Sheet Music

A general framework and related procedures for multiscale analyses of DInSAR data in subsiding urban areas

A general framework and related procedures for multiscale analyses of DInSAR data in subsiding urban areas
of 26
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
  See discussions, stats, and author profiles for this publication at: A general framework and related procedures formultiscale analyses of DInSAR data in subsidingurban areas  ARTICLE   in  ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING · JULY 2015 Impact Factor: 3.13 · DOI: 10.1016/j.isprsjprs.2015.04.001 READS 95 6 AUTHORS , INCLUDING:Dario PedutoUniversità degli Studi di Salerno 25   PUBLICATIONS   219   CITATIONS   SEE PROFILE Settimio FerlisiUniversità degli Studi di Salerno 30   PUBLICATIONS   175   CITATIONS   SEE PROFILE Gianfranco FornaroItalian National Research Council 213   PUBLICATIONS   4,008   CITATIONS   SEE PROFILE Diego RealeItalian National Research Council 55   PUBLICATIONS   429   CITATIONS   SEE PROFILE Available from: Dario PedutoRetrieved on: 04 February 2016  A general framework and related procedures for multiscale analyses of DInSAR data in subsiding urban areas Dario Peduto a, ⇑ , Leonardo Cascini a , Livia Arena a , Settimio Ferlisi a , Gianfranco Fornaro b , Diego Reale b a Department of Civil Engineering, University of Salerno, via Giovanni Paolo II, 132, 84084 Fisciano (SA), Italy b Institute for Electromagnetic Sensing of the Environment (IREA – CNR), via Diocleziano 328, 80124 Naples, Italy a r t i c l e i n f o  Article history: Received 6 November 2014Received in revised form 23 January 2015Accepted 1 April 2015 Keywords: Multipass DInSAR techniquesSubsidenceDetectionMappingMonitoring a b s t r a c t Inthelast decade Differential Synthetic Aperture Radar (DInSAR) data weresuccessfully tested inanum-ber of case studies for the detection, mapping and monitoring of ground displacements associated withnaturaloranthropogenicphenomena.Morerecently,severalnationalandregionalprojectsallaroundtheworldprovidedrichdataarchiveswhoseconfidentuse,however,shouldrelyonmultidisciplinaryexpertsin order to avoid misleading interpretations. To this aim, the present work first introduces a generalframework for the use of DInSAR data; then, focusing on the analysis of subsidence phenomena andthe related consequences to the exposed facilities, a set of srcinal procedures is proposed. By drawingamultiscale approach thestudyhighlights thedifferent goals tobepursued atdifferent scales of analysisvia high/very highresolution SAR sensors andpresents the results withreference tothe case study of theCampania region (southern Italy) where widespread ground displacements occurred and damages of dif-ferent severity were recorded.   2015International SocietyforPhotogrammetryandRemoteSensing, Inc. (ISPRS). PublishedbyElsevierB.V. All rights reserved. 1. Introduction In the last decade images acquired by Synthetic Aperture Radar(SAR) sensors and processed via Differential Interferometry algo-rithms (DInSAR) have been increasingly applied by the scientificcommunity to study the measurable effects of natural or anthro-pogenic phenomena (or dangers) in different fields of Geosciences – including Geology, Geophysics and Glaciology(Crosetto et al., 2003) – as well as in the Civil and EnvironmentalEngineering. This is essentially due to several advantages offeredbyDInSARtechniques,suchas:thepossibilityofmeasuringgroundsurface displacements with sub-centimeter accuracy by exploitinglarge datasets of SAR images acquired over more than 20years;affordable costs for monitoring large areas.As a result the scientific community analyzed a number of casestudies which successfully investigated potential and limits of theDInSAR techniques (for instance a comprehensive overview of applications to slow-moving landslides was recently provided byWasowski and Bovenga, 2014).For what concerns the categorization of the data based on thespatial resolution, it is important to remark since now that in thiswork we refer to the GMES data warehouse specification (Brefort,2011) where sensors with spatial resolutions higher than 4m areclassified as very high resolution (VHR) systems, whereas datawith spatial resolutions in the range 4–30m are referred to highresolution. In many publications it is, however, possible to findthese two categories referred to as high and medium resolution,respectively.Asfornaturaloranthropogenicsubsidencetheavailablestudies– carried out by using data acquired via European Space Agency(ESA) and Canadian Space Agency (CSA) high resolution (Brefort,2011) ERS-1/2, ENVISATSARandRADARSAT-1/2sensors – demon-stratedthecapabilityofDInSARtechnologytomonitorgroundsur-face displacements induced by either mining (Carnec et al., 1995;Haynes, 2000; Kircher et al., 2003; Raucoles et al., 2003; Crosettoet al., 2005; Wang et al., 2009; Herrera et al., 2010b) or waterextraction (Haynes, 2000; Galloway et al., 2000; Cascini et al.,2006; Herrera et al., 2009b; Calderhead et al., 2011; Cigna et al.,2012; Sanabria et al., 2014; Tomás et al., 2014) or undergroundconstruction works (Giannico et al., 2012).Recently, a breakthrough for the monitoring of subsiding urbanareas at detailed scale (>1:5000) was provided by the last genera-tion X-Band very high resolution (Brefort, 2011) SAR sensors   2015 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights reserved. ⇑ Corresponding author. Tel.: +39 089 964120, mobile: +39 3286935656. E-mail addresses: (D. Peduto), (L. Cascini), (L. Arena), (S. Ferlisi), (G. Fornaro), (D. Reale).ISPRS Journal of Photogrammetry and Remote Sensing 105 (2015) 186–210 Contents lists available at ScienceDirect ISPRS Journal of Photogrammetry and Remote Sensing journal homepage:  TerraSAR-X/TanDEM-X (TSX/TDX) mission of the GermanAerospace Center (DLR) and the COSMO-SkyMed (CSK) constella-tion of the Italian Space Agency (ASI) (Herrera et al., 2010b;Cascini et al., 2013; Fornaro et al., 2014). Such sensors are charac-terized by a higher revisiting time than that pertaining to the ESAsensors (35days); in particular, in the case of CSKthe use of a con-stellation of satellites allows bringing the revisiting time down to4days (on average) thus providing exceptional capability of col-lecting stacks of data useful for interferometric analysis (about30 SAR images) in a shorter time interval, i.e. few months insteadof few years. Furthermore, the resolution improvement allowsmore details of single facilities to be observed and hence their pre-cisemonitoring,as testifiedbytheavailableliteraturereportingontheuseofTerraSAR-Xdata(Gernhardtetal.,2010;ZhuandBamler,2010; Realeet al., 2011a;Fornaroet al., 2012, 2013). Moreover, therecently launched (April 2014) Sentinel mission of the EuropeanSpace Agency will provide continuity of ERS-1/2 and ENVISAT dataarchives with reduced revisiting times and larger coverage swath.Thanks to the reliability of DInSAR data, a number of interna-tional (TERRAFIRMA, 2013; Del Ventisette et al., 2013), nationaland regional (Meisina et al., 2008; Terranova et al., 2009; RisknatProject, 2012) projects all over the world were developed in thelast years so that the end-users are currently provided with exten-sive datasets. An example in Italy is the ‘‘Piano Straordinario diTelerilevamento Ambientale – PST Project’’ (MATTM, 2010) offer-ing a nation-wide coverage of high resolution SAR sensors from1992 to 2010.Inspiteoftheapplicationsandthefutureimprovementsofferedby new satellites, standardized procedures allowing the reliableuse of DInSAR data and their interpretation at all scales of analysis(i.e.small,medium,largeanddetailedaccordingtoFelletal.,2008)are still lacking. To this aim, the present work wants to provide acontribution by first introducing a general framework to be fol-lowed when using DInSAR data to study natural or anthropogenicphenomena. Then, srcinal procedures for the analysis of DInSAR data–derivingfromdifferentsensorsoralgorithms–areproposedwith particular reference to the investigation of subsidence-in-duced ground displacements in urban areas. The proposed proce-dures are tested in well-documented sample areas of theCampania region (southern Italy) at four scales of analysis (small,medium, large and detailed according to Fell et al., 2008) by usingDInSAR data resulting from the rich archive dataset of the ItalianPST Project as well as from an available dataset of CSK data. 2. SAR sensors and DInSAR techniques  2.1. SAR sensors The basic concepts and the first experimental demonstration of interferometric techniques were presented by Zebker andGoldstein (1986) for topographic mapping purposes and byGabriel et al. (1989) for the detection and mapping of small eleva-tion changes by using SEASAT SAR sensor operating in the period June–October 1978. However, the beginning of the developmentthat brought interferometry to operational applications is associ-ated with the launch of ERS-1 satellite in 1991, followed by thetwin sensor ERS-2 launched in 1995. Starting from these sensorsand boosted by evidence of the potential for applications of inter-ferometric techniques, several other sensors were launched in thefollowing years with a significant growth since 2007; many othersare planned in the early next years. Fig. 1 provides a picture of thehistorical evolution of the availability of SAR sensors since 1992when ERS-1 became operative. The sensor working frequency aregenerallyallocatedintheL, CandXbands,correspondingtowave-lengths of about 18 cm, 5.6cm and 3.1cm, respectively.SAR satellites are placed in the so called Lower Earth Orbits(LEO), between 500km and 800km, and follow polar orbits inorder to provide a global coverage. Accordingly, scenes may beobserved by SAR sensors over ascending or descending orbits typ-ically with opposite ground looking directions.Earlier SAR sensors were characterized by fixed swaths (i.e.fixed illumination beam width); modern sensors offer differentoperative modes (e.g. fine, strip, spot, scan) which allow a tradeoff between spatial coverage and spatial resolution. In other words, itis possible to have higher resolutions (e.g. up 1 m for civil applica-tions) with limited coverage (e.g. 10km) and coarser resolutions(e.g. 10m) with wider coverage (e.g. a hundred of kilometers oreven more).FurthermorepastgenerationSARsensorswerecharacterizedbya fixed illumination direction, commonly referred to as Line Of Sight (LOS), whereas modern sensors allow the LOS to be conve-niently set. For instance, the LOS angle with respect to the verticaldirection was 23   for ERS-1/2 whereas for most of the currentlyoperational sensors can range from 20   up to 50  . This ability tovary the illumination angle is important when imaging mountain-ous areas. Moreover, the possibility to change the LOS is alsoimportant for monitoring purposes because displacements mea-sured by DInSAR techniques are components of the 3D grounddisplacement.TakingintoaccountthatSARsatelliteorbitsarepolarandradarsimage the scene typically in the broadside direction – which isorthogonal to the orbit direction – the LOS is almost belonging tothe east–west/vertical plane. In particular, LOS direction is down-ward and being SAR sensors generally right looking, the LOS direc-tion is eastward in ascending passes and westward in descendingpasses. Ground displacements occurring along North and Southdirections are, therefore, almost undetectable unless MultipleAperture Techniques (MAI) – exhibiting significantly coarser reso-lutions – are used ( Jung et al., 2011). For instance, the so-calledsensitivity unit vector (Massonet and Feigl, 1998; Colesanti et al.,2003) has components that for ERS/ENVISAT sensors (in absolutevalues)are:0.38fortheeastcomponent,0.08forthenorthcompo-nent and 0.92 for the vertical component. Depending on the sub-swath, for recent satellites the east and the vertical componentcan be traded off whereas typical orbital inclinations do not allowchanging significantly the north component sensitivity.The time taken for a satellite to repeat the same orbit is calledthe ‘revisiting time’ (RT in Fig. 1); it determines the highest possi-ble temporal sampling rate of signal (time series) associated withthe measured target displacement. Depending on the orbit heightaswellasonthenumberofsensorsforeachsatelliteconstellationsthe revisiting time ranges between tens of days to a day (Fig. 1).Another key issue is that SAR sensors are coherent systems butthe implementation of interferometric techniques, which areintrinsically characterized by a diversity factor of angular or tem-poral srcin, requires that the scattering on the scene is coherentas well, and similarly that the scattering properties must notchange‘significantly’overtime.IntheSARjargonitisrequiredthatthe measured echoes from the scene scattering centers are coher-ent(FornaroandPascazio,2014).Thecoherence,whichisroutinelyestimated on the available interferometric SAR dataset, can bequantified in terms of statistical characterization of the radarechoes: it is a number between 0 and 1; lower coherence meansa higher interferometric phase noise contribution and vice versa.Vegetated areas are prone to temporal coherence losses whereasarid and built up areas are typically associated with higher coher-ence values, i.e. a better quality of the interferometric signal. Thewavelength impacts the coherence; generally the larger the wave-length the higher the coherence. Therefore lower frequencies aremore suitable for the observation of vegetated areas (Fornaro andPascazio, 2014). D. Peduto et al./ISPRS Journal of Photogrammetry and Remote Sensing 105 (2015) 186–210  187   2.2. DInSAR techniques Since early 2000s several multi-pass DInSAR algorithms havebeen developed and widely used to retrieve information on dis-placements of the topographic surface, among them: thePersistent Scatterers (PSInSAR) technique (Ferretti et al., 2000,2001), the Small Baseline Subset (SBAS) technique (Berardinoet al., 2002), the Coherent Point Target Analysis (CPTA) (Moraet al., 2003), the Stable Point Network (SPN) (Arnaud et al.,2003), the Interferometric Point Target Analysis (IPTA)(Wegmüller et al., 2005), the Spatio-Temporal UnwrappingNetwork (STUN) (Kampes and Adam, 2005), the Stanford Methodfor Persistent Scatterers (StaMPS) (Hooper et al., 2004) and theEnhancedSpatialDifferences(ESD)(Fornaroetal.,2007).Theavail-able techniques for the analysis of phase signals in interferometricstacks can be grouped in two classes: Persistent ScatterersInterferometry (PSI) (Ferretti et al., 2000, 2001; Costantini et al.,2008; Crosetto et al., 2008) and Small-Baseline techniques(Berardino et al., 2002; Fornaro et al., 2009).PSI techniques operate at full spatial resolution and identifyreliable scatterers by measuring their multitemporal coherencerelated to the phase stability; monitored scatterers correspond toman-made structures (buildings, roads, bridges) or bare rockswhose size is smaller compared to the system resolution.Conversely, the SBAS techniques are tailored to decorrelatingscatterers (i.e. scatterers that may be distributed in the resolutioncell or characterized by slow temporal change of the scattering)and to measure ground deformations over large areas (Berardinoet al., 2002). These techniques take benefit of a spatial averaging(multilook) to improve the quality of the phase signal thus per-forming a change of the scale of analysis. Moreover, the interfero-grams are generated according to small temporal and spatialseparation constraints in order to further reduce decorrelationeffects associated with possible coherence losses typically presentin the analysis of rural areas. The SQueeSAR algorithm (Fumagalliet al., 2011) has been recently developed to perform a mixed scaleof analysis which includes in PSInSAR processing the multilookoperation and a data dependent weighting of interferograms inorder to improve the capabilities of the srcinal algorithm in themonitoring of decorrelating scatterers.SBAS techniques are suitable for investigating ground deforma-tions at medium scale but can be useful also to calibrate data forfull resolution analysis. At the medium scale, the techniqueexploits averaged (multi-look) interferograms and allows the gen-eration of mean velocity maps and associated time series for areasextending for some thousands of square kilometres (up to100km  100km), with a ground resolution typically of the orderof80  80m(ERS-1/2andENVISAT).SBAScanbeusedalsoatlargescale by exploitingsingle-lookinterferograms, i.e. generated at fullspatial resolution (typically of the order of 10  10m for ERS-1/2and ENVISAT; 3 m  3 m for very high resolution sensors) (Lanariet al., 2004b); however the limitations on the baseline can leadto detection and localization accuracy loss with respect to PSIwhen dealing with persistent scatterers (PS) typically associatedwith man-made targets.With reference to full resolution multitemporal DInSAR analy-sis, particularly suited for investigating single structures with highresolution sensors (Schack and Soergel, 2014), a recent Fig. 1.  The available SAR sensors, the archives and the achievable resolutions.188  D. Peduto et al./ISPRS Journal of Photogrammetry and Remote Sensing 105 (2015) 186–210  advancement is represented by the introduction of tomographyapproaches (Fornaro et al., 2005) and, more specifically, theMulti-Dimensional Imaging technique (MDI) (Lombardini, 2005;Fornaro et al., 2009) which allows the identification, localizationand monitoring of scatterers at full resolution with improved per-formanceswithrespecttoclassicalPSIapproaches.MDItechniquesfostered single building monitoring by retrieving better detectionand estimation accuracy with respect to PSI also allowing the lay-overeffectsonverticalstructures(e.g.buildings)–particularlyevi-dent in VHR systems – to be solved. MDI is currently used at anoperational level by the German Aerospace Center (DLR) for theanalysis of urban areas (Zhu, 2011; Wang et al., 2012) withTerraSAR-X data. Furthermore, the resolution improvement allowsthe capture of more details of single facilities under observationandhencetheirprecisemonitoring,astestifiedbytheavailablelit-erature mainly focused on the use of TerraSAR-X data (Gernhardtet al., 2010; Zhu and Bamler, 2010; Reale et al., 2011a,b; Fornaroet al., 2012, 2013).A last comment is dedicated to the achievable accuracy of theDInSAR data which depends on a number of factors such as thewavelength,thecoherence,thenumberofimages, theoveralltem-poral span and, last but not least, the confidence level of the pro-cessing algorithm and implementation issues which may dependalso on the processed dataset. Quantitative assessments carriedout by comparisons with independent measurements, such as inColesanti (2003), Casuetal. (2006) andHerreraet al. (2009a), indi- catethat,asaruleofthumb,theaccuracyisof1–2mm/yearfortheaverage displacement velocity and of 5–10mm for the single timeseriesdisplacement. Anexperimental evidenceof the possibilitytoachieve an accuracy up to the order of 1mm on a single displace-mentmeasurementwasprovidedinFornaroetal.(2013),althoughthe results refer to the monitoring of a specific structure (a rein-forced concrete bridge). 3. The proposed approach The definitionof a methodological approachin problems whereinput dataderive frominnovativetechnologies representsa neces-sary activity in order to enhance a good practice and to promotetheir diffusion. Accordingly, the present section first introduces ageneral framework for the use of DInSAR data; then, focusing onthe application to subsidence phenomena, it briefly outlines thestate of the art and finally proposes srcinal procedures allowingthe pursuit of different objectives at different scales.  3.1. General framework Owing to the complexity of DInSAR data, as described inCrosetto et al. (2005), the most advanced applications and the bestresults are usually achieved through a close cooperation betweenDInSAR specialists and people able to use DInSAR data in orderto analyse the danger at hand. In this context DInSAR specialistscan play a fundamental role in helping the end-users to be fullyaware of limits and potentials of the techniques. These conceptsinspire the general framework proposed in Fig. 2 which summa-rizes the sequence of activities involving the experts of both radarimage processing (specific activities are framed in red) and thedanger(specificactivitiesareframedinblue).Inparticular,thefirststage includes the definition of the problem, namely: the type of the danger (e.g. subsidence, slow-moving landslides) to be ana-lyzed; the extent of the area to be investigated and the descriptionof its specific features which most directly influence the use of satellite data (i.e. vegetation cover, presence of urbanized area,topographical and historical information about any changesoccurred in the area during the period of observation, etc.).Once the problemis defined, it is necessaryto arrange the anal-ysestobeperformedbyidentifying:theaimandthescale(e.g. Felletal.,2008)oftheanalysis;theperiodofobservation;themappingunits (Guzzetti, 2005) – e.g. grid cell for subsidence phenomena orterrainunitsforslow-movinglandslides–onwhichtocarryouttheanalysis in a GIS environment. After this, there is an ‘‘exit/problemredefinition’’ optionsince insome cases DInSARtechniques cannotrespond (or just partially) to the end-user’s initial problem (e.g. if the expected displacement rate is too high, for instance some dm/yr). So a decision has to be taken whether DInSAR will be used orit would rather need to be associated with other techniques (e.g.opticalimages,groundbaseddata,etc.).Onthecontrary,inthecasethat DInSAR analysis can proceed, with the contribution of bothexperts it is possible to select the image dataset according to thecharacteristics of the available sensors such as: band, orbital data,geometryof acquisition, dataresolution, stacks of availableimageson the study area, revisiting time and swath. In this regard, it isworth pointing out that together with the wavelengths the revisit-ing time directly impacts the maximum measurable displacementover the time; more specifically it bounds the maximum spatialvariation of the displacement between two consecutive observa-tions (i.e. the bound acts on the displacement spatial variationsandnotonthedisplacementofasinglepointbecauseDInSARtech-niquesworkbyintegratingtemporalandspatialvariationbetweencoherent pixels over the time and the space). It is for this reasonthat displacements measured by DInSAR techniques are referredto the known displacement of a single point, commonly referredtoasreferencepointortiepoint,generallyassumedasstablepoint.Moreover,alsothebandinfluencesthesensorselectionsincelowerfrequency bands (L band, i.e. longer wavelength sensors) are moresuitablefor themonitoringof faster dynamics, withtheadvantage,as previously pointed out, of penetrating the vegetation coverage;whereas higher frequencies (X band) provide a higher sensibilityof the sensors to smaller displacements such as those due to ther-mal variation of the built up environment. A good compromise isprovided by C band.OnceSARimagesarechosentheyundergotheprocessingphasevia specific algorithms whose selectionis in charge of both expertsdepending on the available input data and the expected outputs tobe used in the analysis (e.g. velocity, time series, coherence, refer-ence point).The final step concerns the analysis of DInSAR data which, as itis shown in next sections for subsidence phenomena, can be car-ried out at different scales pursuing different aims.  3.2. State of the art of the applications to subsidence phenomena The number of applications of DInSAR techniques to detect andmonitorsubsidingareasatdifferentscalesrapidlyincreasedduringthe last decade. Table 1 lists some case studies available in the sci-entific literature grouped according to the scale of analysis in turnrelated to the extension of the typical area to be zoned as sug-gested by Fell et al. (2008) (see also Section 3.3). For instance, with reference to the Piedmont region (northernItaly), Meisina et al. (2008) developed at small scale (<1:100,000)a methodological approach for the interpretation of PSInSAR datavia the preliminary detection of the so called ‘‘anomalous areas’’wherein significant movements are recorded (clusters with a min-imumof 3 PS havinga maximummutual distance of 50manddis-placement rates exceeding±2mm/year); then, the interpretationis first carried out via the overlap, in a GIS environment, of anoma-lous areas with other layers (i.e. topographic map, geological map,orthophoto) that might have relevance in explaining the patternsof motions of PS points. At small scale, Vilardo et al. (2009) dis-criminated the vertical and east–west displacement componentsin Campania region (southern Italy) thanks to the availability of  D. Peduto et al./ISPRS Journal of Photogrammetry and Remote Sensing 105 (2015) 186–210  189
Similar documents
View more...
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks