Scientist’s Idealism versus User’s Realism onRadarsat-2 HR Stereo Capability without GCP: TwoCases over North and Arctic Sites in Canada
tHierrY toUtin, kHaliD oMari, enriqUe blonDel, Ottawa, Daniel ClaVet, SHerbrooke
& Carla VaneSSa SCHMitt, Ottawa, Canada
Keywords: Radarsat-2, SAR, high-resolution, radargrammetry, DSM, Canadian Arctic
Summary: Digital surface models (DSMs) are ex-tracted from high-resolution Radarsat-2 (R2) stereoimages using our new hybrid radargrammetricmodeling without ground control point developedat the Canada Centre for Remote Sensing. They arethen evaluated over two Canadian northern andArctic study sites: the irst for the scientiic valida-tion and the second in the Arctic (steep relief andglaciated surfaces) for the operational evaluation.For the validation site, the bias and elevation linearerrors with 68 percent conidence level (LE68) ofR2 stereo-extracted DSM compared to lidar datawere computed over bare surfaces: LE68 of 3.7 mand no bias were achieved. For the Arctic studysite, a large negative bias of 18–20 m and LE68 of21–30 m were computed versus ICESat data overand outside ice ields, respectively. In additionLE68 of 15 m with equivalent bias was obtainedover ice ields with 0°–5° slopes, which generallyoccurred in ice ields. The large biases certainlysuggest a bias in the R2 stereo-model and thus inthe metadata used in our hybrid model computa-tion.
schaftlers und die Realität des Nutzers hinsichtlich
der Möglichkeit von Radarsat-2 zu hochaufgelös-
ten Stereoaufnahmen ohne Passpunkte: Zwei Fall-
beispiele für nördliche und arktische Gebiete in
Kanada. Digitale Oberlächenmodelle (DOM), dievon hochaufgelösten Radarsat-2 (R2) Stereobildernmit Hilfe eines neuen am Kanadischen Fernerkun-dungszentrum entwickelten hybriden Radargram-metrie-Modells abgeleitet wurden, welches ohnePasspunkte auskommt, werden anhand von zweiTestgebieten in nördlichen bzw. arktischen Regio-nen Kanadas evaluiert. Die Evaluierung im erstenTestgebiet dient der wissenschaftlichen Validie-rung, während jene im arktischen Testgebiet (stei-les Gelände, Eis als Untergrund) der Untersuchungoperationeller Gesichtspunkte dient. Für die Vali-dierung wurden im Gebiet der systematische Fehlerund die linearen Höhenfehler des DOMs aus R2-Daten mit dem DOM aus Laserscanning-Daten aufvegetationslosem Boden verglichen und das 68%Konidenzniveau (LE68) bestimmt: LE68 betrug3.7 m. Ein systematischer Fehler konnte nicht fest-gestellt werden. Für das arktische Testgebiet ergabsich an bzw. außerhalb von Eisfeldern ein großernegativer systematischer Fehler von 18–20 m undein Wert des LE68 von 21–30 m im Vergleich zuden Daten von ICESat. Darüber hinaus ergabensich ein Wert für LE68 von 15 m und ein äquivalen-ter systematischer Fehler über Eisfeldern, die mit0°–5° geneigt waren, was eine typische Neigungvon Eisfeldern darstellt. Die großen systematischenFehler legen einen Zusammenhang mit systemati-schen Fehlern im R2 Stereomodell und somit in denMetadaten, die in unserem hybriden Modell ver-wendet wurden, nahe.
In summary, the choice of SAR stereo-im-ages should resolve the following issues:● to have an operational SAR system for ac-quiring the planned data from shallow lookangles at the best season corresponding tothe end of the melt season with no snow orlittle soaked snow and supraglacial debrisdue to freezing/melt process for soaked ice/snow,
● to reduce temporal resolution and then tem-poral radiometric changes with the orbitconvergence and a larger choice of appro-priate look angles (from 35°– 50°),
● to reduce illumination and radiometric dif-ferences with images having close look an-gles,
● to increase the SAR backscatter on ice withshallow look angles due to higher surfaceroughness with supraglacial debris after themelt season,
● to increase the matching performance andreduce the mismatched areas due to less ra-diometric variations with small intersectionangle, and
● to reduce the SAR depth penetration overthe “wet or soaked” 1st-ice layer covered bysupraglacial debris with shallow look an-gles (Dall et al. 2001).
Due to the remote and harsh environmentsof the Canadian Arctic, the 3D geometricprocessing of SAR images should require noground control points (GCPs) collected by us-ers for the operational applications. A new hy-brid radargrammetric model (toUtin&oMaRi2011) was recently developed for Radarsat-2(R2) at the Canada Centre for Remote Sens-ing and thus used for the stereo-modelingand the generation of digital surface models(DSM) without GCP. This hybrid model com-bine the deterministic toUtin’s model (toUtin& CHénieR 2009), the rational function model(RFM) empirical model and the Radarsat-2associated metadata. In order to irst evaluatethe performance of the stereo-radargrammet-ric process in a well-controlled research envi-ronment (scientist’s idealism) and after in anoperational environment (user’s realism), twoCanadian northern and Arctic study sites wereused: the irst one for the scientiic validationand the second one for the operational evalu-ation in Arctic. This last experiment was the
1 Introduction
Because Canadian Arctic suffers from poorly-known relief and the surface state of glaciat-ed regions is rapidly evolving due to snowfall,snow transport by wind and/or surface melt,remote sensing data (synthetic aperture radar(SAR) or optical) used to retrieve the topogra-phy must be acquired with the shortest possi-ble time interval to maximize their coherenceor correlation. This is also important becausethe low of the glacier (up to a few metres perday) can bias the topographic measurementduring this time interval.Stereo radargrammetry using SAR data, as
developed since 1960’s by la pRaDe (1963),RosenFielD (1968), la pRaDe & leonaRDo
(1969), lebeRl (1972, 1978) and many otherafterwards, can thus be an appropriate solu-tion in Canadian Arctic, even if there was noprevious radargrammetric experiment overglacierized region, to the best of our knowl-edge (toUtin 2011). Because the state-of-the-art of radargrammetry and DSM generationusing SAR data can be found in many detailsin lebeRl (1990) and toUtin& gRaY (2000), itis thus more important to focus on the winningconditions of stereo-radargrammetry in thechallenging conditions of a glacierized region.The multi-date across-track capability is notso problematic due to different SAR advantag-es speciic to ice regions. First, the backscatterof SAR sensors is more dependent on the ru-gosity or the dielectric component, which en-able more radiometric contrast and less depthpenetration over ice ields covered with su-praglacial debris, rock glaciers, moraines, etc.Second, the SAR sensors are operated in all-weather conditions and not dependent on thesolar illumination conditions, which thus can-celled the large shadowed areas with opticaldata in high latitudes. Third, the convergenceof heliosynchronous orbits to North/Southpoles combined with a large range of look an-gles (20°–60°) gives thus a strong advantageto drastically reduce the temporal variationsto few days between the multi-date stereo-im-ages acquired in the highest latitudes. Last butnot least, the new satellite SAR sensors havenow high-resolution (HR) capabilities (sub tofew metres), and are dedicated toward opera-tional applications.
Thierry Toutin et al., Scientist’s Idealism 387
The elevation ranges almost from 10 m in thecity in the southeast to around 1000 m in theCanadian Shield in the north. The northernpart is a hilly to mountainous topography (5°–30° slopes) mainly covered with forests (de-ciduous, conifer and mixed) while the southpart is a semi-lat topography (0°–5° slopes)with urban and residential areas. This site is tovalidate the quality output products in a well-controlled research environment: the scien-tist’s idealism.The second study site (operational site here-
after) located in the Bafin Island, Nunavutat approximately 70° 50’ N and 71° 30’W(Fig. 2, left). There is no vegetation cover, ex-cept small plants. Even if there is some residu-al snow in the highest altitude on the Landsatortho-image acquired at the beginning of themelt season in August (Fig. 2, right), the iceields are similar to the exposed ice surfacesof Jakobshavn Isbrae, west Greenland, wherea depth penetration of 1±2 m with C-band ra-dar was found (Rignot et al. 2001). More than80% is covered by ice ields with cirque gla-ciers (permanent ice-covered mountains), out-lets and valley glaciers and glaciers tonguessurrounded by spectacular fjords with 70°–90° cliffs of 500–800 m height. Bare surface
irst application of stereo-radargrammetry onglacierized region.
2 Field Study Sites
The irst study site (validation site hereafter) islocated north of Québec City, Québec, Canadaand spans different environments: urban andresidential, semi-rural and forested (Fig. 1).
validate our new hybrid model accuracy. Inaddition, DSM (20 km by 20 km, 1 m spac-ing) were obtained from a irst-echoed returnlidar survey collected by GPR Consultants inOctober 2009 (Fig. 3), and covered the full R2stereo pair. When compared to the previousDGPS, the positioning and altimetric accura-cy (1 σ) of the lidar was computed to be betterthan 0.3 m and 0.1 m, respectively.For the operational site, the R2 stereo im-
ages were acquired in 2009 from descend-ing orbit with the ultra-ine mode (U modewith 3 m resolution) in HH polarization: U12(Fig. 4) and U26 (20 km by 20 km; 1.6 m by2.5–3.0 m pixel spacing) on September 28with 38.83°–39.84° right look angles and Sep-tember 9, 2009 with 48.12°–48.93° look an-gles, respectively. This acquisition period cor-responded (i) to the end of the melt season and(ii) to -3°/+3° temperature range generating afreezing/melt process during the night/day, re-spectively. It results, especially with shallowangles that the 1st layer of the exposed “wet”ice ields covered by supraglacial debris andof the remaining soaked snow in the highestaltitudes reduced the SAR depth penetrationto its minimum (less than 1 m) due to a highwater content and dielectric component (Dallet al. 2001, gRaY 2011). The radiometry ofthese stereo data are, however, dominated bythe geometric issues due to high relief with no
mountains also with steep slopes surround theice ields and glaciers. The valley glacier in thesouth-east is about 1-km wide with up-to-4°slopes surrounded by 600 m height bare sur-face and cirque glaciers. The elevation rangesfrom sea level to 1840 m and the slopes varyfrom 0°– 90° at fjord cliffs, illustrating a verychallenging environment (in terms of landcover and relief). This site is to validate thequality output products in operational map-ping environment: the user’s realism.
3 Data Acquisition andProcessing
For the validation site, the R2 SAR dataset in-cluded two stereo images (20 km by 20 km)acquired September 10 and 14, 2008 with theC-band ultra-ine (U) mode (1 by 1 look; 1.6–2.4 m by 3 m resolution) in VV polarizationfrom descending orbits with view angles of30.8°–32° (U2 Fig. 1) and 47.5°–48.3° (U25) atthe near-far edges, respectively.The reference cartographic data included
ground points, mainly road intersections andelectrical poles, collected from a differentialglobal positioning system (DGPS) survey inNovember 2008 with 3-D ground accuracy of10–20 cm. The collected points were used asindependent check points (ICPs) to quantify/
Fig. 3: Digital surface model (20 km by 20 km, 1-m spacing) from a lidar survey.
Thierry Toutin et al., Scientist’s Idealism 389
grid spacing (around 25 m at 70° latitude), toremove potential elevation anomalies result-ing from clouds or valley fog.The processing steps for DSM generation
with HR SAR stereo-images were previouslyaddressed and documented (toUtin 2010). Thenew hybrid toUtin’s model developed for theradargrammetric processing of R2 at CCRSdoes not require any GCP collected by theuser (toUtin & oMaRi 2011): it only uses theinformation in the metadata of the images forcomputing its parameters. The hybrid modelhas been proven to be 25-cm precise (toUtin& CHénieR 2009), and the accuracy (1 σ) of theresults in stereoscopy was better than one pix-el with one-pixel biases in the three axes. Themain processing steps, which used CCRS sci-entiic software (steps 1–3 & 6) and commer-cial OrthoEngineSE software of PCI Geomat-ics1 (steps 4 & 5), are:1. Acquisition and pre-processing of the SLCSAR images (speckle iltering with adap-tive Frost SAR ilter using 11 by 11 window)and metadata;
2. Collection of 60 ICPs from the DGPS sur-vey, only for the validation site;
3. Computation of our hybrid models and theirvalidation with ICPs (systematic and ran-dom errors) (toUtin & oMaRi 2010);
4. Elevation extraction using a 7-step hierar-chical grey-level image matching (meannormalized cross-correlation method withsub-pixel computation of the maximum ofthe correlation coeficient) performed on thequasi-epipolar stereo-images, (ostRoWski& CHeng 2000). Different matching param-eterization for both sites were tested to in-crease the matching success and to reduceelevation interpolation (toUtin 2010);
5. Edition of the quasi-epipolar DSM (blun-ders removal, gap interpolation, water bod-ies), and geocoding of the edited quasi-epipolar DSM.
6. Evaluating (systematic and random errors)the geocoded DSMs with the lidar elevationdata (toUtin 2010).
1 In 2012, all steps can be operationally performedwithin PCI Geomatics.
vegetation cover: more severe layover in U12over the east-oriented slopes and more occlud-ed areas over the west-oriented slopes in U26.In fact, the south-north curved land-waterboundary of the left island represents the clifflayover over the ocean, and not the “smooth-er” coast line of a glacial-eroded fjord, as wellas part of the low lands along this coastlinecliff thus “disappeared” due to severe layover.On the other hand, it is almost impossible todiscriminate the true water-land boundary forthe opposite coastline cliffs due to the SARshadow/occluded areas.We can notice on Fig. 2 and Fig. 4 that all
coastline cliffs and most cirque glaciers ashaving strong slopes (60°–90°) while the iceields with their outlet glaciers have in gen-eral lower slopes (0°–20°). However, steep20°–90° slopes also occurred in some iceields. The lidar ICESat data (ascending anddescending tracks) over the 27F13 map sheetwas extracted from GLA14 product (L2 glob-al land surface altimetry data) Release 28, 29and 31 over the full life cycle of the mission2003–2008. Conversion to Canadian refer-ence systems was applied to translate ICESatGLAS data into orthometric heights. ICESatdata points were spatially iltered: (1) horizon-tally to remove redundant values within 100-m radius, and (2) vertically within 1-arcsec
water bodies), a basic cubic interpolation wasperformed to ill the gap without introduc-ing signiicant additional errors. Visually, theDSM in the quasi-epipolar geometry beforethe geocoding (Fig. 5) is smooth and well de-scribes the macro-topography and the macrolinear trends with mountains and valleys, en-hancing the structural geological frameworkin the northwest-southeast direction. Themountains and valleys are generally smooth,being a good representation of a Precambriangeomorphology (smoothed-glacial and erodedtopography).The quantitative evaluation was performed
over the coverage of the lidar data and DSM,approximately 20 km by 20 km correspondingto 100% of the stereo-pair and the elevationdifferences between the lidar and DSM aregiven in Tab. 2. However, the results comput-ed over the full lidar/R2 overlap (Tab. 2, Allsurfaces) do not relect the true DSM accuracysince other sources in the error budget comesfrom: (1) the footprint and penetration in thevegetated cover are different for both sensors(SAR and lidar); and (2) the compared stereoSAR and lidar points are not at the same eleva-tion in the vegetated cover (70% of lidar cov-erage). These errors are thus relected in the-1.9 m bias due to more depth penetration oflidar in forested areas and 4.7 m LE68 due todifferential depth penetration of lidar depend-ing of tree species and densities. The negativebias is consistent with Z-bias error computedover ICPs (Tab. 1). To have the true elevationaccuracy, the error evaluation was thus per-formed only on surfaces where the lidar andstereo SAR points were at the same ground el-evation (Tab. 2, two last lines). The bare sur-faces included the bare soils and the imper-vious areas: in the country (ields, highways,roads) and in the urban parts (streets, private/public parking and gardens, etc.) and are rep-
4 Results and Discussions
Results are related to the systematic and ran-dom errors (1) of the hybrid radargrammetricmodels computed over ICPs and (2) of the ste-reo-extracted DSMs computed over the lidarelevation data.
4.1 Radargrammetric Hybrid ModelEvaluation
Because there was no control data in the op-erational site, Tab. 1 only summarizes the re-sults of the radargrammetric modelling com-putation for the validation site and dataset pre-viously described: the errors (bias and stand-ard deviation (Std) in metres) computed over60 DGPS ICPs providing independent andunbiased evaluations of the modelling accu-racy. Biases of one pixel (or half SAR resolu-tion) are obtained. Similarly, Std results in theorder of one pixel are better for X-direction.It is certainly due to the better knowledge ofthe range direction than the azimuth directioncorresponding to the satellite displacement.Both results in Z-direction are also very goodversus the SAR resolution and the same-sidestereo geometry. These results are compara-ble (10% difference), but a little worse in theY-direction, to the original radargrammetricmodel computed with user-collected GCPs(toUtin & CHénieR 2009). On the other hand,the small lost in accuracy for the hybrid modelis compensated by the gain of processing thestereo images without GCP.
4.2. DSM Evaluation
Validation site
Due to the few percents (less than 3%) ofsparse and small mismatched areas (mainly on
Tab. 2: Differences between lidar data and R2DSM over different surface types: Bias andLE68 in metres.
Surfaces Bias (m) LE68 (m)
All surfaces −1.9 4.7
Bare surfaces −0.9 3.7
Urban surfaces −0.5 3.2
Tab.1: Modelling results over 60 DGPS inde-pendent check points (ICPs) for the validationsite: Bias and standard deviation (Std) in metres.
Bias (m) Std (m)
X Y Z X Y Z
1.8 2.6 −2.7 0.93 1.33 2.34
Thierry Toutin et al., Scientist’s Idealism 391
lages in the high lands and slopes in the north.No signiicant difference was noticed insideor outside the urban parts, when the match-ing was successful: almost no bias and LE68of 3.7 m and 3.2 m are thus obtained for baresurfaces and urban surfaces, respectively. The
resentative of the full terrain relief becausethey occur not only on low lands and slopesbut also in the high lands and slopes (mainly,in the northeast). The urban surfaces includedthe buildings only and occurred generally onlow land and slopes in the south and few vil-
Fig. 5: Shaded-relief DSM (18 km by 16 km) of the validation site in the R2 quasi-epipolar geom-etry before the geocoding; the main city on low elevations and slopes is in the southeast part.
Fig. 6: Left: Geocoded DSM of the validation site with the 12-strip lidar overlaid: Strip 1 at far edgeand Strip 12 at near edge. Right: LE68 (in metres) for each strip.
the end of the melt season decreased the depthpenetration (Dall et al. 2001), increased theroughness and thus the radiometric contrast.During this period, the wet/soaked ice due tofreezing/melt process covered by supraglacialdebris and dust offered its maximum degreeof surface texture and a high backscatter coef-icient with shallow SAR look angles. It is theother reason of few percent mismatched areas,which were illed in with the same interpola-tion without introducing signiicant additionalerrors in ice ields. Conversely, the DEM looksvery strange along all coast cliffs displayingover 50° slopes and large geometric and ra-diometric differences between the two imag-es. Instead of generating mismatched pixels,wrong pixels were correlated. The combina-tion of these geometric and radiometric distor-tions, which only occurred in such a challeng-ing Arctic study site, would certainly impedeany image matching system and would havegenerate either mismatched or wrongly corre-lated pixels.The quantitative evaluation was per-
formed with ICESat lidar. The height meas-urements of land surface will be the prime in-terest for DEM quality assessments (zWallYet al. 2002). While the total number of ICE-Sat data from ascending and descending or-bits is limited to thousand points (Fig. 8, blueand red footprints), it will be more limited be-cause ICESat accuracy strongly degrades withslopes. Because the primary goal of ICESat isto measure inter-annual and long-term varia-tions in the polar ice-sheet elevation and vol-ume of Greenland and Antarctica, there wereno absolute validation results over more than5° slopes. Consequently as a function of theexpected accuracy for R2 DSM, we only con-sidered for R2 DSM evaluation the ICESatdata on slopes less than 30° (Fig. 8, blue foot-prints). Tab. 3 gives the computed differencebetween ICESat data and R2 DSM outside(326 points) and over ice ields (387 points).Because there was no signiicant bias in
the validation site using the same processing,the two biases in this operational site suggesta systematic error in the stereo-model of R2and thus in the metadata used for this stereomodel computation and not from the stereo-radargrammetric processing. While uncer-tainty about the metadata reliability could
slightly better results over urban surfaces canbe due to the fact that most of them generallyoccurred in low slopes.The last accuracy evaluation was performed
as a function of the SAR range. The lidar datawas cut into 12 regular strips (except strip 1),more and less parallel to SAR image azimuth(Fig. 6, left). Strip 1 was at the far edge andstrip 12 at the near edge. LE68 (in metres) wasthus computed for each strip (Fig. 6, right) andshows there is no correlation between the el-evation accuracy and the SAR range, whichconirms that the radargrammetric model isnot range or look-angle dependent. The resultof strip 1 is less signiicant because there is notenough data when compared to other strips(less than 40%).
Operational site
The R2 DEM is displayed in Fig. 7 with theice ield and glaciers boundaries (in red) andsupraglacial debris and moraines boundaries(in blue). The DEM looks relatively smoothover the ice ields, even with the backscatterhomogeneity in ice covered areas, mainly dueto the choice of the matching parameteriza-tion. In addition, the planned R2 acquisition at
Fig. 7: R2 DSM of the operational site with iceield and glacier boundaries (in red) and su-praglacial debris and moraine boundaries (inblue) overlaid.
Thierry Toutin et al., Scientist’s Idealism 393
models and R2 metadata was evaluated withR2 stereo data for DSM generation over twostudy sites in the north and Arctic of Canadawithout GCP. The validation site having ac-curate control data enabled DSM accuracy of3.7 m (LE68) with no signiicant bias (-1 m)over bare surfaces to be computed. Even bet-ter LE68 (3.2 m) was obtained over buildings,due to their location generally in low slopes. Itis certainly the scientist’s idealism!The operational site, a challenging environ-
ment with glaciated surfaces, fjords and steeprelief, was used to evaluate the mapping po-tential of the method in the Canadian Arcticwithout control data. The R2 data acquisitionwas planned to reduce the SAR depth penetra-tion to its minimum. In this remote and harshenvironment, DSM LE68 of 21 m with largebias (-18 m) was achieved over ice ield. Thismajor part of this bias is certainly due to a sys-tematic error in the metadata, which can eas-ily be corrected with water bodies or the sealevel, which frequently occurred in ice ields.However, the uncertainty of the metadata reli-ability could slightly impede the operationalexploitation of this method due to a possibleZ-bias to be corrected: it is maybe the user’srealism!While other methods using optical and SAR
systems could achieve similar and even bet-ter results, this application demonstrated thecapability of R2 to generate DSM with bet-ter than 21 m LE68 without collecting controldata over ice bodies depending of the terrainslopes (0°–30°) or around 18 m over ice sheets(slopes less than 5°) when the radar penetra-tion is small (less than few metres). This newmethod increases the applicability of R2 to re-mote and harsh environments even if there isa slight loss in accuracy when compared to thesolution using control points. It is largely com-pensated by the gain of no control data.
impede the exploitation of the hybrid modelwith R2 data without GCPs, water bodies orthe sea level, frequently occurring in ice re-gions, can be used for correcting the resultingZ-bias. Because the bias is negatively largerover the ice ields, it is good indication thatthere was (i) no signiicant or measurable el-evation lowering or ice thinning between thetwo data acquisitions and (ii) there is no depthpenetration. The LE68 over ice ields is a lit-tle better (21 m) than outside (30 m) due toless steep slopes over the ice ield. In addition,most of ice bodies have low slope relief, otherstatistical results of elevation differences be-tween ICESat and R2 DEM show that LE68 isstrongly correlated with slopes: LE68 of 18 mwith an equivalent bias was then computedover ice ields (103 ICESat points) within 0°–5° slopes.
Conclusions
A new hybrid radargrammetric model com-bining toUtin’s deterministic and empirical
Tab. 3: Differences between ICESat blue footprints and R2 DSM over different surface types:Bias, LE68 and minimum/maximum (Min./Max.) in metres.
toUtin, t., 2010: Impact of RADARSAT-2 SARUltraine-Mode Parameters on Stereo-Radar-grammetric DEMs. – IEEE Transactions onGeoscience and Remote Sensing 48 (10): 3816–3823.
toUtin, t., 2011: Digital elevation model generationover glacierized regions. – singH, V.p., singH, p.& HaRitasHYa, U.K. (eds.): Encyclopedia ofSnow, Ice and Glaciers: 202–213, Springer.
toUtin, t. & CHénieR, R., 2009: 3-D Radargram-metric Modeling of RADARSAT-2 UltraineMode: Preliminary Results of the GeometricCalibration. – IEEE Geoscience and RemoteSensing Letters 6 (2): 282–286 & 6 (3): 611–615.
toUtin, t. & gRaY, a.l., 2000: State-of-the-art ofextraction of elevation data using satellite SARdata. – ISPRS Journal of Photogrammetry andRemote Sensing 55 (1): 13–33.
toUtin, t. & oMaRi, k., 2011: A new hybrid model-ing for geometric processing of Radarsat-2 datawithout user GCP. – Photogrammetric Engineer-ing and Remote Sensing 77 (6): 601–608.
zWallY, H.J., sCHUtz, b., abDalati, W., absHiRe, J.,
bentleY, C., bRenneR, a., bUFton, J., Dezio, J.,
HanCoCk, D., HaRDing, D., HeRRing, t., Min-
steR, b., qUinn, k., palM, s., spinHiRne, J. &tHoMas, R., 2002: ICESat’s laser measurementsof polar ice, atmosphere, ocean, and land. –Journal of Geodynamics 34 (3–4): 405–445.
Dr. Daniel ClaVet, Centre for Topographic Infor-mation, 2144 King Street West, Sherbrooke, Que-bec, J1J 2E8, Canada, Tel.: +1-819-564-5600, ext:255, Fax: +1-819-564-5698, e-mail: [email protected]
Manuskript eingereicht: Oktober 2011Angenommen: Februar 2012
Acknowledgements
The authors would like to thank paUl bRianDand the Canadian Space Agency for support-ing and inancing this research under theirSOAR and GRIP programs. They also thankDr. pHilip CHeng and PCI for the adaptationand integration of CCRS math model and al-gorithms in OrthoEngineSE of PCI Geomat-ica.
References
Dall, J., MaDsen, s.n., kelleR, k. & FosbeRg, R.,2001: Topography and Penetration of the Green-land Ice Sheet Measured with Airborne SARInterferometry. – Geophysical Research Letters28 (09): 1703–1706.
gRaY, A.L., 2011: personnel communication.la pRaDe, g., 1963: An analytical and experimen-tal study of stereo for radar. – PhotogrammetricEngineering 29 (2): 294–300.
la pRaDe, g. & leonaRDo, e., 1969: Elevationsfrom radar imagery. – Photogrammetric Engi-neering 35 (42): 366–371.
lebeRl, F.W., 1972: On model formation with re-mote sensing imagery. – ÖsterreichischeZeitschrift für Vermessungswesen 2: 43–61.
lebeRl, F.W., 1978: Satellite radargrammetric. –156 p., Deutsche Geodätische Kommission, Mu-nich, Germany, Serie C 239.
ostRoWski, J.a. & CHeng, p., 2000: DEM extrac-tion from stereo SAR satellite imagery. – IEEEGeoscience and Remote Sensing Symposium(IGARSS) 2000 (5): 2176–2178, Honolulu, Ha-waii.
Rignot, e., eCHelMeYeR, k. & kRabill, W., 2001:Penetration depth of interferometric syntheticaperture radar signals in snow and ice. – Geo-physical Research Letters 28 (18): 3501–3504,doi: 0094-8276/01/2000GL012484.
