Deformation Rate Estimation On Changing Landscapes Using Temporarily Coherent Point INSAR

In areas undergoing large scale constructional developments, such as cities in most of developing countries, there are abundant scatterers that are only partially coherent in some observation period. In fact these scatterers still carry high-quality phase information at least in a subset of interferograms allowing us to estimate the deformation rates from them. Here the partially coherent scatterers as well as the persistently coherent scatterers are termed as Temporarily Coherent Points (TCPs). In this paper we provide two approaches for TCP identification. One is based on offset deviations during image pair coregistration procedure and the other one is based on Ambiguity Mad Median Ratio (AMMR). Since the TCPs might keep coherent in a subset of interferograms, we propose two deformation parameter estimators that can either be performed only with coherent phases of TCPs or have the ability to suppress the effect of decorrelated phases and phase ambiguities that are considered as “outliers” when taking all interferograms as observations. We apply the techniques to map land surface deformation over Macau, China. The results from our TCPInSAR processing have been confirmed by ground measurements. 1. INTRUDUCTION Multi-temporal InSAR (MT-InSAR) [1][2] is a useful tool for remote sensing of ground deformation. As the basic observations of current MT-InSAR techniques, persistently coherent scatterers can usually be densely identified from radar images over well-developed urban areas where the townscapes have evolved into a stable stage. With the dense coherent scatterers the current MT-InSAR techniques can be successfully applied. However there are many urban areas, especially in developing countries, which are undergoing rapid constructional development. Urbanization makes it difficult to identify abundant persistently coherent scatterers and thereby hampers us from achieving a better risk assessment over these areas. Over these changing landscapes, many scatterers cannot maintain consistently coherence during the whole observation time span even though they still carry high-quality phase signals at least in a certain period to allow for an estimate of land surface deformation [3][4]. The coherent points on changing landscapes can basically be classified into two types. One type is the persistently coherent scatterers (e.g., PS) and the other type is partially coherent points. Both of these points hereafter are referred to as Temporally Coherent Points (TCPs). In this paper we present a novel MT-InSAR analysis technique termed as TCPInSAR to identify the TCPs and retrieve ground deformation rates from these points. Regarding the identification of TCPs, besides the method shown in [5] which is an image pair based method, we also propose an image (amplitude) based method. The method is similar to the amplitude dispersion index [1], but based on the Ambiguity Mad Median Ratio (AMMR) that has the ability to isolate partially coherent points. To estimate deformation from these TCPs, we propose two parameter estimators. One is modified from [6] by involving a coherence index to remove interferograms where the point pair is not coherent simultaneously. The other estimator is especially designed for the TCPs selected by AMMR. Since AMMR dose not identify interferograms where the selected TCPs are coherent, when taking all interferograms as the phase times series for a given arc, some decorrelated phases might exist. Therefore we need to design a robust estimator that can suppress the effects of low-quality phases and the potential phase ambiguities at arcs to retrieve the deformation reliably. Here we select L1 norm (also called least absolute deviations) estimator to meet our purpose, which has been used in [7] to improve the robustness of SBAS method. The basic observations of the proposed estimators are differential phases at arcs (point pairs) in multi-master interferograms with small spatial baselines, short temporal baselines, and small Doppler separations. One significant advantage of our estimators is that the deformation parameters can be estimated directly from the wrapped phases. In other words, there is no need of phase unwrapping. To evaluate the performance of the proposed TCPInSAR method, we choose the southern part of Macau as the test site. The area experienced rapid development from 2003 to 2010. Comparison with ground measurements has confirmed the validity of the results achieved through TCPInSAR method. _____________________________________________________ Proc. ‘Fringe 2011 Workshop’, Frascati, Italy, 19–23 September 2011 (ESA SP-697, January 2012) 2. TEMPORARILY COHERENT POINT INSAR 2.1. TCP identification The identification of TCPs is the first core step in TCPInSAR processing. Two methods i.e., image pair based method and image based method, are introduced respectively in this section. 2.1.1 Image pair based method The TCPs can primarily be identified based on the offset deviation in range and azimuth directions. The equation derived by Bamler and Eineder [8] indicates that standard errors of the estimated offsets from stronger scatterers is less sensitive to the window size and oversampling factor used in the image crosscorrelation compared with those from distributed scatterers. Therefore it is possible to distinguish the strong scatterers from distributed scatterers by offset statistics. The detailed analysis and test of the method can be found in [5]. Here we propose an improved processing strategy which can significantly accelerate the TCP selection. Using the master image, we first identify points that can keep almost the same backscattering intensity when processed with fractional azimuth and range bandwidth as the TCP candidates. Second, TCP candidates are further evaluated by changing the size of patches and the oversampling factor in image cross-correlation. For the sake of simplicity, a fixed oversampling factor can be used. We can then obtain an offset vector ( j OT ) for a given TCP candidate ( j ) which includes the offsets ( , 1, , ji ot i N   ) estimated from N windows with changing sizes as shown in Eq.1. Points whose standard errors of offsets are less than a threshold are selected as TCPs. The threshold can be set as 0.1 considering the fact that when the precision of calculated offsets reaches 0.1 pixels or better, the coregistration error would be negligible [9].