POLinSAR 2005 Summaries & Recommendations

 

These recommendations from the POLinSAR 2005 workshop are based on round table discussions initiated by a set of seed questions proposed by the session chairpersons. The chairpersons summarized the round table discussions and presented the summaries in the dedicated summary sessions. In the summaries and recommendations the session seed questions are indicated in italic.

 

SAR Polarimetric Interferometry (Pol-InSAR)		Chair: W. M. Boerner/S. Cloude

1.      Pol-InSAR was developed originally for tree height estimation. How mature is this product now and are there any gaps in the understanding of the performance of Pol-InSAR in different forest types? Is this the best way to get at biomass from radar coverage? There appear to be no saturation effects but how accurate is biomass from height and how robust are the allometric relations used?

 

The most mature Pol-InSAR Application is forest height, there are more than 6 different forest test sites with in-situ comparisons.

 

The results support idea of an L-Band Spaceborne Pol-InSAR for global biomass mapping. Key features:

1.      No saturation means high biomass with L-band space technology (> 300 T/Ha)

2.      Robust performance across diverse forest systems

3.      Key development is demonstrated capability in Tropical Forests (to be confirmed)

4.      Temporal decorrelation is an error source in inversion (needs to be further assessed)

5.      Height-to-biomass is an error source (need clarification on product requirements from users (‘global’ Forestry)) This is a shared problem with Lidar. Radar provides direct estimate of ‘forest’height, not ‘tree’ height…a good thing?

 

2.      What are the prospects for exploiting Pol-InSAR using existing/near future satellite sensors? What are the key limitations on performance? Do we need to wait until the next generation of multi-satellite clusters? Are there sufficient programs in place to try and fully exploit Pol-InSAR for ALOS-PALSAR, Envisat ASAR, Radarsat II and Tandem-X?

 

There are 2 levels of application to consider:

 

1.      Low Frequency repeat pass (L-band) quadpol to combat temporal decorrelation + optimise parameter estimation. Some operational performance possible even for dualpol (HH/HV) systems like ALOS/PALSAR

2.      High Frequency (C and X band)

Vegetation operation seems possible for low biomass (forestry and agriculture) but requires single-pass or short temporal baselines…but

 

Urban applications (Persistent Scatterer InSAR)…have good prospects in 2 ways:

1.      Increasing the density of PS for better performance in D-InSAR

2.      New physical parameters (rotations and twists)…

3.      What is the role of full vs. partial polarimetry in Pol-InSAR products? Can dual pol systems be used for height/biomass estimation with wide swath coverage for example?

Quadpol provides more robust performance but Pol-InSAR information is ‘column’ rather than ‘diagonal’ based in the S matrix (unlike POLSAR) so single TX pol dual RX  can provide inversion products (forestry height). An assessment is needed to decide if degraded product quality is acceptable for applications.

4.      What is the role of advanced signal processing issues based on coherence optimisation, coherent target decomposition etc. Are these crucial to product development or can simpler processing schemes based on standard linear polarisations, Pauli channels etc. be used just as well?

Mature development of algorithms has demonstrated improvement in robustness but optimum algorithm is application specific.Therefore there is a need to provide tools to users with guidance and recommendations for different applications. This is input to development of software tools like POLSARPRO/RAT…etc

5.      What are the prospects for wide area vector tomographic imaging from space? Are there any fundamental problems in processing, data collection etc. or are the requirements well within the limits of existing technology and algorithm maturity? What would be the advantage of tomographic imaging over interferometry? What new products would emerge

·        Processing techniques demonstrate practical application to air/space borne data…

·        Polarimetry is demonstrated to be of importance for image interpretation

Need large number of baselines so well suited to future constellations. ‘system of systems’ An interesting area is the overlap between MBPI and PolTomo.

6.      What about exploiting polarimetry in differential or D-InSAR applications? Are there any clear application areas where this will be important? What extra performance can we expect to get for D-Pol-InSAR vs. classical  D-InSAR systems?

Still early developments, more data sets are needed. But urban applications (Persistent Scatterer InSAR)…have good prospects in 2 ways:

1.      Increasing the density of PS for better performance in D-InSAR

2.      New physical parameters (rotations and twists)…

 

 

Forestry Session		Chairpersons: P. Dubois-Fernandez, D. Hoekman

Seed Questions

1.      Which are the most promising approaches for forest biomass estimation, and what are the limitations?

2.      Is it possible to estimate forest height with L- and/or P-band over tropical forest?

3.      Can we identify a set of forest types over which the different techniques should be tested, validated? Have tests being conducted over these types of forests? What are the estimated limits of each of the identified techniques over these different biotopes?

4.      Can we develop robust biophysical parameter retrieval techniques for complex targets such as heterogeneous, hilly and textured natural forests? What are the needed biophysical parameters? Biomass, tree height, forest density, bhd...?

5.      Most tests have been conducted with airborne data. Is this representative for spaceborne data?  Is resolution an issue for this type of applications? Is the typically steeper spaceborne incidence angle a difficult issue?

6.      Tropical forests are disappearing at alarming rates. Could ENVISAT ASAR be utilized better to mitigate this problem?

 

Comments

·        Estimation of forest height using Pol-InSAR data and the RVoG model was given much attention. Results of validation experiments over a number of European test sites (Papathanassiou) showed consistent results and fair performance over a wide range of forest types, forests heights and biomass levels. The appropriateness of the h100 measure was discussed. Since results seem to depend on vertical forest structure either more complicated height measures or more complicated models may be necessary to improve results further. Since straightforward application of the RVoG model and the use of a single baseline may limit possibilities other approaches were considered.

·        Possible sources of error were indicated such as temporal decorrelation, too much attenuation, non-optimal baseline, and orientation effects in plantations. Also canopy roughness was indicated as an error source (Garestier). Slope effects were not considered to be a major problem (Papathanassiou). Dual polarization InSAR in principle allows forest height estimation and, thus, may be applicable on a larger scale from space, however is not very accurate. Dual baseline space observation (PALSAR polarimetric mode) is expected to meet accuracy requirements. Most of the Pol-InSAR inversion presented in this session was performed at L-band. One presentation however, presented results at X-band. Garestier showed significant penetration at X-band and differential penetration with polarisation, providing evidence that Pol-InSAR biomass inversion can be performed at these higher frequencies. This is of great importance for the future planned system like TerraSAR-X, TandemXSAR or the future C-band system from ESA.

·        Besides dual baseline approaches to mitigate problems hybrid methods involving both classification and parameter estimation were proposed (Lee, Ferro-Famil). For example, prior classification of forest area using PolSAR data, followed by unsupervised classification of forests by Pol-InSAR data yielded promising results. Though classes can be separated in terms of structure, height and biomass it is still not completely clear how to link unsupervised results to groundtruth. Also the use or additional use of X- and P-band DEM and DTM were discussed (Aulinger). In this paper, the authors identified a bias to the estimate of the ground height obtained from the HH P-band interferometry information clearly correlated to biomass. Brian Mercer indicated that previous studies by Infoterra showed that using the optimised polarisation does not have such a bias.

·        The applicability of allometry (Mette) as a means to estimate biomass from height was discussed for a wide range of forest types. Besides errors and unknowns in allometry, such as ecosystems without light competition or tropical forests it was remarked that the Pol-InSAR height estimation may not work accurate in discontinuous open forests or heterogeneous forests.

·        Repeat-pass C-band coherence was shown to be exploitable in the Northern Taiga regions of Siberia, under favourable conditions (Wiesman). A larger range for biomass retrieval is expected from PALSAR.

·        Knowledge on the relation between biomass and P-band intensity data at high incidence angles is of importance for space application. It was shown that results from previous research, at medium incidence angles, are comparable. The only exception being that low biomass areas are confused with bare areas, but this problem could be mitigated by prior classification. This positive result is of importance for the BIOMASCA proposal (Dubois).

·        For applications, which parameters are needed, and at which accuracy? Is forest height a meaningful parameter? Soil moisture, for example, is hard to describe with a single number because in general the soil moisture vertical profile is not uniform. In analogy, could we derive a height distribution? Application areas are not limited to forestry. Numerical weather prediction models or climate models need the aerodynamical roughness of forests to describe the atmospheric boundary layer. Ecologists need data such as crown closure, species clumping, gap dynamics, disturbance. If we only invert for average height and consider canopy roughness as a source of error, we throw away very useful data.

·        And what about biomass? One issue is clearly the global estimation of biomass for which SAR could provide useful information. However biomass information is crucial for other purposes. In carbon offset trading issues, for example, very accurate carbon sequestration figures are needed, above and below ground. This can only be achieved in small permanent sample plots. The role of SAR in this case would be assessing for which areas such sample plots are representative or to detect disturbances.  The temporal resolution of this information is also extremely important. Clearly more discussion with user communities is required.

 

Applications Session		Chairpersons: E. Pottier, M. Sato

 

 

In this session, we discussed on some original and new applications for POL-SAR and Pol-InSAR, such as:

• Urban area (segmentation, coherent points +  PS)
• GBSAR
• Target detection (dual-freq + dual aspect angle, FOPEN)
• Snow monitoring
• Ocean (surface RCS, oil spill detection, Oyster farming)

 

In each application: Obviously and Definitively:

• Polarimetry and Polarimetry + Interferometry play an important key role

• Result improvement
• Physical interpretation

 

Seed Questions

1.      In each measurement, what kind of “unique” calibration and ground truth will be required?

2.      For further development of new technical approaches, what kind of validation data (which include SAR data, but also other physical measurements) should be needed? Are these data or measurements already available?

3.      Who should be involved in order to achieve the final aims of the developed technique?

4.      Is multi-frequency + polarimetry very important for future applications, and what would be the limitations and gains.

 

Comments

Ground Truth - Very detailed ground truth is necessary to:

·        explain the variability observed on data

·        be adapted to end-user needs

 

At the same time, in many cases, we still do not know what the SAR system is looking.

 

Calibration - Calibration is playing more and more an important rule:

• Interferometry + Polarimetry -> effects on correlation and interferometric terms.
• More sensitivity (dynamic range) is required for forest studies
• Same comments as CEOS - CalVal

 

Data Exchange - A priori information is always useful in classification

• Data sets -> yes
• Ground truth: more difficult
• Photographs of the test site
• Information about the calibrator

 

A list of available data sets will be quite useful to obtain new users.

Multi-Frequency:

·        In general, multi frequency will be useful, however, in practice, it has to be limited.

·        We have to continue the discussion on the useful frequency.

·        For a given application :

o       What can we do with multi-channels SAR data (Polarimetry + Interferometry, Polarimetry + Multi-Frequency)?

o       Best configuration?

·        Solution (?) - For a given application <-> select a test site:

o       Collect data from Space borne sensors

-         Radarsat2 (C-Band)

-         Alos (L-Band)

-         TerraSAR-X (X-Band)

-         Envisat – ASAR (C-Band)

-         + auxiliary data (optical data, multi-spectral data …)

 

o       POLSAR / Pol-InSAR / POLfreSAR analysis -> configuration

-         Comparison / Test / Validation coherent / incoherent models

-         Ground truth (?)

 Possibility in the frame of a general / common SO between agencies?

 

Land-Agriculture Applications Session		Chairpersons: P. Lombardo, M. Shimada

 

Seed Questions

1.      How do spatial resolution and revisit time affect the usefulness of SAR remote sensing data for agricultural application? Can recommendations be made for future missions or experiments, especially based on polarimetric interferometry?

2.      Is it still general opinion that X-band carries lower information content on agricultural areas compared to L-band and C-band?

3.      What is the relative importance of multi-frequency vs. polarimetric SAR/InSAR data for agricultural applications?

4.      In terms of importance for the agricultural application, in which order would you list the following items and their combinations: Spatial resolution, radiometric resolution, multifrequency, polarimetry, interferometry, revisit time?

5.      Are data fusion techniques (with reference both to multiple types of SAR images and multiple sensors) for the agricultural applications completely assessed? Would you consider useful to have an opportunity to compare techniques developed by different institutions in an open contest? (in which case what should be the rules to make it useful?)

 

·        A reference single-pass Pol-InSAR data set would be extremely useful to assess processing techniques and compare results among different research centres

 

·        However, the data set would consist of a significant amount of images to represent many different cover types, each one collected with short revisit times (the definition of a significant data set might be hard work itself)

 

·        A potential solution to mitigate difficulty in collecting such data set would be to make a large use of ground-based sensors

 

·        However time is becoming very short for new recommendations on next spaceborne C-Band system, there is a general agreement that the Pol-InSAR community should encourage building fully polarimetric sensors, because:

o       Not having it would be a step back!

o       Even X-band has a penetration useful for biomass estimation, and C-band is expected to allow useful measurements (something in between L and X)

o       Non-polarimetric C-band has now only simple applications (except for PS)

 

·        A document has been prepared as a result of POLinSAR 2005 that motivates this recommendation of polarimetry at C-band from a scientific point of view. 

 

·        Another important issue that should be pushed for, together with C-band polarimetry, is the shorter repetition time (possibly 6 days). In this direction potential interactions with other Space Agencies could be welcome.

 

Theoretical Modelling Session		Chairpersons: I. Hajnsek, T. Le Toan

Strong evolution in theoretical modelling since last POLinSAR 2003

 

1.      Is there a benefit to introduce (more) physical modelling in algebraic modelling in Pol-InSAR ? How to link the two communities?

 

·        There is already a strong use of structural models in Pol-InSAR forest application

·        A link between the direct modelling community is partially established and is strongly recommended to be further developed

 

2.      How reliable are incoherent / coherent forest scattering models today in terms of polarimetric and interferometric observables at X, C, L and P bands? Where are the main limitations?

 

·        There are a set of models established and validated for coherent modelling, but there is still a strong need for further work to be done (forest, soil, agriculture, ice)

 

3.      What is the potential of coherent agriculture soil and vegetation modelling and where are the main limitations?

 

·        The potential is high, but it needs more development

 

4.      Are the existing (coherent) ice scattering models capable for Pol-InSAR applications? What are the required developments?

 

·        No presentation concerning land-ice or sea ice was present

·        A need for more understanding, coherent model and experimental data

 

5.      For mission operation design and performance analysis, data sets with relevant ground data are required for scattering modelling. How large and reliable are the existing data sets and how to make them available?

 

·        There exist forest data for initialisation of direct models (available for all?)

·        There is a need also for agricultural data for initialisation

 

6.      What was the progress with respect to the recommendations of the POLinSAR 2003 WS about:

• Modelling of vertical extinction in tree and forest canopies
• Introduction of higher order scattering for soil surfaces
• Polarisation behaviour of microwaves measurements of crops
• Modelling of polarimetric phase difference in sea ice

 

·        In summary there is a evolution recognised in direct modelling

·        There would be the need to focus the work and to compare the different models(in terms of sensor close development)

 

 

Summary

·        Development of coherent modelling for land (forest, soil, agriculture, ice) is still needed

·        Effort in development of theoretical modelling dedicated to relevant topics (e.g. applications of high scientific priority, critical gaps)

·        Observation/ database is crucial for theoretical modelling, Data need to be available to the community

·        Emphasis on well defined ground data (discussion on common data collection procedures)

Airborne Pol-InSAR Campaigns Session		Chairpersons: S. Hensley, H. Skriver

1.      Is there a need for an airborne campaign where multiple sensors collect Polarimetric Interferometric and/or Polarimetric data at multiple frequencies with associated in situ ground truth data for cross-calibration and algorithm comparisons purposes? What sites would be recommended for such studies?

 

·        No specific recommendation was put forward.

 

2.      Can any significant gaps be identified in Polarimetric Interferometric and/or Polarimetric data sets for validation and development of existing and new techniques, i.e. is there a need for campaigns that fill these gaps for different applications (e.g. forest, hydrology, agriculture, and ice).  If so, what applications, modes, frequencies, sites etc. would be recommended for such campaigns.

 

• Several areas were recommended for dedicated measurement campaigns- urban (use square loop), soil moisture in vegetated regions,
• Any suggested campaign must be linked to the context of planned missions with specific applications.

 

3.      Is there a need for Single Pass Polarimetric interferometric systems to help quantify the influence of temporal decorrelation on repeat pass results? If so what frequencies would be the most useful to implement?

 

• Yes, was suggested in the context of biomass applications. Certainly would he useful at all frequencies but particularly at L-band.

 

4.      Are the geophysical or biological systems that would benefit from persistent polarimetric interferometric observations that may be possible with UAV type systems?  What would be the desired operating characteristics of such a system?

 

• Persistent monitoring of volcanoes, pre and post seismic observations, landslide prone an anthropogenic induced deformations could benefit from such systems.

 

5.      Are there synergistic opportunities for spaceborne/airborne joint campaigns similar to ones previously conducted for the ERS, JERS and SIR-A, B-C missions?

 

• No specific recommendations were put forward. One suggestion would be with ALOS PALSAR. Make repeat observations of temporal intervals of time intervals up to 46 days with increased resolution compared to PALSAR to better understand temporal decorrelations and resolution limitations of PALSAR.
• Care must be made in going from airborne to spaceborne applications to properly compensate for resolution and SNR.

 

6.      Are there any new ideas about new Polarimetric Interferometric observation techniques, such as bi- and multistatic techniques and multi-frequency techniques that require new campaigns.

 

7.      Are there any new applications areas for Polarimetric Interferometry that requires new campaigns for proof of concept, demonstration and validation.

 

Spaceborne Missions for Pol-InSAR		Chairpersons: A. Moreira, J.-C. Soyris

Seed Questions

1.      What future missions have a great potential for Pol-InSAR applications? What recommendations can be given in this respect?

2.      Pol-InSAR vs. "Single-pol multi-baseline" : Pol-InSAR & "Single-pol multi-baseline" interferometry are different acquisitions configurations  (for e.g. for DEM extraction over vegetated areas). What are the advantages/drawbacks of such architectures in the framework of spaceborne missions for vegetation studies?

3.      Spaceborne Pol-InSAR for a global biomass product: Future spaceborne Pol-InSAR missions are expected to provide global biomass estimation with an unprecedented accuracy. What accuracy range, in terms of altimetric & planimetric resolution, localisation, is necessary to allow new applications, in the field of biomass estimation (carbon cycle) ?

4.      Pol-InSAR and bi-static effects: One possibility of achieving low-cost space-borne Pol-InSAR missions is to use a constellation of low cost passive receivers in the vicinity of an illuminating SAR partner (cartwheel, pendulum, ... principles). In this case, the Pol-InSAR acquisition is conducted under a bistatic configuration. What kind of alteration of the received signal is expected? If any, give recommendation on the maximum bistatic angle that can be accepted.

 

Summary

·        Future spaceborne SAR missions will have polarimetric capabilities and will allow the further development of polarimetric and Pol-InSAR applications

·        Temporal decorrelation will be a limiting issue for Pol-InSAR applications. Exception:

o       Polarimetric and D-InSAR (potential to be further explored)

·        Need for a single-pass L-band Pol-InSAR mission (A. Moreira, J.C. Souyris, S. Cloude, K. Papathanassiou and?)

·        Single-pass polarimetry will be possible by utilising a passive micro-satellite constellation together with an active SAR-satellite.

Justification for including a fully polarimetric Capability for future C-Band Satellite Missions

Crop Classification

·        In [1], for a particular classification experiment in the Flevoland agricultural area, it is shown that single polarization (HH or VV) achieves an accuracy under 40 %, dual polarization achieves a result of 60%-70%, and full polarimetry achieves 90%.

[1] Hoekman, D.H., and M.A.M. Vissers, 2003. A new polarimetric classification approach evaluated for agricultural crops, IEEE Transactions on Geoscience and Remote Sensing, Vol.41, No.12, December 2003, pp.2881-2889.

·        Classification accuracy using fully polarimetric data increases by ca. 20% when compared with dual-pol [2]

[2] J.S. Lee, M. R. Grunes and E. Pottier, “Quantitative Comparison of Classification Capability: Fully polarimetric versus Dual- and Single polarization SAR,” IEEE TGRS, November 2002

·        Fully polarimetric information is required for an unsupervised crop classification scheme providing robust classification of important crops (ref. Skriver, Quegan, LeToan)

·        Supervised Wishart classification of multitemporal C-band Pol-InSAR data showed an improvement of ca. 15 % from dual-pol to full polarized data (Skriver et al, 2000).

·        A key parameter for discriminating crop types and monitoring the state of crop development is the height of the crops. Simulations, laboratory measurements and actual data have shown that it is possible to measure crop heights using polarimetric interferometric techniques at high frequencies. By incorporating short temporal baselines, C-band polarimetric interferometry proves to be a potential operational tool for agricultural applications (Scott Hensley et al).

 

Snow Monitoring

·        Fully polarimetric data is essential for accurate snow classification since an optimization of the polarimetric contrast and other polarimetric parameters are needed for discrimination between dry snow and uncovered areas. Algorithm is also robust against topographic effects (ref. A. Martini, L. Ferro-Famil, E. Pottier) 

 

Sea Ice

·        Phase difference is important for discrimination between thin ice and open water (H. Skriver, W. Dierking)

 

Strategic & Programmatic Issues

·        A full polarimetric capability for C-band satellites represents a new dimension for product development

·        Radarsat-2 SOAR statistics demonstrate the strong request for full polarimetric data that will foster the development of a variety of new applications

·        Complementarity and compatibility to RADARSAT-2 and RADARSAT-3 satellites with full polarimetric capabilities