Prof. Jin A. Kong
Department of Electrical Engineering
Massachusetts Institute of Technology
Cambridge, MA 02139





SIR-C Polarimetric Radar Image Simulation and Interpretation Based on Random Medium Model


OBJECTIVES
Demonstrate the applicability of the random medium model in simulating SIR-C imagery.

Analyze and interpret SIR-C imagery for remote sensing applications.

Investigation of seasonal variations and atmospheric effects.


PROGRESS
The work during this period has been focused on the use of spaceborne polarimetric radar measurements for monitoring, mapping, and retrieving the above ground vegetation biomass. Fully polarimetric radar data obtained from the SIR-C/X-SAR missions in April and October 1994 over the Landes Forest in Southwestern France have been analyzed in detail. The Landes forest is the largest plantation forest in France, and covers nearly one million hectares of flat topography. This forest is almost totally formed by maritime pine (pinus pinaster), and it has been managed in such a way that the forest is divided into various areas of large and statistically homogeneous tree stands of the same age. The acquired SIR-C data has been compared with the previous AIRSAR campaign (L- and C-band, fully polarimetric, 40-50 degree incidence angle), ERS-1 data (C-band, VV, 23 degree incidence angle), and JERS-1 data (L-band, HH, 35 degree incidence angle) to assure the consistency of measurement.

In the investigation of the application of SIR-C data to vegetated terrain classification and biomass inversion, the measured backscattering coefficients [Sigma]0hh, [Sigma]0vv, and [Sigma]0hv,), the derived complex correlation coefficient ([rho]) of HH and VV polarizations as well as the ratio between cross- and co-polarization ratio ([Sigma]0hv / [Sigma]0vv) are fully utilized. A validated pine forest scattering model, which is based on the radiative transfer theory with the specific branching structure of pine tree taken into account, is used to interpret the SIR-C/X-SAR polarimetric backscattering measurements from the Landes forest. From the analysis of measured data and the theoretical simulation, the cross-polarization backscattering coefficients at L-band and the correlation between HH and VV backscattering returns at both L- and C-band are found to be most useful for the biomass retrieval. Bayesian classifications using data with known ground truth and with theoretical simulation are applied to classify the forest for biomass up to 50 tons per hectare with the available data at this time (26 degree incidence angle). With the use of pine forest scattering model, biomass inversion has been shown to be feasible over a wider biomass range (up to 100 tons per hectare) for angles of incidence around 45 degrees. In addition to the analysis of SIR-C/X-SAR data, we have refined our forest scattering model by taking into account the double scattering mechanism between trunk and branches which shows more effects on the cross-polarized backscattering return. We have also studied the collective scattering and absorption effects of clustered objects like the branches and leaves in a vegetation canopy. A new approach for studying the polarimetric response of various types of forest is also developed by using the L-systems technique to generate different kinds of plants.

In this work, we have collaborated closely with Dr. Le Toan's research group at the Center d'Etudes Spatiale De La Biosphere (CESBIO) of France. During the different flights of SIR-C/X-SAR over the test site, extensive ground truth data had been collected by Dr. Le Toan's team. These consist of an updated biomass map which provides the location and ages of more than 50 stands of maritime pines, as well as the statistical information about the densities and sizes of trees and branches. In addition, a clear-cut map is also available with some ground truth measurements including soil moisture and surface profiles. These valuable descriptions provide the key input parameters for our theoretical pine forest scattering model.


SIGNIFICANT RESULTS
The fully polarimetric backscattering measurements over the Landes forest from the first SIR-C flight was taken at a 26 degree incidence angle. In Figures 1 and 2, we compare the backscattering coefficients and the correlation coefficients between the SIR-C measured and model predicted data for both forest stands and clear-cuts at L- and C-band frequencies. The comparison between experiment and theory shows good agreement. The clear-cuts are areas where the biomass is lower than 5 tons/ha. For the area with bare soil surface, the magnitude of the correlation coefficient is close to 1 either at L- or C- band. When the area with forest stands is considered, the magnitude of rho drops to a value of 0.35 for older stands.

During the second SIR-C flight, measurements with different incident angles (18 and 51 degrees) were obtained. We then performed theoretical simulations to examine the scattering mechanisms involved in the angular variations with backscatter from forest. It is found that at a higher incident angle (51°), where the scattering from tree crown dominates for both L- and C-band, the cross-polarized return at L-band has larger dynamic range than lower incident angles. For small incident angles, the copolarized return from ground is more important for forest stands with low biomass. This suggests that the results may differ with surface conditions.

Classification and Biomass Estimation of the Landes Forest

With the ground truth from biomass map, the supervised classification of forest stands between 0 tons/ha and 50 tons/ha has been performed with a set of multi-look (5 x 5), fully polarimetric data at 26 degrees. The data are divided into 5 classes: from 0 to 7 tons/ha, from 8 to 20 tons/ha, from 21 to 33 tons/ha, from 34 to 50 tons/ha, and the one with more than 50 tons/ha. Using Bayes classification algorithm, an accuracy of 86% has been achieved. The accuracy of classification with backscattering coefficients only is 62%. Unsupervised classification of forest stands using theoretical models gives an accuracy of 70%. It is also found that for the classification of bare soil with forest areas, the ratio [Sigma]0hv / [Sigma]0vv and the magnitude of [rho] yield the best results. For both L and C-band frequencies, the forest can be classified with ratio higher than -11 dB or the magnitude of [rho] lower than 0.85. As for the biomass retrieval, the magnitude of [rho] gives the best performance.

Collective Scattering and Absorption Effects

For a locally clustered medium, the scatterers are clumped together like branches and leaves in a vegetation canopy. In such cases, the scatterers will scatter collectively. Collective scattering effects include correlated scattering, which takes into account the relative phase of scattered waves from the scatterers and their neighbors. The mutual coherent wave interactions between scatterers are also included. The locally clustering structure has important effects in determining the cluster's electromagnetic properties. We have shown that, in locally clustered media, the absorption of the cluster can be several times greater than the incoherent sum of the absorption of its components. This suggests that in random media problems the effects of the clustered geometry on both scattering and absorption must be considered.

Theoretical Modeling of Forest Using L-Systems

The Monte Carlo approach has also been applied to study the scattering of electromagnetic waves by plants that are grown using the L-systems technique. The position, size, and orientation of every element in a generated tree can be obtained from the computer simulation. The scattering fields from all tree elements are added coherently to calculate the total scattering field. The results are further averaged over many tree realizations. Simulations of different types of trees show that the polarimetric backscattering behavior of forest is affected by the inner structure of plants.


FUTURE PLANS
The overall objective of this study is to investigate the application of spaceborne polarimetric data for classifying, mapping, and parameter retrieval of vegetated terrain using SIR-C/X-SAR polarimetric data. Based on our research goals, the planned investigations include:

Classification of Various Types of Forests

We will investigate the capabilities of radar data to achieve the classification of forests in a worldwide scale into major types according to the tree architectural forms. For each type of forest, the corresponding SIR-C/X-SAR images will be requested, and the backscattering coefficients will be evaluated. We will also investigate the relationships between polarimetric discriminations and tree structures and species. In addition to the Landes pine forest, the forest of eucalyptus in Congo, and the rain forest in Amazon will be studied. For different types of trees, the corresponding tree architecture model and the backscattering model will be set up with the use of L-systems technique.

Monitoring Rice Growth

In this study, the identification of rice fields at various growth stages, as well as the classification into different species will be investigated. At present, a theoretical scattering model for rice field at different growth stages has been developed at MIT, it will be applied to interpret the radar measurements. The test sites will be some rice fields at Indonesia and Japan. These sites have been the subject of remote sensing project using ERS-1 data to monitor rice growth. ERS-1 data will also provide us with samples of measurement at various rice growth stages for comparison.

Soil Moisture and Roughness Inversion

In order to assess the inversion of soil moisture from polarimetric data, a forward rough surface scattering model has been developed at MIT. This model is using the Monte Carlo approach which solves a 3-D random rough surface scattering problem numerically. A neural network is trained based on the direct scattering model and will be subsequently used to invert the SIR-C/X-SAR data. The selected sites with supporting ground truth are the Landes forest in France and Matera in Italy.


PUBLICATIONS
C. C. Hsu, J. A. Kong, T. Le Toan, S. Paloscia, and P. Pampaloni, "Microwave Emission and Backscattering from Crops," Proc. PIERS '94 Symp. , Noordwijk, Netherlands, July 1994.

F. Scire-Scappuzzo, C. C. Hsu, L. Wang, J. A. Kong, and T. Le Toan, "Biomass Inversion for Pine Forest using Polarimetric Backscattering Models," Proc. PIERS '94 Symp. , Noordwijk, Netherlands, July 1994.

L. Tsang, G. Zhang, K. H. Ding, C. Hsu, and J. A. Kong, "Microwave Scattering by Vegetation Based on Wave Approach," Proc. PIERS '94 Symp. , Noordwijk, Netherlands, July 1994.

L. Tsang, K. H. Ding, G. Zhang, C. Hsu, and J. A. Kong, "Backscattering Scattering Enhancement of Volume-Surface Interaction of a Layer of Scatterers Overlying a Homogeneous Dielectric Half Space," AP Dig. Joint Symp. IEEE-APS/URSI, Seattle, WA, June 1994, pp. 2338-3341.

L. Tsang, Z. Chen, K. H. Ding, C. Hsu, and G. Zhang, "Collective Scattering Effects in Vegetation Canopies at Microwave Frequencies Based on Monte-Carlo Simulations," Proc. IGARRS '94 Symp. , Pasadena, CA, August 1994, pp. 548-550.

C. C. Hsu, H. C. Han, R. T. Shin, J. A. Kong, A. Beaudoin and T. Le Toan, "Radiative Transfer Theory for Polarimetric Remote Sensing of Pine Forest at P-band," Int. J. Remote Sensing , Vol. 15, pp. 2943-2954, 1994.

A. Beaudoin, T. Le Toan, S. Gozc, E. Nezry, A. Lopes, E. Mougin, Hsu, C.C., H.C. Han, J.A. Kong, and R.T. Shin, "Retrieval of Forest Biomass from SAR Data," Int. J. Remote Sensing , Vol. 15, pp. 2777-2794, 1994.

W. C. Au, L. Tsang, and J. A. Kong, "Absorption Enhancement of Scattering of Electromagnetic Waves by Dielectric Cylinder Clusters," Microwave Opt. Tech. Lett. , Vol. 7, 454-457, 1994.

L. Tsang, J. A. Kong, Z. Chen, K. Pak and C. Hsu, "Theory of Microwave Scattering from Vegetation on the Collective Scattering Effects of Discrete Scatterers," Proceedings of the ESA-NASA Workshop on Passive Microwave Remote Sensing of Land-Atmosphere Interaction, } VSP Press, Netherlands, 1994.

J. C. Souyris, T. Le Toan, C. C. Hsu, and J.A. Kong, "Assessment of SIR-C/X-SAR Polarimetric Data for the Estimation of Forest Parameters," Proc. Third International Workshop on Radar Polarimetry , Nantes, France, March 1995, pp. 636-645.

L. Tsang, K. H. Ding, G. Zhang, C. C. Hsu, and J.A. Kong, "Backscattering Scattering Enhancement and Clustering Effects of Randomly Distributed Dielectric Cylinders Overlying a Dielectric Half Space Based on Monte-Carlo Simulations," IEEE Trans. Antennas Propagat. , Vol. AP-43, pp. 488-499, 1995.

J. C. Souyris, T. Le Toan, J. A. Kong, and C. C. Hsu, "Inversion of Landes Forest Biomass Using SIR-C/X-SAR Data: Experiment and Theory," Proc. IGARRS '95 Symp. , Firenze, Italy, July 1995, pp. 1201-1203.

Wang, L., K. H. Ding, C. C. Hsu, Y. E. Yang, and J. A. Kong, "Electromagnetic Scattering Model for Vegetation Based on L-Systems," Proc. PIERS '95 Symp. , Seattle, WA, July 1995, pp. 278.

Souyris, J. C., T. Le Toan, Y. Zhang, C. C. Hsu, and J. A. Kong, "Inversion of Biomass with Polarimetric Data from SIR-C/X-SAR, " Proc. PIERS '95 Symp. , Seattle, WA, July 1995, pp. 902.
Hsu, C. C., J. A. Kong, J. C. Souyris and T. Le Toan, "Application of Radiative Transfer Modeling to the Polarimetric Backscattering of Forest," Proc. PIERS '95 Symp. , Seattle, WA, July 1995, pp. 903.

Hsu, C. C., L. Wang, J. A. Kong, J. C. Souyris and T. Le Toan, "Theoretical Modeling for Microwave Remote Sensing of Forest," accepted for the International Symposium on the Retrieval of Bio- and Geophysical Parameters from SAR Data for Land Applications,} Toulouse, France, October 1995.

Souyris, J. C., T. Le Toan, C. C. Hsu, and J. A. Kong, "Inversion of Landes Forest Biomass Using SIR-C/X-SAR Data: Experiment and Theory," accepted for the International Symposium on the Retrieval of Bio- and Geophysical Parameters from SAR Data for Land Applications, Toulouse, France, October 1995.

Le Toan, T., F. Ribbes, N. Floury, L. Wang, K. H. Ding, C. C. Hsu, and J. A. Kong, "On the Retrieval of Rice Crop Parameters from SAR Data," accepted for the International Symposium on the Retrieval of Bio- and Geophysical Parameters from SAR Data for Land Applications, Toulouse, France, October 1995.

Au, W. C., L. Tsang, R. T. Shin, and J. A. Kong, "Collective Scattering and Absorption Effects in Microwave Interaction with Vegetation Canopies," submitted to Progress in Electromagnetics Research, 1995.

Le Toan, T., J. C. Souyris, C. C. Hsu, and J. A. Kong, "Inversion of Landes Forest Biomass Using SIR-C/X-SAR Data: Experiment and Theory," to be submitted.
[Figures 1 and 2]

Table of Contents

Converted to HTML by Alvin Wong, al.wong@jpl.nasa.gov

The Jet Propulsion Laboratory
4800 Oak Grove Drive
Pasadena, Cailfornia 91109