RAPID - A JOINT PROPOSAL USING InSAR

ABSTRACT

RAPID is an approach to implement an end to end system, from the interferometric processing to application of SAR and InSAR products. It is carried out in the framework of the Announcement of Opportunites for ERS-1/2 at different Universities in Germany and Austria and the DLR. InSAR is the basic part of the project. In this CO'I project the DFD is implementing an pre-operational system to produce necessary interferometric products for some selected applications. At the current implementation level it is possible to generate coherence maps, unwrapped phase information, single Digital Elevation Models and geocoded DEM. The paper will present first results from the test area in Hungary. DEM's and coherence maps will be an essential input to the application projects in RAPID. With ERS tandem mission data coherence information are used for a monitoring project of the rainforest in Brazil. In Africa geomorphological studies uses also information from different tandem pairs to extract information like e.g. movments of sand dunes. In Germany the SAR data are input for the monitoring of two selected areas, one in the Rhine Valley and the other near the lake Steinhude. In some very hilly testareas in the Austrian alps it is planed to generate DEM with InSAR and compare them with available DEM's. In a second step a combination of ascending and descending pathes will be used to reduce geometric effects. An extraction of linear features from the SAR will show first interesting results.

Keywords: ERS-SAR, Interferometry, Processing, Applications

1. INTRODUCTION

The joint proposal RAPID (Radar Amplitude, Phase and Interferometric Detection) gathers nine research groups ( from Germany, Austria, Hungary and Phlippines) with the intention to use not only the ERS SAR amplitude information but also the interferometric possibilities of the mission. Rapid is considered as innovative scientific proposal exploiting ERS data either alone or in synergy with other data for land surface processes and microwave interaction mechanisms, including also climatological purposes. The tandem mode of ERS-1 and ERS-2 will be used as well as the synergetic use of ERS data from more than one instrument on ERS-1 or ERS-2.The main reason for a joint proposal is to establish a solid interaction between researchers developing algorithms for new data types from the imaging SAR sensor (i.e. phase information, coherence maps, interferometric parameters) and those who apply these data in different regions of Europe, Africa and South America, for different research application purposes and in synergy with different remote sensing and ancillary data. The application will then give feedback in an iterative way to the group developing new software for basic SAR data.

Figure 1. RAPID Project Scheme

2. DEVELOPMENT OF PRE-OPERATIONAL PROCESSING CHAIN AND GENERATION OF A LARGE AREA DEM

In the InSAR part of RAPID a pre-operational processing chain for large areas was implement and a weighted least square approach to phaseunwrapping was implemented and tested. In a test area of Hungary these approaches were tested with sufficient results. The internal accuracies of different derived DEM are in a dimension of 20 m to 40 m.

A mosaicking tool specially for terrain models was developed and implemented in a UNIX/C configuration. It was possible to generate a mosaic of 15000 square kilometers. The validation bases on national topographic maps.

Figure 2. InSAR processing system

3. LINEAR STRUCTURE IN SAR COHERENCE DATA

The goal of the subproject from the University of Munich is to simplify mapping linear structures with Synthetic Aperture Radar (SAR) data. This is necessary as the specific conditions of SAR such as speckle and the dependence on incidence angle have resulted in either a high interactive effort or a rather insufficient output of previous approaches, especially when compared with results of mapping with optical imagery. Therefore, a specific approach is developed which accounts for SAR characteristics. It integrates several data sources such as SAR intensity images as well as coherence maps from an interferometric evaluation of two SAR scenes. Also given geographic information can be used to support the extraction.

A new approach to detect lines in noisy images using a Markov random field (MRF) model and Bayesian classification is proposed. The unobservable object classes of single pixels are assumed to fulfill the Markov condition, i.e. to depend on the object classes of neighboring pixels only. The influence of neighboring line pixels is formulated based on potentials derived from a random walk model. Locally, the image data is evaluated with a rotating template.As SAR intensity data is deteriorated by multiplicative noise, normalized intensity ratio is used as the response of the local line detector resulting in a constant false alarm rate. Besides maximum a posteriori (MAP) and iterated conditional modes (ICM) estimation of the object parameters, an implementation of local highest confidence first (LHCF) estimation is used. It is initially applied to the sites which are most probably structures in object space, and is then allowed to progress to regions less promising for line detection depending on the results of previous iterations. In this way processing times are substantially reduced.In an testarea located in Siberia it was possible to detect different linear features, probably pipelines.

4. MATCHING OF DISSIMILAR OVERLAPPING IMAGES

The primary goal is the use of interferometric methods. The component of the University of Graz is to compare results from interferometry with other 3D surface reconstruction methods, and to match dissimilar (opposite side) imagery. These investigations are assessed to evaluate the feasibility of combining data from different look angles and possible quality increasement in 3D surface reconstruction. These tasks depend on data provided by partners, which are going to be compiled. For the project data from the testsite "Hartberg" in south-east Styria/Austria were used. In particular the geocoding of ERS-data with a newly developed geocoding tool which is based on SAR simulation was done.

By matching the simulated and the actual image, the simulation is validated. If the discrepancies are below a certain threshold (usually 1 to a maximum of 3 pixels), the geocoding can be performed by resampling using the same geometric model. Remaining geometrical errors are decreased by applying a higher order polynomial adjustment between the simulated and the actual radar image. Other developments were in the field of surface reconstruction i.e. matching same-side stereo radar images with varying look-angles of ERS-1/2 with area-based template matching methods and matching dissimilar opposite side radar images of ERS-1/2 with a feature-based approach. Edge detection and grouping edge strength locations to solid edge contours together with a weighted edge-neighborhood graph makes opposite side matching feasible if a considerable amount of structural content is contained in the opposite side radar image pair. Finally, the University of Graz started on quality increasement in same-side radar image matching by detecting layover in multi-lookangle radar images of ERS-1 (roll-tilt). Matching points bordering layover areas contradict the fundamental epipolar restriction common matching software depends on. An extension capable to cope with layover arising in same-side radar image matching is currently being developed.

5. LONGTERM MONITORING

The detection and separation of ecologically important areas in the surroundings of the test area "Steinhuder Meer" is one major topic of the research part from the University of Hannover. For this investigation several ERS-1 scenes between 1992 and 1995 have been processed and classified. All interpretation and classification methods could show that it is possible to detect for environmental studies important areas in multitemporal ERS images (GTC), but a separation into refined classes like greenland, partially covered greenland with forests and forests is very difficult. Therefore in future developments coherence maps from ERS-1 and ERS-2 tandem mission shall be considered.

The second topic was to investigate the backscatter behavior of farmland and wasteland in ERS-1 multitemporal images. The student could collect ground truth data in a test area near "Göttingen" and was able to separate the main agricultural fields. In the time range from 1992 to 1994 only a few characteristic ERS-1 scenes could be investigated. Scenes between September and December were not useful for the multitemporal interpretation methods. Nevertheless the candidate was able to detect wasteland areas without prior information. A ground truth comparison could verify his results. In the future investigations with multitemporal and multi band mission shall be carried out (X-SAR).

In the DFG-funded project "Semantic modelling of remote sensing data" also ERS-1 and ERS-2 image shall be investigated. As the main developments are in the scale 1 : 5 000 to 1 : 25 000 additional radar data sets will be considered for topographic maps in the scale of 1 : 100 000 or 1 : 200 000. For this purpose following ERS-2 coherence maps have been ordered at DLR-DFD, depending on the season there are different priorities for the data application.

6. RAINFOREST MONITORING

Embedded in a cooperation with the Brazilian Space Research Institute INPE, testsites in the state of Acre, Brazil in the Western Amazon are to be investigated for the use of complex ERS data products in rainforest monitoring.

Three tandem data products (quarter scenes SLC) have been ordered in July 96. The scenes from May 1996 cover areas in the North of Rio Branco, the capital of Acre, and an area along the Brazil Rodovia BR-364 towards Sena Madureira and Peru. The May data should be promising for using coherence data because of beginning dry season. Other tandem datasets of the Rio Branco area have been registered in October 1995. Two full scenes of ERS-1 of that orbit are already available. The scenes cover Rio Branco and surroundings with different patterns and stages of deforestation and regrowth as well as a large undisturbed primary rainforest area North of it. This data has been used for a forest / nonforest classification, also trying to separate fresh pasture areas from old pasture / regrowth areas. For classification, the EBIS texture classification approach (after Lohmann, 1994) was used. This evidence based classifier is based on texture characterization by the co-occurrence matrices in a certain window environment. Although the October scene was registered in beginning wet season, most of the known deforestation areas have been separated. In comparison to available Landsat TM reference data from July 1994, several new deforested areas have been registrated, especially Southwest of Rio Branco. On the other hand, main misclassification resulted in the peripheral areas of Rio Branco and adjacent environments. Here an overlay of backscatter signals of settlements and old pasture land exits, but also of different plantations, reaching similar levels of backscatter amplitudes as in rainforest areas. Those areas could be interesting for integrating coherence data. Based on the two ERS-1 scenes from October 1995 and two adjacent ERS-1 scenes from August 1995 West of it (also available since July 1996), a classification has been performed. The mosaic dataset on 50 m resolution covers an area of about 200 km by 170 km and is to demonstrate large-area mapping. An accuracy assessment is under work. The data products are to be used to supplement time series data produced by the former ERS-1 rainforest pilot project. They are also to be used in preparation of a combined amplitude / phase investigation.

A combination from amplitude images and coherence results of the InSAR system shows a much higher variety in the deforestet than multitemporale ampltide images.

7. AUTOMATED LANDUSE MAPPING WITH ERS-1/ERS-2 SAR- AND INTERFEROMETRY DATA FOR FLOOD MONITORING

The Institute of Photogrammetry and Remote Sensing (IPF), University of Karlsruhe, The test area of this subproject from the Institute of Photogrammetry and Remote Sensing (IPF), University of Karlsruhe is located at the upper Rhine value with the coordinates UL N 48 5 29'05'' E 7 545'00' LR N 48525'00'' E 7547'50'' where several water retention areas were established. These areas may be flooded to avoid disasters at the lower Rhine valley, where several flood events caused immense damage during the last few years. In addition the water retension areas are being retransferred into a natural state of fauna and flora for ecological reasons. This process is enforced by recurrent intentional flood events, called ecological flooding. In this context, several points of interest may be served by remote sensing techniques. For example the monitoring of flooded areas during ecological flooding, the distribution of water within the retention area, the water volume an area may hold and so on.

In a first step several data sets were acquired to meet the requirements of a context based SAR data analysis. From the German weather service (DWD) weather data were ordered: especially the wind conditions within our test area are of interest. On water surfaces, the wind condition is an essential mapping condition for SAR images, and can profitably be explored by an automated interpretation method for the SAR image data. Two SAR image pairs from the ERS-1/ERS-2 tandem mission have been ordered by our RAPID partner DLR for the purpose of updating the map information we gathered so far. One of the pairs was acquired during an ecological flooding within a retension area. So the IPF will study the mapping of the flooded areas by the intensity SAR image as well as the interferometric data sets to demonstrate the capabilities of this new techniques. By comparing the list of past flooding events with the available ERS-1/ERS-2 images in the archive of DLR - DFD, we found another match for a 1995 event. Since this SAR imagery has been acquired in 1995, only two pass interferometry will be possible and thus, the evaluation of interferometric products for this event is not of interest. Nevertheless a continued observation of such events by SAR intensity images would be of interest in this context. Digital terrain models are one focal point within this RAPID subproject. They are used to determine the volume of retension areas and they serve as context for an automated interpretation of actual SAR information. Therefore, the precision of different sources of DEMs is also of interest in this project.

In order to enable a higher degree of automation for the data evaluation we aim at the creation of a 'fuzzy' expert system, which holds all rules describing the relations between the actual SAR input data, the known mapping conditions such as wind force and the final event (e.g. flooded areas) the IPF want to reason about. In addition there are a lot of unknown mapping conditions influencing the final SAR signal. These conditions may not be considered, but their influence on the SAR signal may be extracted by means of GIS technology directly from the SAR image. For example, the roughness of a water body may be estimated by the SAR signal within all water areas of a given minimum size, shown in a thematic base map. For that purposes a classification scheme was established and an evaluation of tools and libraries for the implementation of the classificator was realized.

8. IMPROVEMENT OF SATELLITE DATA INTERPRETATION THROUGH INTERFEROMETRIC ANALYSIS

Different datasets over the Philippines were acquired and processed since the official start in 1996. In the testareas groundstudies were made and a vegetation classification has been completed. The vegetation types were grouped into three clusters:

- vegetation types that can be identified on Spot/Landsat images,

- vegetation type that can be identified on aerial photographs,

- vegetation type that can only be identified in the field.

The work is carried out by the University of San Carlos, Phlippines.

9. CONCLUSION

The preoperational InSAR system as basic element is capable to produce coherence maps and DEM. The availability of tandem data from the foreign station is at the moment a bottleneck, because a lot of required frames were rejected the the satellite system. For first interpretation the available ERS-products are sufficient.

The rainforst monitoring the coherence information is a very usefull dataset for interpretation and classification and will be integrated to the standard classification process. Also the detection of linear features with In SAR is an sucessfull approach.

The application institutes got and will get different dataset from their testarea und will continue with their investgation.

Keywords: ESA European Space Agency - Agence spatiale europeenne, observation de la terre, earth observation, satellite remote sensing, teledetection, geophysique, altimetrie, radar, chimique atmospherique, geophysics, altimetry, radar, atmospheric chemistry