Interferometric SAR parameters estimation is a very important and challenging problem. The InSAR2InSAR method previously proposed is one of the few self-supervised methods that aims to estimate InSAR parameters. This method has proven to outperform state-of-the-art methods on simulated synthetic data. However, it has to be extended on real data. In this letter, we demonstrate that Sentinel-1 images acquired in the Interferometric Wide Swath mode possess the necessary properties to train and apply InSAR2InSAR effectively. In this paper, we demonstrate the ability of InSAR2InSAR to process across-track Sentinel-1 interferometric images with state-of-the-art performances.

In recent decades, oil palm plantations have expanded significantly in Indonesia. Palm oil production currently relies on two plantation types with different economic, social and environmental impacts: (1) industrial plantations and (2) smallholder plantations. While efforts have been made to characterize these plantation types, this objective remains challenging for the remote sensing community. Consequently, this study assesses the potential of Sentinel-1 textural metrics in distinguishing between industrial and smallholder plantations, offering potential solutions to this persistent challenge. We used machine learning algorithms (Random Forest vs. eXtreme Gradient Boosting Tree) with textural features calculated by the Grey Level Co-occurrence Matrix. The results confirmed the potential of Sentinel-1 textural metrics to discriminate OP plantation types. The XGBTree model achieved a higher Kappa (0.71) than the Random Forest model (0.63). Moreover, the contrast, dissimilarity, GLCM-Mean, and GLCMVariance metrics were the most explanatory for discriminating industrial and smallholder plantations. This study possesses limitations, especially concerning its applicability on a broader scale. However, leveraging cloud computing tools like Google Earth Engine could aid in scaling up the methodology.

SAR time series are powerful assets for forest monitoring. In recent years, they were involved in various classical forest applications such as forest mapping [9] . These applications largely benefited from the advances of Deep Learning, particularly Unsupervised Learning, using Convolutional Autoencoders in applications such as wildfire detection [4] . Not only did purely temporal approaches offer high prediction performance compared to spatiotemporal variants, but the unsupervised autoencoder rivaled its supervised counterparts. The monitoring of forests also involves the detection of under-canopy targets, which could disturb protected environments. The literature mostly relies on classical SAR approaches PolSAR change detection [8] . A recent shift towards the usage of SAR time series displayed promising performance [10] . Thus, to fully exploit the potential of multi-temporal SAR imagery, this paper proposes the usage of unsupervised Deep Learning, particularly the Convolutional Autoencoder, to detect under forest cover objects.

International audience

This study compares two approaches for simulating synthetic aperture radar (SAR) images. The first approach uses a conditional Generative Adversarial Network (cGAN) to learn statistical image distributions from optical images. In a second approach, we generate SAR images using a electromagnetic simulator taking into input material maps obtained by segmenting optical images. We propose two metrics to evaluate the quality of the simulation. We evaluate the methods on existing Sentinel-1 SAR images of France using the DREAM database. The results suggest that the physical simulator with automatically created material maps is better suited for generating realistic SAR images compared to the cGAN approach, even if a lot of work remains to be done on the complexity of the description of the scene.

The C-Band radiometric signature of boreal forests is highly seasonal, with apparent correlations to temperature changes. Within these seasonal components, we assume that information related to tree height can be extracted. We apply a one-dimensional Convolutional Neural Network to assess this assumption, intending to retrieve tree height measured by Airborne Laser Scanning from C-Band Sentinel-1 time series. A study site in the Parc National des Grands Jardins, in Québec, Canada, was selected for this analysis. Prediction-wise, we reach an R2 score of 0.45 and an RMSE of 1.84m, following a 4-fold cross-validation, which exhibits a non-negligible influence of the tree height parameter on boreal forest radiometric response in C-Band Synthetic Aperture Radar, despite the presumed fast saturation of this wavelength, when observing forested environments. In addition to performance metrics, we use a gradient-based explainability tool to diagnose the most contributing periods of the input time series to predict tree height to better correlate the seasonal conditions of this parameter’s influence on the forests’ radiometry.

This paper introduces a tool for explaining the latent space generated by applying convolutional autoencoders to satellite image time series, entitled Grad-SLAM. We rely on backpropagated gradient interpretation combined with network activation localization. We use the proposed formula for multiple layers of the encoder, then scale and merge the results to generate a single date contribution metric for the generation of the latent space. We illustrate the potential of this method with the study of the unsupervised classification of agricultural Sentinel-1 time series. We show that critical characterizing dates for unsupervised retrieval of a given class are conditioned by the crop type’s radiometric signature and class count. We also present how Grad-SLAM can be used to enhance the understanding of unsupervised classification confusion.

Numerous glacier velocity observations, derived from spaceborne imagery, are available online, but it remains difficult to analyze them because they are measured with different temporal baselines, by various sensors. In this study, we propose a novel formulation of the temporal closure to fuse multi-temporal multi-sensor velocity observations without prior information on the displacement behavior and the data uncertainty. We establish a system of linear equations between combinations of displacement observations and fractions of estimated displacements. The proposed approach provides a velocity time-series with a regular and optimal temporal sampling, the latter representing a compromise between the temporal resolution and the signal-to-noise ratio. The proposed approach is first evaluated on synthetic datasets and second on Sentinel-2 and Venμs velocity observations over the Fox glacier in New Zealand. The results show the intra-annual variability of Fox glacier surface velocity with a reduced uncertainty and complete temporal coverage.

Today, more and more velocity observations are available online or on-demand. However, this amount of data is complex to analyze since velocity observations span different temporal baselines. Velocities obtained from a small temporal baseline are close to the derivative of the displacement but are more likely to be contaminated by noise. Velocities obtained from a long temporal baseline approximate the mean velocity between two dates but can be affected by temporal decorrelation. Having short and long temporal baselines provides a data redundancy that needs to be properly considered. In this article, we propose a method that aims to extract short-term velocity time series with regular temporal sampling from all available displacement observations. The proposed method relies on a temporal inversion based on an improved temporal closure of the displacement observation network. Two criteria are proposed to determine the optimal temporal sampling to study short-term variations. To take the unequal data uncertainty into account, the temporal inversion is done by an iterative reweighted least square using a well-established weighting function, without preprocessing. The proposed method results in velocity time series with an optimal temporal sampling, improved temporal coverage, reduced uncertainty, and no redundancy. The studied area is the Kyagar glacier, in the North of the Karakoram range that is characterized by strong velocity variations originated from a glacier surge and additional seasonal variability.

We present a fully unsupervised learning pipeline, which involves both a projection method and a clustering algorithm dedicated to the pixel-wise classification of multitemporal SAR images. We design a Convolutional Autoencoder as the method to project our time series onto a lower dimensional latent space, where semantically similar temporal signals are placed close together. The additional use of convolutional layers as feature extraction steps allows us to exploit the sequential nature of time series, exhibiting higher representation performance than fully connected layers. The extracted clusters can encapture different semantic levels to either separate classes or extract outlying temporal signals. The application of this method to crop-types mapping enables the extraction of major crop-types within a scene, without supervision. In a labeled context, this method also allows for the extraction of outlying profiles which can lead to the discovery of mislabeled time series.

This article proposes a method based on the temporal closure of the displacement measurement's network. The aim is to extract short-term glacier velocities and to use data redundancy to reject outliers and reduce uncertainty. By using all the available displacement measurements, we retrieve a displacement time series between consecutive observation dates by means of an inversion. The proposed inversion method is an Iterative Weighted Least Square (IWLS) with a regularization on the discrete derivative of displacements. We apply our method to a glaciers velocity data-set covering Fox Glacier in the Southern Alps of New Zealand.

This paper describes a multimodal remote sensing database, called DREAM (Data Rang for EArth Monitoring), developed from open-source data. The method of development of the database is described, as well as its specificity. The database includes radar, optical, rasterized Open Street Map datasets, and DEM data for two countries containing a variety of landscapes: France and the United States. We present results of preliminary work using these datasets, including multimodal coregistration, and radar simulation from optics by image to image translation.

Temporal Convolutional AutoEncoders are used as feature extractors to project time series onto a latent space where similarity detection can be easily performed. This model can generate accurate descriptors of the temporal profile of the input time-series. We apply this algorithm to PolSAR S1 uncoherent SAR time series where the model learns highly discriminative data representations. This reduction method is compared to others such as PCA or Temporal Averaging and is shown to outperform them when leveraging the learnt representation using K-Means clustering.

We introduce a multi-branch Deep Learning architecture that allows for the extraction of multi-scale features. Exploiting the data multi-modality structure through the combined use of various feature extractors provides high performance on data fusion tasks. Furthermore, the representation of the multi-temporality of the data using sensor-specific 3D convolutions with custom kernel size extracts temporal features at an early computation stage. Our methodology allows reaching performance up to 0.8876 F1 Score on the development phase dataset and around 0.8798 on the test phase dataset. Finally, we demonstrate the contribution of each sensor to the prediction task with the design of data-focused experiments.

Confusion between the polarimetric radar signatures of rotated buildings and vegetation has been widely studied to compensate for the effect of rotation. We propose in this study to have a new look on this problem. To do this, we have selected different areas in the San Francisco bay and in New Orleans. These selected zones present different orientations with respect to the sensor illuminating the scene, RADARSAT-2 (C-band) for San Francisco or UAVSAR (L-band) for New Orleans. The variations with the orientation angle of the VV/HH and HV/HH responses collected over these areas are almost identical whatever the sensor. These two quantities first grow and rapidly reach a plateau, at a level which is similar to the responses of forests we collected throughout literature. Actually, it seems that, when the rotation angles grows, the polarimetric radar responses of urban areas tend to limit values which are similar to the radar responses of forests.

This paper proposes a method of denoising SAR images, using a deep learning method, which takes advantage of the abundance of data to learn on large stacks of images of the same scene. The approach is based on the use of convolutional networks, used as auto-encoders. Learning is led on a large pile of images acquired on the same area, and assumes that the images of this stack differ only by the speckle noise. Several pairs of images are chosen randomly in the stack, and the network tries to predict the slave image from the master image. In this prediction, the network can not predict the noise because of its random nature. Also the application of this network to a new image fulfills the speckle filtering function. Results are given on Sentinel 1 images. They show that this approach is qualitatively competitive with literature.

This paper presents the overall strategy of fusion of full waveform LIDAR and L-band Polarimetric and Interferometric radar (PolInSAR) images of forests in order to predict the forest vertical profile where there is no LIDAR information. The images considered are the radar dataset collected by the Uninhabited Vehicle Synthetic Aperture Radar (UAVSAR) and the Lidar vertical full waveforms acquired by LVIS over boreal forests in the Canadian province of Quebec. We propose dataset descriptors and fusion methods in order to predict lidar from radar. The first challenge is to find features that go beyond the difference of geometrical configurations between the two types of information, and also that compensate the effect of the incidence angle on radar observables. This has been studied in previous work [1] and will be used here in order to focus on the fusion methods. In this paper, we aim to predict the vertical structure of a forest canopy from PolInSAR images. We assume the Lidar waveforms are a good descriptor of structure and use 3 decomposition methods to qualitatively characterizes these waveforms: 1-Relative Height (RH) metrics, 2-Legendre decomposition and 3-spectral clustering. The prediction will be obtained by a neuronal network which requires an input vector representing the PolinSAR data. We extracted 7 parameters from the PolinSAR images: [θmean.hα, θ0.hα, γmean, λmeax, λmin, Rp, Ra]. Prediction of the RH produced a root mean square error (RMSE) of 3.6 m for the top height (I.e. RH100). On the other hand, the Legendre coefficients were predicted with an accuracy of 15%, and the spectral clustering classes were obtained with accuracies better than 80% with 2 classes, but rapidly decreasing as the number of classes (waveform shapes) is increased.

In this study application results are presented derived from multi-parametric SAR observations covering five different thematic domains: forest, agriculture, ocean, urban and cryosphere. In total 21 application products have been selected and described. Their application on different data sets, space- and airborne sensors, was demonstrated and can independently be reproduced by any scientist. The results and algorithms are available soon through Springer.

In the context of rapid global urbanization, urban environments represent one of the most dynamic regions on earth. Even in developed countries the yearly conversion of natural or agricultural space into residential, industrial or transport areas frequently exceeds 100 ha. The current increase in population has resulted in widespread spatial changes, particularly rapid development of built-up areas, in the city and its environs. Due to these rapid changes, up-to-date spatial information is requisite for the effective management and mitigation of the effects of built-up dynamics. Various studies have shown the potential of high resolution optical satellite data for the detection and classification of urban area. Nevertheless optical satellite imagery is characterized by a high dependency in weather conditions and daytime. Thus, particularly in case of regional and national surveys within a short period of time, disaster management, or when data have to be acquired at specific dates, radar systems are more valuable. Thus, the new generation of civil space borne Synthetic Aperture Radar (SAR)-Systems with short revisit can serve as a valuable instrument. Furthermore, there is today a growing access to available data, with revisit times shorter and shorter, improved available resolutions, and a diversification of polarimetric modes. Thus, promising approaches towards the classification of urban area include the analysis of multipolarized image analysis.

This letter focuses on the analysis of layover effects in interferometric synthetic aperture radar (SAR) data of urban areas. In particular, we derive two formulations to express the backscattering coefficient and the interferometric coherence in this case. These equations show that the backscattering coefficient and the interferometric coherence in layover areas can be seen as a combination of the backscattering coefficients and interferometric coherences of the individual scattering mechanisms. These formulations are then tested on interferometric SAR (InSAR) data and analyzed statistically.

In this letter, a fast and robust optical-flow estimation algorithm is investigated for synthetic aperture radar (SAR) images' coregistration. The principle of the initial algorithm is described, as well as its adaptation to the case of radar images. A performance evaluation method is proposed to fix the choice of the parameters of the algorithm. Promising results in change detection or interferometry between SAR images of different resolutions are presented. They offer the opportunity to use this kind of algorithm in the case of high-resolution images containing many structural elements as in urban areas.

The objective of this paper is to propose an alternative measurement device at optical scale to help understanding of bistatic polarimetric SAR images of forests. The device is employed to measure nanoscale trunk forests with a 106scale factor. Considering the scale invariant rule in electromagnetic scattering this device would enable to predict the Mueller matrix of a whole tree forest for P band radar measurements. The device presents the advantages to be low cost and to get a complete set of bistatic configurations at once. Moreover, the measurements are very fast and would potentially be done on infinitely diverse and well controlled forest structures. Multiplicative decompositions of the Mueller matrix are presented. We underline the trunk density influence for an entire scope of bistatic configuration.

This paper investigates the use of polarimetry to improve the estimation of the height of buildings in high-resolution synthetic aperture radar (SAR) images. Polarimetric coherence optimization techniques solve the problem of layover effects in urban scenes by allowing a phase separation of the scatterers sharing the same resolution cell. Bare soil elevation estimation is also improved by the polarimetric phase diversity. First, we present an analysis of the statistical modeling of the generalized coherence set. A building height estimation method is then derived from this analysis. Finally, the method is tested and quantitatively validated over an X-band polarimetric interferometric SAR (PolInSAR) airborne image acquired in a single-pass mode, containing a set of 140 different buildings with ground truth.

In this paper, our concern was to help understanding the different scatterers responses in an interferometric image of a urban area by studying the different mechanisms that can occur inside the urban canyon. This study relies on a geometric code we specifically developed for this kind of scene. In this example, because of the reflections on the walls, we highlight the impact of the Brewster angle on the interferometric signature. For this reason, we recommend the use of different polarization and incidence angles for urban areas analysis of InSAR image.

In this paper, we present numerical simulations of forest scattering in bistatic configuration. The aim is to compare the influence of the structure of the forest on typical measurement parameters in a bistatic and polarimetric framework. Bistatic configuration pledges advantages of cost and stealth. The scattering model COBISMO has been developped to simulate such configurations. Simulations are performed on four forest structures with different dispositions of branches. Results show an important influence of branch inclinations but not a real influence of their position.

This paper deals with the behavior of the depolarization of natural environments in bistatic settings. In bistatic radar imaging, we seek simulation tools capable of predicting this depolarization for any geometric configuration. As we lack actual real data to validate such tools, we propose an alternative measurement method, at the optical scale, on depolarizing media consisting of carbon nanotubes. We present the first results of such measures, and we offer a number of phenomenological interpretations using also a simulation tool.

In this paper, we are interested in identifying all areas inside a urban canyon that can be illuminated by a radar in NLOS (non line of sight) configuration. We developed a simple model to identify, according to each canyon configuration, the ground canyon areas illuminated by the radar and also the non illuminated areas. We test the results of our algorithm for a specific canyon. To validate our theoretical results we finally present some measurements that will be performed in the anechoic chamber of ONERA on a scaled urban canyon in the case of far field.

Within the frame of bistatic polarimetry, this paper discusses the entangled effects of bistatic geometry and target features on polarimetric measurements. Three different geometrical effects are distinguished: antenna rotations, target orientation, and bistatic angle. Antenna rotations are addressed through the use of polarimetric bases taking the scattering plane as the reference plane. Target orientation effects are not considered since only spheres are studied. This paper focuses on the bistatic angle effect through a bistatic polarimetric analysis on classical parameters. Targets consisting of single or multiple spheres in the resonance region are investigated. Finally, the results of indoor polarimetric measurements on such targets are presented and discussed.

In this study we give short overview of polarimetric SAR image visualization with colors. By studying the color models and polarization visualization models we propose basic principles which should be followed when presenting polarimetric information in color. We show that for different polarimetric parameters, different color models should be used, and give guidelines for color model selection. We present also two visualization schemes which are suitable for interpretation and browsing of large polarimetric SAR images.

The purpose of this paper is to discuss two polarimetric parameters which are widely used in synthetic aperture radar (SAR) polarimetry, namely, target entropy and alpha angle. We propose alternative parameters based on our analysis on how they are connected to covariance matrix similarity invariants and how they can be physically interpreted in optical polarimetry. The proposed alternatives can be computed by a fairly simple algorithm and even by the use of software without complex mathematics abilities. As an example, a NASA/Jet Propulsion Laboratory Airborne SAR L-band image of the San Francisco Bay is used to compare the proposed parameter schemes with the original entropy and alpha. A coherent rationale for these alternative parameters is formulated in order to provide insight to polarimetric parameter interpretation.

Here are presented issues about building or urban area high resolution imaging using synthetic aperture radar (SAR). Promises of circular imaging are assessed. Indeed, this acquisition mode, though available only to airborne sensors, has the potential to solve shadow and overlay problem through rotation of the slantrange projecting direction. It has also an intrinsic elevation dependency that would allow "along track stereo" reconstruction. This is illustrated from several recent circular/spotlight acquisitions over urban areas.

The airborne radar system RAMSES collected data over the Sweden forest to investigate the capabilities of detection at P-band and influences of different SAR parameters like resolution, central frequencies, and look angle. During this campaign, circular trajectories have been performed in order to analyze the presence of anisotropic scattering from the targets at P-band. We show the treatment of two circular trajectories using the Flashlight imaging mode and its ability to collect several SAR data with different look angles. The scope of this paper includes a general description of the operating mode of flashlight SAR images in the interferometric mode and the use of it for FOPEN purpose. The capabilities of detection of such polarimetric images have already been investigated, but without using the interferometric mode [1]. It had been shown that using circular trajectories at P-band was necessary to detect targets: first in our configuration detection was very difficult because the forest was very dense, secondly the targets were very sensitive to the orientation angle of the radar, which should not be the case at lower frequencies. Moreover, polarimetry and polarimetric interferometry are useful tools once the acquisition conditions ensure that detection is possible. Now, the full polarimetric and interferometric information are used in order to explore the potential of the POLINSAR circular mode to yield high detection rate.

SAR interferometry, requires the sensor separation to be below a critical baseline above which the coherency is lost as the ground projected frequency ranges do not overlap (frequency band shifting of one sensor -the delta-k system- is not addressed here). Indeed, as critical baseline increases with both wavelength and bandwidth, low frequency and/or wide band SAR systems have less stringent constraint for interferometric coherence, which is especially worthy for airborne acquisitions. However, as mentioned in earlier publications, the number of fringes apparent on an interferogram is (roughly) limited to twice the inverse of the relative bandwidth. Hence, wide band interferometry requires range migration evaluating for maintaining coherency, this range migration evaluation is similar to a stereogrammetric measure. As relative bandwidth grows, the two techniques merge, which can be illustrated from FOI Carabas VHF SAR data (of which the relative bandwidth is 111%).

After a first series of full circle SAR acquisitions in L and P-bands during a 2004 joint FOI-ONERA campaign in Sweden, ONERA experimented in 2006 high resolution (15 cm) polarimetric, full circle acquisitions in France and Germany using its X-band sensor. In order to cope with narrower antenna pattern and aircraft attitude fluctuations, a steerable antenna was used. Furthermore, an experimental setup for retrieving high accuracy trajectory was installed. This paper describes the processing of this signals.

Radar polarimetry aims to determine the scattering properties of a target or scatterer. For this purpose, the scattering matrix can be analyzed and represented in several ways using various techniques to extract information about the scattering mechanisms. Polarimetry and ellipsometry are techniques which both study the properties of the polarization of the scattered waves but traditionally refer to different wavelengths: optical wavelengths for ellipsometry, and high-frequencies radio waves for radar polarimetry. This paper deals with natural targets in the general bistatic case, for which the 16 parameters of the Mueller matrix are independent. We try to answer the following questions: how to deduce the radar polarimetric parameters from the optical measurements of a Mueller ellipsometer? What are the polarimetric parameters traditionally devoted to optical images, and in which extent is it possible to apply and interpret them in the case of SAR images?

The objective of this paper is to examine the link between the attenuation coefficients and the interferometric phase center heights, for several frequencies from P- to L-band, and to study the extent to which it depends on the canopy architecture and description. This paper relies on the use of a coherent and full polarimetric scattering model, which simulates the fields backscattered by a forested area. In the first part, we study the behavior with a frequency of the interferometric phase center heights, and in the second part, we focus on the attenuation coefficients. Then, we compare the behaviors of these two quantities, and we propose to empirically derive a relation between these two quantities and the mean forest height. Finally, we investigate if a change in the initial forest or radar configuration has an impact on the determination of this relation.

The objective of this paper is to examine the link between the attenuation coefficients and the interferometric phase center heights, for several frequencies from P-band to L-band, and to study in what extent it depends on the canopy architecture and description. This study relies on the use of a coherent and full polarimetric scattering model, which simulates the backscattered fields by a forested area. In a first part, we study the frequential behavior of the interferometric phase center heights and in the second part, we focus on the attenuation coefficients. The behaviors of these two quantities are compared and in a third part, we propose to empirically derive a relation between these two quantities and the mean forest height. Finally, we investigate if a change of the incidence angle has an impact on the determination of this relation.

This paper investigates to what extent a new interferometric coherence optimization in radar polarimetry allows the separation of point scatterers located in the same resolution cell according to their interferometric phases. An interferometric coherence definition called the single-mechanism coherence is introduced, and the corresponding optimization method is briefly discussed. This method was first validated theoretically when no volume decorrelation occurs. Then, it has been applied to simple target measurements acquired in an anechoic chamber, and to an X-band polarimetric and interferometric synthetic aperture radar image containing man-made targets. In both cases, the single-mechanism coherence optimization enables to resolve the interferometric phases of several scattering centers inside the same resolution cell.

The interpretation of radar data would ideally require extensive and numerous observations. However, the number of observations is limited by the difficulty and the cost of acquiring ground truth and radar data. On the other hand, numerical models can provide a wide range of situations, both in inputs and in outputs. More precisely, they have to provide radiometric, polarimetric, and interferometric simulations and be applicable to various forested areas (high density, high/low moisture, inhomogeneous area, etc.) and radar configurations (low/high frequency, bistatic observation, etc.). This paper is dedicated to the presentation of the capabilities of a descriptive coherent scattering model (COSMO) applied to the electromagnetic study of the backscattering by forested areas. Improvements have been implemented in order to produce in output a radar image, which can be treated with the same polarimetric and interferometric tools as those applied to real synthetic aperture radar images. Thus, comparisons are possible. COSMO has been widely tested from P- to L- bands, over temperate and tropical forests and applied to radiometry, polarimetry, and interferometry. It appears finally as an efficient simulating tool to carry out parametric studies and to analyze how the total scattered field is built from canonical mechanisms and individual scatterer contributions.

The interpretation of SAR data remains particularly difficult in case of forests, which makes the use of modelling very helpful. On the other hand, the potential of combined use of interferometric and polarimetric data has been already demonstrated to provide key forest parameters. In this paper, we present comparisons between interferometric polarimetric images of a forest simulated by a coherent scattering model derived from L. Thirion (2003) and L. Thirion (2004), and real airborne and full polarimetric interferometric P-band data measured over maritime pine trees. Results are presented that demonstrate the efficiency of the model to retrieve polarimetric and interferometric parameters on this forest. Finally, a first step of simplification and inversion of the model, relying on the analysis of the scatterers contributions, is proposed

The interferometric coherence optimization methods are based on the definition of projection vectors, which is the way to combine the information occurring from the polarimetric channels of both interferometric images. Optimization algorithms have already been proposed. In order to know in which cases these algorithms really improve the scalar interferometric result, a new entropy parameter H' calculated with the eigenvalues found by the second optimization method and called IFPOL entropy is introduced. The physical significance of this entropy parameter is first explained. Particular cases where this parameter is equal to 1 are studied. Finally, we explain how this parameter H' can be used in order to obtain more information on the scene and, for example, to improve terrain classification methods.