Interferometric Synthetic Aperture and Radargrammetry for

Accurate Digital Model Generation in New South Wales,

Australia

By

Jung Hum Yu

A thesis submitted to The University of New South Wales

in partial fulfilment of the requirements for the degree of Doctor of Philosophy

Geodesy and Earth Observing Systems Group

School of and Spatial Information Systems

The University of New South Wales

Sydney NSW 2052, Australia

March, 2011 ORIGINALITY STATEMENT ‘I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgement is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project’s design and conception or in style, presentation and linguistic expression is acknowledged.’

Signed

Date COPYRIGHT STATEMENT ˄I hereby grant the University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all proprietary rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation. I also authorise University Microfilms to use the 350 word abstract of my thesis in Dissertation Abstract International (this is applicable to doctoral theses only). I have either used no substantial portions of copyright material in my thesis or I have obtained permission to use copyright material; where permission has not been granted I have applied/will apply for a partial restriction of the digital copy of my thesis or dissertation.'

Signed

Date

AUTHENTICITY STATEMENT ‘I certify that the Library deposit digital copy is a direct equivalent of the final officially approved version of my thesis. No emendation of content has occurred and if there are any minor variations in formatting, they are the result of the conversion to digital format.’

Signed

Date ABSTRACT

A is a prerequisite to the thematic map of geological and environmental information. The Digital Elevation Model (DEM) has become an important source of topographical information for many scientific and uses, such as hydrological and geological studies, infrastructure planning and environmental applications. Digital Elevation Models (DEMs) can be generated by using various techniques and by using a range of data sources, including ground surveying, , optical , radar, and laser scanning.

Where topographical data is unavailable, global coverage elevation data sets, typically DEMs based on remotely sensed data, can be the main source of information.

Remote sensing techniques are a rapid means of acquiring elevation information over extensive . In particular, processing remotely sensed data collected by Earth observation satellites is a very efficient and cost-effective mean of acquiring up-to-date and relatively accurate land cover and topographic information. Active remote sensing sensors (such as radar), which can operate in almost all weather conditions and also in darkness using their own illumination have become an important remote sensing technique.

In radar remote sensing systems, DEM generation methods are based on the analysis of Synthetic Aperture Radar (SAR) images, and include , radargrammetry, radarclinometry, and polarimetry. The two most common methods for generating DEMs from SAR images are: (1) radargrammetry, a technique derived from

i photogrammetry and based on the stereoscopic principle, (2) interferometry, based on the phase differences between identical imaged points in two SAR images.

This thesis describes the methods or techniques of DEM quality improvement and reduction in elevation errors, which are generated by various SAR techniques and imagery. InSAR DEM generation relies on the measurement of phase difference between two sets of complex radar signals, i.e. the range difference between the satellite-borne radar instrument and the ground targets reflecting the radar transmissions. In InSAR DEM generation, the so-called “master image” parameters, such as signal wavelength, incidence angle, and SAR image relationship (i.e. perpendicular baseline), affect the final DEM products.

Furthermore, the different orbit direction (ascending and descending) provides a different representation of terrain and multi-temporal observation, leading to a more detailed representation of terrain over the same target area. Hence, ways of improving the quality of InSAR DEMs include using images collected by the satellite sensor from different orbit directions and multi-epoch data acquisitions. Also, in the case of InSAR

DEM generation, major issues related to long satellite repeat-cycle times and low resolution DEM updating are discussed and solutions are proposed for ground distortion excluding method and selective elevation updating.

Depending on the data acquisition conditions, the InSAR technique is less robust and more difficult to implement, particularly because the InSAR technique often produces poor results caused by poor coherence, atmospheric differences between two

ii processed images and conditions. These three factors are influenced by the incidence angle and the Doppler similarity, which are quite stringent.

InSAR requires the expectations of a certain baseline while interferometry is sensitive to the direction of sensor movements and some other factors. For this reason, the radargrammetry technique is an important alternative for DEM generation.

In summary, the major contributions of this thesis are to analyse research into the following;

• Wide coverage and weather independent elevation data generation method

using active remote sensing

• Stable data updating of various ground information by different capability

satellite remote sensing

• Geometric distortion exclusion in side-looking observation systems

• Elevation accuracy improvement by atmospheric disturbance reduction

• Robust DEM generation techniques using InSAR and radargrammetry

iii ACKNOWLEDGEMENT

This research was performed in the Geodesy and Earth Observing Systems (GEOS) group in the School of Surveying and Spatial Information Systems (SSIS) at the

University of New South Wales (UNSW), Sydney, New South Wales, Australia.

Many people supported me in various ways during the journey of my research. First and foremost, I would like to thank my supervisor, Associate Professor Linlin Ge, for supervising and supporting my doctoral research. I would also like to express my thanks to my co-supervisor Chris Rizos, Professor and Head of SSIS at UNSW, for his patience and inspiration on supporting my research. I would not have learned and achieved so much without the support and encouragement from my supervisors. I would also like to thank both of my supervisors again for their careful reading of this work.

Thanks also go to all the people at the SSIS for numerous discussions, particularly relating to remote sensing and the philosophy of life. Especially, I am so grateful to our current and former GEOS group members, including Dr. Xiaojing Li, Dr. Hsing-Chung

Chang, Alex Hay-Man Ng, Kui Zhang, Wendi Peng, Wing Yip Lau, Dan Meng, Zhe

Hu, Li Gou, Hai Tung Chu, Rattanasuda Cholathat, Wang Xin, Yuanyuan Zhang.

I would like to also acknowledge Dr. Sungheuk Jung, Dr. Yusen Dong, Dr.

Mahmoud Salah, Mr. Cemal Ozer Yigit, visiting scientists at UNSW from Chungbuk

National University in Republic of Korea, China University of Geosciences in China,

Banha University in Egypt, Selcuk University in Turkey.

iv The European Space Agency is acknowledged for providing data from the ERS SAR missions. The Earth Remote Sensing Data Analysis Centre is acknowledged for providing the ALOS PALSAR data. The Land and Property management Authority is acknowledged for providing the data.

I would like to also acknowledge Korea-Australasia Research Centre members; Prof.

Chung-Sok Suh, Dr. Gi-Hyun Shin, Dr. Seung-Ho Kwon, Dr. Sungbae Ko, Ms.

Hyunok Ke, Mr. Joseph Kim.

And, my friends Mr. Daehui Oh and his wife Ms. Jungeun Kim, Dr. In Jun and Ms.

Mie Yang, and Mr. Chaelqu Yang, also my colleagues Ms. Jasmin Kim, Mr. Hongju

Park.

Last, but far from least, I give my most heart-felt appreciation and thank my parents, my sister, brother-in-low and nephew for their emotional supports and understanding during my PhD studies.

v LIST OF ABBREVIATIONS

ACRES Australian Centre for Remote Sensing

ALOS Advanced Land Observing Satellite

ALS Airborne Laser Scanning

AMI Active Microwave Imager

CCRS Canada Centre for Remote Sensing

COSMO-SkyMed Constellation of small Satellites for the Mediterranean basin

Observation

CRCSI Cooperative Research Centre for Spatial Information

CSA Canadian Space Agency

DEM Digital Elevation Model

DInSAR Differential Interferometric Synthetic Aperture Radar

DLR Deutsches Zentrum für Luft- und Raumfahrt (German Aerospace Centre)

EORC Earth Observation Research Centre

ESA European Space Agency

ENVISAT Environmental SATellite

ERS-1/2 1st/2nd European Remote Sensing Satellites

ERSDAC Earth Remote Sensing Data Analysis Centre

GA Geoscience Australia

GCP Ground Control Point

GEOS Geodesy and Earth Observing Systems Group

GIS Geographic Information System

GPS Global Positioning System

IFSAR see InSAR

InSAR Interferometric Synthetic Aperture Radar vi LIST OF ABBREVIATIONS

IMU Inertial Measurement Unit

JAXA Japan Aerospace Exploration Agency

JERS Japanese Earth Resources Satellite

JPL Jet Propulsion Laboratory

LiDAR Light Detection and Ranging

LOS Line of Sight

NASA National Aeronautics and Space Administration

PRISM Panchromatic Remote-Instrument for Stereo Mapping

RADAR Radio Detection And Ranging

PALSAR Phased Array type L-band Synthetic Aperture Radar

RAR Real Aperture Radar

RMSE Root Mean Square Errors

RTK-GPS Real Time Kinematic Global Positioning System

SAR Synthetic Aperture Radar

SLAR Side Looking Airborne Radar

SLC Single Look Complex

SLR Side Looking Radar

SRTM Shuttle Radar Topography Mission

TanDEM-X TerraSAR-X add-on for digital elevation measurements

vii TABLE OF CONTENTS

ABSTRACT ...... i

ACKNOWLEDGEMENT ...... iv

LIST OF ABBREVIATIONS ...... vi

TABLE OF CONTENTS ...... viii

LIST OF FIGURES ...... xii

LIST OF TABLES ...... xxi

CHAPTER 1 INTRODUCTION ...... 1 1.1 Introduction ...... 1 1.2 Motivation and Objectives ...... 4 1.3 Overview ...... 9

CHAPTER 2 RADAR AND SYNTHETIC APERTURE RADAR ...... 10 2.1 SAR Imagery ...... 11 2.1.1 Spotlight ...... 11 2.1.2 Strip-map ...... 12 2.1.3 ScanSAR ...... 12 2.2 SAR Platforms for the Digital Elevation Model Generation ...... 13 2.2.1 Airborne SAR systems for DEM generation ...... 14 2.2.2 Space-borne SAR systems for DEM generation ...... 16 2.2.3 The Shuttle Radar Topography Mission ...... 20 2.3 SAR Geometry ...... 25 2.3.1 Radarclinometry ...... 25 2.3.2 Interferometry ...... 26 2.3.3 Radargrammetry ...... 27 2.3.4 Polarimetry ...... 28 2.4 Concluding remarks ...... 29

viii CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION ...... 30 3.1 Interferometric SAR for DEM generation ...... 30 3.1.1 Introduction of InSAR ...... 32 3.1.1.1 Conditions for interferometry ...... 35 3.1.1.2 Critical baseline of InSAR ...... 36 3.2 InSAR DEM processing ...... 37 3.2.1 Raw image ...... 40 3.2.2 Coregistration ...... 41 3.2.3 Coherence ...... 42 3.2.4 Interferogram ...... 44 3.2.5 Altitude of ambiguities ...... 45 3.2.6 Phase unwrapping ...... 45 3.2.7 Phase to height ...... 46 3.2.8 Geocoding ...... 47 3.3 Sources of InSAR noise ...... 48 3.3.1 Decorrelation ...... 49 3.3.2 Thermal noise and Processing error ...... 51 3.3.3 Atmospheric Noise ...... 53 3.3.4 Orbit and Geometry error ...... 54

CHAPTER 4 C- & L-BAND InSAR DEM GENERATION ...... 58 4.1 C-band InSAR DEM Generation ...... 59 4.2 L-band InSAR DEM Generation ...... 67 4.3 Assessment of different wavelength InSAR data DEM generation ...... 75 4.4 Ascending and Descending InSAR generated DEM fusion ...... 76 4.4.1 Terrain Response Differences for Ascending and Descending SAR data .... 78 4.4.2 Merging DEMs from Ascending and Descending InSAR DEM ...... 85 4.5 Surface deformation exclusion in InSAR DEM using differential interferogram 91 4.5.1 Surface deformation detection using differential InSAR ...... 92 4.5.2 Area detection of surface deformation ...... 93 4.5.3 InSAR DEM generation without surface deformation ...... 97 4.6 SRTM DEM updating using selected InSAR DEM based on coherence value selection method ...... 104 4.6.2 Elevation updating using selected InSAR DEM height value ...... 106

ix 4.6 Concluding remarks ...... 113

CHAPTER 5 MULTI-PASS InSAR DEM GENERATION ...... 115 5.1 SAR data information ...... 116 5.1.1 Meta-data information at raw SAR data ...... 117 5.1.2 Relationship of DEM grid location and Orbit information ...... 119 5.2 Multi-pass InSAR DEM generation ...... 121 5.2.1 Multi-pass InSAR DEM processing using single master image ...... 123 5.2.2 Multi-pass InSAR DEM processing using different master image ...... 130 5.3 Concluding remarks ...... 141

CHAPTER 6 RADARGRAMMETRY FOR DEM GENERATION ..... 143 6.1 Geometry of Radargrammetry ...... 146 6.1.1 Perpendicular radargrammetry ...... 149 6.1.2 Oblique radargrammetry ...... 150 6.1.3 Same-side stereo ...... 151 6.1.4 Opposite-look stereo ...... 152 6.2 Radargrammetry DEM generation using different orbit SAR images ...... 153 6.3 Concluding remarks ...... 166

CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND

DIFFERENT SOURCES ...... 167 7.1 Relative accuracy ...... 167 7.2 Absolute accuracy ...... 168 7.3 Different sources for DEM generation ...... 179 7.3.1 Optical photogrammetry ...... 180 7.3.2 Airborne laser scanning ...... 182 7.3.3 Topographic maps ...... 183 7.4 Concluding remarks ...... 187

CHAPTER 8 CONCLUSIONS AND DISCUSSIONS ...... 188

8.1 Conclusions ...... 188 8.1.1 InSAR DEM accuracy improvement techniques in single-baseline observation ...... 188 x 8.1.2 InSAR DEM accuracy improvement techniques in multi-baseline observation ...... 190 8.1.3 SAR DEM generation using intensity image ...... 192 8.1.4 Product quality assessment and other DEM generation sources ...... 193 8.2 Future research ...... 193

REFERENCES ...... 196

xi LIST OF FIGURES

Figure 1. 1 The research area in New South Wales, Australia ...... 7 Figure 1. 2 A panoramic view of the Wollongong research area (left-Wollongong’s northern suburbs, right-Wollongong’s southern suburbs) ...... 8 Figure 2. 1 The different imaging geometry of ALOS/PALSAR (EORC, JAXA) ...... 12 Figure 2. 2 The imaging modes of RADARSAT (CSA) ...... 13 Figure 2. 3 The Airborne SAR system (E-SAR) ...... 14 Figure 2. 4 Airborne SAR system (UAVSAR, NASA) ...... 15 Figure 2. 5 Artist’s view of airborne F-SAR system (DLR) ...... 16 Figure 2. 6 The image of ERS satellite (ESA) ...... 17 Figure 2. 7 The RASARSAT satellite (Canadian Space Agency) ...... 18 Figure 2. 8 The image of ALOS satellite (JAXA) ...... 19 Figure 2. 9 The image of TerraSAR-X satellite (DLR) ...... 19 Figure 2. 10 The mapping sequence of the SRTM...... 21 Figure 2. 11 The geometry of radarclinometry (Toutin and Gray 2000) ...... 25 Figure 2. 12 The geometry of interferometry...... 26 Figure 2. 13 The geometry of ...... 27 Figure 2. 14 The geometry of polarimetry ...... 28 Figure 3. 1 Geometry of repeat pass interferometric configuration (Hanssen 2001) .... 33

Figure 3. 2 Coherence image of ALOS/PALSAR ...... 36

Figure 3. 3 The flow chart of InSAR DEM generation...... 40

Figure 3. 4 The intensity image of ALOS/PALSAR ...... 41

Figure 3. 5 Geometry of phase difference due to the height difference with an identical slant range (left) and geometry of phase difference due to the slant range difference with an identical height (right) (Liao et al. 2007)...... 53

Figure 3. 6 The different determination of baselines between sensors ...... 55

Figure 3. 7 The Geometrical limitations of SAR images (Layover, Foreshortening and Shadowing) ...... 55

Figure 3. 8 The geometry of orbit errors ...... 57

Figure 4. 1 The waves and frequency ranges used by radar (radartutorial) ...... 58

xii Figure 4. 2 The research areas in Appin (left) and Wollongong (right), NSW, Australia ...... 61 Figure 4. 3 Height ambiguity due to phase changes of C-band...... 62 Figure 4. 4 C-band coherence (29/10/95-30/10/95-left) and C-band coherence (03/12/95-04/12/95-right) images from ERS Tandem ...... 63 Figure 4. 5 C-band InSAR generated DEM (29/10/95-30/10/95-left) and C-band InSAR generated DEM (03/12/95-04/12/95-right) using ERS Tandem ...... 64 Figure 4. 6 The InSAR DEM and DEM errors (red box) surrounding the coastline in Wollongong (29/10/95-30/10/95) ...... 64 Figure 4. 7 The profile lines of the test area in Appin and Wollongong...... 65 Figure 4. 8 The profile lines (No.1) of C-band InSAR generated DEM in Appin ...... 65 Figure 4. 9 The profile lines (No. 2) of C-band InSAR generated DEM in Wollongong ...... 66 Figure 4. 10 The research area in Appin, NSW, Australia ...... 68 Figure 4. 11 Height ambiguity due to phase changes of L-band...... 69 Figure 4. 12 L-band coherence (14/11/07-30/12/07-left) and L-band coherence (14/02/08-31/03/08-right) images from ALOS/PALSAR ...... 70 Figure 4. 13 L-band InSAR generated DEM (14/11/07-30/12/07-left) and L-band InSAR generated DEM (14/02/08-31/03/08-right) using ALOS/PALSAR ...... 70 Figure 4. 14 Large coverage L-band InSAR generated DEM (31/03/08-01/07/08) ...... 71 Figure 4. 15 The profile lines in Appin ...... 72 Figure 4. 16 The profile (No.1) of L-band InSAR generated DEM in Appin ...... 72 Figure 4. 17 The profile (No.2) of L-band InSAR generated DEM in Appin ...... 73 Figure 4. 18 GPS RTK Surveying points in the Appin area ...... 74 Figure 4. 19 The elevation comparison of C and L-band generated DEMs ...... 75 Figure 4. 20 Geometry of ascending and descending SAR imaging...... 77 Figure 4. 21 Different orbits of ascending and descending SAR sensors ...... 77 Figure 4. 22 The intensity images of ascending (14/11/2007: left) and descending (17/07/2008: right) orbits in test area ...... 79 Figure 4. 23 Boundaries of ascending and descending pairs in the Appin area ...... 79 Figure 4. 24 ALOS/PALSAR ascending InSAR generated DEMs (14/11/07-30/12/07: left and 31/03/08/-01/07/08: right) ...... 80

xiii Figure 4. 25 ALOS/PALSAR descending InSAR generated DEMs (01/06/08-17/07/08: left and 17/07/08-01/09/08: right) ...... 80 Figure 4. 26 The profile lines of ascending and descending ALOS/PALSAR ...... 81 Figure 4. 27 Profile line (No.1) of ascending and descending ALOS/PALSAR ...... 82 Figure 4. 28 Profile line (No.2) of ascending and descending ALOS/PALSAR ...... 82 Figure 4. 29 Elevation differences of LiDAR DEM-Ascending (14/11/07-30/12/07) DEM ...... 83 Figure 4. 30 Elevation differences of LiDAR DEM-Ascending (31/03/08-01/07/08) DEM ...... 83 Figure 4. 31 Elevation differences of LiDAR DEM-Descending (01/06/08-17/07/08) DEM ...... 84 Figure 4. 32 Elevation differences of LiDAR DEM-Descending (17/07/08-01/09/08) DEM ...... 84 Figure 4. 33 The merged InSAR generated DEM using mean method from Ascending (14/11/07-30/12/07 and 31/03/08/-01/07/08) and Descending (01/06/08-17/07/08 and 17/07/08-01/09/08) DEMs ...... 86 Figure 4. 34 The merged InSAR generated DEM using baseline weight mean method from Ascending (14/11/07-30/12/07 and 31/03/08/-01/07/08) and Descending (01/06/08-17/07/08 and 17/07/08-01/09/08) DEMs ...... 87 Figure 4. 35 The merged InSAR generated DEM using coherence weight mean method from Ascending (14/11/07-30/12/07 and 31/03/08/-01/07/08) and Descending (01/06/08-17/07/08 and 17/07/08-01/09/08) DEMs ...... 87 Figure 4. 36 The part of SRTM DEM (left) and mean merged DEM (right) in test area ...... 88 Figure 4. 37 The part of baseline weight merged DEM (left) and coherence weight merged DEM (right) ...... 88 Figure 4. 38 Descending (left) and ascending (right) InSAR DEMs in test area...... 89 Figure 4. 39 Elevation differences of LiDAR DEM-Mean merged DEM ...... 89 Figure 4. 40 Elevation differences of LiDAR DEM-Baseline weight merged DEM .... 90 Figure 4. 41 Elevation differences of LiDAR DEM-Coherence weight merged DEM . 90 Figure 4. 42 Examples of differential interferogram of different pairs in DInSAR (14/11/07-30/12/07: left and 14/11/07-14/02/08: right). The relative phase value is

xiv changed between –pi to +pi and the differential interferogram illustrates the relative phase value of each interferogram...... 93 Figure 4. 43 Window of 16-neighborhood pixels for deformation detection in differential interferogram...... 94 Figure 4. 44 Histograms of number of pixels (14/11/07-30/12/07: left) and (14/11/07- 14/02/08: right) ...... 95 Figure 4. 45 The mask of a deformation area from DInSAR...... 96 Figure 4. 46 Flowchart of deformation exclusion of InSAR DEM using DInSAR ...... 97 Figure 4. 47 Height ambiguities due to the phase change and perpendicular baseline .. 98 Figure 4. 48 Subsidence map generated by DInSAR processing (left) and InSAR- generated DEM (right) at the Appin test site ...... 99 Figure 4. 49 InSAR-generated DEM and DEM with deformation removal (14/11/07- 30/12/07) ...... 100 Figure 4. 50 InSAR-generated DEM and DEM with deformation removal (14/11/07- 14/02/08) ...... 100 Figure 4. 51 Profile lines (No.1 and 2) on the ground deformation map at the test area ...... 101 Figure 4. 52 Profiles (No.1) of InSAR-generated DEM and ground deformation removed InSAR-generated DEM compared with SRTM DEM: (14/11/07-30/12/07) 102 Figure 4. 53 Profiles (No.1) of InSAR-generated DEM and ground deformation removed InSAR-generated DEM compared with SRTM DEM: (14/11/07-14/02/08) 102 Figure 4. 54 Profiles (No.2) of InSAR-generated DEM and ground deformation removed InSAR-generated DEM compared with SRTM DEM: (14/11/07-14/02/08) 103 Figure 4. 55 Profiles (No.2) of InSAR-generated DEM and ground deformation removed InSAR-generated DEM compared with SRTM DEM: (14/11/07-30/12/07) 103 Figure 4. 56 Flowchart of SRTM DEM updating using selected elevation ...... 106 Figure 4. 57 A selected coherence image (left) and a selected elevation (right) from InSAR generated DEM using threshold value: 0.2 ...... 107 Figure 4. 58 A selected coherence image (left) and a selected elevation (right) from InSAR generated DEM using threshold value: 0.5 ...... 107 Figure 4. 59 A selected coherence image (left) and a selected elevation (right) from InSAR generated DEM using threshold value: 0.7 ...... 108

xv Figure 4. 60 SRTM DEM (left) and updated SRTM DEM (right) using selected elevation: threshold 0.2 ...... 109 Figure 4. 61 The updated SRTM DEM using selected elevation: threshold 0.3 (left) and updated SRTM DEM using selected elevation: threshold 0.4 (right) ...... 109 Figure 4. 62 The updated SRTM DEM using selected elevation: threshold 0.5 (left) and updated SRTM DEM using selected elevation: threshold 0.6 (right) ...... 110 Figure 4. 63 The updated SRTM DEM using selected elevation: threshold 0.7 (left) and updated SRTM DEM using selected elevation: threshold 0.8 (right) ...... 110 Figure 4. 64 The SRTM DEM (left) and updated SRTM DEM (right) using selected elevation from ALOS (14/02/08-31/03/08): threshold 0.5 ...... 112 Figure 5. 1 The acquisition area by ALOS in the World (ERSDAC) ...... 116 Figure 5. 2 Scene definition of single beam ...... 117 Figure 5. 3 The geometry of Doppler centre ...... 120 Figure 5. 4 The geometry of multi-pass InSAR ...... 122 Figure 5. 5 Grid location relationship of multi-pass InSAR ...... 123 Figure 5. 6 The test area boundary for identicla master multi-pass InSAR DEM generation ...... 124 Figure 5. 7 The same master InSAR generated DEM (14/11/07-27/12/06: left and 14/11/07-11/02/07: right) ...... 125 Figure 5. 8 The same master InSAR generated DEM (14/11/07-30/12/07: left and 14/11/07-14/02/08: right) ...... 125 Figure 5. 9 The same master InSAR generated DEM (14/11/07-31/03/08) ...... 126 Figure 5. 10 The mean merged InSAR generated DEM using same master multi-pass method ...... 127 Figure 5. 11 The mean merged InSAR generated DEM using (14/11/07-30/12/07 and 14/11/07-14/02/08: left) pairs and (14/11/07-11/02/07 and 14/11/07-31/03/08: right) pairs ...... 127 Figure 5. 12 The merged InSAR generated DEM using coherence value weight ...... 128 Figure 5. 13 The merged InSAR generated DEM using baseline distance weight ...... 128 Figure 5. 14 The elevation differences between LiDAR DEM and Mean merged InSAR generated DEM ...... 129 Figure 5. 15 The elevation differences between LiDAR DEM and coherence weight InSAR generated DEM ...... 129

xvi Figure 5. 16 The elevation differences between LiDAR DEM and baseline weight InSAR generated DEM ...... 130 Figure 5. 17 Sketch of SRTM and InSAR elevation interpolation ...... 132 Figure 5. 18 ERS-1 Master (29/10/09) and Slave (30/10/09) images ...... 133 Figure 5. 19 ALOS-PALSAR Master (14/11/07) and Slave (30/12/07) images ...... 133 Figure 5. 20 Height ambiguity due to phase changes...... 134 Figure 5. 21 Grid points in different InSAR DEMs ...... 135 Figure 5. 22 The research areas in Appin and Wollongong, NSW, Australia ...... 135 Figure 5. 23 ALOS/PALSAR InSAR generated DEM (14/02/08-31/03/08; left) and (14/11/07-30/12/07; right) ...... 136 Figure 5. 24 ALOS/PALSAR InSAR generated DEM (27/12/06-11/02/07, left) and (31/03/08-01/07/08, right) ...... 136 Figure 5. 25 ERS InSAR generated DEM (29/10/95-30/10/95; left) and (03/12/95- 04/12/95; right) ...... 137 Figure 5. 26 Merged InSAR generated DEM using mean value from a different master ...... 138 Figure 5. 27 Merged InSAR generated DEM using coherence weight method from a different master ...... 138 Figure 5. 28 Merged InSAR generated DEM using baseline distance weight method from a different master ...... 138 Figure 5. 29 The elevation differences between LiDAR DEM and Mean merged InSAR generated DEM ...... 139 Figure 5. 30 The elevation differences between LiDAR DEM and coherence weight method merged InSAR generated DEM ...... 140 Figure 5. 31 The elevation differences between LiDAR DEM and baseline weight method merged InSAR generated DEM ...... 140 Figure 6. 1 The difference between terrain response between photogrammetry and radargrammetry ...... 144 Figure 6. 2 Examples of different terrain responses and projections between optical and radar images: optical images (left: ) and TerraSAR-X intensity image (right) in Tokyo, Japan ...... 145 Figure 6. 3 The geometric projection differences of radar system (left) and photogrammetry system (right) ...... 146

xvii Figure 6. 4 Radargrammetry geometry of look angle difference and image overlapping ...... 147 Figure 6. 5 The different observation positions and geometry for radargrammetry (Maitre 2008) ...... 148 Figure 6. 6 Principles of perpendicular radargrammetry (Massonnet and Souyris 2008) ...... 150 Figure 6. 7 The geometry of same side direction of radargrammetry ...... 151 Figure 6. 8 The parallax difference in same side radargrammetry ...... 152 Figure 6. 9 The geometry of opposite side direction of radargrammetry ...... 153 Figure 6. 10 The parallax difference in opposite side radargrammetry ...... 153 Figure 6. 11 Flow chart of radargrammetry processing ...... 154 Figure 6. 12 The radargrammetry intensity images of 31/03/08-05/04/08 radargrammetry pair (31/03/08: reference image-left, 05/04/08: matching image-right) ...... 155 Figure 6. 13 The coregistration processing of reference image and matching image in radargrammetry ...... 156 Figure 6. 14 The correlation image between a reference image and a match image after the matching process ...... 156 Figure 6. 15 The radargrammetry DEM generated from ALOS/PALSAR (31/03/08- 05/04/08: left) and (01/07/08-06/07/08: right) ...... 157 Figure 6. 16 The radargrammetry DEM generated from ALOS/PALSAR (07/04/08- 05/04/08: left) and (23/05/08-21/05/08: right) ...... 158 Figure 6. 17 Radargrammetry DEM generated from /ASAR (08/02/10-05/02/10: left) and (15/03/10-12/03/10: right) ...... 158 Figure 6. 18 Radargrammetry DEM generated from Envisat/ASAR (02/04/10-31/10/09: left, 18/12/09-26/09/09: right) ...... 159 Figure 6. 19 The elevation differences between LiDAR DEM and ALOS InSAR DEM (14/02/08-31/03/08) ...... 160 Figure 6. 20 The elevation differences between LiDAR DEM and ERS InSAR DEM (03/12/95-04/12/95) ...... 161 Figure 6. 21 The elevation differences between LiDAR DEM and Envisat radargrammetry DEM (08/02/10-05/02/10) ...... 161

xviii Figure 6. 22 The elevation differences between LiDAR DEM and Envisat radargrammetry DEM (15/03/10-12/03/10) ...... 162 Figure 6. 23 The elevation differences between LiDAR DEM and Envisat radargrammetry DEM (02/04/10-31/10/09) ...... 162 Figure 6. 24 The elevation differences between LiDAR DEM and Envisat radargrammetry DEM (18/12/09-26/09/09) ...... 163 Figure 6. 25 The elevation differences between LiDAR DEM and ALOS radargrammetry DEM (31/03/08-05/04/08) ...... 163 Figure 6. 26 The elevation differences between LiDAR DEM and ALOS radargrammetry DEM (07/04/08-05/04/08) ...... 164 Figure 6. 27 The elevation differences between LiDAR DEM and ALOS radargrammetry DEM (23/05/08-21/05/08) ...... 164 Figure 6. 28 The elevation differences between LiDAR DEM and ALOS radargrammetry DEM (01/07/08-06/07/08) ...... 165 Figure 7. 1 The profile line for comparing DEMs generation using SAR techniques . 169 Figure 7. 2 The profiles of DEM generation using Radargrammetry, C-band InSAR(ERS), L-band InSAR (ALOS) and same mater multi-pass InSAR (ALOS) ... 170 Figure 7. 3 The profiles of DEM generation using different master multi-pass InSAR (C, L-band) and ascending and descending InSAR (ALOS) ...... 170 Figure 7. 4 Elevation differences between SRTM and ERS InSAR DEM (03/12/95- 04/12/95) ...... 172 Figure 7. 5 Elevation differences between SRTM and ERS merged DEM (different master) ...... 172 Figure 7. 6 Elevation differences between SRTM and ALOS InSAR DEM (14/02/08- 31/03/08) ...... 173 Figure 7. 7 Elevation differences between SRTM and ALOS merged InSAR DEM (same master) ...... 173 Figure 7. 8 Elevation differences between SRTM and ALOS merged InSAR DEM (different master) ...... 174 Figure 7. 9 Elevation differences between SRTM and ALOS merged InSAR DEM (Ascending and Descending) ...... 174 Figure 7. 10 Elevation differences between SRTM and Radargrammetry (01/07/08- 06/07/08) ...... 175

xix Figure 7. 11 Elevation comparison between SRTM and ERS InSAR DEM (03/12/95- 04/12/95); Red line: Linear least square, Blue line; extension line ...... 176 Figure 7. 12 Elevation comparison between SRTM and ERS merged DEM (different master); Red line: Linear least square, Blue line; extension line ...... 176 Figure 7. 13 Elevation comparison between SRTM and ALOS InSAR DEM (14/02/08- 31/03/08); Red line: Linear least square, Blue line; extension line ...... 177 Figure 7. 14 Elevation comparison between SRTM and ALOS merged InSAR DEM (same master); Red line: Linear least square, Blue line; extension line ...... 177 Figure 7. 15 Elevation comparison between SRTM and ALOS merged InSAR DEM (different master); Red line: Linear least square, Blue line; extension line ...... 178 Figure 7. 16 Elevation comparison between SRTM and ALOS merged InSAR DEM (Ascending and Descending); Red line: Linear least square, Blue line; extension line ...... 178 Figure 7. 17 Elevation comparison between SRTM and Radargrammetry (01/07/08- 06/07/08); Red line: Linear least square, Blue line; extension line ...... 179 Figure 7. 18 Stereo geometric condition of an optical system ...... 180 Figure 7. 19 IKONOS and SPOT-5 satellite image data collection (SIC) ...... 181 Figure 7. 20 ALOS/PRISM forward mosaic image in Appin NSW, Australia ...... 181 Figure 7. 21 The airborne LiDAR system (USGS) ...... 183

xx LIST OF TABLES

Table 2. 1 Wavelengths and frequencies of remote sensing signals in microwave ...... 11 Table 2. 2 F-SAR technical characteristics (Horn et al. 2008) ...... 16 Table 2. 3 The characteristics of the space-borne and airborne SAR systems ...... 22 Table 3. 1 A comparison of critical baseline of radar from Sandwell et al. report (2008) ...... 37 Table 3. 2 Parameters influencing the InSAR data (Crosetto 2002; Kyaruzi 2005) ...... 48 Table 4. 1 Spacing and resolution of radar images (unit: m) ...... 59 Table 4. 2 The characters of ERS system ...... 60 Table 4. 3 ERS data pair information...... 60 Table 4. 4 Altitude of ambiguity for each data pair ...... 61 Table 4. 5 Height ambiguity due to phase changes of C-band...... 62 Table 4. 6 The Root Mean Square Error of C-band InSAR DEMs ...... 66 Table 4. 7 The characters of the ALOS/PALSAR system ...... 67 Table 4. 8 ALOS/PALSAR Data pair information ...... 68 Table 4. 9 Altitude of ambiguity for data pairs ...... 68 Table 4. 10 Height ambiguity due to phase changes of L-band...... 69 Table 4. 11 The Root Mean Square Error of L-band InSAR DEM ...... 73 Table 4. 12 The Root Mean Square Error of L-band InSAR DEM compared with GPS measurement...... 74 Table 4. 13 The Root Mean Square Error of different bands InSAR DEMs ...... 76 Table 4. 14 The state vector of ascending and descending images ...... 78 Table 4. 15 ALOS/PALSAR data information ...... 79 Table 4. 16 The Root Mean Square Error of merged DEM ...... 89 Table 4. 17 The Root Mean Square Error of merged DEM compare with GPS measurements...... 91 Table 4. 18 ALOS/PALSAR image information ...... 98 Table 4. 19 Height ambiguities due to the phase change and perpendicular baselines . 99 Table 4. 20 The information of ALOS/PALSAR Data ...... 106 Table 4. 21 The RMSE of SRTM, update SRTM by different coherence values and InSAR DEM (ALOS: 31/03/08-01/07/08) with LiDAR DEM...... 111 Table 4. 22 The RMSE of SRTM, update SRTM by different coherence values and InSAR DEM (ALOS: 14/02/08-31/03/08) with LiDAR DEM...... 111 xxi Table 4. 23 The RMSE of SRTM DEM, InSAR DEM, update SRTM DEM and merged SRTM DEM capering with LiDAR DEM ...... 112 Table 5. 1 Meta-data of raw SAR ...... 118 Table 5. 2 ALOS/PALSAR Data pairs ...... 124 Table 5. 3 The Root Mean Square Error of merged DEMs ...... 129 Table 5. 4 The root mean square error of identical master InSAR DEMs compared with GPS measurement...... 130 Table 5. 5 ALOS/PALSAR and ERS Data ...... 133 Table 5. 6 Altitude of ambiguity at each DEM pairs ...... 134 Table 5. 7 Height ambiguity due to phase changes ...... 134 Table 5. 8 The Root Mean Square Error of merged DEM ...... 139 Table 5. 9 The Root Mean Square Error of different master InSAR DEMs compared with GPS measurement ...... 141 Table 6. 1 The information of radargrammetry images ...... 155 Table 6. 2 The Root Mean Square Error of radargrammetry DEMs ...... 160 Table 6. 3 The Root Mean Square Error of radargrammetry DEMs compared with GPS ...... 165 Table 7. 1 The Root Mean Square Error of SAR generated DEMs ...... 171 Table 7. 2 Satellite data prices of radar and optical images (GA, SOPT IMAGE) ..... 182 Table 7. 3 Topographic maps at different scales (Li et al. 2004) ...... 184 Table 7. 4 Map scales and commonly used contour intervals (Li et al. 2004) ...... 184 Table 7. 5 Comparison of InSAR, ALS and Photogrammetry for Digital Elevation Model generation ...... 185

xxii CHAPTER 1 INTRODUCTION

1.1 Introduction

Geodesy is a scientific discipline that deals with the measurement and representation of the Earth’s surface in a three-dimensional, time-varying space. A topographic map is essential for the thematic mapping of geological and environmental information.

A Digital Elevation Model (DEM) is defined as “any digital representation of the continuous variation of relief over space” (Burrough 1986). A DEM is a representation of terrain elevation at regularly spaced grid intervals. DEMs are often used in three- dimensional graphics applications such as to display the slope, aspect, and terrain profile between selected positions. Hence, DEMs have become a vital source of topographical data for scientific investigations hydrological, geological, infrastructure planning and environmental applications. In most geological applications it replaces or complements traditional data sources and formats like conventional hardcopy maps. A

DEM is a digital file consisting of terrain elevations for ground positions at regularly spaced horizontal intervals. Generally, DEMs are generated using elevation data derived from existing contour maps at varying scales ranging from 1:25,000 to

1:100,000. In addition, digital stereo capture provides a terrain surface representation with a horizontal resolution of 20 to 50 (Land-Victoria 2002). For some areas of the world where topographic data is unavailable (such as in significant parts of the tropical and equatorial zones) DEM from remotely sensed data can be the main source of information (Bourgine and Baghdadi 2005; Walker and Willgoose 1999).

DEMs can be generated using different techniques and different data sources, including ground surveys, levelling, photogrammetry, optical and radar remote sensing,

1 CHAPTER 1 INTRODUCTION and laser scanning. In particular, satellite remote sensing techniques provide an efficient, regular schedule and cost-effective means of acquiring relatively accurate topographic information over extensive areas of terrain (Nitti et al. 2009). This method of DEM generation offers additional advantages, including its utility in difficult environments such as pre- or post-earthquake events, water resources, volcano hazard studies and ice cap measurement (Dall et al, 2000; d’Ozouville et al. 2008; Hirano et al.

2003; Jenson 1991; Moore et al. 2006; Wang et al. 2007).

Remote sensing techniques can be an efficient approach to acquire and process elevation information over an extensive terrain. Generally, DEMs have been generated from optical stereo or aerial stereo photographs. Furthermore, stereo- pairs of mixed sensors are suitable for elevation extraction if a suitable base-height ratio is exists (Eckert et al, 2005). However, optical systems have operation limitation depending on the sun illumination and weather conditions (such as cloud, smog, and storm). Radar systems (active systems) can operate in almost all weather conditions, as well as during darkness using their own illumination.

Radar was investigated by A.H. Taylor and L.C. Young in 1922 (Wilkinson 1987).

Side Looking Airborne Radar (SLAR), which has a continuous strip mapping capability, was developed in the 1950s. SLAR has two major types: real aperture radar and synthetic aperture radar, where the word “aperture” means antenna. Real aperture radar uses a fixed length antenna ranging in length from 1~2m. Synthetic aperture radar also uses similar length antenna however it can synthesize a much longer antenna which has an improved resolution. Radio interferometry was developed after World

War Two. Real and synthetic aperture SLAR were used for resource reconnaissance

2 CHAPTER 1 INTRODUCTION after the 1960s, (Jensen 2007). The first airborne radar interferometry experiment for topographic mapping was conducted by the U.S. military and a patent for obtaining elevation from phase difference images was filed in 1971 (Hanssen 2001).

In radar remote sensing systems, DEM extraction methods are based on the analysis of Synthetic Aperture Radar (SAR) images, and include: (1) interferometry, (2) radargrammetry, (3) radarclinometry, and (4) polarimetry. The two most common methods for generating DEMs from SAR images are: (1) radargrammetry, a technique derived from photogrammetry, based on the stereoscopic principle; (2) interferometry, a technique based on the phase differences between identical imaged points in two SAR images. Interferometric Synthetic Aperture Radar (InSAR) is used in many applications, including topographic surveys, estimating ocean currents, land monitoring, ground deformation, detecting and locating moving targets, snow-covered terrain (Dall et al.

2000; Ge et al. 2007; Hongda et al, 2002; Janssen et al. 2004). Furthermore, the various InSAR techniques operated for improve DEM accuracy and reconstruct elevation such as multi-pass imagery, multi-channel reconstruction and stereo technology (Baselice et al. 2009; Ferretti et al. 1997; Nonaka et al. 2009).

InSAR DEM generation is an interesting method in terms of advantages of the

InSAR, which has high spatial resolution (up to 10m), height precision, with a residual error generally less than 10m, automated processing, and produces high quality results over remote or not easily-accessible regions at a low cost per square kilometre.

However, the main limitations of InSAR generated DEM accuracy is the problematic height estimation from severely sloped and forested areas, where low coherence

3 CHAPTER 1 INTRODUCTION condition and layover effects are present (Pizzi and Pugliese 2004). To supersede the limitation of InSAR, the Radargrammetric SAR can be used to DEM generation.

Radargrammetry is based on stereogrammetry, which is the classical method for relief reconstruction using remotely sensed images (Paillou and Gelautz 1999). One advantage of radargrammetry is that it remains less affected by atmospheric influences compared with interferometry. The effect of the atmosphere on SAR imagery is essentially the same in either radargrammetry or in the InSAR. However, radargrammetry utilises the magnitude value, as opposed to the InSAR which uses the phase value. As magnitude is less affected by the atmosphere than the phase value, radargrammetry is more robust against atmospheric influences than InSAR (Massonnet and Souyris 2008).

1.2 Motivation and Objectives

The objectives of this research encompass quality improvement and DEM error reduction, which are generated by various SAR techniques and imagery.

Repeat-pass satellite SAR is especially well suited for cost-effective, precise and large-coverage surfaces. The orbits of SAR satellites are such that they revisit a similar path typically every 11~46 days. InSAR is a technique for extracting three-dimensional information on the terrain topography from the phase interferogram of two SAR data images. InSAR DEM generation is based on the processing of at least two complex

SAR images covering the same area which are acquired from slightly different points of view (or “look angle”) and at different times and with different perpendicular baselines (i.e. different orbital positions) (Eckert et al. 2005; Li et al. 2007; Rocca

2007).

4 CHAPTER 1 INTRODUCTION

In order to generate a DEM, the SAR data have to undergo several processing stages, such as image registration, interferogram calculation and filtering, phase unwrapping, and phase-to-height transformation. Coordinate information and other parameters related to the InSAR DEMs are determined according to the parameters of the master image, and depending on the wavelength of the SAR image, the terrain responses of

DEM are subsequently changed (Crosetto 2002; Ferretti et al. 2001; Zhou et al. 2005).

To improve the quality of DEMs, multi-pass InSAR DEM generation (DEM reiteration) is used. In normal cases, one fixed master is selected for easy processing of elevation matching and coverage overlap DEMs, which are generated by different pairs.

However, multi-pass and multi-wavelength InSAR DEMs, which are generated with different master images, consist of different grid sizes and locations, even though the coordinate systems are identical. These differences provide diverse terrain information.

Furthermore, each pair of InSAR is made up of a different perpendicular baseline and acquisition conditions such as water vapour delays and measurement noise. In addition, the overlapping of different DEMs increases the terrain detail and reduces vertical errors.

The satellite systems are widely used and have been developed rapidly and applied to remote sensing as a result of improvements in sensor capabilities and vehicle stability. However, a major shortcoming of repeat-pass satellite InSAR DEM processing is the assumption that there is no deformation phase. This shortcoming is due to the fact that the time interval between SAR image acquisitions is comparatively short. Hence, if there are ground deformation signatures in the interferometric fringes

5 CHAPTER 1 INTRODUCTION during InSAR processing, this limits the accuracy of the resulting DEM, the phase unwrapping process, and, ultimately, the representation of terrain reality.

Differential InSAR (DInSAR) is a well-known method of determining ground deformation. DInSAR processing requires the removal of the interferometric phase that topography contributes towards. That is, DInSAR results isolate the area of ground deformation without any topographical effect being recorded. This deformation measurement method has been applied to , earthquakes, and urban and mining subsidence (see, e.g., Amelung et al. 1999; Chang et al. 2008; Ge et al. 2008;

Massonnet and Feigl 1998). The basic processing requirements for DInSAR are essentially the same as for InSAR processing.

The low resolution DEM updating is an interesting topic in DEM research.

Normally, it is processed using a higher resolution DEM for foundation DEM (updated

DEM) updating. Generally, foundation DEMs had large coverage and low spatial resolution. However, for various applications in current research, the DEMs require large coverage and high spatial resolution. Space-borne InSAR generated DEM achieves a higher degree of accuracy than SRTM DEM, and the InSAR generated

DEM, which has a large-coverage and a high resolution, is the most suitable source for elevation updating. However, some incorrect elevations aggravate the elevation accuracy. Thus, prior to the elevation update process, elevation verification or proven elevation selection has to be performed. This thesis focuses on elevation selection based on coherence value. The elevation, which is calculated by InSAR processing, can be improved by a coherence pixel value interlocked with an elevation pixel because the coherence represents the correlation between two SAR images.

6 CHAPTER 1 INTRODUCTION

The author proposes the following methods to improve the accuracy and reality of

DEM through the deployment of various SAR techniques.

(1) Different wavelength InSAR DEM generation and merging

(2) Ascending and descending InSAR DEM generation and fusion

(3) Surface deformation exclusion in InSAR DEM using differential interferogram

(4) Low resolution updating using selective elevation with coherence value

(5) Multi-pass InSAR DEM generation using single master and different masters

(6) Radargrammetric DEM generation

The areas researched in this thesis are Appin and Wollongong, New South Wales,

Australia. Their locations extend from a longitude of 150˚ 31' E to 150˚ 55' E and a latitude of 34˚ 6' S to 34˚ 39' S. The general landscape of the study area includes the mountains, lakes, rivers, grass plains, agricultural fields, steep slopes, urban areas, and the coastline. The local elevation varies between 0m (sea level) to about 800m. Figure

1.1 illustrates the research area.

Figure 1. 1 The research area in New South Wales, Australia

7 CHAPTER 1 INTRODUCTION

Figure 1. 2 A panoramic view of the Wollongong research area (left-Wollongong’s

northern suburbs, right-Wollongong’s southern suburbs)

Figure 1.2 illustrates the landscapes of the Wollongong area. The northern part of the city consists of a narrow urban area located between a forested escarpment and the

Pacific Ocean. The southern part of the city is urbanised. Lake Illawarra occupies a large section of the southern suburbs.

To conduct this study, the following software programs were used: ArcGIS, which is a common software package with GIS capabilities and simple image processing tools;

MATLAB, which provides a numerical computing method and fourth-generation programming language; ERDAS Imagine 9.3, a remote sensing analytical tool with spatial modelling creation and CRC-SI DEM generation package, which was developed by Geodesy and Earth Observing Systems Group (GEOS), a research group supported by the Cooperative Research Centre for Spatial Information (CRC-SI). This package has been used to process all SAR interferometric data in this thesis.

The Shuttle Radar Topography Mission (SRTM) DEM (Farr et al. 2007) was used as external DEM for the InSAR DEM generation process, airborne LiDAR DEM and

Global Positioning System (GPS) surveying data was used as a reference data for data

8 CHAPTER 1 INTRODUCTION assessment. LiDAR DEM is generated and verified by Land and Property Management

Authority (LPMA) and GPS surveying is performed by UNSW.

1.3 Overview

This thesis consists of eight chapters. Basics of radar and SAR imagery are reviewed in Chapter Two. This includes a brief overview of imagery acquisition modes and platforms, and the various geometries of SAR systems.

Chapter Three discusses the principle of InSAR technique, InSAR DEM generation processes and sources of noise which affect InSAR processing.

Chapter Four discusses the InSAR DEM generation and DEM merging using different wavelengths (C-, L-band) and imagery from different observation directions

(ascending and descending). Also, elevation errors due to the ground deformation are discussed, which is one of the major shortcomings of InSAR DEM generation, exclusion method using DInSAR. And low resolution DEM updating method using selected elevation value.

Chapter Five discusses multi-baseline and multi-temporal imagery for InSAR DEM generation and merging DEMs.

Chapter Six discusses the radargrammetry technique which uses the intensity value of SAR imagery, whereas the aforementioned techniques use the phase value of SAR imagery.

Chapter Seven discusses product quality assessment and different remote sensing sources for DEM generation.

Chapter Eight concludes the study.

9 CHAPTER 2 RADAR AND SYNTHETIC APERTURE RADAR

RADAR an acronym for “Radio Detection And Ranging”, is an active sensor involving the transmission of an electromagnetic signal and recording of the backscattered response from the target or surface of the illuminated terrain (Ahmad

2008). Radar operates in the microwave region of the electromagnetic spectrum. This region generally includes wavelengths ranging from 1-1000mm (Henderson and Lewis

1998). A multitude of RADAR systems have been developed, including the Real

Aperture Radar (RAR), synthetic Aperture Radar (SAR), Side-Looking Radar (SLR), and Active Microwave Imager (AMI). SAR systems employ a short antenna, however, through improved data recording and processing techniques, spatial resolution similar to that of long antennae can be achieved.

Background

Synthetic Aperture Radar

Synthetic Aperture Radar (SAR) is a new microwave sensor developed in the 1950s.

Offering the advantages of consistent multi-temporal imagery acquisition independent of weather conditions and solar illumination, remotely sensed SAR has become an important technique for data acquisition. In particular, remotely sensed SAR has been successfully employed in regions where it is difficult for optical sensors to acquire high-quality imagery such as cloud-prone areas of high precipitation and polar regions.

SAR utilises electromagnetic waves in the microwave portion of the spectrum, which are outlined in Table 2.1. The shortest wavelengths (in K-band) are close to the size of raindrops, especially in tropical regions.

10 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

Consequently, it is more sensitive to interference by atmospheric conditions. Longer wavelengths (in P-band) tend to be delayed by the ionosphere. Accordingly, X- to L- bands are preferred for most space-borne SAR systems (Snoeji et al. 2001).

Table 2. 1 Wavelengths and frequencies of remote sensing signals in microwave Band Wavelength (cm) Frequency (GHz)

K 1.1-1.7 26.5-18.5

X 2.4-3.8 12.5-8

C 3.8-7.5 8-4

S 7.5-15 4-2

L 15-30 2-1

P 30-100 1-0.3

2.1 SAR Imagery

The SAR imaging geometry can be distinguished into three modes depending on the image acquisition mode such as Spotlight, Strip-map and ScanSAR modes.

2.1.1 Spotlight

The antenna in Spotlight mode is continuously steered toward a fixed area on the ground in order to maintain the view of the area over a longer time period.

Theoretically, the SAR azimuth resolution depends on the length of time that the scatterer is illuminated by the radar. A higher resolution in the azimuth direction can be achieved by increasing imaging time for the target region. The SAR spotlight mode is limited in that only selected regions may be imaged in isolation, whereas Strip-map and

ScanSAR have unlimited operation times, if power and data storage issues are excluded.

11 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR Spotlight mode is available on current satellite SAR systems, such as TerraSAR-X,

COSMO-SkyMed and RADARSAT-2 (Gough and Hawkins 1997).

2.1.2 Strip-map

In the Strip-map mode, the antenna is directed perpendicular to the flight path. It provides a continuous imaging swath with standard resolution. The ground swath is illuminated with a continuous sequence of pulses while the antenna beam is fixed in elevation and azimuth. Strip-map is the most commonly utilised imaging mode and all

SAR data illustrated in this dissertation are acquired in this mode.

2.1.3 ScanSAR

ScanSAR has a wider swath coverage in slant range than Strip-map. ScanSAR uses the burst technique to capture large swaths at the expense of the azimuth resolution.

The wider coverage of ScanSAR mode is achieved by switching the antenna look- angles and scanning different swaths (Wiesmann et al. 2006). In the interval between bursts, the look angle of the antenna beam is switched to illuminate a swath parallel to the previous one. The different swaths are then combined to give a much wider beam width than the Strip-map mode. The enormous coverage comes at the cost of reduced resolution due to the burst mode operation.

Figure 2. 1 The different imaging geometry of ALOS/PALSAR (EORC, JAXA) 12 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

Figure 2. 2 The imaging modes of RADARSAT (Imaging notes)

2.2 SAR Platforms for the Digital Elevation Model Generation

Airborne SAR data are preferred for the InSAR DEM generation because it is able to minimise the phases contributed to by land displacement, atmospheric disturbance and decorrelation noise. The Airborne SAR images are acquired simultaneously by two antennae installed at a fixed-distance on the aircraft. Therefore, the spatial decorrelation is very limited. Because there is no time difference between acquisitions of the two images, phase contributions from land displacement and temporal decorrelation can be ignored. Furthermore, since the radar signals travel through the same atmospheric conditions, the atmospheric disturbance is equal in both images. As a result, the atmospheric phase will cancel each other out in the interferogram.

Disadvantages of airborne SAR systems include regular site revisibility, surface coverage and the higher costs associated with arranging each fly-by compared to satellite SAR systems. In the case of space-borne SAR systems, pre-requisite criteria for the selection of suitable SAR images for DEM generation include sensitivity to the variation of terrain, zero or negligible land deformation phases developed between the

13 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR two radar acquisitions and minimum phase noise due to the spatial and temporal decorrelation between the master and slave acquisitions.

2.2.1 Airborne SAR systems for DEM generation

A number of airborne SAR systems have been developed for DEM generation and other applications. The following are three influential examples.

E-SAR

The E-SAR sensor system is a German Aerospace Center (DLR) SAR system which operates on DLR Dornier DO 288 aircraft. This system has four frequency bands (X, C,

L and P-bands). The system is polarimetrically calibrated in the L and P bands.

Interferometric SAR is operated in the X band (DLR).

Figure 2. 3 The Airborne SAR system (E-SAR)

UAVSAR

UAVSAR is designed to acquire airborne (Grumman Gulfstream III) repeat pass

SAR data for differential interferometric measurements from NASA. The radar is fully polarimetric with a range bandwidth of 80MHz, which supports a 16km range swath.

The UAV radar is a compact, pod-mounted polarimetric, L band radar for repeat-pass

14 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR observations. The radar instrument is made up of three major subsystems: the RF electronics subsystems, the digital electronics subsystems, and the antenna subsystems.

(UAVSAR, NASA)

Figure 2. 4 Airborne SAR system (UAVSAR, NASA)

F-SAR

The F-SAR system was created by Deutsches Zentrum für Luft- und Raumfahrt

(DLR) after the success of its former E-SAR system. The system is currently under development at the Microwaves and Radar Institute. The first F-SAR flight was in

November 2006 by DLR. A new development which utilises the most modern hardware and commercial off–the-shelf components, F-SAR is designed to operate in X,

C, S, L, and P bands with full polarimetric capability in all bands. For Earth observation purposes, the radar covers an off-nadir angle range from 25 to 60 degrees at elevations of up to 6000m. A new mount which is designed to fix planar array antennae to the DO-228 aircraft is currently under development. Being fully-fledged in multi-frequency configuration, it holds seven right-looking dual polarised antennae: X-

(3), C-(1), S-(2) and L-band (1).

15 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

Figure 2. 5 Artist’s view of airborne F-SAR system (DLR)

Table 2. 2 F-SAR technical characteristics (Horn et al. 2008) X C S L P RF [GHz] 9.60 5.30 3.25 1.32 0.35 Band width [MHz] 800 400 300 150 100 Range resolution [m] 0.3 0.6 0.75 1.5 2.25 Azimuth resolution [m] 0.2 0.3 0.35 0.4 1.5 8 Bit real; 1GS/500MS selectable; Sampling max number of samples 64K per range line; 4 recoding channels

2.2.2 Space-borne SAR systems for DEM generation

Satellites involved in space-borne SAR for DEM generation can operate in a variety of Earth orbits, including geostationary orbits, polar orbits and inclined orbits. Polar orbit satellites provide a more global observation of Earth, circling at near-polar inclinations at an altitude of 700 to 800km. Most SAR sensors, which are mounted on satellites in polar orbits, look to the right of the satellite.

European Remote Sensing

European Remote Sensing (ERS) satellite 1 and 2, launched respectively on the 17th of July 1991 and 21st of April 1995 were the European Space Agency’s (ESA) first

Earth observation satellites. Both were mounted with SAR sensors, radar altimeters and other powerful instruments to observe ocean surface temperature and winds at sea.

16 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR ERS-1 and 2 operate at a C-band frequency (57mm), 23˚ look angle, 100km swath width, and with a spatial resolution exceeding 30m. After the launch of ERS-2 in 1995, the two satellites cooperated in the first ever ‘Tandem’ mission for duration of nine months. During the mission, the temporal resolution for both satellites was just 24 hours. Radar images acquired in this way can provide high coherence interferograms.

More than 140,000 image pairs are available for InSAR applications (ESA; Rignot and

Zyl 1993).

Tandem Phase

The ERS-2 had the same orbit with a 30-minute offset from ERS-1 from 1995 to

2000. This made it possible to observe the same ground target, flown over by ERS-1 and revisited 24 hours later by ERS-2. During this period, the increased frequency and level of data available provided a unique opportunity to observe changes over a short period of time.

Figure 2. 6 The image of ERS satellite (ESA)

17 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

Figure 2. 7 The RASARSAT satellite (Canadian Space Agency)

RADARSAT

The Canadian Space Agency’s (CSA) RADARSAT (RADAR SATellite) mission was designed to allow the satellite to rotate 180˚ in orbit and to observe Antarctica.

This is based on a radar antennae consisting of eight identical panels in the azimuth direction, each divided into thirty-two sub-antennaes. Seven angle values are available in standard mode. The RASARSAT-2 has left- and right-looking modes to reduce the revisit time by half while doubling the accessibility swath.

Advanced Land Observing Satellite

Advanced Land Observing Satellite (ALOS) Phased Array type L-band Synthetic

Aperture Radar (PALSAR) is an active microwave sensor using L-band frequencies.

ALOS was launched by Japan Aerospace Exploration Agency (JAXA) and placed in a sun-synchronous orbit at 692km elevation, with 98.16° inclination and 46 days of the revisit period. The main advantages of the L-band over the C-band include the strong penetration in areas of vegetation, leading to reduced temporal decorrelation and thus enabling interferograms to have longer temporal baselines. Also, the longer wavelength increases the critical baseline, producing more useful interferometric pairs. PALSAR

18 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR consists of three image acquiring modes: Fine resolution, Polarimetric and ScanSAR models (ALOS Users Handbook; JAXA; Sandwell et al. 2008).

Figure 2. 8 The image of ALOS satellite (JAXA)

TerraSAR-X

TerraSAR-X, the first satellite implemented in a public/private partnership in

Germany, was launched in June, 2007 by DLR. The 5m long and 2.4m wide satellite bus consists of a hexagonal cross-section. One of the six sides mount the 5m by 0.8m radar antenna. The TerraSAR-X is a 1-meter level resolution radar satellite, which delivers Earth observation data for scientific, institutional and commercial use. The satellite consists of an active phased array X-band SAR, which is capable of acquiring imagery in three different modes including Spotlight, StripMap and ScanSAR, and in single, dual and quad polarisation. The satellite was launched into a sun-synchronous orbit at an altitude of 514 km, and provides repeat coverage of any site on the Earth’s surface every eleven days.

Figure 2. 9 The image of TerraSAR-X satellite (DLR)

19 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

2.2.3 The Shuttle Radar Topography Mission

The Shuttle Radar Topography Mission (SRTM) provides virtual global coverage with high resolution digital elevation data using radar images. Launched in February

2000, the mission is as a joint endeavour of NASA, the National Geospatial

Intelligence Agency, and the German and Italian Space Agencies. A SRTM DEM was acquired using C-band (SIR-C) and X-band (X-SAR) InSAR for almost 80 percent of the Earth’s land surface between 60˚N and 56˚S. Two antennae, one located in the shuttle’s cargo bay, the other at the top of a 60m-long mast, simultaneously received reflected radar pulses, NASA’s Jet Propulsion Laboratory (JPL) performed preliminary processing of the raw C-band SRTM data. A 1-arc-second (~30m) DEM and a downgraded 3-arc-second (~90m) DEM were then derived from this data by the

National Geospatial Intelligence Agency (NGA) and its contractors. The 1-arc-second original data have been made available only to the North American public. The X-band

SRTM data was processed by the DLR (Deutches Zentrum für Luft- und Raumfahrt,

Germany) and the ASI (Agenzia Spatiale Italiana). Due to the nature of InSAR technology, one of the major limitations of SRTM is the presence of a non-data area, combined with the interaction of radar energy with the atmosphere and ground targets

(Bourgine and Baghdadi 2005; Falorni et al. 2005; Farr et al. 2007; Rodriguez et al.

2005).

20 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

Figure 2. 10 The mapping sequence of the SRTM.

X-band and C-band InSAR data were acquired over 11 days in February, 2000, between 60° N and 57° S. Gaps are observed in the observation plan due to the 45 km swath width of the X-band SAR (Rabus et al. 2003).

Table 2.3 summarises the characteristics of the space-borne and airborne SAR systems.

21 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

Table 2. 3 The characteristics of the space-borne and airborne SAR systems Wavelength Repeat cycle Organisation Mission Year Band Incidence angle Resolution Polarization Swath Width Altitude (cm) (days) (Country)

Space-borne systems

NASA SEASAT 1978 June L 23.5 3 23˚ 25m×25m HH 100km 800km (USA)

1991- ALMAZ* S 10 - 30˚-60˚ 15m×15-30m HH 30-60km 270km USSR 1992

17 July 1991-10 ESA ERS-1 C 5.67 35 23˚ 26m×30m VV 100km 785km March (EU) 2000 21 April ESA ERS-2 1995- C 5.67 35 23˚ 26m×30m VV 100km 785km (EU) current

11 Feb JAXA JERS-1 1992-12 L 23.5 44 38˚ 18m×18m HH 75km 568km (Japan) Oct 1998

9-20 April 301,5.8, SIR-C* X,C,L - 17˚-63˚ 30m×30m Full 15-90km 225km USA 1994 23.5

4 Nov CSA RASARSAT-1 1995- C 5.77 24 20˚-49˚ 8m×100m HH 45-500km 792km (Canada) current

11 Feb 50km, NASA SRTM* X,C 3,5.6 - 30˚-45˚ 30-90m Dual 233km 2000 225km (USA)

22 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

1 Mar ESA ENVISAT 2002- C 5.67 35 20˚-50˚ 30m Selectable 100-500km 794km (EU) current 24 Jan ALOS/ 40km- JAXA/METI 2006- L 23.5 46 8˚-60˚ 7-100m Selectable 568km PALSAR 350km (Japan) current 7 Jun 2007- COSMO- current SkyMed1 1m×1m 10km 16 3m×3m 9 Dec 30km COSMO- (revisit time 5m×5m ASI 2007- X 3.1 20˚-60˚ Selectable 40km 620km SkyMed2 less than 12 30m×30m (Italy) current 100km hours) 100m× 200km COSMO- 100m 25 Oct SkyMed3 2008- current 15km×10km 15 Jun 1m, 30km×1500 DLR TerraSAR-X 2007- X 3.1 11 15˚-60˚ 3m, Full km 514km (Germany) current 16m 100km×150 0km 14 Dec 20km×500k CSA RADARSAT-2 2007- C 5.55 24 20˚-60˚ 3-100m Full 798km m (Canada) current 3-6m 20 April 30km×240k ISR RISAT-2 C 5.60 25 41˚ 9-12m Dual linear 610km 2009 m (India) 25m, 50m

Air-borne systems

AIRSAR/ C, L, 5.6cm, NASA 1988 - 5m, 10m Full 10-15km 8000m TOPSAR P 24cm, 60cm (USA)

23 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

4.3km- X: Full 11.9km JAXA/NICT Pi-SAR 1997 X, L 3cm - 10˚-75˚ 1.5m, 3m L: 12000m 19.6km- (Japan) polarimetric 42.5km 12km ESA EMISAR 1995 C, L 5.6cm, 24cm - 31˚, 42˚ 2,4 or 8m Full 24km 12500m (EU) 48km 25cm, 30cm, X,C,L 3cm,5cm, 3km×20km DLR E-SAR 1988 - 27˚-60˚ 40cm, selectable ,P 23cm,85cm 5km×20km (Germany) 150cm VV(X), 1-3m, NASA GeoSAR 2000 X, P 3cm, 86cm - HH&HV or 20km 2-5m (US) VV&VH

Rg re: 0.3m, 0.6m, 0.75m, 3cm, 5cm, 1.5m, 2.25m X,C,S DLR F-SAR 2006 9.2cm, - 25˚-55˚ Full 12.5km 6000m , L,P (Germany) 23cm, 85cm Az re: 0.2m, 0.3m, 0.35m, 0.4m, 1.5m

NASA UAVSAR 2007 L 24cm - 25˚-60˚ 10cm 16km 2000-18000m (USA) P~W ONERA RAMSES band 3cm - 30˚~85˚ Full (France) (8) Note: * denotes space shuttle missions.

24 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

2.3 SAR Geometry

Four DEM extraction methods are possible based on SAR images: (1) radarclinometry, (2) interferometry, (3) radargrammetry, and (4) polarimetry (Chen et al. 2007).

2.3.1 Radarclinometry

Radarclinometry is used to build an elevation chart based on SAR imaging and uses a single image unlike interferometry and radargrammetry, which use two images. This technique is based on the shape from shading principles. The main phenomena of SAR images such as shadows, layover and occluded conditions can be used to measure the size and shape of objects. Shadow and layover effect due to the relative position of the sensor position and the illuminated objects on the ground. Depending on the SAR incidence angle, steep terrain can appear to be shadowed. The shadow and layover lengths can be measured from vertical structures, such as buildings, towers and trees.

The usage of clinometry based on SAR data is less obvious due to the sensitivity of shading to the reflected target on the terrain. (Tison et al. 2004; Toutin and Gray 2000)

A1

H R1

ǻșR

Azimuth P h

Range

Figure 2. 11 The geometry of radarclinometry (Toutin and Gray 2000)

25 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR 2.3.2 Interferometry

If two antennae separated by a fixed distance are used, the slight variation in distance between each antenna and the target on the ground can be used to calculate the elevation of the terrain. The principle of interferometry exploits differences between two radar images. The interferometric phase is calculated using two SAR images, which are acquired either by different antennae or by repeat-pass antennae. Phase measurements can be used to observe relative distances as a fraction of the radar wavelength, while the difference in the sensor locations enables observation of angular differences necessary for topographic mapping. The InSAR technique uses multiplicative interferometry. The applications of interferometry are numerous in geophysics, topography, geology, glaciology, , volcanology, , gas or oil exploitation, or the management of mineral resources in general.

A2

Bs A1

R2

R1 H

ǻșI

P Azimuth h

Range

Figure 2. 12 The geometry of interferometry

26 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

2.3.3 Radargrammetry

The stereoscopic methods were first applied to radar images in the 1960s, to derive ground elevation, leading to the development of radargrammetry (La Prade 1963;

Rosenfield 1968). During the 1980s, improvements in SAR systems have allowed the demonstration of stereoscopic radar with same or opposite-side looking. Stereoscopic observation technique improves the interpretation of photographic and radar imagery by creating the illusion of depth which permits the shape and height of objects to be determined (Kaupp et al. 1983). To obtain good geometry for stereoscopy, the intersection angle between the two images should be large in order to increase the stereo-exaggeration factor, or equivalently, the observed parallax. This is then used to calculate the elevation of terrain. The different processing steps to produce DEMs using stereo images can be described in broad terms: (1) acquiring stereo data; (2) stereomodel refinement; (3) image matching and (4) stereo intersection. (Toutin and

Gray 2000; Li et al. 2006)

A1 A2 Bs

R H 2 R1

ǻșs

P Azimuth h

Range

Figure 2. 13 The geometry of stereoscopy

27 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

2.3.4 Polarimetry

Radar polarisation is the orientation of the electric vector in an electromagnetic wave in radar systems. Polarimetric SAR measures the amplitude and phase terms of the complex scattering matrix. Polarimetry’s capacity to detect changes of shape, orientation, and dielectric properties of scatterers has enabled it to be used for thematic classification studies involving both natural terrain and man-made objects. This allows for the identification and separation of scattering mechanisms in natural media employing differences in the polarization signature for purposes of classification and parameter estimation. (Papathanassiou and Cloude 2001; Toutin and Gray 2000).

Polarimetry provides a large amount of scattering information embedded in a Mueller matrix, or equivalently, in a polarimetric covariance matrix (Lee et al. 1994). The range and azimuth directions which are derived from polarimetric SAR data present the terrain slopes. These data are used to generate the elevation information and direct measure of terrain slopes (Chen et al. 2009; Schuler et al. 1996).

A1

H R1

Azimuthal Azimuth Surface normal P

90˚-SA Range

Figure 2. 14 The geometry of polarimetry

28 CHAPTER 2. RADAR AND SYNTHETIC APERTURE RADAR

2.4 Concluding remarks

This chapter provides historical and background information in relation to synthetic aperture radar, imaging modes and acquisition systems. The airborne systems were developed initially stage and have provided the high resolution images while satellite observation systems rapidly increased the number of sensors over the last decade.

These provided the stable and periodic information about Earth. Different data acquisition modes lead to various applications and analysis methods.

The mathematical illustration and more detailed information in relation to interferometry will be provided in the following chapter in order to present concepts applied in this thesis.

29 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION

An interferogram is generated after image registration by comparing the phase difference between images, caused by disparity in position or acquisition time. Hence, interferograms contain information on relative geometry. Interferometric Synthetic

Aperture Radar (InSAR) is a highly potential space technique used to measure the geodetic information of terrain and to map three-dimensional topography with a high degree of spatial resolution and height accuracy. It has been successfully applied to acquire almost global digital elevation models (DEMs) (Zou et al. 2009). The first applications of radar interferometry were in the observation of the and Venus

(Campbell et al. 1970; Shapiro et al. 1972). The first Earth-orbiting SAR satellite

“Seasat” was designed for the remote sensing of the Earth’s oceans and was operated by Jet Propulsion Laboratory (JPL).

When terrain information is acquired using an InSAR system, the three-dimensional object space is projected onto the cylindrical two-dimensional coordinate system of the

InSAR system. The SAR image assumes that images are focused to zero Doppler geometry. In the projected two dimensional SAR image, the first dimension (azimuth) is a scan along the flight direction of the sensor. The second dimension (range) is the radial distance from the flight path of the sensor (Elineder and Adam 2005).

3.1 Interferometric SAR for DEM generation

The use of radar involves measurement in only a single dimension: the range from the radar to a targeted object on the ground. Two-dimensional measurement of targets is possible using a moving platform mounted with radar instruments and measuring location by Doppler frequency shift. This SAR produces two dimensional images 30 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION

which are resolved in a range proportional to the reciprocal of the radar bandwidth, and in an azimuth equal to half the antenna length in the direction of flight. To generate a three-dimensional image, range differences between the two radar images are required and this is realized most accurately and efficiently using principles of interferometry.

The generation of DEM using InSAR was first suggested by Graham in 1974

(Graham 1974) and the first application of this technique took place in 1986 by Zebker and Goldstein (Zebker and Goldstein 1986). Development of this method was made by

Li and Goldstein (1990), Rodriguez and Martin (1992), Zebker et al. (1994a,b,c),

Crosetto (2002), Eckert et al. (2005), Liao et al. (2007), Li et al. (2007), and

Lombardini and Pardini (2008).

SAR data from satellite sensors (e.g ERS-1/2, JERS-1, RADARSAT, ENVISAT,

ALOS, TerraSAR-X and COSMO-SkyMed) are available and many researchers are utilising this data to study InSAR DEM generation (Rocca 2007; Nitti et al. 2009). One of the most important applications of InSAR technology is the generation of DEMs

(Liao 2007). Traditionally, airborne InSAR systems have a sufficient resolution and prompt response to events. However, in global or large coverage DEM reconstruction, a space-borne InSAR system is more efficient (Holecz et al. 1998; Li et al. 2007).

Space-borne InSAR has been increasingly applied in the last two decades for the measurement of the topography, surface deformation, environment and velocities

(Zebker et al. 1994c; Zebker and Rosen 1997; Kwok and Fahnestock 1994; Rignot et al.

1994; Massonnet et al. 1994; Peltzer et al. 1994; Peltzer and Rosen 1995).

31 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION InSAR has enabled fine spatial resolution, wide coverage, and competitive accuracy measurement of terrain information. For successful two-pass interferometry, two conditions of achievement must be met. The sensor which is mounted on the satellite or aircraft has to revisit the target position closely enough to maintain coherence, and to ensure the terrain condition has not been unduly changed during the revisited periods.

If the terrain undergoes significant change in the interim period, the coherence value will be low.

In particular, space-borne InSAR DEM generation is an excellent method to regularly update global scale DEMs. Merging the SRTM and InSAR DEMs, or combinations of different InSAR DEMs, is useful for updating DEMs.

3.1.1 Introduction of InSAR

SAR images are constructed by recoding the phase (time delay) and amplitude

(energy intensity) of microwaves which are backscattered from the Earth terrain.

Geographic coordinates are calculated from the observed ranges and Doppler central frequencies, which are calculated in that the target points are focused in SAR imaging processing. InSAR technique uses the phase values from two radar images (Li et al.

2007; Smith 2002).

The geometry of repeat pass InSAR configuration is illustrated in Figure 3.1. The distance from a resolution pixel on the ground surface is represented by R1 and R2, the altitude of the satellite is Hs and the pixel elevation on the ground is h. The look angle of the antenna is ș and the angle of the baseline B with respect to the horizontal, is Į.

The perpendicular baseline between the two acquisitions is Bŏ.

32 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION

Satellite 2 B Satellite 1 Bŏ ș R2 R1 H

P Azimuth h

Range

Figure 3. 1 Geometry of repeat pass interferometric configuration (Hanssen 2001)

Using fundamental trigonometry, the elevation of the resolution pixel can be determined using Equation (3.1)

−= θ RHh 1 cos (Equation 3.1)

SAR images record the amplitude and phase values of the back-scattered signals from the resolution pixel using complex values. Therefore, two image acquisitions (S1 and S2) can be represented in exponential form with the modulus a1 and a2 and argument b1 and b2 as:

= = aS 11 exp(jb ;) aS 221 exp(jb2 ) (Equation 3.2)

The phase difference between S1 and S2, the so-called interferogram, is an interference pattern of fringes containing all the information on relative geometry. It can be computed by using conjugated multiplication of S1 and S2 as:

* = − 21 21 exp( ( bbjaaSS 21 )) (Equation 3.3)

2π b −= 2 R + φ 1 λ 11 scattering (Equation 3.4)

2π b −= 2 R + φ 2 λ 22 scattering (Equation 3.5) 33 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION

The contributions of the scattering phases in both SAR images are φ and 1 scattering

φ . 2 scattering

If S1 and S2 are highly coherent, then φ and φ can be treated as 1 scattering 2 scattering identical, hence:

4π 4π φ RRbb )( Δ−=−−=−= R 21 λ 21 λ (Equation 3.6)

Assuming R1 and R2 are perfectly parallel, then ǻR can be solved using the cosine rule as follows:

22 2 −+=Δ+ 0 +− θα 2 )( RBRR 2 2BR2 cos(90 ) (Equation 3.7)

2 2 It can be simplified by ignoring ΔR as (ǻR/R1,2) <<1, so:

B 2 BR θα )sin( +−−=Δ (Equation 3.8) 2R2

For space-borne systems, the second term at the right hand side of the equation can be ignored as B is normally less than one kilometre while R is about 500~900km, such that:

αθθα =−=−−≈Δ BR sin( ) Bsin( ) B// (Equation 3.9)

Where B// is the parallel baseline of the interferometric configuration.

Therefore, from Equations (3.1), (3.6) and (3.9):

θ ∂=∂ θ Rh 1 sin (Equation 3.10)

4π φ Δ∂−=∂ R λ (Equation 3.11)

−=Δ∂ BR cos( θ α ) ∂− θ (Equation 3.12)

34 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION Combining Equations (3.10), (3.11) and (3.12):

∂φ π θ − α π = 4 B cos( ) = 4 B⊥ ∂ λ θ θλ (Equation 3.13) h R1 sin R1 sin

Equation (3.6) shows that the ambiguity of a 2ʌ phase change to the height displacement along the radar line of sight is Ȝ/2.

If the baseline is zero, that is, the antenna position of two SAR images is absolutely the same, each interferometric phase is ascribed to ground displacement and distortion.

Therefore, short baseline conditions are optimal for detecting ground movement monitoring. However, longer baselines increase sensitivity to topography because it generates more cycles of the interferometric phase in the interferogram for the same altitude of terrain. The long baseline interferogram is a critical for DEM generation (Li et al. 2007; Smith 2002).

3.1.1.1 Conditions for interferometry

Before InSAR processing, it is important to understand the image and image-related conditions for interferometry, and be able to select pairs of images that may have the necessary properties for the generation of useful terrain information. The imaging geometry of the first image-pass must be almost replicated in the second image-pass.

There are three further conditions of interferometry that also must be fulfilled: 1) no change in terrain backscatter, 2) stable viewing geometry, and 3) an SAR sensor that preserves the inherent phase information in the motion compensated signal data. These conditions affect the coherence value of two images. The coherence value measures the similarity level of two images. SAR interferometry only works under coherent conditions, where the received signals are correlated between master and slave SAR images. The coherence of two images is the most important condition for 35 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION interferometry. Loss of coherence, known as decorrelation, can be due to a number of driving mechanisms (Hanssen 2001; Toutin and Gray 2000). The coherence image is generated using both the phase and amplitude from the SAR data pair. Figure 3.2 shows the coherence image of ALOS/PALSAR data.

Figure 3. 2 Coherence image of ALOS/PALSAR

3.1.1.2 Critical baseline of InSAR

The essential issue of interferometry is the baseline determination between the two sensors. The useful distance between the two sensors, which is measured perpendicularly to the look direction, is referred to as the perpendicular or effective baseline because the sensor is unable to directly measure the angular differences. This determination needs to be derived from the distance measurements between both sensors and the resolution pixel on terrain (Hanssen 2001).

When the InSAR technique is applied to DEM generation, the spatial baseline remains as a factor in the following considerations: 1) the terrain height measurement accuracy is proportional to the spatial baseline because elevation sensitivity increases with the length of the baseline, 2) however, when the spatial baseline increases, the

36 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION correlation of the SAR images decreases and height measurement accuracy decreases.

Therefore, there must be an optimal spatial baseline that balances these two opposing factors. Table 3.1 is the critical baseline of different radar.

The critical baseline (Bcritical) is calculated:

λR sin θ B = B (Equation 3.14) critical c

where, B is the spatial baseline, ș is the incidence angle, R is the distance between sensor and target, Ȝ is the radar wavelength, c is the light speed.

Table 3. 1 A comparison of critical baselines of radar from Sandwell et al. report

(2008).

Look angle 23˚ 34˚ 41˚

ERS/ENVIST 1.1 km 2.0 km 2.9 km 16MHz ALOS FBD 3.6 km 6.5 km 9.6 km 14MHz ALOS FBS 7.3 km 13.1 km 18.6 km 28MHZ

Note. ERS/ENVISAT: altitude=790km, wavelength=56mm; ALOS: altitude=700km, wavelength=236mm; Shaded boxes represent the most commonly used modes for interferometry (Sandwel et al. 2008)

3.2 InSAR DEM processing

Phase information gained from pixels of a single SAR image is insufficient to generate DEMs. In order to obtain essential elevation information, the relative phase

37 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION from at least two independent SAR images of the same site is required. The SAR images are selected from the same area based on two adjacent passes, and the phase difference of the two SAR images is used to generate the interferogram. Elevation information is then obtained from the interferogram by unwrapping the interferometric phases.

The phase difference of two coherent SAR images identifies the interferometric phase contribution due to the terrain morphology:

4π φ −= RR )( λ 21 (Equation 3.15)

where Ȝ is the radar wavelength, R1 and R2 are the distances from master and slave to ground targets. φ φ φ φ φ ++++= φ InSAR flat topo defo atm noi (Equation 3.16)

φ φ φ where, InSAR is the interferometric phase, flat is the flat earth phase, topo is the

φ φ topographic phase, defo is the deformation phase, atm is the atmospheric delay phase

φ and noi is the noise.

The atmospheric delay can be identified using the fact that its fringe structure is independent over several interferograms, or can be modelled by using a GPS network or interferogram stacking.

φ φ In Equation (3.16), the “flat earth” phase flat and noise noi can be removed by using the orbit information correction and applying an interferogram filtering method.

The flat earth phase is:

4π x φ = 0 Δx flat λ − (Equation 3.17) rr sm )(

38 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION

where, ȟx is the displacement range direction distance from swath, x0 is the shortest distance from ground trace to swath, rm is the shortest slant range distance from master image, rs is the shortest slant range distance from slave image (Huang et al. 2000). φ When the imaging interval is sufficiently short, there is no deformation phase defo .

φ If the atmospheric delay phase atm can be ignored, then equation (3.16) reduces to:

φ = φ InSAR topo (Equation 3.18)

The DEM is obtained by “unwrapping” the phase φ , converting it to a height quantity for the pixel. The pixel is then geo-coded. The coordinates and individual parameters of the InSAR DEMs are determined according to the parameters of the master image (Ferretti et al. 2001).

The essential relationship used to derive the topographic height from the interferometric phase is the following:

λ R sin θ h = φ (Equation 3.19) 4π B

Where, B is the baseline, ș is the incidence angle and R is the range distance of master image. The accuracy of the interferometrically derived DEM depends on the interferometric configuration and on the noise level of the fringe pattern.

Interferometric processing requires a comparatively short temporal baseline between image acquisitions in order to ensure maximum coherence (d’Ozouville et al. 2008;

Rabus et al. 2003; Sandwell et al. 2008).

Figure 3.3 shows the processing steps for InSAR generation.

39 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION

External DEM Master SLC Slave SLC

Simulated SAR

Coregistration Coregistration (DEM to Master) (Slave to Master)

Interferogram

Coherence

Denoise Interferogram

Phase Unwrapping

Phase to Height

Geocoded Products

Figure 3. 3 The flow chart of InSAR DEM generation

InSAR is a useful technique for the generation of DEMs and for the mapping of displacement fields to the order of radar wavelengths. For processing, at least two spatially or temporally separated images are required to form an interferometric pair.

3.2.1 Raw image

Raw SAR data is the term generally used for the original data from a SAR sensor.

Raw data are digitized on a low number of bits. For example, the ERS and ALOS data are coded on 5 bits for the real part and another 5 bits for the imaginary part (Maitre

2008; JAXA). The raw data is processed in pulse compression (range) and synthesis

(azimuth) and is corrected for motion compensation and other processes to produce 40 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION complex look data. The Single Look Complex (SLC) data are arrays of complex numbers containing amplitude and phase information. The image amplitude represents the energy reflectivity of the Earth’s surface back to the sensor. The reflected energy is related to the reflection pattern (diffuse and specular reflection), which is dependent on the wavelength and surface roughness. Most natural objects are diffuse reflectors.

Long wavelengths (L-band) penetrate a relatively thin tree canopy more deeply than short wavelengths (C-band). Electromagnetic signals, such as those used in Global

Positioning Systems (GPS) and radar, carries an amplitude and phase. Phase values are distributed between –ʌ and +ʌ. The phase has a useful meaning when these values are isolated by comparing radar images (Massonnet and Feigl 1998). Figure 3.4 illustrates the intensity image of SAR.

Figure 3. 4 The intensity image of ALOS/PALSAR

3.2.2 Coregistration

Before the interferogram is generated, very precise coregistration is required to increase the coherence of the two images, and the accuracy of the final products. If the

41 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION coregistration errors are to the order of the geometric resolution, the coherence of the interferogram is significantly reduced and the phase noise is increased.

Traditional coregistration methods for SAR images were developed in the late 1980s and early 1990s, and are based on the estimation of the azimuth and range warp- functions, that is, the functions that map the slave onto the master image in both in azimuth and range directions.

InSAR requires a pixel-to-pixel match between common features in SAR image pairs (master and slave). Due to the different look geometry in each pass, co- registration is an essential first step to align the two SAR images and thus obtain an accurate determination of phase difference to reduce noise in InSAR processing. The typical SAR coregistration procedure is as follows: 1) pixel level accuracy co- registration, including a search for image offsets and shift of the slave image, 2) sub- pixel level accuracy coregistration, including a search for sub-pixel tie points, fitting transformation equations, and re-sampling the slave image. The commonly reported level of co-registration accuracy in the literature is to the order of 0.05 image samples

(Li and Bethel 2008; Li and Goldstein 1990; Scheiber and Moreira 2000)

3.2.3 Coherence

Coherence is a measurement of similarity between two SAR images. Generally, the

SAR ground scene is not stationary. Therefore, the coherence in the acquired scene changes from area to area or even largely from pixel to pixel. The quality of the interferograms is predominantly and intuitively estimated by the coherence image. The phase values, measured from backscattering, are composite values formed by the complex summation of the contributions from individual scatterers within a single pixel.

The range value is defined between 0 (low coherence) and 1 (high coherence) (Huang

42 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION et al. 2000). Coherence quality depends on consistency of the scattering characteristics at the surface during radar acquisition time. Coherence is affected by local slope, condition of surface (vegetation, grass, river, and moving objects), the perpendicular baseline distance, and other technical problems (Earthnet online ESA, 2009). The coherence image can be used as a measure to confirm the accuracy of the interferometric phase and to determine the potential use of SAR imagery for DEM generation.

Coherence depends on three elements: 1) decorrelation from spatial baseline of two images, 2) decorrelation of physical changes in the terrain during the period between image acquisitions, 3) decorrelation due to the ground motion of the individual scattering centres within each resolution element.

The amplitude of the complex correlation coefficient, relative to the two images involved in the interferogram is as follows:

SSE * ][ r = 21 (Equation 3.20) 2 ⋅ 2 1 SESE 2 ][][

* Where, S1S2 represents the conjugated multiplication between master and slave and

E [•] is the statistical expectation. By this definition, a coherence value can be assigned to every pixel in the interferogram. While coherence is more relative to the backscattering coefficient than to topography, in the mountainous areas, the coherence is also affected by topography (Wagner et al. 2003).

Coherence information is independent from the image amplitude information and enables many applications such as in coastline detection, erosion or flooded area detection, and in classification of forests.

43 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION 3.2.4 Interferogram

The interferogram of two SAR images (master and slave) is generated after co- registration of two SAR images. The basic principle of interferometry is that the phase of SAR images is an ambiguous measure of the sensor-target distance. Therefore, distance variations can be determined by computing the phase difference between two

SAR images on a pixel level. Complex interferograms are calculated by multiplying each complex pixel of the master image by the complex conjugate of the corresponding pixel in slave image (Colesanti and Wasowski 2006). The intensity of the interferogram is a measure of cross correlation between the images. Interferometry only works under coherent conditions, where the received reflections are correlated between the two SAR images.

The phase difference of two coherent SAR images identifies the interferometric phase contribution due to the terrain morphology:

(4 B sin − B sin θθπ ) 4π φ = h v −= RR )( (Equation 3.21) λ λ 21

where Ȝ is the radar wavelength.

However, the receiving system can only measure phase modulo-2ʌ, the “wrapped phase”.

φ ϕ += k 2π (Equation 3.22)

Where φ is the true phase value (the so-called absolute value), ϕ is the measured

(wrapped) modulo-2 ʌ phase value, and k is integer number of wavelengths (Bioucas-

Dias and Valadao 2005, 2007).

44 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION 3.2.5 Altitude of ambiguities

The altitude between next to two phase discontinuities after interferogram flattening is called the ‘altitude of ambiguity’ and can be computed from the interferometer parameters:

R ×× sin θλ = m ha (Equation 3.23) 2B ⊥

where, Rm is the distance from sensor to target in master image, Ȝ represents the wavelength, ș is the look angle and Bŏ is the perpendicular baseline.

According to Equation 3.23, a large perpendicular baseline produces high topographic sensitivity in the interferogram, but a low, unambiguous height range.

Moreover, perpendicular baselines that are too large can lead to increased phase errors due to signal spatial decorrelation at the baseline extremities (Ge et al. 2006; Gini and

Lombardini 2005; Kotsis et al. 2008; Massonnet and Souyris 2008).

The relative elevation accuracy as a function of the altitude of ambiguity is defined as: Δ φ =Δ hh (Equation 3.24) a 2π

3.2.6 Phase unwrapping

The purpose of phase unwrapping is to resolve the 2ʌ ambiguity and the process of recovering the absolute phase from the wrapped phase. The interferogram, which is directly related to the topography, is only measured in modulo 2ʌ. To calculate the elevation of each point, it is necessary to add the correct integer number of phase cycles to each phase measurement. Phase unwrapping refers to converting the measured phase

45 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION to the absolute phase. Conventional phase unwrapping methods, such as the branch-cut and the least square one method, require 1/10 to 1/100 resolution cell size accuracy (Li et al, 2006).

w W φφ }{ mod{ }2, −+== πππφ (Equation 3.25)

ΔR −= 4πφ 2 k +=+ φπφ λ N N (Equation 3.26)

where W{·} is the wrapping operator and

π W }{ <⋅<− π (Equation 3.27)

Many new phase unwrapping algorithms have been developed since InSAR became a research . More popular methods include the minimum cost matching (MCM) method, which is based on the existence of discontinuity sources or residues. The

Residue-Cut Method (RCM) is a comparatively fast method, and the Least Squares

Method (LSM) uses the minimisation of discrete gradients difference squares. These differences are taken between the wrapped phases calculated discrete gradients and supposed unwrapped phase discrete gradients (Buckland et al. 1995; Chen and Zebker

2000; Crosetto 2002; Goldstein et al. 1988; Gutmann and Weber 2000; Hanssen 2001).

After the phase unwrapping process, phases are unflattened by adding the phase contribution of the earth curvature (flat earth phase). The absolute phase is a measure for the range difference between the sensors which will be exploited for the phase to height conversion.

3.2.7 Phase to height

The geometry of the interferometric height determination is presented in figure 3.1.

The height (h) above a reference datum and the range direction (position y) of the target

46 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION are computed with the known sensor elevation (H) and range distance (R), both depending on the local incidence angle:

−= RHh cosθ (Equation 3.28)

= Ry sin θ (Equation 3.29)

The unwrapped phases have to be converted in calculation of terrain heights.

Phase to height conversion is the procedure that relates the unwrapped phase to topographic height. After phase unwrapping, the height of the terrain can be evaluated using equation 3.30.

λ R sinθ λ R sinθ h =Δ m φ =Δ m Δφ π −αθ π (Equation 3.30) 4 Bs cos( 4) B⊥

where Rm is range distance in the master image, Bs is the baseline and Bŏ is the perpendicular baseline (Liao et al. 2007; Wimmer et al. 2000).

3.2.8 Geocoding

Geocoding is the geometric transformation of an image into a cartographic map projection. The image coordinates (line/sample) are changed to geographic coordinates

(latitude/longitude) using image geometry (Bayer et al. 1991; Schwabisch 1998).

In the process of geocoding, the geometric distortions of radar images due to the side-looking SAR sensor observation technique and terrain shape are corrected and translated into a geodetic coordinate system. The SAR data is acquired from backscattered signals from ground targets, and converted into an image in range- azimuth coordinates from the radar receiver’s perspective. This range-azimuth projection is also commonly called the slant range projection. As per former discussions about processing, all works are organised in the slant range projection.

Therefore, the SAR processing results are re-projected to a common Earth-related

47 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION reference frame. The process of transforming SAR results from slant range projection into a geographic coordinate system is called geocoding or georeferencing.

3.3 Sources of InSAR noise

InSAR generated DEM accuracy is determined by system parameters (incidence angle, frequency etc.), the geometry of InSAR (baseline) and surface conditions (vegetation, terrain relief etc).

Table 3. 2 Parameters influencing the InSAR data (Crosetto 2002; Kyaruzi 2005)

Incidence angle Spatial resolution Internal clock drift Satellite system Approximated focusing Image mis-coregistration System error

Determination precise range Baseline (coherence, geometry decoration) Orbit Repeat phase (temporal decorrelation) Non-parallel orbits

Frequency Polarization Signal Bandwidth Noise/speckle

Flat terrain Phase and range difference Topography Direction of slope Surface characteristics

Wind Weather conditions Snow coverage Thermal noise

Low coherence Processing Phase unwrapping difficulties

48 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION As InSAR generated DEMs utilise two SAR images acquired at different times or in different orbits the phase accuracy of interferograms are affected by 1) spatial, temporal and volume decorrelation 2) thermal noise and processing error, 3) atmospheric noise,

4) orbit and geometry error. Table 3.2 summarise the noises.

3.3.1 Decorrelation

Usually called ‘Decorrelation’ or ‘Incoherence’, this phenomenon confuses the arranged fringe pattern in an interferogram. For every pixel affected by randomly changed phases, an area of random speckles appears in the interferogram (Massonnet and Feigl 1998). Decorrelation depends on local weather conditions, vegetation cover, complexity of terrain, land use, etc. The decorrelation effect is more severe in areas with dense vegetation cover, complex terrain or constantly changing weather conditions.

In vegetation areas, the short wavelength (C-band) is expected to have higher decorrelation than the longer wavelength (L-band). Decorrelation can be classified into three main categories: spatial decorrelation, temporal decorrelation and volume decorrelation. They can be presented as (Alberga 2004; Zebker and Villasenor 1992);

γ γ γ ⋅⋅= γ Total Temporal Spatial SNR (Equation 3.31)

γ In this formula, Temporal represents the decorrelation of a temporal scene and is defined as the ratio of temporarily stable scattering contributions to the total scattering

γ intensity transferred to the SAR images. Spatial is the decorrelation due to the different

γ processing performed on the SAR signals. SNR is the decorrelation due to additive noise.

49 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION Spatial decorrelation

The spatial decorrelation is caused by the physical distortion between the locations of the two sensors. This physical separation is referred to as the baseline. This decorrelation is particularly evident on the sensor facing terrain slope (foreshortening, layover) and it may be misinterpreted as rapid random changes. In particular, the spatial decorrelation due to the terrain slope and the baseline dominates any other decorrelation factors in mountainous areas (Lee and Liu 2001; Massonnet and Feigl

1998).

The spatial baseline decorrelation function is given by

2 cos θ BR ⊥ γ 1 −= gr (Equation 3.32) spatial λR

Where, Bŏ is perpendicular baseline, Rgr is the ground range resolution, Ȝ is the wavelength and R is the slant range. In the equation, the spatial decorrelation increases according to the baseline.

The ground resolution is a function of the local terrain slope as

c R = (Equation 3.33) r − αθ 2 B w sin( 0 )

Where c is light speed, Bw is the frequency bandwidth of the transmitted chirp signal, and Į is the local slope measured upward from the horizon away from the radar direction.

Temporal decorrelation

Temporal decorrelation is caused by physical change to the dielectric properties of a surface between two observations in the same orbit. Repeat pass InSAR is affected by the decorrelation of ground resolution factors which derive from the surface and 50 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION backscatter properties. In InSAR DEM generation, successive interferograms are processed under two conditions: terrain stability of two repeat pass SAR images and radar scattering characteristics. These conditions are not met by the terrain properties of agricultural area. Not only harvested areas, but also crop movements in the presence of wind make it impossible to recover the topography.

Volume decorrelation

Volume decorrelation is also related to the phase stability of the imaging pixel over time, especially multiple scattering over the forest canopy. The dielectric properties

(surface moisture film, freezing, thawing) and alignments of leaves and branches of trees in the canopy may result in variation of the backscattered radar signals. For agricultural crops, the mechanical cultivation associated with farming activities such as harvesting, ploughing, mowing and tillage cause complete decorrelation as all scatterings are changed. Furthermore, volume decorrelation is also affected easily by local weather conditions, such as rainfall and wind. Therefore, it is desirable to have a finer SAR imaging resolution in order to minimise the variation of the random scattering phases (Balzter 2001).

3.3.2 Thermal noise and Processing error

Thermal noise is caused by characteristics of the system as well as decorrelation noise, before they are combined to form interferograms. Thermal noise is dependent on the radar cross section of reflections and is reduced by using transmitters with the greatest power and receivers with the lowest noise.

51 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION Processing error

When a DEM is constructed from an interferogram, the phase errors also convert to elevation errors. The phase value is the most important element in DEM generation using the InSAR method because the phase expresses the relevant height value between neighbouring pixels. The phase error is mainly due to 1) phase trends caused by orbital errors in the flattened interferogram, 2) decorrelation and thermal noise, and 3) atmospheric phase error between master and slave images (Liao et al. 2007).

σ σ The height standard deviation ( H ) is related to the phase standard deviation ( φ ) according to the first order approximation formula:

λ R sin θ σ = σ φ (Equation 3.34) H 4π B cos( − αθ )

In the above equation, to keep a low height standard deviation, the reflected phase value must be stable between the two images.

The geometry of phase difference is influenced by two conditions: 1) the phase difference from the height difference with an identical slant range, and 2) the phase difference from the slant range difference with an identical height (Liao et al. 2007).

Figure 3.5 shows the phase difference resulting from the height difference with an identical slant range and the slant range difference with an identical height.

As shown in the following Figure 3.5,

4π B cos(θ α )Δ− H φ −= (Equation 3.35) H λ R sin θ

4π B cos(θ α )Δ− R φ −= (Equation 3.36) flt λ R tan θ where ǻH is the height difference within an identical slant range and ǻR is the slant range difference within an identical height. 52 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION Equation (3.36) represents the flat Earth effect in phase change. In InSAR processing, the flat Earth effect can be modelled and removed as reference phase in the

‘flat Earth’ step by orbital information.

Į Į

ș ș

ǻșH ǻșflt R R

P'

P ǻH P ǻR P'

Figure 3. 5 Geometry of phase difference due to the height difference with an identical

slant range (left) and geometry of phase difference due to the slant range difference

with an identical height (right) (Liao et al. 2007).

3.3.3 Atmospheric Noise

The atmospheric component is the major limitation of the interferometric method for

DEM generation. The atmospheric conditions may severely affect the information content of the InSAR phase. It is well known that electromagnetic signals are affected when the signals pass through the atmosphere (Toutin and Gray 2000).

Electromagnetic signals travel at slightly different speeds in media with different refractive index values. In InSAR, the atmosphere fluent may cause a phase shift on the sensed signal. The phase shift of the signal is also dependent on the refractivity in different air strata - which would result in a mismeasurement in elevated terrain. The

53 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION condition of the atmosphere is not identical in acquired images. Any difference of the troposphere or the ionosphere between the two images can change the distance of the path between the radar and the ground (Crosetto 2002; Ding et al. 2008; Massonnet et al. 1994). This is shown as phase variation between two SAR images even for flat terrain without ground deformation.

φ The atmospheric phase component, atm , is primarily due to the fluctuations of atmospheric water vapour between the sensor and the surface. It can cause a phase variation equivalent to about 9cm in several irregular patterns in interferograms over a distance of several kilometres. The atmospheric noise can be identified using the fact that its fringe structure is independent over several interferograms, or it can be modelled by GPS network data (Khiabain et al. 2008).

3.3.4 Orbit and Geometry error

Satellite InSAR processing requires the orbital satellite ephemerides and baseline between two images information for accurate DEM generation. As such, orbit errors propagate directly into errors in topographic height or deformation maps (Hanssen

2001; Li and Bethel 2007).

The baseline is also conveniently represented using the parallel and perpendicular baseline, the horizontal and vertical baseline, as well as the baseline length and angle.

Figure 3.6 shows the different baseline determination.

& = ()= ()()= α || ⊥ vh BBBBBB ,,, (Equation 3.37)

The perpendicular baseline can be regarded as a scaling factor for the altitude of ambiguity (Hanssen 2001).

54 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION

BŒ

Bv

Bŏ

Bh

Į

B

Figure 3. 6 The different determination of baselines between sensors

Image plane Image plane B' A' A' C' B' C'

B C B A A C Layover Foreshortening

Image plane

A' B'

B A Shadowing C

Figure 3. 7 The Geometrical limitations of SAR images; Layover, Foreshortening and Shadowing (Hanssen, 2001)

55 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION The geometrical limitations of SAR images (figure 3.7) are affected by the relationship between the look angle and terrain relief. These limitations are classified into layover, foreshortening, and shadowing.

Layover is a terrain induced distortion of the SAR image and occurs in very steep terrain on the sensor-facing direction of mountainous areas. Layover occurs when the terrain slope angle is larger than the SAR incidence angle. The layover area can be detected by the above condition, and is extracted by inspection of SAR images. This is represented by areas of relative brightness due to the superposition of response signal power.

terrain slope angle > incidence angle (Equation 3.38)

Shadowing is similar to solar illumination. The SAR signal may not detect the backside of a mountain in case of

terrain angleslope ≤ incidence angle − 90 0 (Equation 3.39)

The receiver is able to receive weak or no response signals from the shadowed area.

Shadowed areas in DEM are non-elevation areas. The intuitive way to fill the DEM gap is DEM generation using opposite side images of the shadowed area (Honikel 1998).

Orbit error

Another error source in the InSAR DEM generation is related to sensor orbit ambiguities that correspond to errors in the interferometer baseline length and tilt angle evaluation. Therefore, interferograms require orbital information or vectors of satellites for the determination of the interferometric baseline and to refer the interferometric products to a reference. Orbital errors may cause additive phase patterns in the interferogram and propagate directly into errors of topographic elevation. Orbital errors

56 CHAPTER 3 INTERFEROMETRIC SAR DEM GENERATION are systematic and relative to SAR coordinates in azimuth and range directions

(Hanssen 2001; Liao et al. 2007; Sansosti et al. 1999).

R sin 2 θ σ = σ (Equation 3.40) Hx B cos( − αθ ) Bx

R sin θ con θ σ = σ (Equation 3.41) Hy B cos( − αθ ) By

where σ and σ are the standard deviations of the baseline components. Bx By

Figure 3.8 illustrates the orbit error elements. An orbit error is expressed in the coordinate system rotating with the satellite and consists of three components: 1) the along-track, 2) the across-track, 3) the radial direction component.

Radial error

Orbit

Across track error Along track error

Figure 3. 8 The geometry of orbit errors

For SAR interferometry, the along-track errors are sufficiently corrected during the coregistration of the two images. Along-track positioning errors can also be regarded as timing errors. According to Hanssen (2001), only radial and across-track errors will propagate as systematic phase errors in the interferogram.

57 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Currently, X-, C- and L-band are used by satellite systems. Each microwave band has a different advantage for Earth observations. Over forested or vegetated areas,

C-band data is expected to have higher decorrelation than the comparatively longer wavelength of L-band, as the C-band wavelength is more comparable to the size of tree leaves. L-band is affected by large vegetation components such as trunks of trees and major structures of forests (Pulliainen et al. 1999). X-band observation can detect smaller particles and be utilised for high-resolution applications. To increase the azimuth resolution, a shorter wavelength, a wider beamwidth or a lower orbiting spacecraft may be employed (Hanssen 2001).

Figure 4.1 illustrates the wavelength and frequency range for radar. The division of waves and frequency has varying determinations by a range of international organisation such as North Atlantic Treaty Organization (NATO) and The Institute of

Electrical and Electronics Engineers (IEEE).

Table 4.1 details the spacing and resolution of ERS (C-band) and ALOS (L-band) radar images.

Figure 4. 1 The waves and frequency ranges used by radar (radartutorial)

58 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Table 4. 1 Spacing and resolution of radar images (unit: m)

ERS ALOS/PALSAR Pixel spacing 7.904890 4.684257 Pixel resolution 9.639629 5.353437 Line spacing 4.038725 3.228661 Line resolution 5.048417 4.566016

4.1 C-band InSAR DEM Generation

IEEE defines the C-band in the microwave range of frequencies of the electromagnetic spectrum ranging from 4~8 GHz. The European Space Agency (ESA) launched the ERS-1 satellite in 1991, which operated in the C-band microwave range.

After the launch of ERS-2 in 1995, both ERS-1 and 2 cooperated over nine months in the first ever ‘tandem’ mission. During the mission, both satellites’ temporal resolution was just 24 hours. They were used to monitor natural disasters such as severe flooding or earthquakes throughout remote parts of Earth. They also monitored agricultural areas, forests, coastlines and marine pollution. For InSAR DEM generation, the radar images acquired by this tandem mission can provide high coherence interferograms. More than

140,000 image pairs are available for InSAR techniques (ESA; Rignot and Zyl 1993).

In this study, ERS tandem systems were used for C-band InSAR DEM generation.

One can expect that for ERS tandem systems, the temporal coherence should be quite high provided the terrain conditions are stable. Table 4.2 contains the characteristics of the ERS satellite system and Table 4.3 details data information of ERS tandem pairs

(Davidson and Cumming 1997; Williams 2000).

59 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Table 4. 2 The characteristics of ERS system

Wavelength 5.6 cm (C-band: 5.3 GHz)

Polarisation VV

Bandwidth 15.5±0.06MHz

Swath width 100km × 100km

Table 4. 3 ERS data pairs information

Sensor Master date Slave date Bperp (m) Btemp (days)

ERS 29/10/1995 30/10/1995 49 1

ERS 03/12/1995 04/12/1995 145 1

The test site was in the Appin and Wollongong areas in the state of New South

Wales, Australia. SRTM DEM was used as the reference and external DEM.

According to the pixel and line spacing of the azimuth and range direction in Table

4.1, the ERS has extensive coverage compared with other radar sensors. Therefore, the

Appin and Wollongong areas were processed simultaneously to take advantage of the large coverage.

Figure 4.2 represent Landsat images of the test sites. The Appin area was covered by mountainous forest, grassland, and rivers. The Wollongong area is composed of mountainous areas, steep sloped areas between the mountains and city, coastline and city. Each test area is made up of a variety of component elements and different scattering mechanisms. These different characteristics affected the interferogram and unwrapping processes.

60 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 2 The research areas in Appin (left) and Wollongong (right), NSW, Australia

In InSAR DEM processing, the relationship between interferogram fringes and elevation is an important factor as the terrain height is calculated from the fringes. The altitude between two adjacent discontinuities is called the ‘altitude of ambiguity’. The altitude of ambiguity of the interferogram fringes is calculated for comparison of the height sensitivity in each DEM. For the ERS, the wavelength was 5.6 cm, the incidence angle was 23˚ and the distance between the antenna and target on the ground was 850 km. In principle, the longer baseline provides a more accurate altitude measurement.

However, it is vulnerable to the interferometric signal decorrelation and may lead to no fringes. There is an optimum perpendicular baseline that maximises the signal to noise power ratio. Table 4.4 lists the altitude of ambiguities of the test image pairs.

Table 4. 4 Altitude of ambiguity for each data pair

Pair of SAR data Bperp (m) Height ambiguities ha (m)

29/10/1995-30/10/1995 49 187m

03/12/1995-04/12/1995 145 63m

61 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION The height ambiguity due to the phase changes are changed by the perpendicular baseline and wavelength. Table 4.5 and Figure 4.3 present the elevation ambiguity due to phase changes in the C-band. Each fringe is represented by the 2-pi phase difference.

Table 4. 5 Height ambiguity due to phase changes of C-band.

Phase change (Δφ ) Bperp(m) ʌ/4 ʌ/2 3ʌ/4 ʌ 5ʌ/4 3ʌ/2 7ʌ/4 2ʌ

100 11.54 23.24 34.78 46.47 58.16 69.71 81.25 92.94

200 5.77 11.62 17.39 23.24 29.08 34.85 40.63 46.47

400 2.89 5.81 8.70 11.62 14.54 17.43 20.31 23.24

600 1.93 3.88 5.80 7.76 9.71 11.63 13.56 15.51

800 1.44 2.90 4.35 5.81 7.27 8.71 10.16 11.62

1000 1.15 2.32 3.48 4.65 5.82 6.97 8.13 9.29

200.00

150.00

100.00 Height error (m)

100.00 200.00 50.00 400.00 600.00 800.00 1000.00

0.00 ʌ/4 ʌ/2 3ʌ/4 ʌ 5ʌ/4 3ʌ/2 7ʌ/4 2ʌ phase error

Figure 4. 3 Height ambiguity due to phase changes of C-band.

62 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 4 C-band coherence (29/10/95-30/10/95-left) and C-band coherence

(03/12/95-04/12/95-right) images from ERS Tandem

Figure 4.4 shows the coherence images of the research areas using ERS images. The coherence represents the relationship between two processed images (master and slave).

In the figure 4.4, 03/12/95-04/12/95 coherence image had low value. However, the weather information in test area at same data acquisition did not provide from government. So I cannot comment any reason about the low coherence value in the mountainous area. The coherence mean value is 0.32 in the 29/10/95-30/10/95 pair and

0.19 in 03/12/95-04/12/95 pair. Each mean value is lower than expected because the research area includes the water surface. Water surfaces such as oceans, rivers, and reservoirs have low coherence values due to their constant surface movement.

Figure 4.5 is the C–band InSAR generated DEMs in the Appin and Wollongong area in NSW. Coverage is determined by the spacing and resolution of images. According to the Table 4.1, the ERS sensor has a 7.9m pixel spacing and 4.0m line spacing. This means that the same numbers of pixels cover a larger area than the L-band sensor and hence more volume decorrelation. Furthermore, the processing time per coverage is reduced and the processing hardware capacity is improved. In Figure 4.5, both DEMs

63 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION have increased noise, or error, near the coast due to movement at the water surface.

Elevation errors due to movement at the surface of the ocean are related to low coherence levels and unwrapping processing. Figure 4.6 shows InSAR DEM and elevation errors caused by water movement effects.

Figure 4. 5 C-band InSAR generated DEM (29/10/95-30/10/95-left) and C-band InSAR

generated DEM (03/12/95-04/12/95-right) using ERS Tandem

Figure 4. 6 The InSAR DEM and DEM errors (red box) surrounding the coastline in

Wollongong (29/10/95-30/10/95-left) and Wollongong city (correspond red box-right)

64 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Figure 4.7 illustrates the profile lines for verification of the accuracy of the C-band

InSAR generated DEM. Profile line No. 1 crosses from a mountainous area to a flat grassy area. Profile line No. 2 was selected to include the city and the rapidly sloping terrain of the adjacent escarpment.

ID: 0

ID: 1600 ID: 2900

ID: 0

Figure 4. 7 The profile lines of the test area in Appin and Wollongong

350

300

250 Elevation (m) 200

150 SRTM (No.1)

ERS (29/10/95-30/10/95: No.1)

ERS (03/12/95-04/12/95: No.1)

100 1 65 129 193 257 321 385 449 513 577 641 705 769 833 897 961 1025 1089 1153 1217 1281 1345 1409 1473 1537 1601 ID

Figure 4. 8 The profile lines (No.1) of C-band InSAR generated DEM in Appin

Figure 4.8 and 4.9 show the terrain cross section follows the profile lines in Appin and Wollongong. In figure 4.8, the elevation difference between two InSAR DEMs

65 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION (ID: 520~1150) is related with low coherence condition. Image pair 03/12/95-04/12/95 has low coherence value in that area in figure 4.4. Coherence value of two images presents the confidence of products.

700

SRTM (No.2) 600 ERS (03/12/95-04/12/95: No.2)

ERS (29/10/95-30/10/95: No.2)

500

400

Elevation (m) 300

200

100

0 1 109 217 325 433 541 649 757 865 973 1081 1189 1297 1405 1513 1621 1729 1837 1945 2053 2161 2269 2377 2485 2593 2701 2809 ID

Figure 4. 9 The profile lines (No. 2) of C-band InSAR generated DEM in Wollongong

Table 4.6 is the Root Mean Square Error (RMSE) of single pass ERS InSAR DEMs compared with SRTM DEM.

Table 4. 6 The Root Mean Square Error of C-band InSAR DEMs

RMSE (unit: m)

SRTM DEM-ERS InSAR DEM (29/10/95-30/10/95: No.1) 32.1

SRTM DEM- ERS InSAR DEM (03/12/95-04/12/95: No.1) 18.0

SRTM DEM- ERS InSAR DEM (29/10/95-30/10/95: No.2) 27.2

SRTM DEM- ERS InSAR DEM (03/12/95-04/12/95: No.2) 27.4

66 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION 4.2 L-band InSAR DEM Generation

The IEEE defines the L-band in the microwave range of the electromagnetic spectrum, ranging from 1~2 GHz. The main advantages of the L-band compared with the C-band include being able to penetrate deeply into areas of vegetation, which allows for less temporal decorrelation. This enables interferograms to have a longer temporal separation and a longer wavelength, which increases the critical baseline and provides more useful interferometric pairs. In this research, ALOS/PALSAR images were used for DEM generation by the L-band sensor. The ALOS/PALSAR Fine Beam

Single (FBS) has a 40~70km observation swath and can provide global scale image coverage three times per year (Furuta et al. 2005; Jarvis et al. 2004).

The test site is the Appin and Wollongong area in the state of New South Wales,

Australia. The SRTM DEM was used as the external DEM.

ALOS systems are used for L-band InSAR DEM generation. Table 4.7 details the features of ALOS/PALSAR satellite system and Table 4.8 presents the data information of PALSAR pairs. Figure 4.10 shows the test area for L-band InSAR DEM generation.

Table 4. 7 The characters of the ALOS/PALSAR system

Wavelength 23.5 cm (L-band: 1.2 GHz)

Polarisation HH/VV

Band width 28MHz

Swath width 70km × 70km

67 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Table 4. 8 ALOS/PALSAR Data pair information

Sensor Master date Slave date Bperp (m) Btemp (days)

ALOS 14/11/2007 30/12/2007 757 46

ALOS 14/02/2008 31/03/2008 659 46

Table 4. 9 Altitude of ambiguity for data pairs

Pair of SAR data Bperp (m) Height ambiguities ha (m)

14/11/2007-30/12/2007 757 85m

14/02/2008-31/03/2008 659 98m

Figure 4. 10 The research area in Appin, NSW, Australia

The altitude of ambiguity of the interferogram fringes is calculated for comparison of the height sensitivity in each DEM. For PALSAR, the wavelength was 24cm, the incidence angle was 38˚ and the distance between the antenna and target on the ground

68 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION was 883km. Table 4.9 lists the elevation ambiguity of test data pairs. Table 4.10 and

Figure 4.11 present the elevation ambiguity due to phase changes in the L-band.

Table 4. 10 Height ambiguity due to phase changes of L-band.

Phase change (Δφ ) Bperp(m) ʌ/4 ʌ/2 3ʌ/4 ʌ 5ʌ/4 3ʌ/2 7ʌ/4 2ʌ

100 91.10 183.38274.48 366.75 459.02 550.13 641.23733.50

200 45.55 91.69 137.24183.38 229.51 275.06 320.62366.75

400 22.78 45.84 68.62 91.69 114.76 137.53 160.31 183.38

600 15.19 30.57 45.75 61.14 76.52 91.70 106.89 122.27

800 11.39 22.92 34.31 45.84 57.38 68.77 80.15 91.69

1000 9.11 18.34 27.45 36.68 45.90 55.01 64.12 73.35

800.00

700.00

600.00 100.00 200.00 400.00 500.00 600.00 800.00 1000.00

400.00 Height error

300.00

200.00

100.00

0.00 ʌ/4 ʌ/2 3ʌ/4 ʌ 5ʌ/4 3ʌ/2 7ʌ/4 2ʌ Phase error

Figure 4. 11 Height ambiguity due to phase changes of L-band.

69 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 12 L-band coherence (14/11/07-30/12/07-left) and L-band coherence

(14/02/08-31/03/08-right) images from ALOS/PALSAR

The coherence represents the relations of two processed images (master and slave).

The coherence mean value is 0.36 in the 14/11/07-30/12/07 pair and 0.83 in 14/02/08-

31/03/08 pair. This mean value difference between two coherence images is related with weather and seasonal effects in acquisition time. Figure 4.12 shows the coherence images of two pairs in Appin. The coherence value is lower in the surrounding river and farming areas.

Figure 4. 13 L-band InSAR generated DEM (14/11/07-30/12/07-left) and L-band

InSAR generated DEM (14/02/08-31/03/08-right) using ALOS/PALSAR. 70 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Figure 4.13 illustrates the InSAR generated DEMs using L-band ALOS/PALSAR images. The test area, which consists of mountainous and flat ground, covers approximately 40km2. In the results, small parts (red dotted line circles in figure 4.13) have an elevation error due to the ground deformation between two different acquisition times. This effect is also one of problem in InSAR DEM generation. It is discussed in chapter 4.5.3.

Figure 4. 14 Large coverage L-band InSAR generated DEM (31/03/08-01/07/08)

Figure 4.14 shows a large coverage L-band InSAR generated DEM. It covers the

Appin, Wollongong, and part of the Blue Mountain areas in NSW. The result coverage is approximately 55 km × 60 km and consists of a variety of terrain conditions. This

DEM was generated from one pair of ALOS/PALSAR images. Comparing with

ALOS/Panchromatic Remote-Instrument for Stereo Mapping (PRISM), the image coverage is 35km×35km under non-cloud weather condition with 2.5m spatial resolution. (EORC, JAXA).

71 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

ID: 0

ID: 600

ID: 0 ID: 730

Figure 4. 15 The profile lines in Appin

Figure 4.15 shows profile lines for the analysis of final DEM. Profile line (No.1) traverses a river and flat ground. Profile line (No.2) was incorporates a forested and mountainous region.

200

180

160

140

120

100 Height (m) 80

60 LiDAR

ALOS (14/11/07-30/12/07) 40 ALOS (14/02/08-30/03/08) 20

0 1 19 37 55 73 91 109 127 145 163 181 199 217 235 253 271 289 307 325 343 361 379 397 415 433 451 469 487 505 523 541 559 577 595 Points

Figure 4. 16 The profile (No.1) of L-band InSAR generated DEM in Appin

Figure 4.16 illustrates the profile (No.1) in the Appin area. In this graph, two profile lines have similar terrain patterns and elevation fluctuations. However, some points have higher elevation than the reference DEM (LiDAR). This elevation difference is

72 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION caused by relation between sensor direction and terrain shape effects. The SAR image has difficulty in valley which is hidden from sensor viewing range.

Figure 4.17 illustrates profile (No.2) in the Appin area. The profile lines were shifted compared with the reference DEM, due to the look direction of the sensor. The terrain appearance was more distinct than the reference DEM.

Table 4.11 represents the Root Mean Square Error (RMSE) of single pass

ALOS/PALSAR InSAR DEMs compared with LiDAR DEM.

350

300

250 Height (m) 200

150 LiDAR

ALOS (14/01/07-30/12/07)

ALOS (14/02/08-30/02/08) 100 1 23 45 67 89 111 133 155 177 199 221 243 265 287 309 331 353 375 397 419 441 463 485 507 529 551 573 595 617 639 661 683 705 727 Points

Figure 4. 17 The profile (No.2) of L-band InSAR generated DEM in Appin

Table 4. 11 The Root Mean Square Error of L-band InSAR DEM

RMSE (unit:m)

LiDAR DEM-PALSAR InSAR DEM (14/11/07-30/12/07: No.1) 11.3

LiDAR DEM-PALSAR InSAR DEM (14/02/08-30/03/08: No.1) 18.0

LiDAR DEM-PALSAR InSAR DEM (14/11/07-30/12/07: No.2) 9.7

LiDAR DEM-PALSAR InSAR DEM (14/02/08-30/03/08: No.2) 10.9

73 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 18 GPS RTK Surveying points in the Appin area

Figure 4.18 shows the Global Positioning System (GPS) Real Time Kinematic

(RTK) surveying points in the Appin test area. The GPS points were surveyed along the roads. GPS do not covered forested area, because the GPS technique is unable to survey under covered areas such as forests and tunnels. GPS measurement can provide true ground data in the test area. However, the survey coverage area and technical capability is limited.

Table 4.12 presents the RMSE of comparison between the GPS surveying data and

L-band InSAR DEM and LiDAR DEM.

Table 4. 12 The Root Mean Square Error of L-band InSAR DEM compared with GPS measurement.

RMSE (unit:m)

GPS - PALSAR InSAR DEM (14/11/07-30/12/07: No.1) 6.7

GPS - PALSAR InSAR DEM (14/02/08-30/03/08: No.1) 8.1

GPS - LiDAR DEM 0.9

74 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION 4.3 Assessment of different wavelength InSAR data DEM generation

The different wavelength band SAR images are allowed different terrain responses in the same area. Furthermore, pixels in X, C, and L-band SAR images contain different features which provide a variety of terrain information from different wavelength data. For example, due to the relationship between the wavelength and the size of the ground object, the short wavelength data is expected to have more detailed ground information, but it appears the lower correlation between image pairs than the comparatively longer wavelength band data. Therefore, using various wavelengths in the same target area is a useful method for InSAR DEM generation and for other purposes requiring geometric information.

In the profiles of C- and L-band InSAR generated DEM (figure 4.19), the C-band

InSAR generated DEM had higher elevations than the L-band InSAR generated DEM and a smoother terrain response. The height differences were affected by the relationship between the observation wavelength and the terrain canopy, while the shape differences were affected by the relationship between pixel size and line space.

350

300

250 Height (m) Height 200

LiDAR 150 ALOS (14/01/07-30/12/07)

ALOS (14/02/08-30/02/08)

ERS (03/12/95-04/12/95)

100 1 23 45 67 89 111 133 155 177 199 221 243 265 287 309 331 353 375 397 419 441 463 485 507 529 551 573 595 617 639 661 683 705 727 Points

Figure 4. 19 The elevation comparison of C and L-band generated InSAR DEMs in

Appin. The profile line uses the No.2 line in figure 4.15. 75 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Table 4.13 is the Root Mean Square Error (RMSE) of single pass ALOS/PALSAR and ERS Tandem InSAR DEMs compared with LiDAR DEM.

Table 4. 13 The Root Mean Square Error of different bands InSAR DEMs

RMSE (unit:m)

LiDAR DEM-PALSAR InSAR DEM (14/11/07-30/12/07) 9.7

LiDAR DEM-PALSAR InSAR DEM (14/02/08-30/03/08) 10.9

LiDAR DEM-ERS InSAR DEM (03/12/95-04/12/95) 24.2

4.4 Ascending and Descending InSAR generated DEM fusion

A satellite can acquire an image when it passes from south to north (ascending orbit) or from north to south (descending orbit). Images from ascending and descending orbits utilise different imaging geometry. Due to the structure of SAR signal transmitters and receivers, SAR images contain slant ranges. This feature of SAR images is a limitation of image acquisition, especially for the extraction of terrain information, due to problems with foreshortening, layover and shadowed areas. In the case of foreshortening, the ground resolution is slightly degraded, although the heights can still be calculated. However, no information can be recovered if there is layover or shadowing. If both ascending and descending passes are used, foreshortening, layover and shadowing in one orbit direction’s InSAR pair may be recovered from other passes

(Crosetto 2002; Yun et al. 2006). In airborne InSAR applications, Murata and

Miyawaki (2003) used different incident angles and directions (ascending and descending) for observation of a volcanic area. Figure 4.20 shows the different geometry of ascending and descending orbits SAR systems.

76 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Ascending orbit Descending orbit

Figure 4. 20 Geometry of ascending and descending SAR imaging.

Figure 4. 21 Different orbits and observed location of ascending and descending SAR

master images from table 4.14 coordinate value

Figure 4.21 illustrates the satellite orbits and observed location of ascending and descending SAR images according to the state vector coordinates of raw images.

Ascending and descending orbits derived from different locations. In the illustration, the look direction, which observes the targets from the sensor, is set in the opposite direction. Also, if the target area contains mountainous terrain, the scattered information of different side orbit sensors is acquired from the opposite side slope of terrain. If the orbit direction is the same, the perpendicular baseline between each

77 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION sensor is small compared to the sensor from the opposite orbit direction. Table 4.14 contains the coordinates of the SAR sensor at master image acquisition.

Table 4. 14 The state vector coordinates of ascending and descending images

(Cartesian coordinate system)

14/11/07 31/03/08 01/06/08 17/07/08 (ascending) (ascending) (descending) (descending) x: -2856373.802094 x: -3040335.831031 x: -6707162.857529 x: -6701692.082298 y: 1097104.822033 y: 1279255.936505 y: 1899414.157338 y: 1904276.312634 z: -6385403.343382 z: -6265549.035622 z: -1218993.206261 z: -1243920.082046 : : : : : : : : x: -4157288.447444 x: -4288210.356967 x: -5877443.151306 x: -5864312.114955 y: 2487202.205073 y: 2652202.757768 y: 2136941.414350 y: 2140731.511050 z: -5163304.559381 z: -4970522.105656 z: -3316185.929882 z: -3338135.812850

x: -4364151.483460 x: -4482810.663974 x: -5636350.780352 x: -5621847.469556 y: 2745951.776000 y: 2905162.523010 y: 2154058.261601 y: 2157551.683165 z: -4851866.984389 z: -4646598.148115 z: -3701880.070780 z: -3722964.365773

x: -4551266.427681 x: -4657235.576525 x: -5372483.374387 x: -5356674.953779 y: 2995398.293813 y: 3148067.603499 y: 2160300.703511 y: 2163474.029433 z: -4520801.568977 z: -4303871.713414 z: -4072572.314679 z: -4092709.654788 : : : : : : : : x: -5170285.601537 x: -5208430.981699 x: -3755355.679372 x: -3734200.183358 y: 4063030.309414 y: 4172613.854257 y: 2022833.170602 y: 2024127.196886 z: -2620137.318473 z: -2358835.016940 z: -5651227.321364 z: -5665572.813421

x: -5228035.224469 x: -5252162.219761 x: -3382228.785101 x: -3360268.158141 y: 4233246.960352 y: 4332363.981366 y: 1961287.918233 y: 1962166.015242 z: -2201679.887730 z: -1934548.110552 z: -5903117.710904 z: -5916129.478502

x: -5263305.325485 x: -5273405.483688 x: -2996130.515500 x: -2973460.429460 y: 4386815.161994 y: 4474939.420301 y: 1888558.360151 y: 1889012.695232 z: -1774286.178125 z: -1502407.266720 z: -6131167.593247 z: -6142800.095161

4.4.1 Terrain Response Differences for Ascending and Descending SAR data

This research used the ascending and descending ALOS/PALSAR SAR images (L- band) to generate DEMs. The ascending pairs correspond to 1-repeat (46 days) and 2- repeat cycle orbits, while the descending pairs are from 1-repeat cycle orbits. In order to improve the coherence of image pairs, short temporal baselines were selected for

DEM generation. Comparing each baseline of ascending and descending data pairs,

78 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION descending pairs had similar distances but ascending pairs recorded some variation.

This was also provided a suitable combination to test the relationship between baseline distance and height sensitivity.

Table 4. 15 ALOS/PALSAR data information

Orbit direction Master date Slave date Bperp (m) Btemp (days) Ascending 14/11/2007 30/12/2007 757 46 Ascending 31/03/2008 01/07/2008 2992 92 Descending 01/06/2008 17/07/2008 2809 46 Descending 17/07/2008 01/09/2008 2245 46

Figure 4. 22 The intensity images of ascending (14/11/2007: left) and descending

(17/07/2008: right) orbits in test area

Figure 4. 23 Boundaries of ascending and descending pairs in the Appin area

79 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Table 4.15 provides the test data information and Figure 4.22 shows the intensity images of the opposite side orbit. Figure 4.23 shows the range of ascending and descending pairs. Each pair covered slightly different areas because the master image of each pair had different orbit parameters. Moreover, the ascending and descending orbit images had different flight directions. Therefore, the overlapping area, which can be generated by the orbit merging method, is reduced. Figures 4.24 and 4.25 illustrate the DEM products from different orbit pairs.

Figure 4. 24 ALOS/PALSAR ascending InSAR generated DEMs (14/11/07-30/12/07:

left and 31/03/08/-01/07/08: right)

Figure 4. 25 ALOS/PALSAR descending InSAR generated DEMs (01/06/08-17/07/08:

left and 17/07/08-01/09/08: right) 80 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Figure 4.26 shows the profile lines of ascending and descending InSAR generated

DEMs. Profile lines were selected for verification of the accuracy of the final InSAR generated DEM.

ID: 919 ID: 0

ID: 595 ID: 0

Figure 4. 26 The profile lines of ascending and descending ALOS/PALSAR

The profile direction was in a similar direction to the range direction. The range direction was clearer than the azimuth direction for comparison with the different orbit effect. Figures 4.27 and 4.28 represent the profiles of ascending and descending InSAR generated DEMs in the test area. In the ascending (14/11/07-30/12/17) example, some elevation errors are evident in certain parts of the area due to ground deformation between the two acquisition times. In the Figures 4.27 and 4.28, ascending and descending orbit DEM results were shifted to each sensor.

81 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

350

300

250 Height (m) 200

LiDAR Ascending (14/11/07-30/12/07) Ascending (31/03/08-01/07/08) 150 Descending (01/06/08-17/07/08) Descending (17/07/08-01/09/08)

100 1 19 37 55 73 91 109 127 145 163 181 199 217 235 253 271 289 307 325 343 361 379 397 415 433 451 469 487 505 523 541 559 577 595 Points

Figure 4. 27 Profile line (No.1) of ascending and descending ALOS/PALSAR

300

250

200

150 Height (m)

100

LiDAR Ascending (14/11/07-30/12/07) 50 Ascending (31/03/08-01/07/08) Descending (01/06/08-17/07/08) Descending (17/07/08-01/09/08)

0 1 28 55 82 109 136 163 190 217 244 271 298 325 352 379 406 433 460 487 514 541 568 595 622 649 676 703 730 757 784 811 838 865 892 919 Points

Figure 4. 28 Profile line (No.2) of ascending and descending ALOS/PALSAR

82 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 29 Elevation differences of LiDAR DEM-Ascending (14/11/07-30/12/07)

DEM

Figure 4. 30 Elevation differences of LiDAR DEM-Ascending (31/03/08-01/07/08)

DEM

83 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 31 Elevation differences of LiDAR DEM-Descending (01/06/08-17/07/08)

DEM

Figure 4. 32 Elevation differences of LiDAR DEM-Descending (17/07/08-01/09/08)

DEM

Figures 4.29, 4.30, 4.31 and 4.32 show the elevation differences of ascending and descending InSAR DEMs compared with LiDAR DEM.

84 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION 4.4.2 Merging DEMs from Ascending and Descending InSAR DEM

The orbit direction (ascending or descending) provides a different representation of terrain over the same object area. Hence, using images collected by the satellite sensor from different orbit directions is way of improving the quality of InSAR-generated

DEMs (Crosetto 2002). The reconstructed DEM take advantage of the different InSAR

DEM altitudes of ambiguity to fill in the elevation gap of low resolution and other

DEMs. This interpolation is affected by the perpendicular baseline and different orbits of each master image. Parameters of the InSAR DEMs are affected by the master image parameters even if master and slave image are swapped for DEM generation, the resulting products will be slightly different DEMs. Therefore, each InSAR DEM has different grid locations and altitude of ambiguity. These factors cause different elevation contour and grid density.

Three methods are used to merge the InSAR DEMs. The first method is the “mean stacking DEMs’ method:

n ¦ i yxh ),( i =1 H = (Equation 4.1) n

where x and y are the coordinates of the pixel/point in the InSAR DEM, n is the number of InSAR DEMs, and hi is the height value of point (x,y).

The “coherence weight stacking” method uses the coherence value as the weight of the merging value:

+ chch H = iiii ++ 11 + (Equation 4.2) cc ii +1

th where, ci is the coherence value in i DEM.

85 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Tables 4.5 and 4.10 present the relationship between the height ambiguity and the phase changes. In the table, the longer baseline has less height error in the final DEM.

The “baseline distance weight stacking” method uses the baseline distance as the weight of the merging value:

n ¦ Bh ii = i = 1 (Equation 4.3) H n ¦ B i i = 1

where Bi is the baseline distance

Figure 4.33, 4.34 and 4.35 show the merged InSAR DEMs using different merging methods.

Figure 4. 33 The merged InSAR generated DEM using mean method from Ascending

(14/11/07-30/12/07 and 31/03/08/-01/07/08) and Descending (01/06/08-17/07/08 and

17/07/08-01/09/08) DEMs

86 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 34 The merged InSAR generated DEM using baseline weight mean method

from Ascending (14/11/07-30/12/07 and 31/03/08/-01/07/08) and Descending

(01/06/08-17/07/08 and 17/07/08-01/09/08) DEMs

Figure 4. 35 The merged InSAR generated DEM using coherence weight mean method

from Ascending (14/11/07-30/12/07 and 31/03/08/-01/07/08) and Descending

(01/06/08-17/07/08 and 17/07/08-01/09/08) DEMs

87 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Figure 4.36, 4.37 and 4.38 illustrate the part (red box in figure 4.35) of DEMs which are generated by different merging methods and different orbit direction DEM. This area has large elevation difference and some geographical distortions of SAR. In this condition, the elevation errors are increased. Therefore, it is good test area for DEM re- generation techniques.

Figure 4. 36 The part of SRTM DEM (left) and mean merged DEM (right) in test area

Figure 4. 37 The part of baseline weight merged DEM (left) and coherence weight

merged DEM (right)

88 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 38 Descending (left) and ascending (right) InSAR DEMs in test area

Table 4.16 is the Root Mean Square Error (RMSE) of merged DEM compared with

LiDAR DEM.

Table 4. 16 The Root Mean Square Error of merged DEM

RMSE (unit:m) LiDAR DEM-Mean merged DEM 9.0 LiDAR DEM-Baseline weight merged DEM 9.1 LiDAR DEM-Coherence weight merged DEM 8.9

Figure 4. 39 Elevation differences of LiDAR DEM-Mean merged DEM

89 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 40 Elevation differences of LiDAR DEM-Baseline weight merged DEM

Figure 4. 41 Elevation differences of LiDAR DEM-Coherence weight merged DEM

Figures 4.39, 4.40 and 4.41 graph elevation differences with reference DEM based on figure 4.33~4.35. Baseline and coherence weight merged DEMs have less difference than mean merged DEM. However, the noise or elevation errors were not sufficiently removed after the merging process. These were increased in some parts.

Figure 4.18 shows the profile points of GPS RTK surveying in the test area. Table

4.17 is the RMSE compare with GPS surveying data and merged InSAR DEMs.

90 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Table 4. 17 The Root Mean Square Error of merged DEM compare with GPS measurements.

RMSE (unit:m) GPS - Mean merged DEM 4.5 GPS - Baseline weight merged DEM 4.7 GPS - Coherence weight merged DEM 4.5

4.5 Surface deformation exclusion in InSAR DEM using differential interferogram

Repeat-pass satellite SAR interferometry is a useful method for generating relatively low-cost, wide-coverage DEM coverage with minimum delay, at -level accuracy.

The orbits of SAR satellites are such that they revisit a similar path typically every 11-

46 days (Table 2.3). An interferometric phase comparison of two SAR images acquired at different times and perpendicular baselines (i.e. different orbital positions) allows for the extraction of three-dimensional terrain topography information (Ferretti et al. 2001).

However, a major shortcoming of InSAR DEM processing is the assumption that there is no deformation phase, because the time interval between SAR image acquisitions is comparatively short. Hence, if there are ground deformation signatures in the interferometric fringes during InSAR processing, this limits the accuracy of the resulting DEM, the phase unwrapping process, and, ultimately, the representation of terrain reality.

Differential InSAR (DInSAR) is a well-known method of determining ground deformation. DInSAR processing requires the removal of the interferometric phase that is contributed to by the topography. That is, DInSAR results isolate the area of ground deformation without topographical effect. This deformation measurement method has 91 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION been applied to landslides, earthquakes, and urban and mining subsidence (Amelung et al. 1999; Chang et al. 2008; Ge et al. 2008; Zebker et al. 1994b). The basic processing requirements for DInSAR are essentially the same as for InSAR processing.

4.5.1 Surface deformation detection using differential InSAR

Differential InSAR is often utilised to measure ground deformation using ‘repeat- pass interferometry’. The main issue associated with DInSAR processing is the discrimination between noise and phase difference due to topography, ground displacement, and atmospheric delay in the interferometric phase. For the detection of ground deformation, the topographic and “flat Earth” phase values are calculated and removed using an external DEM and precise satellite orbit data. Using DInSAR, surface deformations can be measured with a very high degree of accuracy and spatial measurement density.

φ In Equation (4.4), the topographical fringes topo are calculated from simulated fringes, which are generated from an “external” (or known) DEM, and subtracted from the original interferogram.

+−++++= φφφφφφφφ DInSAR flat topo defo atm noi ( topo_ sim flat _ sim ) (Equation 4.4)

After all fringes are removed due to ground elevation, only the fringes representing ground surface displacements remain (see, e.g., Crosetto 2004; Ge et al. 2002):

φ ≅ φ (Equation 4.5) DInSAR defo

Since the development of the DInSAR technique, it has proved to be a very useful technique of mapping ground deformation.

In differential interferogram processing, the areas that have no ground deformation have a common phase value whereas the deformation-affected areas are represented by 92 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION repeat cycle fringes between –ʌ and ʌ. The number of repeat cycle fringes directly reflects the magnitude of ground deformation which is related to Equation 3.30. The ground deformation fringes are distinct from the stable ground phase and are an important indicator for highlighting the deformation-affected areas. In particular, the differential interferogram is represented by the relative phase value. Therefore, the phase value for the area unaffected by ground deformation is always altered. Figure

4.42 plots the relative phase value of differential interferograms.

Figure 4. 42 Examples of differential interferogram of different pairs in DInSAR

(14/11/07-30/12/07: left and 14/11/07-14/02/08: right). The relative phase value is

changed between –pi to +pi and the differential interferogram illustrates the relative

phase value of each interferogram.

4.5.2 Area detection of surface deformation

Candidate areas classified as being affected by ground deformation are selected from the DInSAR differential interferograms using a specific phase value, which is determined by the most common phase values in the interferogram. Consequently, the phase value selected for detection of deformation is unique to each interferogram.

Areas affected by deformation are distinguished from stable areas using the threshold

93 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION method. The identified candidate areas are dispersed throughout the binary images and are roughly located surrounding, but not fully covering, deformation areas. Hence, the candidate areas are expanded to cover all deformation areas.

The deformation area exclusion is processed at the interferogram level in InSAR processing. The deformation area is extracted using the deformation mask. The removed deformation area is replaced by an interpolated area derived from surrounding pixel values and external DEM information. After deformation exclusion processing, the following steps are the same as for standard InSAR DEM processing.

i-2, j-2 i-2, j i-2, j+2

i-1, j-1 i-1, j i-1, j+1

i, j-2 i, j-1 i, j i, j+1 i, j+2

i+1, j-1 i+1, j i+1, j+1

i+2, j-2 i+2, j i+2, j+2

Figure 4. 43 Window of 16-neighborhood pixels for deformation detection in

differential interferogram.

The search window size is determined by consideration of processing time and the relationship between image ground resolution and image pixel size. In this research,

16-neighbourhood windows were used for the detection of the deformation area and its expansion in the MATLAB. Figure 4.43 shows the 16-neighbourhood windows. The proceeding steps were: 1) Selection of specific phase value using threshold method

(each DInSAR has different value), 2) Expansion of selected area (covering the

94 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION deformation area), 3) Exclusion of noise pixel (unexpected areas), and 4) extraction of deformation area and fringes.

Firstly, the specific phase value (for deformation areas), which is distinguished from non-deformation area, is determined for selection according to deformation area pixel candidates. Figure 4.44 present histograms of the number of pixels according to the selected phase value for identification of potential sites of deformation. Each histogram has a different common phase value range (stable area). It was useful to select a pixel value for deformation candidate areas, as determined by a low number of pixels. For example, in the 14/11/07-30/12/07 pair at Figure 4.44, the highest numbers of pixels have a phase value between –pi to 1, but several pixels have a phase value between 1 to

2. These were the selected pixel values for the deformation candidate area.

Each differential interferogram has a different candidate pixel value; in the first pair

(14/11/07-30/12/07) the candidate pixel value is between 1 to 2, and in the second pair

(14/11/07-14/02/08) the candidate pixel value is around –pi to -2.5 and 2.5 to pi.

Figure 4. 44 Histograms of number of pixels (14/11/07-30/12/07: left) and (14/11/07-

14/02/08: right)

The deformation mask was developed based on the candidate pixels using 16- neighborhood windows. In particular, the mask expansion processing performed calculations from two different image directions because the mask shape patterns are 95 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION affected by the image processing direction. After calculating the mask area, the extended mask was filtered to remove the noise (detection error). Figure 4.45 is the deformation area mask from DInSAR processing. The mask included some topographic noise area.

Figure 4. 45 The mask of a deformation area from DInSAR

The deformation fringes in interferogram of InSAR were extracted using a deformation mask. In this processing, the deformation mask and interferogram of

InSAR had to have the same image coordination and size. The mask provided the deformation pixel location, which was applied to the interferogram of InSAR. The removed deformation area was replaced by the interpolated elevation values using surrounding pixel values and the external DEM information. Firstly, surrounding pixels which are not affected by ground deformation were collected and averaged to fill into the removed area information. Secondly, external DEM was used to elevation interpolation with replaced elevation information. After deformation exclusion processing, post-processing steps followed the general InSAR DEM processing

96 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION 4.5.3 InSAR DEM generation without surface deformation

InSAR DEM generation without surface deformation processing combines two different SAR data processing techniques. InSAR processing is used for DEM generation while DInSAR processing is used for surface deformation area detection.

Figure 4.46 presents a flowchart of InSAR DEM generation without ground deformation using InSAR and DInSAR techniques.

InSAR DInSAR

Differential interferogram Interferogram Specific value selection

Deformation area detection

Deformation area mask

Deformation exclusion

Elevation interpolation

Denoise interpolation

Unwrapping

DEM generation

Figure 4. 46 Flowchart of deformation exclusion of InSAR DEM using DInSAR

In DInSAR processing, master and slave images were selected in the same order as

InSAR DEM generation to ensure the correspondence of the two results. Table 4.18 summarises data related to deformation exclusion in InSAR DEM generation.

97 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Table 4.19 and Figure 4.47 illustrate the terrain elevation ambiguities caused by phase change and perpendicular baselines. Fringe changes (-pi~+pi) with a 100 meter perpendicular baseline affected an elevation error approximately of 733m in final DEM.

That is, a small phase change in the interferogram increased the error in the DEM products.

Table 4. 18 ALOS/PALSAR image information

Master date Slave date Bperp (m) Btemp (days)

14/11/2007 30/12/2007 757 46

14/11/2007 14/02/2008 742 92

800

700

600

500 100 300

500 700 400 900 1100 Height error (m) error Height 300

200

100

0 ʌ/4 2ʌ/4 3ʌ/4 ʌ 5ʌ/4 6ʌ/4 7ʌ/4 2ʌ Phase change

Figure 4. 47 Height ambiguities due to the phase change and perpendicular baseline

In the test area, ground subsidence was detected in two underground mining areas.

Figure 4.48 is a ground subsidence map generated from DInSAR processing and DEM from InSAR processing in same test area. The ground deformation area is determined using the DInSAR result. When comparing the InSAR-generated DEM with the

DInSAR deformation map, it can be seen that the ground deformation area is correlated with the DEM height error area.

98 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Table 4. 19 Height ambiguities due to the phase change and perpendicular baselines

Δφ Bperp Phase change ( ) (m) ʌ/4 2ʌ/4 3ʌ/4 ʌ 5ʌ/4 6ʌ/4 7ʌ/4 2ʌ 100 91.71 183.41 275.13 366.84 458.54 550.26 641.96 733.68 300 30.57 61.13 91.71 122.28 152.84 183.42 213.98 244.56 500 18.34 36.68 55.02 73.36 91.71 110.05 128.39 146.73 700 13.11 26.21 39.31 52.41 65.51 78.61 91.71 104.81 900 10.19 20.37 30.57 40.76 50.94 61.14 71.32 81.52 1100 8.33 16.67 25.01 33.34 41.68 50.02 58.36 66.69

Figure 4. 48 Subsidence map generated by DInSAR processing (left) and InSAR-

generated DEM (right) at the Appin test site

Figures 4.49 and 4.50 illustrate the peer-InSAR processed DEM and proposed-

InSAR processed DEM at 14/11/07-30/12/07 pair and 14/11/07-14/02/08 pair. The ground deformation effects were removed from the figures in the proposed InSAR

DEM generation.

99 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 49 InSAR-generated DEM and DEM with deformation removal (14/11/07-

30/12/07)

Figure 4. 50 InSAR-generated DEM and DEM with deformation removal (14/11/07-

14/02/08)

Figure 4.51 details the profile lines of the test area. The profile lines were selected across two areas of subsidence.

100 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

ID: 0

ID: 0

ID: 280 ID: 280

Figure 4. 51 Profile lines (No.1 and 2) on the ground deformation map at the test area

Figures 4.52, 4.53, 4.54 and 4.54 show profiles comparing InSAR-generated DEM, deformation removed DEM, and reference DEM (SRTM). Height difference was observed between two InSAR-generated DEMs in the ground deformation area. Also, elevation errors of the deformation area changed according to the perpendicular baseline of the master and slave images. Generally, terrain elevation without ground deformation presents the same height value without the effect of the perpendicular baseline. However, if ground deformation occurs in the interferogram, the terrain elevation changes depending on the interferogram fringes and perpendicular baseline.

101 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

190.00

170.00

150.00

130.00

Height (m) 110.00

90.00 DEM (14/11/07-30/12/07)

70.00 Re-DEM (14/11/07-30/12/07) Reference DEM

50.00 1 10 19 28 37 46 55 64 73 82 91 100 109 118 127 136 145 154 163 172 181 190 199 208 217 226 235 244 253 262 271 280 ID

Figure 4. 52 Profiles (No.1) of InSAR-generated DEM and ground deformation removed InSAR-generated DEM compared with SRTM DEM: (14/11/07-30/12/07)

300.00 DEM (14/11/07-14/02/08) Re-DEM (14/11/07-14/02/08) Reference DEM 250.00

200.00 Height (m) Height

150.00

100.00

50.00 1 10 19 28 37 46 55 64 73 82 91 100 109 118 127 136 145 154 163 172 181 190 199 208 217 226 235 244 253 262 271 280 ID

Figure 4. 53 Profiles (No.1) of InSAR-generated DEM and ground deformation removed InSAR-generated DEM compared with SRTM DEM: (14/11/07-14/02/08)

102 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

440.00 DEM (14/11/07-14/02/08) Re-DEM (14/11/07-14/02/08) 390.00 Reference DEM

340.00

290.00 Height (m) Height

240.00

190.00

140.00 1 16 31 46 61 76 91 106 121 136 151 166 181 196 211 226 241 256 271 286 301 316 331 346 361 376 391 406 421 436 451 466 ID

Figure 4. 54 Profiles (No.2) of InSAR-generated DEM and ground deformation removed InSAR-generated DEM compared with SRTM DEM: (14/11/07-14/02/08)

400.00 DEM (14/11/07-30/12/07) Re-DEM (14/11/07-30/12/07) 350.00 Reference DEM

300.00 Height (m) 250.00

200.00

150.00 1 16 31 46 61 76 91 106 121 136 151 166 181 196 211 226 241 256 271 286 301 316 331 346 361 376 391 406 421 436 451 466 ID

Figure 4. 55 Profiles (No.2) of InSAR-generated DEM and ground deformation removed InSAR-generated DEM compared with SRTM DEM: (14/11/07-30/12/07)

103 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION 4.6 SRTM DEM updating using selected InSAR DEM elevations based on coherence value selection method

Currently, low resolution SRTM DEM can be updated and interpolated using other high resolution DEMs which are generated by different sources or technique such as

InSAR, photogrammetry, and laser scanning. In many DEM generation techniques, the

InSAR DEM generation technique is suitable with SRTM coverage and uses the same technique. SRTM DEM acquired almost 80 percent of the Earth’s land surface located between 60˚N and 56˚S. A 1-arc-second (~30m) DEM and a downgraded 3-arc-second

(~90m) DEM were then derived from this data by the National Geospatial Intelligence

Agency (NGA) and its contractors (Bourgine and Baghdadi, 2005; Rodriguez et al.

2005; Farr et al. 2007).

InSAR generated DEM have approximately 10m resolution depending on the terrain shape and slope. However, the lower correlation between master and slave image is leaded the lower accurate elevation result at InSAR DEM generation. This quality of correlation between two images is measured in ‘coherence’ in the InSAR processing. A high coherence pixel leads to the high accurate elevation pixel that is highly accurate.

This thesis proposes a method for low resolution DEM updating using reliable elevation points selected from InSAR generated DEM based on their respective coherence value. Using these coherence values, the quality of elevation points are proven and selected, resulting in their potential replacement with old DEM data.

104 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

4.6.1 InSAR DEM elevation selection using coherence value

This processing is based on the coherence value of the InSAR DEM generation.

Figure 4.56 outlines the processing steps. According to the coherence value (0~1), the elevation pixels of InSAR generated DEM were selected and replaced in the SRTM

DEM. The threshold value of coherence could be selected depending on the quality expectation of updated DEM. During the process, only selected coherence pixels were used and they converted the same weighting value. A low coherence threshold value causes selection of many elevation pixels however the selected pixels have low credibility of elevation while high coherence value leads to high elevation credibility.

Using selected coherence pixels, selected elevation pixels were extracted from InSAR generated DEM. Also, selected coherence pixels were used for pixel location selection and substitution in SRTM DEM which will be replaced by selected elevation pixel from InSAR generated DEM. Before applying the pixel extraction, the oversampling processing is necessary for pixel size matching between SRTM DEM and InSAR generated DEMs. The SRTM DEM pixel size is reduced to InSAR DEM pixel size.

Finally, selected elevation pixels from InSAR generated DEM were replaced into the extracted pixels of SRTM DEM. Figure 4.56 shows the processing steps of SRTM

DEM updating using selected elevation. To improve the accuracy, the updated SRTM

DEMs using coherence values were merged using the mean merging method.

105 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

InSAR process SRTM DEM

Coherence image InSAR generated DEM

Coherence selection Pixel oversampling

Selected coherence

Selected elevation Removed SRTM DEM

Update SRTM DEM

Figure 4. 56 Flowchart of SRTM DEM updating using selected elevation

4.6.2 Elevation updating using selected InSAR DEM height value

The three pairs of ALOS/PALSAR SAR images (L-band) were used to generate the

DEMs. The SRTM DEM was used as the external DEM for InSAR processing and updated DEM. LiDAR DEM was used as “ground truth” DEM for the purpose of accuracy validation. Table 4.20 summarises SAR data information.

Table 4. 20 The information of ALOS/PALSAR Data

Sensor Master date Slave date Bperp (m) Btemp (days)

ALOS 14/11/2007 30/12/2007 757 46

ALOS 14/02/2008 31/03/2008 659 46

ALOS 31/03/2008 01/07/2008 2992 92

106 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION The coherence threshold values of 0.2 to 0.8 were determined. Selected coherence pixels were changed to the same weighting value (1) and non-selected coherence pixels had a zero value (0). Figure 4.57 shows the selected coherence image and selected elevation from InSAR generated DEM using a threshold value of 0.2. In the 0.2 case, many pixels are selected around the whole image.

Figure 4. 57 A selected coherence image (left) and a selected elevation (right) from

InSAR generated DEM using threshold value: 0.2

Figure 4. 58 A selected coherence image (left) and a selected elevation (right) from

InSAR generated DEM using threshold value: 0.5

107 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Figure 4.58 shows the selected coherence image and selected elevation from InSAR generated DEM using a threshold of 0.5. In the 0.5 case, non-selected pixels are located in the surrounding river, farming areas and some mountains.

Figure 4.59 shows the selected coherence image and selected elevation from InSAR generated DEM using a threshold of 0.7. In the 0.7 case, few pixels are selected from the coherence image. The selected pixels present stable terrain and elevation information. However, limited information can be updated in the final product.

Figure 4. 59 A selected coherence image (left) and a selected elevation (right) from

InSAR generated DEM using threshold value: 0.7

Figures 4.60, 4.61, 4.62 and 4.63 show updated SRTM DEMs using selected elevation from InSAR generated DEM by different threshold values (0.2~0.8). The zoom-in area is located in north-west direction from Appin city. This area has large elevation difference and various ground surface conditions. Under these conditions, coherence value is affected by many different parameters. Therefore, it is suitable test area for the coherence selection processing.

108 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 60 SRTM DEM (left) and updated SRTM DEM (right) using selected

elevation: threshold 0.2

Figure 4. 61 The updated SRTM DEM using selected elevation: threshold 0.3 (left) and

updated SRTM DEM using selected elevation: threshold 0.4 (right)

109 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 62 The updated SRTM DEM using selected elevation: threshold 0.5 (left) and

updated SRTM DEM using selected elevation: threshold 0.6 (right)

Figure 4. 63 The updated SRTM DEM using selected elevation: threshold 0.7 (left) and

updated SRTM DEM using selected elevation: threshold 0.8 (right)

Tables 4.21 and 4.22 outline the RMSE of SRTM DEM, updated SRTM DEM according to two selected elevations from InSAR generated DEMs and InSAR generated DEMs compared with LiDAR DEM. 110 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Table 4. 21 The RMSE of SRTM, update SRTM by different coherence values and

InSAR DEM (ALOS: 31/03/08-01/07/08) with LiDAR DEM.

RMSE (unit: m) Selected pixels SRTM DEM 9.8 Update SRTM DEM (coherence:0.8) 9.9 28,884 Update SRTM DEM (coherence:0.7) 9.8 200,645 Update SRTM DEM (coherence:0.6) 9.6 838,823 Update SRTM DEM (coherence:0.5) 9.3 2,099,513 Update SRTM DEM (coherence:0.4) 8.9 3,721,680 Update SRTM DEM (coherence:0.3) 8.4 5,292,729 Update SRTM DEM (coherence:0.2) 7.8 6,534,080 ALOS (31/03/08-01/07/08) 7.6 (total) 7,362,883

Table 4. 22 The RMSE of SRTM, update SRTM by different coherence values and

InSAR DEM (ALOS: 14/02/08-31/03/08) with LiDAR DEM.

RMSE (unit: m) Selected pixels SRTM DEM 9.9 Update SRTM DEM (coherence:0.8) 9.6 648,902 Update SRTM DEM (coherence:0.7) 9.7 2,134,055 Update SRTM DEM (coherence:0.6) 9.9 3,813,196 Update SRTM DEM (coherence:0.5) 11.0 5,214,518 Update SRTM DEM (coherence:0.4) 11.6 6,198,480 Update SRTM DEM (coherence:0.3) 12.1 6,811,424 Update SRTM DEM (coherence:0.2) 12.8 7,160,847 ALOS (14/02/08-31/03/08) 13.1 (total) 7,347,718

Coherence threshold selection is considered about the balance between high selected pixels numbers and low RMSE value. According to the numbers of selected pixels and

RMSE result from table 4.21 and 4.22, a coherence threshold value of 0.5 was selected for SRTM updating. Figure 4.64 shows the oversampled SRTM DEM and updated

SRTM DEM using selected elevation from InSAR DEM (14/02/08-31/03/08).

111 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION

Figure 4. 64 The SRTM DEM (left) and updated SRTM DEM (right) using selected

elevation from ALOS (14/02/08-31/03/08): threshold 0.5

Table 4. 23 The RMSE of SRTM DEM, InSAR DEM, update SRTM DEM and merged

SRTM DEM compare with LiDAR DEM

RMSE (unit: m) SRTM DEM 10.4 ALOS (14/11/07-30/12/07) 9.8 ALOS (14/02/08-31/03/08) 15.7 ALOS (31/03/08-01/07/08) 7.2 Update SRTM DEM by ALOS (14/11/07-30/12/07) 8.6 Update SRTM DEM by ALOS (14/02/08-31/03/08) 12.5 Update SRTM DEM by ALOS (31/03/08-01/07/08) 9.1 Mean merged SRTM DEM 7.4

Table 4.23 is the RMSEs between SRTM DEM, InSAR DEMs, updated SRTM

DEMs and merged SRTM DEM compared with LiDAR DEM. In the table 4.23, update

SRTM DEMs by InSAR DEM has lower RMSE value than pure SRTM DEM. It is affected by elevation data replacement from SRTM to InSAR DEM. Generally, updated SRTM DEMs have improved elevation information. However, if the updating elevation had low quality information, the updated DEM is contaminated by wrong elevation information in the update SRTM DEM by ALOS (14/02/08-31/03/08) case. 112 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION Furthermore, merged SRTM DEM using updated SRTM DEMs appears the improved elevation quality. InSAR DEM which was generated by ALOS (31/03/08-

01/07/08) leads the highest elevation quality than other DEMs in this case.

4.6 Concluding remarks

DEM generation using different wavelength SAR images using the InSAR technique is a useful method in areas of vegetation, forest, mountains and rapid slopes as different wavelengths produce different ground representations. In this research, C-band InSAR generated DEM measured the canopy of dense forest or trees. Therefore, the average elevation was higher than the L-band InSAR generated DEM. C-band InSAR DEM has approximately 20m RMSE and L-band InSAR DEM has approximately 10m RMSE compared with LiDAR DEM. L-band SAR image provided more stable images with reduced atmospheric effects and other noise compared to C-band SAR images. L-band

InSAR generated DEM provided more reliable DEM products. It provides about 7m

RMSE compare with GPS surveying along the road. ALOS/PALSAR data supplied higher resolution SAR images. These can be used to produce more accurate DEMs.

Opposite-side observation (ascending and descending) provided more reality ground information than same-side observation. This can improve the quality of DEM in areas of rough terrain. Furthermore, these InSAR DEMs can be merged to improving the accuracy with increased ground reality. In this paper, three kinds of merging methods were used and RMSE of merging methods is approximately 4.5m in all conditions.

Due to the ground deformation during different acquisition times of the master and slave images, the DEM elevation error was removed in InSAR DEM generation using the DInSAR technique. The elevation error due to ground deformation was 20~200m in

113 CHAPTER 4 C- & L-BAND InSAR DEM GENERATION test area. After ground deformation area detection and elevation interpolation processing, the elevation errors was removed and the DEM quality was improved. This method can provide more accurate and reliable DEM products generated by satellite systems.

Finally, the overall InSAR DEM has higher accuracy, however, some parts of

InSAR DEM have low accuracy because of the image pair conditions such as short baseline, noise and low coherence. Therefore, selected elevation data with accurate information is required to improve the updated DEM quality.

By selecting elevation by coherence value improves the elevation confidence in

DEM and important information for SRTM DEM updating. The updated SRTM DEM has higher degree of accuracy compared with single InSAR generated DEM and original SRTM DEM. Furthermore, merged SRTM DEM using each updated SRTM

DEMs also improve the vertical accuracy.

114 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

InSAR DEM generated by a single-baseline is limited by its poor height sensitivity and unwrapping process related to interferometric fringes and the geological layover phenomenon, which limits mapping of complicated topography contained in steep terrain. In the single-baseline InSAR, the relative interferometric phase is obtained from the phase unwrapping step. So far, one effective way of converting the relative interferometric phase to the absolute phase has required more than one ground control point (GCP). However, it is difficult to acquire GCPs in practical applications. The multi-baseline technique allows for the procurement of additional information from

SAR data.

Multi-baseline techniques can be usefully exploited to generate a DEM, which is less affected by errors by averaging the uncorrelated atmospheric contributions in the single-baseline interferogram.

The multi-baseline InSAR technique is capable of overcoming the weaknesses of single-baseline InSAR systems. Multi-baseline techniques are an advancement that enables interferometric techniques to be applied to not only the surface mapping, but also to monitor temporal evolution. Multi-baseline InSAR can be used by multiple- passes air-borne/space-borne InSAR sensors, or distributed InSAR satellites.

Consequently, multi-baseline methods have been more widely investigated throughout the last decade (Ferraiuolo et al. 2009; Fornaro et al. 2005, 2007; Li et al. 2007).

115 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

5.1 SAR data information

SAR signals require that a focusing process be performed on the received raw signal to produce a high-resolution image. The spatial resolution of the SAR is dependent upon the bandwidth of the source, the width of the transmitter and receiver beams, and the distance between the ground target and sensor. High resolution in the azimuth direction is achieved by processing a step aimed at synthesizing an aperture, where the dimension is much larger than the size of the azimuth of the real antenna mounted on the platform. In contrast, a high resolution in the range direction is obtained by focusing a transmitted by large bandwidth linear frequency modulated pulse. Figure 5.1 shows the acquisition area by the ALOS/PALSAR.

Figure 5. 1 The acquisition area by ALOS/PALSAR in the World (ERSDAC)

The PALSAR SLC data (level 1.1, Vexcel standard) is produced in 32-bit float format, slant range coordinate system, with product file size 535~1650MB, and azimuth look bandwidth 1584Hz. The scene is based on ‘scene fixed points’. The first

116 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION scene fixed point is + 0.25 degrees from the ascending node and the next points are placed at satellite footprints at every 0.5 degree interval.

Based on the fixed points, the scene of a single beam is defined such that the range of the scene is dependent on the swath width. In the fine and dual-polarimetric mode cases, Figure 5.2, the swath width is 70km (off-nadir angle of 34.3 degrees on the equator), and the quad-polarimetric mode has a 30km swath width (off-nadir angle of

21.5 degrees on the equator). Furthermore, the azimuth length of the scene is defined as

10 seconds from time of scene fixed point: –5 seconds to time of scene fixed point +5 seconds. This length is about 70km (ERSDAC). Swath width

Time of scene fixed point Time of scene fixed point . .

5sec 5sec

Azimuth length of scene: 10sec 5sec 5sec

Azimuth length of scene: 10sec

Figure 5. 2 Scene definition of single beam

5.1.1 Meta-data information at raw SAR data

Raw SAR data contains the meta-data (input-data) sets which can be identified by the source, the sensor, the orbit, the frame and the date of acquisition date. The raw data format (single look complex) indicates the corrections which have been generated by the processing and archiving facility (PAF). The PAF use different processors to produce SLC data, which might lead to slightly different results (Gens 1999). The orbital parameters are derived from precise state vectors, while the other parameters are

117 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION included in the auxiliary data of each SAR image. The parameter accuracy directly affects the quality of the DEMs.

Table 5.1 is the meta-data of raw ALOS/PALSAR data. The meta-data has source and sensor, instrument mode, polarity, incidence angle, latitude and longitude of image centre, and product level, etc.

Table 5. 1 Meta-data of raw SAR

PASL1100711141311120712050003 ALOS PALSAR PALSAR_GDS FBSH 7 HH 34.3000000 38.700 9619 370 649.00 Ascending 1 2114.1649049 0.0000000 88.7313310 1595901912 64 OFF 2007/11/14 13:11:07.853 2007/11/14 13:11:17.853 2007/11/14 13:11:12.853 -34.3683260 150.6458120 -34.1613070 150.1688290 -34.7551260

118 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

150.3702220 -34.5775460 151.1085950 -33.9850460 150.9019870 -33.9850460 -34.7551260 151.1085950 150.1688290 -4591117 2585959 -3581707 -4750954 3284072 -4094353 PASL1000711141311120711190018 L1.1 9440 21132 PASL10S0711141310230711190001 0.0000 1.3600 High Accurate Orbit 2007/12/05 05:33:10.000 Vexcel 3dSAR 6.5.2 normal process 4.6842570

5.1.2 Relationship of DEM grid location and Orbit information

The adaptation of radar data for InSAR processing and SAR applications depends on the SAR geometry during data acquisition. The most important parameter of SAR geometry is the baseline between the two SAR data. This baseline can be represented by its length and orientation angle. Alternatively, the vertical and horizontal baselines

119 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION or the parallel and perpendicular components of the baseline may be given to express the baseline geometry of images (Gens 1999).

The relative motion between sensor and target is the supposed to be known in the

SAR process. Common SAR focussing processes evaluate the back-scatter coefficient of an elementary cell by picking up samples from the received data with an appropriate sequence of time and phase delays.

The target coordinates of a pixel are calculated using the slant-range, the Doppler and the interferometric equations for pixels of the interferogram. It requires precise orbit information of master and slave satellites (Crosetto 2002). Figure 5.3 shows the

Doppler centre geometry.

M S

Doppler centre plane

Object• (x,y,z)

Figure 5. 3 The geometry of Doppler centre

For interferogram pixels, the azimuth coordinates ‘y’ of the master and slave images are known. From this, the acquisition time T is calculated through:

−Δ+= 0 ytTT )1( (Equation 5.1)

Δ where, T0 is the acquisition time of first line of image, and t is the time interval in azimuth direction. Terrain elevation is calculated from the phase difference equation, which depends on orbital information and particularly on the interferometric baseline

(Abdelfattah and Nicolas 2002). 120 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

5.2 Multi-pass InSAR DEM generation

InSAR DEM generation using satellite system is especially well-suited for cost- effective, precise and large-coverage DEM generation. However, the single pass

(single-baseline) InSAR DEM method is limited by its terrain sensitivity, errors and problematic data acquisition. Furthermore, different DEMs are affected by different systematic vertical and horizontal errors, as well as random noise. Each DEM generally has its own spatial resolution and coverage. To improve the reliability of individual

InSAR DEM, the multi-pass (multi-baseline) is the most useful method. The multi-pass method aims to combine information from several single-pass products in order to extract common information. This technique involves interferometric phase comparison of two images acquired at different times and with different perpendicular baselines

(Hanssen et al. 2003; Raucoules et al. 2008; Van Leijen and Hanssen 2004).

The external DEM, unlike the conventional InSAR method, is used as the known topographic phase and subtracted from the interferogram. Image pairs (master and slave) are selected according to temporal baseline criteria to increase the coherence.

Interferometric processing requires optimally short time-delay between image acquisitions to ensure maximum coherence between acquisitions and sufficiently short perpendicular baselines (d’Ozouville et al. 2008; Rabus et al. 2003; Sandwell et al.

2008).

Moreover, the height variation of each DEM is related to the baseline difference during the DEM merging process. That is: −=Δ BBB jin (Equation 5.2)

R λ sin θ h =Δ m a Δ (Equation 5.3) 2 B n 121 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Each data pair has a different perpendicular baseline and slightly different orbit and look angle. This means that each interferogram, generated by different pairs of SAR images, have specific fringe patterns for the same target terrain. In the phase unwrapping step, various fringe patterns affect the process difficulty and sensitivity of the derived DEM differently. Therefore, DEMs processed with different perpendicular baseline pairs can replace the height gap of other DEMs (Massonnet and Feigl 1998).

The term “scale” refers to a combination of both spatial extent and detail or resolution. Multiple scales allows for a greater amount of information to be extracted from DEMs about the terrain. The SRTM DEM has a spatial resolution of 90m in

Australia, while the single ALOS and ERS DEMs have approximately 20m-30m spatial resolution. Moreover, merged InSAR DEMs have superior spatial resolution compared with single processed DEMs (Kiamehr and Sjoberg 2005; Rabus et al. 2003; Rufino et al. 1998). Figure 5.4 shows the geometry of multi-pass interferometric SAR system.

A B C

Figure 5. 4 The geometry of multi-pass InSAR

Generally, for the multi-baseline InSAR DEM generation process or data stacking method, one fixed master image is selected for high co-registration between slave images. However, in this research, both same and different master multi-pass methods were utilised. 122 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figure 5.5 shows the geometry of multi-baseline InSAR DEM generation using multi-master image processing. Each DEM is generated using different master images and the DEM elements are determined by master image parameters.

S1 S2 S3

DEM 1: DEM 2: S1: DEM 1 master S2: DEM 1 slave & DEM 2 master S3: DEM 2 slave

Figure 5. 5 Grid location relationship of multi-pass InSAR

5.2.1 Multi-pass InSAR DEM processing using single master image

Data merging techniques are generally used to improve data accuracy and information detail. Normally, in multi-pass InSAR DEM generation, the same master is selected to easily match the processing area and grid location of other generated DEMs.

Table 5.2 presents the information of SAR data. In order to improve the coherence of image pairs, short temporal baselines were selected for InSAR DEM generation.

123 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Table 5. 2 ALOS/PALSAR Data pairs

Master date Slave date Bperp (m) Btemp (days)

14/11/2007 27/12/2006 945 322

14/11/2007 11/02/2007 1444 276

14/11/2007 30/12/2007 757 46

14/11/2007 14/02/2008 742 92

14/11/2007 31/03/2008 1400 138

Figure 5. 6 The test area boundary for identical master multi-pass InSAR DEM

generation

Figure 5.6 shows the research area for the same master multi-pass InSAR DEM generation. They are covered in the same processing area due to the master image parameters. Five pairs of ALOS/PALSAR data are used for DEM generation.

124 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figure 5. 7 The same master InSAR generated DEM (14/11/07-27/12/06: left and

14/11/07-11/02/07: right)

Figure 5. 8 The same master InSAR generated DEM (14/11/07-30/12/07: left and

14/11/07-14/02/08: right)

125 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figure 5. 9 The same master InSAR generated DEM (14/11/07-31/03/08)

Figures 5.7, 5.8 and 5.9 show the InSAR generated DEM using the same master image. Each DEM reveals a different terrain appearance in some parts of the research area. This is affected by the temporal baseline between the two images (master and slave). If a long temporal baseline is present, the probability of ground movement and changes in ground conditions due to seasonal or geological effects will increase. Three methods are used to merge the InSAR DEMs. One is the “mean stacking DEMs” method:

n ¦ i yxh ),( H = i =1 (Equation 5.4) n

where x and y is the coordinate of the pixel/point in the InSAR DEM, n is the number of InSAR DEMs, and hi is the height value of point (x,y).

The “coherence weight stacking” method uses the coherence value as the weight of the merging value:

+ chch H = iiii ++ 11 + (Equation 5.5) cc ii +1

th where ci is the coherence value in i DEM. 126 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

The “baseline distance weight stacking” method uses the baseline distance as the weight of the merging value:

n ¦ Bh ii = i =1 H n (Equation 5.6) ¦ B i i =1

Where Bi is the baseline distance

Figure 5. 10 The mean merged InSAR generated DEM using same master multi-pass

method

Figure 5. 11 The mean merged InSAR generated DEM using (14/11/07-30/12/07 and

14/11/07-14/02/08: left) pairs and (14/11/07-11/02/07 and 14/11/07-31/03/08: right)

pairs

127 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figure 5.10 shows the mean merged InSAR generated DEM using same master multi-pass method. Figure 5.11 presents the merged InSAR generated DEM using similar baseline distances pairs. One of mean merged DEM (14/11/07-30/12/07 and

14/11/07-14/02/08) reveals the ground deformation affected elevation error. Figure

5.12 and 5.13 show the merged InSAR DEM using the coherence value weight and the baseline distance weight.

Figure 5. 12 The merged InSAR generated DEM using coherence value weight

Figure 5. 13 The merged InSAR generated DEM using baseline distance weight

128 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Table 5.3 is the root mean square error of each merged DEM comparing with

LiDAR DEM of the test area. Figure 5.14, 5.15 and 5.16 is the elevation difference graph compare with LiDAR DEM and re-generated InSAR DEM in sample area.

Table 5. 3 The Root Mean Square Error of merged DEMs

RMSE (unit: m) LiDAR - Mean merged DEM 9.7 LiDAR - Baseline weight merged DEM 11.6 LiDAR - Coherence weight merged DEM 15.6

Figure 5. 14 The elevation differences between LiDAR DEM and Mean merged InSAR

generated DEM

Figure 5. 15 The elevation differences between LiDAR DEM and coherence weight

InSAR generated DEM 129 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figure 5. 16 The elevation differences between LiDAR DEM and baseline weight

InSAR generated DEM

Figure 4.18 presents GPS surveying points in test area. Table 5.4 shows the RMSE compare with GPS surveying data and the identical master processing InSAR DEM.

Table 5. 4 The Root Mean Square Error of identical master InSAR DEMs compared with GPS measurement.

RMSE (unit:m) GPS - Mean merged DEM 4.8 GPS - Baseline weight merged DEM 7.9 GPS - Coherence weight merged DEM 14.4

5.2.2 Multi-pass InSAR DEM processing using different master image

In general, the spatial resolution of a DEM is inversely proportional to the area covered. If all terrain areas had the identical, high spatial resolution, the size of the resulting DEM would be huge. However, if only certain areas of interest require high spatial resolution, such as CBD areas, industrial areas, or disaster-affected areas, then other areas, such as forested or mountainous terrain, may only need a low resolution

130 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

DEM. Nevertheless, commercial DEM products provide the same spatial resolution throughout the scene. For example, the SRTM DEM has a 90m spatial resolution in

Australia, while the single Advanced Land Observing Satellite (ALOS/PALSAR) has approximately a 10m spatial resolution. Therefore, the DEM merging technique, which uses different resolution DEMs or DEMs generated from different sensors is necessary.

Moreover, merged InSAR DEMs increase the spatial resolution compared with single processed DEMs (Kiamehr and Sjoberg 2005; Rabus et al. 2003).

InSAR DEMs generated with different master images have different grid sizes and locations, despite having the same coordinate systems. Moreover, each InSAR pair has a different perpendicular baseline and a slightly different orbit, look angle, and acquisition conditions such as water vapour delays and measurement noise. This means that each interferogram has specific fringe patterns for the same target terrain. In the phase unwrapping step, individual fringe patterns influence the process and the sensitivity of the derived DEM differently. Therefore DEMs processed with different perpendicular baseline pairs can, in principle, fill in the “height gap” of other DEMs

(Massonnet and Feigl 1998). The overlapping of the different DEMs would furthermore increase the terrain detail and reduce the vertical errors.

A method that exploits the information contained in the area of overlap between different master InSAR DEMs is proposed. This not only reduces the total DEM data size and processing time through the removal of excess data in areas of reduced importance, but also reduces vertical systematic errors despite the use of a variety of pairs with different perpendicular baselines.

131 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

SRTM DEM grid First InSAR DEM grid Second InSAR DEM

Figure 5. 17 Sketch of SRTM and InSAR elevation interpolation

Figure 5.17 shows the reconstructed DEM taking advantage of the different InSAR

DEM altitude of ambiguity to fill in the elevation gap of the low resolution DEM and other DEMs. This interpolation is affected by the perpendicular baseline, and the slightly different orbit, of each master image. Parameters of the InSAR DEMs are affected by the master image parameters - even if the master and slave images were to be swapped for DEM generation, the produced DEMs would exhibit slight differences.

Therefore, every InSAR DEM generated by a different master image has a different grid location and altitude of ambiguity. These factors provide variance in the elevation contour and grid density.

Table 5.5 lists the data information of multi-pass InSAR DEM generation using different master images. In this case, data pairs are selected by their short temporal baseline to increase the coherence between the master and slave images. In particular the ERS data uses Tandem images.

132 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Table 5. 5 ALOS/PALSAR and ERS Data

Sensor Master date Slave date Bperp (m) Btemp (days) ALOS 27/12/2006 11/02/2007 549 46 ALOS 14/11/2007 30/12/2007 757 46 ALOS 30/12/2007 14/02/2008 78 46 ALOS 14/02/2008 31/03/2008 659 46 ALOS 31/03/2008 01/07/2008 2992 92 ERS 29/10/1995 30/10/1995 49 1 ERS 03/12/1995 04/12/1995 145 1

Figure 5. 18 ERS-1 Master (29/10/95) and Slave (30/10/95) images

Figure 5. 19 ALOS-PALSAR Master (14/11/07) and Slave (30/12/07) images

Figures 5.18 and 5.19 show the intensity images of the test area for the multi-pass

InSAR DEM generation using different master images. The ERS image has a larger coverage than the PALSAR image because it has larger pixels and line spacing.

Therefore, the ERS could be processed in the Appin and Wollongong areas in one data scene. 133 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Table 5. 6 Height ambiguity at each DEM pairs

Pair of SAR data Bperp (m) ha (m) 27/12/2006-11/02/2007 549 133.55 14/11/2007-30/12/2007 757 96.91 30/12/2007-14/02/2008 78 938.61 14/02/2008-31/03/2008 659 111.22 31/03/2008-01/07/2008 2992 24.52 29/10/1995-30/10/1995 49 187 03/12/1995-04/12/1995 145 63

Table 5.6 outlines the altitude of ambiguity for each SAR data pair of L-, C-band image. Table 5.7 lists the elevation ambiguity caused by phase changes. Figure 5.20 shows the elevation ambiguity. It is an important element for processing pair selection.

Table 5. 7 Height ambiguity due to phase changes (ALOS/PALSAR)

Phase change ( Δφ ) Bperp(m) ʌ/4 ʌ/2 3ʌ/4 ʌ 5ʌ/4 3ʌ/2 7ʌ/4 2ʌ 549 16.57 33.37 50.16 66.74 83.53 100.26 116.98 133.71 757 12.03 24.21 36.40 48.43 60.61 72.75 84.89 97.03 78 116.51 234.53 352.54 469.06 587.08 704.64 822.21 939.77 659 13.80 27.79 41.77 55.58 69.56 83.49 97.42 111.35 2992 3.04 6.12 9.21 12.25 15.33 18.40 21.47 24.55

Figure 5. 20 Height ambiguity due to phase changes. 134 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figure 5. 21 Grid points in different InSAR DEMs

The locations of grid points in each DEM are determined by information from the master image. Figure 5.21 shows the locations of grid central point of different master

InSAR DEM products. Each grid points is expressed by different elevation values and can be interpolated with other points. It is also useful for improving the terrain reality or detail.

Figure 5. 22 The research areas in Appin and Wollongong, NSW, Australia

135 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

In Figure 5.22, the boundaries of each processed area are different because each area has a different master image. The parameters of the final DEM products are determined according to the master image parameters. Furthermore, the coverage of ALOS and

ERS vary because of the parameters of the master image and the capability of the sensor coverage. This is described in table 4.1.

Figure 5. 23 ALOS/PALSAR InSAR generated DEM (14/02/08-31/03/08; left) and

(14/11/07-30/12/07; right)

Figure 5. 24 ALOS/PALSAR InSAR generated DEM (27/12/06-11/02/07, left) and

(31/03/08-01/07/08, right)

136 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figures 5.23 and 5.24 present the L-band PALSAR InSAR generated DEMs in the test areas. Each DEM appears to detect slightly different ground information. This is related to data parameters such as the image acquisition date, the perpendicular baseline between master and slave, target area conditions, and local weather conditions. The perpendicular baseline is the parameter with the strongest influence on the product.

Figure 5. 25 ERS InSAR generated DEM (29/10/95-30/10/95; left) and (03/12/95-

04/12/95; right)

Figure 5.25 shows the ERS C-band InSAR generated DEM in the test area.

Compared to the L-band InSAR generated DEM, the results appear less detailed due to the inherent characteristics of the observation sensor and wavelength. However, the coverage of C-band data and less processing times are larger while the L-band is more detailed.

137 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figure 5. 26 Merged InSAR generated DEM using mean value from a different master

Figure 5. 27 Merged InSAR generated DEM using coherence weight method from a

different master

Figure 5. 28 Merged InSAR generated DEM using baseline distance weight method

from a different master 138 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figures 5.26, 5.27 and 5.28 show the merged InSAR DEMs using various merge techniques such as mean, coherence weight method and baseline distance weight method from a different master.

Table 5.8 is the root mean square error of each merged DEM compared with LiDAR

DEM in the test area. Figures 5.29, 5.30 and 5.31 graph the elevation difference graph compared with LiDAR DEM and re-generated InSAR DEM in the sample area.

Table 5. 8 The Root Mean Square Error of merged DEM

RMSE (unit: m)

LiDAR - Mean merged DEM 10.5

LiDAR - Baseline weight merged DEM 8.4

LiDAR - Coherence weight merged DEM 10.3

Figure 5. 29 The elevation differences between LiDAR DEM and Mean merged InSAR

generated DEM

139 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Figure 5. 30 The elevation differences between LiDAR DEM and coherence weight

method merged InSAR generated DEM

Figure 5. 31 The elevation differences between LiDAR DEM and baseline weight

method merged InSAR generated DEM

Figure 4.18 displays GPS surveying points in the test area. Table 5.9 is the RMSE compared with GPS surveying data and different master processing InSAR DEM.

140 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

Table 5. 9 The Root Mean Square Error of different master InSAR DEMs compared with GPS measurement

RMSE (unit:m)

GPS - Mean merged DEM 5.7

GPS - Baseline weight merged DEM 3.8

GPS - Coherence weight merged DEM 5.0

5.3 Concluding remarks

The multi-baseline technique can be usefully exploited to generate InSAR DEM.

This method is less affected by data errors as it averages uncorrelated atmospheric contributions in the single-baseline interferogram. The multi-baseline InSAR technique has the capacity to overcome the weaknesses of single-baseline InSAR systems. The multi-baseline technique is an advancement that enables interferometric techniques to map the surface and monitor temporal evolution.

Generally, the multi-baseline method provides more terrain information than the single-baseline method. Two kinds of multi-baseline InSAR DEM generation were researched in this paper. These were the single (fixed) master multi-baseline method and the different master multi-baseline method. Each method has its own strengths in data processing, target area selection, data arrangement, and data matching.

In the single master multi-baseline case, the mean merged DEM attains smallest

RMSE 9.7m for LiDAR in table 5.3 and RMSE 4.8m for GPS in table 5.4.

141 CHAPTER 5 MULTI-PASS InSAR DEM GENERATION

The different master multi-baseline method produces more accurate results and improved ground reality than the single master multi-baseline or single-baseline methods. Using different master multi-baseline case, baseline weight merged DEM attains smallest RMSE 8.4m for LiDAR and RMSE 3.8m for GPS in table 5.8 and 5.9.

Furthermore, different merging techniques were used for the multi-baseline method.

The coherence weighting and baseline distance weighting methods were more powerful and uncomplicated than the mean method.

After the merging process, some areas of merged DEMs decreased the accuracy of elevation due to low coherence or elevation noise. This will be a focus for future research in multi-baseline InSAR DEM generation. However, practicable coverage produced by the single master multi-baseline method is larger than the different master multi-baseline method due to the overlapping of different master images.

142 CHAPTER 6 RADARGRAMMETRY FOR DEM GENERATION

There are two major methods of DEM generation used by SAR imagery. One is based on the InSAR technique and the other on the Radargrammetry (stereo-SAR) technique. Both techniques measure and detect the same geometrical deviations.

Generally, highly accurate DEMs can be generated using InSAR, so this has been a major research area. However, InSAR DEM generation is less robust and more difficult to implement particularly because the InSAR technique often leads to poor results caused by poor coherence, atmospheric differences, sensitivity to direction of sensor motion and ambiguous elevation due to fringe wrapping. Furthermore, there are stringent conditions on incidence angles and Doppler similarities (Massonnet and

Souyris 2008). InSAR also requires the expectations of a certain baseline. For this reason, the radargrammetry technique is an important alternative for DEM generation

(Chen and Dowman 2001). The major difference between the two techniques is that radargrammetry calculates the image range offset using position matching of the same ground targets in two images, while InSAR calculates the phase difference between two images (Kyaruzi 2005; Sansosti 2004).

Normally, if a radargrammetry image pair meets the base-height ratio condition, a

DEM may be generated (Huang et al. 2004). Radargrammetry is based on stereogrammetry, which is the classical method for relief reconstruction using remotely sensed images. This technique can be applied to radargrammetry images to generate

DEMs of good quality (Paillou and Gelautz 1999). One advantage of radargrammetry is that it remains less affected by atmospheric influences compared with interferometry.

The effect of the atmosphere on SAR imagery is essentially the same in either

143 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION radargrammetry or in the InSAR. However, radargrammetry utilises the magnitude value, as opposed to the InSAR which uses the phase value. As magnitude is less affected by the atmosphere than the phase value, radargrammetry is more robust against atmospheric influences than InSAR (Massonnet and Souyris 2008).

Radargrammetry also utilises the amplitude of SAR images that photogrammetry used with optical images. Radargrammetry is used with stereoscopic pairs acquired from different incidence angles. At optical wavelengths, radargrammetry is distinct from photogrammetry in several aspects (Schanda 1985): (1) the interaction effects with the surface at radar wavelengths are different from those at optical wavelengths,

(2) radar measures the distance between sensor and target, therefore the parallax appears reversed when compared with optical images, (3) the long wavelengths cause poor angular resolution at a given size of ‘optics’, therefore a useful stereo base cannot be established simultaneously from one simple platform.

Photogrammetry Radargrammetry

Figure 6. 1 The difference between terrain response between photogrammetry and

radargrammetry

Figure 6.1 shows the different terrain responses obtained from photogrammetry and radargrammetry due to geometric characteristics and inherent difference in the

144 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION mechanism of information acquisition. In the optical case, the target images lie opposite to the sensor while radar targets lie on the same positional side as the sensor.

Figure 6.2 illustrates optical and SAR images which have different terrain responses at same area. The optical image is taken from vertical direction and SAR image is taken from left side-looking direction. Therefore, the acquired image has different response in same target area.

Figure 6. 2 Examples of different terrain responses and projections between optical and

radar images: optical images (left: Google Earth) and TerraSAR-X intensity image

(right) in Tokyo, Japan

Since the 1980s, improvements in SAR techniques have allowed the demonstration of stereoscopic radar with same-side or opposite-side looking image geometry (Toutin and Gray 2000). Stereoscopic pairs for the radargrammetry technique should take into account that the geometry and parallax produced from a particular system configuration and the awareness of the image pairs by the interpreter. Furthermore, input image pairs considered for radargrammetric extraction of terrain elevation must meet stringent conditions, including approximately 10~20 degrees incidence angle difference and overlapping of two images (Mercer 1995; Kaupp et al. 1983). DEM generation using radargrammetry was improved by Mercer (1995), space-borne Thermal Emission and

145 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION Reflection Radiometer (ASTER) (Hirano et al. 2003), RADARSAT and airborne data in mountainous areas (Huang et al. 2004; Hu et al. 2006).

Figure 6.3 shows a one-dimensional relief displacement due to the side looking in radar geometry as well as projection difference with photogrammetry. Ground distortion is perpendicular to the direction of flight orbit resulting in tall objects being displaced toward the sensor, unlike photogrammetry, where vertically-oriented objects are displaced in a radial direction away from the nadir.

Figure 6. 3 The geometric projection differences of radar system (left) and

photogrammetry system (right)

6.1 Geometry of Radargrammetry

In the 1960s (La Prade 1963), stereoscopic methods were first applied to radar images to generate ground elevation leading to the development of radargrammetry

(Toutin and Gray 2000).

SAR images are strongly distorted by terrain both radiometrically and geometrically.

Radiometric distortion, due to the direct reflection of the antenna signal, is such that slopes facing the sensor have a strong value (bright), while slopes facing away from the

146 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION sensor have a low value (dark), or are even shadowed. Geometrically, foreshortening and layover are exhibited by slopes facing the sensor, while slopes facing away are stretched. Geometric and radiometric distortions make matching of stereo-SAR images more difficult and less accurate than matching of optical images (Ostrowski and Cheng

2000).

To generate DEMs using radargrammetry, there are three commonly applied models:

1) The model based on range and Doppler equations.

2) The equivalent line central projection models based on the photogrammetry theory.

3) The parallax and elevation relation model which uses the relation between parallax and elevation to first calculate the elevation difference and then the plane coordinates.

S2 Image1 S1

Image2

H1 H2

Figure 6. 4 Radargrammetry geometry of look angle difference and image overlapping

Radargrammetry requires image acquisitions with varying incidence angles. Figure

6.4 shows the geometry of radargrammetry. The quality of radargrammetry DEMs depend on the base-to-height ratio or intersection angle of the radargrammetry pair. To acquire good geometry for radargrammetry pairs, the intersection angle between the 147 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION two SAR images should be sufficient to produce observed parallax, which is then used to determine the terrain elevation. However, to have a good stereo-viewing, almost identical images are necessary in processing. A small intersecting angle (approximately

10-20 degrees) and a shallow look angle (i.e. angle between the vertical and the beam direction >20˚) are usually considered optimal configurations for medium to high relief areas (d’Ozouville et al. 2008). Thus, a compromise has to be reached between better stereo-viewing and more accurate elevation determination (Toutin and Gray 2000).

z

S1

S2 Bx H Bz

R2 R1 y

x2 P P 2 h x1 dp P1 x

Figure 6. 5 The different observation positions and geometry for radargrammetry

(Maitre 2008)

In Figure 6.5, target P is seen as P1 and P2 in the SAR images. This is called

‘disparity’ distance P1 P2. If the ground elevation is zero (h=0), disparity will be zero with increments for increasing heights h. It is expressed by:

2 22 2 2 2 −+−−++−−−−+= dp x z )()()()( BBHhBHBxHhHx xz (Equation 6.1)

148 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION More important for the elevation calculation is the reverse relation which gives elevation h for the disparity dp.

2/142 ++++Δ+−+ 2 BH x dpdp Bx dp zx xBB dp)(2)()4()(2 Bzdp h = (Equation 6.2) 22 ++− 2 BB zx x dpdpB )2(2

with

x dpdp 322 2 BHB dp)( ++++−=Δ B dp 222 x x x (Equation 6.3) + 222 ++ ++− + HB z dp Bx z xB dp xx xB dp HB zx xB dp)(2)2(

The above equation is simplified when the sensor location is horizontal (Bz=0):

HB Hdp −+ 422 BH 22 dpΛ+ h = x x + (Equation 6.4) dp B x

with

22 22 22 +−+++++−=Λ x (8 xHB xBx dp Bx x dpdpBdp (4)44() xH xBx ) (Equation 6.5)

These equations are simplified even further if the images are acquired from satellite sensors (sensor elevation H is significant compared to target altitudes h). Therefore, the equation is a simple expression in accordance with angles:

dp h = θ − θ (Equation 6.6) cotan 2 cotan 1

where, și is the look angle of i observation position.

6.1.1 Perpendicular radargrammetry

Two radar images taken from different orbits have different incidence angles on targets of the terrain. These response range pixels are projected onto the ground with varying sizes.

149 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION Figure 6.6 shows that acquired SAR images have a different scale of terrain in single pixels, even on flat terrain. The elevations of terrain regenerate this difference in scale and lead to stereoscopic effects, allowing reconstruction of the topography from a detailed analysis of the images’ geometric distortions (Massonnet and Souyris 2008).

Image 2 B2

A2

B1

A1 B Image 1

Ԧ1 Ԧ2

A

Figure 6. 6 Principles of perpendicular radargrammetry (Massonnet and Souyris 2008)

6.1.2 Oblique radargrammetry

Two SAR images acquired from different viewpoints display not only local range but also local azimuth differences, which are affected even if the two SAR images are referenced at zero Doppler. The local azimuth differences are consequences of the angle of orbital paths. In zero Doppler geometry, any points on the terrain are located in a plane perpendicular to the velocity at a given range and each plane corresponds to the time t when the point was at its closest to the radar. In a single image, the location of the target cannot be determined as the targets locate in the same distance area (circle).

However, the target location can be determined using two images with a stereoscopic baseline and different orbit path. The target will change its corresponding time on the 150 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION second image as a function of its elevation, and the change in elevation is detected from its azimuth shift (Massonnet and Souyris 2008)

6.1.3 Same-side stereo

Two stereo-images taken from the same side of a ground target has a smaller intersection angle than two images taken from opposite sides. Acceptable stereo was observed by an interpreter using a standard stereo scope for intersection angles of 15˚ or larger from each model area. Most stereo radar images are acquired by same-side orbits (Kaupp et al. 1983). Figure 6.7 presents the sensor location in same side radargrammetry and Figure 6.8 demonstrates the parallax difference according to the sensor location in same side radargrammetry (Toutin 1996; Toutin and Gray 2000).

H

h

Figure 6. 7 The geometry of same side direction of radargrammetry

151 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

S1

S2

Para1

Para2

Figure 6. 8 The parallax difference in same side radargrammetry

6.1.4 Opposite-look stereo

This image configuration is obtained by imaging to one side of the terrain target on one pass and the opposite side on a succeeding pass. Consequently, the back slopes and shadowed areas of one radar image are the fore-slopes in the other. As the slope and relative relief of the terrain increase, one image increasingly fills in missing data of the other, rather than creating parallax (Kaupp et al. 1983). Figure 6.9 shows the relationship between sensors and the target in opposite side radargrammetry. Figure

6.10 demonstrates the parallax difference in opposite side radargrammetry (Toutin

1996; Toutin and Gray 2000).

152 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

H

h

Figure 6. 9 The geometry of opposite side direction of radargrammetry

S 1 S2

Para1

Para2

Figure 6. 10 The parallax difference in opposite side radargrammetry

6.2 Radargrammetry DEM generation using different orbit SAR images

The radargrammetry DEM processing steps can be described as follows: (1) acquiring stereoscopic images; (2) subset in the areas of interest; (3) despeckle to remove the noise; (4) coregistration of two subset images; (5) matching between coregistered images; (6) height calculation; (7) geocoding and DEM generation.

Figure 6.11 shows the processing flowchart using different orbit SAR images for radargrammetry.

153 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION Currently, there are two methods principally used to extract the elevation parallax using image matching: computer-supervised or automatic. Computer-supervised matching is an extension of the traditional photogrammetry method to measure elevation data using stereoscopic plotter (Tountin and Gray 2000). Furthermore, correlation and feature-based methods have been applied successfully in photogrammetry image matching.

Reference image Match image

Subset image Subset image

Despeckle Despeckle

Coregistration

Matching

Height calculation

DEM generation

Figure 6. 11 Flow chart of radargrammetry processing

In this research, four pairs of ALOS/PALSAR and four pairs of ENVISAT/ASAR images were used. Radargrammetry allowed for larger coverage processing than the interferometry method. The research areas included Appin and Wollongong in NSW,

Australia. Images were acquired from three different orbits paths with different incidence angles. Table 6.1 lists the image information used in radargrammetry for this study. The same-side stereo method was used in radargrammetric DEM generation.

154 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION Table 6. 1 The information of radargrammetry images

Reference image Match image Incidence Incidence Sensor (Path No.) (Path No.) angle_Re angle_Ma ALOS 31/03/08 (370) 05/04/08 (373) 38.7˚ 47.3˚ ALOS 07/04/08 (365) 05/04/08 (373) 24.0˚ 47.3˚ ALOS 23/05/08 (365) 21/05/08 (373) 24.0˚ 47.3˚ ALOS 01/07/08 (370) 06/07/08 (373) 38.7˚ 47.3˚ ENVISAT 18/12/09 (152) 26/09/09 (467) 33.6˚ 43.8˚ ENVISAT 02/04/10 (152) 31/10/09 (467) 33.6˚ 43.8˚ ENVISAT 15/03/10 (402) 12/03/10 (359) 22.9˚ 33.6˚ ENVISAT 08/02/10 (402) 05/02/10 (359) 22.9˚ 33.6˚

Figure 6. 12 The radargrammetry intensity images of 31/03/08-05/04/08

radargrammetry pair (31/03/08: reference image-left, 05/04/08: matching image-right)

Figure 6.12 shows the intensity images for the radargrammetry technique. The reference and matching images have incidence angles from the satellite sensor of 38.7˚ and 47.3˚ respectively. The terrain shapes have different appearances such as in the distance and width between rivers and the size of the grassy area. These phenomenons generate the stereoscopy of radargrammetry. The reference and matched images were registered for pixel offset calculation. Figure 6.13 illustrates the coregistration between reference and matched images.

155 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Figure 6. 13 The coregistration processing of reference image and matching image in

radargrammetry

Automatic matching methods require a correlation between two images in order to generate a disparity map. Matching can be performed using initial gray-level images, edge images, or other image features such as linked-edge elements or regions. The main problems encountered when matching radar images for radargrammetry are speckle noise and the difference between two stereo partners, as the backscattered radar signal mainly depends on the local incidence angle (Paillou and Gelautz 1999; Tountin and Gray 2000).

Figure 6. 14 The correlation image between a reference image and a match image after

the matching process 156 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION Figure 6.14 shows the correlations between the reference and the matched images.

The correlation number gives a measure of how strong a match exists between the reference and matched points. The correlation scale ranges from 0 to 1, where better matches were indicated by results closer to 1. Higher correlations were detected in city areas such as Appin and Wollongong, and also in surrounding rivers.

Figure 6. 15 The radargrammetry DEM generated from ALOS/PALSAR (31/03/08-

05/04/08: left) and (01/07/08-06/07/08: right)

Figure 6.15 shows the DEM generated by radargrammetry conducted between

31/03/08 (370) - 05/04/08 (373) and 01/07/08 (370) - 06/07/08 08 (373). The intersection angle between the 370 path and 373 path is 8.6˚.

157 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Figure 6. 16 The radargrammetry DEM generated from ALOS/PALSAR (07/04/08-

05/04/08: left) and (23/05/08-21/05/08: right)

Figure 6.16 shows the DEM generated by radargrammetry conducted between

07/04/08 (365) - 05/04/08 (373) and 23/05/08 (365) - 21/05/08 (373). The intersection angle between the 365 path and 373 path is 23.3˚.

Figure 6. 17 Radargrammetry DEM generated from Envisat/ASAR (08/02/10-05/02/10:

left) and (15/03/10-12/03/10: right)

Figure 6.17 shows the DEM generated by radargrammetry conducted between

08/02/10 (402) - 05/02/10 (359) and 15/03/10 (402) -12/03/10 (359). The intersection angle between the 402 path and 359 path is 10.7˚. 158 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Figure 6. 18 Radargrammetry DEM generated from Envisat/ASAR (02/04/10-31/10/09:

left, 18/12/09-26/09/09: right)

Figure 6.18 shows the DEM generated by radargrammetry conducted between

02/04/10 (152) - 31/10/09 (467) and 18/12/09 (152) - 26/09/09 (467). The intersection angle between the 152 path and 467 path is 10.2˚.

Table 6.2 is the root mean square error of InSAR DEMs and radargrammetry DEMs compared with LiDAR DEM (reference) in the sample area. Figure 6.19 and 6.20 are the elevation difference graphs of InSAR DEMs compared with LiDAR DEMs. Figure

6.19-6.28 are calculated by random check points in test area. The random check points cover the whole DEM area. Figures 6.21~6.28 illustrate the elevation difference graph of radargrammetry DEMs compared with LiDAR DEM.

159 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Table 6. 2 The Root Mean Square Error of radargrammetry DEMs

SAR techniques RMSE (unit: m) ALOS InSAR DEM (14/02/08-31/03/08) 16.97 ERS InSAR DEM (03/12/95-04/12/95) 22.04 Envisat radargrammetry DEM (08/02/10-05/02/10) 48.54 Envisat radargrammetry DEM (15/03/10-12/03/10) 36.03 Envisat radargrammetry DEM (02/04/10-31/10/09) 62.46 Envisat radargrammetry DEM (18/12/09-26/09/09) 44.95 ALOS radargrammetry DEM (31/03/08-05/04/08) 64.26 ALOS radargrammetry DEM (07/04/08-05/04/08) 24.92 ALOS radargrammetry DEM (23/05/08-21/05/08) 26.44 ALOS radargrammetry DEM (01/07/08-06/07/08) 56.44

Figure 6. 19 The elevation differences between LiDAR DEM and ALOS InSAR DEM

(14/02/08-31/03/08)

160 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Figure 6. 20 The elevation differences between LiDAR DEM and ERS InSAR DEM

(03/12/95-04/12/95)

Figure 6. 21 The elevation differences between LiDAR DEM and Envisat

radargrammetry DEM (08/02/10-05/02/10)

161 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Figure 6. 22 The elevation differences between LiDAR DEM and Envisat

radargrammetry DEM (15/03/10-12/03/10)

Figure 6. 23 The elevation differences between LiDAR DEM and Envisat

radargrammetry DEM (02/04/10-31/10/09)

162 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Figure 6. 24 The elevation differences between LiDAR DEM and Envisat

radargrammetry DEM (18/12/09-26/09/09)

Figure 6. 25 The elevation differences between LiDAR DEM and ALOS

radargrammetry DEM (31/03/08-05/04/08)

163 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Figure 6. 26 The elevation differences between LiDAR DEM and ALOS

radargrammetry DEM (07/04/08-05/04/08)

Figure 6. 27 The elevation differences between LiDAR DEM and ALOS

radargrammetry DEM (23/05/08-21/05/08)

164 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION

Figure 6. 28 The elevation differences between LiDAR DEM and ALOS

radargrammetry DEM (01/07/08-06/07/08)

Table 6.3 lists the RMSE compares with GPS survey data and radargrammetric

DEMs.

Table 6. 3 The Root Mean Square Error of radargrammetry DEMs compared with GPS

SAR techniques RMSE (unit: m)

Envisat radargrammetry DEM (08/02/10-05/02/10) 38.13 Envisat radargrammetry DEM (15/03/10-12/03/10) 25.14 Envisat radargrammetry DEM (02/04/10-31/10/09) 30.42 Envisat radargrammetry DEM (18/12/09-26/09/09) 47.25 ALOS radargrammetry DEM (31/03/08-05/04/08) 50.96 ALOS radargrammetry DEM (07/04/08-05/04/08) 34.02 ALOS radargrammetry DEM (23/05/08-21/05/08) 35.18 ALOS radargrammetry DEM (01/07/08-06/07/08) 62.46

165 CHAPTER 6. RADARGRAMMETRY FOR DEM GENERATION 6.3 Concluding remarks

Radargrammetry has superseded InSAR as the DEM generation method. This superseded condition is related with atmosphere effect, object ground condition and/or data acquisition condition in sensor. Under some condition, The InSAR technique cannot generate any DEM information even low quality DEM because of low coherence between two images. However, radargrammetry can generate DEM information in same condition. Therefore, radargrammetry can supersede InSAR method to generating DEM.

Radargrammetry is comparable to photogrammetry in that a stereo parallax found in a stereo image pair corresponds to the terrain elevation. The major difference is that

SAR images are used instead of optical images. Radargrammetry provides a large coverage considering the processing time and the software capability while the disadvantage of InSAR processing is the phase unwrapping step. However, the main problem of radargrammetry is that the DEM products are of low quality in the test area due to the spatial resolution of SAR images and the terrain slope. The disparity and convergence of objects are two key cues when viewing stereo imagery. For radar images, disparity predominates but the shade and shadow cues also have a strong and cumulative effect.

166 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

Quality is a key element for information users to consider when evaluating whether the provided dataset fits the intended application. In DEM products, the term quality describes how faithfully the elevation model reflects ground morphology as expressed by the input.

Many methods have been proposed for the assessment of the DEM quality.

According to the Desmet (1997), DEM accuracy is defined as a compromise between

“precision” and “shape reliability”.

DEM accuracy is assessed through a comparison with the DEM elevation values and by contrasting many ground true elevations. The pair-wise comparisons allow for the calculation of the mean error, Root Mean Squared Error (RMSE), standard deviation and similar statistics. In contrast, “shape reliability” is assessed through statistical analysis of a parameter set characterising the spatial properties of a surface such as roughness, slope and curvature (Cuartero et al. 2005).

7.1 Relative accuracy

Relative accuracy is a measurement of each point’s accuracy within the DEM. For many applications, relative accuracy is more important than absolute accuracy. A DEM with good relative accuracy is one that models the shape and dimensions of the terrain accurately, but may not necessarily be accurately registered to real geographic coordinates. The relative accuracy is expressed as the standard deviation of the vertical error.

167 Chapter 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

− ZZ )( 2* σ = ii (Equation 7.1) n

where n is the number of check points of the DEM, Zi is derived elevation at check

* point i and Zi is the terrain elevation at check point i.

In the assessment, all real elevations in the test area cannot be measured because the test area covers large area. Therefore, we selects the check points in test area and they use to accuracy assessment.

7.2 Absolute accuracy

The absolute accuracy is a measurement of the error between a DEM and the geographic coordinates of the terrain. The absolute accuracy is important in some applications such as . To produce a DEM with good absolute accuracy, reliable ground control can be used to remove bias. Absolute accuracy is expressed as the vertical RMSE.

Discrepancies between the InSAR or radargrammetry generated DEM elevations

(ZDEM) and the reference DEM elevation (Zreference) were condensed into summary statistics, such as minimum, mean and maximum values, standard deviation and the

RMSE

n − 2 ¦ ( Z reference Z DEM ) i RMSE = i=1 (Equation 7.2) n

where ZDEM is the DEM elevation and Zreference is the reference DEM elevation at the i-th test location of interest, and n is the number of test locations (Vazquez and Feyen

2007).

168 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

An elevation profile line provides an effective way of illustrating differences between digital elevation models.

Figure 7.1 shows the profile line for the comparison of DEMs from SAR techniques in this study. This line was selected according to the terrain shape, condition and slope direction because the radargrammetry method and the ascending and descending orbits merging techniques are more strongly related to observation direction than other SAR

DEM generation and merging techniques. Therefore, the profile line was laid in the two range directions and one azimuth direction. These combinations of two directions can show the relationship between the terrain slope and the observation direction while image distortion is also affected. Furthermore, this line located both forest and flat ground areas to show the terrain condition effects in SAR DEMs.

ID: 0

ID: 3490

Figure 7. 1 The profile line for comparing DEMs generation using SAR techniques

169 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

400.00

350.00

300.00

250.00

200.00 Elevation (m) Elevation

150.00

100.00 SRTM Radargrammetry (01/07/08-06/07/08) ERS (03/12/95-04/12/95) 50.00 ALOS_Same_Master ALOS (14/02/08-31/03/08) 0.00 1 131 261 391 521 651 781 911 1041 1171 1301 1431 1561 1691 1821 1951 2081 2211 2341 2471 2601 2731 2861 2991 3121 3251 3381 ID

Figure 7. 2 The profiles of DEM generation using Radargrammetry, C-band

InSAR(ERS), L-band InSAR (ALOS) and same master multi-pass InSAR (ALOS)

400.00

350.00

300.00

250.00

200.00 Elevation (m) Elevation

150.00

100.00 SRTM

ERS_Different_Master

50.00 ALOS_Different_Master

ALOS_Ascending_Descending

0.00 1 132 263 394 525 656 787 918 1049 1180 1311 1442 1573 1704 1835 1966 2097 2228 2359 2490 2621 2752 2883 3014 3145 3276 3407 ID

Figure 7. 3 The profiles of DEM generation using different master multi-pass InSAR

(C, L-band) and ascending and descending InSAR (ALOS)

Profiles of DEMs generated by different techniques using SAR imagery are illustrated in Figures 7.2 and 7.3. In these figures, most InSAR generated DEMs have similar terrain representation with the exception of radargrammetry. Using

170 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES radargrammetry, some locations (ID: 1550~1700, 2900~3000, 3300~3400) appeared to have reverse terrain elevations. These phenomena can be attributed to matching errors in radargrammetry. One of the main difficulties in radargrammetry processing is the matching process which associates pixels between the reference and matched images.

Due to the speckle and noise phenomenon, this step is particularly difficult with SAR images.

Table 7. 1 The Root Mean Square Error of SAR generated DEMs for SRTM DEM

SAR techniques RMSE (unit: m)

Radargrammetry (01/07/08-06/07/08) 30.9 ERS (03/12/95-04/12/95) 20.3 ERS merging (different master) 33.6 ALOS (14/02/08-31/03/08) 19.9 ALOS merging (same master) 23.3 ALOS merging (different master) 18.4 ALOS Ascending and Descending 16.4

The RMSE of DEMs is shown in Table 7.1. The RMSE calculation was based on the SRTM DEM (reference). Therefore, the RMSE results are higher than normal cases because of the spatial resolution (3-arc-second) of SRTM. In the RMSE results, some merging techniques lead to poor results by increasing the noise value. In the ALOS cases, merging of the results using the same master InSAR DEM generation recorded poorer results due to the increase in the temporal baseline between the master and slave images. A long temporal baseline leads to a low correlation between master and slave images. Also, the atmospheric effect in the same master method is not removed during merging processing. The atmospheric problem of the single master method can be

171 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES circumvented by the multi-master method. The multi-master method can minimise the perpendicular and temporal baselines between acquired images. Therefore, different master methods can be used for elimination of atmospheric effect.

100

80

60

40

20

0 0 200 400 600 800 1000 1200 1400 1600 1800 2000

-20 Elevation difference (m)

-40

-60 SRTM-ERS (03/12/95-04/12/95)

-80

-100

Figure 7. 4 Elevation differences between SRTM and ERS InSAR DEM (03/12/95-

04/12/95)

100

80

60

40

20

0 0 200 400 600 800 1000 1200 1400 1600 1800 2000

-20 Elevation difference (m)

-40

SRTM-ERS merging (different master) -60

-80

-100

Figure 7. 5 Elevation differences between SRTM and ERS merged DEM (different

master) 172 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

100

80

60

40

20

0 0 200 400 600 800 1000 1200 1400 1600 1800 2000

-20 Elevation difference (m) difference Elevation

-40

-60 SRTM-ALOS (14/02/08-31/03/08)

-80

-100

Figure 7. 6 Elevation differences between SRTM and ALOS InSAR DEM (14/02/08-

31/03/08)

100

80

60

40

20

0 0 200 400 600 800 1000 1200 1400 1600 1800 2000

-20 Elevation difference (m)

-40

-60

SRTM-ALOS merging (same master) -80

-100

Figure 7. 7 Elevation differences between SRTM and ALOS merged InSAR DEM

(same master)

173 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

100

80

60

40

20

0 0 200 400 600 800 1000 1200 1400 1600 1800 2000

-20 Elevation difference (m)

-40

-60 SRTM-ALOS merging (different master)

-80

-100

Figure 7. 8 Elevation differences between SRTM and ALOS merged InSAR DEM

(different master)

100

80

60

40

20

0 0 200 400 600 800 1000 1200 1400 1600 1800 2000

-20 Elevation difference (m)

-40

-60 SRTM-ALOS ascending & descending

-80

-100

Figure 7. 9 Elevation differences between SRTM and ALOS merged InSAR DEM

(Ascending and Descending)

174 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

100

80

SRTM-Radargrammetry (01/07/08-06/07/08) 60

40

20

0 0 200 400 600 800 1000 1200 1400 1600 1800 2000

-20 Elevation difference (m)

-40

-60

-80

-100

Figure 7. 10 Elevation differences between SRTM and Radargrammetry (01/07/08-

06/07/08)

Figures 7.4~7.10 show the elevation differences between the SRTM DEM and

InSAR generated DEM or regenerated DEMs using merged techniques. In most cases, the merged DEMs improved upon the single InSAR generated DEM while the usage of different master images lead to better results than the usage of the same master images.

Furthermore, Figures 7.11~7.17 show the elevation comparison between SRTM DEM

(reference DEM) and InSAR and radargrammetry DEMs using different data and techniques. In these figures, the SRTM DEM appears to have a higher height than any other DEMs because of the short wavelength of the SRTM sensor and the rough spatial resolution. The point distributions in these figures show that the ALOS/PALSAR provides more stable ground information in the test area. The elevation comparison analysis is based on the relative value analysis than absolute value analysis. Also systematic effects are considered in the analysis. However, the constant components are not perfectly removed.

175 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

400.0

350.0

300.0

250.0

200.0 SRTM (unit:m) 150.0

100.0

50.0

0.0 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 ERS (03/12/95-04/12/95) (unit:m)

Figure 7. 11 Elevation comparison between SRTM and ERS InSAR DEM (03/12/95-

04/12/95); Red line: Linear least square, Blue line; extension line

400.0

350.0

300.0

250.0

200.0 SRTM (unit:m) 150.0

100.0

50.0

0.0 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 ERS merging (different master) (unit:m)

Figure 7. 12 Elevation comparison between SRTM and ERS merged DEM (different

master); Red line: Linear least square, Blue line; extension line

176 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

400.0

350.0

300.0

250.0

200.0 SRTM (unit:m) 150.0

100.0

50.0

0.0 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 ALOS (14/02/08-31/03/08) (unit:m)

Figure 7. 13 Elevation comparison between SRTM and ALOS InSAR DEM (14/02/08-

31/03/08); Red line: Linear least square, Blue line; extension line

400.0

350.0

300.0

250.0

200.0 SRTM (unit:m) 150.0

100.0

50.0

0.0 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 ALOS merging (same master) (unit:m)

Figure 7. 14 Elevation comparison between SRTM and ALOS merged InSAR DEM

(same master); Red line: Linear least square, Blue line; extension line

177 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

400.0

350.0

300.0

250.0

200.0 SRTM (unit:m) 150.0

100.0

50.0

0.0 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 ALOS merging (different master) (unit:m)

Figure 7. 15 Elevation comparison between SRTM and ALOS merged InSAR DEM

(different master); Red line: Linear least square, Blue line; extension line

400.0

350.0

300.0

250.0

200.0 SRTM (unit:m) 150.0

100.0

50.0

0.0 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 ALOS Ascending & Descending (unit:m)

Figure 7. 16 Elevation comparison between SRTM and ALOS merged InSAR DEM

(Ascending and Descending); Red line: Linear least square, Blue line; extension line

178 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

400.0

350.0

300.0

250.0

200.0 SRTM (unit:m) 150.0

100.0

50.0

0.0 0.0 50.0 100.0 150.0 200.0 250.0 300.0 350.0 400.0 RAdargrammetry (01/07/08-06/07/08) (unit:m)

Figure 7. 17 Elevation comparison between SRTM and Radargrammetry (01/07/08-

06/07/08); Red line: Linear least square, Blue line; extension line

7.3 Different sources for DEM generation

DEMs may be generated using different techniques and from different data sources, including ground surveys, photogrammetry, optical remote sensing, radar, and laser scanning.

Field data and GPS are used by incorporating a significant number of elevation points surveyed in the field. Contour lines are measured by maps using manual and semi-automated digitisers and automatic raster scanners. The hypsometry is provided by photogrammetry and satellite imagery (Taud et al. 1999).

179 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

7.3.1 Optical photogrammetry

Earth observation started with photographic sensors used for national security.

Generally, optical images are dependent upon a cloud free view and the availability of sufficient light conditions. However, these offer the advantage of high resolution.

Photogrammetry is a principle of measurement that uses a pair of stereo images to reconstruct the original shape of three dimension objects. Digital airborne and space- borne photogrammetry is a useful tool for DEM generation. The photogrammetric technique defines shape, size and position of targets using the distortion of images which are taken from different viewing directions. Images can be acquired using analogue or digital sensors. Figure 7.18 illustrates the geometry of photogrammetry.

a' b" a b

B A

Figure 7. 18 Stereo geometric condition of an optical system

Digital photogrammetry can acquire many three-dimensional points for high spatial resolution DEM generation. The accuracy of digital photogrammetry depends on the camera-target distance, image quality and resolution. The critical parameters are the

‘base to height’ ratio and the size of the pixel, where smaller pixels have better

180 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES distortion measurements (Massonnet and Souyris, 2008; Fabris and Pesci 2005; Kornus et al. 1998). The high resolution systems IKONOS (US), QuickBird (US), EROS A

(Israel), SPOT (France), KOMPSAT (Korea), and ALOS (Japan) are currently in use around the world. Figure 7.19 shows the US-operated IKONOS satellite system. Figure

7.20 is the satellite panchromatic images from ALOS/ Panchromatic remote sensing instrument for stereo mapping (PRISM) from JAXA. Table 7.2 lists data prices of radar and optical satellite images.

Figure 7. 19 IKONOS and SPOT-5 satellite image data collection (SIC)

Figure 7. 20 ALOS/PRISM forward mosaic image in Appin NSW, Australia

181 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

Table 7. 2 Satellite data prices of radar and optical images (GA, SOPT IMAGE)

Sensors Price (AUD) Scene size ALOS/PALSAR 440 30×30km ~ 350×350km ERS-1/2 590 100×100km RADARSAT 4950 50×50km ~ 500×500km TerraSAR-X 2700 ~ 4800 5×10km ~ 100×100km SRTM Free Latitude 60N~57S Landsat 7 ETM 1100 ~ 1500 185×185km ~ 185×530km Landsat 5 TM 1100 ~ 1200 185×185km ~ 225×185km ALOS/PRIMS 440 ~ 550 35km×35km ASTER 145 ~ 580 60×60km SPOT 2600 ~ 11400 60×60km Quickbird 280 16.5×16.5km

7.3.2 Airborne laser scanning

Airborne laser scanning (ALS) or Light detection and ranging (LiDAR) is an active remote sensing technique providing direct range measurements between the laser scanner, which is aircraft-mounted, and topography, which is shown in Figure 7.21.

LiDAR operates on the same principle as radar. Distance measurements are mapped into three dimensional point clouds. Several backscattering signals can be recorded for a single pulse emission. Since 2004, a new ALS system called “Full waveform LiDAR” has recorded the complete waveform of the backscattered signal echo. Common airborne LiDAR systems consist of a laser transmitter and receiver, a mechanical scanner, a hybrid positioning system (inertial measurement unit and GPS), storage media, and an operating system for signal digitization and on-line data acquisition. This

182 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES unit monitors and synchronizes measurements, and processes data in real-time to extract geo-referenced points (Mallet and Bretar 2009; Hodgson and Bresnahan 2004).

Figure 7. 21 The airborne LiDAR system (USGS)

One of the most desirable characteristics of LiDAR is that it has high vertical accuracy which enables it to represent the topography with a high degree of accuracy.

LiDAR is one of the few systems that is able to collect data from all points while also producing DEM with a 1-2m horizontal resolution.

7.3.3 Topographic maps

Most countries have topographic maps and these are used as another main data source for DEM modelling. The topographic maps cover the entire country using good quality contours. The scale of a topographic map is decided according to the specific purpose of the topographic map. Table 7.3 shows characteristics depending on the map scale. The contour density is measured by the vertical contour interval, “contour

183 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES interval”. The intervals according to the map scales are shown in table 7.4

(Li et al. 2004).

Table 7. 3 Topographic maps at different scales (Li et al. 2004)

Topographic Map Scale Characteristics

Large-to medium scale >1:10,000 Representation true to plan maps

Medium- to small-scale Representation similar to 1:20,000-1:75,000 maps plan

High degree of General topographic map < 1:100,000 generalisation or signature representation

Table 7. 4 Map scales and commonly used contour intervals (Li et al. 2004)

Scale of the Topographic Map Interval between contour lines (m)

1:200,000 25-100

1:100,000 10-40

1:50,000 10-20

1:25,000 5-20

1:10,000 2-10

Table 7.5 is the comparative list of DEM generation derived from different sources such as InSAR, ALS and Photogrammetry.

184 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

Table 7. 5 Comparison of InSAR, ALS and Photogrammetry for Digital Elevation Model generation

Active Remote Sensing Passive Remote Sensing

InSAR Photogrammetry Sensor Airborne Lidar (ALS50) Satellite SRTM Airborne Satellite Airborne (ADS80) (ALOS: fine mode) C-band X-band (UAVSAR) (LANDSAT7) Vertical discrimination 1 arc sec 3 arc sec Resolution Spatial Resolution: 10m 15 to 30m 10cm to 1m distance: 3.5m (30×30m) (90×90m) Vertical accuracy: Vertical accuracy: 16m (90%) within a Lateral placement :7~64cm 16m (90%) Accuracy 255×255km Vertical: 10cm Vertical placement:8~24cm Horizontal accuracy: Horizontal accuracy: ~60m ~20m Maximum slant Maximum slant range: 233km orbit with 57’ of 233km orbit with 57’ range: 7563m 7563m inclination of inclination Altitude 691.65km 1.8~2 km 705km ±5km Minimum slant range: 30 degree off nadir 30degree off nadir Minimum slant 200m 58 degree of nadir 58 degree of nadir range: 200m Range swath : 225km Range swath : 225km Max:9000m Observation swath: Swath width: 185km Coverage X-band: Range swath: 16km Field of view: 64˚ 306m~11606m 70km 183×170 km C-band:119 million km2 58 million km2 Size of image: Max overage: 70×70km 11 days 11 days Efficiency - 16 days - 11.6×444=5150.4 km2/h Max overage: (in 2000) (in 2000) 1890350 km2/h Repeat cycle: Ground speed 46 days Repeat cycle: 16days (GS)=90kts for Field of view(FOV): Latitude 60N~57S. Ground speed range: Limitations Period: 98.7 min 400 scenes/day GSD of 1.2``/3cm 0~75 degree 159 orbits were used for operational mapping 110-250m/s Speed: 450.09km/m 10% cloud cover 240kts for GSD of 27,005.4km/h 4``/10cm

185 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

Datum WGS 84 WGS 84 WGS84 WGS84 WGS 84

Size of antenna: (Azimuth:8.9m 6-10 Ground ×elevation:3.1m) Control points 421C: Cessna aircraft (1981) (VH-NSW) Peak power: 2kW Velocity: 7.5km/s Height ambiguity per 6-10 Ground Control Internal orientation, Look angle: 25-60 Max speed:444km/h Look angle range: 30-60 fringe : 175m points external orientation, Range: 2,756km Look direction: Right Absolute orientation, Centre frequency: Point extraction 1270MHz/23.6cm

Deformation error SNR decorrelation: 0.2mm Bandwidth(MHz) to Vertical accuracy is Geometric/Temporal Resolution (m) and RMSE of decorrelation 1.0mm 400:0.37 H/9000 Pulse width range from 6 to Flight path 300:0.5 12 ns uncertainty:1.0mm 160:0.95 Maximum error of this pulse width range Slant range distance at scene centre: 392km 80:1.9 3RMSE translates into pulse lengths Instrument phase: 40:3.7 of approximately 1.8-3.6 m 0.6mm 20:7.5 C-factor=flying 10:15.0 height/contour Topography interval uncertainty: 5.5mm

Atmospheric distortion: 10mm

186 CHAPTER 7 DIGITAL ELEVATION MODEL ASSESSMENT AND DIFFERENT SOURCES

7.4 Concluding remarks

All generated data sets were placed in the GIS software for quality assessment using profiles and RMSEs. The RMSE is the basic method that was deployed to produce quality estimation and approximately two thousand check points were selected in this chapter. The product quality is improved by multi-baseline and elevation error elimination methods. The profile lines of products illustrate the terrain shape and curve.

In particular, it is able to show the amount of location shift of responded terrain in a

DEM. Generally, InSAR generated DEM has location shift forward to the sensor direction. These phenomena are removed using different orbit observation methods and merging methods.

In addition, this chapter summarised other data acquisition techniques for DEM generation. Photogrammetry is a method with a long technical history, which is able to provide high resolution information. However, this method is strongly affected by atmospheric effects and is limited by observation time. LiDAR is able to generate high resolution 3D points and can be used directly for DEM generation. However, the huge amount of data generated encumbers data processing. DEM generation that uses a topographic map is a basic method which enables the supply of stable terrain information.

187 CHAPTER 8 CONCLUSIONS AND DISCUSSIONS

8.1 Conclusions

Synthetic Aperture Radar (SAR) interferometry is a geodetic imaging technique increasingly used for measuring terrain geometry. It exploits the phase information of radar signals backscattered from the ground surface. In this thesis, characteristics of past and present airborne/satellite SAR systems and SAR data acquisition methods and types were summarised to provide a quick reference. This summary, which is based on a selection of appropriate SAR data for various remote sensing applications, is a useful reference.

Interferometric SAR generates the terrain elevation information using phase information from two SAR images. The interferometric phases are represented in an interferogram, which blend topography with atmospheric heterogeneity, noise and ground deformation between the two image acquisitions. The applications of interferometric SAR, such as cartographic mapping, land cover characterisation, measurement of hazards, Digital Elevation Model (DEM) generation, land deformation mapping and atmospheric studies, use only one of the phase components at a time by eliminating or minimising the impact of the other components.

8.1.1 InSAR DEM accuracy improvement techniques in single-baseline observation

The accuracy and quality of InSAR DEM is influenced by observation direction, wavelength of sensor and relationship of two acquired images. In this thesis, various techniques are developed for accuracy improvement such as combinations of opposite- side observation, differential InSAR (DInSAR) and elevation selection using coherence values.

188 CHAPTER 8 CONCLUSIONS AND DISCUSSIONS

The SAR images which are generated by different wavelength provide useful and various terrain information such as vegetation, forests, mountainous and rapidly sloping areas because different wavelengths appear to have different ground representations.

The short wavelength SAR signals respond best to the canopy of dense forests or to trees. Long wavelength SAR images can penetrate forests and vegetation and are less affected by vegetation and forest movement. Therefore, long wavelength SAR image provids more stable information than short wavelength SAR images despite surface canopy effects and other sources of noise. ALOS/PALSAR data (L-band) supplies higher resolution SAR images. These can be used to produce more accurate and reliable DEM products.

The geometrical distortion of side-looking observation systems is resolved by combination of opposite-side orbit observation. The one directional side looking systems have terrain distortion forward to sensor direction. Opposite side orbit observations (ascending and descending) provide more accurate terrain information and has less geometrical distortion than the same side orbit observations as they improve the quality of DEM in areas of rough terrain such as mountains and valleys.

In the case of InSAR DEM processing, a problem was encountered in the elevation error caused by ground deformation between images acquired by a repeat-pass satellite sensor. The ground deformation influenced the interferometric phase values, which was converted into elevation error in the DEM product.

This problem can be solved using the DInSAR technique for ground deformation area detection and elevation error region exclusion. Furthermore, the excluded elevation value was interpolated by the surrounding height value of the InSAR DEM

189 CHAPTER 8 CONCLUSIONS AND DISCUSSIONS and reference DEM. This provided more accurate and reliable DEM products without ground deformation errors, which were generated by repeat-pass satellite remote sensing systems.

Although the overall InSAR DEM has a higher accuracy, some parts of InSAR

DEM have a low accuracy because of the image pair conditions such as short baseline, noise and low coherence. Therefore, selected elevation data with accurate information is required to improve the updated DEM quality.

The elevation confidence can be secured by the coherence value which is generated by master and slave images. In the processing results, the coherence threshold values from 0.5 to 0.6 offer the most reasonable value when between the amounts of elevation data and elevation confidence are compared.

Elevation selection by coherence value improves the elevation confidence in DEM and important information for SRTM DEM updating. The updated SRTM DEM has a higher accuracy compared with single InSAR generated DEM and SRTM DEM.

Furthermore, merged SRTM DEM using each updated SRTM DEMs also improves accuracy. However, in the special case (InSAR DEM: 31/03/08-01/07/08), InSAR

DEM has a higher accuracy than any other DEMs, which are generated in SRTM DEM updating case study. InSAR DEM quality pre-assessment is required for SRTM DEM updating. Thus, the selection of suitable InSAR DEM should be the focus of future work.

8.1.2 InSAR DEM accuracy improvement techniques in multi-baseline observation

InSAR DEM generation using multi-baseline and multi-temporal SAR image pairs has improved DEM quality compared with the single-baseline method. The multi- 190 CHAPTER 8 CONCLUSIONS AND DISCUSSIONS baseline technique was useful for precise DEM generation supported by a repeat-pass satellite system. It is capable of overcoming the weaknesses of single-baseline InSAR systems. The multi-baseline and multi-temporal techniques are advancements that enables interferometric techniques, which are not only able to perform surface mapping to be deployed, but are also able to monitor temporal evolution. Multi-pass interferograms are less affected by atmospheric contributions, which are more related to elevation errors, than single-pass interferograms. Furthermore, it can improve the accuracy of DEMs by using accumulated terrain information from many data pairs.

In this thesis, two kinds of multi-baseline InSAR DEM generation were used by L- band imagery. These are the single (fixed) master multi-baseline method and the different master multi-baseline method. Each method has its own strengths in data processing, target area selection, data arrangement, and data matching.

Different master multi-baseline method produces more accurate results and improved ground reality than the single master multi-baseline method. Furthermore, different merging techniques were used for the multi-baseline method. For example, the coherence weighting and the baseline distance weighting method were the more powerful and adaptive merging methods than the mean method. The baseline distance weighting method is considered to be the image geometric relationship of the whole processed area between two SAR images. It can provide the unifying condition

(weighting) to the whole imaging processed area. However, it does not consider the pixel by pixel relationship between two SAR images. The coherence weighting method is better at relating each pixel condition than the baseline distance weighting method.

However, it is more associated with surface conditions than the terrain height of two

SAR images.

191 CHAPTER 8 CONCLUSIONS AND DISCUSSIONS

After the merging process, some areas within merged DEMs experienced decreased accuracy of elevation due to low coherence or elevation noise. This can be a key focus of future research in multi-baseline InSAR DEM generation. However, the practical coverage of the single master multi-baseline method is larger than the different master multi-baseline method because of the overlaying area of different master images. Also, atmospheric errors in the same master method remain during merging processing while atmospheric errors in the different master methods are minimised during merging processing. Further, different master methods can be used for atmospheric effect elimination.

8.1.3 SAR DEM generation using intensity image

Radargrammetry is comparable to photogrammetry in that a stereo parallax found in a stereo image pair corresponds to terrain elevation. The major difference between radargrammetry and photogrammetry is that SAR images are used instead of optical images. Furthermore, another difference exists in that radargrammetry uses the intensity value while the InSAR method uses the phase value of SAR data.

Taking into account processing time and software capabilities, radargrammetry can generate DEMs with a larger coverage compared with InSAR, while InSAR processing has the disadvantage of the unwrapping step. However, the main problems of radargrammetry are that the DEM products are of low quality due to the low spatial resolution of SAR intensity imagery and the terrain slope conditions. Disparity and convergence of objects are the two key considerations when viewing stereo imagery.

Disparity predominates when viewing radar images, but shade and shadow cues also have a strong and cumulative effect.

192 CHAPTER 8 CONCLUSIONS AND DISCUSSIONS

8.1.4 Product quality assessment and other DEM generation sources

The quality assessment of all DEMs, which are generated using InSAR and radargrammetry, is discussed in the final section of this thesis. Quality assessment offers useful information for dataset selection. The RMSE and profile line of each

DEM product represent the product quality. Single baseline InSAR DEM processing can supply a large coverage area and lead to the rapid product generation because it uses only one pair of SAR images. However, multi-baseline InSAR DEMs can provide improved accuracy and terrain details. Various merging techniques were used and each technique appears advantageous when used for different purposes and in different conditions. Furthermore, other data acquisition techniques for DEM generation such as photogrammetry, LiDAR and topographic maps were summarised in the this thesis.

This thesis proposes that improvement in DEM accuracy depends on the problems encountered in each case, such as sensor characteristics, corresponding terrain shift due to the observation location, low resolution and lack of terrain information acquired by a single pair, ground deformation between different acquisitions, uncertain elevation information due to a low correlation between different acquisitions and surface canopy effects. Each method is recovered each case problem. Final products which were generated by each method can be combined and cooperated with other products to generate precise DEM products.

8.2 Future research

As discussed in Chapter Four, elevation errors caused by ground deformation were not fully omitted from this paper. The proposed method is able to detect the ground 193 CHAPTER 8 CONCLUSIONS AND DISCUSSIONS movement area and subsequently exclude elevation errors. However, after omission of the elevation errors, the elevation interpolation or refilled data did not connect with the surrounding elevation. In future studies, the height interpolation method using different height sources and more precise ground movement area detection methods will be proposed for elevation error exclusion.

The atmospheric effect, which is ignored during InSAR processing, has to be considered and calculated for precise DEM generation. Generally, the InSAR technique is not affected by atmospheric effects, including dense cloud and water vapour.

However, for precise InSAR processing, atmospheric effects should be taken into account and resolved. A potential method of calculating atmospheric effect is by the

GPS network. The GPS network can cover a large area, similar to InSAR DEM coverage.

Current and new satellite SAR observation systems provide less than 3-meter spatial resolution and various observation modes such as RADARSAT-2, RISAT-2,

TerraSAR-X and COSMO-SkyMed. The repeat cycles range from 11 to 25 days, and these systems can acquire more detailed information from the Earth’s surface. However, this creates difficulties in data processing and in particular, filtering of acquired information.

Using short wavelength (X-band) SAR imagery, the radargrammetry is a useful method for higher accuracy DEM generation in vegetated and in urban areas while short wavelength InSAR DEM experiences processing difficulties due to low coherence and dense phase change in the same regions. Ascending and descending

194 CHAPTER 8 CONCLUSIONS AND DISCUSSIONS stereo pair combinations will be topics of interesting study for radargrammetry applications.

TerraSAR-X add-on for digital elevation measurements (TanDEM-X) is the first bistatic SAR mission to TerraSAR-X and flying the two satellites nearly controlled formation with typical distances between 250 and 500m. It can provide a high degree of accuracy and stable global digital elevation data. This bistatic mission operates two sensors simultaneously in space. It leads different concept with previous satellite SAR remote sensing while it is similar with airborne SAR operating systems. This mission will offer interesting and useful research opportunities.

195 REFERENCES

Abdelfattah, R. & Nicolas, J. M., 2002, ‘Topographic SAR interferometry formulation for high-precision DEM generation’, IEEE Transactions on geoscience and remote sensing, 40(11), 2415-2426.

Ahmad, B. B., 2008, ‘Assessment and Correction of DEM Generation from Airborne and Space Borne RADAR Systems with Reference to Geo-Hazard Identification in the Cameron Highlands, Malaysia’, PhD thesis paper, UNSW.

Alberga, V., 2004 ‘Volume decorrelation effects in polarimetric SAR interferometry’, IEEE Transactions on geoscience and remote sensing, 42(11), 2467-2478.

Amelung, F., Galloway, D. L., Bell, J. W., Zebker, H. A., & Laczniak, R. J., 1999, ‘Sensing the ups and downs of Las Vegas: InSAR reveals structural control of land subsidence and aquifer-system deformation’, Geology, 27(6), 483-486.

Balzter, H., 2001, ‘Forest mapping and monitoring with interferometric synthetic aperture radar (InSAR)’, Progress in , 25(2), 159-177.

Baselice, F., Budillon, A., Ferraiolo, G., & Pascazio, V., 2009, ‘Joint SAR imaging and DEM reconstruction from multichannel layover-affected SAR data’, IEEE Geoscience and Remote Sensing Symposium proceedings, Cape Town, South Africa, 12-17 July, III-25-III-28.

Bayer, T., Winter, R., & Schreier, G., 1991, ‘Terrain influences in SAR backscatter and attempts to their correction’, IEEE Transactions on geoscience and remote sensing, 29(3), 451-462.

Bioucas-Dias, J. M., & Valadao, G., 2005, ‘Phase unwrapping via graph cuts’, Pattern recognition and image analysis, Springer Berlin Heidelberg, 3522, 360-367.

196 Bioucas-Dias, J. M., & Valadao, G., 2007, ‘Phase unwrapping via graph cuts’, IEEE Transactions on image processing, 16(3), 698-709.

Bourgine, B., & Baghdadi, N., 2005, ‘Assessment of C-band SRTM DEM in a dense equatorial forest zone’, Comptes Rendus Geosciences, 337(14), 1225-1234.

Buckland, J. R., Huntley, J. M., & Turner, S. R. E., 1995, ‘Unwrapping noisy phase maps by use of a minimum-cost-matching algorithm’, Applied optics, 34(23), 5100-5108.

Burrough, P. A, 1986, ‘Principles of geographical information systems for land resources assessment’ Oxford Clarendon Press.

Campbell, D. B., Jurgens R. F., Dyce, R. B., Harris F. S., & Pettengill, G. H., 1970, ‘Radar interferometric observations of Venus at 70-centimeter wavelength’ Science, 170(3962), 1090-1092.

Chang, H. C., Ge, L., Ng, A. H. M., Rizos, C., Wang, H., & Omura, M., 2008, ‘Combination of multiple repeat orbits of ENVISAT for mine deformation monitoring’, IAG symposia Series, 133, 631-638.

Chen, C. W., & Zebker, H. A., 2000, ‘Network approaches to two-dimensional phase unwrapping: intractability and two new algorithms’, Journal of the optical society of america, 17(3), 401-414.

Chen, P. H., & Dowman, I.J., 2001, ‘A weighted least squares solution for space intersection of spaceborne stereo SAR data’, IEEE Transactions on geoscience and remote sensing, 39(2), 233-240.

Chen, X., Wang, C., & Zhang, H., 2009, ‘DEM generation combining SAR polarimetry and shape-from-shading techniques’, IEEE geoscience and remote sensing letters, 6(1), 28-32.

197 Chen, Y., Shi, P., Deng, L., & Li, J., 2007, ‘Generation of top-of-canopy Digital Elevation Model (DEM) in tropical rain forest regions using radargrammetry’, International journal of remote sensing, 28(19), 4345-4349.

Colesanti, C., & Wasowski, J., 2006, ‘Investigating landslides with space-borne synthetic aperture radar (SAR) interferometry’, Engineering geology, 88, 173- 199.

Crosetto, M., 2002, ‘Calibration and validation of SAR interferometry for DEM generation’, ISPRS Journal of photogrammetry and remote sensing, 57(3), 213- 227.

Crosetto, M., 2004, ‘Deformation measurement using interferometric SAR data’, ISPRS congress, Istanbul, Turkey, 12-23 July, 164-169.

Cuartero, A., Felicisimo, A. M., & Ariza, F.J. ,2005, ‘Accuracy, Reliability, and Depuration of SPOT HRV and ASTER digital elevation models’, IEEE Transactions on geoscience and remote sensing, 43(2), 404-407.

Dall, J., Madsen, S.N., Keller, K., & Forsberg, R., 2000, ‘Using airborne SAR interferometry to measure the elevation of a Greenland ice cap’, IEEE Geoscience and Remote Sensing Symposium proceedings, Honolulu, HI, USA, 24~28 July, 3, 1125-1127.

Davidson, G.W., & Cumming, I., 1997, ‘Signal properties of spaceborne squint-mode SAR’, IEEE Transactions on geoscience and remote sensing, 35(3), 611-617.

Desmet, P. L. L., 1997, ‘Effects of interpolation Errors on the Analysis of DEMs’, Earth surface processes and , 22(6), 563-580.

Ding, X., Li, Z., Zhu, J., Feng, G., & Long, J., 2008 ‘Atmospheric effects in InSAR measurements and their mitigation’ Sensors, 8, 5426-5448.

198 d’Ozouville, N., Deffontaines, B., Benveniste, J., Wegmuller, U., Violette, S., & de Marsily, G., 2008, ‘DEM generation using ASAR (ENVISAT) for addressing the lack of freshwater ecosystems management, Santa Cruz Island, Galapagos’, Remote sensing of environment, 112(11), 4131-4147.

Earth observation research and application centre, 2008, ‘ALOS data users handbook: revision C’, Japan aerospace exploration agency, March 2008.

Earth remote sensing data analysis centre, 2010, ‘ERSDAC PALSAR product characteristics’, ERSDAC-VX-PALSAR-001 Rev.1.7, Tokyo, Japan, February 2010.

Eckert, S., Kellenberger, T., & Itten, K., 2005, ‘Accuracy assessment of automatically derived digital elevation models from aster data in mountainous terrain”, International journal of remote sensing, 26(9), 1943-1957.

Eineder, M., & Adam, N., 2005, ‘A Maximum-Likelihood Estimator to Simultaneously Unwrap, Geocode, and Fuse SAR Interferograms From Different viewing Geometries Into One Digital Elevation Model’, IEEE Transactions on geoscience and remote sensing, 43(1), 24-36.

Fabris, M., & Pesci, A., 2005, ‘Automated DEM extraction in digital aerial photogrammetry: precisions and validation for mass movement monitoring’, Annals of geophysics, 48(6), 973-988.

Falorni, G., Teles, V., Vivoni, E. R., Bras, R. L., & Amaratunga, K.S., 2005, ‘Analysis and characterization of the vertical accuracy of digital elevation models from the Shuttle Radar Topography Mission’, Journal of geophysical research, 110, F02005.

Farr, T. G., Rosen, P.A., Caro, E., Crippen, R., Duren, R., Hensly, S., Kobrick, M., Paller, M., Rodriguez, E., Roth, L., Seal, D., Shaffer, S., Shimada, J., Umland, J.,

199 Werner, M., Oskin, M., Burbank, D. & Alsdorf, D., 2007, ‘The Shuttle Radar Topography Mission’, Rev. Geophys., 45, RG2004, doi:10.1029/2005RG000183.

Ferraiuolo, G., Meglio, F., Pascazio, V., & Schirinzi, G., 2009, ‘DEM Reconstruction Accuracy in Multichannel SAR Interferometry’, IEEE Transaction on geoscience and remote sensing, 47(1), 191-201.

Ferretti, A., Prati, C., & Rocca, F., 2001, ‘Permanent scatterers in SAR interferometry’, IEEE Transactions on geoscience and remote sensing, 39(1), 8-20.

Ferretti, A., Prati, C., Rocca, F., & Monti Guarnieri, A., 1997, ‘Multibaseline SAR Interferometry for automatic DEM reconstruction’, Third ERS Symposium on Space at the service of our Environment, Florence, Italy, 14-21 March, 1809.

Fornaro, G., Guarnieri, A. M., Pauciullo, A., & Tebaldini, S., 2005, ‘Joint Multi- Baseline SAR Interferometry’, EURASIP Journal on applied signal processing, 20, 3194-3205.

Fornaro, G., Pauciullo, A., & Serafino, F., 2007, ‘Multipass SAR Processing for Urbanized Areas Imaging and Deformation Monitoring at Small and Large Scales’, 2007 Urban remote sensing joint event, Paris, France, 11-13 April, 1-7.

Furuta, R., Shimada, M., Tadono, T., & Watanabe, M., 2005, ‘Interferometric capabilities of ALOS PALSAR and its utilization’, Fringe 2005 workshop, Proceeding of the conference in Frascati, Italy, 34-44.

Graham, L. C., 1974, ‘Satellite interferometer radar for topographic mapping’, Proceedings of the IEEE, 62(6), 763-768.

Ge, D. Q., Guo, X. F., Zhang, L., Liu, S. W., & Fan, J. H., 2006, ‘Linear deformation rate derivation from multi-baseline differential interferogram stacks’, IEEE Geoscience and Remote Sensing Symposium proceedings, Denver, Colorado, 31 July-4 August, 763-766.

200 Ge, L., Chang, H. C., & Rizos, C., 2007, ‘Mine Subsidence Monitoring Using Multi- source Satellite SAR Images’, Photogrammetric engineering and remote sensing, 73(3), 259-266.

Ge, L., Tsujii, T., & Rizos, C., 2002, ‘Tropospheric heterogeneities corrections in differential radar interferometry’, IEEE International Geoscience and Remote Sensing Symposium, Toronto, Canada, 24-28 June, 3, 1747-1749.

Ge, L., Zhang, K., Ng, A. H. M., Dong, Y., Chang, H. C., & Rizos, C., 2008, ‘Preliminary results of satellite radar interferometry for the co-seismic deformation of the 12 May 2008 Ms 8.0 Wenchuan earthquake’, Journal of geographic information science, 14(1), 12-19.

Gens, R., 1999, ‘SAR interferometry: software, data format, and data quality’, Photogrammetric engineering and remote sensing, 65, 1375-1378.

Gini, F., & Lombardini, F., 2005, ‘Multibaseline cross-track SAR interferometry: A signal processing perspective’, IEEE Aerospace and electronic systems magazine, 20(8), 71-93.

Goldstein, R. M., Zebker, H. A., & Werner, C. L., 1988, ‘Satellite radar interferometry: two-dimensional phase unwrapping’, Radio science, 23(4), 713-720.

Gough, P., & Hawkins, D., 1997, ‘Unified framework for modem synthetic aperture imaging algorithms’, International journal of imaging systems and technology, 8(4), 343-358.

Gutmann, B., & Weber, H., 2000, ‘Phase unwrapping with the branch-cut method: role of phase-field direction’, Applied optics, 39(26), 4802-4816.

Hanssen, R., Moisseev, D., & Businger, S., 2003, ‘Resolving the acquisition ambiguity for atmospheric monitoring in multi-pass radar interferometry’, IEEE

201 International Geoscience and Remote Sensing Symposium, Toulouse, France, 21- 25 July, 2, 1202-1205.

Hanssen, R. F., 2001, ‘Radar Interferometry: Data Interpretation and Error Analysis’, Kluwer Academic Publishers, 113-130, 15.

Henderson, F. M., & Lewis, A. J., 1998, ‘Principles and applications of imaging radar Manual of remote sensing’, John Wiley and Sons, Inc., Third edition, Vol. 2.

Hirano, A., Welch, R., & Lang, H., 2003, ‘Mapping from ASTER stereo image data: DEM validation and accuracy assessment’, ISPRS Journal of photogrammetry and remote sensing, 57, 356-370.

Hodgson, M. E., & Bresnahan, P., 2004, ‘Accuracy of airborne lidar-derived elevation: empirical assessment and error budget’, Photogrammetric engineering and remote sensing, 70(3), 331-340.

Hogda, K. A., Guneriussen, T., & Lauknes, I., 2002, ‘Synthetic aperture radar for DEM generation in snow-covered mountain terrain’, IEEE International Geoscience and Remote Sensing Symposium, Toronto, Canada, 24-28 June, 4, 2193-2135.

Holecz, F., Pasquali, P., Moreira, J., Meier, E. & Nuesch, D., 1998, ‘Automatic generation and quality assessment of digital surface models generated from AeS- 1 InSAR data’, European Conference on Synthetic Aperture Radar, Friedrichshafen, Germany, 25-27 May, 57-60.

Honikel, M., 1998, ‘Improvement of InSAR DEM accuracy using data and sensor fusion’, IEEE International Geoscience and Remote Sensing Symposium, Seattle, USA, 6-10 July, 2348-2350.

Horn, R., Nottensteiner, A., & Scheiber, R., 2008, ‘F-SAR-DLR’s advanced airborne SAR system onboard DO228’, European Conference on Synthetic Aperture Radar, Friedrichshafen, Germany, 2-5 June.

202 Hu, D., Li, J., Chen, Y., Deng, L., & Ramli, k., 2006, ‘Radargrammetry DEM from RADARSAT imageries and accuracy validation’, IEEE Geoscience and Remote Sensing Symposium proceedings, Denver, Colorado, 31 July-4 August, 3119- 3122.

Huang, G. M., Guo, J. K., Zhao. Z., Xiao, Z., Qiu, C. P., Pang, L., & Wang, Z. Y., 2004, ‘DEM generation from stereo SAR images based on polynomial rectification and height displacement’, IEEE Geoscience and Remote Sensing Symposium proceedings, Anchorage, Alaska, 20-24 September, 6, 4227-4230.

Huang, Y., Xu, H., Chen, J., Li, C., Li, J., & Zhou, Y., 2000, ‘An algorithm for interferometric SAR data processing’, Journal of electronics, 17(3), 202-207.

Janssen, V., Ge, L., & Rizos, C., 2004, ‘Tropospheric corrections to SAR interferometry from GPS observations’, GPS solutions, 8(3), 140-151.

Javis, A., Rubiano, J., Nelson, A., Farrow, A., & Mulligan, M., 2004, ‘Practical use of SRTM data in the tropics-comparisons with digital elevation models generated from cartographic data’, International Centre for Tropical Agriculture, Cali, Colombia, Working document, 198, 5-10.

Jensen, J. R., 2007, ‘Remote sensing of the environment: an Earth resource perspective’, Person Education, Inc. 2nd ed. 291-334.

Jenson, S. K., 1991, ‘Applications of hydrologic information automatically extracted from digital elevation models’, Hydrological processes, 5, 31-44.

Kaupp, V. H., Bridges, L. C., Pisaruck, M. A., MacDonald, H. C., & Waite, W. P., 1983, ‘Simulation of spaceborne stereo radar imagery: Experimental result’, IEEE Transactions on geoscience and remote sensing, GE-21(3), 400-405.

203 Khiabain, S.A., Valadan Zoej, M.J., Mobasheri, M.R. & Dehghani, M., 2008, ‘Atmospheric effects removal of ASAR-derived InSAR products using meris data and GPS’, the International archives of the photogrammetry, remote sensing and spatial information sciences, XXXVII, B1, 151-155.

Kiamehr, R., & Sjoberg, L. E., 2005, ‘Effect of the SRTM global DEM on the determination of a high-resolution geoid model: A case study in Iran’, Journal of geodesy, 79(9), 540-551.

Kornus, W., Lehner, M., Ebner, H., Froba, H., & Ohlhof, T., 1998, ‘Photogrammetric point determination and DEM generation using MOMS-2P/PRIRODA three-line imagery’, International archives of photogrammetry and remote sensing, 32, 321- 328.

Kotsis, I., Kontoes, C., Paradissis, D., Karamitsos, S., Elias, P., & Papoutsis, I., 2008, ‘A methodology to validate the InSAR derived displacement field of the September 7th, 1999 Athens earthquake using terrestrial surveying. Improvement of the assessed deformation field by interferometric stacking’, Sensors, 8, 4119- 4134.

Kwok, R., & Fahnestock, M., 1996, ‘Ice sheet motion and topography from radar interferometry’, IEEE Transactions on geoscience and remote sensing, 34(1), 189-200.

Kyaruzi, J. K., 2005, ‘Quality assessment of DEM from radargrammetry data’ International institute for geo-information science and Earth observation enschede, The Netherlands, master of science.

La Prade, G., 1963, ‘An analytical and experimental study of stereo for radar’, Photogramm. Eng., 35(2), 294-333.

Land-Victoria, 2002, Production description, Vicmap Elevation, Land Victoria, Department of Sustainability and Environment.

204 Lee, H., & Liu, J. G., 2001, ‘Analysis of topographic decorrelation in SAR interferometry using ratio coherence imagery’, IEEE Transactions on geoscience and remote sensing, 39(2), 223-232.

Lee, J. S., Hoppel, K. W., Mango, S. A., & Miller, A. R., 1994, ‘Intensity and phase statistics of multilook polarimetric and interferometric SAR imagery’, IEEE Transactions on geoscience and remote sensing, 32(5), 1017-1028.

Li, F., & Goldstein, R., 1990, ‘Studies of multibaseline spaceborne interferometric synthetic aperture radars’, IEEE Transactions on geoscience and remote sensing, 28(1), 88-97.

Li, J., Huang, H., & Liang, D., 2007, ‘A Multi-baseline InSAR DEM Reconstruction Approach without Ground Control Points’ IEEE Geoscience and Remote Sensing Symposium proceedings, Barcelona, Spain, 23-27 July, 4509-4512.

Li, Z., Bao, Z., & Suo, Z., 2007, ‘A joint image coregistration, phase noise suppression and phase unwrapping method based on subspace projection for multibaseline InSAR systems’, IEEE Transactions on geoscience and remote sensing, 45(3), 584-591.

Li, Z., & Bethel, J., 2008, ‘Image coregistration in SAR interferometry’, International Archives of the photogrammetry, remote sensing and spatial information sciences. XXXVII, part B1.

Li, Z., Liu, G., & Ding, X., 2006, ‘Exploring the generation of digital elevation models from same-side ERS SAR images: topographic and temporal effects’, The photogrammetric record, 21(114), 124-140.

Li, Z., Zhu, Q., & Gold, C., 2004, ‘Digital terrain modelling: principles and methodology’, CRC press, 34-35.

205 Li, Z. T., & Bethel, J., 2007, ‘DEM Alignment and Registration in Interferometric SAR Processing and Evaluation’, IEEE International Geoscience and Remote Sensing Symposium, Barcelona, Spain, 23-27 July, 4890-4893.

Liao, M., Wang, T., Lu, L., Zhou, W., & Li, D., 2007, ‘Reconstruction of DEMs from ERS-1/2 Tandem Data in Mountainous Area Facilitated by SRTM Data’, IEEE Transactions on geoscience and remote sensing, 45(7), 2325-2335.

Lombardini, F., & Pardini, M., 2008, ‘3-D SAR tomography: the multibaseline sector interpolation approach’, IEEE Geoscience and remote sensing letters, 5(4), 630- 634.

Maitre, H., 2008, ‘Processing of synthetic aperture radar images’, ISTE Ltd and John Wiley & Sons Inc.

Mallet, C., & Bretar, F., 2009, ‘Full-waveform topographic lidar: State-of the-art’, ISPRS Journal of photogrammetry and remote sensing, 64, 1-16.

Massonnet, D., & Feigl, K. L., 1998, ‘Radar interferometry and its application to changes in the Earth’s surface’, Review of geophysics, 36(4), 441-500.

Massonnet, D., Feigl, K., Rossi, M., & Adragna, F., 1994 ‘Radar interferometric mapping of deformation in the year after the Landers earthquake’, Nature, 369, 227-230.

Massonnet, D., & Souyris, J. C., 2008, ‘Image with synthetic aperture radar”, EPFL Press, 189-191.

Mercer, J. B., 1995, ‘SAR technologies for topographic mapping’, Photogrammetric Week’ 95, 117-126.

206 Moore, D., Grayson, R. B., & Ladson, A. R., 2006, ‘Digital terrain modelling: a review of hydrological, geomorphological, and biological applications’, Hydrological processes, 5(1), 3-30.

Murata, M., & Miyawaki, M, 2003, ‘DEM generation of a volcano using airborne InSAR’, NEC Res Dev, 44(2), 157-160.

Nonaka, T., Hayakawa, T., Griffiths, S., & Mercer, B., 2009, ‘DEM production utilizing stereo technology of TerraSAR-X data’, IEEE Geoscience and Remote Sensing Symposium proceedings, Cape Town, South Africa, 12-17 July, IV-157- IV-160. Nitti, D. O., Nutricato, R., Bovenga, F., Refice, A., Chiaradia, M, T., & Guerriero, L., 2009, ‘TerraSAR-X InSAR multi-pass analysis on Venice (Italy)’, Image and Signal Processing for Remote Sensing XV, Berlin, Germany, 31 August, 7477, 1-12.

Ostrowski, J. A., & Cheng, P., 2000, ‘DEM extraction from stereo SAR satellite imagery’, IEEE Geoscience and Remote Sensing Symposium proceedings, Honolulu, Hawaii, 24-28 July, 5, 2176-2178.

Paillou, P., & Gelautz, M., 1999, ‘Relief reconstruction from SAR stereo pairs: the “Optimal Gradient” matching method’, IEEE Transactions on geoscience and remote sensing, 37(4), 2099-2107.

Papathanassious, K. P., & Cloude, S. R., 2001, ‘Single-baseline polarimetric SAR interferometry’, IEEE transactions on geoscience and remote sensing, 39(11), 2352-2363.

Peltzer, G., Hudnut, K. W., & Fiegl, K. L., 1994, ‘Analysis of coseismic surface displacement gradients using radar interferometry: new insights into the Landers earthquake’, J. Geophys. Res. 99(B11), 21971-21981.

207 Peltzer, G., & Rosen, P., 1995, ‘Surface displacement of the 17 May 1993 Eureka Valley, California earthquake observed by SAR interferometry’, Science, 268(5215), 1333-1336.

Pizzi, A., & Pugliese, G., 2004, ‘InSAR-DEM analyses integrated with geologic field methods for the study of long-term seismogenic fault behaviour: Applications in the axial zone of the central Apennines (Italy)’, Journal of seismology, 8(3), 313- 329.

Pulliainen, J. T., Kurvonen, L., & Hallicainen, T., 1999, ‘Multitemporal behaviour of L- and C-band SAR observations of boreal forests’, IEEE Transactions on geoscience and remote sensing, 37(2), 927-937.

Rabus, B., Eineder, M., Roth, A., & Bamler, R., 2003, ‘The shuttle radar topography mission-a new class of digital elevation models acquired by spaceborne radar’, ISPRS Journal of photogrammetry and remote sensing, 57(4), 241-262.

Raucoules, D., Parcharidis, I., Feurer, D., Novalli, F., Ferretti, A., Carnec, C., Lagios, E., Sakkas, V., Le Mouelic, S., Cooksley, G., & Hosford, S., 2008, ‘Ground deformation detection of the greater area of Thessaloniki (Northern Greece) using radar interferometry techniques’, Natural hazards and earth system sciences, 8, 779-788.

Rignot, E. J. M., & Zyl, J. J., 1993, ‘Change detection techniques for ERS-1 SAR data’, IEEE Transactions on geoscience and remote sensing, 31(4), 896-906.

Rignot, E., Way, J. B., McDonald, K., Viereck, L., Wlliams, C., Adams, P., Payne, C., Wood, W., and Shi, J., 1994, ‘Monitoring of environment conditions in Taiga forest using ERS-1 SAR’, Remote sensing of environment, 49(2), 145-154.

Rocca, F., 2007, ‘Modeling interferogram stacks’, IEEE Transactions on geoscience and remote sensing, 45(10), 3289-3299.

208 Rodriquez, E., & Martin, J. M., 1992, ‘Theory and design of interferometric synthetic aperture radars’, IEE proceedings F: Radar & Signal processing, 139(2), 147-159.

Rodriguez, E., Morris, C. S., Belz, J. E., Chapin, E. C., Martin, J. M., Daffer, W., & Hensley, S., 2005, ‘An assessment of the SRTM topographic products’, Technical Report JPL D-31639, Jet Propulsion laboratory, Pasadena, California, 143.

Rosenfield, G. H., 1968, ‘Stereo radar techniques’, The Photogrammetric engineering, 34(6), 586-594.

Rufino, G., Moccia, A., & Esposito, S., 1998, ‘DEM generation by means of ERS tandem data’, IEEE Transactions on geoscience and remote sensing, 36(6), 1905- 1912.

Sandwell, D. T., Myer, D., Mellors, R., Shimada, M., Brooks, B., & Foster, J., 2008, ‘Accuracy and Resolution of ALOS Interferometry: Vector Deformation Maps of the Father’s Day Intrusion at Kilauea’, IEEE Transactions on geoscience and remote sensing, 46(11), 3524-3534.

Sansosti, E., 2004, ‘A simple and exact solution for the interferometric and stereo SAR geolocation problem’, IEEE Transactions on geoscience and remote sensing, 42(8), 1625-1634.

Sansosti, E., Lanari, R., Fornaro, G., Franceschetti, G., Tesauro, M., Puglisi, G., & Coltelli, M., 1999, ‘Digital elevation model generation using ascending and descending ERS-1/ERS-2 tandem data’, International Journal of remote sensing, 20(8), 1527-1547.

Schanda, E., 1985, ‘A radargrammetry experiment in a mountain region’, International Journal of remote sensing, 6(7), 1113-1124.

209 Scheiber, R., & Moreira, A., 2000, ‘Coregistration of interferometric SAR images using spectral diversity’, IEEE Transactions on geoscience and remote sensing, 38(5), 2179-2191.

Schuler, D. L., Lee, J. S., & Grandi, G. D., 1996, ‘Measurement of topography using Polarimetric SAR images’, IEEE Transactions on geoscience and remote sensing, 34(5), 1266-1277.

Schwabisch, M., 1998, ‘A fast and efficient technique for SAR interferogram geocoding’, IEEE International Geoscience and Remote Sensing Symposium, Seattle, USA, 6-10 July, 2, 1100-1102.

Shapiro, I. I, Slade, M. A., Zisk, S. H., Rogers, A. E. E., & Thompson, T. W., 1972 ‘Lunar Topography: global determination by radar: delay-Doppler stereoscopy and radar interferometry yield high-resolution three-dimensional views of the moon’, Science, 178, 939-948.

Smith, L. C., 2002, ‘Emerging applications of interferometric synthetic aperture radar (InSAR) in and hydrology’, Annals of the association of american geographers, 92(3), 385-398.

Snoeji, P., van der Valk, L., Boom, E., & Hoekman, D., 2001, ‘Effect of the ionosphere on P-band spaceborne SAR images’, IEEE Geoscience and Remote Sensing Symposium proceedings, Sydney, Australia, 9-13 July, 1, 132-134.

Taud, H., Rarrot, J. F., & Alvarez, R., 1999, ‘DEM generation by contour line dilation’, Computers & Geosciences, 25(7), 775-783.

Tison, C., Tupin, F., & Maitre, H., 2004, ‘Retrieval of building shapes from shadows in high resolution SAR interferometric images’, IEEE Geoscience and Remote Sensing Symposium, Anchorage, Alaska, USA, 3, 1788-1791.

210 Toutin, T., 1996, ‘Opposite side ERS-1 SAR stereo mapping over rolling topography’, IEEE Transactions on geoscience and remote sensing, 34(2), 543-549.

Toutin, T., & Gray, L., 2000, ‘State-of-the-art of elevation extraction from satellite SAR data’, ISPRS Journal of photogrammetry and remote sensing, 55(1), 13-33.

Van Leijen, F., & Hanssen, R., 2004, ‘Interferometric radar meteorology: resolving the acquisition ambiguity’, CEOS SAR workshop, Ulm, Germany, 27-28 May.

Vazquez, R. F., & Feyen, J., 2007, ‘Assessment of the effects of DEM gridding on the predictions of basin runoff using MIKE SHE and modelling resolution of 600m’, Journal of Hydrology, 334, 73-87.

Wagner, W., Luckman, A., Vietmeier, J., Tansey, K., Balzter, H., Schmullius, C., Davidson, M., Gaveau, D., Gluck, M., Le Toan, T., Quegan, S., Shvidenko, A., Wiesmann, A., & Yu, J, J., 2003, ‘Large-scale mapping of boreal forest in SIBERIA using ERS tandem coherence and JERS backscatter data’, Remote sensing of environment, 85(2), 125-144.

Walker, J. P., & Willgoose, G. R., 1999, ‘On the effect of digital elevation model accuracy on hydrology and geomorphology’, Water resources research, 35(7), 2259-2268.

Wang, H, Xu, C., & Ge, L., 2007, ‘Coseismic deformation and slip distribution of the 1997 Mw 7.5 Manyi, Tibet, earthquake from InSAR measurements’, Journal of geodynamics, 44(3-5), 200-212.

Wiesmann, A., Werner, C., Santoro, M., Weqmuller, U., & Strozzi, T., 2006, ‘ScanSAR interferometry for land use applications and terrain deformation’, IEEE Geoscience and Remote Sensing Symposium proceedings, Denver, Colorado, 31 July-4 August,, 3103-3106.

211 Wilkinson, G.R., 1987, ‘Book review: the technical history of the beginnings of Radar’, IEE proceedings A science, measurement & technology, 134(4), 349.

Williams A. K., 2000, ‘The role of satellite exploration in the search for new petroleum reserves in South Asia’, SPE-PAPG Annual Conference, Islamabad, Nov, 9-10.

Wimmer, C., Siegmund, R., Schwabisch, M., & Moreira, J., 2000, ‘Generation of high precision DEMs of the Wadden sea with airborne interferometric SAR’, IEEE Transactions on geoscience and remote sensing, 38(5), 2234-2245.

Yun, S., Segall, P., & Zebker, H. 2006, ‘Constraints on magma chamber geometry at Sierra Negra Volcano, Galapagos islands, based on InSAR observation’, Journal of volcanology and geothermal research, 150, 232-243.

Zebker, H. A., Farr, T. G., Salazar, R. P., & Dixon, T. H., 1994c, ‘Mapping the world’s topography using radar interferometry: the TOPSAT mission’, Proceedings of the IEEE, 82(12), 1774-1786.

Zebker, H. A., & Goldstein, R., 1986, ‘Topographic mapping from SAR observations’, Journal of geophysical research, 91(B5), 4993-4999.

Zebker, H. A., & Rosen, P., 1997, ‘Atmospheric artifacts in interferometric SAR surface deformation and topographic maps’, Journal of geophysical research solid earth, 102(B4), 7547-7563.

Zebker, H. A., Rosen, P. A., Goldstein, R. M., Gabriel, A., & Werner, C., 1994b, ‘On the derivation of coseismic displacement fields using differential radar interferometry: the Landers earthquake’, Journal of geophysical research-solid Earth, 90(B10), 19617-19634.

Zebker, H.A., & Villasenor, J., 1992, ‘Decorrelation in interferometric radar echoes’, IEEE Transactions on geoscience and remote sensing, 30(5), 950-959.

212 Zebker, H. A., Werner, C. L., Reson, P., & Hensley, S., 1994a, ‘Accuracy of topographic maps derived form ERS-1 radar interferometry’ IEEE Transactions on geoscience and remote sensing, 32(4), 823-836.

Zhou, C., Ge, L., E, D., & Chang, H.C., 2005, ‘A case study of using external DEM in InSAR DEM generation’, Geo-Spatial information science, 8(1), 14-18.

Web reference

Earth observation research centre, http://www.eorc.jaxa.jp/ALOS/en/about/palsar.htm, ALOS PALSAR, JAXA, [Last access: 24/06/2010].

Earth observation research centre, http://www.eorc.jaxa.jp/ALOS/en/index.htm, ALOS PRISM, JAXA, [Last access: 24/02/2011].

Earth remote sensing data analysis centre, http://www.palsar.ersdac.or.jp /e/guide/pdf/worldcoverage091027.pdf, [Last access: 15/03/2010].

Earthnet online, http://earth.esa.int/, European space agency, [Last access: 19/03/2010].

Geoscience Australia, http://www.ga.gov.au/remote-sensing/get-satellite-imagery- data/ordering/pricing/index.jsp, [Last access: 02/03/2010].

German Aerospace Center (DLR), http://www.dlr.de/hr/en/desktopdefault.aspx/tabid- 2326/3776_read-5679/, [Last access: 11/06/2010].

Imaging notes, http://www.imagingnotes.com/go/article_free.php?mp_id=147, [Last access: 28/12/2010].

Observing the earth, http://www.esa.int/esaEO/GGGWBR8RVDC_index_1.html, European space agency, [Last access: 24/06/2010].

213 Radartutorial.eu, http://www.radartutorial.eu/07.waves/wa04.de.html, [Last access: 24/06/2010].

Satellite imaging corporation, http://www.satimagingcorp.com/svc/ikonos-stereo- satellite-images.html, [Last access: 03/03/2010].

Spot Image group, http://www.spotimage.com/web/en/684-terrasar-x.php, [Last access: 24/06/2010].

UAVSAR, http://uavsar.jpl.nasa.gov/index.html, NASA, [Last access: 24/06/2010].

U.S. geological survey, http://gulfsci.usgs.gov/tampabay/data/1mapping/lidar/ images/Eaarl1.gif, [Last access: 11/03/2010].

Wollongong, http://www.thegong.com.au/wollongong.php, [Last access: 24/06/2010]. www.eorc.jaza.jp/ALOS/en/about/palsar.htm, [Last access: 21/08/2009] www.esa.int/esaEO/SEMGWH2VQUD_indes_0_m.html, [Last access: 21/08/2009]

214