GEOLOGIJA 50/1, 77–96, Ljubljana 2007 PSInSAR and DInSAR methodology comparison and their applicability in the field of surface deformations - A case of NW Slovenia Primerjava uporabe metodologije PSInSAR in DInSAR za opazovanje premikov povr{ja - primer SZ dela Slovenije Krištof OŠTIR1 & Marko KOMAC2 'Znanstvenoraziskovalni center SAZU, Novi trg 2, 1000 Ljubljana, Slovenia; kristof@zrc-sazu.si 2Geološki zavod Slovenije, Dimičeva ul. 14, 1000 Ljubljana, Slovenia; marko.komac@geo-zs.si Key words: permanent scatterers, surface deformation, differential interferometry, InSAR, Slovenia Ključne besede: permanentni sipalci, deformacije površja, diferencialna interferometri-ja, InSAR, Posočje, Slovenija Abstract In the recent years radar interferometry (InSAR) has become an important tool in various studies. It can be used to produce accurate digital elevation models and observe small surface displacements. Differential interferometry (DInSAR) can detect movements in the radar look direction that are in the order of wavelength used, i.e. less than one centimetre with ERS data. In the presented study DInSAR has been used to observe surface movements in western Slovenia. Three ERS radar images have been supplemented with an external digital elevation model to produce three differential interferograms that temporally covered the Posočje earthquake, which happened on April 12, 1998. For the area around Bovec a land subsidence of approximately 0.5 cm has been observed; the largest movements detected exceeded 2 cm. DInSAR has been compared to the permanent scatterers interferometry (PSInSAR). Both methods are complementary and both have individual advantages and disadvantages. Izvle~ek Radarska interferometrija (InSAR) je razmeroma nova tehnika, ki se je v zadnjih nekaj letih uveljavila v najrazličnejših študijah. Najpomembnejše področje njene uporabe je izdelava digitalnih modelov višin in opazovanje majhnih premikov. Zelo uporabna je tudi diferencialna interferometrija (DInSAR), s katero lahko opazujemo premike tal velikostnega reda valovne dolžine uporabljenega radarskega valovanja, kar znaša pri satelitih ERS približno pol centimetra. V predstavljeni študiji so bili s tremi podobami območja Posočja in uporabo zunanjega modela višin ustvarjeni trije diferencialni interferogrami. Z upoštevanjem dejstva, da so modeli, dobljeni iz različnih interferogramov, odvisni, so bili določeni premiki, nastali ob potresu v zgornjem Posočju, 12. aprila 1998. Interferometrija je pokazala, da se je okolica Bovca v povprečju posedla za 0,5 cm, največji opaženi premiki pa znašajo več kot 2 cm. Opravljena je bila tudi podrobna analiza potenciala metode DInSAR in primerjava z metodo permanenthih sipalcev (PSInSAR). Metodi sta komplementarni, vsaka pa ima svoje prednosti in pomanjkljivosti. DInSAR namreč daje ploskovne rezultate, PSInSAR pa točkovne, vendar omogoča daljše časovno opazovanje, kar je še posebej pomembno v območjih pokritih z vegetacijo, kjer dekorelacija onemogoča uporabo tehnike DInSAR. PSInSAR predstavlja odlično alternativo tudi klasičnim geodetskih tehnikam, saj jih v mnogočem prekaša. Glavna prednost pred slednjimi je velika gostota merskih točk, dolgo časovno opazovanje ter možnost opazovanja brez predhodne namestitve instrumentov. V študiji zahodnega dela Slovenije je bilo mogoče opazovati več kot 20 točk na kvadratni kilometer v obdobju skoraj deset let z natančnostjo desetnike milimetra. Pokazalo se je, da so premiki, določeni z metodama DInSAR in PSInSAR, enakega velikostnega reda, a PSInSAR omogoča njihovo opazovanje skozi daljše časovno obdobje. Raziskava, predstavljena v tem prispevku, je pokazala, da je interferometrija permanentnih sipalcev zelo uporabna metoda, saj predstavlja nadgradnjo “klasične” DInSAR metode in se obnese bolje od nje povsod, razen na gosto naseljenih območjih, kjer je stopnja korelacije visoka tudi skozi daljše obdobje. Največje omejitve PSInSAR tehnike so zapletena interpretacija, nezveznost podatkov (DInSAR lahko služi kot dodatna informacija), neuporabnost metode za opazovanje poraščenih območij, omejeno obdobje ponovitve vzorčevanja, ki je vezano na povratno dobo satelitskega snemanja in neuporabnost metode za opazovanje hitrih premikov oziroma deformacij. 78 Introduction Radar interferometry (InSAR) is a relatively new technique based on stereo pairs of aerial or satellite imagery high resolution images of the Earth's surface. Its primary fields of application are the production of digital elevation models and detection of minor displacements or deformations in vertical direction. The latter is especially useful in areas where the deformations are hard to measure with classic methods (i.e. geodetic measurements). The accuracy of the digital elevation models is about 10 meters in the horizontal (location) and several meters in the vertical (elevation) direction. A special InSAR method, the differential interferometry, is a very useful method that can be applied in ground deformation detection and measurements in the range of the radar wavelength. For ERS satellites the radar wavelength is 5.6 cm resulting in displacement accuracy of approximately half a centimetre (O{tir, 2000, 2006). The radar interferometry approach uses complex satellite radar images that are composed of the amplitude and of the phase of the backscattered signal. The phase is dependent upon the surface's characteristics and the travelling distance of the radar signal (between the emitting antenna, the surface, and the receiving antenna). The advantage of having two images of the same area, taken from slightly different orbits, can be, considering the viewing geometry, exploited for creating the link between the interferogram (the phase difference of the two images) and the surface elevations. This principle can only be used if both images meet the requirements of interferometric analyses, which are 1) they both have to be acquired from orbits that are close to each other, which means that they have similar image acquisition geometry, and 2) the phase reflectance or geophysical properties of the surface must not change substantially between the acquisitions, which means that the time between acquisitions of the image pairs must be short enough to guarantee minimal distortion of the image (O{tir, 2000). Similarly, this principle can be used in differential interferometry (DInSAR) for the detection of small relative displacements or deformations from the set of three images of the ground area with similar image acquisition geometry. With this method, two inter-ferograms can be calculated and with their Kri{tof O{tir & Marko Komac comparison the differential interferogram is produced. One interferogram is created from the first two images and the second interferogram from the last two images. If no changes occurred, the differential inter-ferogram is equal to zero. If the phase reflectance has changed or if the surface has undergone deformations the differential interferogram will not be equal to zero. The phase reflectance represents noise and its elimination from further analyses is necessary, while the second factor enables accurate displacement detections (O{tir, 2000; Hanssen, 2005). Permanent (also persistent) Scatterers InSAR (PSInSAR) technique is an upgrade of DInSAR. For analytical purposes this method uses coherent radar targets that can be clearly distinguished in all images and do not vary in their properties (Ferretti et al., 2001). Based on their permanent properties they are called permanent scatterers. By using permanent scaterers the atmospheric effects can be filtered out and the temporal and geometrical decorrelation can be eliminated. The drawback of this method is a loss of data continuity. The data are a set of points with a density depending on the form and coverage of the surface. These coherent radar targets are abundant in urban areas, but are very scarce in the vegetated and mountainous areas. The theoretical background of interfer-ometry has been known for more than two decades and over fifteen years ago the first successful interferometric analysis was conducted (Zebker in Goldstein, 1986). The real breakthrough in the field of interferom-etry came in 1991 with the launch of the first European satellite for Earth observation, the ERS-1. Since then this technique has been applied in many fields of terrestrial research, from hydrology (Borgeaud & Wegmüller, 1997; Goldstein et al., 1989; Rodriguez et al., 1996), seismology (Massonnet et al., 1993; Massonnet et al., 1994; Dixon, 1995; Peltzer et al., 1996; Massonnet et al., 1996; Peltzer et al., 1999; Carnec & De-lacourt, 2000), glaciology (Mohr & Mad-sen, 1997), ecology (Dixon, 1995; Ludwig et al., 2000), volcanology (Massonnet et al., 1995, Salvi et al., 2004), subsidence (Ferretti et al., 2000), and slow-landslide detection (Ferretti et al., 2001; Hilley et al., 2004). Despite the wide range of applications, the interferometry still hasn't reached its full operational stage, either due to the PSInSAR and DInSAR methodology comparison and their applicability in the field of surface ... 79 lack of data processing standards, due to the complex software or because of the difficulties in combining interferograms. Since the PSInSAR technique is a relatively young method, which is only gaining its recognition among a wider user domain, its application is still very limited. This paper will in short present the theoretical background of interferometry, the differential interferogram analyses and the permanent scatterer technique. The results of the DInSAR and PSInSAR analyses in the upper So~a valley in north-western Slovenia will be compared and evaluated. Study area and data used The study area was defined as an intersection of DInSAR and PSInSAR data acquisition range (Fig. 1). The north-western part of Slovenia was chosen as the study area due to its neotectonic activity (Poljak et al., 2000; Zupan~i~ et al., 2001; Grenerczy et al., 2005) and due to a number of landslide, rockfall and debris flow occurrences (Komac et al., 2005). Prior to the PSInSAR data acquisition in the NW part of Slovenia, the DInSAR analyses were conducted in the Bovec basin in order to analyse the influence of radar interferogram combination on digital elevation and movement accuracy (O{tir, 2000). For the DInSAR analyses of surface deformations, related to the earthquake of 12th April 1998 in the So~a valley area, three ERS-2 satellite images were used. Images were acquired before and after the earthquake. The digital elevation model of the area was calculated from two images, taken before the event in the so-called tandem acquisition where satellites ERS-1 and ERS-2 acquired images with a 24-h delay. One day difference in image acquisition enables good coherence between images. The description of images used for the DInSAR analyses is shown in Table 1. Fig. 1. The study area is located in the NW part of Slovenia. Slika 1. Obravnavano obmo~je se nahaja v SZ delu Slovenije. 80 Kri{tof O{tir & Marko Komac Table 1. Description of images used for the DInSAR analyses of the study area Satellite Orbit Path Frame Date Time Use ERS-2 15235 351 2673 20.3.1998 9:56 deformations ERS-2 15736 351 2673 24.4.1998 9:56 deformations ERS-2 16237 351 2673 29.5.1998 9:56 deformations ERS-1 24888 351 2673 18.4.1996 9:56 elevations ERS-2 5215 351 2673 19.4.1996 9:56 elevations As a part of Interreg III B project Climate Change, Impacts and Adaptation Strategies in the Alpine Space (ClimChAlp) permanent scatterers in the north-western part of Slovenia, between Tolmin in the south and Kranjska Gora in the north, were analysed. The area extends over 700 km2. The primary goal of the research was to determine the slope mass movements using PSInSAR technique and at the same time to determine the use of this method for other geologically based applications. For this purpose 57 images from the descending orbits of ERS-1 and ERS-2 satellites were used. Images were acquired in the period between April 21st 1992 and December 29th 2000. As the reference image, the image taken on September 26th 1997 was selected. Based on the preliminary data analyses and geological prospection for the reference point (stable or a “zero“ displacement point), the location near the town of Tolmin was chosen. The location of the reference point is 46°11´3.44´´N, 13°44´45.12´´E, the velocity of the point – 0.13 mm/year and the overall coherence 0.84. The average density of permanent scatterers is 23 per square kilometre, and the minimum density required for analysis is 15 per square kilometre. Average displacements in the line of sight were determined for the whole population of targets with a coherence higher than 0.5. Altogether 16304 permanent scatterers were used. For approximately ten most reliable percent of the population (1646 PS with a coherence higher than 0.74) the displacement data of all 57 acquisitions were calculated. For these targets, time series of displacements from 1992 to 1994 and again from 1995 to 2000 was derived. Methodology Radar interferometry Radar interferometry is a technique that has been successfully applied in different fields. The Earth’s topography can be ob- served with interferometry by using two approaches, with either one or two passes (overflights). In the first approach emission and reception antennas are placed on the same platform (airplane or satellite), while in the second approach, which is usually used in satellite acquisition, the same or identical platform overflies the same area with a time lag from slightly shifted orbits (O{tir, 2000, 2006). All equations are taken from O{tir (2000). In Fig. 2 the two radar antennas, located in points O1 and O2, simultaneously observe the surface. Vector B is called baseline has a distance B, which represents the distance between the radar antennas, is inclined at angle L in respect to the horizontal plane. The first antenna (O1) is located at elevation H above the selected reference plane (h = 0). The distance between the antenna O1 and the observed surface is defined by r, while the distance between the antenna O2 and the same point on the observed surface is defined by r + d. The phase of the backscat-tered wave is: 2tu Eq. 1 where ? represents phase, indexes t and r represent emission and reception. The inter-ferogram of the images with a common emitter represents only the reception part of the phase since the distance from the emitting antenna to the target is the same for both receptors. This is due to the same location of emitting source and hence the difference is equal to zero. In the described case one antenna emits and receives the radar signal and the other only receives the signal. Both antennas are placed on the same platform, plane, space shuttle or satellite. A different situation occurs when there are two emitters located on the same platform or when the same antenna system images the same area twice. In this case the interferogram represents the difference of two distances between the antenna and the target. PSInSAR and DInSAR methodology comparison and their applicability in the field of surface ... 81 Fig. 2. The geometry of radar interferometry. Slika 2. Geometrija radarske interferometije. The interferogram is derived through an exact coregistration of images and through a complex multiplication of pixel values from the first image with the conjugated pixel values of the second image. The phase difference is defined as: ¦*-A-f, 8 = 2np Eq. 2 Eq. 3 where p stands for the number of emitting antennas (p = 1 for one source and p = 2 for two sources/antennas). The phase that is shown on the interferogram is a modulus of absolute phase. The procedure of defining the absolute phase ? from the measured phase ?M is called phase unwrapping. The phase difference between both signals depends upon the geometry of observation and upon the height (h) of the target above the reference plane (h = 0). Considering the fact that observation geometry can be influenced or defined with adequate accuracy, the elevation of targets h(x,y) can be determined. The achieved accuracy can be in the range of several metres, with the precondition of phase uncertainty elimination. From Fig. 2 the following two rules can be concluded: h = H -rcosd, and Eq. 4 y = rsmO, Eq. 5 where 6 represents the incidence angle of the radar signal to the Earth's surface and hence the angle of observation. From Eq. 2, 3 and 4 the desired elevation of the target or the surface can be derived: Ky> ^) = H - y x ctg lg - aresin sin \2jcpB Eq. 6 Elevations derived from radar interferom-etry represent the average elevation of the pixel (or the basic resolution element) in the image. Usually the size of the pixel is ten or several tens of metres. With the ERS satellite data the resolution achieved can reach between 20 and 25 metres. Displacement observation using DInSAR What happens with interferograms if deformations of the surface occur between two acquisitions? Is it still possible to define the topography of the surface? The answer to these questions depends upon the dimen- 82 sion of the difference or deformation. If the relative position of scatterers within a pixel changes for more than the wavelength of used radar signal, the measurements can’t be conducted. In that case the phase correlation is lost and the comparison of images or production of topography is impossible. The only information available is that differences are greater than the wavelength, not even knowing in which direction they occurred. The differences of two elevation models which were calculated one from the image-pair acquired before and the other from the image-pair after the event (i.e. earthquake, subsidence…) can be used to determine surface deformation. The accuracy of such surface deformation model is identical to the accuracy of primary elevation models. In case of ERS satellites the accuracy would be several metres (O{tir, 2000; Hanssen & Ferretti, 2002; Hanssen, 2005). Deformations can also be observed when the displacements on the surface are coherent and spread over several neighbouring pixels. In this situation the user assumes that the scatterers’ locations within a pixel haven’t changed, while the whole area of pixel and its neighbours has shifted upwards, downwards or to the side. In this case the phase comparison of images can be conducted. The so-called differential phase contains the information on the change in the direction of the observation or in the line-of-sight (LOS). The displacement measurement accuracy that can be achieved with differential interferometry is in the range of less than the radar signal wavelength. Usually the dimensions vary from several millimetres to several centimetres. With systems ERS-1 and ERS-2 that have the wavelength of ? = 5.6 cm measurement accuracy of half a centimetre can be achieved (O{tir, 2000). The high accuracy is the consequence of observing the difference of interferograms and not the actual elevation models or their changes. This enables the highly accurate target motion detection. To achieve such high accuracy of motion detection, a good knowledge about topography and the position and direction of the antennas is necessary. Phase difference measurements based on two successive radar images enable the definition of only one component of the movement vector in space, in the sensor-target direction (LOS). Only one-dimension measurements are a substantial drawback of this Kri{tof O{tir & Marko Komac technique, while the big advantage is the possibility of spatial coverage of the observed area. Combining the radar data from ascending and descending orbits in analyses would enable the definition of two components of movement, which is usually sufficient for analyses. Motion measurements with radar inter-ferometry depend upon the nature of the motion. There are two basic conditions for satisfactory results (O{tir, 2000): • Changes during the acquisition of images must not be too big. This applies especially to their gradient, which should not be too big within a pixel. • Radar scattering within a pixel at the time of acquisition must be as equal as possible. More precisely, the position of emitters within the observed resolution cell should not change more than 20% of the wavelength of the used microwave radiation. The first condition is generally not a major problem. If large changes occurred during image acquisition – e.g. due to a volcano outburst or a destructive earthquake – the elevation model before and after the event is simply subtracted. In this way changes within the range of several metres can be detected. Of course the production of a precise elevation model before and after the observed event may present a limitation of the described method. The second condition is considerably more problematic. When it is not fulfilled, we speak about time decorrelation. Time decorrelation is one of major problems in the use of radar interferometry, because it renders difficult or even impossible comparison of two phases of radar images. Decor-relation – partial or complete – can be observed in images obtained at a few hours’ intervals on areas covered with vegetation and exposed to wind. On the other hand a good phase correlation can be achieved even among images taken several months or even years apart. The condition for a high correlation between images is the observation of the surface which is not covered with vegetation, e.g. desert or urban areas. In general bare areas are more adequate than vegetated, dry areas are better than wet and radars with a larger wavelength are more appropriate than those with smaller. The difficulties with decorrelation can be solved by permanent scatterers technique, which takes into PSInSAR and DInSAR methodology comparison and their applicability in the field of surface ... 83 account only those areas (points) which are coherent (i.e. phase stable). In order to understand differential inter-ferometry we can imagine two radar antennas observing the Earth’s surface at different time intervals (O{tir, 2000; Hanssen, 2005; Hanssen & Ferretti, 2002). This is the so called repeat pass interferometry. The phase of an individual pixel in the radar image is equal to the sum of the travelling part (contribution due to the double path between the satellite and the observed area) and the radiation part (due to the interaction between the radar wavemotion and the ground). In case that the surface properties have not changed between image acquisitions, the radiation part may be removed by subtracting the phases of two images. Only the part remains that is directly related to the geometry of observation. If the approximation of parallel signals is again taken into account, the following equation is obtained S = Bsm(š-9) = B. Eq. 7 Here BB is a component of the baseline in the radar signal movement direction. If it is assumed that there is another interferomet-ric pair where one of the images is equal to the image in the first interferogram. In this case r and 6 remain unchanged, thus enabling the comparison of phases of individual pixels. The other interferogram has a different baseline B' and its orientation L'. If the equations are combined, the result for an interferogram pair is 4;?
) B'
dh ~ 36 dh ~ Asin0 r'
Eq. 11 dfi _ An
Since the baseline B is much smaller than the slanting distance r, phase difference is much more sensitive to surface movements than to changes in elevation. Radar interferometry enables the measuring of absolute elevation within some metres' accuracy, while movements can be determined with the precision of one centimetre or even millimetre. Satellite ERS, orbiting at the elevation H = 770 km and facing the Earth's surface at an average angle 6 = 23° with the wavelength X = 5.6 cm, has at a baseline B' = 300 m and the signal-noise relation SNR = 10 dB and phase error a = 0.6, an elevation error ah = 3.3 m. Under the same circumstances the inaccuracy of movement measurements