SP-406 Volume II December 1997 ISBN 92-9092-307-5
'Fringe 96' Workshop
ERS SAR Interferometry
Zurich, Switzerland 30 September - 2 October 1996
GEN65
European Space Agency Agence spatiale europeenne 2
Corrigenda to SP-406 - 'Fringe 96•
• Printed version only:
CROSS-COMPATIBILITY OF ERS-SLC PRODUCTS A. Barmettler & al., Remote Sensing laboratories, CH (colour images printed on wrong page: 214 instead of 314; the complete and correct version is reproduced in this volume on pages 129-138)
• Printed version & CD-ROM
• listed under co-author in Table of Contents:
THE 1995 GREVENA (NORTHERN GREECE) EARTHQUAKE: FAULT MODEL CONSTRAINED WITH TECTONIC OBSERVATIONS AND SAR INTERFEROMETRY (Abstract) J.B. de Chabalier & al., lnstitut de Physique du Globe, France (listed as B. Meyer & al.)
• listed under incorrect name in Table of Contents:
OBSERVATION AND MODELLING OF THE SAINT-ETIENNE-DE-TINEE LANDSLIDE USING SAR INTERFEROMETRY 8. Fruneau & al., lnstitut de Physique du Globe, France (listed as F. Benedicte)
• listed under thematic session instead of 'Opening Session':
THE SHUTTLE RADAR TOPOGRAPHY MAPPER (Abstract) T.G. Farr & M. Kobrick, JPL, California Institute of Technologv, USA (listed under Session 1)
THE DIGITAL ELEVATION MODEL MARKET: CURRENT SITUATION & PERSPECTIVES (Abstract) L.-F. Guerre & al., Spot Image, France (listed under Session 2)
Addenda
This volume contains 18 late papers missing from the first issue of SP-406.
ESA SP-406 (Volume II): Proceedings of the 'Fringe 96' Workshop on ER<;SAR Interferometry
Published by: ESA Publications Division ESTEC, Noordwijk, The Netherlands Compiled by: T.-D. Guyenne & D. Danesy Price Code: 80 Dtl. Copyright: © 1997 European Space Agency ISBN 92-9092-307-5 Printed in The Netherlands 3
Fringe 96 - Supplement
Session 1 - Geology & Hazards Applications Chairman: H. Laur MONITORING OF SMALL MOTIONS IN MINING AREAS BY SAR INTERFEROMETRY L. Timmen & al., GeoForschungsZentrum. Germany 5
APPLICATION OF SAR INTERFEROMETRY TO THE IMAGING AND MEASUREMENT OF NEOTECTONIC MOVEMENT APPLIED TO MINING AND OTHER SUBSIDENCE/ DOWNWARP MODELLING R. Stow, Doncaster College, UK 9
Session 2 - DEM Applications Chairman: S. Coulson DEM GENERATION BY MEANS OF ERS TANDEM DATA A. Moccia & al., Dipartimento di Scienza e Ingegneria de/lo Spazio "L.G. Napolitano", Universita di Napoli, Italy (listed on CD-ROM under G. Rufino; no link to Contents page) 15
THE EFFECTS OF DIFFERENT LAND COVERS ON THE ACCURACY OF INTERFEROMETRIC DEM A. Wehr & al., Institute of Navigation, University of Stuttgart, Germany (listed on CD-ROM under K-H. Thiel; no link to Contents page) 33
VALIDA TION OF HEIGHT MODELS FROM ERS INTERFEROMETRY D. Small & D. Ntiesch, Remote Sensing Laboratories CRSL),Switzerland 43
Session 3 - Forest & Landcover Applications Chairman: M. Borgeaud INITIAL TESTING OF TANDEM QUALITY FOR INSAR APPLICATIONS, EXAMPLES FROM TAIWAN, MADAGASCAR, ZAIRE, MALI, IVORY COAST AND GREENLAND (Abstract) G. Solaas & F. Gatelli, RSIE Data Utilisation Section, ESRIN, ESA, Italv 55
INTERFEROMETRY FOR FOREST STUDIES N. Floury & al., Centre d'Etudes Spatiales de la Biosphere, France 57
ANALYSIS OF ERS-SAR TANDEM TIME SERIES USING COHERENCE AND BACKSCATTERING COEFFICIENT 0. Stebler & al., Remote Sensing Laboratories (RSL), Switzerland 71
PHASE SHIFT INTERPRETATION ON ERS-1 INTERFEROGRAMS AND LABORATORY MEASUREMENTS J.-P. Rudant & al., UMLV & UPMC, France 83
LAND APPLICATIONS USING ERS-1/2 TANDEM DATA U. Wegmtiller & C. L. Werner, GAMMA Remote Sensing AG, Switzerland 97 4
Session 4 - Processors & Products Chairman: J-P. Guignard AN INTEGRATED METHODOLOGY FOR DEM COMPUTATION THROUGH FUSION OF INTERFEROMETRIC, RADARGRAMMETRIC AND PHOTOGRAMMETRIC DATA (Abstract) I. Tannous & F. Le Goff, SYSECA, France 113
THE UCL 3D IMAGE MAKER SYSTEM FOR AUTOMATED DIFFERENTIAL SAR INTERFEROMETRY (Abstract) M. Upton & al., University College London, UK 115
A WORKSTATION FOR SPACEBORNE INTERFEROMETRIC SAR DATA (Abstract) M.W.A. van der Kooij & al., Atlantis Scientific Systems Group Inc., Canada 116
Session 5 - Algorithms & Techniques Chairman: E. Attema COMPARISON OF REPEAT TRACK INTERFEROMETRIC CORRELATION FROM ERS-I, ERS TANDEM, SIR-C AND JERS-I (Abstract) C. L. Werner & al., Jet Propulsion Laboratory, USA 117
NEW METHODS OF PHASE UNWRAPPING AND BASELINE ADJUSTMENTS IN SAR INTERFEROMETRY H. Tarayre-Oriot & D. Massonnet, ONERA. France 119
ATMOSPHERIC ARTIFACTS ON INTERFEROGRAMS H. Tarayre-Oriot & D. Massonnet, ONERA, France 125
Session 6: Validation Chairman: G. Solaas CROSS-COMPATIBILITY OF ERS-SLC PRODUCTS A. Barmettler & al., Remote Sensing Laboratories, CH 129
Session 7 - Ice & Glaciers Chairman: Y-L. Desnos THE USE OF TANDEM DATA IN THE ANTARCTIC AREA X. Wu & al., Institute of Navigation, University of Stuttgart, Germany (not listed on CD-ROM Contents page) I39
GLACIOLOGICAL STUDIES IN THE ALPS AND IN ANTARCTICA USING ERS INTERFEROMETRIC SAR H. Rott & A. Siegel, lnstitut filr Meteorologie und Geophysik UniversitiitInnsbruck, Austria (not listed on CD-ROM Contents page) I49 5 Monitoring of Small Motions in Mining Areas by SAR lnterf erometry
Ludger Timmen GeoForschungsZentrum Potsdam Telegrafenberg Al 7, D-14473 Potsdam [email protected] Xia Ye GeoForschungsZentrum Potsdam Telegrafenberg Al 7, D-14473 Potsdam [email protected] Christoph Reigber GeoForschungsZentrum Potsdam Telegrafenberg Al 7, D-14473 Potsdam [email protected] Rolf Hartmann Jena-Optronik GmbH Prussingstr. 41, D-07745 Jena Thomas Fiksel Jena-Optronik GmbH Prussingstr, 41, D-07745 Jena W. Winzer Jena-Optronik GmbH Prussingstr. 41, D-07745 Jena Jochen Knoch-Weber WISMUT GmbH Jagdschankenstr. 29, D-09117 Chemnitz
Abstract
In the former WISMUT uranium ore mining and processing area (Thuringia, Germany) controlling of small surface movements with "cmt'-accuracy is required. A promising tool is the D-INSAR technology, which could reduce man power spending, compared with other geodetic techniques (GPS, precise levelling). The GeoForschungsZentrum Potsdam (GFZ), the DASAJena-Optronik company (DJO) and the WISMUT company participate in a joint project (sponsored by the German Space Agency DARA) to apply INSAR in the WISMUT area, using a stable corner reflector (CR) reference net and measuring CRs. The phase difference between the CRs are investigated with the objective to monitor CR movements. The network has been initially positioned by GPS and is also controlled by terrestrial geodetic measurements. Experiments are presently being carried out by shifting CRs out of position. The project status and first results are presented in this paper. Keywords: corner reflectors, mining area, small motions Introduction
Under the supervision of GeoForschungsZentrum Potsdam (GFZ), a SAR interferometry (INSAR) application project for monitoring surface deformations in the WISMUT mining area started in September 1995.
Over more than 40 years the former Soviet-German corporation WISMUT was intensively occupied with exploitation and processing of uranium ore in Thuringia (Eastern Germany). In 1990, the uranium ore production stopped and the WISMUT Ltd. was established to control or perform the redevelopment measures in the 37 km2 large area.
Seeking for cost-effective geodetic measurement methods to monitor anticipated surface movements, the spacebome SAR interferometry is obviously a promissing tool. The capability of differential INSAR (D-INSAR) was already shown in the "Bonn-Experiment", performed by ESA-ESRIN, INS-University of Stuttgart and Politecnico de Milano (Prati et al.
Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Sept. - 2 Oct. 1996 (ESA SP-406,Vol. II, December 1997) 6
performed by ESA-ESRIN, INS-University of Stuttgart and Politecnico de Milano (Prati et al. 1993). Several drawbacks for SAR applications are inherent in the WISMUT area: vegetation. limited space between mining facilities, abrupt terrain height changes in the precincts of slagheaps and of sludge settlement basins. To overcome some of the terrain deficiencies and to achieve the best possible accuracy, corner reflectors (CRs) are employed. A CR represents a defined geometrical reference point in the radar image and shows favourable reflection characteristics with respect to intensity and stability. The detection of surface movements with "cm-accuracy" is required.
In a joint project the GFZ, the DASA Jena-Optronik company (DJO) and the WI SMUT company are developing procedures to derive point movements from a CR network analysis. This includes the elaboration of scientific background as well as practical testing. The methodology is evolved on the principle of the division of labour by GFZ and DJO. In addition. GFZ is mainly focusing on the project assessment where as the partner from industry, DJO will put the technique into practice. The WISMUT defines the user demands, and assesses the project continually from its point of view. Project MODIFI
Within the project MODIFI (Monitoring of Displacement Fields by Radar Interferometry) the field work has reached a level of full availability of a corner reflector network (10 CRs) for experiments. The array is deployed over an area of l 5x6 km2.
All CRs, with an edge length of 1.50 m, were installed on concret platforms of 2.5x2.5x0.8 m3 dimension, see Fig. I. They were centered above GPS groundmarkers with an accuracy of 1 mm, which means that the CR apex and the groundmarker is in one vertical line. The CRs can be rotated allowing an orientation on descending or ascending orbits of ERS-1/2. The axis of rotation coincides with the vertical line defined by apex and groundmarker. The CR base is fixed to the concret foundation in a way that no CR displacement by wind is possible. The CR funnels may be removed to perform GPS measurements in a specific height above the ground marker using a centering rod with a top for a GPS antenna. The CR points are tied to stations of the International GPS Service for Geodynamics (IGS), Potsdam and Wenzell, to ensure an absolute accuracy of better than 1 m. This is comparable to the accuracy of the ESA precise orbits for ERS-112, generated at GFZ/D-PAF, Oberpfaffenhofen. The coordinate differences between CRs are controlled within 5 mm. All INSAR calculations are performed within the ITRF. the International Terrestrial Reference Frame.
Fig. I: GFZ corner reflector installed in the former WISMUT mining area (Thuringia/Germany) 7
The WISMUT mine surveyors integrated the CR/OPS points in their terrestrial geodetic nets. This ensures an additional control of platform stability. Because the center markers are continuously occupied by CRs, eccenters were placed in 3 comers of each platform. Small movements like vertical shifts or tilt effects are detectable by precise levelling. The height differences between center and eccenters are determined once by levelling, the Gauls-Kruger coordinate differences have been derived by simple distance measurements.
2 CRs are equipped with translation devices to perform defined shifts out of position of the funnel in any direction. The radar echo of such a "measuring" CR has to be applied to the signals of the 8 reference CRs. Experiments are presently being carried out. 2 flexible "mobile" CRs are under preparation to test different network configurations (optimization). Real surface deformations are anticipated after flooding the underground mines in a few years. If MODIFI ends successfully, a new application project with a densified CR network is scheduled afterwards. Data evaluation strategy and first results
DJO as well as OFZ develop a methodology which allows the detection of small point movements within a CR network using SAR interferometry. The diagram in Fig. 2 describes the OFZ concept. A "flat earth" corrected interferogram is derived from a SAR SLC image pair and the corresponding precise orbit data. For conversion of interferogram phases into height values, the knowledge of accurate imaging geometry is required. This is obtained from the orbit data and the OPS heights of the CRs depicted in the radar image. An optimization is performed to adjust the interferogram to the heights of the reference CRs. In the following step the phase values of the measuring CRs are investigated. They should be in agreement with phase values calculated from OPS heights. The difference is a measure for a CR motion.
Si•R Image 1 I ISAR Image 2 I IERS-1 /2 Orbits GPS-Heights
"Flat Earth" INSAR Fringes Geometry
Interferogram
Phase/Height Optimized Motion Interf erogram Detection
Fig. 2: Data evaluation strategy to detect relative point movements between corner reflectors (project MOD/FI)
In a first experiment, an ERS-112 tandem data pair, acquired at 8.2. and 9.2.1996 with base line length of 142 m, was evaluated. At that time 5 CRs were available. The objective was the determination of the heights of 4 CRs relative to a CR with fixed height. Tab. 1 compares the results, obtained at OFZ and at DJO, with the "true" OPS results. DJO attained a very good agreement within 3 m. The OFZ result shows one larger discrepancy of 10 m for point CR9. An accuracy of 10 m corresponds to an error of 50° in phase or 4 mm in wave lenght. Considering that a vertical CR shift out of position of only 4 mm causes a misinterpretation of 8
10 m, the overall result is satisfying.
Tab.1: Comparison of corner reflector results, obtained at GFZ and at DJO, with "true" GPS results.
Difference II GFZ II Difference I Station JI OPS-Height II INSA~=~eight (DJO-GPS) INSAR-Height (GFZ-GPS) CRl 11 306.33 m 306.33 m II I II 306.33 m I CR2 II 296.09 m 294.48 m II 1.61 m II 296.41 m +0.32 m CR3 II 335.79 m 334.99 m II 0.80m II 334.36 m 1.43 m CR4 II 328.36 m I 331.20 m II +2.84 m II 325.87 m 2.49 m CR9 II 379.12 m 11 381.12 m II +1.40 m II 370.31 m 9.41 m Encouraged by this first result a SLC image from 4.1.1996 was included in the radar data analysis. After the acquisition of that ERS-1 image, the CR2 position was moved by 2 cm in range direction. Now the objective was the detection of that motion. The corresponding interferogram (4.1./8.2.96, baseline 272 m) was disarranged in most of its parts, the low coherence made it impossible to derive any result for the terrain and also for the CR points. The reason may be found in the different environmental conditions at the two epochs. During the acquisition time on 4.1.1996 (10:00p.m.) the weather was misty with an overcast sky, 95% relative humidity and -7°C, the earth surface was dry without any snow coverage. On 8.2.1996 (similar at 9.2.1996) the weather was clear with no clouds, 90% rel. humidity and -l4°C, the surface was covered by snow (no grass visible). In this special case, the employment of CRs for point motion detection may be restricted due to: 1) perturbations in the radar propagation delay through the troposphere, and/or 2) changing properties of the surrounding terrain surface (radar penetration into the frozen soil and the crusted snow coverage, influence of vegetation). Conclusions
The first result demonstrates the accuracy potential of D-INSAR technique. A subcentimeter accuracy may be possible. But the described result is obtained under favourable environmental conditions. The evaluation of the SLC image pair, acquired in the ERS-1 35 day repeat cycle, gave no result. This is understandable considering that the radar measurements of the CRs are not independent from the environment. In additional experiments, using images with negligible atmospheric perturbations, it has to be assessed how strong the surrounding terrain with its changing surface conditions may affect the CR observations. The CR reflectivity may not be as dominant as expected, which would require a wider area surrounding the CRs, with a cleared plain surface. Acknowledgements
The authors thank the German Space Agency DARA GmbH for sponsoring the project MODIFI through the research grant 50 EE 9429, and ESA for providing the SAR SLC images (project code A02.D133). References
Prati, C., Rocca, F., Monti Guarnieri, A., 1993: SAR Interferometry Experiments with ERS-1. Proceedings First ERS-1 Symposium - Space at the Service of our Environment, Cannes, France, 4-8 November 1992, ESA SP-359, pp. 211-218. 9 Application of SAR Interferometry to the Imaging and Measurement of Neotectonic Movement Applied to Mining and other Subsidence/Downwarp Modelling.
R. Stow Doncaster College, Waterdale, DNl 3EX Doncaster, UK, Phone 44 130 2553553, Fax 44 130 2553559
Abstract
This collaborative project Between Doncaster College,RJB Mining(UK), GEC Marconi Research Centre, and Matra Marconi Space, will integrate Synthetic Aperture Radar (SAR) ERSl and ERS2 satellite remotely sensed images and their derived interferograms(INSAR), differential interferograms(DifflNSAR) and associated Digital Elevation Models (DEM), with mining (and other types of), subsidence mathematical models. The subsidence models are used to determine the maximum mineral extraction, within the subsidence constraints imposed by conditions of permit, granted in the UK under the Town and Country Planning Act. The objective of the project is to use SAR derived elevation data to feed back to the subsidence model to improve subsidence prediction, mine productivity, conformity with regulatory consent and minimised environmental and economic impacts. With many complex and inconsistent variables, the dynamics of mining subsidence are difficult to model. Reliability of modelling is extremely variable, reported actual to predicted subsidence in a range 48 - 773%. There is a general recognition of the inadequacy in both the empirical approach to subsidence modelling (SEH) and the more rationalist computer models. Remotely sensed data has the potential to improve subsidence modelling accuracy by significantly increasing the quantity of feed back data, compared to the quantity of data practicable by conventional land surveys. While land surveys conducted along road and canal bank lines produce a relatively small, highly accurate data base, diff DEMs have the potential to provide elevations at 1cm vertical resolution and 50m horizontal resolution. The significantly increased quantity of feed back data by INSAR should thus provide a more rational, statistically significant, scientific basis for the Mineral Surveyor to adjust the subsidence model. The interferometric processor developed at MRC will be applied to demonstrate the monitoring of subsidence for two thoroughly validated and contrasting sites in the Czech Republic and the UK. The aims of the demonstration will be to assess the vertical accuracy and precision with which subsidence can be measured and to comment on whether the user specified horizontal location accuracy requirements are achieved. As an engineered movement, mining subsidence offers the best opportunity to develop the remote sensing techniques, further applicable to natural tectonic events, since the motions can be predicted spatially and temporally (albeit within the current range of error), are routinely land surveyed and are generally intensively geologically surveyed. The mining activity, being continuous, can provide a regular stream of dynamic statistical data for INSAR and Model validation, under near optimal controlled conditions, not possible under natural conditions.In the context of liability for economic and environmental impacts, and risk to major investment and production however, the technique has to be developed and demonstrated to a high level of dependability before it could be considered ready for application demonstration. Once validated, the extended INSAR techniques and improved subsidence modelling could be expected to contribute to wider opportunities and understanding in geological modelling
Proceedmgs of the 'Fringe 96' Workshop on ERS SAR Interferometry, lunch. 30 Sept - 2 Oct. 1996 (ESA SP-406.Vol. II. December 1997) 10
expected to contribute to wider opportunities and understanding in geological modelling generally, such as earthquake, landslip, heave, isostatic uplift and downwarp, volcanoes, glacial, fluvial and coastal geomorphology. Introduction
The BONN Experiment I in 1992 by ESA/ESRIN, University of Stuttgart and Politechnico di Milano, demonstrated the capability of SAR differential interferometry to measure surface movement to centimetre resolution under experimental conditions using comer reflectors and transponders. The NAPEX Experiment 2 is currently extending this work on the area around Vesuvius, Naples, where surface relaxation of 3cm/annum is occurring. Due to the relative predictability and continuity of mining subsidence compared to natural tectonic movement and the availability of precision ground truth survey data, this proposal represents an unequalled opportunity to validate INSAR techniques and advance them a step towards operational application. The current development status of SAR Interferometry is discussed further. Mining subsidence is typically limited by planning consent in the UK to 1 metre. The dynamics of subsidence involve vertical movement spreading beyond the worked panel at around 37 degrees from the vertical to a maximum depth nominally around 70% of the thickness of the worked panel. Surface strain results in regions of tension over the general area and compression creating a hump along the centre line. Variation in subsidence rate occurs due to inconsistency of geological formations, rock mechanical properties, and the overall uniqueness of each situation, e.g. depth of workings, mining technique, seam inclination, adjacent/subjacent working, face ends etc. RJB Mining UK, as the end user partner in this proposal, is making available its land survey data and subsidence modelling, particularly for the Selby Coalfield. RJB will be collaborating with Doncaster College on land surveying, ground truth testing of differential DEMs, adjustment of DEMs and identification of anomalies, statistical analysis, satellite data interpretation and analysis, mining subsidence computer program modification and data feedback to subsidence modelling. Doncaster College through partnership with VSB University, Czech Republic, also has access to subsidence data and modelling for the Silesian Coalfield in North Moravia where 40 metre subsidence has been experienced in the recent past. This opportunity will further enable testing of experimental repeatability and the effects of different elevations, ground movement range and topography. The area is a focus of attention for environmental improvement/investment/monitoring, attracting both EC (EBRO) and US aid funding. The dynamics of mining subsidence are difficult to model. The Subsidence Engineer's Handbook (SEH) 3, revised by the National Coal Board in 1975, is still the standard reference text, applied largely by use of generalised graphs, usually empirically derived from observation, the results adjusted by local experience. Computer models are generally based on input parameters of seam thickness, depth of mining, amount and direction of seam gradient, adjusted in line with SEH, estimated on a system of annular zones in three dimensions. Reliability of modelling is extremely variable.CR Ferrari 4 of British Waterways (responsible for the security of undermined canal banks), has reported actual to predicted subsidence in a range 48 - 773%. It can be argued that the empirical nature of the models that form the basis of subsidence modelling is, on the one hand appropriate for the inconsistent conditions experienced. On the other hand the refinement of empirical models is dependent on quantity of data. Currently Mineral Surveyors conduct line surveys of subsidence along roads and canal banks, this represents a very small proportion of the area of influence of the mine working. This small sample size will contribute to the wide variation in actual to predicted subsidence currently experienced. Remotely sensed data has the potential to improve subsidence modelling accuracy by significantly increasing the quantity of feed back data, compared to the quantity of data practicable by conventional land surveys. While land surveys conducted along road and canal bank lines produce a relatively small, highly accurate data base, SAR 11 along road and canal bank lines produce a relatively small, highly accurate data base, SAR differential DEMs have the potential to survey the whole area providing elevations at 1cm vertical resolution, at 12.5m horizontal resolution. The significantly increased quantity of feed back data by SAR Interferometry should thus provide a more rational, statistically significant, scientific basis for the Mineral Surveyor to adjust the subsidence model.
By use of near real time images such as the Rapid Information Dissemination System (RAIDS) service being developed by MMS, it should be possible to monitor subsidence comprehensively at (currently) 35 day intervals, enabling rapid response to variance from modelled forecasts with clear economic and environmental benefits.
In order to exploit remote sensing data in monitoring subsidence and subsequently re-engineering mine workings, a clear program of technique development and demonstration is required. The remote sensing techniques are not proven, especially at the levels of confidence necessary when, environmental protection, major property damage liability, stability of railways, roads, canal banks and river and coastal flood protection are at risk, along with multi million pound mine investment and productivity.
Once the remote sensing techniques have been developed, it will be necessary to demonstrate application of these techniques through to the point of production of demonstration (calibrated) subsidence maps. The project will not end at this point. It is very important that the improved subsidence maps derived from remote sensing data are linked in to the appropriate subsidence model. The project will only be complete when the end user(s) are shown clear evidence that the subsidence model predictions are a clear and reliable improvement over those available without remote sensing data. SAR Interferometry technique development
SAR interferometry is a relatively new technique for the generation of topographical height information. As such there are a number of issues which need to be explored in terms of image coherence and DEM calibration. There are two main factors which determine image coherence, firstly the perpendicular baseline separation of the image acquisitions (Bperp) and secondly changes in ground scattering characteristics between image acquisitions. Coherence analysis
In order to extend the range of useful image pairs for INSAR, it is intended to experiment with generating elevation data from images of varying coherence. To establish to what extent useful elevation data can be extracted, less than optimum images will be filtered for the most coherent areas. In the extreme this may be limited to stable linear features such as roads, large buildings, hard surfaces etc.
As a routine process, filtering to remove the least coherent areas (caused by vegetation change, cultivation and variability in soil moisture etc) should improve the overall resolution and correlation with ground truth of the remainder of the image. Operational ranges of 'minimum useful coherence area' and 'maximum tolerable incoherence' should result, supported by correlation values with ground truth, to enable utilisation under the fullest possible image availability range. This is crucial for increasing the operational availability of INSAR imagery to commercial demand. The ability to specify a coherence coefficient for an image pair, (on a similar basis to specification of cloud cover on LANDSAT images) or pre filtered products to 'coherence values' and/or 'coherent area' specifications would advance the whole INSAR technique. Bperp to resolution relationship
The physical separation of repeat orbits has to be within a perpendicular baseline (Bperp) orbit separation tolerance to satisfy interferometric geometry. Bperp up to 600m is the limit for INSAR but precision deteriorates beyond 300m. There is ambiguity over the relationship between Bperp and INSAR resolution and application. With the volume of data available for 12
between Bperp and INSAR resolution and application. With the volume of data available for the UK and Czech test areas, it is intended to derive a relationship between Bperp and DEM/Differential DEM resolution, in order to establish repeatable Bperp values for optimum, median and minimum useful resolution of surface change/elevation, under various surface/topographical conditions. SAR subsidence demonstration.
The interferometric processor developed at MRC will be applied to demonstrate the monitoring of subsidence for two thoroughly validated and contrasting sites in the Czech Republic and the UK. The aims of the demonstration will be to assess the vertical accuracy and precision with which subsidence can be measured and to comment on whether the user specified horizontal location accuracy requirements are achieved.
Consideration will be given as to how the accuracy of results varies as a function of type of terrain, weather conditions and vegetation conditions. On the basis of this, comments will be made on the circumstances under which the application will be viable, given a relationship between coherence and differential DEM precision and accuracy. Advancement of Subsidence Engineering Science
There is a general recognition of the inadequacy in both the empirical approach to subsidence modelling (e.g. SEH) and the more rationalist computer models. There a high expectation that this new opportunity in remote sensing, should enhance the understanding of the dynamics of subsidence, the neotectonic imaging and modelling techniques then also being applicable to other geological processes. As an engineered movement, mining subsidence offers the best opportunity to develop the remote sensing techniques, further applicable to natural tectonic events, since the motions can be predicted spatially and temporally (albeit within the current range of error), are routinely land surveyed and are generally intensively geologically surveyed. The mining activity, being continuous, can provide a regular stream of dynamic statistical data for INSAR and Model validation, under near optimum controlled conditions, not possible with unpredictable natural events. In the context of liability for economic and environmental impacts, and risk to major investment and production however, the technique has to be developed and demonstrated to a high level of dependability before it could be considered ready for application demonstration. 1. It is at this stage hypothesised, that the increased statistical significance of INSAR DEM feedback will improve the reliability and precision of the SEH modelling methodology. 2. Computer models are deemed by Mineral Surveyors to oversimplify subsidence dynamics, but clearly are the appropriate tool for dynamic modelling generally. Incorporation oflNSAR DEM data could provide well researched, statistically significant model adjustment, for specific typical geological structures, or localised field conditions, but subject to rigorous validation testing. A general technique for generation of geological model adjustment could be developed, given opportunity for sufficient trials ofrepeatability. 3. Hybrid modelling combining SEH, Computer Model and INSAR derived factors could be expected to draw on the strengths of each subsidence modelling technique. 4. The scale of data collection theoretically possible through INSAR, in 3 Dimensions and multi temporally by satellite repeat cycle, should enable 3D animation of tectonic movement and multivariate spatial/geostatistical analysis. There is a prospect of thus identifying dynamics and explanations of motions, not previously apparent or adequately scientifically explained, thereby improving the science and understanding of subsidence. Once validated, the extended INSAR techniques and improved subsidence modelling could be expected to contribute to wider opportunities and understanding in geological modelling generally, such 13 as earthquake, landslip, heave, isostatic uplift and downwarp, volcanoes, glacial, fluvial and coastal geomorphology. Progress to Date
Research funding from the British National Space Centre started in July 1996. Preliminary results indicate that mining subsidence can be detected by INSAR but that availability of INSAR data is less than optimum. The ERS Tandem Mission has produced fewer opportunities for INSAR pairs at 1 day repeat under ideal ground conditions of minimum vegetation than expected, due to the Tandem Mission only spanning one winter. There were also gaps in availability of INSAR pairs for the UK test site during the Mission. There is thus the challenge of producing interferograms and DEMs from sparse data after filtering for coherent areas from longer repeat cycle INSAR pairs. This challenge is considered to be important because commercial application of the technique depends on being able to produce useful results with available data, which is not neccesarily of optimum specification. References
1. SAR Interferometry Working Group, Proceedings of 1st Workshop. ESA. 12.10.92 2. SAR Interferometry with ERS 1. S.N Coulson. Earth Observation Quarterly. no.40. April 1993
3. Subsidence Engineers Handbook. National Coal Board Mining Department. 1975. 4. The Case for Continuing Coal Mining Subsidence Research. C. R Ferrari. UK Minerals Industry Conference (proceedings). University of Leeds. 3-5 April 1995. 5. Cordey, R.A. and Walker, N.P., 1994, "A demonstration of terrain height mapping from satellite SAR images", GEC-Marconi Technical Report MTR 93/14A. 6. Partington, K.C., Walker,N.P., Cordey, R.A. and Hines D., 1993, "DEM Generation from SAR images", GEC-Marconi Technical Report MTR 93/03A 7. Partington, K.C. Rye, A.J., Wright, P.A. and Smith, P.J., 1995, "High level software design for the production of synthetic aperture radar-derived digital elevation models", GEC-Marconi Technical Report MTR95/52A. Solaas, G.A., 1994, "ERS-1 interferometric baseline algorithm verification", ESA Report ES-TN-DPE-OM-GS02, version 2.0 8. Wright, P.A. 1994, "An investigation of interferometric SAR techniques for DEM generation in the intertidal zone", GEC-Marconi Technical Report MTR 94/54A. 9. Stow, R.J., 1995, "Application of SAR interferometry to survey neotectonic movement due to mining subsidence", Proceedings of SAR Interferometry Workshop, 27.11.96, GEC-Marconi Research Centre,
15 DEM generation by means of ERS tandem data
Giancarlo Rufino and Dipartimento di Scienza e Ingegneria dello Spazio "L.G. Antonio Moccia Napolitano", Universita di Napoli "Federico II", Piazzale Tecchio 80, 80125 Napoli, Italy [email protected] Salvatore Esposito CO.Rl.S.T.A., Piazzale Tecchio 80, 80125 Napoli, Italy
Abstract
This paper presents an application of ERS tandem data to Digital Elevation Model (DEM) generation. The selected test-site is the Sannio-Matese area (35 x 40 kmz) in Southern Italy, where several Corner Reflectors (CRs) were deployed to be used as Ground Control Points (GCPs) for height measurement accuracy validation. First of all an analysis of the CR response in radar images is presented. Then the procedure for interferogram formation, from image pair geometric registration to phase noise reduction, is described in details. A quantitative analysis is also performed by comparing these interferograms to the corresponding products obtained by using the ISAR software, officiallydistributed by ESA. Reported coherence values show that only tandem pairs allow an efficient interferometric processing to be performed, thanks to their short time baseline (one day), whereas coherence adequate for differential interferometry could not be achieved. The method adopted for the computation of the interferometric baseline components on the basis of satellite orbital data is described, including the GCP-based corrections. The procedure has been applied to obtain DEMs of a quite high coherence 10 x 10 km2subarea and the best attained values of the GCP height measurement accuracy have been about 4 m. Finally the DEMs are compared, giving rms differences less than 20 m in the best case. Keywords: Topography derivation, Orbital data. Introduction
The tandem operation of ERS-1 and ERS-2 is the first space mission aimed at Synthetic Aperture Radar (SAR) interferometric coverage on a global scale and with a short temporal baseline (1 day). The availability of such data certainly deserves great interest because several authors have pointed out the need of spacebome missions for global topographic mapping by means of SAR interferometry (Topsat Working Group,1994)and the problems connected to DEM production when temporal decorrelation effects are significant (Zebker et al., 1992).
In the last few years we have been studying the above aspects, focusing our interest also on ERS tandem data. In particular, CO.Rl.S.T.A. (Consortium for the Research on Advanced Remote Sensing Systems) was one of the investigators of the research "Earthquakes prediction in tectonic active areas using space techniques" founded by Commission of the European Communities contracts. The CO.Rl.S.T.A. responsibility was related to the detection of small crustal motions by means of space-based techniques (e.g. SAR interferometry and GPS) (CO.Rl.S.T.A., 1995, 1996). The area of interest was located in Southern Italy (Sannio-Matese region, about 35 x 40 krn-',centered on 41°l 5'N 14°25'E) and was characterized by quite heterogeneous land use: wide agricultural extensions, forests, bare soils, urban and industrial sites, lakes and river basins. Furthermore the terrain elevation extended from 100 to 1100 m, including large flat areas and steep reliefs. Several CRs were deployed on the test area, to be used as reference targets for interferometry, by the research
Proceed1ngs of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zuricn, 30 Sept. - 2 Oct. 1996 (ESA SP-406,Vol. II, December 1997) 16
co-investigators headed by prof. P. Murino (University of Naples, Department of Space Science and Engineering). Consequently, and besides successive applicative interests, this area certainly represents an interesting test bed for interferometric processing.
This paper describes the procedure we developed and applied to the Sannio-Matese area to produce DEMs starting from Single-Look Complex (SLC) tandem data. After selection of interferometric pairs by means of the baselines available at the ESA/ESRIN server, the procedure consists of CRs identification, geometric registration of the pair and interferogram production, phase unwrapping and baseline estimation. Quantitative evaluations were performed by means of coherence analysis and DEM accuracy assessment. To this end, the "ISAR-Interferogram Generator" software, officially distributed by ESA, was applied to repeat the geometric registration, for coherence validation. Whereas, the CRs heights were used to refine baseline computation and as GCPs to check the height accuracy. CR detection
The CRs were deployed over a quite large area without following a particular geometric pattern in along or across-track direction. Furthermore additional man-made point targets were present in the densely inhabited area. As a consequence, the CRs were not immediately identified in available SLC images. Therefore, a more thorough radiometric analysis was required. Starting from the knowledge of the CR geographic coordinates, the first step of the procedure consisted in the extraction of a small area, approximately centered in each corner, by means of visual inspection. Then the criteria proposed by Bruzzi et al. (1982) were applied to evaluate quantitatively CR the location. The best point targets were identified by means of two coefficients: test_A, related to the ratio between the background and the pixel amplitude, and test_B, a measure of the peak width. Since at this stage it was not yet possible to identify univocally all the CRs, a more accurate analysis was carried out by computing the broadenings, the Integrated Sidelobe Ratios (ISLRs) and the Peak Sidelobe Ratios (PSLRs) both in range and azimuth directions (Moccia et al.,1994b). Table I summarizes the overall results of the above procedure applied to the considered SLC images (table 2). Unfortunately, the radiometric quality was poor and the number of satisfactory CRs was limited. Geometric registration and interferogram production
The selected area for interferometric processing was a subset of a quadrant consisting of 9000 single-look azimuth pixels x 1900 single-look slant range pixels, including all the 9 CRs listed in table 1. A preliminary coarse registration at pixel accuracy of each interferometric pair was performed by means of a 2D rigid translation based on the previous localization of the 9 CRs. Our procedure for fine registration (Moccia et al., l 994a) is based on the automatic identification of a large number of GCPs in addition to the already identified CRs. This task cannot be easily accomplished with SAR data covering large areas (Dowman, 1992) and requires a careful strategy. To this end each image was divided into 512 x 128 pixels subareas and for each subset the brightest point target was assumed as GCP. Subsequently, each subarea was I0 times oversampled by means of cubic B-splines and the subpixel shifts between homologous areas were computed by using the cross-correlation of the GCP amplitudes (table 3). Finally, the geometric registration was performed by means of bicubic polynomials, whose coefficients were computed with least square approximation, using as input the subpixel shifts. Shaded regions in table 3 show decorrelated areas where the procedure was not able to find out the shifts. This result will be verified also by the coherence analysis shown in the next paragraph. The range and azimuth shifts computed for the 3-4 tandem data are quite regular. In particular the azimuth shifts are nearly constant over the whole image, while the range shifts increase of approximately 1 pixel from near to far range. This is probably due to slight orbit misalignments and/or attitude differences. Furthermore, the two images were focused independently, each one with its Doppler centroid frequency and bandwidth varying from near to far range. Differently, the 1-2pair shifts are quite large and exhibit a rotation between
-----·-- 17 the images. This is due to an orbit misalignment which determines a not negligible baseline variation, as it will be demonstrated later.
The resulting interferogram was oversampled 4 times by using the same technique and, finally, a coherent multilook (20 azimuth looks, 4 range looks) was executed averaging the complex values in order to give the maximum likelihood estimation of the phase (Rodriguez, 1992; Werner et al., 1992). Coherence analysis
Several authors showed that coherence is a significant parameter to measure the interferogram quality and the capability of applying efficiently phase unwrapping procedures (Zebker et al., 1992; Small et al., 1995). Figure 1 shows the coherence map corresponding to the 1-2tandem pair. Large decorrelated areas exist, as expected considering the results pointed out in the previous paragraph (shaded subsets in table 3a). Low coherence regions correspond mainly to steep reliefs. Furthermore decorrelated lakes and rivers can be easily identified. Satisfactory coherence coefficients are obtained over flat vegetated areas. The coherence map of the other interferogram is quite similar and it is not reported for the sake of brevity. The average values over the entire image are listed in table 4. To avoid decorrelation problems and speed up the processing, we decided to limit our successive activities to a square subset (about 10 x 10 km-, corresponding to 512 x 512 pixels after the previously described processing), put in evidence in the lower right comer in figure 1, where adequate coherence coefficients were available (table 4). The selected area comprises 3 CRs (numbers 6, 7 and 8 in table 1) and 3 more GCPs, exactly located also on topographic maps and uniformly distributed over the scene. Although the CRs radiometric quality was limited and obviously the 3 additional GCPs were not exactly bright man-made point targets, the availability of reference points will be extremely useful to check the computed height quantitatively and to refine the baseline estimation, as it will be shown in the following. Finally, figures 2a and 3a depict the obtained interferograms. In order to check our results, we selected as reference the product of the "ISAR-Interferogram Generator" software, officially distributed by ESA (Version 3.0 717/95).This software performs an iterative geometric registration, produces and filters the interferograms and is also able to remove the contribution of flat Earth without using orbital data. Since we do not apply any filtering to the interferograms and our final product is a DEM, we could only compare an intermediate result: that is, the coherence of an interferometric pair after geometric registration. Due to disk space and time limitations we performed the comparison using only the selected subset as input. In fact, the ISAR software requires an amount of disk space much larger than the dimensions of the SLC images, which in our case were 140 MB large. After geometric registration, a coherent multilook was applied to obtain the same pixel size in the complex images processed by using ISAR routines and our procedure. Finally, we computed the coherence coefficients, confirming exactly the results listed in table 4. The next step of processing is the phase unwrapping. Our procedure consists of two phases. First, in order to avoid error propagation, it is necessary the identification of residuals, i.e. local errors in interferometric phase (Goldstein et al., 1988).Then we group them by enlarging a 2D window centered on each residual. The adaptive research ends when a single rectangular area contains an equal number of positive and negative residuals and, consequently, it is cancelled (figures 2b, 3b) (Alberti et al., 1994). Baseline estimation
Our procedure for baseline estimation is based on ERS-1 and ERS-2 orbital data, in particular we used the Propagated State Vectors (PSVs), i.e. 5 satellite state vectors computed for each 18
SLC image at approximately 2.S second interval and listed in the CEOS formatted SLC header (leader file, platform position data record). The elements of the state vector are the components of the satellite position and velocity vectors with respect to a geocentric, Earth-fixed, right-handed reference frame. A fourth order polynomial is then computed for each component as a function of time. The interferometric baseline is given by the satellite separation when the antennas view the same target, that is when the target is in the range elevation planes of both antennas. We compute the corresponding orbital times at 1/PRF accuracy by using the SLC images after coarse registration. The satellite state vectors are calculated by using the polynomials and, then, the baseline components are computed. To this end, we adopt a right-handed reference frame, whose origin is coincident with one satellite position, vertical z-axis directed towards the Earth center and lateral y-axis perpendicular to the plane defined by spacecraft position and velocity vectors (figure S). In addition, we compute also the baseline in-plane components, i.e. parallel and perpendicular to the line-of-sight (table S). The x, y, z components will be used as input of the procedure to compute terrain elevation, taking account of the baseline azimuth variation. The baseline estimation procedure allows also to gain further insight into some aspects previously pointed out. In particular, table 3a puts in evidence the need of an image rotation to form the interferogram. Furthermore, figure 3a shows the fringes over an almost flat area not aligned to the ground track. As already mentioned, these aspects are consequent to a baseline variation (figure Sa). In particular, the 7 m variation of baseline y-component during the 9000 lines can be related to the variation of round-trip time. With reference to figure 4, the slant range difference .D.R.can be computed as a function of the baseline components. Assuming=
18°, R1 = 860000 m and Bx, By, B, as plotted in figure Sa, it is immediate to relate the .D.R. variations along the image, measured in wavelength, to the 2 phase rotations over a flat area at constant range bin. In figure 3a we can measure 6.7 null lines along SOO single-look azimuth pixels, corresponding to 100 lines in the plotted image. Dem generation
With reference to figure 4, the height (h) of each pixel above a local spherical Earth (radius re) is given by:
(I)
where a is the orbit semi-major axis. The side looking angle is related to the interferometric phase difference as follows:
21t 4n where is the wavelength and the slant range difference was approximated by baseline second order Ta¥lor series. Of course, a GCP must be used to solve the 2 ambiguity after phase unwrappmg. As shown by Moccia et al. (1994a), one of the most significant height error sources is the baseline uncertainty, which in our procedure is consequent to an inaccurate estimation of satellite positions. In particular, a constant error in baseline estimation causes an incorrect solution of the 2 ambiguity and, consequently, an error propagation in across-track direction., while an erroneous baseline time derivative introduces an inexistent slope in along-track direction. Several authors proposed original techniques to remove the above inaccuracies. As an example, Zebker et al.(1994) modified the interferometric phase by means of a constant ------··---- 19 and a linear term computed by using tie points. In 1993, Small et al. introduced a method for computing a phase constant, an azimuth convergence factor and two constant baseline components for precise baseline estimation by using tie points well-distributed across the scene. Our procedure is aimed at the improvement of the 2 ambiguity solution and at the refinement of the estimation of one baseline time-varying component by using the GCP heights. As shown by equation (2), the most relevant contribution is given by the B, and By inaccuracies, whereas the Bx component is less important. Furthermore, the computed Bz variations are negligible with respect to the By ones (figures 5a, b). Therefore our strategy was based on the minimization of the GCP height errors in a least square sense, by adding a linear term to By in azimuth direction and a constant term to the 2 cycles. Table 6 shows the results obtained before and after the application of the procedure. The root mean square error was computed using only the 4 GCPs not included in the regions cancelled out by the phase unwrapping procedure. In both cases the procedure worked properly, requiring limited corrections and offering results comparable to the ones presented by Zebker et al. (1994). Then we performed a comparison of the whole area (figures 6a, b), obtaining a root mean square difference of 18.7 m . This less effective result is obviously due to temporal decorrelation of extended areas with respect to point targets, as pointed out also by Small et al. 1995, who attained 2.7 mas rms error on a comparably large area when coherence was instead greater than 0.8. For the sake of clarity, in figures 6a, b 16 x 16 DEM pixels were averaged to plot a single height point. Figure 6 shows also the areas cancelled by the unwrapping procedure. The procedure for DEM production was also applied to the 01 and 02 August 1995 tandem pair. Unfortunately the average coherence coefficient was 0.50 on the whole scene and only 0.57 on the selected subset. Consequently, the phase unwrapping procedure identified a large number of residuals and cancelled large areas, making impossible an adequate DEM production. To avoid this problem an additional coherent multilook was applied (2 azimuth looks and 2 range looks). As a result, a quite large root mean square height error on the GCPs was obtained (30 m) with respect to the 1-2 and 3-4 interferometric pairs, in spite of the significant correction applied in this latter case (39 cycles). Of course also the root mean square differences between this DEM and the previous two are considerable (larger than 30 m). Differential interferometry To make possible differential interferometry it is necessary to identify a third coverage offering high coherence coefficients and satisfactory baseline with at least one of a tandem pair, provided that the tandem interferogram was satisfactory. To this end, first of all, we selected the tandem and non-tandem pairs with adequate baseline. Unfortunately we had to exclude the 1-2 pair, since it never met the baseline requirement, as demonstrated by the Interferometric Orbit Listings provided by ESA/ESRIN. Table 7 reports the considered pairs and the computed baselines. We used also an additional tandem pass (02 and 03 April 1996) to get a larger data set. At this stage we evaluated the coherence coefficients but we did not identify any satisfactory area, even considering relatively small subsets. Consequently it was not possible to apply phase unwrapping and to produce differential interferograms, apart from very small and irregular areas, where neither GCPs nor geographic references were found. For the sake of completeness, we computed the coherence coefficients over small subsets after registration performed by means of ISAR routines, obtaining comparable results. Conclusions This paper presented an end-to-end procedure for DEM generation using ERS-1/ERS-2 20 tandem data. The method was applied to a test area located in Southern Italy, where 9 CRs were deployed and three tandem pairs were available. The point target height measurement accuracy was validated on a small subset of GCPs, obtaining satisfactory rms errors of the order of 4 min the best cases (November 1995 and February 1996 tandem pairs), values comparable to the results presented by other authors in different conditions (Zebker et al., 1994). The worst value was 30 m, obtained using as input the August 1995 tandem pair. These results put in evidence the product instability consequent to unpredictable time decorrelation. With reference to extended targets, the DEMs exhibited significant variances, also in presence of high coherence. Consequently, we think that it is unlikely to obtain high resolution and accuracy DEMs (i.e. that meet map accuracy standards) by means of repeat-track interferometry. On this issue, the future Shuttle Radar Topography Mission will certainly play an outstanding role thanks to the simultaneous use of transmitting and receiving antennas. Although boom oscillations and baseline measurement must be carefully considered in this case. On the other hand, ERS tandem data allow low resolution, global scale DEMs to be accomplished, provided that adequate baseline estimation accuracy is available. This paper reported also on the impossibility to perform differential interferometry on the test area by using at least one non-tandem pair. In spite of the availability of six passes, time decorrelation caused coherence values always less than 0.40. Our future research activity will deal with high precision phase preserving processing of SAR raw data by using orbital and attitude inputs, for interferometric applications. Our main targets will be image quality and coherence improvements. To this end we submitted to the ESA Announcement of Opportunity for the Scientific Exploitation of the ERS Tandem Mission the proposal "Use of ERS-1/ERS-2 Tandem Data for Earthquakes Prediction in Tectonic Active Areas" that was accepted (ERS-1/2 experiment code AOT.1302). References Alberti G., Esposito S. and Vetrella S., 1994: The Vesuvius DEM: a test case for the TOPSAR system. Proc. of the Final Workshop "MAC-Europe 1991". Bruzzi S., Guignard J.P. and Pike T., 1982: Quality Assessment of Remote-Sensing Data: The SAR Case. ESA Journal, vol. 6, no. 3, pp. 371-281. CO.RI.S.T.A., 1995, 1996: Earthquakes prediction in tectonic active areas using space techniques. Technical Reports, Contract no. EV5V-CT94-0461 of the Commission of the European Communities . Dowman I., 1992: The geometry of SAR images for geocoding and stereo applications. Int. J Remote Sensing, vol. 13, no. 9, pp. 1609-1617. Goldstein R.M., Zebker H.A. and Werner C.L., 1988: Satellite radar interferometry: Two-dimensional phase unwrapping. Radio Science, vol. 23,no.4,pp. 713-720. JPL SIR-C Team, DLR NE-HF X-SAR Team and I-PAF X-SAR Team, 1990: Data Products and Image Quality Specifications for the SIR-C/X-SAR Mission. JPL Document D-7193. Moccia A., Esposito S. and D'Errico M. 1994a: Height Measurement accuracy of ERS-1 SAR Interferometry. EARSeL Advances in Remote Sensing, vol. 3, no. 1, pp. 94-108. Moccia A., Vetrella S. and Ponte S., 1994b: Passive and Active Calibrators Characterization by using Reference Reflectors. IEEE Trans. on Geoscience and Remote Sensing, vol. 32, no. 3, pp. 715-721. Rodriguez E.R., 1991: Maximum likelihood estimation of the interferometric phase from distributed targets. IEEE Trans. on Geoscience and Remote Sensing. Small D., Werner C.L. and Nuesch D., 1993: 21 Baseline Modeling for ERS-1 SAR Interferometry. Proc. of Int. Geoscience and Rem. Sens. Symp., Tokyo, Japan, pp. 1204-1206. Small D., Werner C.L. and Nuesch D., 1995: Geocoding and validation of ERS-1 InSAR-derived digital elevation models. EARSeL Advances in Remote Sensing, vol.4, no. 2, pp. 26-39. Topsat Working Group, 1994: Scientific Requirements of a Future Space Global Topography Mission. Werner C.L., Goldstein R.M., Rosen P. and Zebker H.A., 1992: Techniques and applications of SAR Interferometry for ERS-1: Topographic Mapping, Slope Measurement and Change Detection. Proc. of the 1s1 Workshop ERS-1 Fringes Working Group, ESA ESRIN, Frascati, pp. 11. Zebker H.A. and Villasenor J., 1992: Decorrelation in Interferometric Radar Echoes. IEEE Trans. on Geoscience and Remote Sensing, vol. 30, no. 5, pp. 950-959. Zebker H.A., Werner C.L., Rosen P. and Hensley S., 1994: Accuracy of Topographic Maps Derived from ERS-1 Interferometric Radar. IEEE Trans. on Geoscience and Remote Sensing, vol. 32, no. 4, pp. 823-836. !Comer Reflector lltest_A lltest_B llrange IIazimuth I EJlsite Ibroadening ~IPSLRll(dB) broadenmg. ~IPsLRI I (dB) [JIFormicola 110.14 11-0.81 11-1% 11-12 11-20 1116% 11-8 ll-16 u!Dragoni 110.13 11-0.74 114% 11-9 11-22 1113% 11-11 11-15 !IJILetino 110.17 11-0.75 1114% 11-6 ll-25 1115% 11-7 ll-19 l!J!Castelpizzuto ll0.16 11-0.60 112% 11-8 11-17 1110% 11-9 ll-18 LJICantalupo 110.04 11-0.75 1111% 11-14 11-25 1112% 11-10 11-22 ~]Baranello 110.10 11-0.71 1116% 11-7 ll-19 1120% 11-9 ll-19 ?uardiaregia [JI EJl-0.73 116% IEJEJl1% IEJEJ EJl~uardiaregia 0~11 % IEJEJl13% IEJEJ ~ICerreto S. 110.18 11-0.73 11-2% 11-5 11-15 1112% 11-5 11-14 I Table 1: CR radiometric analysis. Satisfactory values of test A are positive and much less than unity, andfor test_B are small and negative. Satisfactory broadenings are less than 20 %. Satisfactory !SLR and PSLR are less than -14 dB and-17 dB, respectively (JPL SIR-C team et al., 1990). ~I Satellite llOrbit no. IIDate I DIERS-I 1122662 ll14Nov 1995 I D\ERS-2 l\2989 1115Nov 1995 I DIERS-I 1124165 1127Feb 1996 I LJIERS-2 114492 1128Feb 1996 I Table 2: Processed ERS tandem data:frame 819, track 129, quadrant 2. 22 (a) 23 (b) Table 3: Sub-pixel range and azimuth shifts computed in the 1-2 (a) and 3-4 (b) interferometricpairs. !Interferometric llAverage coherence coefficient !pair l!whole scene l!selected subset 11-2 110.54 110.62 13-4 110.61 110.68 Table 4: Average coherence coefficients. jinterferometric !!Baseline components (m) I Ipair IDLIDEJLI 11-2 1174.0 ll-80.9 1139.1 1112.2 ll-89.0 I 13-4 11-9.3 11218.7 1140.6 11106.0 11195.5 I Table 5: Baseline components computed at scene center. 24 Preliminary DEM 2 ambiguity Interferometric Refined DEM rms GCP height By maximum correction pair correction (cm) rms GCP height error (m) (cycles) error (m) 11-2 1112.6 ll-1.8 11-4 114.8 I 13-4 11s.9 114.4 11-5 113.4 I Table 6: Height errors on the GCPs and applied corrections. Table 7:Selected interferometricpairs and coherence coefficients. 25 00 1 0 Figure 1. 26 27 (b) Figure 2: (a) Interferogram of the subset extractedfrom 1-2pair; (b) the white areas correspond to subset regions cancelled by thephase unwrappingprocedure after connecting the residuals. 28 (a) 29 (b) Figure 3: same asfigure 2for 3-4pair. 30 x a e Figure 4: Geometry of observation. -5 ,,...... g -10 :.: ;::~ i::i::i 70 -15 0 3000 6000 9000 0 3000 6000 9000 -75 225 ,--.... ,,...... _.,a -80 .._.,8 220 c-, :>.. i::i::i i::i::i -85~ 215 0 3000 6000 9000 0 3000 6000 9000 ' 41 ,,-....s <::« 40 N ;::1 i::i::i 38 1 39 0 3000 6000 9000 0 3000 6000 9000 azimuth position azimuth position (a) (b) Figure 5: Baseline components.for 1-2 (a) and 3-4 (b)pairs. 31 (a) Height (m, 1100 900 700 500 (b) Figure 6: DEMs of the interferometricpairs: 1-2 (a) and 3-4 (b). Triangles show the available GCPs. 33 The Effects of different Land Covers on the Accuracy of Interferometric DEM Xiaoqing Wu, Karl-Heinz Thiel, Institute of Navigation, University of Stuttgart, Stuttgart, Aloysius Wehr Germany [email protected] Abstract Different land covers on the ground have different backscatter properties, different coherences and therefore different DEM accuracies in the same interferometric scene. In this paper, four types of land covers are considered. These are corner reflectors, rural-, urban- and forest areas. Seven INSAR scenes of the same area with different baselines from 107m to about 1135mare used to compare their coherences and the accuracies of the DEMs measured by INSAR. The results were compared against an official DEM. Keywords: SAR-interferometry, ERS-1, coherence, DEM Introduction SAR-interferometry allows to compute DEMs. The accuracy of the DEMs depends on many factors. These factors include the accuracies of such parameters as the slant range, the incidental angle, the baseline and the crossing angle of the two orbits used for an interferogram, and the accuracy of the interferogram phase. The effect of these parameters on the DEM accuracy can be eliminated, if enough tie points are available. The interferogram phase, however, is a statistic variable and is completely determined by the coherence if the echo of the SAR target and the thermal noise are modeled as complex Gaussian processesl' 1.The coherence of two SAR images will be affected by temporal scene decorrelation, spatial orbit decorrelation, radar receiver noise and phase aberrations introduced during data acquisition and processing. In this paper, four different SAR-target types: corner reflectors, rural-, urban- and forest areas, their coherences and their effects on the DEM accuracy will be studied with the three day interval ERS-1 data collected in March 1992 and compared with the real DEM. These eight scene data provide different baselines from about 107m up to 1135m. These data are also used to determine empirically the largest critical baseline for interferometry. Test Site The chosen test site for the experiment is located near Bonn, Germany. Fig. 1 is the digital elevation model of this area provided by the German Measurement Agency with an accuracy of about 1m and a resolution on the ground of 5050 m2. Fig. 2 shows the ERS-1 SAR intensity image from March 17, 1992. In the test site there are two large regions of forest, large rural areas and several small towns. In order to study the DEM accuracy computed by INSAR methods, 19 comer reflectors were installed during the data acquisition of ERS-1 on March 1992. The heights of the 19 comer reflectors were measured with an accuracy better than 2cm by using GPS. Eight scene data from March 1992 are used in this study. Fig. 3 shows graphically the dates and the baselines of these data. In the following, the seven combinations shown in Fig. 3 with baselines of about 107m up to 1135m will be used to study the coherence performance of four kinds of land covers: comer reflectors, rural-, urban- and forest areas. Six combinations of them will be used to compare the accuracies of the DEMs achieved by ERS-1 INSAR. Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Sept. - 2 Oct. 1996 (ESA SP-406, Vol. II, December 1997) 34 Fig 1 DEM of test site Bonn with an accuracy of about Im and an area of 2612 knr: Fig 2 f,~RS-1SAR intensity image of test site Bonnfrom March 17, 1992. The red circles show the positions of the tie points used/or orbit parameter estimation. 35 1800 1600 -~'"-..__ _..,.,.._,.,...,. --....._,, 1400 ..~ ,~ 1200 ~E / '"' <. / QJ 1000 c 800 / <, / ~ 4 .:' lO 600 / ro -. ~II> .D 400 / 200 / 0 v Mar Mar Mar Mar Mar Mar Mar Mar Mar Mar 02 05 08 11 14 17 20 23 26 29 Fig 3 Baseline distribution of the used ERS-1 SAR data of Bonn in March 1992. Coherence All the combinations used have the same three day interval, except for the last one Pair 7. which has 6 days time interval. The weather of the Bonn region in March 1992 did not change very much. Therefore the temporal effect can be neglected for the 6 interferometric scenes with three day interval. As all these 7 interferometric scenes are processed by the same interferometric processor, the receiver noise and the phase aberrations caused by data acquisition and processing should be also nearly the same. The coherence difference for the evaluated region results mainly from the spacial orbit decorrelations for the first 6 interferograms, Due to the nearly parallel orbits, the coherence of the same area is determined eventually by the baseline. On the other hand, different land covers have different surface properties and reflectivities, which result in different coherence performances. Within a certain incidental angle, a corner reflector will have a stable and very high coherence value. On the contrary, the coherence of a forest area is variable and very bad because of the volumn reflectivity of forest and the continous swing of trees caused by wind, airflow and so on. The city area should have good coherence if there are no building-site, no traffic and no passengers on the street Nevertheless, the coherence of city areas is much better than that of forest areas though not homogeneous. If there is no vegetation, the rural areas will have homogeneous and good coherence (better than that of city areas). Fig. 4 is an example of a typical coherence map of the test site Bonn. Fig. 4 Coherence image of Bonnfrom March I 4 and I 7, I992. The 6 rectangular regions give the positions of the considered subregions offorest, cities and rural areas. 36 In order to have some quantitative ideas about the coherences of the four kinds ofland covers. we have calculated the the average correlation values and the corresponding standard deviations for the corner reflectors, two subregions of rural-, urban- and forest areas, whose locations are marked in Fig. 4. In the correlation calculation, a window size of 3(slant range) l 5(azimuth) was used. Table 1 shows the results. The average correlation values are shown graphically in Fig. 5. Table 1 Coherence comparison between different land covers calculated with a window size of 315. : mean, : standard deviation !Interferometric pair 111 112 Jl3 Jl4 Jl5 Jl6 Jl7 JBaseline (m) Jll07 Jl293 11405 IJ493 Jl657 Jl842 111135 J0.275 JI0.193 JI0.166 JI0.173 JI0.154 JI0.152 JI0.159 J0.106 110.077 IJ0.065 IJ0.069 IJ0.059 IJ0.056 JI0.057 J0.296 JI0.223 JI0.190 JI0.182 JI0.171 JI0.143 JI0.160 10.114 JI0.089 JI0.079 IJ0.074 JI0.068 JI0.052 JI0.057 10.643 ll0.437 JI0.362 JI0.332 JI0.242 JI0.205 JI0.230 10.108 JI0.113 JI0.104 JI0.104 JI0.095 JI0.089 JI0.106 J0.596 IJ0.429 IJ0.340 IJ0.315 110.243 110.218 110.224 10.111 JI0.105 110.115 JI0.103 JI0.097 IJ0.088 JI0.098 I 10.838 IJ0.667 JI0.58 JI0.525 JI0.357 JI0.222 JI0.159 I J0.048 JI0.066 JI0.073 JI0.078 JI0.088 JI0.076 JI0.055 I 10.819 JI0.636 JI0.589 JI0.504 JI0.343 JI0.218 JI0.156 I 10.049 JI0.086 JI0.064 JI0.081 110.089 JI0.078 JI0.054 I J0.95 JI0.87 JI0.81 JI0.82 JI0.83 JI0.78 JI0.78 I 10.68 JI0.47 JI0.45 JI0.35 JI0.25 JI0.19 JI0.16 I 1.0 0.9 ~---- .ih ~ --w---.._~ <1> 0.8 0.7 __.._ Forest 11.) 0.6 ,....u -+- City itl 0.5 -m- F~m:Jarea ..c:~ Q ~ Comer refl. (..) 0.4 0.3 0.2 0.1 0.0 I 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 baseline (m) Fig. 5 Coherence comparison between different land covers as afunction of baseline. 37 It is amazing to see that the coherence of the comer reflectors is not only much better than those of other kinds of land covers but also decreases much more slowly as the baseline increases. According to the decorrelation theory[2,3], the coherence decreases linearly with the increase of the baseline and will reach zero when the baseline is equal to the critical value, which is approximately equal to about 1030m for ERS-1. But the result in Fig. 5 shows that the coherence of the corner reflectors remains good even when the baseline is equal to 1l 35m. This result means that the normal decorrelation theory can not interpret the coherence performance of comer reflectors. In fact, the statistic Gaussian model of the echoes is not correct for comer reflectors. On the contrary, the coherence of the other kinds of land covers decreases almost linearly with the baseline, if the baseline is smaller than the critical baseline 1030m. According to Fig. 5, the coherences of the total scene, the forest and the rural areas will approach zero when the baseline is greater than I OOOm. This critical baseline value of about 1OOOmdetermined from the experimental data agrees very good with the theoretic critical baseline value of about 1030m. However, the coherence of urban areas seems to decrease more slowly than that of forest and the rural areas, because many points in city areas have very good coherence. Fig. 6 shows the coherence of the 7th INS AR scene with 6 day time interval and about 113 Sm baseline. In addition to the 19 corner reflectors, which have great correlation values, many points (bright points) in city areas have also high correlation values. This result shows that many objects in cities have similar coherence performance as the comer reflectors. It seems that these objects may be large buildings. Contrary to comer reflectors here the decorrelation theory can be used to interpret the decorrelation of the INSAR. And the statistic Gaussian model is also appropriate to describe the SAR echoes of city areas, rural areas, forest areas and the general land covers composed primarily of cities, forest and rural areas. If the baseline is greater than the critical baseline, many objects in cities remain well correlated, like the corner reflectors. Fig. 6 Coherence image of Bonn from March 20 and 26, 1992 with baseline of l l 35m. Furthermore it can be conduced from Fig. 5 that the coherence values of the four studied land covers for all the baselines have the following increasing order: forest areas, city areas, rural areas and corner reflectors. However, this conclusion is only valid, if the acquisition time interval is small and the baseline is smaller than the critical baseline. If the time interval and /or 38 the baseline are great, the coherence order will be different. For example, city areas will have better coherence than rural areas with vegetation for a time interval of more than several months and/or a baseline greater than 1OOOm. Accuracy of DEM The DEM in Fig. 1 has a ground resolution of 5050 m2 and an accuracy in elevation of about lm. In order to compare this DEM with the DEMs derived from INSAR, the INSAR-DEMs must have the same resolution order. Therefore, a multi-look processing with looks number of 3 l 5(slant range azimuth) was done, so that the resulting ground range resolution of the INSAR-DEMs from ERS-1 SAR images with the ground range spacing of about 20m and the azimuth spacing of about 4m will be also about 5050 m2. Six tiepoints (red circles in Fig. 2) were used to adjust the orbit parameters. The registration between the DEM in Fig. 1 and the DEM derived from IN SAR was made with the help of several significant points. Six INSAR pairs with smaller baselines were considered, because the 7th interferogram with a baseline of 1135m provides phases with so bad quality, that the DEM can not be derived from it. The means and the standard deviations of elevation errors between the INSAR-DEMs and the precise DEM shown in Fig. 1 for the four land covers considered and the total scene with an area of 2612 km2 are compiled in Table 2. The standard deviations are plotted in Fig. 7 as a function of baseline. Table 2 Accuracies of INSAR-DEMsfor corner reflectors,forest-, city-, rural- areas and the whole scene. : mean, : standard deviation !Interferometric pair 111 112 113 114 115 116 !Baseline (m) 11107 11293 11405 11493 11657 11842 ICm) 113.128 115.507 11-2.128 112.059 114.401 ll-16.285 ICm) 1113.604 117.759 116.993 116.315 1110.854 1112.688 !Cm) 110.981 114.944 11-2.005 ll-0.295 11-2.124 11-3.530 !Cm) 119.218 114.933 114.555 119.075 117.886 1111.291 !Cm) 112.901 112.392 112.041 113.537 ll-0.451 11-3.250 ICm) 113.107 112.433 111.953 112.075 112.576 112.599 !Cm) ll0.370 113.031 4.563 113.886 111.105 112.160 !Cm) 113.282 111.855 1.803 111.685 112.104 112.743 !Cm) 111.115 111.038 1.401 112.343 110.044 llI.355 ICm) 112.742 113.046 1.740 112.116 111.703 112.251 !Cm) II1.071 ll-2.171 11-2.593 llI.176 11-0.975 110.018 !Cm) 112.054 112.025 llI.450 llI.236 111.295 111.596 !Cm) 113.520 ll-0.480 11-0.79 110.59 11-0.48 110.84 !Cm) 112.40 llI.98 111.87 llI.80 llI.78 111.20 l(m) ll0.41 111.73 11-0.77 111.35 11-0.25 11-0.74 I !Cm) 117.05 114.81 114.29 114.53 114.54 116.82 I According to Table 2 and Fig.7, it is obvious that the forest areas have the worst elevation accuracy among the four kinds of land covers because of the volume reflectivity effect of forest, the worst coherence resulting mainly from the continuous movement of the trees and the phase unwrapping errors, which arise often in forest areas. City- and rural areas have similar elevation uncertainties. Although the accuracy of rural areas is 39 normally a little bit better than that of city areas. Theoretically, the elevation errors of city areas should be much greater than those of rural areas, because unlike the INSAR-DEMs, the reference DEM does not include the building heights. We attribute the small elevation errors to the sparse, small and lower buildings of the small towns. The better accuracy of the rural areas results from the better coherence. The 19 corner reflectors, which were distributed in east-west direction with a length of about 13 km, exhibit stable and very good elevation accuracy. Extrapolating the plot for corner reflectors in Fig. 7 to longer baselines than 1OOOm,an increase in elevation accuracy is still possible. The elevation accuracy of the total scene is determined by several factors. First, the used reference DEM do not have probably 1 m elevation precision. This DEM model represents only the topographical elevation. The elevation of buildings and trees is ignored. Besides, the elevations in some regions are the results of the interpolation, which are obviously not the right ones. These errors in the reference DEM will increase errors in INSAR-DEMs for large area regions in the comparison. This is one reason why the elevation accuracy of the whole scene is worse than that of city- and rural- areas. Second, registration between the reference DEM and the INSAR-DEMs is another reason for the greater deviations of the total scene. Besides, some ,,moving fields" will also produce errors oflNSAR-DEMs in the whole scene. Phase unwrapping errors in some local forest regions is another error contribution to the uncertainty of the whole scene. If the errors resulting from non-INSAR factors can be removed, the accuracy achieved through INSAR would be in the same order of that for subregions of rural- and urban- areas, i.e., 23m. From Fig. 7, we can also determine the approriate baselines for ERS-1 INSAR DEM generation. They amount about 1OOmto 800m. Of course, the approriate baselines will be smaller, if the the temporal decorrelation is great. On the contrary, if the acquisition time interval is very small or even zero (airplane INSAR), the appropriate baselines may be greater. 40 __..__ Forest 14 ---+--- City 13 --EB- Ri.aal area ~ Corener refl. 12 11 10 9 '8' '-' q 0 8 "+l 0$ ·~ 7 'O"' -e ~ "t:l 6 ....,s l'.l:l 5 4 3 2 0 100 200 300 400 500 600 700 800 900 1000 baseline (m) Fig 7 Standard deviations betwen INSAR-DEMs and theprecise DEM/or corner reflectors, two subregions offorest-, city-, rural- areas and the whole scene. Conclusions From the interferometric results of test site Bonn, the following conclusions can be drawn: 1. Corner reflectors due to their construction are special radar targets and therefore behave different from forest, cities and rural fields. They have the best coherence and DEM accuracy. The coherence performance and the accuracy of DEMs derived by INSAR can be interpreted by the decorrelation theory and the echoes can be modeled with the statistic Gaussian Model, if the SAR targets are mainly forest, cities and rural fileds. The decorrelation theory failed, if the SAR targets are corner reflectors. 2. For small acquisition time interval and baselines smaller than the critical baseline, rural areas have better coherence than cities. Forest has always the worst coherence among the three land covers. 3. The critical baseline for ERS-1 INSAR is about 1OOOm.Baselines more than 1OOOmwill make the coherence of general land covers nearly uncorrelated. The appropriate baselines for ERS-1 INSAR DEM generation are 100 to 800m. In the most approriate situation, an elevation accuracy of 2 3 m can be expected by INSAR. 41 Acknowledgement The authors would like to thank the European Space Agency, who have provided the ERS-1 SLC data. References [1] Just, D. and Bamler R., Phase statistics of interferograms with application to synthetic aperture radar, Applied Optics, Vol. 33, No. 20, July 1994, pp.4361-4368. [2] Li, F., Goldstein, R.M., Studies of multibaseline spacebome interferometric synthetic aperture radars, IEEE Trans. Geosci. Remote Sens., vol. 28, no. 4, Jan. 1990, pp. 88-97. [3] Zebker, H.A., Villasenor, J., Decorrelation in interferometry radar echoes, IEEE Trans. Geosci. Remote Sens., vol. 30, Sept. 1992, pp. 950-959. 43 Validation of Height Models from ERS Interferometry D. Small and D. Remote Sensing Laboratories (RSL) Nuesch University of Zurich, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland [email protected] http.z/www.geo.unizh.ch/e-daves/ Abstract Among the many applications of ERS interferometry, the production of digital elevation models is often preeminent. The "end product" of an InSAR processing chain is typically a geocoded InSAR elevation model. For the generation of a height model, many processing steps must followthe generation of the interferogram and coherence image. Although other applications (e.g. investigation of coherence) can be studied using only the interferogram and coherence information (in slant range) available at an early processing stage, generation of height models requires phase unwrapping, determination of absolute phase, refinement of geometry, phase to height conversion, as well as forward geocoding into map coordinates. The processing steps (as implemented at RSL) are reviewed, with special attention devoted to the transformation from the radar system geometry into map coordinates. Results are shown to demonstrate expected accuracies using ERS data. The ERS height models are validated through comparison with reference models obtained through independent sources. Keywords - SAR-Interferometry, ERS Tandem, JnSAR DEM, Geocoding, Validation 1. Introduction The initial processing steps within the InSAR processing chain have become increasingly better documented and well understood in the years since the widespread availability of ERS-1 data first opened up the field of SAR interferometry. Image coregistration (Small D., et. al., l 993a), interferogram generation (Gatelli F., et. al., 1994), and baseline estimation have matured while phase unwrapping is undergoing refinement (Davidson G. and Bamler R., 1996), (Ghiglia D.C. and Romero L.A., 1994). The steps following phase unwrapping are now the subject of increasing research: absolute phase determination, phase to height conversion, forward geocoding, and height model validation (Small D., et. al., l 995b). This paper summarizes some of the latest results at the University of Zurich in the field of height model validation using ERS-1 /2 tandem data. Obstacles to successful height model generation are investigated: coherence maintenance, suboptimal phase unwrapping. In order to test the accuracy of ERS repeat-pass interferometry, test sites with high quality reference DEMs have been chosen. The InSAR processing algorithm and validation methodologies are briefly described, and applied to data from the test sites. Results are presented and evaluated, and conclusions are drawn. 2. InSAR Processing After SLC image coregistration (Small D., et. al., 1993a), interferogram generation, and flattening (Small D., et. al., l 995b), the image geometry must be refined to ensure successful Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, lunch, 30 Sept. - 2 Oct. 1996 (ESA SP-406.Vol. II. December 1997) 44 transfer from slant range geometry into the chosen map reference system. The accuracy of the precise ERS orbits delivered by the D-PAF is approximately 30cm along and cross-track with a slightly higher radial accuracy (8cm) (Massmann F.-H., 1995). However, successful phase to height conversion requires better accuracy (Li F.K. and Goldstein R.M., 1990). The geometry must therefore be refined using tiepoints. Two refinements are performed using two types oftiepoints. Height tiepoints are measured in areas of uniform unwrapped phase, while position tiepoints (Northing, Easting) are measured in parts of the image that can be localized with high geometric accuracy (e.g. bridges, road crossings). Height tiepoints distributed well across range and azimuth are used to refine the baseline model (Small D., et. aL l993a). An iterative non-linear least squares fit algorithm is used to estimate a set of parameters including the phase constant, the cross-track baseline component and its trend along the azimuth dimension. The radial orbit component is usually left at its nominal value, as it is known more accurately. Before such refinement (open loop) height estimates have errors on the order of kilometres. Once the phase constant (only) has been localized, height errors are typically reduced by an order of magnitude. Further refinement of the cross-track baseline component (and trend in azimuth) reduces the height estimation error to still lower values. Position tiepoints are used to refine the SAR image acquisition geometry (as during conventional GTC production (terrain-geocoding)). As with the baseline geometry above, several approaches are also possible here. In conventional ERS GTC production, the azimuth starting point, pixel spacing, near range value, and range pixel spacing are all refined based on tiepoints well distributed across the scene. One may also use the tiepoints to refine the orbit itself. Given a polynomial description of the orbit, one typically refines the constant and linear terms, leaving higher order terms at their nominal values. Based on the refined baseline and the imaging geometries, the interferometric height of each point in the image with a valid unwrapped phase may then be calculated (Small D. et. al., 1996). The interferometric height map (and possibly also other by-products) are finally geocoded to a map projection through solution of range, Doppler, and ellipsoid equations: · Slant Range Sphere "') ") ,. "') (Sx - Px )- + (Sy - P,s + (Sz - Pz )- = R; ·Doppler Cone 2 S-P Xx IS-Pl. (Vp- Vs) = f ref ·Height above oblate ellipsoid ., ., ., (P x - Xo)- ( P ~·- Yo)- (P r: Zo)------=.,-+ - ., + ., = 1 (a+ h)- (a+ h)- (b+ h)- The third equation above is the simplified version appropriate for a 3-pararneter datum shift transformation (translation only: (xo, Yo· Zo)). The more general 7-parameter datum shift transformation is somewhat more complicated, as it adds a scale and three rotation parameters. Backward geocoding is employed in conventional ERS GTC production (Meier E., et. al., 45 1993). Height model positions are transformed into WGS84 Cartesian coordinates (Frei U., et. al.. 1993), providing a reference frame common with the orbit data. Given precise knowledge of the satellite orbit and each DEM position, one may solve for the satellite position that satisfies the Doppler equation above. The associated range value is then easily calculated as the distance from the satellite to the DEM position, and the (range, azimuth) position within the slant range image is now localized. Resampling into map coordinates follows. Forward geocoding presumes knowledge of the local height values in a slant range matrix (see the ellipsoid equation above). An initial starting point p Xo' P}'o' Pzo within the scene is selected and then an iterative method (e.g. Newton-Raphson) is used to solve the set of simultaneous non-linear equations. That solution then serves as the starting point for the solution of the equations defining the next point. A set of irregularly-gridded points P x· P }" P z in WGS84 geometry are the result. As the points do not coincide with a regular grid in any map geometry, a regridding step follows the solution of the equations. Both backward and forward geocoding methodologies have been implemented at RSL that directly support full 7-parameter datum shift transformations. Both implementations geocode an ERS quarter scene in less than an hour on a modern UNIX workstation. Provided that the reference and interferometric height models are of comparable accuracies, the quality of the geocoding from the two methods is nearly identical (Small D. et. al., 1996). The forward method is necessary when one has no reference elevation model, i.e. in the DEM generation application. Other applications are more satisfactorily served by the backward method, as one can generate a DEM of an area once with the forward method, and then use the backward method later successively to terrain geocode all types of slant-range information (e.g. backscatter values, coherence, differential phase). Using the backward method, one is not restricted to height information from a single InSAR pair: the reference elevation model used could be an amalgam formed from several acquisitions, even a mosaic formed through combination of DEMs from different InSAR sensors. 3. Validation The accuracy achievable using the ERS InSAR system for height model generation was estimated using three different approaches: ·DEM Flattening using synthetic interferogram · Backward geocoding of slant-range height model ·Forward geocoding of slant-range height model The advantages and disadvantages of each method break down as follows. Comparison via a synthetic interferogram avoids the need for phase unwrapping, but direct juxtaposition is not possible in the final user map geometry. Given a high quality reference elevation model, backward geocoding enables such comparison, but it is not a "blind" test of the end-to-end InSAR elevation model generation, as the quality of the geocoding is non-representative. It does however supply quick feedback on coherence, local incidence angle, height model error interdependencies. It also allows use of a refined height model assembled from heterogeneous sources. Since it uses the reference height values during geocoding, it cannot be part of a "blind" test of InSAR-generated DEMs. Only a "blind" forward geocoding from slant range into map geometry, with reference heights introduced exclusively at the final height comparison, enables a true end-to-end validation of the InSAR height model generation process. Forward geolocation of areas with poor phase information will be erroneous, and may not register well with ground truth information. 46 The relative success of the forward geocoding can be tested through overlay of the ERS images in map geometry, geocoded using the forward and backward methodologies. Areas where features do not overlap indicate disagreement in the height models (resulting in planimetric shifts during geocoding). Validation of the height models is carried out principally through calculation of the difference between the forward geocoded height model and the reference model. RMS difference values together with histograms of the distribution of the height differences provide quality measures. Note that foreshortened and layover areas undergo extended interpolation when transformed into map geometry - the local slope (and also coherence) therefore influences the locally achievable height accuracy. Combination of at least one ascending and descending pair can be used to mitigate this influence. Consistency checks can be performed between several forward-geocoded height models to increase the height model accuracy. 4. Results 4.1 Test Site Bonn 4.1.1 Introduction The Bonn test site provides an example of a vegetated scene with moderate topography. A variety of baselines and repeat pass intervals are available, as the data was acquired during the first ERS-1 ice phase. Figure 1 shows the location of the Bonn test site. The Rhine river crosses through the northeast comer of the scene. One finds arable land, mixed deciduous and coniferous forests, open-pit mines, and densely populated areas within the scene. Two interferograms: 14.03.92 I 17.03.92 and 14.03.92 I 29.03.92 provide a good combination of baseline variety and repeat-pass temporal interval. Precise orbit products were ordered from the D-PAF for use in geocoding and baseline estimation. A reference DEM (originally produced by digitizing map sheet contours) was provided by the GEOS group at the D-PAF for use in geocoding research. Figure 1: Bonn, Germany - Geographic Location - ERS-1 SLC Quarter Scene - March 14, 1992 47 4.1.2 Height Model Validation DEM flattening was used to investigate the potential accuracy of the phase values and the scope for systematic accuracy achievable given a correctly unwrapped interferogram. .. Figure 2: DEM Flattened ERS-1 lnterferogram - Bonn, Germany 14.03.92 - 29.03.92 Figure 2 shows a 15-day repeat DEM-flattened interferogram. With the exception of areas where the reference elevation model was no longer up-to-date (some mining pits, gravel pits), topographic fringes are no longer visible. The fringes that remain are likely of atmospheric origin. Note that this interferogram has a time interval of 15 days - the relatively long time interval increases its susceptibility to such differential effects. The 14-17.03.92 interferogram has a shorter 3-day repeat and a much more sensitive baseline. Regional phase anomalies such as those seen in Figure 2 are not present. The interferogram was filtered, the phase was unwrapped, and the baseline geometry refined. Finally, the digital elevation model was calculated and geocoded. An area within the scene "Bonn-Norvenich" with terrain variation of approximately 80m, and relatively high coherence values was selected for more detailed study. Geocoded interferometric heights were compared point-by-point with the reference elevations, with an elevation-difference map being produced (Small D., et. al., 1995b). The difference map showed that the interferometric height estimates agree remarkably well with those from the reference DEM. A quantitative representation of the height difference distribution is presented in Figure 3. The histogram shows the frequency of each height difference value within the area. Note that all pixels within the region (of all coherences) were used to compute the histogram, and that no significant systematic height bias is visible. A global RMS error of 2.7m was calculated over the 12x13km area for the interferogram calculated from the March 14/17 1992 pair. Note that the reference DEM was quantized to Im intervals - the high accuracy was achievable due to the slowly rolling topography together with the large baseline (>400m). 48 Results were not so satisfying for the March 14/29 1992 pair. The longer time interval both reduced mean coherence over the scene and introduced more coherent phase shifts, distorting the height estimates, and resulting in an RMS error of about 12m over the same region. The interferometric height model from the Bonn scene was geocoded using both the forward and backward methodologies. Comparison of the geocoded height models showed that planimetric geocoding differences occur where the interferometric height estimates differ (Small D. et. aL 1996). Examples are gravel pits (where the reference height model was no longer current), and forested areas (where the interferometric height estimate was noisy). Jl01111-No••.•,~Aidi J.!Dl.92/ !7.0J.92Ki""griai 2.0°105.--~~~~-.-~~~~...-~~~~ .•...•.~~~~ ...•...• l&.n.bia- -0.11 (m.) ~ !5·105 lll!. •••• - !bl>[m.) ~ 1.0·105 ~ 5.0•10.i ~ 0<-~~~~~"""''--~~~~~-=-~~~~~~-' .zo -10 0 10 t~l DEM DJ&r.ac-e [.m.] Figure 3: Histogram of height differences between interferometrically calculated heights and reference elevations for all pixels within Bonn-Norvenich region 14-17.03.1992 4.2 Test Site Bern 4.2.1 Introduction Figure 4: Bern, Switzerland - Geographic Location - ERS SLC Quarter Scenes - Nov. 1991, Oct. 1995, Nov. 1995 Bern was chosen as a test site to investigate the influence of a variety of slopes as well as vegetation states on the DEM generation process. Within the tandem data inventory the Oct. 95 and Nov. 95 pairs were selected as having the best combination of simultaneously high height sensitivity and coherence (Stebler 0., et. al., 1996). The more extreme topography and numerous hilly forested areas make phase unwrapping more challenging in comparison to the Bonn sc~ne. Larger height errors are the result. Many areas are rendered inaccessible to phase unwrapping. 49 4.2.2 Height Model Validation DEM flattening was performed on the interferograms to investigate the consistency of their phase behaviours, and the potential accuracy achievable (assuming an optimal phase unwrapping algorithm). No noticeable global phase trends remained after this step. Aside from lakes, rivers, and layover regions, most areas showed remarkably flat phase values. An exception was found in forest stands. Figure 5 shows an extract from the Nov. 95 DEM-flattened interferogramjuxtaposed with a scanned extract from a Swiss cartographic map, courtesy of the Swiss Federal Office of Topography. Note how the forest stands are clearly distinguishable in the phase image. The behaviour is consistent for all three interferograms studied (Nov. 91, Oct. 95, and Nov. 95). The phase difference can be explained by the fact that the reference height model (originally derived through aerial stereo photography) is based on ground height measurements, while the C-band interferometric height estimates are based upon radar returns echoed from the tree canopy (Small D.. et. al., 1995a). Figure 5: Differential phase from Forest stands - (a) Extract from DEMflattened interfer black=O, white=Zt; - Bern, Nov. 1995 - (b) Extract of cartographic data from National Switzerland 1:25'000, courtesy: ©Swiss Federal Office a/Topography 9.96 Backward geocoding may be used to validate interferometric height models. However the forward geocoding method provides a more authentic "blind" test of end-to-end system accuracy. A region-growing phase unwrapping algorithm was used to recover absolute phase difference values. Inspection of height difference maps calculated from the resulting single-interferogram height models showed that phase unwrapping errors had corrupted some calculated height values. Although more robust phase unwrapping methods (Davidson G. and 50 Bamler R., 1996) might have yield improved results, in the presence of variable temporal decorrelation it is inadvisable to generally assume a 100% successful phase unwrapping. We therefore employ a scheme to minimize the impact of such errors. The accuracy of the final forward geocoded InSAR height model product can be improved through combination of the results from individual interferograms. Areas with inconsistent height values are marked as "no data" areas,joining the areas that weren't within reach of the phase unwrapping algorithm. The distributions of the single-scene forward-geocoded height differences are shown in Figure Q. Note that phase unwrapping errors produce "islands" of incorrect height values, corrupting the height model. Consistency checking substantially reduces this error source, resulting in the height difference distributions seen in Figure 7. Figure 7(a) shows the height difference histogram for all unwrapped areas. If one restricts the area under study to parts of the image with a local incidence angle in the range from 14-30°, the distribution of height differences improves slightly - see Figure 7(b). >.lo:nb.iD - 1.o.olO[mj XI.IS.cs,..• J!.IO[mj -'200 -!00 0 100 200 JOO He.$.t. DiH=.ac-e [m) -'200 -!00 0 100 200 JOO lk.$.t. DiH=.ac-e [m) Figure 6: Histogram of height differences within all unwrapped areas - Bern, Switzerland- (a) Oct. 1995, (b) Nov.1995 -JOO -WO -!00 0 !00 Jkjgil: Difl>moace [m) '.?5·!05 ~'.?.o·w5r------~--=-----:1.1cn m... 1.•nl•l ~ •••••-11.101-1 ~ u-io5l 0 1,0•!05 ~ 5.0•!0.i 0 --:":""""-~- .ro -20 0 Jkjgbc D~.ac-e [m) Figure 7:Histogram of height differences - Bern, Switzerland - Combination DEM forward geocodedfrom Oct. 1995 and Nov. 1995 - (a) All areas, (b) Areas where 14° < 0 < 30° 51 Figure 8: Height difference (mean± standard deviation) vs. local incidence angle - Bern.forward geocoded DHM combination, Oct. 1995 + Nov. 1995 Forward geocooed lnSAR.heigbr model withDHM25 f0 :z 580 600 620 640 Ea.sting Figure 9: Forward geocoded ERS-1 intensity and height model on 300 m colour cycle with reference DHM25 heights showing where no InSAR height model is available - Bern, Switzerland- Combination Oct. 1995 +Nov. 1995 height model - reference digital height model DHM25 courtesy: (f1 Swiss Federal Office of Topography 9.96 Figure 8 shows the dependence of the height accuracy achieved in the final end-product (forward geocoded) height model on local incidence angle. For local incidence angles ranging from 0 to 45°, the figure shows the mean height error, together with its standard deviation at each LIA. Improvement is significant in comparison to results from a single interferogram. The final height accuracies are respectable: the remaining problem is their lack of omnipresence. Efforts to increase the robustness of the phase unwrapping step are aimed at remedying this problem. 52 The combined forward-geocoded InSAR height model, with the reference DHM25 visible where no consistent InSAR height estimate was available, is shown in Figure 9. The juxtaposition demonstrates that the InSAR height estimates (where they are available) agree with the reference, but that many regions remain off limits to successful phase unwrapping. Given additional data sets that met the criterion of a large enough baseline (for good height sensitivity) further refinement would be possible (Massonnet D., et. al., 1996). The addition of at least one ascending tandem pair (unfortunately unavailable for this study) would increase the accuracy of the end result by increasing the local resolution in areas foreshortened in the descending Oct. 1995 I Nov. 1995 geometries. 5. Conclusions Areal validation of height maps generated by repeat-pass ERS InSAR provides confidence in the InSAR technique. For a 12x13 km area near Bonn, Germany, RMS accuracies of 2.7 m were achieved, with no observable systematic biases over the 40x50 km quarter scene. Height accuracy increases with longer baselines, and decreases drastically where slopes become extreme, as well as in areas oflow coherence (e.g. forest). Spectral-shift filtering dramatically decreases phase variance and is of critical importance for the large baselines that are optimal for the extraction of topography. Phase unwrapping errors must be either manually corrected, or mitigated through consistency checks with data from other interferograms. Improvement of the height estimate through combination of multiple tandem ERS-112pairs is advisable - combination of ascending/descending pairs is necessary to offer a consistent ground resolution across the scene. Experience with our Bern scene observing the seasonal variation of I-day repeat coherence suggests that the winter season appears to be optimum. The minimum coherence useful for mapping is dependent on the slopes within the scene, the required accuracy, and the scene's baseline. However, as a rule of thumb, requiring that the mean coherence be above 0.5 can be a useful discriminator. The optimum baseline for mapping purposes depends on the slopes that can be expected within the scene. Given a relatively flat scene, with gently rolling topography, a baseline of 300-400m provides the best height accuracy while not excessively sacrificing spatial resolution during spectral shift filtering. Scenes with more significant slopes must make do with smaller baselines and less accuracy, lest they fall prey to unsuccessful phase unwrapping. Consistency checks using multiple interferograms can be used to improve accuracy. Given the right conditions, repeat-pass ERS interferometry can produce height models with respectable accuracy. The weakness of ERS-1 InSAR height derivation lies in hilly forested areas, where low coherences combine with topography to render height estimation problematic. However, new techniques continue to emerge that work to increase the robustness of InSAR-based height estimation in the face of all known limitations. 6. Acknowledgements This work was supported by subcontract 11318/95/1-HGEwith Matra-Cap Systemes, France within a larger contract to ESA-ESRIN. Development of the InSAR processing chain was supported by a contract with ESA under the supervision of the German Remote Sensing Data Centre (DLR-DFD). ERS SAR data and precise orbit ephemeris information were provided courtesy of ESA-ESRIN. The Swiss reference digital elevation model was provided courtesy of the Swiss Federal Office of Topography. The reference elevation model for the Bonn area was provided 53 courtesy of the D-PAF, for use in SAR geocoding research. 7. References [1] Davidson G. and Bamler R., 1996A Multiresolution Approach to Improve Phase Unwrapping, Proc. oflEEE-IGARSS'96, Lincoln, USA, pp. 2050-2053. [2] Frei U., et. al., 1993 Cartographic Reference Systems, chapter in SAR Geocoding: Data and Systems, ed. G. Schreier, Herbert Wichmann Verlag GmbH. [3] Gatelli F., et. al., 1994 The WavenumberShift in SAR Interferometry, IEEE Trans. on Geoscience and Remote Sensing, Volume 32, Number 4, pp. 855-865. [4] Ghiglia D.C. and Romero L.A., 1994 Robust Two-Dimensional Weighted and Unweighted Phase Unwrapping that uses Fourier Transforms and Iterative Methods, J. Opt. Soc. Am., Vol.All,pp.107-117. [5] Li F.K. and Goldstein R.M., 1990Studies of Multibaseline Spaceborne Interferometric Synthetic Aperture Radars, IEEE Trans. on Geoscience and Remote Sensing, Volume 28, Number 1, pp. 88-97. [6] Massmann F.-H., 1995Informationfor ERS PRLIPRC Users, GeoForschungsZentrum Potsdam Technical Note. [7] Massonnet D., et. al., 1996 Reduction of the Needfor Phase Unwrapping in SAR Interferometry, Volume 34, Number 2, pp. 489-497. [8] Massonnet D., 1995 Limitations to SAR interferometry due to instrument, climate, or target geometry instabilities, EARSeL Advances in Remote Sensing, Volume 4, Number 2, pp. 19-25. [9] Meier E., et. al., 1993Precise Terrain Corrected Geocoded Images, chapter in SAR Geocoding: Data and Systems, ed. G. Schreier, Herbert Wichmann Verlag GmbH. [1O] Small D. et. al., 1996A Comparison of Phase to Height Conversion Methodsfor SAR Interferometry, Proc. oflEEE-IGARSS'96, Lincoln, Nebraska, USA, pp. 342-344. [11] Small D., et. al., 1995a Combination of Ascending I Descending ERS-1 lnSAR Datafor Calibration and Validation, Proc. oflEEE-IGARSS'95, pp. 553-555. [12] Small D., et. al., 1995b Geocoding and Validation of ERS-1 lnSAR-derived Digital Elevation Models, EARSeL Advances in Remote Sensing, Volume 4, Number 2, pp. 26-39, pp. 1-11. [13] Small D., et. al., 1994Applications ofGeocoded ERS-1 lnSAR-derived Digital Terrain Information, Proc. of CEOS SAR Calibration Workshop, Ann Arbor, Michigan, USA, pp. 184-190. [14] Small D., et. al., l993a Baseline Modelling/or ERS-1 SAR Interferometry, Proc. of IEEE-IGARSS'93, Tokyo, Japan, pp. 1204-1206. [15] Small D., et. al., 1993b Registration of ERS-1 SLC Productsfor Interferometry, Proc. of Fourth GEOSAR Workshop, Loipersdorf, Austria, pp. 63-66. [16] Stebler 0., et. al., 1996Analysis of ERS-SAR Tandem Time-Series Using Coherence and Backscattering Coefficient, Proc. of ESA-FRINGE'96, University of Zurich-Irchel, Zurich, Switzerland, 55 Initial testing of ERS Tandem data quality for InSAR applications; Examples from Taiwan, Madagascar, Zaire, Ivory Coast, Mali and Greenland Gaute Aa ESA/ESRIN/ RS/E Data Utilisation Section Solaas Fabio Gatelli Politecnico di Milano, Dipartimanto di Elettronica e lnformazione, Under contract to ESA/ESRIN RS/P ERS Section Abstract A preliminary investigation has been performed in which interferometric products were generated from selected ERS tandem data. The data covered a number of areas throughout the world and represented a variety of land cover types ranging from glaciers to tropical forests, savannah and deserts. A prototype interferometric quick-look processor was used in an ad-hoc operation where relatively large amounts of data were processed in a short time period (over 50 standard frames of ERS data). In addition to demonstrating the potential of quick-look InSAR processing we could draw the following preliminary conclusions: o Tandem InSAR coherence over dense rain-forest is low and results in highly fragmented fringe-areas. This can be exploited in several applications concerned with land use monitoring. o Over mountain rainforest the fragmentation is even more pronounced due to the effect of foreshortening and layover on SAR imagery. This can partly be remedied by combining interferogrammes from ascending and descending passes, but remains a serious obstacle for operational use of ERS InSAR over mountainous terrain. o Maximum NDVI, as derived from AVHRR over a two month period (to eliminate cloud and haze signatures), can be an important tool for establishing the expectancy of high coherence areas. This can help the user in identifying the areas where a specific application has highest probability of success and can be a major cost saver if e.g. large scale use of ERS InSAR in mapping is going to be carried out. 0 Tandem InSAR has high potential in arctic and Antarctic areas due to the high coherence obtained over dry snow and ice and sparse vegetation. Since there are large amounts of data acquired over these areas the user applications should be assured in most instances. The results presented here are available from ESRIN under the reference: RS/ED96.D002 Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zuticn, 30 Sept. - 2 Oct. 1996 (ESA SP-406. Vol. II, December 1997) 57 Interferometry for Forest Studies Nicolas FLOURY Centre d'Etudes Spatiales de la Biosphere Thuy LE TOAN 18 avenue Edouard Belin, bpi 2801 Jean-Claude SOUYRIS 31401 Toulouse Cedex 4, France [email protected] Kuldip SINGH Centre for Remote Imaging, Sensing and Processing Nicolas STUSSI National University of Singapore Lower Kent Ridge Road, Singapore 119260 Chih Chien HSU Department of Electrical Engineering and Computer Science Jin Au KONG Research Laboratory of Electronics Massachusets Institute of Technology Cambridge, MA02139, USA Abstract This paper presents the use of ERS-1 repeat-pass interferometric data for forest environment studies. The overall objective is to assess the use of interferometric information - degree of coherence and phase difference - for forest I non-forest mapping and for retrieving forest parameters. Interferometric data acquired over a reference test site of temperate forest (Landes, France) and over a tropical forest site (South Sumatra) have been analysed. Coherence is shown to be a good discriminator between forest stands and bare soil surfaces. However, the conditions for optimal use of coherence depend on the forest and environment characteristics. In the specific case of the Landes forest (flat and homogeneous forest stands), phase difference statistics are shown to be linked to forest stands heights. The knowledge of the penetration depth of the wave into the forest canopy, obtained from theoretical modeling, is shown to be necessary to get a good estimate of stands height from interferometric phase difference. Keywords: Interferometry, Forest, Mapping, Forest Biomass, Forest Height Introduction Several studies have shown the potentialities of repeat-pass interferometry for the extraction of DEMs over temporally stable terrains or for the detection of small terrain movements. The interest in interferometry as a tool to study forests is more recent [1]-[2]. The objective of this paper is 1) to analyse the relations between interferometric data and forest parameters over a well known site, 2) to interpret the underlined physical phenomena using a theoretical model, and 3) to discuss on the influence of environmental and temporal conditions on the results. Test-sites and Datasets The reference test site is the Landes forest, located in South-Western France. It is the largest plantation forest in France, constituting nearly one million hectares on flat topography. This artificial forest is almost totally formed of maritime pine (pinus pinaster) and is managed in a consistent fashion, which ensures the canopy to be homogeneous. The ground data consist in a biomass map which provides information about location and age of more than 50 stands of maritime pine, covering about 20 ha each. The age of these stands ranges between 2 and 50 years, while the corresponding biomass is between 5 and 150 tons I ha. Clear Proceedmgs of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Sept. - 2 Oct. 1996 (ESA SP-406. Vol. II, December 1997) 58 cuts and agricultural fields are also present in the area. Other parameters such as tree densities and dendrometric information are also available. Interferometric data consists of 3 interferometric couples based on ERS-1 SLC images acquired in 1991 and processed by the CNES. The characteristics of the interferometric data are summarized in Table1. An alternative test-site is chosen in a tropical environment, at Selatan, South Sumatra, Indonesia. This site comprises a mix of primary forest, plantations and deforested areas. Concurrent data consists of a SPOT XS image of the area acquired within one month of the ERS images. Two tandem (ERS-l/ERS-2) and two 35 days (ERS-1 and ERS-2) interferograms are available over the area. Interferometric data has been processed by CRISP. Date of acquisition II Time interval II Baseline Interferometric Set II (days) given by ESA (m) Landes A 15 oct 91/ 20 nov 91 35 I 10 I Landes B 20 nov 91/02 dee 91 12 142 I Landes C 15oct 91 I 02 dee 91 47 130 I Sumatra D I 12 apr 96 / 13 apr 96 1 II - Sumatra E II 17 may 96 / 18 may 96 1 135 I Sumatra F II 12 apr 96 I 17 may 96 35 II - I Sumatra G II 13 apr 96 / 18 may 96 35 II - I Table 1 : Characteristics of available interferometric data. Theoretical modeling An accurate interpretation of the phase information contained in the interferometric data would need coherent electromagnetic modeling. Until now, the development of coherent models for such natural medium is very limited. Most of the existing models for radar backscatter of forests are based on the Radiative Transfer Theory which cannot be used to interpret the phase information. However, some insights on the dependency of interferometric data to forest parameters can be derived from the these theoretical models. This section summarizes the four-layer Radiative Transfer (RT) model developed at MIT for the modeling of pine forest [7] which is used in this study. The four layers (Fig. 1) include a crown layer, a trunk layer, an understory layer, and a ground interface. Forest canopy and ground surface characteristics are provided by experimental measurements. The Kirchhoff approximation is used to compute the scattering from the ground modeled as a random rough surface. The trunk is modeled as a tilted circular cylinder, branches and needles are modeled as circular cylinders where finite cylinder approximation is applied. The scattering properties of structured pine trees are taken into account in the model by incorporating the branching model [8] into the phase matrix of the RT equation: the crown is modelled as a 4-scale cluster constituted of trunk, primary branches, secondary branches and needles. The vector radiative transfer equation for the specific intensity in each scattering region is of the form - dI(B ¢ z) == - f -== - cosB d~' = -1ce(B,¢,z). I(B,¢,z)+ ac» P(B,¢,B',¢').I(B',¢ 4n- where the Stokes vector contains information regarding field intensity and phase relation of the two orthogonal polarisations and is defined as : 59 ii ~ ..rl \\ 1h v-1, 1 )1 l I r • J = I t, = !__ \I Ev IJ) (2) u v .' I , 'JR l~ *'· VJ - e\l!..-l,Eh)I 2 Im (E "ii'*) \ V""""'h I In (2), the subscripts h and v represent the horizontal and vertical polarisations, respectively. The bracket < > denotes ensemble averaging over the size and orientation distributions of scatterers '11--J µo !8 o . . . . . k= and is the free space impedance. The extinction matnx e represents the attenuation due to both scattering and absorption, and can be obtained through the optical theorem in terms of forward scattering functions. The phase matrix P({}'¢,{! '¢J ) characterizes the scattering of the Stokes vector from (9',cp') direction into (8,cp) direction. The phase matrix can be formulated in terms of the scattering functions of the randomly distributed discrete scatterers. 60 - - / / '\. I I I I ' Figure I: Radiative Transfer Approach Complex Coherence The complex degree of coherence of the two co-registered complex image values s1 and s2, is given by: where p is the degree of coherence and ~ the phase difference between the two signals. 61 Degree of Coherence Landes Test-Site Fig. 2 presents the variations of the coherence versus stand age for the 3 interferometric sets. High temporal coherence is obtained for clear cuts and open fields whereas it decreases with stand age. o.e 0.8 3Sd¥ acquisition inteival •• (15Oct -20 Iov 91) co ,,!. f: 0.6 • ;!! 0 0 0 0.5 •• 0.4 G)e a 0.3 0.2 0.1 0 5 10 15 20 25 30 35 40 45 Stand ~e bears) A O.~ 0.8 12 d~s acquisition i'nleival (20 Iov - 02Ilfc 91) ••o 0 7 c t: '1• .. :!L 0.6 0 c 0 o.s T ~ •• •· e 0.4 Q t3 0.3 0.2 0.1 0 5 10 15 20 25 30 35 40 45 Stand Age('ears) B 0.3 ClS f7 d¥ acquisilionintenral '1• o 7 (15Oct-02 Dec 91) 0 i c e 0.6 ;!: 0 0 0.5 0.•. f: 0.4 Q '5 0.3 l . . .~ 0.2 0.1 0 5 10 15 <.Q ES ::1) 35 40 4S Stand ~e (tears) c Figure 2: Variations of the degree of coherence with stand age 62 Fig. 3 shows the result of theoretical modeling, where different scattering mechanisms at C-band, VV and 23° of incidence are presented. I ...-.... • Crown I N ~ l \ • Ground 1 E I ••••••• \.\. N ! Total -J \ .1, ! \ <, I E l r-, I l m \l ~------l ·- ""C I <, • -..;:;;::··-·--:·--·-···-- • - •••••••••• '1 -----·--- - \~-~- - ___...--1!1······-- 0 -1> b ~ I \ I -11 l l ', l \. l -12 - I 0 5 15 Forest Age (years) Figure. 3: Decomposition of scattering mechanisms derivedfrom theory described in [7} For clear-cuts and very young stands, the backscattered signal results mainly from the soil contribution. As vegetation grows, this high contribution from the soil is attenuated by the crown layer, and the backscattering from the crown increases. Consequently, three distinct regions can be defined. For very young stands, the soil contribution is dominant in the total signal (region I), for older stands (3-12 years in the case of the Landes forest), backscattered signal is a sum of ground and crown contributions (region 2), for larger biomasses(> 12 years), most of the backscattered signal comes from the crown layer contribution (region 3). In terms of degree of coherence, bare surfaces present a high degree of coherence, if they do not undergo any modification in their characteristics (geometry, dielectric, vegetation regrowth) between the two acquisitions. Volume scatterers such as needles or branches are more sensitive to structure variations due to vegetation growth or wind effect. In the case of repeat-pass interferometry, these scatterers have a high probability to move between acquisitions. Thus the volume scattering from vegetation corresponds to a low degree of coherence. The degree of coherence, as a function of forest age or biomass, can be interpreted using the knowledge of the scattering mechanisms as follows. In the region where the soil contribution is dominant (region I), the degree of coherence is high. On the other hand, the region where most of the backscatter comes from the volume contribution (region 3) shows a low degree of coherence. In the intermediary region (region 2), the degree of coherence decreases with stands age/biomass, with a slope depending on soil/vegetation parameters. ~------63 The overall coherence of sets A and C is shown to be lower than coherence of set B. This could be the result of two effects: (a) a decrease of the degree of coherence as a function of time interval between acquisitions as shown in [4] and (b) a drop of coherence due to the strong precipitation (34 mm) which occured on October 1S. Lower coherence of A and C compared to B observed over clear cuts can in addition be explained by changes in the remaining vegetation cover (growth of herbaceous, cleaning of a stand after a cut) or by strong modifications of the roughness state (by harrowing or plough). For long intervals between acquisitions, the degree of coher~nce obtair;ie~ over _anarea can read~ the lowest stable values when the time period is sufficient to statistically mtegrate all possible (non anthropic) temporal changes. As a consequence, the degree of coherence between two separate acquisitions can be a good discriminator between forests and bare surfaces (or surfaces with low vegetation cover). A comparison with the intensity of the backscattered signal of ERS-1 as forest I non-forest discriminator can be made. At C-band, the intensity of backscattered signal from bare soil surfaces depends on the soil parameters (moisture, roughness). Consequently, bare soil surfaces can present a large range of responses (Fig 4). These possible variations of the soil responses may impede the forest I non-forest discrimination because of the possible confusion between some vegetated and non-vegetated areas. On the contrary, the degree of coherence of a bare surface is in most cases higher than the degree of coherence characterizing forested areas, and this independently of the soil moisture and roughness parameters. * Forest t:4" .5 Soil surfaces E N"' -6 I 1=8cm, s=1.5cm, er=20 mE .7 "'C -8 I 1=12cm, s=O.Scm, er=10 '-" > .g > t=acm. s=1.5cm, er=3 'b -10 ~· 1=12cm' s=0.8cm. er=3 •11 ~r e ·12 0 33 65 95 130 150 • Confusion area Total Biomass (tons/ha) Figure 4: Impact of soil parameters on backscattered intensity Fig. Sb shows a map of forest I non-forest obtained by thresholding the degree of coherence of A and B interferometric sets (Sa for set A). The resulting map is in good agreement with the forest map established from ground data. White areas are ground surfaces which did not change between the two extreme acquisition dates (1S oct I 2 dee). Black areas are forest stands characterized by a low correlation. Areas in grey are fields or stands which correlation state has changed. This could have been caused by plough, harvest (on agricultural fields), growth of herbaceous or cleaning of stands. 64 (b) Figure 5: Map of (a) coherence of set B and (b)forest I non-forest areas based on the degree of coherence. Nezer test-site. Concerning the study of young forest biomass, in the case of the Landes forest, the characteristics of the underlying ground and vegetation do not differ much from one stand to another, thus reducing the dispersion of the degree of coherence. Inversion of the degree of coherence into young forest biomass for monitoring is then possible. Selatan Test-Site In the case of the tropical test site, the forest environment is drastically different. The comparison of a tandem and 35 days interval interferogram shows that the latter is useless to discriminate vegetated from non vegetated areas (Fig. 6b-c). In particular, the quick growth of vegetation over former deforested surfaces brings a fast temporal change of target areas. A much shorter time interval is necessary between acquisitions to minimize this effect. The analysis of the tandem interferogram concurrently to the SPOT image of May 1996 (Fig. 6a) shows that under these environmental conditions, the degree of coherence can be helpful in discriminating heavy vegetated areas (plantations and forests) from sparsely covered surfaces. A more thorough study on the possibilities of discriminating other classes of vegetation (primary forest, different kinds of plantations, deforested areas) using a combination of intensity and interferometric data has been undertaken by CRISP [1O]. 65 (a) (b) (c) Figure 6: (a) SPOT image of the test-site. Comparison of the degree of coherence over the same area for (b) tandem and (c) 35 days interval interferograms. Previous studies [11] have shown the possibility to discriminate forested from non-forested areas in tropical environment using multi-temporal intensity ratio information. A first qualitative comparison between this technique and the use of the degree of coherence of a tandem interferogram shows similar results (Fig. 7). A quantitative analysis of the two algorithms in terms of classification accuracy and effective final resolution is to be undertaken. (a) (b) Figure 7 (a) Degree of Coherence ERS-1 I ERS-2 96104112 - 96104113 (b) Intensity Ratio Image ERS-1 93112101- 94108105 Phase Difference The interferometric phase shift between forest and ground responses is related to the integrated height of the vegetation scatterers. Thus, penetration of the wave into the medium must be taken into account. Consequently, the interferometric estimated height is an" effective height" function of the real height of the trees and of the penetration depth (Fig. 8). 66 The Landes forest test-site is well suited to the study aimed at retrieving the forest height for the following reasons: -the stands are large: large number of samples for statistical estimation -the terrain is flat: no topographic effect -a validated theoretical model is available and provides theoretical extinction coefficient and penetration depth Since the phase difference rms decreases with an increase of the degree of coherence, the measurements were extracted from the interferometric set B, which presents the highest overall coherence. The larger baseline of B is also well suited to reducing therms-height errors [5]. penetration depth op A;z=oJ P(z = 0) e .o \ ~ Figure 8: Determination of canopy heightfrom interferometry The general topography of the test-site is flat (as a rule, the slope is less than 0.5%). In addition, we chose to restrict the study to a small area around a clear cut which was considered as the ground reference level. The mean value of the phase difference is extracted from each stand and the corresponding height difference is computed. This leads to the interferometric measured heights displayed in Fig. 9. We can observe on Fig. 9 that the variation of the mean value of the phase difference estimated from interferometry is an increasing function of the stand age. However, discrepancies are observed between the estimated and the actual tree heights. 67 20 18 ..-.... 16 l ~ ...... _,..E 14 ..c:~ 12 8P [ C) ••• ·- 10 Q) I :c 8 op[: Q) Q) 6 ~ op c• • I- 4 2 0 0 10 20 30 Tree Age (years) Estimated Height from Interferometry • Estimated Height from Interferometry + Simulated Penetration Depth (8p) ~ Actual Height Figure 9: Estimation oftree heightfrom interferometry Discrepancies exist between estimated and actual heights, as the measured height is not the height of the top of the trees, but the height of scatterers distributed over a thickness equal to the penetration depth, and is smaller than the actual height. One way to correct these estimated stand heights is to use theoretical modeling to compute the penetration depth of the wave into the medium. For each stand under study, the penetration depth d defined by: P(z ~ o) 1 P(z ~ 0) e (where P(z) is the transmitted power at a depth z below the top of the crown layer, and where the conventions are those of Fig. 8 ) is estimated using the modified radiative transfer model 68 described above. In the case of the Landes forest, young stands are characterized by a high density of trees and a thin crown layer. As the trees grow older, the crown becomes thicker, but the tree density decreases as a consequence of thinning practice. As a first approach, the tree crown is modeled as homogeneous layers; a young stand will be represented by a thin slab of homogeneous medium, an older stand will be represented by a thicker slab of homogeneous medium with a lower extinction coefficient, as depicted in Fig. 10. 8p Figure I 0: Variations ofpenetration depth with tree stand age In the case of the Landes forest, and for ERS configuration, the resulting penetration depth increases with the age of the trees. A corrected estimated height, sum of the interferometric derived height and of the simulated penetration depth is shown to be a good estimate of the actual height of the trees (Fig. 9). The remaining errors could be reduced (a) by using interferometric pairs with a larger baseline to 69 reduce rms height dependence on rms phase difference, and (b) by enhancing the overal degree of coherence using interferometric pairs with a smaller time interval. If the different penetration depths are not available, is should be noted that a classification of relative forest heights is still possible using phase difference information alone. For a given forest, the relation between interferometric height from ERS and real height can be derived once and used (as a look up table) to monitor the forest height variations. Conclusions This study has underlined the relations between the interferometric degree of coherence and the forest age (or biomass). Coherence has been shown to be efficient to discriminate forest areas from clear-cuts in a temperate forest environment. For a tropical forest environment, good results are also obtained if the time interval between acquisitions is short enough. In a specific case, the use of phase difference has been undertaken to retrieve forest stands heights. Optimal interferometric pairs with a smaller time interval (tandem acquisitions) and a larger baseline should (a) enhance the use of the degree of coherence to discriminate young forest stands and (b) permit a more quantitative use of the phase difference to extract forest height. However, some important improvements are to be made: - the behaviour of the degree of coherence in the region where backscattered signal comes from both ground and crown contributions is quantitatively unknown. This may impede the retrieval of young forest biomass when the underlayer varies from stand to stand. -the computation of the penetration depth requires knowledge of the crown parameters, and is only an estimate of the real integrated penetration of the electromagnetic wave into the crown. To process further into the study of the relations between interferometric data and forest parameters, theoretical modeling must be improved to account for coherent interactions. Acknowledgments The work is carried out under CNES/CESBIO Contract n°833/2/95/CNES/l 71. The interferometric data have been provided by CNES/QTIS. Nicolas Floury receives a grant from CNES and Alcatel Espace. References [l] J.O. Hagberg, L.M.H. Ulander and J. Askne: "Repeat-Pass SAR Interferometry over Forested Terrain", IEEE TGRS, Vol. 33, No. 2. March 1995,pp 331-340. [2] U. Wegmuller and C.L. Werner: 11 SAR Interferometric Signatures of Forest 11, IEEE TGRS, Vol. 33, No. 5, September 1995, pp 1153-1161. [3] J.C. Souyris, T. Le Toan, C.C. Hsu and J.A. Kong: 11 Inversion of Landes Forest Biomass using SIR-C/XSAR Data: Experiment and Theory 11, Proceedings ofIGARSS'95, July 95, Vol. 2, pp 1201-1203. [4] H.A. Zebker and J. Villasenor: 11 Decorrelation in Interferometric Radar Echoes 11, IEEE TGRS, Vol. 30, No. 5, September 1992, pp 950-959. [5] F.K. Li and R.M. Goldstein: 11 Studies of Multibaseline Spacebome Interferometric Synthetic Aperture Radars", IEEE TGRS, Vol. 28, No. l, January 1990, pp 88-97. [6] J.S. Lee, K.W. Hoppel, A. Mango, A.R. Miller: 11 Intensity and Phase Statistics of Multilook Polarimetric and Interferometric SAR Imagery 11, IEEE TGRS, Vol. 32, No. 5, September 1994, pp 1017-1028. 70 [7] C.C.Hsu, H.C.Han, R.T.Shin, .LA.Kong,A. Beaudoin and T. Le Toan: "Radiative transfer theory for polarimetric remote sensing of pine forest at P band", URS, Vol. 15,No. 14, September 1994, pp 2943-2954. [8] Yueh, S.H., J.A. Kong, J.K. Jao, R.T. Shin and T. Le Toan, "Branching model for vegetation", IEEE Trans. on Geoscience and Remote Sensing, vol.30, no.2, pp. 390-402, March 1992. [9] Tsang, L., .LA.Kong,and R. T. Shin, Theory of Microwave Remote Sensing, Wiley-lnterscience, New York, 1985. [10] N.Stussi, L.K.Kwoh, S.C.Liew, K.Singh, H.Lim: "ERS-1/2 Interferometry: Some Results on Tropical Forests", FRINGE 96, September-October 1996. [11] T.Le Toan, F.Ribbes, T.Hahn, N.Floury, U.R.Wasrin: "Use of ERS-1 SAR Data for Forest Monitoring in South Sumatra", Proceedings of IGARSS'96, May 96, Vol.2, pp 842-844. 71 Analysis of ERS-SAR Tandem Time-Series Using Coherence and Backscattering Coefficient 0. Stebler, P. Pasquali, D. Small, Remote Sensing Laboratories (RSL) F. Holecz, and D. Nuesch University of Zurich, Winterthurerstrasse 190, CH-8057 Zurich, Switzerland [email protected] Abstract Interferometric synthetic aperture radar (InSAR)' is a powerful method that enables both estimation of topography and derivation of thematic information. The purpose of this paper is to analyse a year long time-series of coherence and backscattering coefficient data from the ERS-1/2 Tandem mission. The data set was collected in Switzerland (Bern region) over varying terrain (from flat to mountainous) with a variety of ground cover types, including agricultural fields, forests, lakes, and urban areas. For a physical interpretation of the backscattering coefficient and a correct estimation of coherence, elevation data is required. This height information can be derived from InSAR or existing elevation models. Only after these calibration and estimation steps can one directly compare coherence and backscattering values within the time-series in an absolute way. Both qualitative and quantitative analyses are presented and discussed. Keywords - InSAR DEM, SAR data calibration, backscattering coefficient. coherence estimation, time-series 1. Introduction For a qualitative and quantitative analysis of SAR time-series data, the backscattering coefficient and coherence need to be properly normalized to correct for biases introduced by topography, baseline, and processing. For a test site in an area surrounding Bern, Switzerland, we present examples showing why such normalization is necessary, followed by presentation and interpretation of the time-series. 2. Data Sets Our ERS Tandem time-series consists of eight complete tandem acquisitions, dating from June 1995 to April 1996. Two pairs are incomplete due to satellite dropouts in September 1995 and February 1996. The perpendicular components of all tandem baselines are below 210 meters. Precise orbits were used during the interferometric data processmg. The acquired data cover an entire vegetation cycle, including a variety of ground cover types, such as agricultural fields, forests, lakes, and urban areas. The data set was collected over varying terrain, ranging from flat to mountainous. Meteorological conditions during the acquisition were measured hourly by the Swiss Meteorological Office. Ground truth samples were taken at different sites within the scene. Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Sept. - 2 Oct. 1996 (ESASP-406, Vol. II, December 1997) 72 Figure 1: Bern, Switzerland - Geographic Location of SLC Quarter Scenes (ERS Tandem data, Frame 2655) 3. Estimation of Backscattering Coefficient From the radar equation for distributed targets it is known that the received power is modulated with the 2-way-antenna gain G~(Si) and with the reciprocal value ofsin(S;), where 9; is the local incidence angle. For each pixel these quantities are therefore dependent on the radar look angle e L, the depression angle of the antenna, the sensor position and attitude, the position of the backscatter element, as well as on the processed pixel spacing in range and azimuth, (Holecz F. et al., 1994), (Holecz F. et al., 1995). Since SAR processing does not include topographic information, these radiometric corrections are omitted during the processing step, and need to be considered in a postprocessing step. Additionally, for ERS SLC products two further factors have to be taken into account, namely the calibration constant (dependent on sensor and processing facility) and the R3 correction. 73 Figure 2: Extract from geocoded uncalibrated (above) and calibrated (below) ERS-1 amplitude data (26.11.95). 74 Figure 3: Applied calibration factor for the complete ERS-1 quarter scene of November 26th, 1995. Calibration factors range from -67 dB to -42 dB. 4. Estimation of Coherence Coherence is defined (Foster M.R. and Guinzy 1., 1967) as the magnitude of the correlation coefficient of the complex signal data; it is directly related to the phase noise that is present in a SAR interferogram. This information can be used to estimate the achievable elevation accuracy during DEM generation, to investigate temporal changes of an observed area or to optimize the processing of the data. .The estimation is carried out on a window basis; it has been shown that this is the Maximum Likelihood Estimator (MLE) (Touzi R. et al.. 1996). The statistical confidence of the estimated coherence is a function of the number of independent samples (degrees offreedom n, (Prati C. et al.. 1994)) and the true coherence Y.The estimator is biased for all values of coherences, especially for areas with low coherence; estimates in these areas show a high variance. Other sources of coherence mis-estimation are introduced by 'baseline decorrelation', processing effects and topographic modulation of the interferometric phase. Therefore, to compare coherence estimates within a time-series, a careful design of the processing has to be carried out, i.e. including spectral shift filtering (Gatelli F. et al., 1994); removal of topography must be performed before the coherence estimation, and in a final step the coherence bias is corrected. Consideration of the local slope improves the estimation of the coherence. Figure 4 shows coherence histograms for the October 1995 pair. The interferograms were flattened with (a) an ellipsoid model, (b) InSAR derived slopes and (c) with a high resolution digital elevation model provided by the Swiss Federal Office of Topography (DHM25) (Small D. et al., 1995). Improvement of the coherence distribution is obtained when the influence of the topography is removed before coherence calculation. The use of pre-existing DEMs (when available) allows an accurate modelling of the topography. With this information, an accurate coherence estimation is possible also in low coherence areas, where the InSAR derived slopes are poor. InSAR-derived slopes can also provide better 75 bias removal - for example, in areas where the independent DEM is no longer current (gravel pits) or where the heights references differ (ground vs. tree height) (Small D. et al., 1995). oozs 0.0:!> !;' O.OJ.S I! lo.ow 0.0QS 0.000 0.0 0.2 OA 0.6 0.11 io Figure 4: Global quarter scene histogram of coherence after flattening with DHM25 (solid line), InSAR derived local slopes (dotted line) and with ellipsoid model, ERS Tandem, October 1995. The last step, correction of the estimation bias, is strictly related to the number of degrees of freedom n used in the estimation. This value is equal to the number of independent samples contained in the moving window (of dimension m) used for the estimation. For several reasons, n and m are different (Joughin I. R. and Winebrenner D. P., 1994): the system resolution and the sampling of the SLC data in general do not correspond, a spectral weighting is present in the data, and defocusing effects can affect the images. In general it is not easy to obtain an analytical value of n: in the present work it has been estimated in an area with known homogeneous coherence (e.g. large water areas, where Y is assumed to be zero). Next, the a posteriori density function for all measured coherences and a given n is computed (Foster M.R. and Guinzv J., 1967).Now the relation between measured and true coherence is known. An example with a given degree of freedom is shown in figure S. The correction for all ERS Tandem pairs was performed based on this relation. 76 Ii 2.1:1 • • g : : .j! 2.0 •••••••••••• ···-i····•·•···•·...... ••••.•..••. ···············+·.... .e : : l 15 ···············+··············· ' ' I1: -----··:::::::t:::::::::::::: •••••••••·--~··---··••••••r' ' " ~ o J-s.___...._._~_.__._.....__....._.__._...__.__._....__._.__._....__. O.t> .i:J.'2 OA 0,6 o.8 1-<> 'lnle mhr~nr:e : : : : L . 'e~ J.: L: :l : l ------i-- --i------~ ------~-·---·~· . i : : : : .g ··············-~---·············j················t···············-~~---········· ~ : : : : 0 ...... ••..•...•..•....•••••••••••.....•...•.•.•.•.•••.••••••••••••••••.•••.••..••.••••••••.••.••.•.•.••••.•.••.•••••••••.••••.•.•.•••• ~ j L l L . C' : : : : E .I .I I I ' . . . 0.1) ll.:Z a.& II.I U> trve aah1rmna1 1.0 ' ' ' ' . '. ' fI:: :::::::J::~~);;::i?~t:::::::---·--:--,.,: ..~·: : 0.2 ~ ,,. -~r--:'.-~--~=:-,·r::._:· +------+------ ~-..L...~.....-~-+..¥- : ! : 0.2 D..4 0.8 1.0 df: 32.44<1-512 Figure 5: Top: A posteriori density functions for measured coherences (0.01-0.99). Middle: Plot of maxima of a posteriori density functions. Bottom: Relation between measured and true coherence (degrees of freedom: 32.4, estimated from a large water area, dotted lines: 10% significance level). 5. Time-Series Analysis Coherence was estimated for five thematic classes: agricultural land, forest, meadow, urban area and water (lake). Figure 8 shows corrected coherence values. Agriculture: In summer there is a lot of agricultural activity, some fields are coherent, while others show no coherence (high variance). In Winter fields are bare and coherence is generally very high (see figures .Q and 1). Forest: Coherence is low for the whole vegetation period. There are only small increases of the coherence in winter (less decorrelation within deciduous forest). Meadow: High variance of coherence during the whole vegetation period. Water: Coherence is almost stable for the whole time-series. 77 Figure 6: Test site 'Buren' during summer (agricultural area with different vegetation cover. enclosed by the Aare river). Tandem pair: August 13-14, 1995 (raw coherence). Figure 7: Test site 'Buren' during winter (agricultural area with no vegetation cover, enclosed by the Aare river). Tandem pair: March 10-11, 1996 (raw coherence). 1.ar------.------.-----..-----.-----~------.-----..-----~---- !I.I ..~· a a ••••••• a " • " a • .•..~ *i<~ Jl, *~ .,.. •••d'ct. ~ .0. ..• Jl, """ i< " Jl, 00 6(,>QQ " ""' o 0 0 <> 0~69" ll.2 0 oc> 09oo 0 0 <><><> 0 o~ oc.tW"> '°' ee"t><> 0 o<,oo()A..>() 0~ x M x x x x o.o~ "..__ .._ .._ ~---~--- " -tAll>liO 9.,/lll'.7. 1~- 17./11!.&. ~10.. 211./27.11. ::n.1:zs:i/1.1.1J11 -t.1).2. 111./11.:S. HJ)::.-t. 11moti...... ,, •-• 1a1om a...µ.ttionic H n•.• Figure 8: Time-series of different test sites (star: agriculture, diamond:forest, triangle meadow, square: urban area, cross: water). Coherence bias was removed. Coherence was lost in the December/January pair due to heavy snow- and rainfall (seefigure 1.Jl) 78 -6 -a -10 ID "0 -12 0 E ·v;"' -1+ -1B -18 2.6., 1. 27.11. tim~ int;orvglbi:twccn tQnd~m 11~quisifam:s;24- hr:s Figure 9: Backscattering coefficientfor five ground cover types (star: agriculture. diamond:forest, triangle: meadow, square: urban area, cross: water). Tandempair: November 26-27, 1995. With a given estimator of coherence, we conclude that correction of the coherence estimation bias is significant only in low coherence areas. For higher coherences, there is little difference between corrected and uncorrected values (see figures 2 and S). Figure 9 illustrates backscattering coefficients and their deviations for several ground cover types. For these two particular acquisition dates (November 26th and 27th) it is evident that backscatter-based discrimination between the selected ground cover types is not possible. Coherence provides a much better discriminator between different ground cover types. However, the class separability depends on the season (i.e. 13./14.8.95, figure S) as well as terrain types and weather conditions (i.e. 31.12.95/1.1.96, figures S and lQ). 79 ';;;' SI\IA Hourly Bem-Liebefeld,Swinctl.md ]:, 25 ::::10 ~ 1.5 ·=> >< 10 :il .5 ~~· v •• -,. I../"'', . ~ ._A I ] 0 L ~ I -·()•>('./'./>.(• • ¢"" J ~ l?fl_±, l2fl...8 12/JO 01/01 01/0l 10 ~ .5 ~ QI 0 !:-< ~ -.5 ~ -10 ~ t3 -1.5 -20 l?j'l.±J 11,n...a 12/30 01/01 01/0l ~ 2.0 ~ 15 3 = o1 LO ·B,. :!:J. OJ ~ 00 l?J'l.±J l?J'l...8 12/JO 01/01 01/0l 199.5-1996 Figure 10: Meteorological data provided by the Swiss Meteorological Office for the BemLiebefeld station. Note the heavy snow- and rainfall between December 31st, 1995 January Ist, 1996 causing the loss of coherence seen in Figure 8. Acquisitions were at approximately 11:20 local time. We show in figure ll for one example within the time-series a combination of backscatter and coherence information. The correction of backscattering intensity and coherence estimation presented in this paper significantly enhances the value of these types of products. Figure 11: ERS Tandem InSAR signatures (26./27.11.95) for the test site 'Buren'. Red: interferometric correlation (raw coherence), green: backscatter intensity of first acquisition, blue: backscatter intensity change. 6. Conclusions The achieved results point to the following conclusions and remarks: 80 1. An accurate calibration of the backscattering coefficient and coherence considering elevation data is a fundamental requirement even for qualitative analysis. 2. For accurate estimation of coherence, the following effects have to be considered during the interferometric processing: (a) Spectral shift filtering due to baseline decorrelation and processing effects. (b) Correction of the estimation bias using the Gamma distribution. (c) Removal of local slope effects. Coherence estimation benefits from DEM flattening of the interferogram, especially when a large number of samples are taken to improve the coherence estimate. 3. After these calibration and estimation steps, one is able to compare backscatter and coherence values within the time-series (i.e. different coherence products). These processing steps are a fundamental prerequisite for further analysis such as classification based on these values. 7. Acknowledgments ERS SAR data and precise orbit ephemeris information were provided courtesy of ESA/ESRIN. The DHM25 elevation model used within some calculations shown here was provided courtesy of the Swiss Federal Office of Topography. Meteorological data was provided courtesy of the Swiss Meteorological Office (SMA). We also would like to thank Dipl. phys. Ruedi Wettstein-Gloor (RSL) for his useful contri butions. 8. References [1] Foster M.R. and Guinzy J., 1967 The Coefficient of Coherence, its Estimation and Use in Geophysical Data Processing, Geophysics, 32, pp. 602-616. [2] Gatelli F. et al., 1994 The Wavenumber Shift in SAR Interferometry, IEEE Transactions on GRS, vol. 32, n. 4, pp. 855-865. [3] Holecz F. et al., 1994 Rigorous Derivation of Backscattering Coefficient, IEEE Geoscience and Remote Sensing Society Newsletter, No. 92. [4] Holecz F. et al., 1995 Topographic Effects on the Antenna Gain Pattern Correction, Proceedings of IGARSS'95 Symposium Florence. [5) Joughin I. R. and Winebrenner D. P., 1994 Effective Number of Looksfor a Multi/ook Interferometric Phase Distribution, Proceedings ofIGARSS'94, Pasadena, pp. 2276-2278. 81 [6] Prati C. et al., 1994 Report on ERS-1 SAR Interferometric Techniques and Applications. [7] Prati C. and Rocca F., 1992 Range Resolution Enhancement with.Multiple SAR Surveys Combination, Proceedings ofIGARSS'92, Houston. [8] Small D. et al., 1995 Combination of Ascending/Descending ERS-1 lnSAR Data for Calibration and Validation, Proceedings ofIEEE-IGARSS'95, Florence, Italy, pp. 553-555. [9] Touzi R. et al., 1996 Estimation of the Coherence Function for Interferometric SAR Applications, Proceedings ofEUSAR'96, pp. 241-244. 83 LABORATORY EXPERIMENTS FOR THE INTERPRETATION OF PHASE SHIFT IN SAR INTERFEROGRAMS Jean-Paul Rudant UMLV & UPMC, tel: 33 1 44275087, fax: 33 1 44275085, Email: jpr @ lgs.jussieu.fr Ali Bedidi, Rodolphe UMV (Universite de Marne La Vallee) Calonne Didier Massonnet CNES, OTIS, 18 ave Edouard Belin, Toulouse Giuseppe Nesti, Dario EMSL I SAi I JRC Ispra, Italy Tarchi Abstract Purpose: Interpretation of SAR interferograms (ERSl) in the light of phase shift laboratory measurements obtained as a result of moisture changes. Results: We show that surface phase changes can be a contribution of both geometric and dielectric effects.The geometric effect result of soil swelling or settling. In our experiments, the dielectric effect is always equivalent to a subsidence when moisture increases and we interpret particular observation made on interferograms as a swelling of the most humid low area. I. Introduction Interferometric data (amplitude, coherence, phase difference) may be used for general or thematic cartographic purposes. They provide information on surface state and surface changes that occur betwen different dates of view. Coherence and phase difference images bring on these points original information that usually used amplitude images do not necessary contain (change of superficial geometry at the wawelenght scale, phase effects of dielectric origin, measure of weak amplitude subsidence of mining exploitation, etc ..). In particular, rotations of phase observed in the interferograms take their origin all both in the displacement of the surface and in changes of dielectric properties that occur between different dates of view. In this study, wetempt to evaluate the various contributions by confronting measures of phase rotation undertaken in the laboratory to interpretations of interferograms obtained on the globally stable site ofNaizin in Brittany. II. Test Site The test site is the Naizin area in South Brittany , France ( size 50km * 50km, Lat 48°N Long 3°W) . Present day, the Naizin area is aseismic, without active deformation, in a standard agricultural area. Fig 1 give also the rain periods over the Naizin basin situated in centre of the study area. The distribution of the rain events is not exactly know for the entire test site and it is expected that the Naizin basin is representative of the entire test site.This hypothesis seems reasonable for this kind of climate, devoided of local microclimate. Proceedmgs of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich. 30 Sept. - 2 Oct. 1996 (ESA SP-406, Vol. 11,December 1997) 84 III. INSAR products The ERS 1 images used were acquired during phase B (3 day repeat orbit) during February and March 1992 from ascending orbits. Out of 12 available images, 6 are in interferometric conditions. The corresponding acquisition dates are the following: 6, 9, 12, 15 February, 13 and 16 March, in a winter period where there is no variation in vegetal cover. For each pair, we have two amplitude images, a coherence image and a phase difference image. All these products were corrected for the effects of distance and of known relief (ref 1). Sign convention for phase differences: The difference provided is algebraic, coded in 8 bits. The interval (0..255) corresponds to a complete 360° phase rotation and the result of the processing in CNES is given by: DIF-PHA1_2 = ( (2I10) * (optical path 2 - optical path 1)* 256 )+128 + (3.1) G 1 -j2)reflection*256/360 + Geometric corrections ) all modulo 256 where 1 indicate the master image and 2 the slave one. The optical path is a one way antenna to target journey, 10the wavelength in vacuum, j the phase effect introduced by ground reflection. The purpose of the geometric corrections is to remove all the predictable effects on the phase (i.e. topography and orbits) from the interferogram. The choi.ceof G 1 -j2) follows from the use of the e jwt convention in the formulea for a progressive wave: e j(wt - k r + j ), with k=(2p/10)*n*u (3.2) where n is the optical index (complex value) and u unitar vector in propagation direction. It results from (3.1) and (3.2) that, with this convention, a negative j is equivalent to an increasing of the optical path. An area raised by a few mm between the two image acquisitions, for example, would look darker than its surroundings on the phase image (for black and white phase scale). IV. interpretations of the phase images In summary, the phase image would be completely uniform ifthe following conditions were fulfilled: - exact geometric corrections, -the atmospheres at the two image acquisition dates have no propagation inhomogeneities (they need not be identical since the differences are to within a constant), - surface conditions unchanged or subject to a uniform movement or dielectric change. In practice, the phase difference images show effects with different origins, related either to the preprocessing, or to changes in the conditions of the surface and the atmosphere. Artefacts related to the precision of the DEM used 85 The correction performed to bring the phase differences to altitude zero takes into account the altitude provided by the DEM used. The imprecisions of the DEM have an impact on the values of the unwrapped phase according to the following algebraic relationship: Signed phase error/ 360 = - DEM error I Ha (4.1) where Ha represents ambiguity in altitude . This results in certain systematic artefacts, in particular for the low points, since the resampling of the DEM used at 40 m smoothes the thalwegs and makes them seem higher than they really are. An example of the preceding phenomenon is shown in Fig 2 for phase image PHA-10, corresponding to an altitude of ambiguity of+ 21 meters. The lowest points in the network stand out in several places by a negative contrast that corresponds to an error in the DEM that is positive by several meters for the thalwegs. Modifications of atmospheric states, dielectric changes in surface states, global displacements The corresponding effects do not depend on the altitude ambiguity. Atmospheric effects are linked with the variability of temperature, pressure and humidity in the air, and dielectric effects are linked with humidity, temperature of the soil and it vegetal cover. A vertical surface displacement +d will induce a negative change in phase that is inversely proportional to the wavelength used, equal to - 360 * 2 * d * cos(a) / l where a is the beam incidence angle. On the interferograms considered here several remarkable effects can be noted: 1-A general variability of 30-40° amplitude, which is neither systematic nor structured either . by the relief or by man-made structures (fields, networks, urban areas), typically extending around 10 km, seems to be due to atmospheric effects (for example Fig 2). This type of effect is found on all the images with variable amplitudes. 2-Structured phase changes highlighting certain fields, in particular when the humidity conditions have changed between two acquisitions. (Fig 3). When the contrast in humidity conditions is smaller (dry period or uniformly wet period), the phase change associated with the fields disappears (see for example Fig 2). There may be various reasons for such positive or negative differentiation between the fields and their environment: - a uniform elevation of the backscattering surface that leads to a shortening of the optical path and therefore a negative phase change (darker area). - an increase in the mean value of the dielectric constant of the vegetation due to humidification, leading to a variation in the optical path and to a positive phase change if the dominant contribution in backscattering is given by the reflexion on the ground. Note that it is unlikely that cutting vegetation would increase the path, because this would make the speckle and the phases incoherent, since the basic reflectors would have changed. Also note that fields probably corresponding to very rough bare soils, wich are characterized by their high radiometry (Fig 3), do not have a specific signature in the interferograms. The contrasts in phase with their environment are in general lower than 10°. Note that for all the fields mentioned above, the phase noise is low and the coherence level high. 86 3- Phase changes on a scale of several kilometers correlated with the general shape of the observed relief when the climatic conditions change from a dry period to a wet one (see figl for rainy events). Comparing the digital terrain model with the unwrapped phases PHA-1 and PHA-2 (6 (dry) and 9 (wet) February on the one hand and 6 and 12 February on the other) (Figs 4, 5) shows that the low areas react in a differentiated way, with phase difference DIFF-PHA, (see equation 3.1) of the order of ( - 80°), with respect to their surroundings (equivalent to moving the backscattering surface closer). This effect is equivalent to a reduction in the optical path, and the possible origins of the phenomen are: dielectric effect related to humidification (phase rotation equivalent to a reduction in the optical path between the second and the first date, with j2 -j 1 = +80° (eq 3.1), ground swelling, localized atmospheric effect . Experimental study on the Naizin site (ref 2) shows that the level of humidity depends on the relative elevation with respect to the river and that the contrasts between low and high areas is larger for a wet period than for a dry period. It thus seems plausible that for the second date used for interferograms PHA 1 and PHA2, the low points in the soil have a higher humidity level , and in consequence the air also. V. laboratory experiments First, let consider what phase rotation the Fresnel formula gives at normal incidence as result of the humidification of a flat homogeneous soil. 5.1. Phase rotation in Fresnel reflection coefficient The Fresnel reflection coefficient for an incident wave normal to the interface plane of two homogeneous media is given by: reij= (1-nr)/(l+nr) (5.1) where nr is the relative index of the 2nd medium with respect to the first andj is the phase rotation associated with the reflection of the wave. This formula can be used for a greatly simplified initial quantitative evaluation of the effect of dielectric changes. We shaw in (ref 3) that the variation is always bounded to 90° and for typical values of dielectric constant is lower that 30°. The evolution of the real and imaginary values of the dielectric constant as a function of the humidity leads, for example, to the following results: Starting from a dry insulated medium (tropical soil) with dielectric constant e = 4.8 (real index m=2.2), subject to great humidification, we obtain: -for dry ground, a phase rotation j = -180° for all frequencies - after humidification, in X-band; e = 15 - j * 9, nr = 4.03 - j * 1.11 (sign - for the imaginary part in accordance with formula 3-2),j= -187.8°, in L-band; e = 24 - j * 4, nr = 4.91 - j * 0.4 ,j= -182°, i.e. phase changes of respectively -7.8° and -2°; the - sign corresponds to a phase rotation equivalent to an increase in the optical path travelled by the wave, in accordance with the formulae 3-2, when this phase rotation is interpreted in terms of time delay. This result shows among other things a lower sensitivity to the effects of humidity when the frequency decreases but in general that a pure dielectric effect at reflection level is always 87 very small and this is in agreement with the observations on the interferograms where for "bare soils" there is not very much differentiation. However, the speckle effect can produce larger variations as pointed out in the above mentioned report. 5.2. Experimental measurements Several experiments have been carried out in the laboratory (at ESIEE and JRC Ispra) in order to approach the preceding interferograms quantitatively. 5.2.1. Experiments at ESIEE At ESIEE, experiments were conducted in order to study moisture effects on the backscattering and phase shift of microwaves by different kind of soil. An HP 851OA spectrum analyser and two X-band antennas were used to perform the measurements. experiments were limitted to X-band waves because of the physical constraints due to the sample size (ref 4). These experiments provided useful orders of magnitude and helped to formulate, in a better way, the questions related to the phenomena under study. A flrst experiment consisted in the measurement of the phase rotation of two samples, sand and humus, at different moisture states. The results show that for sand sample (fig 6) the phase rotation is equivalent to an increase of the distance between the sand and the antennas, the real displacement (sinking) contributes for one third to the equivalent displacement and the dielectric effect for two third. For humus sample (fig 7), an important surface swelling can be observed but the pure dielectric effect is, as for sand sample, equivalent to the surface moving away (increase of the distance sample-antenna). These measurement have been conducted at normal incidence. In a second experiment, the samples wer covered by vegetation to study the hase shift undergone by the wave upon its transmission through vegetation. The vegetation, dry or wet, consiste of numerous thin stems, forming a dense and homogenous (at the wawelenght scale) medium. The results show that the humidification of the leaves induce an attenuation of the amplitude of the backscattered wave and a phase rotation of -20°. An increase of the density of the vegetation (number of shoots) is also followed by a negative phase rotation. (sign - equivalent to an increasing of the optical path). Humidification, like the increase in density, leads therefore on the one hand to attenuation and on the other to a phase rotation equivalent to an increase in the optical path. The magnitude of this results are to be compared with the phase changes observed over fields after rainy periods and allows them to be interpreted when the amplitude decreases and when the phase change corresponds to an increase in the path (positive contrast on the interfeograms). 5.2.2. Experiment at the EMSL Measurements (monostatic, polarimetric) have been performed on two samples (1.0 x 0.7 rn2) of sand in the frequency range 2 -12 GHz and for three incidence angles (18°, 23°, 28°). Progressive humidification of the samples has been performed by gently sprinkling water on the surface, from dry to saturated moist conditions. The experimental procedure and the data analysis is described in (ref 5). The experimental results can be summarised as follows : • In spite of the expected large increase of the backscattered sample, the signal remains generally well correlated. The phase shift is clearly dependent on the moisture level of the upper soil layer (few centimeters). In fact, a saturation level is reached very quickly, when the layer interested by the infiltration is still smaller than 10 cm. • The humidification of the first sample (with a sinusoidal shape perpendicular to the beam, period 10 cm, amplitude 0.5 cm) leads to a phase shift of the backscattered signal at an incidence angle of 18°, reaching -35° in X-VV band and -20° in C-VV band. (Fig 8.a). The jumps in the curves corresponds to repeated measurements after an interval of 88 12 hours without adding water. • The influence of the geometry of the reflecting soil is very important (Fig g.b): the linear dependence on frequency for the first sample could suggests that a real sinking of the surface has occurred (-2.5 mm). However, this hypothesis is not confirmed by the data on the second profile (period 20 cm, amplitude 2.5 cm). where a sort of oscillation is present with an escursion in the positive range (+15° at 4 GHz). This effect shows that in certain configurations the dielectric effect is modulated by the geometry in such a way to result in an apparent swelling of the surface. VI Conclusions The experimental approach employed allows us to appreciate quantitatively some of the causes of the variability in differential interferograms. The main points are recalled below: - For soil, the purely dielectric effects are in general superposed on the geometric effects of sinking or swelling that depend on the nature of the soil. The humidification of sand induces sinking by a fraction of a millimeter while that of topsoil can lead to swelling of several millimeters. In addition, also the effect of surface geometry , needs to be considered. - In most of the experiments carried out, the purely dielectric effect associated with humidification is equivalent to the backscattering surface moving away. In the only case where the opposite observation was made, the phase rotation obtained was limited to 15 °. We would mention that neither the -35_neither the+ 15_can be explained by pure dielectric surface effects. May be a volumetric interaction takes place even if I can not image any volumetric mechanism when the whole soil layer is saturated of water. - In transmission, the effects associated with the humidification of plant cover or an increase in biomass have the same type of signature - weakening of the signal and increase in the optical path. If we compare these results with the observations made on the analysed interferograms from the Naizin site, we can make the following comments and propose the following conclusions: - some important phase changes observed over fields can be interpreted in terms of changes in the optical path due to a modification of the density characteristics of the vegetation traversed. The effects of humidification on the fields are only visible through the associated phase changes when the humidity contrast is large and the delay between image acquisition very small.(3 or 6 days in our exemples).Thise effects are not visible for the bare soil fields. - the phase rotation of the order of 50° to goo(equivalent to moving closer to the surface) observed on the lowest areas when going from a dry period to a wet period is larger than the purely dielectric effects that we found elsewhere and are, furthermore, (for our measurements) generally equivalent to the backscattering surface moving away. The hypothesis of a swelling of the most humid low areas should not therefore be discarded. In our case, this swelling, corresponding to goo in C-band, would be of the order of 7 mm (result obtained by neglecting the purely dielectric effect). For a conventional use of interferograms in geodesy, DTM production for example, the previous effet will be responsible for an error of a quarter of altitude ambiguity. It is clear that at this stage, additional experiments would be welcome to complete this initial approach to measuring the phase changes associated with changes in surface state, in order to expand the use of radar interferometry in the field of environmental monitoring. The humidification experiments carried out with backscattering on different soil types (variable depending on their nature and roughness), in backscattering and transmission over different types of vegetation (variable in structure and biomass) would allow the phenomena involved in the generation of differential interferograms to be better quantified. The EMSL's anechoic chamber at Ispra is ideally suited to this type of measurement. 89 References 1. Discriminating geophysical Phenomena in Satellite Radar interferograms D.Massonnet, K. Feigl Geophys. Res. Lett., 22, 1537-1540, 1995. 2. Effect of saturated areas on backscattering coefficient of the ERS 1 synthetic aperture radar: first result, Merot P., Crave A., Gascuel-Odoux C., Louahala S. Water Resources Research, vol 30, (2), p 175-179, 1994. 3. Mesure au laboratoire d'effets de phase lies a des changements dielectriques d'etats de surface: integration a l'echelle du pixel sol. Rudant J.P. Rapport 92 I CNES I 0421, 1992 4. Caracterisation d'un sol par la rotation de phase d'une onde centimetrique, Calonne R., Arnaud V Memoire d'Ecole d'Ingenieur, ESIEE, March 1995 5. Decorrelation of Backscattered Signal due to soil moisture changes Nesti G., Tarchi D., Rudant J.P IGARSS 95, pp 2026-2028 This work was founded by CNES and PNTS, and the ERS1 data given by ESA for the scientific project AO F 07 (Responsable Jean Chorowicz). Thanks to Patrick Gigord for offering technical assistance. Figures 11 • 11 10 20 30 60 '70 IO 90 Fig 1: Rainy periods over the Nalzba basin in February·March 1992 . day 37, 6 February; day 40, 9 Februuy day 43, 12 February; day 46, 15 Fcbruuy day 73, 13 March; day 76, 16 March Fig 1: Rainy periods over the Naizin basin in February-March 1992. 90 Figs 2: Phase differences obtainedfor an altitude of ambiguity of+ 2Im, before and after unwrapping based on a simple model. Before unwrapping, apoor estimation of the altitude of ambiguity provokedfringes parallel to the satellite's direction of motion. The hydrographical network correspond to a negative contrast since, because of the smoothing performed on the DEM, the thalwegs appear higher than they actually are and thephase correction due to the relief is too large. The thalwegs would correspond to a positive contrastfor a negative altitude of ambiguity. 91 Figure 3 : Specific effects due to the presence of fields Zoom (30km*20km) on phase difference image PHA-2 and amplitude images (6 and 12 February,). Phase is given before (1) and after (2) unwrapping, the geometric structures corresponding to the fields are very clear. These effects are more tenuous in the absence of a contrast in humidity. Bare rough soils are visible on the amplitude images, (3) 6 february and (4) 12 february. No important specific phase shitfs are associated to this bare surface. Figure 4: Representation in grey levels of the topography; lower values at 20 meters (dark) and highest values at 309 meters (white). 92 Figures 5a and 5h: Phase changes structured by the hydrographical network Fig 5 a: Phase image PHA-1 (6 and 9 February) shows the phase changes structured by the network in the center and to the east (The darkening is equivalent to a reduction in the antenna to ground distance). The fact that such structuring is not found on all the data and is not linked to the altitude of ambiguity excludes an artefact related to DEM imprecisions. Fig 5 b : Phase image PHA-2 (6 and 12 February) shows, less clearly, the same type of effects that are seen in Fig 5 a. 93 Phase rotation - . ·1 !~j j 1--- __I ___.~ i ' : 1 ir-1' ----i-' --·-~_!_ L_-1~ i l i : I ! -t ! I t II I ----·---- 0 l ~ t r =-· -=- "" • • --i· -I I I I i I 1 i j , I Sinking in mm e ~ e ~ .,_M •..__..,.__..,.4 ,_.__.•....•.• --..•.....•---•....•...~ ..•.~...,.~~~~---. "'I : ; 8 -r---...~--..---··-- .•. I I I j • i' El--_,_;.-----.--· __ ,,I •• =- I I f!l__~ J__ ~r--- i i LJ.-----.---; ! I I I I 1• !I r' 94 Phase rotation ~~ ~ IJQ ~ ::::'. ~""'I •. s::i ~ ""~ .'-I l ·.··... ' i (") ::-:, ~~ ::: ~ §t~-··············-···· !--··· ". J IJQ (") ·'"-t·--···-··-:------·-- ~ :::::". ~;;:: r;;· § ; • l i :::: ~ 5 ~ §" ·········-···-···"'": ~----···-·-······-...... •~------..~-···· - , ······-. j ! ~..~ -· ~"" .•. : ~~::: I -.~. i 'g ~ ~ ~::::'. s::i...• Hoi ...• .... -········· : ~ 0 0-· §~· ...• ::: -----····-+------·-- ---~=~·~.:·:~~---r· - -....- .... . ~ s::i •• ~ ::: .....• ~ ; i ••• 1 ...,'- :; ~ -· e -e "" ···-····--····~.• ~ ::::'. §;-....- - --..--·+· ····· -...... ~ -····-····- :~~;.:__ ·--······-···'-. ·• :::::: ~ ~.. ::: s::i IJQ ~(") ~;;. -rj'e:. ~L.. J __-·······-··J -·_J_~~~~~~~~,~:..----...... 1i ""~ s::i - (")- ! ' ' . .\:. --S.s::i' ~•.•• "~. ...•:::~ il_._ ...____.....-----1....--~------'------~_._~~~...___; ~ g_ ~ "" ~ ~ - <:: ~ ~- ~ Swelling in nun ~~ ~ ""! ;:;· ~ ~ ::::'. .. U' (") S! :::;- 13; ;::;·~ .~ ~ 1? S;;:i Vj J :0 ~ ~ hI i' ;;-. -~ I:=, ~ ~ : 2- iii' ~ ~ i ; -·<:: ~ -I •• s::i i !. e..~ ;::s •• .....~~ ..•.0 ~ a ~ ::::'."" 1i ~s::i ~(") -.•...• S! .i ' ········----T··-- .. v.----·- H 0 ,~ - , (l <:: .: :· ~· ; '" I:: : . ~-;- . . '\• . :0 s::i ~ Il i i ·· .~ll ~ ' : ~- - ' . .! ~' I ; ·~ ~ ' ' ' ' ~ I 95 PHASE SH:FT (d~) . ~~(,,)c...;)J\). .Joo [,J'1 0 (11 0 (J"I 0 0 .., ~ ~ ~ to i::::i (') ~ (') i::::i •...•. (I:> ""'l ~~ ' (I:> )> i::::i... """' -I ~ rn "=l~ ::n i::::i - ...., (I:> ""'l •...•.c •...•.i::::i -·a ....,::s -·::s ? (') (I:> i::::i... ~ PHASE SHIFT (deg) PHASE SHIFT (deg) I . I I . I ' A (J) ~ ....• (A) ~ -..i m (,.) ...•. c,., C> 0 Ul 0 N :r·l··-r0bl···1J CA) • .,, U'I ii U1 1-· :::c • :0 m m ocn ·- ·- ·- o en c: c mz .....• z-...!m 0 () -< - -< c;Q;l .'5 (X> :::r: . i::t:~"'.•• ,. - ":,"i N I ,. .• -..; N ' -.·~ . . •. •.• i~...._...U;) - U'.) - I ...... •. r 0 ....•...• .....••. N "ti c:r :::tia- 0- N 97 Land Applications using ERS-1/2 Tandem data Urs Wegmuller GAMMA Remote Sensing AG, Thunstrasse 130, CH-3074 Muri BE, Switzerland tel. +41-31-9517005, fax +41-31-9517008, e-mail:[email protected] http://www.primenet.com/-gamma/gamma.html Charles L. Werner GAMMA Remote Sensing AG, Thunstrasse 130, CH-3074 Muri BE, Switzerland tel. +41-31-9517005, fax +41-31-9517008, e-mail: [email protected] http://www.primenet.com/-gamma/gamma.html Abstract The potential of ERS-112 data for landuse classification and change detection and monitoring is discussed. For repeat-pass interferometer systems, such as the ERS satellites, low interferometric correlation indicates random dislocation of the individual scatterers between the two acquisitions of an interferometric image pair. This additional information significantly improves the potential of SAR data for landuse classification, change detection, and the retrieval of geophysical and biophysical parameters. Results obtained with ERS-1/2 Tandem data will be presented and compared to results of the ERS-1 Commissioning and Ice Phases. Finally, a short overview on GAMMA Remote Sensing's products and user services will be given. Keywords: SAR interferometry, landuse classification, change detection. SAR processing and interferometricprocessing software, GAMMA's user services. Introduction Based on ERS-1 data acquired during 3-day repeat orbits a good potential of repeat-pass SAR interferometry for land applications such as landuse classificationand change detection was identified(Wegmuller et al., 1995a, 1995b). It was found that the interferometric correlation is not just a measure of the phase noise of the interferogram but a valuable source of informationon scene properties. With the ERS-1/2 Tandem missionrepeat-pass SAR data useful for interferometric analysisbecame widely available:the I-day acquisition time interval Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Sept. · 2 Oct. 1996 (ESA SP-406,Vol. II, December 1997) 98 and the precise orbit control which allowed to almost permanentlyobtain short interferometric baselines for Tandem pairs resulted in data ideal for interferometry. In addition the 35-day repeat-orbits of the two satellites allowed to achieve nearlyglobal coverage. With the available data interferometric techniques can now be widely applied. In this contribution it will be shown that Tandem data is very useful for landuse classification.In addition potential and limitations of Tandem data for change detection and monitoring will be discussed. The Interferometric Correlation The interferometric correlation is a measure of the phase noise of the interferogram. It depends on sensor parameters (wavelength, system noise, slant range resolution), parameters related to the imaginggeometry (interferometric baseline, local incidence angle), and target parameters. Volume scattering and temporal change (i.e. random motion of the scatterers, change of the scatterers) decrease the interferometric correlation. The system and geometry dependent effects are pretty well understood and can be controlled by appropriate interferometric processing, as long as the system parameters are within a certain interval. The baseline dependence of the interferometric correlation, for example, may be eliminatedin many cases by common spectral band filtering of the range spectrum. The dependence of the interferometric correlation on target parameters is used to retrieve information on the target characteristics. The informationavailablethrough the interferometric correlation is complementary to the informationcontent of the backscatter coefficient as investigated and documented by Wegmulleret al. l 995a, l995b. The maximumlikelihoodestimator used to estimate the interferometric correlation tends to be biased at low interferometric correlation values. For an unbiased estimation a large estimation window is required at low interferometric correlation values as shown in Figure 1. 1.0 I '' • '''I fd o.a - ~ wamr i ._ ~ra~ ~ 0.6 - -- -- urban ,.0 ·..·· ··-.. -- fiel"""'~ JS!- 04. -··--,.v, ! - ····... ~ 0.2 ·····-.··.... ------0.0 . .. ···' 10 100 number of looks Figure 1: Dependence of interferometric correlation estimate on number of looks used in the estimationfor areas representativefor law to high interferometric correlation. 99 Landuse Classification The landuse classificationscheme presented is based on the normalized interferogram and the two backscatter intensity images of the interferometric image pair. In a first step these data are used to estimate (1)the interferometric correlation, (2) the average backscatter intensity, (3) the backscatter intensity change, and (4) the texture of one of the backscatter images. For a wide applicabilitythe classificationis done on a per pixel level. Therefore, appropriate estimation schemes and filtering are required to retrieve useful estimates at per pixel level. As mentioned above the estimation of the interferometric correlation requires a sufficient number of looks. As a compromise between maintaininga high spatial resolution and accurate estimation an estimator with adaptive estimator window size was used. In a first step the interferometric correlation was estimated with a fixed, relative smallestimator window size. In a second step the estimator window size was determined based on the first correlation estimate. To estimate lower correlation values larger estimation windows are used. Usually, the estimator size was varied between 3 x 3 and 9 x 9 pixels of a 5-look interferogram. In addition, a weighting function, decreasing linearlywith increasing distance, was applied in the estimation. For test areas representing areas of low to high interferometric correlation the adaptive estimator is compared to non-adaptive estimation with estimation window sizes of 1 x 1 and 9 x 9 pixels of the 5-look interferogram (Figure 2). • I R an b e WatC'f ..•' tmest I' ulban I j I I fidd& I I . ,-·.... ,.• - . I I I ' ·...... •··.,~• I·. "•'\,, ••:~.:~ 7°-T~~:.:··'."'"...\ 1.0 a.o o.s 1.0 ir'ltcrf.correlation intcrl. corralation Figure 2: Comparison of histogram of estimated interferometric correlationfor areas representativefor low to high interferometric correlation using (a) non-adaptive 3 x 3pixel (b) non-adaptive 9 x 9pixel and (c) adaptive 9 x 9pixel estimators. In order to reduce speckle effects and obtain a backscatter intensity estimate at the pixel level which is representative for the ensembleaverage of the area around that pixel the two registered SAR images of the interferometric pair are averaged. MinimumMean Square Error (MMSE) filtering, as described by Frost et al., 1982, was then applied to the averaged image. Typically,the filter was applied to areas of 7 x 7 pixels of a 5-look image (5 azimuth looks). For the same test areas as used for Figure 2 the histogram ofMMSE filtered backscatter intensityis compared to average filtered data (Figure 3). 100 ' I ' 0.5 I ' ' ' I a - b - c; ~ 0.4 All a W1'1ter .B -- lcresl ~ 0.3 -- urban 1S tiolds .- '.-·'. ~ ()2 ' I ~. ~ ,.·. •' ',.• .·_.·-.:.. ', : -~.• i I / •.. ~ o I ·r\ ....• .'• -, ..- I ...__.,;:.:~-. .·.. ·:··:.- - 1.:.',••. ,··· ·.·:-·:.l .··;-:,, '···~·-··- D.O -·-2ri -10 0 -20 -10 0 -20 -10 n irih!Jll&:ily [dB] iulerrsity [dB] int.11&.ily [dB] Figure 3: Comparison of histogram of estimated backscatter intensity for areas representativefor different landuse classes using (a) averagefiltering with 1 x 1pixel window, (b) averagefiltering with 7 x 7pixel window, and (c) .MMSEfilter with 7 x 7pixel window. The backscatter intensity change between the two images of the interferometric pair is defined as the absolute value of the ratio between the two images expressed in the logarithmic dB scale (1) The brackets { } stand for the ensemble average. Due to the speckle of the individual images it is essential to average sufficiently before the calculation of the ratio. Usually, at least 9 x 9 pixels of the 5-look images were used in the averaging step. In addition, a weighting function, decreasing linearly with increasing distance, was applied in the estimation. In the classification change is only distinguished versus no change. Therefore it is reasonable to take the absolute value of the ratio in dB. If the change is further interpreted this is of course not reasonable as the capability to distinguish backscatter decrease from backscatter increase is wasted. For the same test areas as used for Figure 2 the histograms resulting from different ratio estimators are compared (Figure 4). 0.4 a b e --all i 0.3 -- watCI' £ !.1 ., ·!B torost 1:1 -- urban I\ I~ i 0.2 j' fields /I'' ~ ,I I i i I ,' I ' ~ 0.1 . I 1, _.,,...."".""!•1 ,__ ·• L--': . /f •;'•,- -·s:....- --~"'...,..:.-- - J •.4.·~i-I ~·..:.~: .... D.O "--· • .• -- --' -- -' -. -10 0 10 -10 0 10 -10 0 10 irilE"n&ily raLio [dB) i111AN1sityrill.io (dB] inl.enaily ralio [dB] Figure 4: Comparison of histograms of estimated backscatter intensity ratiosfor areas representativefor different landuse classes using (a) ratio of unfiltered 5-look values, (b) after averagefiltering with 9 x 9pixel window, and (c) after averagefiltering with 9 x 9pixel window and application of linearly decreasing weightingfunction. 101 The texture of the backscatter image is defined as the ratio between the standard deviation and the average (2) Again, the estimation of the ensemble averages requires sufficiently large estimator windows. It turns out that extremely strong scatterers in an image have the unwanted effect the entire high texture is obtained over an area corresponding to the size of the estimator window. This effect can be avoided to some degree if the texture estimation is followed by filtering with a moving average filter of larger size than the texture estimator. As an example we used 15 x 15 pixels of the 5-look image for the initial texture estimation with a subsequent 25 x 25 pixel moving average filtering. In addition, weighting functions, decreasing linearly with increasing distance, were applied in the different steps. For the same test areas as used for Figure 2 the histograms resulting from different texture estimators are compared in Figure 5. I a b c = 0.15 ~ !1 f 0.4 .Ii; i-- 0.3 "-I '•:,! • 10.2 1r,.,.11·· b I 't J \ .•..• , -, - 0.1 11.: ~ :-.. ·, I• 0.0 i ;,-, - ..•• • •. • -10 0 10 -10 Q 10 r.w••••(dB] &..xl.u,. (dB) Figure 5: Comparison of histograms of estimated backscatter intensity texturesfor areas representativefor different landuse classes using (a) texture estimate using 5 x 5 pixel window, (b) texture estimate using 15 x 15pixel window, and (c) texture estimate using 15 x 15pixel window with application of linearly decreasing weightingfunction. Based on the interferometric signatures a simple landuse classification algorithm was developed. A hierarchical decision tree algorithm using the criteria listed in Table 1 allowed to generate a landuse map. In order to account for the specific conditions under which the data was acquired the classification scheme needs to be slightly adapted. Table 1. Decision rules of landuse classification algorithm. The criteria are applied hierarchically, in the order as listed The value ranges used are indicatedfor the interferometric correlation ("'{), the average backscatter intensity of the two images rcr0 ), the backscatter intensity change between the two images, and the texture of thefirst backscatter image. 102 For comparison results based on ERS-1 data acquired during 3-day repeat orbits in November 1991 are used. The November 1991 data is ideal for the presented approach because of the relatively short 3-day acquisition time interval, the short 58 m baseline, and the acquisition during the winter season when the forest can best be distinguished from agricultural fields because the fields are bare or only sparsely covered with vegetation. The RGB color composite of the interferometric correlation (red), the backscatter intensity (green), and the backscatter change (blue) is shown in Figure 6, and the resulting landuse classification in figure 7. These data are compared with results achieved using Tandem data in November 1995, April 1996, and July 1995. The exact dates and baselines are listed in Table 2: Table 2: Bern (CH) test site: Dates and baselinesfor interferometric pairs used Baseline t Sensors Dates [m] [days] ERS-1 & ERS-1 24.11.91 & 27.11.91 58 3 ERS-1 & ERS-2 26.11.95 & 27.11.95 138 1 ERS-1 & ERS-2 14.4.96 & 15.4.96 93 1 ERS-1 & ERS-2 9.7.95 & 10.7.95 27 1 ERS-1 & ERS-2 4.6.95 & 5.6.95 117 1 The RGB composite (Figure 8) and the landuse classification(Figure 9) of the November 1995 Tandem data confirm the expected usefulness of Tandem data for landuse classification. Unlike with 3-day repeat data the approach lead to reasonably good results during spring (Figure 10,11) and summer (Figures 12,13) period, too. The shorter acquisition time interval results in an increase of the interferometric correlation of fieldswith grass or crops, improving the potential to distinguish fields from forest. For the November 1991 data the result was validated with a conventional forest map. An accuracy for the forest/non-forest classificationof around 90% was reported by Wegmuller et al. 1995c. Classification accuracy of the same order may be expected with Tandem data. These examples allow to conclude that landuse classificationbased on interferometric signatures from ERS-1/2 Tandem data is feasible and has a high potential not only because of the quality of the results which may be achieved, but also because of the good spatial and temporal coverage with appropriate image pairs. Figure 6: Bern (CH), ERS-1, 2-127 Nov. 1991: RGB composite of interferometric correlation (red), backscatter intensity (green), and backscatter change (blue). Color cooing used: • WU.le! 11 urban area • forest l (dcnsc/corutcrous) 8 forest 2 (opcu/decjduous) • spars;.~v ~~~ctatitu1 m; moisture chanae.. /frec.1i11Q... • mechanical cultivarion layover UH,'U Figure 7: Bern (CH), ERS-1. 2-127 Nov. 1991: Landuse classification based on SAR interferometric signatures. 104 Figure 8: Bern (CH). ERS-1 2, 26 27 Nov. 1995: RGB composite of interferometric correlation (red). backscatter intensity (green). and backscatter change (blue). Color coding used: I! 'A"lllcI m urhau area • forest l (dense/coniferous) II forest :.'! (open/deciduous) • sparse v•~gc:taLic..in E'. moisture change / frec.7il1Q, • mechanical culnvauon layover WXll Figure 9: Bern (CH). ERS-1 2, 26 27 Nov. 1995: Landuse classification based on SAR interferometric signatures. Figure JO: Bern (CH). ERS-1, 1./ 15 Apr. 1996: RGB composite of interferometric correlation (red), backscatter intensity (green). and backscatter change (bluet. Color coding used: • water m urban area • forest l (dcnsc.conitcrous) E 11par~'. \'~'.l".t'tatic..111 •E moi~n1~chanae / fre.c7inii mechanical cultivunon • ht)'{'"" er urea Figure 11: Bern (CH). ERS-1, ]./ 15 Apr. 1996: Landuse classification based on SAR interferometric signatures. 106 Figure 12: Bern (CH), ERS-1, 9 lOJul. 1995: RGB composite of interferometric correlation (red), backscatter intensity (green), and backscatter change (blue). Color coding used: • WI.LI.er II urban area • forest l (dcnsc/comtcrous) • for, ..-sL":! (up;...,1/dociduo~• • ~tr.;;.~ v~:_gct:uic.11 m moi~tu~ chanae / free.1i11~ • T11«h:m1Calcuhivanon layover 111..:u Figure 13: Bern (CH), ERS-1, 910Jul. 1995: Landuse classification based on SAR interferometric signatures. Change Detection and Monitoring Repeat-pass SAR interferometry is very sensitive to temporal change. To characterize the change the interferometric correlation and the backscatter intensity change are used. The ground resolution and the local incidence angle of corresponding areas in the image pair are 107 identical. Therefore, the backscatter change can be reliably estimated, even without applying elaborate radiometric calibration algorithms that take into account the local pixel size and terrain slope. The backscatter intensity is a function of the geometry and permittivity of the scatterer. Different types of temporal change can be distinguished with repeat-pass SAR interferometry as reported by Wegmuller et al., l 995a, l 995b. Coherent scatter intensity change (high interferometric correlation together with a significant positive or negative backscatter intensity change) results from an unchanged scatterer geometry in combination with a permittivity change. The permittivity of soil and vegetation is dominated by the high permittivity of liquid water. Coherent backscatter intensity change is observed as a result of changing soil moisture, freezing, and thawing. Decreases of about 3 dB as a result of freezing were observed. Smaller changes occur due to changes in soil moisture and plant water content. Changing scatterer geometry causes a loss of coherence. Mechanical cultivation (ploughing, furrowing, harvesting) is an example for incoherent change. Quite often a loss of interferometric correlation is observed in combination with a backscatter intensity change. Nevertheless, it also occurs that no backscatter intensity change is observed in spite of geometric changes. In particular, for the case of random dislocation of the individual scatterers in a resolution cell the interferometric correlation is reduced without backscatter intensity change. Such a behavior is typically observed for forest stands and other dense vegetation (sugar beets, corn, potatoes). If only a part of the scatterers within a resolution cell move or if the displacements are small compared to the radar wavelength partial coherence is maintained. This behavior is observed for canopies with a significant contribution of ground surface scattering. For certain fields a decay of the interferometric correlation due to vegetation growth was observed. If multiple acquisitions are available over a test-site interferometry allows to monitor change (Wegmuller et al. l 995a. To monitor agricultural fields information is required at intervals of one to two weeks. In this respect ideal data was acquired by ERS-1 during the· 3-day repeat-orbit phases. The interferometric correlation of consecutive interferometric pairs allows to monitor change on agricultural fields. An example is shown in Figure 14 for an agricultural area in Middle Zeeland (NL). The temporal behavior of the fields together with knowledge about local crop calendars allows to map crop types and to detect key processes such as sowing and harvest for the different crop types. As discussed by Wegrnuller, 1996, this technique also allows to improve the potential of SAR data for hydrological applications as the interferometric correlation allows to distinguish backscatter changes resulting from geometric change, i.e. changing surface roughness and vegetation cover changes, from permittivity change, i.e. soil moisture change and freezing. Change occurring between the two data acquisitions of Tandem pairs can be detected as described. Nevertheless, monitoring applications are restricted with the acquisition scheme of the Tandem mission. The 1-day I 35-day acquisition intervals do not allow a continuos monitoring of change on agricultural fields due to the too long 35 day interval. Except for bare and very sparsely vegetated fields the interferometric correlation decreases too much after a 35 day period. In addition the interferometric baselines of the 35 day pairs are very often too large for this type of interferometric analysis. As a result no or very little additional information is obtained from the interferometric correlation. 108 Of course multi-temporal composites of the interferometric correlation of Tandem pairs may be generated and interpreted, for instance to study seasonal changes in the land cover. An example combining June 1995, November 1995, and April 1996 data is shown for the Bern test site in Figure 15. Low correlation is permanently observed over water and for certain forests. Over agricultural fields and certain forest stands (predominantly deciduous forest) seasonal changes related to changes in the canopy are observed. Figure 14: Change monitoring on agriculturalfields in Middle Zeeland (NL) winter 1994 with ERS-1 repeat-pass SAR interferometry. The multi-temporal image of the interferometric correlation (red: 6. 15. Jan., green: 15. 27 Jan., blue: 27. Jan. 5. Feb.) allows to monitor farming activities. 1 ( )Ll Figure 15: Multi-temporal image of interferometric correlation over Bern (CH). based 011 Tandempairs on .J. 5. Jun. 1995 (red), 26. 27. Nov. 1995 (green) and l .J. 15. Apr. 1996 (blue). GAMMA Remote Sensing Products and Services GAMMA Remote Sensing Research and Consulting AG (GAMMA) is a Swiss corporation (Aktiengesellschaft - AG) located near Bern, Switzerland. It was founded in January 1995 by Charles Werner and Urs Wegmuller. The objectives of GAMMA are to conduct research studies and provide consulting services in the field of microwave remote sensing. In the following GAMMA's products and services in SAR and SAR interferometry are shortly introduced. The key personnel of GAMMA have extensive experience in remote sensing techniques, theoretical and empirical modeling, and application development. This experience has been gained during their work at the Universities of Bern and Zurich. Switzerland, and the Jet Propulsion Laboratory, Pasadena, USA. Products GAMMA provides licenses for its SAR processing and interferometric processing software. The software is of high quality, portable, efficient (patch processing), user-friendly, and reliable (used at leading institutes). Overall, the design philosophy has been to achieve high quality processing of the data while still permitting processing of the data on a workstation computer in a reasonable amount of time. The software was successfully applied to spaceborne (SEASAT, ERS- 112, JERS, SIR-C, RADARSAT) and airborne data. The SAR Processor is a modular software package to process synthetic aperture radar images from SAR raw data. The design philosophy was to achieve accurate range-Doppler processing of the data (phase conservative, radiometric calibration, well defined geometry, autofocus, motion compensation) while still permitting processing of the data on a workstation computer in a reasonable amount of time. The processor consists of a suite of ANSI-C programs. The ANSI-C language was chosen for its portability and efficiency in 110 processing oflarge data sets such as full frames ofERS (lOOkmx lOOkm).The main modules are prefiltering and range compression, autofocus, azimuth compression and multi-looking. For the processing of airborne data a motion compensation module is available. The interferometric processing software includes the main steps of interferometric processing, i.e. baseline estimation from orbit data, precision registration of interferometric image pairs, interferogram generation (including common spectral band filtering},estimation of interferometric correlation, removal of flat Earth phase trend, adaptive filtering of interferograms, phase unwrapping, precision estimation of interferometric baselines, generation of topographic height, and rectification and interpolation of interferometric height : and slope maps. The display of the final and intermediate products is supported with display programs and programs to generate easily portable images in SUN rasterfile format. Processing related parameters and data characteristics are saved as text files that can be displayed using commercialplotting packages. Services GAMMA supports customers in the development of their applications. GAMMA conducts SAR processing, interferometric processing, and data analysisfor customers, providing end to end support starting with the selection of appropriate data to the interpretation of the resulting SAR images and interferometric products. Conclusions SAR interferometric analysisof ERS-1 SAR data pairs acquired during the 3-day repeat-orbits of the commissioningand ice phases have shown a good potential for landuse classificationand change monitoring. The main goal of this contribution was to investigate the usefulness ofERS-112 Tandem data for these applications. For landuse classificationTandem data has a very good potential. The shorter time interval and the permanently short baseline lead to good results with a much wider applicabilityof the approach due to the different acquisition mode. ERS-1/2 Tandem data is less useful for change monitoring. Change occurring during the I-day acquisition interval of the Tandem pairs can be detected. Nevertheless, continuous monitoring is not possible for most vegetated land surfaces because of a too strong decrease of the interferometric correlation during the 35 day interval to the next data pair. In addition the interferometric baselines of the 35 day pairs are very often too large for this type of interferometric analysis. The main differencesbetween application ofERS-1 repeat-pass data and ERS-1/2 Tandem data for landuse classificationand change detection were summarizedin Table 3. In a quick overview over Gamma Remote Sensing'sproducts and services it was announced that Gamma uses its know-how and software to support customers in their research and application activities in SAR and SAR interferometry. Table 3: Differences between ERS-1 repeat-pass data and ERS-1 2 Tandem data and consequencesfor landuse classification, change detection, and change monitoring. ------·-----~· 111 Acquisition Mode ERS-1 Commissioning and ERS-112 Tandem data Ice Phases Acquisition interval multiples of 3 days 1 day, 34,35,36 days Coverage partial coverage almost global coverage Baselines varying baselines short baselines (< 300 m) Differences up to 300 Hz in Squint angle differences small Doppler Centroid slightly increased correlation over forest Interferometric correlation increased correlation over crops and grass good potential good potential Landuse classification limited applicability wide applicability Change detection during good potential good potential acquisition time interval Change monitoring good potential no continuos series Acknowledgments ERS-112 SAR raw and slc data provided under ESA-A02.JRC101 and BSA/Contract 11740/95/NL/PB(SC). This work was supported by BSA ESTEC and the Swiss Federal Office for Education and Science. References Frost V. S., J. A. Stiles, K. S. Shanmugan, and J. C. Holtzman: A model for radar images and its application to adaptive digital filtering of multiplicative noise, IEEE Trans. Pattern Analysis and Machine Intelligence, Vol. PAMI 4, No. 2, pp. 157-165, 1982. Wegmuller U., C. L. Werner, D. Nuesch, and M. Borgeaud, 1995a Land-surface analysis using ERS-1 SAR interferometry, ESA Bulletin, No. 81, pp. 30-37. Wegmuller U. and C. L. Werner, 1995b: SAR interferometric signatures of forest, IEEE Geosci. Remote Sensing, Vol. 33, No. 5, pp. 1153-1161. Wegmuller U., C. L. Werner, and D. R. Nuesch, l 995c: Retrieval of geophysical and biophysical parameters using ERS SAR interferometry,, Final Report, BSA Purchase Order 143 061, European Space Research and Technology Centre (ESTEC) of the European Space Agency (BSA), 2200 AG Noordwijk, The Netherlands, September, 1995. Wegmuller U., 1996: The potential ofERS SAR interferometry for hydrology, in Progress in Environmental Remote Sensing and Applications, Parlow (ed.), Balkema, Rotterdam, (ISBN 90 54 10 5984), pp. 319-324. 113 An integrated methodology for DEM computation through the fusion of interferometric, radargrammetric and photogrammetric data Issam Tannous & SYSECA, 66-68, Avenue Pierre Brossolette, 9224 7 - Malakoff Frederic Le Goff Cedex, France issam. tannous@syseca. thomson. fr Abstract This paper presents a methodology for the fusion of image-based 3D informations, such as interferometric ERS data, stereoscopic (or radargrammetric) ERS data and stereoscopic (or photogrammetric) SPOT data, for Digital Elevation Model (DEM) generation. With the availibility of several imaging sensors exhibiting high cartographic capability, the users which require a DEM of a scene, either by buying it from production companies or by generating it if they have the adequate software, are confronted with the problem of selecting both the sensor (i.e. ERS, RADARSAT, SPOT, ...) and the computational technique (stereoscopy or interferometry). The observed scene being unique, instead of having several possible individual DEMs (i.e. ERS interferometric DEM, SPOT stereoscopic DEM, ...) as commonly found in practice, it seems natural to obtain only one single DEM whatever the number and the diversity of the available source data. The principle is to combine all the data in order to get the benefit of each of the available sensor. For exemple, the interest of combining pairs of interferometric and stereoscopic images is to provide a DEM being more operationnal in terms of the: * Density of reliable informations : - to get less "holes" than with the individual DEMs (stereoscopic information should provide elevation data in the holes of the interferogram and vice-versa); - stereoscopic information will help in phase unwrapping for areas of the interferogram where several rounds of 2Pi occur between two consecutive fringes. * Accuracy : the availability of several measures of the elevation for a given point (several observations) should increase the accuracy of the fused DEM with respect to the individual DEMs. In this context, we have designed a methodology for DEM generation through the fusion of image-based 3D informations. These informations are pairs of interferometric images, pairs of stereoscopic images (radar or optical), existing DEMs and GCPs. The block diagram of the corresponding Processing Chain is shown hereafter (for ERS and SPOT data). The Processing Chain permits to generate first the 3D informations as interferograms and parallax map respectively from the interferometric and the stereoscopic images. The Interferogram Computation Process is based on an original registration process which is fully automatic and very simple. It uses a physical warping function which is deduced from the sensor imaging geometry. By using the ERS Precise Orbit data, the physical warping function permits the Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Sept - 2 Oct. 1996 (ESA SP-406, Vol. 11, December 1997) 114 registration of the pair of interferometric images automatically with enough accuracy so that the phase differences can be computed on a pixel-to-pixel basis (using a "multi- looking" process). The 3D Fusion Process consists of computing the elevation map of the scene by means of all the available 3D informations (i.e. interferograms, parallax map, existing DEM, GCPs). A confidence criterion is set to each 3D information as the standard deviation of the error on the information (ex: the standard deviation of the error on the interferometric information is deduced from the corresponding coherence value). The implementation of the 3D Fusion Process requires to have a single methodology in order to estimate the relief by mean of each of the individual techniques. One can show that a Bayesian formulation is the optimal framework for estimating the elevation data in a fusion approach. The relief estimation is then a bundle estimation problem where one searches for the optimal elevation which minimizes a criterion deduced from the Bayesian formulation. All the geometric problems involved in the Processing Chain are solved by using functions based on the physical modelling of the sensors' imaging geometry, i.e.: * registering the interferometric and stereoscopic image pairs, * moving from image coordinates to ground coodinates, * relating elevation value to interferometric phase and parallax values. Note that if only interferometric data are available as input, the Processing Chain permits to generate the interferogram and the corresponding DEM (the fusion process allows phase unwrapping). In the same way, if only stereoscopic data are available as input, the Processing Chain permits to generate the parallax map and the corresponding DEM. The feasibility of the 3D fusion methodology has been demonstrated with ERS (SLC products) and SPOT (Panchromatic products, level IA) images over the region of Metz (Eastern city in France) which exhibits a moderate relief (maximum height difference of 300 meters). Several DEMs have been computed using: * only the pair of interferometric ERS images with GCPs, * the pair of interferometric ERS images plus a rough DEM with GCPs, * only the pair of stereoscopic (or radargrammetric) ERS images with GCPs, *the pair of interferometric ERS images plus the pair of stereoscopic ERS images with GCPs, * only the pair of stereoscopic SPOT images with GCPs, the pair of interferometric ERS images plus the pair of stereoscopic SPOT images with GCPs. The resulting fused DEMs have been assessed by comparison with both the individual DEMs (interferometric and stereoscopic DEMs) and a reference DEM of the test site. The results are in accordance with the expected gain of operationnality (i.e., higher density of reliable informations, better accuracy). 115 The UCL 3D Image Maker system for automated differential SAR interferometry Mark Upton, Jan-Peter University College London, Gower Street, London, WCIE 6BT, Muller UK [email protected] http://www.ps.ucl.ac.uk/-upton Andy Smith Phoenix Systems Limited, UK [email protected] Abstract The authors are developing a SAR interferometric processing system. This system generates the followingproducts from either raw or focussed SAR input: Focussed complex image with amplitude and phase suitable for interferometry; Amplitude image; Interferogram; Coherence map; DEM (Digital Elevation Model) or ellipsoid flattened interferogram; Differential interferogram; Unwrapped flattened or differential interferogram; Geocodedversions of these; DEM; Map of vertical surface displacement. Under development are the followingproducts: Shadow map; Layover map; Map of 3D surface displacement. Phoenix Systems has provided two of the components, PulSAR and DRAIN. PulSAR is the SAR processor, that focussesraw images. DRAIN is the interferometry processor, that generates interferograms and optionally performs ellipsoid flattening. The other components have been developed and are being developed at UCL (University CollegeLondon), and basically concern the phase unwrapping, and the geometric aspects, such as geocodingand the conversion of unwrapped phase to vertical or 3D surface displacement. A particular concern has been to simplify the interface to the user as much as possible, and so common tasks are reduced to a single command, driven by automatically generated parameter files. The interface between each DRAIN output product and the UCL software has been reduced to a single parameter file, understood by both items of software. The system has been applied to many different datasets now, featuring different types of terrain, and ascending and descending image pairs. The results are promising, and as far as georeferencing accuracy is concerned, it appears that if one uses DLR's PRC (Precise) platform state vectors, then one can probably do without GCPs (Ground Control Points). The system has been delivered to NPA (NigelPress Associates)for use in their CivlnSAR project. Keywords: SAR, lnterferogram, Differential interferogram, Phase unwrap, Geocode, DEM Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Sept. - 2 Oct. 1996 (ESA SP-406, Vol. II, December 1997) 116 A Workstation for Spaceborne Interferometric SAR Data M.W.A. van der Kooij, B. Atlantis Scientific Systems Group Inc., 1827 Woodward Armour, J. Ehrismann, H Drive, Ottawa, Canada K2C OP9, tel: 613-727-1087. fax: Schwichow and S. Sato 613-727-5853 [email protected] Abstract In the last decade SAR interferometry (InSAR) has emerged as a very promising mapping technique that has the potential to routinely provide quantitative information on height, deformation and change, and allowing a full geocoding of SAR imagery. A growing number of applications have become feasible or are researched. Atlantis has developed a commercially available InSAR workstation that provides both research and routine users of spaceborne InSAR data the opportunity to process and interact with data of varying quality and from various platforms and sources. It contains the full chain of steps necessary for interferometric processing e.g. image coregistration, interferogram generation, coherence products generation, interferogram enhancement, phase unwrapping, DEM I deformation map generation and geocoding.The InSAR workstation is based on the ERGOvista image analysis software. The workstation approach features a state of the art interferogram enhancement and filtering. The dataset geometry is characterized in terms of master and slave state vector and orbital propagators. Toolkit functions provide the opportunity to handle inaccurate state vectors. Several phase unwrapping algorithms have been implemented including the Atlantis patented algorithm. Toolkit functions are available to compensate for potential data quality limitations. During the workshop the details of the processing approach and workstation setup will be demonstrated. Results and examples of interferometric products will be shown using JERS-1 interferometric SAR data of the Great Hanshin (Kobe) earthquake and ERS-112and RADARSAT data over mountainous terrain. Proceedmgs of the 'Fringe 96' Workshop on ERS SAR Interferometry, lunch, 30 Sept.· 2 Oct. 1996 (ESA SP-406. Vol. II, December 1997) 117 Comparison of Repeat Track Interferometric Correlation Signatures from ERS-1, ERS Tandem, SIR-C, and JERS-1 Charles L. Werner, Scott Hensley and Paul A. Rosen Jet Propulsion Laboratory 4800 Oak Grove Drive Pasadena, California, USA 91109 [email protected] http://www.jpl.nasa.gov Urs Wegmueller Gamma Remote Sensing AG Thunstrasse 130 CH-3074 Muri/Bern, Switzerland [email protected] http://www.primenet.com/-gamma Abstract The interferometric correlation coefficientis a function of frequency, baseline geometry, temporal separation, scatterer geometry, and sensor parameters such as SNR and resolution. Temporal change and volumetric scattering measured via the correlation is very useful for land use classification, forest studies, agricultural monitoring, natural disaster damage assessment and estimation of topographic errors in interferometrically derived digital elevation models. We compare repeat track interferometric (RTI) data derived from ERS-1, ERS-2ffandem, SIR-C, JERS-1, over a wide range of terrain classes (desert, tropical rain forest, fields, and forest) from regions in USA, South America, and central Europe. Comparison of SIR-C RTI images over the South American rain forest at both L- and C-Band reveal that only at L-Band is there sufficient correlation for forest classification and mapping. Topographic maps generated for an arid region near Fort Irwin in the Mojave Desert using JERS-1, ERS, and SIR-Care presented and compared to a precision reference DEM. The observed height errors are consistent with those predicted from the correlation coefficient. C-Band RTI correlation data from the tandem ERS-112mission are used to generate a land use classification map for a region in Switzerland. This will be compared with SIR-C derived classification utilizing both L- and C-band data for a similar region near Lucens in Switzerland. JERS-1 has demonstrated that long time interval RTI is possible for temperate regions. After the Kobe Earthquake, regions that suffered Iiquifaction decorrelated demonstrating the potential for earthquake damage assessment. The impact of these observations on the design of future interferometric missions is discussed. Keywords: correlation coefficient, classification, interferometry, topography, disaster monitoring Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Sept. - 2 Oct. 1996 (ESA SP-406,Vol. II. December 1997) 119 New methods of phase unwrapping in SAR interferometry Helene Tarayre-Oriot ONERA, DES/STD/IM, 29 Ave de la division Leclerc, 92322 Cedex Chatillon, France, Phone 33 1 46734996, Fax 33 1 46734149 [email protected] Didier Massonnet CNES, 18 av E. Belin, 31055 Toulouse Cedex, France [email protected] · Abstract Digital Elevation Models can be computed from ERS interferometric products. The phase unwrapping step is the main problem of the process. We present a robust-to-noise phase-unwrapping algorithm based on a global analysis of the interferogram. Then, we focus on a technique that eases the phase unwrapping process by using simultaneously several interferograms. Finally, an example of DEM computed thanks to this method is presented. Keywords: Phase unwrapping, Least square optimisation, Digital Elevation Model Introduction Digital Elevation Models can be computed from ERS interferometric products. The phase unwrapping step is the main problem. We present a robust-to-noise phase-unwrapping algorithm based on a global analysis of the interferogram. Then, we focus on a technique that eases the phase unwrapping process by using simultaneously several interferograms. Finally, an example of DEM computed thanks to this method is presented. Phase unwrapping Usually, the phase is unwrapped by computing the actual phase jumps (cf. Goldstein & al 1988, Prati & al 1990). We have developped a method that computes a mathematical model of the unwrapped phase. This model can be used to retrieve the actual phase jumps on the interferogram. We assume that the unwrapped phase is a continuous function ofrange and azimut. We then compute a piecewise linear model of the phase that fits the interferogram (cf. Tarayre and Massonnet 1995). In order to do so, we devide the interferogram into elementary squares and we adjust a linear model on each elementary square using a dichotomic search by minimizing the length of the fringes of the difference image between the elementary interferogram and the elementary model. Once we know each elementary model, we adjust the elementary models one another so that the global model is continuous. This adjustment is done using a least square minimization (Cf. Ancey & al 1994, Guiglia and Romero 1994, Tarayre 1996) on the phase of the summits of the elementary squares. It can be shown that the phase is unwrapped as soon as the difference image between the interferogram and the model (called residue image) does not display any fringes (Cf. Tarayre and Massonnet 1995). The unwrapped phase is then retrieved by adding the residue image to the computed model. Proceedmgs of the 'Fringe 96" Workshop on ERS SAR Interferometry, lunch. 30 Sept. - 2 Oct. 1996 (ESA SP-406.Vol. II. December 1997) 120 Compared to the Guiglia method, Guiglia and Romero 1994, this method has the advantage to reduce considerably the number of points over which the least square minimization is computed. The algorithm can be applied to images with different numbers of looks according to the fringe density. The model is then extended or reduced to the desired resolution. Figure I shows an example of such a phase unwrapping over Mount Etna. Figure 1: Thefirst image is the interferogram that has to be unwrapped. The second one is the model (it is represented with a 2*pi ambiguity so that it can be compared with the actual interferogram). The third image is the difference between the model and the interferogram. We can see that the noisyfringes above the volcanoe have been removed. Somefringes due to overlay below the volcanoe remain. The last image is the integer image corresponding to the unwrappedphase. Multi-interferogram phase-unwrapping The former phase unwrapping algorithm works quite well on noisy data. It can solve some overlay problems but it does not work on data with a lot of fringe discontinuities. Fringe discontinuities occur very often on low ambiguity altitude interferograms. Unfortunately, in order to obtain a Digital Elevation Model with a good precision one has to unwrapp such interferograms. To unwrap these interferograms we use a multi-interferogram scheme. D. Massonnet has shown that it is possible to create a pseudo interferogram of higher ambiguity altitude using a linear combination of two interferograms with integer coefficients (Cf. Massonnet and Yadon 1996). This "pseudo-interferogram" is noisier to unwrapp. We create such an interferogram, we unwrap it with the former algorithm (this algorithm is robust to noise so it is well fitted to such a phase unwrapping). Then, the unwrapped interferogram is used to simplify one of the two low ambiguity altitude interferograms which is unwrapped. This method is suitable to tandem data Figure 2 shows an example of such a multi-interferogram scheme. Two interferograms of 60 and 80 m ambiguity altitude are available. The difference of the two interferograms gives a 1~1 pseudo interferogram of 123 m ambiguity altitude. This interferogram is unwrapped and used to simplify the 80 m ambiguity altitude interferogram which is unwrapped and used to unwrap the 60 m interferogram. Figure 2: Top images: Thefirst and second images represent respectively the 60 and 80 m ambiguity altitude interferograms. The third one is thepseudo interferogram which displays less smallfringes and is easier to unwrapp. Bottom images: Thefirst image is the residue image after unwapping the pseudo interferogram. The second one is the simplified 60 m interferogram. There arefew noisyfringes that can be unwrapped with our algorithm. The last image is the unwrapped 60 m interferogram. Example : Vosges site Having two interferograms of the Vosges area (one of 60 m ambiguity altitude and the other of 80 m ambiguity altitude) we computed the output DEM. The phase unwrapping process has been explained in the last section. We corrected the orbits using 50 altitude points (i.e. points on the interferogram of known altitude). One can show that orbits are best corrected using a large number of non-precise altitude points than with few, but precise, altitude points (Cf. Tarayre 1996). We computed the two DEM and compare them two 200 ground control points collected on 1:25000 map of the area. The best results are obtained with the 80 m ambiguity altitude interferogram. The rms error is 36 m in mountaneaous area and 15 m in valleys (CF Figure3). This error is quite large but it has to be interpreted with caution since we did not have a reference DEM to compute it. 122 Figure 3: Vosges DEM processed with the 80 m interferogram. In order to analyse the artefacts on the scene. we removed from the 80 m interferogram the topography estimated with the 60 m interferogram. The output image (Figure 4) should not display any variation since theoretically the topography is better estimated with a low ambiguity altitude interferogram. Figure 4 displays some phase variations. Therefore. there is an artefact on one of the two interferograms. Having only two interferograms we cannot conclude which interferogram is corrupted by artefacts. Nevertheless, the computation of the rms error between the interferometric DEM and ground control points tends to show that the 60 m interferogram was more corrupted than the 80 m interferogram. Figure s: This image represents the artefact due to the 80 m or 60 m interferogram. Conclusion We have presented a robust-to-noise phase-unwrapping algorithm based on a global analysis of the interferogram. We have focused on a technique that eases the phase unwrapping process by using simultaneously several interferograms. Finally, an example of DEM computed thanks to this method has been presented. Acknowledgement This work has been done at Matra Cap Systemes in cooperation with CNES as part as the ESA project "IN SAR Quantitative Evaluation". References Ancey R.M.,S. Mascle, H. Tarayre 1994: Valorisation des plateformes RSO. cas particulier du deroulernent de franges interferornetriques In : Bulletin SFPT. No 138, pp 65-71, Nov 1994. 123 Goldstein R.M., H.A. Zebker, C.L. Werner 1988: Satellite radar interferometry: Two dimensional phase unwrapping. In : Radio Science,Vol 23, No 4, pp 717-720, August 1988. Ghiglia D.,L. Romero 1994: Robust two-dimensional weighted and unweighted phase unwrapping that uses fast transforms and iterative methods. In: Journal of Optical Society of America A,Vol 11, No 1, pp 107-117, January 1994. Massonnet D., H. Yadon 1996: . Reduction of the need for phase unwrapping In : IEEE Transactions on Geoscience and Remote Sensing,to be published. · Prati C., F. Rocca, A. Guarnieri, E. Damonti 1990: Seismic Migration for SAR Focusing: Interferometrical Applications In: IEEE Transactions on Geoscience and Remote Sensing, Vol28, pp 627-640, July 1990. Tarayre H., D.Massonnet 1995: Noise-robust phase-unwrapping method in radar interferometry. In: Proceedings of Satellite and Remote Sensing 11,Paris,Sept 1995. Tarayre Helene, 1996: Extraction de modeles numeriques de terrain par interferometric radar satellitaire, algorithmie et artefacts atmospheriques. PhD Thesis, MCS 1996. 125 Atmospheric artifacts on interferograms Helene Tarayre-Oriot ONERA, DES/STD/IM, 29 Ave de la division Leclerc, 92322 Cedex Chatillon, France, Phone 33 I 46734996, Fax 33 1 46734149 [email protected] Didier Massonnet CNES, 18 av E. Belin, 31055 Toulouse Cedex.,France [email protected] Abstract The use of spacebome RADAR interferometry has considerably increased since the ERS-1 launch. In this paper we verify the atmospheric refraction hypothesis. We show that the presence of large height variations and a difference of refractive profiles between the two imagings create interferometric artifacts that have to be dealt with in the Digital Elevation Model generation process. Moreover, water content horizontal gradients, clear air turbulences and ionospheric phenomena also create local artifacts on interf erograms. We conclude that one needs more than one interf erogram to solve this problem. Keywords: Interferometry,Atmospheric artifacts, Troposphere, Ionosphere, Refractive index. Introduction The use of spacebome RADAR interferometry has considerablyincreased sincethe ERS-1 launch. In this paper we verify the atmospheric refraction hypothesis. First we explainthe model used. Then we show that a large height variation inside a scene combined with a change of refractive index profiles between the two imagingscreates interferometric artifacts that have to be dealt with in the Digital Elevation Model generation process. Finallywe study the effects of an heterogeneous atmosphere on interferograms. 1-Modelisation 1.1-Ray theory model We model the path delay due to a refractive atmosphere using Snell'slaw (cf Tarayre and Massonnet 1996). The path delay depends on the refractive index profile of the atmosphere and on the angle of incidence of the ray . This refractive index profile depends essentiallyon altitude but it depends, to a lesser extent, on the climaticand meteorological conditions of the atmosphere. Therefore, the differentialatmospheric phase shift depends on the differenceof the refractive index profiles between the two imagingsas well as on the incident angle of the ray. 1.2-Refractive index The atmosphere can be devided into two layers which have different refractive index properties. In the ionosphere (above 90 km), the refractive index depends on the electronic content. The electronic density of the ionosphere depends on the altitude, latitude, solar time and solar Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich, 30 Seot.: 2 Oct. 1996 (ESA SP-406,Vol II, December 1997) 126 activity. It can change temporally, between the two imagings, and spatially inside a scene (cf CCIR 1986). In the troposphere, the refractive index depends on the temperature, pressure and water content. These quantities depend on the climatic and meteorological conditions of the atmosphere at the time of the imaging. These conditions can change from one imaging to the other and inside a scene. We have to study two different kinds of artifacts. The artifacts due to a difference of the refractive atmosphere between the two imagings and the artifacts due to spatial heterogeneities inside a scene. 2-Artifacts due to an homogeneous atmosphere We have seen that the atmospheric path delay depends on the incident angle of the ray. In fact it produces a fringe pattern between near range and far range. This fringe pattern is similarto a fringe pattern due to an orbital error and it is removed when correcting the orbits. It has been shown in Tarayre and Massonnet 1994 and in Tarayre 1996that the residual error due to the interpretation of an atmospheric error as an orbital error is neglectible in most usual cases. Furthermore, misplacingthe satellites leads to an ambiguity altitude error. This error is also neglectible. · Another error exists in mounteneous areas. Most of the refraction occurs near the ground in the layer limit. If the altitude of the ground changes inside the scene the "amount of refraction" crossed by the ray changes. Therefore, when there is a change of the refractive index profile between the two imagings, the path delayvaries with relief In Figure 1, we show such a phenomenon. The first image is a SIRC differentialinterferogram of mount Etna processed with the CNES software. The volcanoe is revealed by contour lines. Two concentric fringes appear near the summit. The second image is a simulation of the refractive index error computed thanks to the real refractive index profiles obtained by sounding balloons. The third image is the residual error after removal of the atmospheric artifact. One can see that the two fringes have disappeared but a larger fringe has appeared. The use of radio-sonde data is not sufficient in this case to remove completelythe atmospheric artifact. 3-Artifacts due to an heterogeneous atmosphere Atmospheric heterogeneities can also produce artifacts. Ionospheric irregularities can be important. The irregularities of the F-layer (around 300 km high) can reach 1.5 fringes in C band. In the troposphere, the refractive index varies essentiallywith water content. Under stable conditions, the wator pressure changes slowly horizontally. It has been observed smooth changes of the order of0.3 fringes on ERS interferograms. Under unstable conditions, the refractive index can change more quickly (cf Panofskv and Dutton 1984) leading to artifacts up to 2 or 3 fringes. Artifacts due to growing cumulushave been observed on the Landers site (Cf Figure 2). Artifacts due to a vertical wind shear have also been observed (Cf Figure 3) Tarayre and Massonnet 1996. l'..,-_ Figure 1: Atmospheric artifact on mount Etna SJRC interferogram. Thefirst image represents a differential interferogram over mount Etna. The relief has been removed. The second image is the simulated artifact and the third image is the difference between the real and simulated artifacts. Figure 2: Fringe pattern due to growing cumulus on the Landers site. Figure 3: Baltimore interferogram. Fringes due to the relief are seen tn the top and bottom of the image. The wavepattern that appears i11blue is due to a vertical wind shear. 128 Conclusion We have shown that the presence of large height variations and a differenceof refractive profiles between the two imagings create interferometric artifacts that can reach one fringe. This phenomenon has to be dealt with when processing data to construct Digital Elevation Models. Moreover, water content horizontal gradients, clear air turbulences and ionospheric phenomena also create local artifacts on interferograms. It seems that these artifacts appear on most interferograms but that their orders of magnitude vary drastically according to the meteorological situation. In order to remove these artifacts on Digital Elevation Models, one needs more than one interferogram to process data. Acknowledgement This work has been done at Matra Cap Systemes in cooperation with CNES. We would like to thank JPL for having provided SIRC images. References CCIR 1986: Propagation dans les milieuxionises. Dubrovnik. Panofsky H.A., J.A. Dutton, 1984: Atmospheric turbulence. John Wiley New York Tarayre H., D. Massonnet, 1994: Effects of a refractive atmosphere on interferometric processing. Proceedings of IGARSS'94, pp 717-719. Tarayre Helene, 1996: Extraction de modeles numeriques de terrain par interferometric radar satellitaire, algorithmie et artefacts atmospheriques. PhD Thesis, MCS 1996. Tarayre H., D.Massonnet 1996: Atmospheric propagation heterogeneities revealed by ERS-1. In : Geophysical Resaerch Letters, Vol23, No 9, pp 989-992, May 1, 1996. 129 Cross-Compatibility of ERS-SLC Products Remote Sensing Laboratories (RSL) A. Barmettler, P. University of Zurich, Winterthurerstrasse 190, CH-8057 Zurich, Pasquali, Switzerland D. Small and D. Nilesch [email protected] http://www.geo.unizh.ch/rsl/ Abstract The SAR image products (ERS-SLC) from the ERS AMI sensor are produced and distributed by various PAFs, such as CPRF, 1-PAFand D-PAF. For high end applications that fully exploit the phase information of the coherent recording system (i.e. interferometric applications I DEM generation) the feasibility of combining products from different processing facilities is important. During this study, the level of cross-compatibility of SLCs was assessed by comparing interferograms and coherence maps which exploit combinations of pairs of SLCs. This included auto-interferograms from different processors. The conclusions is that the tested SLC products can be exchanged without a significant increase of the phase noise. Nevertheless, high precision measurements must take into account the systematic errors - phase offset and trend - introduced into the data. Keywords - SAR-Interferometry, Phase Noise, Image Quality, SLC, PAF Introduction The user community has an interest in performing SAR Interferometry by using existing SLC (Single Look Complex) images, without having to take into account which PAF (Processing and Archiving Facility) produced them. The requirements for SLC images are described in the ERS.SAR.SLC-I Product Specifications (ESA Pub., 1995), where a number of tests are defined, e.g. to confirm the phase preservation during the processing. However, these tests do not ensure the cross-compatibility of the images processed by different SAR processors for high quality SAR interferometry, since they check mainly the quality of the azimuth compression algorithm rather than the overall performance of the processor. The interferometric phase depends heavily on the surface structure. To be representative of as many applications as possible, the chosen test scenes included areas with a large variety of coherences. We used two different tandem scenes, acquired over Bern (Switzerland) and Cairo (Egypt), respectively. The Bern tandem data from 13/14 August 1995 covers agricultural, forest and urban areas as well as lakes: a mid to low level of coherence was observed in this area. The Cairo scene comprises larger regions with rocks and dry areas to provide a completely different coherence distribution. The desert area south-west of Cairo showed high temporal coherence on the tandem I-day repeat data from 19/20November 1995. The SLC images received from SAR processors used at CPRF (Central Processing Reference Facility, ESRIN) - ESA VMP (Verification Mode Processor) - and 1-PAF(ASI) were analysed and their CEOS parameters and Doppler spectra compared. This was important for the interpretation of the interferograms and its statistics, which were processed subsequently from each pair of the SLC products. The InSAR processing was performed using our in-house developed interferometric SAR processor ISP and Zurich InSAR Processor (ZIP). Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, Zurich. 30 Seot.: 2 Oct. 1996 (ESA SP-406,Vol. 11, December 1997) 130 Doppler Centroid The azimuth spectrum contains information about the Doppler shift. and its bandwidth is related to the spatial resolution in the azimuth direction. For the interpretation of the coherence map statistics the knowledge of the relative resolution between the SLC images is required. Figure 1 presents the averaged total azimuth power spectra for the Cairo SLC. It is evident that ERS-1 and ERS-2 spectra have their maxima at a different Doppler frequency (i.e. the Doppler centroid frequency) due to different squint angles. In interferometric processing, one has to avoid relating the not common spectral parts by filtering. However, the systems have the same spectral shape and the same processed and even a corresponding half-power bandwidth of about 605 Hz (both, Cairo and Bern, 36% of the PRF of 1679.9 Hz). On the other hand, the values indicated for the processed Doppler bandwidth in the CEOS header differ significantly, probably using different definitions of the term "bandwidth". ER.~~2 ER.S-1 o.e OA 0.2 -1°000 -eee -eee -<100 '200 e ~o «10 1100 eoo 1eee °"3pler f,.,CJl..,q CHz.) Figure l: Relative azimuth power spectrum of ERS-1 (right) and ERS-2for CPRF (solid line) and 1-PAF (dotted). Bern test site. The Doppler centroid frequency (the frequency of the spectra's maximum) is a main driver for accurate azimuth focusing. Since it depends on the Earth's rotation and satellite yaw steering, it varies in slant range. This dependency is usually approximated by a polynomial of small degree. In Figure 2 the polynomials given in the header files are plotted and compared with the calculated average of the Doppler Centroid frequency for the ERS-2 SLC processed by RSL. All PAFs use polynomials of small degree to represent the range dependency of the Doppler Centroid frequency. The CEOS third order polynomials used by the CPRF and D-PAF show inconsistencies compared to our own estimation from raw data, whereas I-PAF represents an acceptable linear fit. The Centroid shows a high dependence on the mean altitude above sea level. For ERS the azimuthal displacement is in the order of 100 m per 1000 m change in altitude, which corresponds to -30 Hz. Therefore it is important to use an algorithm that takes as many azimuth samples as possible into account to mitigate this topographic influence. If this is not the case, the Doppler Centroid is expected to show a significant dependence on azimuth position, and thus shows a high variability compared to the mean value of the Doppler Centroid over the whole scene. If one estimates the Doppler Centroid by exploiting the final SLC product one finds a good 131 representation of the polynomial used during focusing, as expected since the Doppler Centroid is projected into the data. However, the polynomials used at CPRF and D-PAF show significantly different coefficients and the 1-PAFeven a different polynomial degree, hence the parameters or the software to estimate the Doppler centroid frequency must be different among these PAFs. -··-··~··-··-···-··-··-··-~--·-··-· . . . . . -!'400 500 I000 !!500 3100 2!500 Rlnll" Pb •• Figure 2: Doppler Centroid variation along slant rangefor ERS-2from Cairo processed at the various PAFs. Thepolynomial coefficients are takenfrom the SLC CEOS leaderfiles. Estimation is the JO point average with an FFTsize of8192 samples. Interf erograms More than a dozen interferograms were computed and analysed. To distinguish auto-interferograms and tandem interferograms, the terms Singletrack and Multitrack are introduced. Singletrack means that the identical original raw data from the sensor is processed at different PAFs and that these products are compared. Multitrack is the common tandem configuration ofERS-1 and ERS-2. ERS-1 is considered the master track within this report. Singletrack Singletrack interferograms provide a useful way to compare the SAR processors. To calculate the phase difference of the two images, a coregistration had to be performed. This resulted in a shift in range of none to several pixels, whereas the azimuth offset additionally showed a subpixel offset. The observed range offset must be explained by inaccurate time-referencing of the SAR processors. The azimuth shift of up to thousands of pixels is originated by the non-standardized starting time of the frames and differences in the sensor velocities used for focusing, together with inaccurate azimuth time-referencing. The phase statistics of the coregistered auto-interferograms are compiled in Table 1. The value of the interferometric phase between SLC from the various PAF was almost constant but non-zero. According to the requirement of Phase Preservation of the SLC (ESA Pub., 1995), the zero phase of the azimuth filters are designated to the zero Doppler point in the time domain. Since the auto-interferograms act similar to the interferometric offset processing test proposed in CESAPub.. 1995), we expected a mean phase of less than 0.1 degrees. The observed phase bias leads to the assumption that at least one of the tested SAR processors has problems with phase preservation. Most likely this is due to differences in the Doppler centroid estimation. 132 ICairo llinterferogram IIMean Phase Value (deg) Std Dev Phase (deg) IERS-1 IICPRF I 1-PAF II 306.5 3.50 IERS-1 llCPRFI D-PAF II 189.l 3.04 IERS-2 II CPRF /1-PAF II 155.8 2.98 IERS-2 llCPRFI D-PAF II 10.3 3.21 Table 1: Singletrack interferogram ofSLCproducts. Cairo site. This observed variability of the mean interferometric phase implies that absolute phase measurement (besides the 2-rcambiguity) is still not possible. However, each processor passes the requirement on the standard deviation of the interferometric phase. It lies below the limit of 5 degrees. It was a surprise to observe that the CPRF and D-PAF processors do not show a smaller standard deviation, since both are assumed to use identical source code. Though, if the start time in azimuth is different between CPRF and D-PAF, the final products are not identical, and could show the observed variation on the same level as two different processors do. Therefore it is still possible that the processors have the same source code. This random phase offset could be a problem when mosaicing full frame or quarter scene images: the phase continuity at the border would not be guaranteed. In the following we tested for a systematic error in the phase, i.e. a dependency on range or azimuth position. Figure 3 reveals no phase trend in azimuth for the auto-interferograms. Moreover, the mean phase stays in the interval of 0.1 degrees. 10~r······· i Ti·l····Ti i ..T..."··t·i tr: 10.21!·I·.••I.I...t····:-1-t....l..•....·•····•········11-1··1··1:-······1·1··•······--·•····· ~ ; ~ 10·1 o 1oo 200 300 410 S'.IO mo 100 imo 900 1ooo Azim.it. Pi Figure 3: Interferometric phase variation along azimuth direction for the CPRF I D-PAF interferogram. ERS-2, Cairo area. On the other hand, Figure 4 reveals a systematic phase trend in the range direction on the order of 0.6 degrees. This phase trend in range is most probably due to differences in the polynomial used to approximate the Doppler Centroid frequency as a function of range, and hence the Doppler Centroid estimation software. ~~~~~- -~~ ~~- We believe that it is less important that the Doppler Centroid estimation is a perfect representation of the physical Doppler Centroid frequency than that the various PAFs calculate the data using the same algorithm and polynomial order. o•~--··- ·--·_ .,..·-·-··--,. ··--···--r-·····-. ' Figure 4: Interferometric phase vanatron along range direction (or the CPRF D-PAFinterferogram. ERS-:!.Carro area. Figure 5: Flattened interferometrtc phase o(ERS-1 ERS-:!for Bern. CPRFprocessing was used for both SLC Figure 6: Flattened interferometric phase ofERS-1 ERS-:!.forCarro.CPRF.processing •ms used (or both SLC. 134 Multi track The generation of the multitrack interferogram is the conventional task of producing a tandem interferogram. Figure 5 and .Q show the zn fringes of the flattened interferogram. The resulting phase fringes have a periodicity equivalent to a change in height of about 150 m for the Bern interferogram and of about 40 m for Cairo. The two interferograms confirm the choice of the two test sites: on the one hand there is the accurate and high-resolution interferogram from the flat and dry area around Cairo, which could be used for direct phase unwrapping without any further processing steps; on the other hand, the Bern interferogram appears noisy even in the one-day repeat tandem configuration and the rather short baseline. The production of a DEM would require numerous user interactions during the · unwrapping task. Coherence Maps The coherence histograms and statistics are an indicator for the quality of the various processors. The processing of image pairs from the same raw data in particular reveals the relative phase differences introduced by the different processors. Singletrack Figure 7 shows the coherence that can be achieved when combining products from different SAR processors in the singletrack case. Notice how a difference, even if small, is present in the phase information provided by different focusing programs starting from the same raw data. The level of phase noise is directly related to the correlation coefficient (Bruhler R. and Just D., 1993): the results presented in Figure 7 concerning the coherence distribution confirm the results seen in Table 1 in terms of phase noise. In this case no systematic phase trends need to be considered as a source of coherence loss, since the estimation of the correlation coefficient is performed on relatively small subwindows. No temporal decorrelation effects are present in this combination: only image misregistration, differential defocusing and different processor noise levels are possible sources for this small loss of coherence. ERS-1 CPRF 1211-········l·········ERS-l CPRF . ERS-2 CPRF 1oot-···-....L ERS-2 CPRF 0.9? 0.915 0.98 0.985 0.99 0.995 Cdl•••nae Figure 7: Coherence/or Cairo images: solid ERS-1CPRFI1-PAF; dotted ERSl CPRF I D-PAF; dashed ERS-2CPRFI1-PAF; dashdot ERS-2 CPRF I D-PAF. I l 135 Multitrack The multitrack tandem coherence maps are also affected by the temporal decorrelation. The histograms in Figure 10 and 11confirm again the choice of two completely different sites. by showing distinct distribution and maxima values. Figure 8. Coherence map of the Bern test site. SLCs processed at CPRF. White represents high coherence areas ~··~·. Figure 9: Coherence map of the Cairo test site. SLCs processed at CPRF. White represents high coherence areas For the Bern site (Figure 10), all calculated interferograms show about the same probability distribution function. Very small differences in the estimated coherence values are noticeable only when analysing the numerical results. No reasons appear from this test to suggest that one ~ust necessarily use SLC pairs focused by the same processor to produce interferometric images. 136 2.5.-----,----r--....,.---.----.--..,..---,,----r--,...---. 2~ .....•• -~·-·······;.. •...... ; .;...... •..; ; -. . ~ ·1115~.....•• ~- ~ ~i +"f f""'""'f"""' : ! . . . Q.5~-1...••. ; ....•••.. ; ..••••.•. ; ••••.. ·--~--.. ·····~·-··· ·--~····..•• . . OO 0.1 02 0.3 QA 0.5 0.11 0.7 0.8 0.9 Cah..,.nae Figure JO:Coherence/or ERS-1/ERS-2 Tandem images over Bern. Following combinations are used; 1-PAFICPRF.1-PAF/1-PAF.CPRF!CPRF, CPRF/1-PAF. For the Cairo site (Figure 9) some differences can be noticed between the different coherence maps obtained from the various SLC combinations. Table 2 compiles the figures of statistics. The mean coherence of an interferogram produced using SLCs from different processors is not necessarily lower than that obtained from SLC produced by the same processor: on the contrary, the maximum value of coherence for this site was obtained by exploiting two data sets focused by different processors. Cairo ]IMean Coherence llStd Dev of Coherence IERS-1 llERS-21 ICPRF .,,CPRF II 0.6961 --1IT52o II I CPRF II1-PAF II 0.7025 0.1549 II I 1-PAF II CPRF II 0.6599 0.1445 II 11-PAF111-PAFII 0.6757 0.1476 II Table 2: Coherence statistics for ihe tandem interferograms. Cairo site. 137 zns.i ERS·2 ; !ii-········•·········:······--·:· CPRF O'RF CPRF l-PAF 4 -·······'········· •.....•• I-PAF O'RF i° : I-PAF l-PAF i : . c : : i~3 -······-'········: ·-······ ..••.... .;.:•....••. G. : i 2 •...... • , ....•.... ••····•· ········+-······· : : . . : : 11-·······-~········· T . 0o 0.1 02 0.3 0.4 c.;,.,....,CIO0.5 0.6 0.7 0.8 0.9 Figure//: The coherencefor ERS-llERS-2 Tandem images over Cairo. Starting/ram left: 1-PAFIC-PAF(dashed), 1-PAFll-PAF (dolled), CPRFICPRF (solid), CPRFll-PAF (dash-dolled). The observed coherence differences can be explained in terms of residual defocusing effects (Monti Guarnieri A.. 1996), differences in the focusing algorithms and considering the normal variability of the phase noise introduced by every processor. Conclusions Although this study revealed some remarkable qualitative results, we could not perform further tests to get better statistics for quantitative data extraction. It should be kept in mind that only a small data set has been used within this analysis. The investigated SLC image did not show any anomalies, i.e. sidelobe artefacts or saturated areas. However, the tested processors are not compatible with respect to the Doppler Centroid estimation. The values for this frequency deviates by as much as 50 Hz compared at one specific range distance. The Doppler Centroid frequency dependency on range is represented by polynomials of a degree not consistent among all PAFs. This uncertainty may be the source for a random phase offset introduced into the SLCs. In turn, this may lead to a phase inconsistency when stepping quarter scenes to full frames (or full frame images to larger mosaics), and makes reconstruction of the absolute phase of the interferometric products without the use of ground control points impossible. Auto- and differential interferograms showed a phase trend in the range direction over a quarter scene swath, whereas the mean interferometric phase in the azimuth direction was observed to be constant. This phase trend is probably introduced by the SAR processors and leads to systematic errors in the reconstructed DEMs. The influence of this trend is larger in interferometric applications with small baselines. The product specifications CESAPub .. 1995) point out that SLC-1 products consist of a full frame. To date, most of the operational SAR processors (suitable for SAR interferometry) have only been capable of processing single quarter scenes at a time. Thus, for larger-scale interferometry, a full scene is assembled from four independently processed parts, with the possibility of introducing a phase inconsistency at the borders. Hence, the phase trends and discontinuity will become more severe when considering full frames. 138 Most important, the In SAR processing of SLC data from different PAF s did not increase phase noise. The user can combine SLCs from various PAFs for producing interferograms without a loss of the overall quality. However, there are some secondary restrictions to the cross-compatibility: e.g. areas covered by the SLC are not standardized, especially the azimuth times. This may lead to a shift of up to 25% of the length of corresponding quarter scenes. The parameters included within the CEOS header are defined differently for each PAF (i.e. I-PAF and CPRF even indicate different wavelengths, pulse repetition frequencies, and spatial resolutions). Some of these data seem to be in disagreement with the image positions. ESA has announced its intention to harmonize the time-referencing of the processors and CEOS header format used at ESA PAFs. Acknowledgement This work emerged from an ESA contract study addressed to ESRIN, Rome. We like to thank ESA for the generous provision of the required ERS images. References Bamler R. and Just D., 1993 Phase Statistics and Decorrelation in SAR lnterferograms, Proc. IGARSS '93, pp. 980-984. Barmettler A., et. al., 1996 Cross-Compatibility of ERS-SLC-1Products, Report to the ESA/ESRIN, RSL Zurich. Cattabeni M., et. al., 1994 Estimation and Improvement of Coherence in SAR Interferometry, Proc. IGARSS '94, pp. 720-722. ESA Pub., 1995 ERS.SAR.SLC-1Product Specifications, ESA Doc., Issue 1. Goodman N.R., 1963 Statistical Analysis Based on a Certain Multivariate Complex Gaussian Distribution, Ann. Math. Statist., Vol. 34, No. 152, pp. 152-180. Monti Guarnieri A., 1996 Residual SAR Focusing: An Application to Coherence Improvement, in IEEE Trans. GRS, Vol. 34, No. 1, pp. 201-211. 139 The use of tandem data in the Antarctic area Xiaoqing Wu, Karl-Heinz Thiel Institute of Navigation, University of Stuttgart Stuttgart, Germany Stefan Wunderle Department of Physical Geography, University of Freiburg Freiburg, Germany Abstract ERS SAR interferometry can be used to monitor the movement of glaciers and ice shelves in the remote Antarctic region. The ERS-1/ERS-2 tandem data with only one day acquisition time interval provide an opportunity to study the ice movement in those areas with a high dynamic in the weather conditions and therefore a fast increase in air temperature which lead to a decrease in coherence. In this paper the tandem data are used to estimate the movements of the glaciers on the mountains at Marguerite Bay, near the arg. base San Martin on the Antarctic peninsula and of the shelf ice around Hemmen Ice Rise near Berkner Island of Antarctica. Comparison between the results from ERS-1 data and those from ERS-1/ERS-2 tandem data will be given for the shelf ice areas around Hemmen Ice Rise. Keywords: SAR interferometry, ice movement, DEM, ERS-JIERS-2 Introduction SAR-interferometry proved to be a promising technique for polar research. It can be used to determine the position of grounding lines, to produce three-dimensional topographical maps, to measure the tidal changes and to estimate the ice surface movement [1,2,3,4]_ An interferogram derived from two'single look complex SAR images includes the information about topographical elevations and surface changes. If there is no surface change, the digital elevation model (DEM) can be derived from the phase of the interferogram. For the ERS-1 SAR data in a region near Bonn, Germany, the precision of about 4.Sm for the elevation estimation over the region with an area of 2612 km2 was reachedl-l. If the elevation is known, the information about surface movement included in the interferogram can be used to estimate the surface movement. But only one movement component in slant range is not enough to describe the surface movement in two or three dimensions. The azimuth displacement is used to estimate the azimuth movement component, which provides another movement component of the surface. For mountain glaciers, the ice is supposed to have only the movements in horizon, because the vertical movement caused by the local slope is very small for the relatively flat glaciers. For the floating ice on the water, the vertical movement component is supposed to be constant and its value must be known a priori in order to estimate the horizontal ice movement. In this paper, the ice movement of two test areas will be studied with the ERS-1/ERS-2 tandem data with one day time span. Comparisons between the results from ERS-1 data and those from ERS-1/ERS-2 tandem data will be given for areas of the ice shelf around Hemmen Ice Rise. Proceedings of the 'Fringe 96' Workshop on ERS SAR lnterfefometry, Zuticn, 30 Sept.· 2 Oct. 1996 (ESA SP-406, Vol. II. December 1997) 140 Relative Displacement For two SAR images of the same area, there are relative displacements between them because of different orbital positions of the SAR and the changes of positions and properties of the scatterers on the ground between the two SAR observations. The relative displacement includes slant range displacement and azimuth displacement. They contain information about relative orbital positions, topographical elevations and topographical changes. If the two orbits are nearly parallel and the azimuth displacement caused by system erorrs has been eleminated, the azimuth displacement can be approximated as Azimuth displacement Surface (I) change in Azimuth The azimuth displacement can be estimated by use of the amplitude correlation, the statistic correlationlvl or the spectral methodl/l. The accuracy depends on the coherence between the two SAR images. Fig. 1 gives the comparison between the accuracy of the amplitude correlation method and that of spectral method by computer simulation. The accuracy of spectral method will be better than 0.1 pixel, if the coherence is greater than 0.55. If the coherence lies between 0.55 and 0.7, the accuracy will be 1/10 to 1120pixel. For ERS-1 and ERS-2 with pixel spacing of about 4m in azimuth, the accuracy will be in the order of 0.2 0.4 m. 3 r~ ' ,-.. -0 ' .~ ") 0.. ~ <::« ~ 0 '.;::?~ -~ '' "O '' ' ' ' ' 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 coherence Fig. 1 Standard registration deviation of the amplitude correlation method and the spectral method as afunction of the coherence . After the phase component resulting from reference ellipsoid of the Earth is removed, the slant range displacement, which is equivalent to the phase of the interferogram of the two SAR images scaled by a factor of 14, contains only the topography and the surface changes and can be approximated for nearly parallel orbits: l 141 1 -{'¥(P)- '¥(Po)} 1 . d" 1 . 1 411 = re atrve tsp acement m s ant range (2) Bx cose - By sin0 R ·sin0 z + r where P0 is a reference point, is the interferogram phase, R is the slant range between P0 and the master orbit, is the incidence angle, Bx and By are the horizontal and vertical baseline components, respectively, z is the relative elevation with respect to the reference point and r is the surface change in slant range. Because of the equivalent relation between the interferogram phase (P) and the relative slant range displacement, the slant range displacement can be determined very accurately from the interferogram phase. Generally, after the normal multilook processing an accuracy of 2/10 in the estimation of the interferogram phase can be reached. A phase accuracy of2/10 means that if there is no movement between the two acquisition times, the accuracy for the elevation estimation from the interferogram phase will be 10 percent of the 2 elevation, and that if the topographical elevation is known, the accuracy for surface movement estimation from the phase will be in the order of I0 percent of /2=2.8cm, i.e., 2.8mm in slant range. According to (2), 2.8mm surface change in slant range results either from 7.2mm surface change in ground range or 3 mm surface change in elevation for an incidence angle of 23°. The interferogram phase can be used to estimate the topography. Both the phase and the azimuth displacement are required to estimate the surface changes between the two acquisition times of the two SAR images. In the following, we present some results in two different test areas in Antarctica. Test Site Marguerite Bay (McClary- and Northeast Glacier) -Horizontal movement estimation The test-site McClary- and Northeast Glacier is located in the Marguerite Bay area (Antarctic Peninsula) at 68°S and 67°W. This area is influenced by maritime-continental air masses which often lead to a fast increase of air temperature and therefore for changes in snow properties[9,IOJ.These climatic conditions makes it difficult to obtain an interferogram from repeat pass images. McClary- and Northeast Glacier and the surrounding mountains have a complicated topography with highest elevations up to l .800m a.s.l. The flow velocity and direction of the glaciers differ significantly. Fig.2 shows the digital elevation model of this area with an accuracy of about 1Om,where the arrows stand for the known movement directions of the glaciers and the values of some places mean the glaciers velocities measured by GPS. Fig~ Digiru/ elevation model of.\lcC!w:1·- and .\'orrheo.1r Glacier with an uccuracv of l 0111 rIIA G-Fnmkfurt J. Fig. 3 is the interferogram phase obtained from the SAR images collected on Oct. 15(ERS-1) and Oct. 16 (ERS-2 ). 1995. from which the phase component resulting from the Earth reference ellipsoid is already removed by utilizing the precise orbit data. The baseline is about 113m and the corresponding 2 elevation is about 92111. The phase contains the information of both the elevation and the surface changes. The highest point among the considered glaciers is about 800m higher than the lowest point. An error of about 8 fringes or 56cm surface change in slant range will result. if the elevation information in the interfcrcgram phase is ignored. In order to obtain a more precise glacier movement estimation from the interterogram phase. the phase resulting from the elevation is removed from the phase. The resulting relative phase in Fig. 4 contains mainly the surface movement of the glaciers. If the precise orbit data is absolutely accurate. IOm uncertainty of the DEM and 2110 uncertainty of the phase result in an uncertainty of about 2/5 in phase or /I 0 in slant range displacement. whi ch corresponds to l .5cm uncertainty of surface change in slant range within the acquisition time interval of one day. Therefore an uncertainty of 2 3 cm for the surface change in ground range can be expected. if the errors caused by registration between the DEM and the SAR image are also considered ~~ Fig.3 Relative interferogram phase obtained Fig. .:/.Modified relative phase of Fig. 3 hy from SAR images collected on Oct. 15. 1995 using the DEAi in Fig 2 (ERS-]) and Oct. 16. 1995 (ERS-2) with baseline of. about 113111and 2 elevation of. about 92111. Fig.5 Movement map of McC/ary- and Northeast Glacier estimated by INSARfrom the ERS-1 ERS-2 tandem data on Oct. 15 16, 1995. 144 By using both of the surface change in azimuth obtained from the azimuth displacement and the surface change in ground range derived from the modified relative phase. we obtained the movement map of some glaciers on the test site Marguerite Bay. which is shown in Fig. 5 The background of Fig. 5 is the intensity image of ERS-1 SAR from 15 Oct . 1995. The result agrees very good with the GPS-measured result Test Site Hemmen Ice Rise - Topography and horizontal movement estimation The second test site is located around Hemmen Ice Rise(HIR) and near Berkner Island( BI) of the Antarctic with latitude of S77 9° and longitude ofW48.8°. All the area is covered with ice and snow. The surrounding ofHIR is the Filchner-Ronne-lceshelf The shelf ice drifts towards see with velocities of about 700900m/a west ofHIR and 400500m/a between HIR and Bil81 Fig 6 is an air photograph of the area, where the arrows represent the movement directions of the shelf tee Fig 6 Air photograph of HIR with the movement direction of the shelf ice. Fig 7 shows the relative interferogram phase of the HIR with baseline of about 179m and 2 elevation of about 50.6m, which is obtained from the ERS-1/ERS-2 tandem data on Jan. 16(ERS-l) I l 7(ERS-2), 1996. In Fig. 7, the phase resulting from the earth reference ellipsoid is already removed by utilizing the precise orbit data. The relative phase in Fig. 7 includes information about the elevation and the surface changes. The ice on BI and HIR moves probably very slowly. But the velocity is not known and depends on the weather condition change between the two acquisition times. The ERS-1/ERS-2 tandem data with one day time interval permit us to estimate the topographical elevation of ice Islands such as BI with a better accuracy, if the baseline is not too small. A horizontal velocity of IOm/a corresponds to about 1cm/day slant range change, which results in 1/3 fringe in the relative phase. For the 2 elevation of 50m, 1/3 fringe represents l 7m in elevation. Therefore, if the velocity of the ice on BI is smaller than IOrn/a, and if the precise orbit data are exact, the elevation estimated from the relative phase in Fig. 7 will have an accuracy of about l 7+5=22m, where the uncertainty of Sm results from the phase uncertainty of 2110 Fig 8 gives the 145 30-version of the estimated elevation map. For comparison, Fig. 9 and Fig. 10 give the relative phase obtained from ERS-1 SAR data collected on Jan. 20/23, 1992 and the corresponding 30-version of the elevation map, respectively. Accidentally, the two interferograms in Fig. 7 and . Fig.9 have almost the same baseline of about 180m and the same 2 elevation of about 50m. Comparison between Fig. 7 and Fig. 9 shows that the topography derived from ERS-1/ERS-2 tandem data with one day time span is in good agreement with that obtained from ERS-1 data with three day time span. This result reveals that the ice on BI is relatively fixed. If it moves, it moves very slowly. For ERS-1/ERS-2 tandem data with one day time span, BI can be regarded as fixed land. As the topography of the ice shelf around HIR is very flat with elevation differences smaller than 50m, the fringes of the ice shelf in Fig. 7 and Fig. 9 result mainly from the surface changes of the shelf ice. In addition to the horizontal surface change, the shelf ice has also a vertical change, which is caused mainly by the tide changes. If the tidal variation is known, the horizontal surface change can be derived from the relative phase. According to tidal prediction provided by the British Antarctic Survey, the tidal height ofHIR at time of the ERS-1 SAR data acquisition is 59.6cm at 9:24 of Jan. 20, 1992 and 35.5cm at 9:24 of Jan. 23, 1992, respectively. The tidal change in the meantime is -24. lcm. By utilizing the constant tidal change in the floating shelf ice area, we have derived the ground range component of the surface change. Together with the azimuth displacement, the horizontal movement map of the shelf ice around HIR has been estimated. The result is shown in Fig. 11. If the tidal variation is not known, the horizontal surface change can not be derived from the relative phase. However, if a tie point on the floating shelf ice with known velocity is available, the constant tidal variation can be determined according to the known velocity and the fringes between this tie point and the fixed land. Point A in Fig. 11 with the known velocity of3m/3days is selected as the tie point. With the help of tie point A, the tidal change between 9: 15 of Jan. 16, 1996 and 9:15 of Jan. 17, 1996 is determined from the relative phase in Fig. 7 to be -27cm. The resulting horizontal movement estimated from the relative phase, the azimuth displacement and the tie point is shown in Fig. 12. 1.+6 ?1§.lzj.itlfi 0 500 m Fig 8 JD-version of the topographical elevatio Fig ~Relative phase of H!Rfrom Jan. estimatedfrom Fig 7. lhe background is the JJ 16(ERS-l) l 7(ERS-2). 19<)6with baseline of image of the areatl: Intensity ·~~~\=~:;=::u::'="~ o ~OOm Fig. 9 Relative phase of HIRfrom Jan. Fig l 0 slr-version of the topographical elem I 20 23(ERS-l). 1992 with baseline of about Bl estimated from Fig Y. The background is ti 1?-Simand 2elevation of about ./9.8m. image of the area/I: 1111e11si(i:ofJan. 20. /<)<)2 elevation. Saturation: coherence) 14- Fig. 11 HIR shelf ice horizontal movement Fig. 12 HIR shelf ice horizontal movement estimated from ERS-1 SAR data on Jan. 20 23. estimatedfrom ERS-1 ERS-2 tandem data on 1992 and the tidal data provided by British Jan. 16 Jan. l "'. 1996. Antarctic Survey. Conclusions ERS-1/ERS-2 tandem data with only one day time span can be used to estimate topographical elevation models oflarge islands such as BI in the Antarctic with good accuracy. They can also be used to estimate the movement of glaciers on the mountains and of shelf ice on the see water. For the glaciers on the mountains and thin shelf ices, it is recommended to use the tandem data with one day time interval. The coherence may be lost for such areas. if data sets with great time spans are used. Acknowledgments This work was supported by German Ministry of Science and Research (BMBF) and DARA project ..DYPAG" (FKZ:03PL016A4) and also by European Space Agency. which provided all the SAR data used. We would like to thank lfAG-Frankfurt for the digital elevation model. References [I] Hartl, Ph., Thiel, K.H., Wu, X., "Information extraction from ERS-1 SAR data by means of INSAR and 0-INSAR techniques in Antarctic research", second ERS-1 symposium. Hamburg, Germany, Oct 1993, pp. 697- 702. [2] Goldstein,R.M., Engelhardt, H., Kamb, B., Frolich, RM., "Satellite Radar Interferometry for Mornitoring Ice Sheet Motion: Application to an Antarctic Ice Stream", Science, Vol. 262, Dec 1993, pp. 1525-1530. 148 (3] Thiel K.H., Hartl Ph., Wu X., "Monitoring the ice movements with ERS SAR interferometry in the Antarctic region", Proc. of second ERS Applications Workshop, London, UK, Dec. 1995, pp. 219-223. (4] R. Kwok, M.A. Fahnestock, "Ice sheet motion and topography from radar interferometry", IEEE Trans. Vol-GE 34, No., 1, Jan. 1996, pp. 189-200. (5] Wu, X., Thiel, K.H., Wehr, A., "The effect of different land covers on the accuracy of interferometric DEM", Workshop on ERS SAR Interferometry, Sep., 1996, Zurich, Switzerland. (6] Gabriel, A. K., Goldstein, R. M. ,,Crossed orbit interferometry: theory and experimental results from SIR- B",Int. J. Remote Sensing, vol. 9, No. 5, pp. 857-872,1988. [7] Li, F., Goldstein, R.M., "Studies of multibaseline spacebome interferometric synthetic aperture radars ", IEEE Trans. Geosci. Remote Sens., vol. 28, no. 4, Jan. 1990, pp. 88-97. (8] Bennat, H., Heidrich B. and Sievers, J., "Extraction of Antarctic topographic glaciological features from ERS-1 SAR data", Proc. Second ERS-1 Symposium: Space at the Service of our environment, Oct. 1993, Hamburg, Germany, pp. 141-145. (9] Wunderle, S. (1996): Die Schneedeckendynamik der Antarktischen Halbinselund ihre Erfassung mit aktiven und passiven Femerkundungsverfahren. - Freiburger Geographische Hefte, Heft 48 [10] Wunderle, S. & H. Saurer (1995): Snow properties of the Antarcitc peninsula derived from ERS-1 SAR images. - RSS95, Remote Sensing in Action, Proceedings of the 21st Annual Conference of the Remote Sensing Society, 11.-14. Sept. 1995, pp.1231-1237 149 Glaciological Studies in the Alps and in Antarctica Using ERS Interferometric SAR Helmut Rott and Institut fur Meteorologie und Geophysik Universitat Innsbruck. Andreas Siegel A-6020 Innsbruck, Austria http://info.uibk.ac.at/c/c7/c707I Abstract Applications of repeat pass SAR interferometry and conditions f Spacebome radar interferometry has been recognized as a valuable tool for glaciological research, offering the possibility to monitor the ice motion and to map the surface topography over extended areas. Temporal decorrelation is a major problem for repeat pass interferometry over snow and ice. Furthermore, the penetration of microwaves into dry snow causes difficulties because the return signal originates from the volume and thus shows baseline-dependent decorrelation ( Gatelli et al., 1994). In addition to the relevance for interferometric analysis, investigations on signal coherence provide important insights into the interaction mechanisms of microwaves with snow and ice. In this paper we report on studies of coherence for various snow and ice types in the Alps and in Antarctica, based on interferometric SAR data from ERS-1 in 3-day repeat orbit and from the ERS-1/ERS-2 tandem campaign. The investigations are a contribution to the ERS-1/ERS-2 experiment A02.A 101 "Comparative Investigations of Climate Sensitivity and Dynamics of Glaciers in Antarctica, Patagonia, and the Alps". Coherence Study on Alpine Glaciers During the ERS-l/ERS::2 tandem operation 1995/96 four field campaigns were carried out on glaciers of the test site Otztal, Austrian Alps, to learn about the conditions for signal coherence and temporal decorrelation. Over this area descending and ascending overflights of ERS with 35 day repeat orbits are separated only by 12 hours in time. Five tandem pairs from 23/24 August 1995, 27/28September1995, 6/7December1995, and 14/15 February 1996 were available for this study. Interferograms were generated for all of these image pairs. The complex degree of coherence between the radar signals S1 and S2 was calculated according to Proceedings of the 'Fringe 96' Workshop on ERS SAR Interferometry, lunch. 30 Sept. - 2 Oct. 1996 (ESA SP-406, Vol. 11, December 1997) 150 y J(s1s; ){s2s;) where < > denotes ensemble averaging. For calculating coherence images I0 x I0 pixels averaging was applied, corresponding to about 70 independent looks, and the local fringe frequency was taken into account. Averaging is particularly important in regions of low coherence to obtain reliable numbers (Lee et al., 1994). The degree of coherence is the magnitude of and may assume values in the range from 0.0 to.1.0. As an example. Figure 1 shows the interferogram of 14/15 February 1995 for the ascending satellite pass. The effective baseline was Bn = 136 m, resulting in an elevation difference of 65 m for one fringe cycle. For this image pair the coherence was comparatively high throughout the scene. Only on steep foreslopes and in layover zones the fringe visibility is very poor or fringes are lost due to undersampling. The ice-free surfaces were covered with dry winter snow of 0.5 m to 1.5 m depth. The ice areas and the frozen firn on the glaciers were covered with 1 m to 2 m of winter snow. The arrows in the Figure indicate the flow direction of three main glaciers. The firn areas of Gepatschferner and Kesselwandferner form a comparatively gentle plateau of about 15 km2 in area at elevations between 3000 m and 3350 m. Because the ice on the plateau moves less than 5 cm per day, the fringes are primarily related to topography. Apart from the glacier plateau, the topography in the test site is quite steep, resulting in considerable loss of information due to layover and foreshortening (Rott and Nagler, 1994). The altitudes in Figure 1 are ranging from 2200 m to 3700 m above sea level. Flight. Dir.t t N L__.Look Dir Figure 1: Interfcrogram of the test site Otztal. formed from an ERS-1 image of1../ February 1996 and an ERS-2 image of 15 February 1996. superimposed to an amplitude image. Glaciers: G - Gepatschferner. H - Hintereisferner.. K - Kesse/wandferner.. Figure 2 shows the degree of coherence for different surfaces in the test site Otztal for 4 different tandem pairs of the ERS-1 /ERS-2 mission. The baselines for the four image pairs ranged from Bn = 80 m (23/24 August) to Bn = 274 m (27/28 September). so that geometric decorrelation due to baseline effects is of little relevance. The meteorological and snow conditions for the 24 hour periods between the satellite overflights can be summarized as follows: 23124 August 1995: Air temperature in 3000 m between +3 and +5°C. cloudy, part of the time fog and drizzle, little wind: water on the ice surfaces: in the accumulation areas coarse grained and wet snow with liquid water content of 5% by volume in the top layer. 27128 September 1995: Air temperature in 3000 m dropping from -3 to -8 °C. overcast. fog. passing of a cold front with strong winds, snow drift. several centimeters of snowfall between the two overflights. The ice and fim areas of the glaciers were covered with 20 to 60 cm of snow from early September. The top snow layer (20 to 40 cm in 3000 m) was frozen, the thickness of the frozen layer varied with the altitude and increased between the two overflights. 617 December 1995: Air temperature in 3000 m -l 2°C, low wind velocities in the valleys, but 152 gusty winds from the south on the mountains, no snowfall; lm to l.5m of homogeneous dry winter snow on the glaciers. 14115February1996: Air temperature in 3000 m between-16 and -18°C, cloudy and some wind during the overflight ofERS-1, calm on 15 February; lm to 2m of homogeneous dry winter snow on the glaciers. Firn Ice (l) g 11.(>+------1 SJ> ~ o.4-t------' u0 O.:.i ,,\.!\. 2? .••. ~-1~. 1<-.2 za.e 2? .••. ~.12. 1<.~- Alpine Grassland (.),lj r--- Q) r-- .--- r-- ...... ~ oJ.6 •... asQ) r-- is oJ.4 ~I- ~ r-t I- u O.:L >-I 1-4 •..•• 1-1 11--4 •..•• () I I I I I I I I I I I I I I I I I I I I I I I I I I I '.1"\ >L v r», :"1 I:;, ' !;, ';.","\.k, :; I',". /v, 1-:0" I A.:1. Figure 2: Degree of coherencefor 3 sites injirn areas and 2 sites in ice areas of glaciers, andfor 2 sites in Alpine grassland of the test site Otztal,from ERS-1/ERS-2 tandem data of 4 different dates. As expected, the coherence of the ice areas on the glaciers is high for the two winter dates and low in August and September (Figure 2). However, in spite of strong melting in August, the degree of coherence reaches values up to 0.4 over parts of the ice and fim areas, thus being high enough to obtain fringes. For wet snow and ice the dominating backscatter mechanism is surface scattering. Apparently the geometric structure of the surface was preserved within the 24 hours between the image acquisitions because the overcast sky resulted in homogeneous melt conditions. The coherence of the fim areas on the glaciers shows a more complex behaviour than of the ice areas. The coherence is in general quite low in the August and September interferograms. In September the coherence reaches values above 0.4 only on the high plateau where the freezing depth in the snowpack is larger than at lower elevations. In winter with dry snow, 153 when the dominating part of the radar return originates from the volume, high coherence and stable conditions should be expected. However, between 6 and 7 December 1995 the signal decorrelated over parts of the fim areas, mainly on the high glacier plateau of Gepatschfemer. On other glaciers the coherence was quite high. The only possible explanation for this are major changes of surface structure due to wind erosion and snow drift, effects which may vary significantly at regional and local scales in dependence orography. In the unglaciated high Alpine areas the coherence is in general quite high throughout the year. A significant reduction is only observed for 27/28 September, related to major changes of meteorological conditions within 24 hours. The areas above the timberline are covered with low vegetation, mainly thin sedges and grasses, and some dwarf-shrubs. The surface is rough, and usually interleaved with rocks. In the sub-Alpine forests the degree of coherence is less than 0.4 throughout the year in the 24- hour repeat pass interferograms. Ice Motion in the Heimefrontfjella Mountain Range, Antarctica Investigations on coherence and ice motion were carried out in the Heimefrontfjella mountain range, Dronning Maud Land, East Antarctica. The area around Scharffenbergbotnen (74°35'S, 11°03'W) was selected, because measurements on microwave scattering and emission have been carried out in this region Rott et al., 1993 and because ice motion data are available for selected points (Stroeven and Pohjola. 1991). Seven ERS-1 SLC quarter scenes from January and February I994 in 3 day repeat intervals were available for the analysis. With this data set it was possible to derive absolute ice motion because there is a number of stable targets (nunataks) in the scene which can be used as reference. 154 N meter/year 0 15 30 45 60 75 92 Flight Dir Figure 3: Ice motion in range direction in the region of Heimefrontfjella, Antarctica. derived by means ofdifferential interferometry based on ERS-1 SAR data. Descrete steps ofice motion infalse colors are superimposed to an amplitude image. A - Aubertisen. P - Pionerflaket. S - Scharffenbergbotnen, V - Veststraumen. Figure 3 shows the results of the ice motion analysis for the full quarter scene, derived by means of differential interferometry from the image pairs 29 January/I February (Bn = 167 m) and 114February (Bn = 34 m). These two image pairs were selected for the motion analysis because of good coherence and short baselines. The Figure shows the motion in direction of the radar beam. The horizontal and vertical velocity components are not separated, but on the main ice streams, which show little inclination, the horizontal component is dominating. The ice plateau Pionerflaket, which rises to 2500 m a.s.l. and is visible at the right margin of Figure 3, blocks the iceflow from the inland. The ice on Pionerflaket, as well as the ice around the nunataks in the north-west of it (to the left in the Figure), shows very little motion. Ice motion data are available for Scharffenbergbotnen, 155 which is an ice inlet at 1250 m above sea level surrounded by nunataks with peak altitudes between 1400 m and 2200 m a.s.l. Scharffenbergbotnen contains blue ice fields, which are areas of negative surface mass balance due to sublimation of ice as a result of strong katabatic winds. The ice moves towards the mountains with velocities of less than one meter per year (Stroeven and Pohjola, 1991). Also the interferometric analysis shows that this ice is stagnant. The main transport of ice in the study area takes place through Aubertisen with the main flow direction towards north-west. A mountain ridge closely beyond the upper right corner of Figure 3 is an obstacle for the ice flow from above. Below this region of very low velocity two ice stream, from the south and from the west, are joining. The highest velocities with a maximum of 92 m per year are observed below this confluence. Below the slopes the ice flow turns towards west and joins the Veststraumen which is the major ice stream draining towards Riiser-Larsen Ice Shelf. On Veststraumen the magnitude of the velocity vector is about 30 % higher than the velocity component in range direction shown in Figure 3, because the angle between flow and look directions is about 45 degrees. The interferometric analysis provides remarkable detail on the complex ice motion pattern around the mountains. Coherence Study in Antarctica Target characteristics and temporal variations of coherence were investigated for the area of the motion analysis presented above. Figure 4 shows coherence images for the 3-day repeat interferogram with lowest coherence (23/26 January 1994)and with highest coherence (1/4 February 1994) among the available data. The baseline was small in both cases (54 m and 34 m). The area is almost completely covered with permanently dry firn which is strongly layered and has a one-way penetration depth of about 20 mat C-band Rott et al., 1993. 156 23 Jan / 26 Jan ~l4 B-eff = 54m 1 feb I 4 feb 94 N ' ·r;~ft7"···· 0.0 0.5 1.0 Flight Dir Figure 4: Interferometric correlationfor ERS-1 SAR imagespairs of 23126January 1994 (left) and 114February1994 (right). The grey scale corresponds to the degree of coherencefrom 0.0 to 1.0. Same area as Figure 3. For the firn areas the degree of coherence is less than 0.4 in the 23/26 January interferogram and reaches values above 0.8 in the 1/4 February interferogram. From meteorological observations at the Neumayer station, which is located in a distance of 450 km from the study area, it is known that a low pressure system with strong winds was passing by between 23 and 26 January, and that the weather in early February was comparatively calm. Strong winds are usually strongly modifying the surface roughness of dry firn, as also known from field campaigns in the Heimefrontfjella region. Thus it can be concluded that wind is the main reason for decorrelation of cold polar firn. In the February interferogram the coherence is low only on the steep backslopes of the ice plateau Pionerflaket where the signal to noise ratio is low. The noise signals are uncorrelated. High coherence is observed in all interferograms for the comparatively small areas of bare rock on nunataks, and for the blue ice fields and surface moraines which cover areas of several km2 size below the mountains. 157 Surface Moraine 1 500 CD . I = ;.1- r ;::;- .1100 ·a; 1 1 I I 11111_ u.~ - • -=-· ·- ••• 1:• (.1 - ~ •• ~) 0.6 '- ~ - (J U.4- < ~) u u.~. 0 -r--I 0 100 200 300 400 cff. Baseline Figure 5a: Degree of coherence for different time intervals of repeat pass interferograms (top) and degree of coherence in dependence of baseline in meters (bottom) for Surface moraine. The black dots in the top diagram indicate the effective baseline. 158 Blue Ice 1 son Q) c:: 0.8 - ~ 400 ·;;; Q} - ~ • en u • • r n:i ~ 0.6 '- r •• :;)UU CD Q) <1> > .s: I- • ,__•.. 0 0.4 '--- I• 200 'G (_) ,__ Jg 0 '>··'--- • Q) • iii I 100 ."- ~ " 0 0 3" 6 9 12 15 18 Ti me Difference in Days 1 - 0.8- •l) •• • • (; • • U.U - • •• •• •••• 'd> m 0.4 - -§ <> 0.._-» - u - 0 100 200 300 400 off. B Firn 1 ~·ID 0.8 500 Cl) c c ID ~00 ~ ~~ ro~ 00 0.4 ::soo coCl)::- 0.2 200 'GQ) 0 100 1i5 0 3 6 9 12 15 18 Time Diff~renoe in D~ys • ~ 0.6- rP • • .. - · 1 I 11 i:::I I ·· i ••.· 0-t--~~~r--~~~--~~~-r~~~--t-~ (1 100 200 :\00 cff. Baseline Figure 5c: As Figure 5a. Firn area. The temporal and baseline dependence of coherence for 3 selected sites is shown in Figure Sa , Sb ..2£._. The coherence of the surface moraine, which covers a level area in 159 Scharffenbergbotnen, is temporally stable at least up to the maximum investigated time difference of 18 days and decreases only slightly with increasing baseline. The baseline dependent decorrelation is small because surface scattering is dominating and spectral bandpass filtering according to the dominant fringe frequency was applied previous to the interferogram generation ( Gatelli et al., 1994). The blue ice field shows somewhat higher temporal variability. From surface observations it is known that sastrugi (snow dunes) may cover parts of the blue ice if the wind is not very strong, but the snow is blown away during storms. The baseline decorrelation is higher than for the moraine. Though surface scattering should be important for blue ice, a significant part of the backscattered signal originates from scattering at air bubbles within the volume. The fim areas show a very pronounced decrease of coherence with increasing baseline and with time. The main reason for the temporal variability are changes in surface conditions, as explained above. Though the magnitude of the backscattering coefficient is stable in time, and more than 90 % of the signal originate from the snow volume, the volume-surface interaction term seems to be an important mechanism for phase changes. The importance of the volume scattering contribution and the large penetration depth are reflected in the observed decrease between coherence and baseline, which is in principal agreement with theoretical calculations ( Gatelli et al., 1994). Conclusions The investigations with interferometric ERS data in the Alps and in Antarctica were able to provide insights in the relations between coherence and target characteristics. Melting is the main factor for decorrelation of snow and ice, causing in general almost complete decorrelation within one day, though under certain conditions the coherence may stay high enough for obtaining fringes. For dry snow and polar fim the main factor for temporal decorrelation is wind. Another important factor for reduced coherence in dry polar fim is the baseline-dependent decorrelation. Coherence characteristics enable the separation of different snow and ice regimes, such as blue ice fields and fim. The motion analysis of the Antarctic region confirms the unique capabilities of differential interferometry for studying complex patterns of ice dynamics with remarkable detail. Acknowledgement The investigations were supported by the National Space Research Program of the Austrian Academy of Sciences. The ERS-1 SAR data were made available by the European Space Agency for ERS Experiment A02.Al01. References Gatelli F., A.M. Guarneri, F. Parizzi, P. Pasquali, C. Prati, and F. Rocca, 1994: The wavenumber shift in SAR interferometry. IEEE Trans. Geosci. Rem. Sens., 32 (4), 855-865. J.-S. Lee., K.W. Hoppel, S.A. Mango, and A.R. Allen, 1994: Intensity and phase statistics of multilook polarimetric and interferometric SAR imagery. IEEE Trans. Geosci. Rem. Sens., 32 (5), 1017-1027. Rott H., K. Sturm, and H. Miller, 1993: Active and passive microwave signatures of Antarctic fim by means of field measurements and satellite data. Annals of Glaciology 17, 337-343. Rott H. and T. Nagler, 1994: Capabilities of ERS-1 SAR for snow and glacier monitoring in alpine areas. Proc. of Second ERS-1 Symposium ESA SP-361, 965-970. Stroeven A.P. and V.A. Pohjola, 1991: Glaciological studies in Scharffenbergbotnen. In: The Expedition Antarktis-VJll, Reports on Polar Research, 86, Alfred-Wegener-Institut, Bremerhaven,126-130. I I