InSAR processing for the recognition of landslides

Adv. Geosci., 14, 189–194, 2008 www.adv-geosci.net/14/189/2008/ © Author(s) 2008. This work is licensed under a Creative Commons License. Advances in...
Author: Shonda Powell
0 downloads 1 Views 3MB Size
Adv. Geosci., 14, 189–194, 2008 www.adv-geosci.net/14/189/2008/ © Author(s) 2008. This work is licensed under a Creative Commons License.

Advances in Geosciences

InSAR processing for the recognition of landslides B. Riedel and A. Walther Technische Universit¨at Braunschweig, Institut f¨ur Geod¨asie und Photogrammetrie, Gaußstraße 22, 38106 Braunschweig, Germany Received: 26 July 2007 – Revised: 2 November 2007 – Accepted: 10 November 2007 – Published: 2 January 2008

Abstract. Synthetic Aperture Radar Interferometry (InSAR) is an established method for the detection and monitoring of earth surface processes. This approach has been most successful where the observed area fulfills specific requirements, such as sufficient backscattering, flat slope gradients or very slow changes of vegetation. We investigated the capability of two different InSAR techniques and achieved good results for the recognition of landslides in China and Greece that compared well with geodetic derived movement rates. This demonstrates the strong potential of SAR Interferometry for the detection of landslides and earth surface movements.

1

Introduction

Landslides are one of the most dangerous natural hazards in the world, causing high annual death tolls (Sidle and Ochiai, 2006). On average, landslides annually kill twice as many people as earthquakes and result in high annual damage costs. In the years 2003 to 2006 the European Union (EU) funded a multidisciplinary international project called OASYS (Integrated optimization of landslide alert systems). The scope of OASYS was to set up an integrated workflow for landslide hazard management. This system should lead the practitioner from data acquisition to suggestions of risk management measures. The emphasis of the project was the development of observation methods that allow: – detection of potential landslides on large scale – an efficient and continuous observation of critical areas – a knowledge-based derivation of real time information about actual risks in order to support an alert system (Kahmen et al., 2007).

For spatial optimization of this kind of an alert system a multi scale approach has to be applied, which helps to reduce the area that has to be observed (Niemeier and Riedel, 2006). This multi scale approach or observation concept starts with the processing of remote sensing data, because the early identification of high risk areas is the most important step. On the one hand remote sensed data allow scientists to examine past landslide evidence and on the other hand these data can be used in ongoing and future satellite missions to serve as a base for the detection and monitoring of earth surface processes. For the recognition of surface movements we investigated the capability of Synthetic Aperture Radar Interferometry for different test sites in Greece, Germany, Hungary, Romania and China. Examples from Greece and China are described herein. 2 2.1

SAR Interferometry Basic principles

The use of data acquired from remote sensing systems, especially from active sensor systems like radar, has become of more and more importance in recent years. Since the start of the European Remote Sensing (ERS) satellites ERS 1 in 1991 and ERS 2 in 1995 continuously recording of high quality Synthetic Aperture Radar (SAR) scenes has been accomplished and an extensive data archive is available today. SAR Interferometry gives us two main possibilities. On the one hand it is possible to generate Digital Elevation Models (DEM) and on the other hand there is the possibility to derive changes of the earth surface in the sense of subsidence (vertical displacements) or horizontal displacements, i.e. post seismic movements (Lu et al., 2007).

Correspondence to: B. Riedel ([email protected]) Published by Copernicus Publications on behalf of the European Geosciences Union.

190 2

B. Riedel and Walther: InSAR processing forthe the recognition oflandslides landslides B.RiedelInSAR and A.Walther: InSAR processing forofthe recognition of landslid B.Riedel and A. A.Walther: processing for recognition

function alert sys 3.1

Fig. 1. of master and and slaveslave scenescene with awith shorta Fig. 1. Typical TypicalSAR SARgeometry geometry of master baseline for thefor detection of earthofsurface in the 2-pass proshort baseline the detection earth changes surface changes in the 2cessing approach. A movement of the point P downhill results in a pass processing approach. A movement of the point P downhill distanceinchange to the secondtodata acquisition the satelliteofand results a distance change the second dataofacquisition the leads to and a phase the radar signal. The third scene with a satellite leadsshift to aofphase shift of the radar signal. The third longer baseline serves in relation to the master scene in the 3-pass scene with a longer baseline serves in relation to the master scene processing for the derivation of a height model in SAR geometry. in the 3-pass processing for the derivation of a height model in SAR geometry. InSAR processing for the recognition of landslides Riedel and A.Walther:

ith a short -pass proesults in a tellite and ne with a he 3-pass eometry.

R geome for the e 1. The 20 m in ween the ormation the absond varies

preted as heights or displacements. The typical SAR geometry of master and slave scene with a short baseline for the detection of earth surface changes is shown in figure 1. The ground resolution of the examined C-band data is 20 m in one direction. A prerequisite is the coherence between the two data acquisitions of the backscattered phase information of the ground pixel. The coherence is measured as the absolute value of the correlation between related pixels and varies from 0 to 1. This corresponds from low to high coherence (see examples in figures 4 and 5). The higher the coherence the better the final differential interferogram. The main radar frequency used by SAR satellites, (ERS 1/2, ENVISAT and RADARSAT), is in the C-band with a nominal frequency of 5.3 GHz corresponding to a wavelength of 5.6 cm. The other frequencies are in the L-band (1−2.6 GHz) used by the Japanese satellites JERS and ALOS and the X-band (8.2 − 12.4 GHz) for the new German Radar satellite TerraSAR-X. The higher the radar frequency the Fig. 2.the Flow chart of thedepth main processing schemes of InSAR. The lower in any type of material. AddiFig. 2. Flowpenetration chart of the main processing schemes of InSAR. The difference between both approaches lies in the use of an external tionally tobetween this circumstance therelies is in also dependence on difference both approaches theause of an external DEM (2-pass interferometry) or in the use of a third image (3-pass DEM (2-pass interferometry) or in the content. use of a third (3-pass surface roughness and its moisture The image C-band repprocessing) with a long baseline in relation to the master image for processing) with a longcompromise baseline in relation to the master image for resents a reasonable penetration depth DEM generation in SAR geometry. between The resulting differential inDEM in be SAR geometry. The differential ininterferogram thegeneration canopies of trees andand bushes andresulting backscattering prophas to filtered unwrapped for the derivation of terferogram has to be filtered and unwrapped for the derivation of erties ofdisplacements. the soil in relation to the L- and X-band (Henderson surface surface displacements. and Lewis, 1998; Nolan and Fatland, 2003). The repeat cycle between 2 or more scenes depends on the satellite misited and by the of coherence depends onand the24 vegetation sion is loss 35 days for ERSwhich and ENVISAT days for Generally two SAR data acquisitions, called scenes or imcycle and the growth rate. RADARSAT. The time interval for InSAR processing is limages, of the same area are required to generate interference The other limiting factor for SAR data processing is the length of the perpendicular baseline between the acquired images for data Gens (1998) described the spaAdv. Geosci., 14,processing. 189–194, 2008 tial distribution of baseline lengths for ERS. Here, the practical limit for InSAR processing is restricted to a perpendicular baseline length up to 600 m. For the generation of DEMs

Fig.3.3.The TheBaota Baotalandslide landslide) occurred 1982 bank of Fig. occurred in in 1982 on on thethe leftleft bank of the the Yangtze river (visible in the background). The city of Yunyang Yangtze riverchart (visible in main the background). The city Yunyang Fig. 2. Flow of the processing schemes ofof InSAR. Theis is located in the middle the Three Gorges Reservoir. located in between the middle of of the Three Gorges difference both approaches lies inReservoir. the use of an external DEM (2-pass interferometry) or in the use of a third image (3-pass processing) with a long baseline in relation to the master image for phase signal for each pixel and the coherence. From this real DEM generation in from SAR geometry. The resulting infringes resulting phase differences thatdifferential can be interinterferogram a synthetic interferogram, representing the toterferogram has to be filtered and unwrapped for the derivation of preted as heights or adisplacements. TheModel typical geompographic phase of Digital Elevation forSAR the area of surface displacements. etry of master and slave scene with a short baseline for the

investigation, has to be substracted. To get reasonable input detection earth surface itchanges in Fig. 1. and The values forof data processing is usefulistoshown use precise orbits ground resolution of the examined C-band data is 20 m in a DEM, as coherence the Space which Shuttledepends Radar Topographic Misited by thesuch loss of on the vegetation one direction. A prerequisite is the coherence between the sion and (SRTM) which rate. is available for 90 % of the earth surface cycle the growth two datagrid acquisitions ofofthe90for backscattered phase information with resolution m SAR (Farr data et al.,processing 2005). Multiple The aother limiting factor is the of the ground pixel. The coherence is measured as the absofilter steps are applied to reduce the system and length of the perpendicular baseline between theprocessing acquired lute value correlation between related pixels and varies noise and tothe enhance the searched signal, i.e. the change images for of data processing. Gens (1998) described the spafrom 0 to 1. This corresponds from low to high coherence of the earth surface (Wegmueller and Strozzi, 1998; Werner tial distribution of baseline lengths for ERS. Here, the practi(see examples in Figs. 4the andphase 5). The higher coherence et al., 2002). Thereafter, unwrapping process is excal limit for InSAR processing is restricted to athe perpendicuthe better the final differential interferogram. ecuted to generate the differential interferogram. This kind lar baseline length up to 600 m. For the generation of DEMs The main radar frequency used by SAR satellites, (ERS of processing is called the 2-pass approach. The topograhic baseline length between 150 m and 300 m are most suitable.1phase can also and bebetween removed infor relation /2, ENVISAT RADARSAT), isa third in theimage C-band with Baseline lengths 30by m using and 70 m are useful sur- a to the master image with a long baseline by generating a nominal frequency of 5.3 GHz corresponding to a waveface change detection and baselines smaller than 30 m secare ond real interferogram. The result of this 3-pass approach length ofto 5.6derive cm. The frequencies are in the L-band excellent earthother surface movements. These limitsis thealso final differential interferogram, (figure 2). (1-2.6 GHz) used the Japanese JERS and ALOS are valid for by ENVISAT data, satellites because ENVISAT uses andsame the X-band (8.2-12.4 GHz) for the new German Radar the wavelength as ERS. satellite TerraSAR-X. The higher the radar frequency the 3 Results from the application of InSAR techniques to 2.2 approaches lowerProcessing the penetration depth in any type of material. Addilandslide monitoring tionally to this circumstance there is also a dependence on The image selectionand hasitsamoisture significant influence on the final surface roughness content. Thewe C-band repIn theofframework of the EUprocessing. project OASYS, processed results the interferometric The criteria varydepth acresents a reasonable compromise between penetration data from testspecific sites in several countries. In general,The European cording to the of an investigation. most inRemote the canopies treesobjects and bushes and Sensingofsatellite (ERS 1 and 2) backscattering data were used.propFor important parameters are the sensor type and the availabilerties of the soil of in relation to the and (Henderson the processing the Greek test Lsite inX-band Prinotopa we also ity of data. The andFatland, the spatial distribution of the and 1998;temporal Nolan and 2003). The repeat cyusedLewis, ENVISAT observations. The subsampled 3sec-SRTM baselines and the terrain characteristics are also parameters cle between 2 or more scenes depends on the satellite misDEM of the regions of investigations were imported into the that have great on the data processing. The days spatial sion andprocessing is 35 influence daysapproaches. for ERS and and for 2-pass InENVISAT most cases of24 the 3-pass slope orientation can be taken into consideration by choosing RADARSAT. The data time sets interval for InSAR is limprocess we used derived from a processing ERS 1-/2 tandem descending or ascending satellite tracks. Theon first ited by thewith loss of coherence depends theprocessing mission long baselineswhich for the generation ofvegetation a height step in the processing chain is the co-registration cycle growth rate. Both types of DEMs canbetween modeland in the SAR geometry. serve as two radar scenes. If these scenes are co-registered than it a The useful datalimiting source for modelling purposes and analyzing other factor for SAR data processing is the is possible to calculate the interferogram by multiplying the length of the perpendicular baseline between the acquired

www.adv-geosci.net/14/189/2008/

Ba

The InS on the B Yunyang voir, 223 of the Y the 18th rainfall rock ma live on vestigat and 450 degrees 2003). The E by a sm This ne using th The 2-p of fram 07.01.96 perpend orbits. F rections tation gr area com herence strong d ing prop and veg ferogram figure 4 The c by sequ face for 01.07.19 5 mm. T sensor a part of t at the fo yang we In 19 which c ing poin work w receiver Wild DI and Nov the mid tion area Jiang, 2

4

B.Riedel and A.Walther: InSAR processing for the recognition of landslides

B. Riedel and A. Walther: InSAR processing for the recognition of landslides 4

B.Riedel and A.Walther: InSAR processing for the recognition of landslides

191

Fig. 4. The left side of the figure shows the coherence map for the area of investigation. White pixels corresond to high coherence and grey/black pixels tofigure low orshows no correlation. The right side offor the figure shows resulting fringe pattern on the amplitude radar coherence Fig. side the the map the of White pixels corresond to high high coherence and and Fig.4.4.The Theleft left sideofof thefigure shows thecoherence coherence map for the area area oftheinvestigation. investigation. Whiteoverlayed pixels corresond to image the 2-pass processing. Movement rates areofvisible for the shows landslidethe andresulting for the cityfringe of Yunyang. Theoverlayed inlet figure shows grey/black totofrom low ororno correlation. The right side the figure pattern on the thetheamplitude amplitude radar radar grey/blackpixels pixels low no correlation. The right side of the figure shows the resulting fringe pattern overlayed on unwrapped phase for the middle part of the landslide in detail with a movement rate of 14 mm per 6 months. image Yunyang. The The inlet inlet figure figure shows shows the the imagefrom fromthe the2-pass 2-passprocessing. processing. Movement Movementrates ratesare arevisible visible for for the the landslide landslide and and for for the the city city of of Yunyang. unwrapped unwrappedphase phasefor forthe themiddle middlepart partofofthe thelandslide landslideinindetail detailwith withaamovement movement rate rate of of 14 14 mm mm per per 66 months. months.

sults with the 1997 geodetical derived observations shows a good coincidence under the assumption of linear movement trend for the slope.

images for data processing. Gens (1998) described the spa-a sults with the 1997 geodetical derived observations shows 3.2ofPrinotopa Greece tial distribution baseline lengths for ERS. practigood coincidence under landslide, the assumption of Here, linear the movement cal limitforfortheInSAR trend slope. processing is restricted to a perpendicuThe southern part of the Prinotopa landslide area can be seen lar baseline length up5.to It600 m. For20the in figure is situated kmgeneration north-east ofof theDEMs city of 3.2 Prinotopa landslide, Greece Ioannina. This150 area intersected nationalsuitable. highway, baseline lengths between misand 300 mbyarea most whichbetween has to be resurfaced summer because for of numerBaseline lengths 30 m andevery 70 m are useful surous landslide damages. These landslidesarea are triggered from Thechange southern part of the Prinotopa landslide can be seen face detection and baselines smaller than 30 m are snow melt, heavy rainfall innorth-east spring and additionally from in figure 5. It is situated 20 km of the city of excellent to derive earth surface movements. These limits earthquakes. A part of this national road will be enlarged, Ioannina. This area is intersected by a national highway, and replaced the Egnatia motorway uses in the are also validreconstructed for ENVISAT data, by because ENVISAT which has toPrinotopa be resurfaced every summer because of numerlandslide the same wavelength as ERS. area. Egnatia Motorway extends from andlandslides will be 680 km and consists of ous landslideIgoumenitsa damages.to Kipi These arelong triggered from bridges and 69 in Tunnels and and numerous major earthworks snow melt, 196 heavy rainfall spring additionally from 2.2 Processing approaches (Egnatia Odos, 2003). earthquakes. The A part of this national road will be enlarged, area of InSAR investigation, shown on the coherence reconstructed and replaced the of Egnatia in the map in figure 5, wasby an area 20 km inmotorway West-East direction The image selection has a significant influence on the height final and 10 km area. in South-North with a maximum Prinotopa landslide Egnatiadirection Motorway extends from results of the interferometric processing. The criteria vary differences of roughly 1500 m. Detailed studies, shown in Igoumenitsa to Kipi and will be 680 km long and2 consistsacof figure 6, were applied of to an area of 8 by 8 km . The most cording to the specific objects an investigation. 196 bridges and 69 Tunnels and numerous major earthworks Nine ERSare 1-/2the satellite radartype scenesand for the 1995- 96 important parameters sensor theyears availabil(Egnatia Odos, 2003). and 23 ENVISAT scenes from 2002 to 2005 were used for ity of data. temporal and the distribution oforbits the the InSAR study. Additionally thespatial SRTM-DEM The areaThe of investigation, shown onand theprecise coherence baselines andwere the terrain characteristics also parameters in processing chain. The steepness of the termap in figure 5, used was anthe area of 20 km inare West-East direction rain caused some the SRTM dataThe set that were that great influence ondata the gaps datainprocessing. spatial andhave 10 km infilled South-North direction with maximum height by bi-linear interpolation. The afinal input DEM had a slope orientation can be taken into consideration by we choosing differences of 1500 m. the Detailed studies, shown in gridroughly width of 30 m. For InSAR processing used the 2 above descending or2-pass ascending The and 3-pass mentioned in the subfigure 6, were applied tosatellite anapproaches area tracks. of 8asby 8 kmfirst . processing section ’Processing Approaches’. The of data stepNine in the processing chain is the co-registration ERS 1-/2 satellite radar scenes for thecombination years between 199596 sets for If thethese different approaches was selected by the basetwo radar scenes. scenes are co-registered than it and 23 ENVISAT scenes from 2002 to 2005 were used for line length, time difference between the acquisiton and the isthe possible to calculate the interferogram by multiplying the study. Additionally the SRTM-DEM and coherence precise orbits resulting coherence. An example of good is disphase forthe each pixel andchain. the coherence. Fromofthis weresignal used in processing The steepness thereal terinterferogram a synthetic interferogram, representing thewere torain caused some data gaps in the SRTM data set that pographic phase of a Digital Elevation Model for the area filled by bi-linear interpolation. The final input DEM hadofa investigation, substracted. get reasonable input grid width ofhas 30 to m.beFor the InSARTo processing we used the values for data processing it is useful to use precise orbits and 2-pass and 3-pass approaches as mentioned above in the suba section DEM, such as the Space Shuttle Radar TopographicofMis’Processing Approaches’. The combination data sion (SRTM) which is available for 90% of the earth surface sets for the different approaches was selected by the baseal.,acquisiton 2005). Multiple with grid resolution of 90 mbetween (Farr etthe linealength, time difference and the filter steps coherence. are applied An to reduce and processing resulting exampletheofsystem good coherence is disnoise and to enhance the searched signal, i.e. the change of the earth surface (Wegmueller and Strozzi, 1998; Werner et al., 2002). Thereafter, the phase unwrapping process is exwww.adv-geosci.net/14/189/2008/

Fig. 5. The coherence map for the Prinotopa area shows good coherence for the selected data. The images from January to April 1996 cover the main triggering period from snow melt and heavy spring rainfall and were taken before the vegetation growth period, which starts in June.

Fig. 5. 5.in The The coherence map for forcoherence the shows played figurecoherence 5. The displayed map ofarea frame Fig. map the Prinotopa Prinotopa area shows good good cocoherence for the selected data. The images from January 783 was derived from the scenes of the 27.01.1996 and the herence for the selected data. The images from January to to April April 06.04.1996 with a 27 m baseline. 1996 cover cover the main triggering period 1996 the main triggering period from from snow snow melt melt and and heavy heavy The final resultsand in the linetaken of the radar sensor for the 2spring rainfall were spring rainfall and were taken before before the the vegetation vegetation growth growth period, period, pass and starts 3-passin processing which June. can be seen in figure 6. The left which starts in shows June. side of the figure the fringe pattern for the Prinotopa area overlayed on the amplitude radar image from the 2-pass processing with the SRTM-DEM. The middle part of this played in generate figure 5. the The displayed coherence of frame figure displays the results of the same area derived from 3-map This ecuted to differential interferogram. kind pass processing. Both results were generated with the data 783 was derived from the scenes of the 27.01.1996 and the of processing is called the 2-pass approach. The topograhic set seen in figure 5. The baseline of the to06.04.1996 with a 27perpendicular m baseline. phase can using isa third in relation pographic pairalso usedbe forremoved the 3-passby approach 157 m image and The final in the06.04.1996 the for the 2to the master image with a line longofbaseline by sensor generating a secwas derived fromresults the scenes of and radar 11.05.1996. pass and 3-pass processing can be seen in figure 6. The left The fringe pattern of the left subfigure results from imperond real interferogram. The result of this 3-pass approach is

sidefinal of the figure shows the fringe pattern the differential interferogram, (Fig. 2).for the Prinotopa area overlayed on the amplitude radar image from the 2-pass processing with the SRTM-DEM. The middle part of this displays thethe results of the same area derived from 33figure Results from application of InSAR techniques to passlandslide processing. Both results were generated with the data monitoring set seen in figure 5. The perpendicular baseline of the toIn the framework of the project approach OASYS, is we157 processed pographic pair used for EU the 3-pass m and data from test sites in several countries. In general, European was derived from the scenes of 06.04.1996 and 11.05.1996. Remote Sensing satellite 1 and 2) data were used. For The fringe pattern of the (ERS left subfigure results from imperthe processing of the Greek test site in Prinotopa we also used ENVISAT observations. The subsampled 3sec-SRTM DEM of the regions of investigations were imported into the

Adv. Geosci., 14, 189–194, 2008

192

B. Riedel and A. Walther: InSAR processing for the recognition of landslides

Fig. 6. The left side of the figure shows the fringe pattern for the Prinotopa area overlayed on the amplitude radar image from the 2-pass

Fig. 6. The left side ofwith the the figure shows The the middle fringepart pattern forthethe Prinotopa area the amplitude radar image from the 2-pass processing SRTM-DEM. displays results of the same areaoverlayed derived fromon a 3-pass processing. The enlargement, the SRTM-DEM. right, shows movement rates of 14part mm displays per 3 months, are in with GPS measurements from 2003. processing withonthe The middle thewhich results ofgood the agreement same area derived from a 3-pass processing. The enlargement, on the right, shows movement rates of 14 mm per 3 months, which are in good agreement with GPS measurements from 2003.

2-pass processing approaches. In most cases of the 3-pass process we used data sets derived from a ERS 1-/2 tandem mission with long baselines for the generation of a height model in SAR geometry. Both types of DEMs can serve as a useful data source for modelling purposes and analyzing functions in Geographic Information System (GIS) within an alert system, e.g. for the derivation of slope parameters. 3.1

Baota landslide, China

The InSAR investigations in China were especially focused on the Baota landslide, (Fig. 3). The site is near the city of Yunyang, located in the middle of the Three Gorges Reservoir, 223 km upstream of the Three Gorges dam, on the left bank of the Yangtze river. The Baota landslide event occurred on the 18 July 1982, triggered from a hundred year frequency rainfall and flooding. During this event 2.3 million m3 of rock mass slid into the Yangtze (Cui, 2000). 4000 people live on the landslide, which is still moving. The area of investigation has a length of 1900 m, a mean width of 1000 m and 450 m height difference. The mean surface gradient is 15 degrees and the thickness of the landslide deposits is 70 m (Cui, 2003). The ERS 1-/2 data processing for this test site was limited by a small number of acquired images in the ESA archive. This necessitated the processing with the 2-pass approach using the subsampled and interpolated 30 m SRTM-DEM. The 2-pass processing chain was applied to the image pair of frame 2979 with the scenes 23 424 (date of acquisition 07.01.96) of ERS 1 and 06256 (01.07.96) of ERS 2. The perpendicular baseline length is 23 m derived from precise orbits. Further processing was accompanied by manual corrections to the standard filter parameters, because two vegetation growth periods and the steepness of the terrain in this area complicated the InSAR processing, as visible in the coherence map of the left side of Fig. 4. One can see the strong Adv. Geosci., 14, 189–194, 2008

difference in coherence as a result of the backscattering properties of the city of Yunyang in relation to the soil and vegetation surface of the landslide. The resulting interferogram for Baota landslide can be seen in the right side of Fig. 4. The changes of colour in the interferogram, represented by sequences of colour cycles, show the change of earth surface for a time difference of approximately 6 months (07.01.– 01.07.1996). One colour cycle, called fringe, is scaled to 5 mm. The displayed displacements in the line of sight of the sensor achieves a maximum value of 14mm for the middle part of the landslide and 3–5 mm for the accumulation zone at the foot of the slope. Movement rates for the city of Yunyang were also derived. In 1997 a GPS and terrestrial network was established which consisted of 5 stable control points and 12 monitoring points on monuments in the moving landslide. This network was measured three times in 1997 using Rogue 8000 receivers for the GPS measurements and EDM from the type Wild DI 2002. Based on these GPS-EDM results for January and November 1997 movement rates between 10–25 mm in the middle of the slope and 5 to 11 mm for the accumulation area were derived for a period of 11 months (Zhang and Jiang, 2003). A comparison of our 1996 2-pass InSAR results with the 1997 geodetical derived observations shows a good coincidence under the assumption of linear movement trend for the slope. 3.2

Prinotopa landslide, Greece

The southern part of the Prinotopa landslide area can be seen in Fig. 5. It is situated 20 km north-east of the city of Ioannina. This area is intersected by a national highway, which has to be resurfaced every summer because of numerous landslide damages. These landslides are triggered from snow melt, heavy rainfall in spring and additionally from earthquakes. A part of this national road will be enlarged, www.adv-geosci.net/14/189/2008/

B. Riedel and A. Walther: InSAR processing for the recognition of landslides reconstructed and replaced by the Egnatia motorway in the Prinotopa landslide area. The Egnatia Motorway extends from Igoumenitsa to Kipi and will be 680 km long and consists of 196 bridges and 69 Tunnels and numerous major earthworks (Egnatia Odos, 2003). The area of InSAR investigation, shown on the coherence map in Fig. 5, was an area of 20 km in West-East direction and 10 km in South-North direction with a maximum height differences of roughly 1500 m. Detailed studies, shown in Fig. 6, were applied to an area of 8 by 8 km2 . Nine ERS 1-/2 satellite radar scenes for the years 1995– 1996 and 23 ENVISAT scenes from 2002 to 2005 were used for the study. Additionally the SRTM-DEM and precise orbits were used in the processing chain. The steepness of the terrain caused some data gaps in the SRTM data set that were filled by bi-linear interpolation. The final input DEM had a grid width of 30 m. For the InSAR processing we used the 2-pass and 3-pass approaches as mentioned above in the subsection “Processing Approaches”. The combination of data sets for the different approaches was selected by the baseline length, time difference between the acquisiton and the resulting coherence. An example of good coherence is displayed in Fig. 5. The displayed coherence map of frame 783 was derived from the scenes of the 27.01.1996 and the 06.04.1996 with a 27 m baseline. The final results in the line of the radar sensor for the 2pass and 3-pass processing can be seen in Fig. 6. The left side of the figure shows the fringe pattern for the Prinotopa area overlayed on the amplitude radar image from the 2-pass processing with the SRTM-DEM. The middle part of this figure displays the results of the same area derived from 3-pass processing. Both results were generated with the data set seen in Fig. 5. The perpendicular baseline of the topographic pair used for the 3-pass approach is 157 m and was derived from the scenes of 06.04.1996 and 11.05.1996. The fringe pattern of the left subfigure results from imperfections of the DEM that is overlayed by the deformation pattern. In contrast to this 2-pass result, the middle subfigure shows unique areas of earth surface changes. The right hand enlargement of Fig. 6 clearly shows the fringe pattern and the random noise of the pixels. It also shows that the whole area is covered by numerous landslides. The derived movement rates correspond to a fringe cycle of 2 mm and reach 14 mm of displacement per 3 months. These landslide movement rates are in good agreement with observations of the combined GPS and terrestrial network. Deformations up to 30 mm per year were derived from these observations (Lakakis, 2003).

4

Conclusions

The results demonstrate on the one hand that there is a strong potential for the detection of landslides and possible earth surface movements with SAR Interferometry and on the other hand the high accuracy of interferometric satellite www.adv-geosci.net/14/189/2008/

193

data processing in comparison with ground based GPS observations. The displacements on the Baota landslide of up to 14 mm in six months from the InSAR processing were in very good agreement with the geodetic derived movement rates of up to 25 mm in 11 months. Landslide movement rates derived from InSAR processing for the Prinotopa study site show the same good agreement with observations of a combined GPS and terrestrial network. Both comparisons between InSAR and geodetic results show that InSAR is a powerful tool for the detection and observation of earth surface processes if the observed area fulfills specific requirements, including sufficient backscattering, flat slope gradients and very slow changes of vegetation. Acknowledgements. The OASYS project (EVG1-2001-00061) was supported by the European Commission under the Fifth RTD Framework Program. The satellite data for the InSAR investigations for the Greek test site were supported by the European Space Agency via a category-1 proposal. Edited by: P. Fabian Reviewed by: M. Larsen and D. Keefer

References Cui, Z. Q.: Failure Mechanism and Prediction ideology for natural slopes of the Three Gorges Area of Changjiang, Bureau of Geotechnique of Changjiang Water Resources Commission (CWCR), Wuhan, 2000. Cui, Z. Q.: Brief introduction of BAOTA Landslide –One of the Examination Spots of OASYS, Bureau of Investigation & Survey, Wuhan, 2003. Egnatia Odos: Egnatia Motorway –Informative Report, Egnatia Odos A. E., Ministry of Environment, Physical Planning and Public Works, Thessaloniki, 2003. Farr, T. G., Rosen, P. A., Caro, E., et al.: The Shuttle Radar Topography Mission, Rev. Geophys., 45, RG2004, doi:10.1029/2005RG000183, 2005. Gens, R.: Quality assessment of SAR interferometric data, Wissenschaftliche Arbeiten der Fachrichtung Vermessungswesen der Universitaet Hannover, Nr. 226, Dissertation, 86–88, 1998. Henderson, F. M. and Lewis A. J.: Principles and Application of Imaging Radar, Manual of remote sensing, Vol. 2, 3 ed., John Wiley & Sons, New York, 867 pp., 1998. Kahmen, H., Eichhorn, A., and Haberler-Weber, M.: A Multi-Scale Monitoring Concept for Landslide Disaster Mitigation, in: Dynamic Planet Monitoring and Understanding a Dynamic Planet with Geodetic and Oceanographic Tools, edited by: Tregoning, P. and Rizos, C., IAG Symposium, Cairns, Australia, 22–26 August, 2005 Series, International Association of Geodesy Symposia, Vol. 130, Springer, 769–775, 2007. Lakakis, K.: Surface Landslides Deformation Monitoring in Egnatia Ods S.A. Internal Report of Egnatia Odos A.E., Ministry of Environment, Physical Planning and Public Works, Thessaloniki, 2003.

Adv. Geosci., 14, 189–194, 2008

194

B. Riedel and A. Walther: InSAR processing for the recognition of landslides

Lu, Z., Kwoun, 0., and Rykhus, R.: Interferometric Synthetic Aperture Radar (InSAR): Its Past, Present and Future, Photogramm. Eng. Rem. S., 73, 3, 217–221, 2007. Niemeier, W. and Riedel, B.: Mehrskaliges geodaetisches Beobachtungskonzept fuer die Ueberwachung groflaechiger Rutschungsgebiete, in: Schriftenreihe des Institutes fuer Markscheidewesen und Geodaesie an der Technischen Universitaet Bergakademie, 7. Geokinematischer Tag Freiberg, 23–30, 2006. Nolan, M. and Fatland, D. R.: Penetration Depth as a DInSAR Observable and Proxy for Soil Moisture, IEEE T. Geosci. Remote, 41, 3, 552–537, 2003. Riedel, B. and Niemeier, W.: Results of InSAR processing in the Changjiang (Yangtze river) region, Proceedings of APSG Workshop, Hong Kong, 5 pp., 2005.

Adv. Geosci., 14, 189–194, 2008

Sidle, R. C. and Ochiai, H.: Landslides- Processes, Prediciton and Land Use, AGU Books Board, Washington, 312 pp., 2006. Wegmueller, U. and Strozzi, T.: Characterization of Differential Interferometry Approaches, European Conference on Synthetic Aperture Radar EUSAR’98, 237–240, 1998. Werner, C., Wegmueller, U., and Strozzi, T.: Processing strategies for phase unwrapping for INSAR applications, Procs. EUSAR Conf., Cologne, Germany, 4–6 June 2002, 353–356, 2002. Zhang, J. and Jiang, B.: GPS landslide monitoring of Yunyang Baota, Report of University Wuhan, 14 pp., 2003.

www.adv-geosci.net/14/189/2008/

Suggest Documents