination of reasonable weights is the key issue to be solved. Thus the random height error can be effectively reduced and the StereoSAR DEM can also fill in the void in the original InSAR DEM. The effectiveness of the proposed methods was demonstrated by experimental results with high-resolution TerraSAR-X data pairs for the test site of Mount Song, one of the five sacred mountains in China.

2. Urban subsidence analysis.SAR systems can measure distances and movements with high precision. Using for example PS-InSAR, deformations can be estimated with a very high precision. The long-term surveillance of urban subsidence and the infrastructure stability in Shanghai is our major research goal since Dragon-1. With data starting from ERS-1, over ENVISAT ASAR, ALOS PALSAR, up to modern systems like TerraSAR-X, COSMO-SkyMed, PALSAR-2, and Sentinel-1, we continuously monitor the subsidence over Shanghai for far over a decade now. Furthermore, the subsidence distribution of Wuhan is derived from the long-term Sentinel-1 data stack. The InSAR-derived results also highlight active motions of built-up areas and infrastructures, such as some communities and railways segments. It can benefit safety screening and risk assessment. Furthermore, the Sentinel-1 data stacks are also applied in monitoring the infrastructures such as bridges.

Abstract - 3D SAR imaging by tomographic processing of multibaseline interferometric data has emerged for operational spaceborne monitoring of forest biomass in incoming or next ESA missions. However, a few open issues, or improvement needs, still stand, in particular of radiometric accuracy of the most diffused superresolution processing, the adaptive Capon method. After its introduction in SAR Tomography by University of Pisa a decade and a half ago, despite the widespread experimental use its basic structure, and issues, remained unchanged. In this work, an alternative improved (double) adaptive algorithm for spaceborne forest SAR Tomography is tested, characterized, and tuned by simulations, showing that it can furnish a better tomographic performance trade-off than Fourier and classic Capon Tomography. First corresponding low frequency real SAR data tests are also performed.

3D SAR Tomography (TomoSAR) [1-2] is a well established technique, for which operational interests related to spaceborne SAR missions have emerged, in particular for urban and forest applications. TomoSAR stems from advanced multibaseline (MB) Interferometry, exploiting the MB cross-track array typically constructed by multiple SAR passes for beamforming and steering along the vertical axis, estimating the 3D distribution of the backscattered power in volumetric scenarios. This is typically accomplished by spatial (baseline) spectral analysis [1,2], each scattering component at a given height originating a corresponding spatial frequency component in the MB data vector.

In particular, concerning the incoming ESA mission BIOMASS for forest monitoring, the well know, tested and widespread superresolution Capon method [2,3] is foreseen to be exploited, which in a simple adaptive and light burden manner is able to get a height resolution beyond the overall baseline-related Rayleigh limit, and reduce layer cross-talk i.e. height sidelobes especially for typically non perfectly uniform baselines. This, in parallel to Fourier TomoSAR, that offers limited layer resolution capability and sensible cross-talks. Unfortunately, beyond the attention given to long-term temporal decorrelation [4] affecting all the TomoSAR methods for forest applications that are not based on companion satellite concepts, it is well known that Capon TomoSAR is affected by radiometric issues, presenting in the practical applications, with limited number of looks and residual data miscalibration, power losses in the height-resolved backscatterers, resulting in a non-linear behaviour.

It is thus the goal of this work to tune and test an alternative adaptive method [5] for TomoSAR imaging, offering height superresolution and sidelobe cleaning with improved radiometric capabilities. Both simulated analyses will be developed of the 3D imaging quality and radiometric fidelity, and first low frequency real forest data tests carried out.

Insights in the Capon radiometric issues are first given. In particular, a source of the power losses resulting in the Capon non-linearities is the self-cancellation phenomenon intrinsic in the Capon concept. To get the height superresolution and cross-talk reduction, the Capon algorithm relies on the knowledge of the MB array response (steering vector) and of the spatial (baseline) correlation matrix, to adaptively reject the interfering scattering coming from height directions different from that currently targeted during the height scan [2,3].

In this process, deviations of the actual steering vector from the nominal one, related to residual miscalibrations typically after atmospheric compensation, and imperfect correlation estimates, lead Capon to misinterpret the data component from the targeted height as an unwanted interference to be reduced, so tending to cancel also the signal of interest, resulting in non-linear radiometric sensitivity. This can be only partially controlled by the well-known diagonal loading method, that tends to brake the critical adaptive interference rejection.

The method presented here to cure these issues of Capon TomoSAR is based on a specific preconditioning of the MB data before adaptive spectral estimation [5]. In particular, a pre-estimate of the current targeted component is (partially) compensated in the data to bypass the misinterpretation in the adaptive processing that triggers the self-cancellation. Two different partial compensation strategies are experimented in this work. The new TomoSAR method can be considered to be double adaptive, and advantageously trade-off superresolution, in particular the sub-Rayleigh resolution level, with the reduced radiometric losses i.e. improved linearity.

First simulated analyses are reported for a controlled characterization of the radiometric behaviour of the proposed method, with the Fourier beamforming and (loaded) Capon as comparison methods.

Typical realizations are shown of Tomo profiles for the new method and the reference algorithms. The MB array is composed by 6 almost uniformly spaced passes, which is a typical BIOMASS mission scenario, different residual phase miscalibration levels are applied, and the processed looks follows typical figures for the forest application. The backscattering scenario consists of two both equi and different power speckled sources, height-compact for an easier investigation, with typical forest SNR, and height separation slightly sub-Rayleigh. It is shown how the new method can offer a very good recovery of the expected peaks level, with amplitudes very close to the Fourier ones, overcoming the sensible Capon radiometric loss, still producing a well satisfactory superresolution and low sidelobes.

In the attempt to optimize the global tomographic performance trade-off, height accuracy is also analyzed and the partialization factor of the compensation step in the method tuned, finding an advantageous knee point. A more extended characterization of the Capon radiometric issues and of the new tuned method is also performed, producing sensitivity plots.

First real data trials are also performed of the new double adaptive method, for a line of low frequency airborne MB SAR data, taken over a forest. The proposed advance can be useful in the context of both the BIOMASS and the SAOCOM-CS programs.

[1] A. Reigber, A. Moreira, “First demonstration of airborne SAR tomography using multibaseline L-band data,” IEEE TGRS, 38(5), pp.2142–2152, 2000.

[2] F. Gini, F. Lombardini, and M. Montanari, “Layover solution in multibaseline SAR interferometry,” IEEE TAES, 38(4), pp.1344–1356, 2002.

[3] F. Lombardini, J. Ender, L. Rößing, et al., “Experiments of interferometric layover solution with the three-antenna airborne AER-II SAR system,” Proc. IGARSS 2004.

[4] F. Lombardini, F. Cai, “Temporal decorrelation-robust SAR tomography,” IEEE TGRS, 52(9), pp.5412-5421, 2014.

[5] F. Lombardini, F. Viviani, “Radiometrically robust superresolution tomography: first analyses,” Proc. IGARSS 2016.

Synthetic Aperture Radar (SAR) provides precise range and range-difference measurements. These measurements suffer from speckle noise, when there are more than one dominant scatterer in a resolution cell. However, when focusing on dominant and stable point-like scatterers, often called permanent scatterers (PS), the measurement of the backscattered signal is not affected by speckling and allows for precise measurement of distance differences using the interferometric phase differences. This is the reason for the importance of stable point-scatterers in SAR remote sensing, which are the base for techniques like PS-InSAR, but also for absolute position measurements with SAR geodesy.
Such point-scatterers can be found in rather large numbers in urban areas. The spatial density of these PS points depends on the structure of the area of interest, but also on the used wavelength, with shorter wavelengths providing a denser PS network.
In many areas outside of cities, there are only few stable point-scatterers to be found. Artificial targets, like corner-reflectors, can be an alternative solution for areas without ‘natural’ available point-scatterers. However, as corner-reflectors are large and expensive, they cannot be widely used outside of secured test areas, because they are prone to misuse and theft.
We propose the use of small, inexpensive artificial targets that can be used in large numbers under such circumstances. We demonstrate the use of such targets with TSX. For C-band in Sentinel, larger targets are necessary. We will demonstrate the possibility to also use cheap targets in C-band for co-pol and cross-pol cases.

Mapping forest height makes a great contribution to quantitative estimation of forest above ground biomass, leading to a better knowledge of carbon stocks stored in forests. In recent years, polarimetric SAR interferometry (Pol-InSAR) and SAR tomography (TomoSAR) techniques have become major tools for forest height retrieval based on SAR measurements. In polarimetric SAR interferometry, forest height is retrieved from single baseline polarimetric data, under the assumption of the random volume over ground (RVoG) model. For SAR tomography, instead, fully 3-D back-scattering profiles are reconstructed by jointly focusing data from multiple flights and forest height is then obtained by analyzing the shape of the vertical profiles. In this work, we aim at comparing these two techniques in the context of P-Band SAR retrieval of forest parameters in tropical areas. To accomplish this goal, both techniques are applied to the same SAR dataset at P-band, which is the one acquired by ONERA in French Guiana during the TropiSAR campaign. PolInSAR and TomoSAR forest height maps are then analyzed using Lidar measurements.

Forest above ground biomass (AGB) retrieval by P-band SAR Tomography has largely been studied in recent years, mostly in the frame of studies related to the forthcoming spaceborne Mission BIOMASS. Using SAR Tomography, it has been demonstrated that the backscattered power at the canopy layer is strongly correlated to the forest AGB. In the context of spaceborne missions, however, it is difficult to achieve enough passes for SAR tomography. Interferometric ground notching has recently been proposed as a new method to single out volume scattering contributions. The method takes as input a single pair of SAR images to obtain a ground-notched image by canceling out the backscattered power coming from the ground level. Most interestingly, the correlation between ground-notched intensity and forest AGB has been demonstrated to be very significantly improved w.r.t. the case of single images. In this paper, we evaluate the impact of temporal decorrelation on interferometric ground notching. A model is presented to show the impact of temporal decorrelation, and an experimental assessment is provided by analyzing data from the P-Band campaign BIOSAR-1, where multiple baselines where acquired both on the same day and with a time span of 23, 30, and 53 days. The experimental results show that ground-notched intensity is more stable for tall-forested areas, whereas the low vegetation is more affected by temporal decorrelation. Current work is ongoing to extend the analysis to tropical sites.

The monitoring of line-infrastructure, such as railways and dams, benefits from the synergy of SAR interferometry (InSAR) using multiple satellite missions. Different orbital and instrument viewing geometries, as well as spatial and temporal coverage and resolution, optimize the amount of information that can be extracted from the data. However, InSAR is an opportunistic approach as the location and occurrence of its measurements (coherent scatterers) cannot be guaranteed, and the quality of the InSAR products is not uniform. Therefore, advanced integrated products and generic performance assessment metrics are necessary. Here, we propose several new monitoring products and quality metrics for a-priori and a-posteriori performance assessment using multisensor InSAR, based on the assumption that:

1) coherent scatterers can be found representing the same physical phenomenon,

2) alignment of the datasets in space and time is possible, and

3) the influence of (nonperiodic) longitudinal movements compared to transversal and normal motion is limited.

The methods and metrics address two main operational questions.

1) Can we measure a particular deformation in a specific direction, at a specific location, and how well can we measure that?

Sensitivity values and Sensitivity circles are introduced, leading to a deformation variance as a function of the infrastructure orientation and the orthogonal elevation angle. The particular observability yields Minimal Detectable Deformations (MDDs) that can be observed with a given confidence level.

2) What can a particular combination of sensors produce as deformation products, and how does this compare with another combination of sensors?

We state the method for Line-of-Sight decomposition specifically to an asset-based coordinate system, and provide the variance covariance matrix in this coordinate system. The Dilution of Precision (DoP) is introduced as a scalar-valued quality metric, which is convenient to compare different sensor combinations.

Once InSAR data have been processed into deformation estimates, we introduce two operationally relevant end-products. First, the Significance Deformation Map (SDM) shows all locations on the selected asset where deformation is significant, given a confidence level as agreed with the asset-manager. Second, the Longitudinal Anomaly Profiles (LAPs) are a convenient way for instant information on the occurrence, location, and significance of anomalies along the track.

The proposed methods and metrics are demonstrated on a 125 km railway line-infrastructure asset in the Netherlands. All work contributes to a more structured, repeatable, and generic approach in the operational monitoring of line infrastructure.

This work has been recently published by the IEEE Journal of Selected Topics for Applied Remote Sensing, entitled ‘Monitoring line-infrastructure with multi-sensor SAR interferometry: products and performance assessment metrics’, doi: 10.1109/JSTARS.2018.2803074

Elevated highways, as one of the most important infrastructures, make contributions to a convenient and efficient public traffic, whose operational safety is the foundation of city development. Thus, deformation monitoring is the necessary prerequisite to normal operation of elevated roads. Persistent Scatterer SAR Interferometry (PSI) is a mature tool for land subsidence monitoring in an urban area, and its reliability has been verified by many studies.

In this research, we processed a long time-series of high-resolution TerraSAR-X satellite dataset in Shanghai from 2013 to 2017 to explore the spatio-temporal patterns along the elevated highways. Then with ground leveling data for InSAR accuracy verification, we compared and analyzed results between InSAR and leveling. The spatial distribution and temporal evolution of deformation characteristics of elevated highways were explored with joint analysis of PSI results, regional land subsidence, dynamic loads and the historical construction activities.

According to our results, regional land subsidence is a major factor for the deformation of Elevated highways because foundation of elevated highway is made up of end bearing pile foundation, which can generate frictional resistance between pile body and soil layer. The second factor we may consider is vehicle dynamic loads during the operational stage. We try to build a model between deformation and dynamic loads based on big data. And another one is the time when the elevated roads built. As is known, the land subsidence may undergo similar evolution after the engineering completion. So there are some relations between the completion time and deformation. Moreover, other factors such as groundwater, surrounding projects, etc. may play some but very small roles and we ignore them here.

Synthetic Aperture Radar (SAR) image co-registration is a fundamental but crucial procedure for interferometric SAR applications. Mainstream SAR coregistration algorithms are based on either cross-correlation or geometrical mapping. Both algorithms suffer from high computational expenses. The cross-correlation based coregistration, which is widely applied to conventional stripmap SAR data, requires many sub-image patches and an oversampling operation to derive the robust offsets with one tenth pixel accuracy. While the geometrical co-registration, which is widely applied in S-1 Interferometric Wide Swath (IW) images, is quite time consuming due to wide imaging coverage of TOPS mode and the iterative range-Doppler method. The massive parallelism of Graphic Processing Units (GPUs) can be used to improve the calculation efficiency of the two algorithms. The two new parallel algorithms are developed in NVIDIA’s Compute Unified Device Architecture (CUDA). The parallel cross-correlation based algorithm is implemented on batched processing for small matrices operation. The parallel geometrical processing is optimized by parallel pipelines. The efficiency improvement of the two parallel algorithms can be observed via the contrast experiments on Envisat stripmap and Sentinel-1 IW data.

Sentinel-1 Capability of Surface Deformation Estimation over a Wide Area in North-Eastern Algeria

Abstract

Monitoring ground deformation over a wide area with classical geodetic techniques, such as Geodetic Levelling and Vertical inclinometer, is very time consuming and expensive. On the other hand, interferometric synthetic aperture radar (InSAR) has been successfully used over the last two decades to produce high spatial density displacement maps in centimeter/millimeter accuracy with relatively low cost. InSAR gives the opportunity to study various phenomena, like fault creep, landslides, and subsidence induced by groundwater extraction.
This work is focusing on detecting land deformation and study geohazards such as landslide and subsidence over large areas in northeast Algeria by using Sentinel-1 data. Sentinel-1 data is provided by the Copernicus Program satellites constellation conducted by the ESA. Sentinel-1data is acquired in TOPS (Terrain Observation by Progressive Scans) mode, which is mainly designed for the purpose of wide coverage capability. We used all available data acquired over the study area between 2015 and 2018.
Stacks containing hundreds of SLC images covering the study area in ascending and descending orbits were analysed for land surface deformation using SarProZ and applying Persistent Scatterer Interferometry (PSI) technique.
This study main objectives are: first, increasing the detection capability of active landslides and ground subsidence; second, monitoring and analyzing the temporal evolution of the detected deformations by providing a time series deformation maps and velocities maps over a period of time. In the end, to understand deformations distribution and investigate geomorphological and geological causes.
Results revealed the characteristics of the study area where large areas of sparse vegetation and rocky surfaces, especially at high altitude and around the main mountainous structures, have a low density of measured PS points. On the other hand, areas where the vegetation cover decreases, especially in the urban areas, the density of Persistent Scatterers increases. Obtained deformation maps in these areas are composed of dense PS points. Where, the biggest amount of PSs is found in the urban area, the rest in the monotonous rocky areas.
A deformation time series was calculated for each Persistent Scatterer. Each point associated with the value of the annual linear velocity (mm/yr), estimated over the analyzed period and the displacement accumulated at each sensor acquisition date (mm). The measures are referred to the movement of the ground point in the satellite Line of Sight (LOS) direction.

Recently, the Sentinel-1 data has received extensive attention due to its large coverage and free availability. It is often used for monitoring of large-scale volcanoes and earthquakes, but it also offers an opportunity for subsidence monitoring in built-up areas although it is conventionally not included as high-resolution imagery.

The study area in Wuhan is along the Yangtze River, and has developed rapidly in recent years. Rapid urbanization and extensive carbonate rock strips as well as soft soil layers underground in Wuhan have contributed to land subsidence in most parts of Wuhan. The risks to the safety of buildings and public infrastructures are concerned by municipal departments and citizens. It is crucial to detect land subsidence to facilitate understanding of the evolutionary processes so that proper measures can be taken to carry out effective planning and construction and to mitigate further loss.

The time-series analysis method is adapted to derive the deformation from Sentinel-1 images. In this case, StaMPS is introduced for data processing, which does not need to assume a special deformation model, directly through the three-dimensional phase unwrapping algorithm to obtain surface deformation information ^[1]. It is used to extract deformation on constructions combining with high-resolution imagery ^[2]. Whereas application of Sentinel-1 data in monitoring deformation in built-up areas is relatively few.

In this experiment, totally 44 scenes of the IW mode data of the Sentinel-1A are collected. Using the free Sentinel-1 data, time-series PS-InSAR technology ^[3] is applied to obtain the deformation rate of Wuhan City and there are several areas with severe subsidence, with a maximum subsidence rate of -27mm/y. Furthermore, the InSAR-derived displacement map also highlights active motions of built-up areas and infrastructures, such as some communities and railways segments. It helps with safety screening and risks assessment.

An overall subsidence of -15mm/y to -27mm/y occurs in Anjuyuan Community, with a cumulative deformation of 30mm. From the obtained time-series curve of the PS points, the subsidence rate of the community accelerated from August 2016 to November 2016, during which the subsidence exceeded 10mm, which is equivalent to the subsidence in the previous 16 months. It should be paid more attention to.

Remarkable subsidence occurs in the Fazhanercun and Jingnan Community in Hankou district. The maximum subsidence rate is -27mm/y, and the maximum deformation is even -38mm. According to the survey, since the start of reconstruction project of the Village in the south of the community, the walls of some households in this community were cracked, and the front steps were severely separated from the ground, which is consistent with the experimental results, indicating that the processing results of Sentinel-1A dataset is reliable.

An analysis of a section of a railway passing through Hankou Railway Station shows that the majority railway of this section suffers from different degrees of subsidence, and the deformation rate in the vertical direction is between -11.64mm/y and 6.18mm/y. Besides, the differential subsidence in the railway curve is relatively large, and it is worthy of attention. The deformation rate in the vertical direction of a section of Wuhan-Guangzhou Railway passing through Wuhan Railway Station is approximately -5mm/y to 5mm/y. Except for a slight uplift near the Wuhan Railway station, this section is overall sinking slightly and there is no uneven subsidence.

REFERENCES

[1]. Hooper, A.J., Persistent scatter radar interferometry for crustal deformation studies and modeling of volcanic deformation [D]. Stanford University, 2006.

[2]. Qin. X, Liao. M, Yang. M, and Zhang. L. Monitoring structure health of urban bridges with advanced multi-temporal InSAR analysis [J]. Annals of GIS. 2017, 23(6):1-10.

[3]. FERRETTI A, PRATI C, ROCCA F. Permanent scatterers in SAR interferometry [J]. IEEE Transactions on Geoscience & Remote Sensing, 2001, 39: 8-20

Land reclamation is a well-known solution for land augmentation on coastal areas to serve the population increase, rapid urbanization, and economic development. Many countries have expanded its coastal land by using this method including England, Korea, Germany(Flemming and Nyandwi 1994), Ireland, Netherlands, Spain, Bangladesh, Nigeria, and China. Remote sensing has represented a high capability in detecting environmental changes caused by land reclamation in coastal areas due to its wide coverage, periodical revisit, and different data types and techniques. In this study, we have utilized microwave remote sensing data of Sentinel-1 to estimate the stability of reclaimed areas’ surface in Shenzhen City, Guangdong Province and Hong Kong, Southern China.

This study area located in the southeastern part of the Pearl River Estuary (PRE). The Pearl River is the third largest river in China and the second biggest river measured by mean annual runoff. The river’s water flows through eight different outlets into the South China Sea. Four of which are located in the north and west of the PRE. The coastal area of this area is under an intense pressure of land reclamation and almost can be considered as an artificial environment due to the recent economic development. This area has witnessed an enormous urban encroachment on agricultural land and coastal swamps during the last four decades due to the open-door policies in China since 1978. This urban expansion was a result of high population density and rapid industrialization in the Pearl River Delta since then.

Reclaimed areas in the study area were defined based on the land expansion on the estuary’s water in Shenzhen and Hong Kong coastal areas since 1973. The first available Landsat Multispectral Scanner System MMS images for the PRE which had been acquired in December 1973, were used to map the PRE water surface body in 1973 as the study’s area of interest (AOI). Then, the most recent images of Sentinel-2 were used to define the recent coastal line in our study area. Normalized Difference Water Index (NDWI) was applied to all images to extract the water body of our study area. This method based on the high reflectance of water in the green light wavelength and its low reflectance in the near-infrared wavelength.

After extracting reclaimed areas in the period between 1973 and 2018; Sentinel-1 data was used to detect surface deformation on these areas by applying PS-InSAR technique using SarProZ software. 65 Ascending images between June 2018 and January 2018 were utilized for this purpose.

Results revealed more PS points on the older impervious surfaces land cover of reclaimed areas compared to recent built-up and cultivated areas. The overall results showed stable surface in most reclaimed areas. However, some reclaimed areas represented surface deformation reached in some places to -30 mm/year such as areas southern Bao’an airport in Shenzhen City and some wave breakers in Hong Kong.

InSAR has proved a powerful technique for mapping surface deformation, developing rapidly in recent twenty years. But one-dimensional InSAR LOS measurement has limited its application to retrieve 3-D surface displacements, as it is only sensitive to surface movements towards or away from the satellite. The most straightforward approach is to integrate InSAR LOS measurements with homogenous data (Offset tracking, MAI) or heterogeneous data (GPS data, leveling). In this paper, we reconstruct the three-dimensional deformation field with surface-parallel flow assumption based on the knowledge of DEM information on ground deformation. In addition to, due to the different influence of the errors in different observation data, the iteration method by correcting characteristic value with maximum likelihood estimation is used to literately process the function model to get the accurate random model through prior information, and also the exact weight function. We apply this method on the Jiaju landslide and the results shows, horizontal displacement of Jiaju landslide appears to move along the landslide direction in the east-west direction, vertical deformation rate of the north part is large which exceeding -2cm/y, while the south part is -0.5cm/y.