Low earth orbit provides an ideal operating environment for synthetic aperture radar (SAR) imaging due to well-defined platform motion. Airborne SAR platforms do not benefit from this luxury and suffer unpredictable motions causing image defocusing. Autofocus is a widely acknowledged technique for correcting these motions, causing the image to be properly focused. Accurate position and velocity information is required to process spaceborne SAR scenes with higher spatial resolutions requiring greater knowledge of the satellite's orbit. The availability of position and velocity data is dependent on the efficiency and location of the ground station and could finally lead to delays in processing times. In these cases an orbital propagation model has to be employed for immediate processing. The precision of the image is now dependent on the accuracy of the orbit model used. Using DRA patented autofocus/phase correction techniques on ERS1 raw data it is shown that the point target response of imagery obtained with no prior orbit knowledge is comparable with precision imagery. In addition the technique allows continuous strip map imagery to be produced with no discontinuities. Finally, the need for autofocus for high resolution imagery is discussed.

A radiometric correction method based on a SAR image simulation is presented taking into account the resolution of the SAR image and comparing to the grid size of the available DEM. Once calibrated, another correction using an empirical or theoretical model can be processed. The method is illustrated by an example over hilly terrain and is validated by showing the decorrelation between the SAR image intensity and the slopes after the correction.

The subject of the reported work is the improvement of geometrical models for a SAR scanning in pushbroom, spotlight, scansar or bistatic imaging modes. This research has been motivated by the planetary cornerstone mission of ESA's long term program for European Space Science ('rendezvous' with a comet, and fly-bys of asteroids). In this specific context, the synthetic aperture radar is destined for an important role, but the rules and standard backgrounds of the Cartesian geometry are no longer justified. Several new techniques are proposed to handle with an optimal precision the data relative to celestial bodies with a complex geometry (coherent and non-coherent imagery). On the basis of a mathematical rigor (singleness of solutions, convergence of processes, biunivocity of transformations and generalizations), a lot of scenarios are discussed with key relations established (plane and spherical models, bodies with a symmetry of revolution and general bodies, specific sensor(s) trajectories as fly-bys or flight into orbit with the possibility of an approaching probe). The four methods developed are the tomographic analogy of radar principles (only known, previously, in the usual case of a straight line flight at constant altitude over a plane surface) and Hilbertian techniques for a direct adaptation to the scanned surface geometry, an automated autofocusing which enhances the contrast resulting from a Cartesian reconstruction and the coordinates transformation where the real space is converted into a fictitious space where Cartesian algorithms are fully rigorous. Beyond the fact that an interpolation step is often unavoidable, the major conclusion of the research is that all the prospected techniques are complementary and that the choice between the methods has to be made according to geometry, objectives and time requirements (reconstruction on board or not). In particular, coordinates transformation techniques are worthy of commendation in the case of plane (wavefront curvature balancing) or spherical models in a monostatic situation. Autofocusing methods (judicious ponderation between the usual reconstruction and a reconstruction of the derivative of the key expression of the mathematical formalism with regard to one of its parameters) has proven its validity in the hilly regions east of Belgium with low differences in contrast, while the Hilbertian principles are general methods without any restriction on the paths of the probes, the geometry of the celestial body, the modulation scheme and antennae radiation pattern. On the other hand, the tomographic analogy can be applied in all situations where a correct model of the body relief is available, but there are some approximations in the formalism (no antenna pattern modeling, no balancing of the range migration).

The aim of this work is to develop some algebraic reconstruction techniques for low resolution power SAR imagery, as in the Scansar or QUICKLOOK imaging modes. The traditional reconstruction algorithms are indeed not well fit to low resolution power purposes, since Fourier constraints impose a computational load of the same order as the one of the usual SAR azimuthal resolution. Furthermore, the range migration balancing is superfluous, as it does not cover a tenth of the resolution cell in the less favorable situations. There are several possibilities for using matrices in the azimuthal direction. The most direct alternative leads to a matrix inversion. Unfortunately, the numerical conditioning of the problem is far from being excellent, since each line of the matrix is an image of the antenna radiating pattern with a shift between two successive lines corresponding to the distance covered by the SAR between two pulses transmission (a few meters for satellite ERS1). We'll show how it is possible to turn a very ill conditioned problem into an equivalent one, but without any divergence risk, by a technique of successive decimation by two (resolution power increased by two at each step). This technique leads to very small square matrices (two lines and two columns), the good numeric conditioning of which is certified by a well-known theorem of numerical analysis. The convergence rate of the process depends on the circumstances (mainly the distance between two impulses transmissions) and on the required accuracy, but five or six iterations already give excellent results. The process is applicable at four or five levels (number of decimations) which corresponds to initial matrices of 16 by 16 or 32 by 32. The azimuth processing is performed on the basis of the projection function concept (tomographic analogy of radar principles). This integrated information results from classical coherent range compression. The aperture synthesis is obtained by non-coherent processing of the projection functions following the above outline. A set of simulation examples are depicted, which take into account the synthetic aperture level (defined as the ratio between the antenna footprint on the ground and the required resolution in azimuth), the burst length, the number of bursts, the pulse repetition frequency, the gap between bursts, the azimuthal relative speed and other design parameters such as the number of bits for raw data or the antenna angular opening at reconstruction. The degrees of freedom of the reconstruction scheme are also outlined (possibility of filter effect for a smooth appearance or, on the other hand, a speckle impression.)

The frequency domain approach of Scansar problems suffers, numerically speaking, from the constrained smallness of the unit of azimuth increment. This unit is equal to the displacement of the craft between two consecutive transmission times, whatever the burst length, the latter conditioning the resolution of the coherent reconstruction. Thus a large oversampling occurs. There are three main ways of action to save computational effort: (1) The simplest one is to go out of the frequency domain with a dimension reduced transform operating on a cheap manipulation of the full frequential product. (2) The second one is to build the frequential input from several transforms at a dimension close to the burst, thus taking advantage of the initial zero padding. (3) The third one is to modify the 'static' template pipe in order to avoid repetitive operations in the signal pipe. A simple situation will be met where this policy is immediate. A much more intricate case of application leads to a hybrid algorithm connected with the SPECAN approach. The difference between the hybrid algorithm and Specan rests in the strictly diagonal character of the initial transformation performed on the burst in the Specan case, contrary to a narrow band matrix multiplication for the more general hybrid approach. The main lines of conclusion of this study are that, assorted with carefully designed numerical procedures, the frequency domain approach for Scansar is convenient, mainly, for coherent processing of rather longs bursts. More particularly, it combines well with frequential azimuth multilook. In the case of too short bursts, it becomes weaker than direct algebraic manipulation, but it gives an easy way for preparation of the needed static transformation matrices. The comparison with SPECAN approach is less easy, as the nets of points locations for reconstruction do not match. Nevertheless, a static transformation generalizes the Specan concept to prescribed grid by approximations coming from the hybrid algorithm. That new way could be challenging. The important theoretical part of this paper is clarified by illustrative means and some quicklook alternatives discussed in the function of, among other things, the imaging scenario, the required resolution, and the bursts length.

This paper is intended to compare compression algorithms suitable for synthetic aperture radar (SAR) raw data. BAQ, BFPQ and BFPQ combined with vector quantization have been applied to simulated and actual ERS-1 raw data with output data rates of 2 and 3 bits per sample. The simulated raw data are generated from a terrain scenario (point target and uniform areas of sigma 0) scanned by a SAR simulator set to ERS-1 parameters. The measurements performed on raw data consist of histograms, signal to digital noise ratios and quadratic phase errors. The effects of compression on the resulting SAR processed images are assessed by measuring the impulse response parameters, the radiometric degradation and by displaying difference images. Eventually, the performances reached on simulated and actual data are discussed and compared.

The scattering of electromagnetic waves from random rough surfaces near the grazing incident problem is of widespread interest in various physical situations, in particular coastal and shipborne radar backscatter. It is, however, a very difficult problem still awaiting a satisfactory analytical solution. Two of the features which have been seen experimentally and yet have not been explained analytically are sea spikes, and unexpectedly large hh/vv scattering cross section compared to the analytical solutions available. In this paper we present a numerical study of backscatter cross section for the above situation. The numerical calculation was done using the Petit method, which unlike the better known method of moments (MOM), allows calculation to be done for the case of large angle incidence.

The exact solution to the synthetic aperture radar (SAR) scattering problem from an urban scene, with its collection of object-length scales, requires the solution of Maxwell's equations for the combination of source and scattering objects present in the scene, which for any reasonable size target area is computationally too large to be realistic. The geometrical theory of diffraction (GTD) is based on the fact that the most important contributions towards the scattered field come from an area in the neighborhood of some critical points on the scattering surface. This gives an accurate result with a practical amount of computation. However, the range of problems to which GTD can be applied is limited by the availability of the solutions to the canonical problems from which the diffraction coefficients of GTD are calculated. In this paper a physical optics version of GTD was used in conjunction with a ray-tracing approach to simulate the SAR scattering from a simulated urban scene.

Strong point backscatter can be observed in ERS-1 satellite SAR images taken over urban areas. Several of these point backscatter come from metal surfaces and roofs. An experiment using a continuous wave (cw) radar has been carried out in order to model the radar backscatter from tarred board roofs and corrugated iron surfaces. Test objects are illuminated by the cw-radar using the ERS-1 SAR frequency, but with several polarizations and incidence angles. Specular reflection and Bragg resonance effects are studied in particular. Radar simulations are used to confirm the experimental results. The final results show that both metal and tarred board roofs can give strong radar backscatter. Both first and second order Bragg resonance peaks occurred at the theoretical incidence angles when illuminating corrugated iron plates with the radar. Comparisons are also made to real data by investigating point target backscatter seen in a multitemporal ERS-1 SAR image data set. Further investigations could be carried out when RADARSAT images acquired at different incidence angles are available.

Coherent images of natural scenes formed using synthetic aperture radar (SAR) often possess textural properties associated with the clutter. Due to a recent improvement in the Defence Research Agency (DRA) X band SAR, very high resolution imagery is now available for analysis. This increase in resolution has visibly modified the textural properties of observed clutter forcing a re-examination of the statistical image properties. The results of this study are given. Areas of imagery that appear homogeneous have their single point distribution properties measured. Comparisons with known distributions that often fit similar data are made and shown to give poor agreement. Reasons for poor agreement, such as inhomogeneity, are investigated through the use of clutter simulations. K and lognormal mixture distributions are shown to offer good agreement with the observed distributions and also validate the premise of homogeneity for the regions considered.

We have experimented on the effects on SAR ERS1 scenes of two classes of filters designed for speckle reduction according to the size of the objects to be extracted. The valuation of both classes of filters is done by two criteria: homogeneity and connectivity. The first class of filters is devoted to the extraction of thin features. In this case, the various filters operate within an elementary neighborhood. The filters belonging to the second class enable the extraction of larger elements. These filters are alternated sequential filters obtained by composition of morphological filters using structuring elements of increasing size. The comparison of the results provided by these filters, as well as the parameters quantifying homogeneity and connectivity, will allow the choice of the more accurate filter in order to ensure the extraction of features of various sizes on SAR-ERS1 images.

In order to remove the speckle on radar images while preserving the radiometrical and spatial resolution of the initial image, various filtering techniques have been proposed. These techniques are mainly based on a statistical and/or a morphological approach. The filtering presented here belongs to the first class. The statistical distribution of the backscatter values of the initial image is observed inside a given neighborhood and controlled by three parameters (size of the neighborhood, radiometric range and length of the isolated elements). This technique has been applied on two SAR-ERS1 scenes (Spitsbergen glaciers and Mount Pinatubo in the Philippines). While reducing the speckle, the HK-Filter seems able to underline the linear features.

One of the most difficult and important problems encountered in the automatic digitizing of graphical topographic maps is the identification of different kinds of features. Textures are an important spatial feature useful for identifying objects or regions of interest in a remote sensing image. This work presents a wavelet based algorithm combined with a fuzzy C-means classifier. A single image is preprocessed by a wavelet packed algorithm and divided in subimages, different representation of the same scene. The development of this transform is motivated by the observation that a large class of natural textures can be modeled as a quasi- periodic signal, whose dominant frequencies are located in the middle frequency channels. The subband images are then processed by an envelope signal estimation in order to provide a method for features extraction: different textures have different 'energy' in the detail subband; this energy can be seen as a magnitude of oscillation of wavelet coefficients for each subband. The image can now be seen as a multiband representation of the same scene and this can be considered a multi-dimensional data clustering problem. Fuzzy C-means algorithm is so applied to the image as to have a very efficient fuzzy segmentation of the different textures.

In this paper we derive the maximum likelihood (ML) criterion for splitting (or merging) two regions of single-look SAR imagery as a function of the mean intensity. Two distinct optimization criteria can be postulated: (1) maximizing the total probability of detecting an edge within a window; and (2) maximizing the accuracy with which the edge position can be determined. Initially we derive the ML solution for the first criterion and demonstrate its superiority over an approach based on the Student t test when applied to intensity segmentation. Next we discuss the ML solution for determining the edge position. Finally, we propose a two-stage edge detection scheme offering near optimum edge detection and position estimation.

We present an algorithm that is able to smooth out the speckle from many SAR images and which does not suffer from the drawbacks of multilooking. The algorithm is able to preserve the detail and resolution of the original image while producing a smooth, real-valued output. In many cases the quality of the smoothed image is sufficiently high that it may be used with standard optical post-processing algorithms. We use a global optimization method (simulated annealing) and single point gamma statistics to find the MAP solution for the radar cross- section. However, this method may also be regarded as an ideal adaptive filter that is both computationally efficient and highly parallelizable. Results are presented for airborne, ERS-1 and multi-temporal SAR images.

The PULSAR project, aiming for the parallelization and adaptation of SAR segmentation and filtering codes, is carried out under the EUROPORT 2 activities within the European funded ESPRIT 3 program and running over 2 years. An international consortium consisting of companies working in the informatics and earth observation domain was created to perform the project as an interdisciplinary study. Different actions carried out in the past year and a half were mainly the testing and parallelization of the different codes, their adaptation to selected applications and the benchmarking in terms of speed and reliability of results. A number of codes were improved and speeded-up even before the parallelization and better benchmarking results could thus be achieved. For the detailed testing of the codes three key applications were identified where radar remote sensing will play a key role in the next few years. These applications are the oil spill detection, the surface management of temperate (European) agriculture and the tropical agriculture (rice surface detection and monitoring). For all three application fields different parameters available in the software were tested and results show great improvements of the interpretation capabilities in relation to the initially speckled ERS SAR data. The segmentation specifically allowed us to obtain field boundaries of agriculture fields over the test site Bourges (France) and areas of similar growth conditions for rice areas in Indonesia. The speed of the processing of one 512 by 512 pixel image was at 15 minutes using the serial version and 2 minutes using the parallelized version with 4 processors. The comparison between the serial and parallel results allowed us to investigate the stability of the parallel process which was below 2% inaccuracy. Thus, it can be underlined that the PULSAR project allowed us to develop a new and very useful tool for the interpretation of radar images. For specific applications it is intended to further adapt the tool in order to be able to deliver turn-key solutions to the user.

Polarimetric SAR images will contain extended regions of apparently homogeneous clutter arising, perhaps, from areas of vegetation or uniformly driven expanses of water. Such regions may contain localized clutter variations which indicate the presence of features of interest such as changes in vegetation density due to environmental effects, damping of the sea surface due to the presence of pollutants or the effects of partially concealed land-based or maritime military targets. It is thus of interest to develop techniques which can discriminate localized clutter features from the background clutter. If the distribution parameter values are known for both background and feature then the likelihood ratio method can be used. Frequently, however, the feature parameter values are unknown in which case one option is to use simply the background likelihood. In both cases, quadratic test statistics result which are analyzed to allow a comparison of theoretical performances. A second option is to introduce a probability distribution for the feature parameter values. Optimum performance will result if the assumed and true distributions match exactly but any mismatch may considerably reduce the performance. Simulations are used to investigate this effect. In conclusion, the paper assesses the various discrimination techniques in terms of complexity, prior knowledge requirements and performance.

The aim of this paper is to show that Dempster-Shafer evidence theory may be successfully applied to unsupervised classification in multisource remote sensing. The main advantage of unsupervised classification is that no a priori information is needed. Dempster-Shafer formulation allows the user to consider union of classes, and to represent both imprecision and incertitude. So, it provides better representation of sensor information and more reliable classification results. An unsupervised multisource classification algorithm is applied to Mac- Europe'91 multi-sensor airborne campaign data. Classification results using different combinations of sensors (TMS and AirSAR) or wavelengths L and C bands) are compared. Performance of data fusion has been evaluated in terms of identification of culture types. Particularly, we show that, even if most performing results were achieved using the three data sets, good identification rates could be obtained using less expensive combinations of data.

This paper deals with analysis of statistical properties of multi-look processed polarimetric SAR data. Based on an assumption that the multi-look polarimetric measurement is a product between a Gamma-distributed texture variable and a Wishart-distributed polarimetric speckle variable, it is shown that the multi-look polarimetric measurement from a nonhomogeneous region obeys a generalized K-distribution. In order to validate this statistical model, two of its varied versions, multi-look intensity and amplitude K-distributions are particularly compared with histograms of the observed multi-look SAR data of three terrain types, ocean, forest-like and city regions, and with four empirical distribution models, Gaussian, log-normal, gamma and Weibull models. A qualitative relation between the degree of nonhomogeneity of a textured scene and the well-fitting statistical model is then empirically established. Finally, a classifier with adaptive distributions guided by the order parameter of the texture distribution estimated with local statistics is introduced to perform terrain classification, experimental results with both multi-look fully polarimetric data and multi-look single-channel intensity/amplitude data indicate its effectiveness.

We have developed and we present here a radarclinometry technique partly based on a shape- from-shading method, allowing us to obtain from a 1-look SAR image, a qualitative digital elevation model (DEM). This technique, that does not require a high computational cost, is sufficient enough to produce from a single SAR image a qualitative DEM without important topographic irregularities. It is then possible to obtain either a preliminary DEM that can be used as a support for further improvements by means of another technique such as, for example, interferometry, or to precision some details on a previous DEM obtained another way and thus to underline important topographic changes. Two test zones located in the Philippines have been chosen and this technique has been applied to reconstruct active volcanic structures.

Many emerging SAR applications require imagery which is multi-channel either in time, frequency, or polarization. If SAR segmentation algorithms are going to be of genuine utility they must be applicable to such datasets. Two approaches to multi-channel segmentation are considered: segmenting each channel separately and then recombining the results; and segmenting the multi-channel image as a single entity. In both cases segmentation is based on edge detection and segment growing. The utility of multi-channel segmentation for change detection, classification and analyzing the information content of multi-channel data, is discussed. A multi-temporal ERS-1 image, and a multi-polarized, multi-frequency AIRSAR image of the same agricultural scene, are used for case studies.

The use of the ERS-1 C band SAR for monitoring tropical forest areas is assessed, using three ERS images from the Tapajos region of Amazonia gathered in 1992. Forest areas display a very stable RCS, while non-forest areas in some cases exhibit changes which appear to be associated with soil moisture variations. Discrimination between forest and non-forest is greatest after a dry period. Because of distortions in RCS caused by topography, change detection provides a more useful discrimination approach than RCS differences on single images. A number of automatic change detection techniques are compared and their ability to classify forest and non-forest are quantitatively assessed, assuming that a forest map inferred from a 1992 Landsat TM image is correct. Block averaging followed by image ratioing provides a reasonable approach to detecting the large scale structure of the image, but simulated annealing provides improved performance at a computational cost which is becoming competitive with simpler methods. Approximately 50% of the non-forest region can be detected from the ERS-1 images. This figure may be improved by more frequent image acquisition, but there are fundamental limitations in using C band data, which would be lessened by using longer wavelengths.

To increase the spatial resolution of synthetic aperture radar (SAR) images, we process the received data through super-resolution methods instead of conventional pulse compression. Since the problem of SAR processing is 'ill-posed', we must apply regularization techniques by incorporating prior knowledge about the imaged scene. We review restoration techniques with prior knowledge based on known deterministic properties of the solution or formulated in terms of a priori probability distributions. Moreover, we pose the scattering cross section problem as a maximum-likelihood one, and we maximize the likelihood function through optimization algorithms. Numerical results demonstrate that the classical resolution limit can be overcome by using certain super-resolution techniques.

This paper adapts Pentland's technique of shape from shading using assumed umbilical points, for application to synthetic aperture radar. A new surface archetype suitable for SAR geometry is developed to replace the spherical archetype of the optical technique. Comparisons are drawn between the optical and the SAR case, discussing the problems of ambiguity in the solution surfaces. The accuracy of this method is explored by comparing the surface generated from simulated SAR images with the data used to create the images. The problems of speckle and self shadowing that will be found in real SAR cases are discussed.

The weighted least squares approach to phase unwrapping, an extension of the unweighted least squares approach, is a robust and accurate method for phase unwrapping. This approach zero-weights portions of the phase data to avoid unwrapping across regions of corrupted phase, which are typically caused by the SAR phenomena of layover and radar shadow. Recently the first practical method for accomplishing this weighted phase unwrapping was introduced. This method is based on preconditioned conjugate gradients (PCGs). We present a new method that is an extension of our multigrid algorithm for unweighted least squares phase unwrapping. This new method features a carefully defined operator for transferring the phase weights to the coarse grids of the multigrid scheme. The algorithm converges in only a few multigrid cycles, and on phase data that contain severe discontinuities or shear we demonstrate that it is up to 25 times faster than the PCG algorithm. We also present methods for defining the initial phase weights, thus yielding a completely automated algorithm for fast and accurate phase unwrapping.

We present a multigrid algorithm for unweighted least squares phase unwrapping. This algorithm applies Gauss-Seidel relaxation schemes to solve the Poisson equation on smaller, coarser grids and transfers the intermediate results to the finer grids. This approach forms the basis of our multigrid algorithm for weighted least squares phase unwrapping, which is described in a separate paper. The key idea of our multigrid approach is to maintain the partial derivatives of the phase data in separate arrays and to correct these derivatives at the boundaries of the coarser grids. This maintains the boundary conditions necessary for rapid convergence to the correct solution. Although the multigrid algorithm is an iterative algorithm, we demonstrate that it is nearly as fast as the direct Fourier-based method. We also describe how to parallelize the algorithm for execution on a distributed-memory parallel processor computer or a network-cluster of workstations.

In this paper we exploit a new solution for phase unwrapping specially designed for interferometric synthetic aperture radar applications. The proposed algorithm is based on use of the first Green's identity and provides the unwrapped phase pattern starting from the gradient of the measured phase. Results on simulated as well as on real data are presented and confirm the excellent performances of the method.

Synthetic aperture radar (SAR) interferometry is a technique based on the exploitation of the interference pattern of two SAR images acquired in similar geometric conditions, at two different dates. This technique has a very broad range of applications. It is used to generate digital elevation models (DEM). It also can detect centimeter-size changes on the ground surface. In order to extract topography from interferograms, the interference pattern has to be unwrapped and transformed into altitude. Phase unwrapping is one of the most difficult parts of the process and has become one of the most studied subjects in the interferometric community. In this paper, we present a new phase unwrapping method, based on a global analysis of the interferogram. The interferogram is divided into surface elements that can be modeled by local slopes. A model of the unwrapped phase is computed thanks to these elementary surfaces. We show that the interferogram can be unwrapped using the computed model as soon as the differences between the model and the interferogram are smaller than half a fringe on the entire image. Then, we present two algorithms based on this method, an automatic one and an interactive one. We discuss the performances of each algorithm on test sites. We show that the automatic algorithm is very efficient in noisy areas, and that the interactive one can be of interest in areas with shadow or overlay. Finally, possibilities of improvements are discussed.

We propose a method of phase unwrapping based on residue connection, using coherence measurement as a complementary information source. Coherence measurements are shown to be improved when an approximated phase due to topographic fringes is removed. The coherence map so obtained is used as a mask to guide the connection process.

This paper presents a novel and efficient method, based on a modified version of the Chirp-Z transform algorithm, for performing the co-registration of SAR images as required in SAR interferometry. The method also allows complex interpolation of an entire image, by application, at each point of a bilinear coordinate transformation.

Image registration in interferometric synthetic aperture radar (SAR) applications is necessary to improve the interferogram quality and in turn the resulting DEM accuracy. This paper deals with a new image registration procedure implemented at raw data processing stage: the two complex SAR images are generated with respect to a common reference output system. All the registration task is achieved via scaling and shifting compensation that can be efficiently and easily included in a standard SAR processing code. An algorithm to estimate the correct processing parameters, whenever the orbital information is inaccurate, is also presented and is based on the relation between spectral shift and linear stretch of the two images. Examples on real and simulated data validate the presented procedure.

Repeat-pass interferometry data have been acquired during the first and second SIR-C/X-SAR missions in April and October 1994. This paper presents the first results from X-SAR interferometry on four different sites. The temporal separations were one day and six months. At two sites the coherence requirements were met, resulting in high quality interferograms. A digital elevation model has been derived. The limitations of the X-SAR interferometry are discussed.

In April and October 1994 the X-SAR radar has been flown twice onboard the Space Shuttle, as part of the Space Radar Laboratory (SRL-1 and SRL-2). This radar payload is the first synthetic aperture radar (SAR) system of its kind in space, with three frequencies, multi- polarization, variable incidence angle and variable modes of operation. SIR-C/X-SAR, the new generation of imaging microwave remote sensing sensors from space, demonstrated successfully repeat pass interferometry in all three frequencies with a one day repeat as well as a six month repeat orbit. The major problems with the repeat orbit interferometry are the temporal target decorrelation, unsuitable baseline and different squint angles for the two passes to be processed. Therefore, for the third mission of the Space Radar Lab which shall be called SRTM (shuttle radar topographic mapper), single pass interferometry with a second receive antenna is proposed to generate a topographic map of all land surfaces between +60 degrees and -56 degrees latitude. X-SAR's 12 meter long and 40 cm wide main transmit and receive antenna is mounted directly to a tiltable part of the SIR-C antenna truss structure in the Shuttle cargo bay. The second receive antenna is 6 meters long and is mounted together with the second C-band antenna to a 31 meter long deployable boom structure perpendicular to the velocity direction to build the baseline. X-SAR is not capable of operating in a scan SAR model like SIR-C to allow continuous coverage, but will operate in a high resolution mode with a swath width of 30 - 35 km. The engineering design of the interferometric configuration for X-SAR, the requirement specifications, and the predicted performance as well as the mission operation aspects are described in this paper. SIR-C/X- SAR is a cooperative project. The SIR-C instrument was developed by NASA's Jet Propulsion Laboratory (JPL). The X-band radar (X-SAR) was developed by the Dornier and Alenia Spazio Companies for the German Space Agency (DARA) and the Italian Space Agency (ASI), with the Deutsche Forshcungsanstalt fuer Luft- und Raumfahrt e.V. (DLR), as the major partner in science, operations, and data processing.

Topographic mapping on whole Earth surface is aimed for a lot of applications. SAR interferometry is the technique for accomplishing topographic mapping using a pair of SAR images of the same area taken from two different points of view separated by a distance called 'baseline.' These points of view can be one or two antennas placed on an airborne platform (double antenna technique) or two passes of the same satellite on two different orbits (double passes technique). This method of acquisition causes the phase of each pixel to be different in the two images. This depends, after having removed the contribution of the different geometry of view, only on the height of the observed point on ground. Alenia Spazio has developed a whole processing technique able to obtain three-dimensional images of the observed area using two complex images (SLC -- single look complex) coming from SAR sensors like ERS-1 and X-SAR. This paper describes the whole processing of the complex SAR data to obtain the 3D representation of the image from a couple of SAR images after proper overlapping. In particular some steps have been pointed out: the evaluation of range and azimuth shifts between the two images; the overlapping of these by a particular interpolation; the filtering of the complex data in order to improve the coherence between the images; the extraction and filtering of the interferogram; the phase unwrapping; the phase-to-height conversion. The whole processing was applied to some images coming from scientific missions: in this paper some results are presented.

CSL (Centre Spatial de Liege) developed a complete InSAR processor as part of the Belgian National Remote Sensing Program. This processor makes use of a new 2D complex interpolator used in coregistration of SAR images, an adapted filter and a phase unwrapping method based on residue connection guided by coherency measurements. We sketch the methods and present the results we obtained on a reference interferometric pair made available by the ESA FRINGE working group.

ERS-1 SAR interferometry has a large potential to map out forested areas. Different coherence maps show a stable pattern, even with high wind speeds and melting temperatures on the ground. With a DEM of the ground, ERS-1 SAR interferometry can be used to measure interferometric tree heights in a forest. Measurements yield a tree height even when the coherence is low, but have a tendency to decrease with the area coherence. Two boreal forests have been measured showing an interferometric tree height about half of what was measured in-situ. The fact that interferometric measurements are lower than the real tree heights may have been caused by either high wind speeds lowering the effective stable scattering center in the trees or by existence of a ground contribution to the forest backscatter of the radar return.

SiSAR was planned as a realistic as possible, modular, user-friendly and fast SAR raw data simulator running on ordinary workstations. Interest in (interferometric) SAR products is growing on an international scale. There is a concentration of manpower and financial resources. Dead ends, respectively failures, have to be avoided during design and mission of every SAR project by simulating the system thoroughly before the experiment. Another reason to make use of extensive reproducible simulations during design and development is the reduction of time and manpower costs. As it comes down to verifying and comparing different processing algorithms we see that (interferometric) SAR simulation is an indispensable tool for testing individual processing steps. SiSAR is a modular SAR raw data simulator for realistic description of the functions of a SAR-system. It contains an implementation of diverse models to characterize radar targets, various approaches to describe the trajectory and the motion of the footprint on the target surface and different raw data formation algorithms. Beyond there is a wide supply of tools for manipulation, analysis and user-friendly simulation handling. Results obtained by SiSAR and some first simulated interferometric SAR raw data are shown in the paper.

Since August of 1994, the Defense Meteorological Satellite Program (DMSP) has had on-orbit two satellite platforms (F-11 and F-12) which carry the special sensor microwave water vapor sounder (SSM/T-2). The sensor consists of 5 channels: three located symmetrically about the 183 GHz water vapor absorption line, one at 150 GHz and a 91.65 GHz window channel. Calibration of the SSM/T-2 on both platforms was undertaken to verify the absolute accuracy of the microwave measurements. Initial findings show a discrepancy between measurements from the SSM/T-2 and an independent aircraft-mounted instrument of only 1-1.6 K at 183 GHz. Basic channel information obtained from radiative transfer modeling provides an insight into the surface and atmospheric contributions to the channel observations and the sensitivity of the channels to various atmospheric phenomena. From the modeling and calibration studies a number of interesting channel signatures were observed in the SSM/T-2 measurements including the effects of the underlying surface, fractional cloud coverage and type, and precipitation occurrence. The signatures of clouds and precipitation (over water and land) have been identified and efforts to derive detection algorithms have been made.

From studies of the special sensor microwave water vapor sounder (SSM/T-2) brightness temperature (T_b) measurements, channel signatures were identified for various surface and atmospheric conditions. The sensor consists of 5 channels: three located about the 183 GHz water vapor absorption line, one at 150 GHz and a 91.65 GHz window channel. Additional sensor information was used (specifically SSM/I, OLS and GOES visible and infrared imagery) to determine the presence of clouds and precipitation in the SSM/T-2 field-of-view (FOV). Non-precipitating clouds over water generally display T_b signatures similar to clear FOVs although some differences do occur, especially for the 91 GHz channel. For data collected in the western equatorial Pacific, the presence of light rain over water caused the warmest T_b to shift to 150 GHz. As the rain rate and scattering in the FOV increased, the 183 plus or minus 1 GHz T_b became the warmest of the three atmospheric channels. For the study of the effect of precipitation over land, SSM/I and manually digitized radar (MDR) data were collocated with SSM/T-2 observations. Techniques that examined the distribution of the T_b differences between neighboring pixels appear to provide a robust technique to identify precipitation. This technique also worked over water surfaces.

Mixed-phase stratiform clouds are modelled for radiative transfer at 89 and 157 GHz. The ice water content (IWC) distribution in a vertical section of such a system is estimated from radar measurements. Aircraft microphysics and radar data are combined to derive parameterizations for the liquid water fraction and for the variation in ice density as a function of hydrometeor size. The effect of the reduction of ice density on the refractive index and hence the single particle extinction parameters is calculated using two different approaches which produce significantly different extinction behavior. Using aircraft-measured dropsize distributions from a range of altitudes, the ice extinction parameters are then represented as functions of IWC and temperature. The effectiveness of the parameterizations is tested by populating an Eddington model using the radar data and comparing the downwelling output brightness temperatures at the bottom of the mixed phase layer with simultaneous airborne radiometer measurements.

The GEOSAT follow-on (GFO) WVR provides correction factors to the GFO altimeter range measurements for path delays due to water vapor. Radiometric 'brightness temperatures' of 22.2 and 37.0 GHz are measured with a shared antenna that has pattern footprints that overlap the altimeter footprint. The antenna is a 1.1 m diameter off-set parabolic reflector with a corrugated tri-band feed. Main beam efficiency is 94% and near beam (plus or minus 10 degrees) efficiency is 97%. This total power radiometer system has fully redundant electronics, weighs 12.7 kg and consumes 18 watts. Channel sensitivities are less than 0.18 degrees Kelvin for a 1 sec sample period.

The global positioning system (GPS) is based on a constellation of 24 transmitter satellites orbiting the earth at approximately 21,000 km altitude. The original goal of the GPS was to provide global and all-weather precision positioning and navigation for the military. Since this original concept was developed, several civilian applications have been conceived that are making use of these satellites. GPS/MET is one such application. GPS/MET is sponsored by NSF, FAA, NOAA, and NASA. The goal of GPS/MET is to demonstrate the feasibility of recovering atmospheric temperature profiles from occulting radio signals from one of the 24 GPS transmitters. On April 3, 1995, a small radio receiver was launched into a 750 km low- earth orbit and 70 degree inclination. As this receiver orbits, occultations occur when the radio link between any one of the 24 GPS transmitters and the low-earth orbiting (LEO) receiver progressively descends or ascends through the earth's atmosphere. With the current constellation of GPS transmitters, approximately 500 such occultations occur in each 24-hour period per LEO receiver. Several hundred occultations have been analyzed to date, where some type of confirmational data has been available (i.e., radiosonde, satellite, numerical analysis gridded data). In this paper, we present a brief outline of the method followed by a few typical temperature soundings that have been obtained.

Two new geometrical concepts of scatterometer were investigated in order to reduce the variation of incidence angle on the fore and aft beams. The first one has fore, aft and mid beams nearly parallel and directed across the satellite track. The second one has concentric fore and aft beams. A feasibility study was undertaken to estimate system performances and budget in C band. This study begins with the determination of radiometric resolution requirement from a wind retrieval simulation in order to take into account the variation of azimuth angle along the beams. Finally performances and interface budgets are compared with currently developed systems.

The aim of the small satellite project VAGSAT is to provide measurement of the directional wave spectrum over the globe to improve the climatology models. The system definition is based upon the nadir beam measurement to normalize the wave spectrum, and the spectrum beam measurement to provide the directional wave spectrum. The dual beam antenna sub- system with one rotative beam and the on-board processing design with the pulse compression solution selected are presented. The first VAGSAT design is encouraging; the payload conception demonstrates the feasibility with an overall mass and power consumption compatible with a small satellite platform.

In a 2-year breadboarding activity funded by ESA, ALCATEL ESPACE lead the design, manufacturing, and test of an all digital pulse generation and compression sub-system for a rain radar, capable of real time processing. Despite the low time/bandwidth product waveforms (BT equals 270), the breadboard meets -60 dB side-lobe level which prevents contamination of the rain echo by the surface return. The compression principle is a frequent product of samples with a replica, the result of an automatic error compensation procedure that takes into account affects of distortions representative of the radar hardware (such as filters, phase/amplitude ripple, saturation effects of high power amplifier).

PHARUS (phased array universal SAR) is an airborne SAR concept which is being developed in the Netherlands. The PHARUS system differs from other airborne SARs by the use of a phased array antenna, which provides both for the flexibility in the design as well as for a compact, light-weight instrument that can be carried on small aircraft. The concept allows for the construction of airborne SAR systems on a common generic basis but tailored to specific user needs and can be seen as a preparation for future spaceborne SAR systems using solid state transmitters with electronically steerable phased array antenna. The whole approach is aimed at providing an economic and yet technically sophisticated solution to remote sensing or surveying needs of a specific user. The solid state phased array antenna consists of a collection of radiating patches; the design flexibility for a large part resides in the freedom to choose the number of patches, and thereby the essential radar performance parameters such as resolution and swath width. Another consequence of the use of the phased array antenna is the system's compactness and the possibility to rigidly mount it on a small aircraft. The use of small aircraft of course considerably improves the cost/benefit ratio of the use of airborne SAR. Flight altitude of the system is flexible between about 7,000 and 40,000 feet, giving much operational freedom within the meteo and airspace control limits. In the PHARUS concept the airborne segment is complemented by a ground segment, which consists of a SAR processor, possibly extended by a matching image processing package. (A quick look image is available in real-time on board the aircraft.) The SAR processor is UNIX based and runs on easily available hardware (SUN station). Although the additional image processing software is available, the SAR processing software is nevertheless designed to be able to interface with commercially available image processing software, as well as being able to ingest raw data from other SARs on the input side. The combination of the airborne and the ground segment, augmented by the transfer of technological knowledge needed to operate the system, will provide for an autonomous capability of the system user/owner. The PHARUS project has so far resulted in the construction of a C-band, VV-polarized research SAR (PHARS) with a 1- look resolution of 1.5 multiplied by 5 meter (5 multiplied by 5 meter at 7 independent looks) and a swath width of 6 km. This system has been extensively used for research and application projects in Europe, for purposes of mapping, land use inventory, change detection, coastal bathymetry, ship detection and ocean wave measurement. The next system recently completed is a fully polarimetric C-band system with adjustable resolution and swath width (the latter up to 20 km); this system is expected to be operational autumn 1995. The polarimetric capability will provide for a much enhanced discerning power (discrimination between e.g. forest/cultivated, various forest types, etc.). Discrimination by polarimetric signature is an alterative approach, with different possibilities and limitations, to e.g. the use of several frequencies. This paper gives an overview of the SAR research system and the results obtained with this system. The PHARUS design and use are discussed.

Given a high-quality, distortion free SAR image generated by a SAR processor, we would like to remove the speckle to reveal the underlying cross-section. Traditional filter-based techniques all suffer from the fact that the large filter size required for effective smoothing will cover a highly inhomogeneous region. A CPU intensive algorithmic noise removal process called simulated annealing has been successfully applied to both SAR intensity images and SAR texture images generated using a mean-normalized log texture estimator. To improve the execution time of the smoothing process, we have adopted a neural network based solution which emulates simulated annealing. A factorized neural network was chosen, consisting of a vector-quantizer first stage which is used to select a specific multi-layer perceptron from the second stage. This technique reduces both the training and run times for large neural networks. A further reduction in training times is achieved by the use of self-adjusting training algorithms. Statistical analysis of test data has shown that the network produces a good approximation to the estimated cross-section. Simulated annealing has the advantage of a much larger adaptive input window than the neural network, and a better comparison can be made by restricting simulated annealing to operate on a window with dimensions comparable with that of the neural network. A comparison with alternative techniques based on multi- dimensional lookup tables is also presented, comparing both the quality of the result with the execution time of the algorithm.

Most of the processing/analysis tools for SAR images, and particularly the most usual speckle filters, are based on the use of first order local statistics (local mean and local variance). In order to account for the effects due to the spatial correlation of both the speckle and the scene in SAR images, estimators originating from the local autocorrelation functions (ACF) are used, to refine the evaluation of the non-stationary first order local statistics, as well as to detect the structural elements of the scene. The aim is to enhance scene textural properties, and to preserve the useful spatial resolution in the speckle filtered image. To detect and preserve very thin scene structures in the presence of speckle, an heuristic implementation of these estimators is presented for the case of multilook SAR images. Results obtained on 7-look airborne C-SAR and 3-look spaceborne ERS PRI images with different spatial resolutions illustrate the performance of these estimators, either implemented in the speckle filter, or for texture analysis, or for small/thin scene objects detection. Finally, it is shown how two-points statistics and derived indices can be used as texture analysis tools or as discriminators. Some ERS applications using these techniques, either for speckle filtering, or for texture-based analysis, are briefly presented.