zyxwvu zyxwv zyxw zyxwvutsr zyxwvut zyxwv Generation of Precise Wide-Area Geocoded Elevation Models with ERS SAR Data zyxwvutsr zyxw zyxwvut Marc BARA, Oscar MORA, Miquel ROMERO, Antoni BROQUETAs UniversitatPolitknica de Catalunya (UPC). Campus Nord, c/ Jordi Girona, 1-3,08034Barcelona, Spain. Phone / Fax: +34 93 401 1065/ +34 93 401 7232.E-mail: mabara@voltor.upc.es ABSTRACT In this p a p we present an improved technique for the generation of Digital Elevation Models (DEM), capable of dealing with full scene images (100x100 km) coming from an interferogram obtained with ERS satellite data. Starting from an interferometric processor aimed to the geocoding of smaller areas, now we expound the new improvements based an the use of Ground Control Points (GCP) in order to calibrate some imprecisions which appear in the case of very wide swaths. A generated DEM of the test zone of Tarragona (Spain) and its mor assessment are presented. In this paper we propose the use of GCP on the studied surface in order to calibrate the orbits. If the S A R image is available?it is possible to find several well-known points (like road crossings, lakes, etc.) in the amplitude image and their corresponding height values on the local datum, using other mapping souf%es. Figure 1 shows the process to know the satellite pitions with an accuracy of few cm.It is divided in two parts: in a first step, the orbits are calculated with the PRC files provided by D-PAF; in a second part, starting from the calibratim made with GCP, the orbits are calculated in a much more precise way. GEOCODING INTRODUCTION popular application of satellite S A R interferometry is the generation of high-quality Digital Elevation Models. In order to perform this work,an accurate phase to height conversion as well as an exact geocoding algorithm are required. ?he basis of the interferometric process has been widely described in many papers: image registration, range filtering, interferogram generation, flat Earth correction, h g e filtering and phase unwrapping [2]. After phase unwrapping the interferometric process provides a phase image which has to be converted into height information and geocoded to a The most A geocoding process [4] consists of three fundamental steps: - FWEarthreanoving. - Phase to height conversion. - Ground range projection. standard cartographic system. This is basically a geometric problem which requires a precise knowledge of both the orbit and the m e Earth shape (the local datum). In this paper we remark the importance of precise geometry and timing knowledge in order to minimise the errors when generating geocoded images: a technique which makes use of GCP to refine this information is presented. The usefulness of this methodology to generate wide-area maps has been checked with a full ERS tandem scene. THE ORBIT In or& to calculate the satellite position for ea& range line in the image, the 'Precise Orbits' files of D-PAF have been used. In the precise orbit file, the state vectors are spaced 30 s (225 km) and the area covered by the SLC image is around 100 km. Thus, it is necessary to inteapolate the data firom the two nearest ephemerids [l][3]. The problem is that the achieved precision (15 an in radial direction) is not enough for avoiding systematic errors due to the lack of precision in baseline. Therefore, additional information to refine both orbital positions is needed. 0-7803-5207-6/99/$10.00 Q 1999 IEEE. f 924 I Figure 1. Orbit generation layout zyxwvutsrq zyxwvuts zyxwvutsr zyxwvuts zyxwvuts zyxwvuts zyxwvu zyxwvutsr zyxw zyxwvut The first step demands an accurate knowledge of the azimuth time of the satellite the studied area.Indeed, the subtraction of the reference ellipsoid characterised by a necessary to know the 'Near Range Zero Doppler Time' very precisely. As an example, with a baseline of 100 m, a mistake of 1 p in the 'Near Range Zero Doppler Time' yields an mor of about 100 m in the range direction. The implementaimethod of Fig. 1offers a solution for this kind of errors since it is capable of correcting the times provided m the SLC headers to obtain accurate orbits. Then, the quality of the orbits is maximised and, consequently, the flatEarth subtraction (see Fig. 2) does not introduce additional distortions. The second step consists of converting phase to height in slant-range cocxdinates. This process is carried out accurately with a high or& conversion described in [11[31, using the range, Doppler and ellipsoid equations. However, after completing the conversion, some systematic phase erras can be observed when geocoding wide areas, which lead to height slopes in range and azimuth directions. Tkey appear basically as a result of the atmosphetic propagation and the wrong baseline calculation for each range line. In order to implement a second adjustment of the wave paths to both antenna positions, we propose the use of GCP, which can be useful to estimate these systematic inaccuracies. Then, with this new phase, the process is repeated to get highprecision height i&mnation. The third step is oriented to the conversion of the height information &om slant-range coordinates onto a standard cartographic reference system, like UTM. In this way, we dispose of precise DEM which cau be integrated, for instance, into a GIs. RESULTS WITHERS DATA Using this technique we have geocoded the area of Figure 2. Geocodm . glayout Tarragona (Spain) with data collected by ERS-1 and ERS-2 with a 100-mbaseline. This m e is quite heterogeneous, with ~ v a and , a -an(% the range of w@*c The reference DEM has a 2 m rms vertical accuracy and a features from the flat area in Delta del Ebro to the mountains 30 @-id in the idand 1200 m. Mweoverv some a ~ ~ P h e r i c To perform a detailed study, we have selected a smaller artifacts can be observed at Df?lZU del Ebr-0 Which yield mea which is affected by different kind of erras, since it important height mors, as it is shown in the next s d o n . comprises the Delta del Ebro (flat zone) and some mountains. The w e to height C X X W ~ has S ~ ~ ~ peafarmed using The result is shown in Fig. 4. As we can see, there are some 14 CElntrol points spread over the F m Y , the m P has amospheric artifacts in the flat area, as well as shadowing with a 3 - m spacing*Figure b X i gmoded to a uTh'l and layover phenomenons in the mountainous arm They shows a 3D rendering of the wik-area DEM. produce the most significant errors in the obtained DEM. For example, the atmospheric artifacts at Delta del Ebro generate ERROR ASSESSMENT hills up to 100 m, while in the shadowed zones of the mountains the error reaches the value of 240 m. In this section we present a comparison between the obtained H ~this kind ~ of inamad= ~ ~can not ~be attributed , to DEM with a high accuracy reference provided by the Znsfifut the interferometric since they the Cartogrdjicde Catalunya (ICC). acquisition geometry itself. 1925 zyxw zyxwv zyxw zyxw zy zyxwvuts zyxwvuts zyxwvut zyxwvuts Figure 4. Error between the InSAR DEM and the ICC reference DEM. These new parameters, added to the precise mection of the ellipsoidal-Earth tenn,a high order algorithm of phase to height conversion and the support of control points allow the Figure 3.3D perspective view of the obtained DEM generation of accurate wide-area DEMs from ERS S A R data without the appearance of systematic errors. The methodology has been checked with real data, and a detailed The geocoding process provides the highest precision in the error assessment has been presented in order to discuss the favourable zones of the interferogram. For instance, in Delta performance of satelliterepeat-pass interferometry. del Ebro we have obtained a rms mor of approximately 16 m. In the mountains, as a result of shadowing and layover, ACKNOWLEDGMENT this value degrades to 56 m. Indeed, we can classify two different type of errors: The authors would like to thank the CICYT (Spanish Commision for Science and Technology) and the CIRIT Errors associated to rough topography: Interferometric (Catalan Commission for Research) for the financial support coherence is degraded due to the high slopes of the terrain. of this work. They also acknowledge the ICC (Cartographic Moreover, the topography itself makes difficult the Institute of Catalonia) for providing the DEM used for the interpolation where less data is available. error assessment. Errors associated to atmospheric arttjluts: These artifacts are the responsible for the generation of small hills up to a REFERENCES height of 100 m in Delta del Ebro. [l] D. Carrasco, D. Esteban, F. Upez, J. Tena,0. Rodriguez, CONCLUSIONS M. Bara, A. Broquetas, “Precise geometry techniques for ERS S A R Interferometry”, European Conference on When geocoding large interferograms, the exact knowledge Synthetic Aperture Radar, EUSAR’98,1998. of satellite geometry and time parameters is extremely [2] F. Rocca, “An overview of ERS SAR interferometry”, important. Orbit precisions of centimetres and time accuracies Roc 3d ESA ERS Symposium, Florence, May 1997. of nanoseconds are required. In this paper we have shown an [3] D. Carrasco, “SAR Interferometry for digital elevation operative technique to refine the initial supplied orbit model generation and differential appliations”, WD parameters Dissertation, Marc41 1998. [4] G. Schreider, “SAR Geocoding: Data and systems”, Wichmann ed. 1993. 1926