^{th}, 1999 Athens Earthquake Using Terrestrial Surveying. Improvement of the Assessed Deformation Field by Interferometric Stacking

^{1}

^{2}

^{*}

^{1}

^{1}

^{2}

^{2}

This article is an open-access article distributed under the terms and conditions of the Creative Commons Attribution license (http://creativecommons.org/licenses/by/3.0/).

The primary objective of this paper is the evaluation of the InSAR derived displacement field caused by the 07/09/1999 Athens earthquake, using as reference an external data source provided by terrestrial surveying along the Mornos river open aqueduct. To accomplish this, a processing chain to render comparable the leveling measurements and the interferometric derived measurements has been developed. The distinct steps proposed include a solution for reducing the orbital and atmospheric interferometric fringes and an innovative method to compute the actual InSAR estimated vertical ground subsidence, for direct comparison with the leveling data. Results indicate that the modeled deformation derived from a series of stacked interferograms, falls entirely within the confidence interval assessed for the terrestrial surveying data.

One of the most significant natural disasters to strike Greece in the 20^{th} century was the September 7, 1999, 11^{h} 56^{m} 50^{s} UTC, Mw (moment magnitude) = 5.9 Athens earthquake. It claimed the lives of 143 people, and caused the collapse of several buildings, mainly in the northwest suburbs of the Greek capital. The approximate location of the earthquake epicenter was 38.10°N, 23.56°E, roughly 20 km northwest from the center of Athens [

The vertical displacement field at the surface level caused by this tectonic event was investigated with space born Synthetic Aperture Radar Interferometry (InSAR), using ERS-2 data. InSAR processing showed a significant deformation with the maximum Line Of Sight (LOS) subsidence being of approximately 6 cm [

The region of maximum deformation coincided with the main shock epicenter. This area was very close to the Mornos river open aqueduct, used for water supply to Athens. The distance of the aqueduct pass from the earthquake epicenter was less than 2.5 km. The water supply authority in Athens awarded an aqueduct-leveling project to the National Technical University of Athens/Higher Geodesy department (NTUA/HG), which lasted for two months, from March to April 2001. Prior leveling data along the Mornos aqueduct had been obtained in 1984. No height data were available for the intermediate time interval 1984-2001; however no major seismic event had taken place in that period. The two co-seismic sets of leveling data were considered adequate to investigate the vertical displacement in the affected by the earthquake area and verify the InSAR derived observations.

The scope of this paper is the evaluation of the InSAR derived displacement field caused by the Sept 7, 1999 Athens earthquake, using as reference an external data source provided by terrestrial surveying along the Mornos river open aqueduct. Research works relating to InSAR – leveling interoperability issues have been published in the past, focusing on either verifying the InSAR derived subsidence, or integrating them with the leveling data to increase the reliability of the measurement. In [

This paper is structured as follows: section 2 refers to the preliminary processing of the input data, namely the InSAR and leveling measurements. Section 3 presents in an analytic way the distinct steps in rendering the two data sets compatible. Section 4 outlines the results obtained by applying the proposed processing chain, whilst section 5 investigates more thoroughly the physical meaning of these results and the applicability of the method in verifying InSAR derived subsidence on the basis of terrestrial surveying data.

ERS-1/2 sensor images spanning the period from December 1997 to January 2001 were acquired and processed over the Athens Greater Area. The satellite images were provided by the European Space Agency in the frame of the ESA-GREECE AO project 1489OD/11-2003/72.

Interferometric calculations were done by using the CNES DIAPASON InSAR processing software, and the sixteen most coherent co-seismic interferograms were kept for the purposes of the study. The image pairs used along with their corresponding “altitude of ambiguities” are shown in

The first terrestrial surveying work on the aqueduct was done in 1984, covering its whole length of approximately 200 km. A special trigonometric height technique was used, providing the same level of accuracy as conventional leveling but being significantly faster [

Moreover, a standard geometric leveling was realized in 2001. The total distance surveyed was 40 km, of which 12 km were confined in the area of interest illustrated in

The accuracy of the leveling works was estimated to be of the order of a few millimeters between successive height references [

The differential displacement data derived by the two different techniques were incompatible and consequently a direct comparison was not possible. These incompatibilities may be summarized as follows:

InSAR processing provided wrapped interferograms, consequently only the fractional part mod_{2π}Φ of the full phase difference Φ was known.

InSAR results correspond to the projection of the true vertical deformation along the LOS vector.

The reference systems of the leveling data and the InSAR data were different. InSAR data were referring to ED 50 UTM zone 34 while the leveling data were referring to the mean sea level and the height reference positions to the Hellenic Geodetic Reference System 87 (HGRS 87).

The interferograms were “noisy” mainly due to temporal decorrelation, orbital and tropospheric disturbances.

The following sections describe the procedure used to eliminate the effects of the above types of incompatibility, rendering the two datasets comparable.

The wrapped interferogram underwent a simple filtering procedure. The primary objective of this action was to minimize the probability of phase unwrapping failure, while a secondary goal was the improvement of the wrapped and unwrapped interferogram appearance in order to derive qualitative evaluations more efficiently. The filter used was a simple 2D 3×3 space mean filter (symmetric to match the rectangular pixel dimensions), applied on both the real cos(ψ_{i,j}) and imaginary sin(ψ_{i,j}) parts of a virtual unitary magnitude signal e^{jψi,j} = cos(ψ_{i,j}) + j sin(ψ_{i,j}) . The phase of this signal is the unfiltered interferometric phase ψ_{i,j} In other words, the 2D space filter was applied on a unitary signal to which the phase of the input interferogram was projected. The phase ψ_{flt/i,j}, comprising the filtered interferogram, was extracted through an arctan operation from the filtered real and imaginary parts of the virtual signal. The filtering procedure is best defined by the following formula:

Various 2D phase unwrapping techniques have been developed for resolving the “integer ambiguity” problem of the interferometric phases. In this study “Quality Guided Path Following”, “Least Squares Without Weights”, “Weighted Least Squares”, and “Minimum LP Norm” approaches were implemented [

The unwrapped co-seismic interferograms were all undergone a special processing in order to minimize the existing orbital, tropospheric and DEM disturbances. These errors were lifted by a “tilting” and “shifting” operation, using a number of coherent pixels located outside the deformed area. According to this approach [

The HGRS 87 unwrapped interferogram provides the differential vertical displacements for each target pixel as projected to the LOS vector _{LOS}(E,N) , and not the vertical differential displacements_{dU}(E,N) themselves, as is the case of leveling (

In order to determine the differential vertical displacements from LOS projection displacements, the value of the incidence angle for each target pixel was required. The incidence angle computation procedure was based on satellite trajectory data and the position of the target. Initially, for every target pixel (i, j) the zero - Doppler position of the space born SAR sensor had to be computed. This was achieved through signal processing applied on the “master” (or “reference”) image. Third degree polynomials were fitted with Least Squares to the known satellite position vectors

The map projection coordinates of the target were converted to geocentric Cartesian coordinates in the geodetic terrestrial reference frame in which the satellite orbits were provided (in this particular case from HGRS 87 map coordinates to ITRF 96 geocentric Cartesian coordinates).

The mean Doppler frequency shift was computed by the CNES DIAPASON software and was assumed to be the same for every single pixel target. The Doppler frequency shift f(i,j) was expressed as a function of the satellite position, the satellite velocity vectors and the target position

A total of seven equations were accumulated, and an equal number of unknowns was introduced, three for the satellite position vector, three for the satellite velocity vector and one for the time t_{i}. Hence, a non linear seven-equation system was created for the estimation of the seven unknowns. The system was linearised with Taylor series expansion and solved iteratively.

Knowing the satellite and target position vectors, the unitary LOS vector could be calculated simply from the following vector equation:

The target position ellipsoidal coordinates φ_{i,j}, λ _{i,j} were then calculated on the same geodetic terrestrial frame, which was used to express the orbits and the target coordinates in the previous step.

Knowing the target's latitude and longitude φ_{i,j}, λ _{i,j} the LOS vector components were transformed to the local geodetic reference system (delta north - DN, delta east - DE, delta up -DU) by means of a rotation matrix:

The third component of the

In the framework of this study and due to the fact that reliable verification data were available through the leveling survey, it was possible to evaluate the advantage in using a mean stacked interferogram instead of using only one, that is the “highest-quality” (most coherent) interferogram. For this purpose the sixteen “tilted” and “shifted” unwrapped interferograms were stacked to derive a mean temporal deformation field. This technique produced an image S(i, j) defined as: S(i, j) = mean(I_{1}(i, j), I_{2} (i, j),… I_{n} (i, j)), where n represents the number of the available interferograms and I_{m}(i, j) the unwrapped interferometric phase of the m^{th} interferogram at pixel location (i,j). Consequently the produced interferogram depicting the mean deformation field, was released from high and intermediate frequencies [

It should be mentioned that at this stage alternative stacking methods were implemented as well. They comprised of the formation of A) a weighted mean stacked interferogram, using as weights the pixel coherence values of each contributing interferogram, B) a maximum coherence stacked product, on which each phase pixel value stems from the interferogram with the highest corresponding coherence pixel value and C) a windowed maximum coherence stacked product; here each phase pixel value stems from the interferogram with the highest mean coherence value, calculated inside a 3 by 3 pixels window, centered on the pixel of interest. As is shown in section 4, the above methods returned very similar results compared to the mean stacked approach.

As mentioned the unwrapped interferometric calculations were referring to a UTM map projection on the ED 50 Greek Datum. In contrast the coordinates of the height references were expressed in the HGRS 87 reference system, using the Transverse Mercator map projection on the GRS 80 ellipsoid. To overcome this incompatibility the initial interferograms were converted to HGRS 87 projection system as follows:

The ED 50 UTM map coordinates (Eastings and Northings - E, N) were converted to ED 50 ellipsoidal coordinates (latitude and longitude -φ, λ), assigning to each pixel the corresponding orthometric height (H) derived from the input DEM.

The orthometric heights were converted to geometric ones (h), by implementing a constant additive geoid undulation value (N) for the entire area of interest, since the geoid in this area is relatively “flat” exhibiting a very low gradient. This value was obtained by the Ohio State University OSU 91 Geoid Model, and was recomputed for ED 50.

The ED 50 ellipsoidal coordinates were converted to ED 50 Cartesian coordinates (X, Y, Z).

Subsequently, the ED 50 geocentric Cartesian coordinates were converted to HGRS 87 geocentric ones, assuming only a parallel shift between the two systems. The latter assumption was expected to successfully provide the conversion due to the small size of the area of interest.

Then, the HGRS 87 geocentric Cartesian coordinates were translated to HGRS 87 ellipsoidal (φ, λ) coordinates.

Ultimately, the HGRS 87 ellipsoidal (φ, λ) coordinates were converted to HGRS 87 Transverse Mercator projection coordinates (E, N).

Thorough examination of the unwrapped (stacked and/or “highest-quality”) interferograms, exhibited the presence of “local” phase anomalies in certain areas extending from one to several pixels. The phase values in these pixels deviated from the prevailing values in the surrounding region. These anomalies were survived the filtering procedure described in section 3.1. It is beyond the scope of this paper to explore the origin of such phase “residuals”, but it could be assumed that they stemmed from local temporal decorrelation. It was also observed that the areas affected by these anomalies, presented significantly low coherence values and therefore they should be excluded.

Because the tectonic deformations observed were characterized by low phase gradient and spatial continuity, it was decided to proceed with a phase smoothing operation, by fitting (with the Weighted Least Squares method) a 3D-mathematical surface to the unwrapped interferometric phases _{dU}(E, N). After a series of adjustments, a successful fit according to the chi-square (^{2}) test was achieved, using the value of 6 mm as a-priori standard deviation for the observations. By the application of the error propagation law (given the estimated model parameters and their a-posteriori standard deviation values), it was concluded that the 3D-mathematical surface would provide the vertical deformation estimate for each target pixel (E, N), with an estimated a-priori deviation not higher than 0.2 mm. In order to ensure that the mathematical model represents the best fit to the displacement pattern observed, the most general form of m^{th} degree surface was tested:

After several runs, it was determined that a polynomial surface with degree higher than third would be redundant, as it was not offering any further improvement in terms of a-posteriori variance and measurement residuals. All higher degree coefficients were close to zero. The produced surface is presented in

Based on the 3D surface model produced, it became possible to extract a profile section of the InSAR vertical differential displacements along the leveling traverse. For this an origin had to be defined, and this was decided to be the height reference HR 65. Consequently, its displacement was set to zero. All other vertical displacements were provided in relevance to HR65. Profile data for InSAR and leveling data are presented in

Examining the profiles illustrated in

This research focused on rendering compatible and comparable the InSAR derived displacements, related to the September 7, 1999 Athens earthquake, with leveling survey data. Towards this goal a processing chain was implemented, encompassing an algorithm for orbit and atmospheric disturbances removal, and a methodology for transforming the LOS deformation vector to the true vertical deformation vector. The proposed method used a mean stacked interferogram to get a more consistent representation of the displacement pattern. Finally, an agreement between the deformation values originating from InSAR data with the ones derived from leveling survey data was demonstrated. Only minor discrepancies were identified between the two.

These small differences may be attributed to several types of error sources, such as 1) SAR sensor noise, radiometric instabilities and system aging, 2) surface subsidence model deviations, 3) remaining orbital phase “ramps”, 4) remaining tropospheric artifacts, 5) unwrapping errors, 6) temporal decorrelation effects, and 7) DEM errors. The possibility for a-seismic deformations in the period 1984-1998 could be also considered as a possible contributor to the relative subsidence profile differences. However, this a-seismic tectonic deformation, if it exists, remains unaccounted for, due to the absence of InSAR calculations in that period. The above-mentioned factors, may contribute to the observed total error of the derived relative subsidence values. However, the combined influence of the first three factors is considered to be essentially ignorable, taking into account the orders of magnitude of the resulting relative vertical displacement differences. Moreover, unwrapping errors computed by rewrapping the unwrapped interferogram, were assessed to be to an acceptable level in the area of interest. Therefore factors 6 and 7 namely temporal decorrelation and DEM errors, seem to be the most crucial parameters resulting in InSAR subsidence profile deviations. Temporal decorrelation could not be computed but must be considered as the major contributor to the spatially uncorrelated component of the residuals arising from the Least Squares approximation of the polynomial surface.

However as shown in

As far as the terrestrial surveying derived relative subsidence profiles are concerned, the estimation accuracy was much simpler and more explicit. The leveling data accuracy was estimated to lie in the range from 4 mm to 8 mm, in relative heights between successive height benchmarks. With the above estimations it becomes clear that the deviation of the two relative subsidence profiles (cases (a) and (c) in

We are grateful to the European Space Agency (ESA) for providing us the ERS1/2 data, in the framework of the ESA-GREECE AO project 1489OD/11-2003/72.

_{w}=6.0, 7 September 1999 Athens earthquake

(a) Plots of the Mornos aqueduct (blue) and height network (red) projected on 1:50000-scale map and (b) onto an ERS-2 SAR image interferogram.

Set of interferometric pairs used in the study. The vertical dashed line indicates the date of the earthquake occurrence.

Leveling path legs plot (red) and aqueduct plot (blue) projected onto a wrapped interferogram. For clarity purposes, only the segments connecting the height references are displayed. The actual leveling path follows the channel.

Wrapped interferogram, before (left) and after (right) filtering.

(a) & (b) Wrapped and unwrapped versions of the same interferogram. Note the unrealistic fringe pattern due to inaccuracies in the orbital data used. (c) & (d) The effect of the “tilting” and “shifting” operation on the same interferogram; the orbital fringes are removed.

Relative geometry of the true vertical deformation and the deformation provided by InSAR.

Spectral density of the stacked interferogram. Low frequencies prevail.

Differential vertical displacement model using a third degree mathematical surface.

Differential vertical deformation profiles derived by the, (a) conventional terrestrial surveying, (b) single “highest quality” interferogram, (c) mean stacked interferogram, (d) windowed maximum coherence interferogram. HR65 indicates the starting point of leveling.

Average and standard deviation of the vertical displacement differences between the leveling data and the InSAR methods.

−0,0096 | −0,0016 | −0,0030 | 0,0047 | 0,0020 | |

0,0056 | 0,0048 | 0,0055 | 0,0150 | 0,0056 |