«int. j. remote sensing, 1997, vol. 18, no. 6, 1319± 1332 Decomposition of polarimetric synthetic aperture radar backscatter from upland and ¯ ooded ...»
Decomposed SAR scattering components in forests
Figure 7. Colour composite of scattering powers decomposed from multi-frequenc y AIRSAR data, Altamaha River, GA (28 March 1990 ).
SP-O ( Red ), SP-E (Green), and SP-C ( Blue). (a) C-band (top), (b) L -band (middle), and (c) P-band ( bottom).
1330 Y. Wang and F. W. Davis The result is consistent with the ® ndings in ¯ ooded forest by other researchers (e.g., Richards et al. 1987, Hess et al. 1995 ). Visually, there may not be much di erence between the decomposed scattering power images at L -band and P-band (® gure 7 (b,c)). However, one, examining carefully, may argue that the green in the ¯ ooded forests are slightly purer (or with less mixtures of red and blue) at P-band than at L -band. The purity of the green colour at P-band further shows that there is more double-bounce trunk-ground scattering than other types of scattering such as the direct canopy volume scattering. it should also be noted that the noise a ects the decomposition, especially at P-band. The scattering power decomposed from some portions of the river channels (in middle on the left, ® gure 7 (c)) is SP-E (green).
5. Conclusions and remarks We have used multi-frequency JPL AIRSAR data of Goldstone (CA) to verify a radar target decomposition method. By using this method, we decomposed the AIRSAR backscatter data into the scattering power with an odd number of re¯ ections (SP-O), the scattering power with an even number of re¯ ections (SP-E), and the total cross-polarized scattering (SP-C). Dihedral corner re¯ ectors oriented 0ß ( horizontally) on a ¯ at ground, and 45ß o the ground were used in the veri® cation of this method.
We then applied the method to decompose the multi-frequency AIRSAR backscatter data from two types of forests to help understand scattering mechanism in forested environment, and to evaluate potential application of the decomposition method in forests. For ponderosa pine forest (Mt. Shasta, CA), as the wavelength increased from C-band to P-band, the SP-O return decreased, the SP-E return increased, and there was likely no obvious trend for the cross-polarized scattering.
For ¯ oodplain forest (Altamaha River, GA), the SP-O return dominated the decomposed image at C-band. The SP-E return from the ¯ ooded forests was strong at L -band, and stronger at P-band. This increase of the SP-E return could be explained as a result of the increase of the trunk-ground interactions at long wavelengths. The cross-polarized scattering from ¯ ooded forest was not strong at C-, L -, and P-bands, and its strength became less and less as the wavelength increased from C-band to P-band.
In the imagery derived by the decomposition method, various forest structural classes and land cover types were easily distinguishable. More importantly, the decomposition method provided a means of unraveling complex interactions between radar signals and vegetated surfaces in terms of the scattering mechanisms from targets, but not in terms of the HH and V V polarizations from a radar system con® guration. As shown in this paper, this decomposition method helped understand the scattering mechanisms of polarimetric SAR data.
* * * Sum of k 1hh k1 hh and k 2hh k2 hh was the total HH backscatter, and k1v v k1 v v plus * k 2v v k2 v v was the total V V backscatter (equations ( 1) and ( 6 )). Thus, in addition to traditional HH and V V backscatter from a polarimetric SAR system, one more * * independent backscattering component (k 1hh k1 hh or k 2hh k2 hh ) for HH, and one more * * component (k1v v k1 v v or k 2v v k2 v v) for V V have been derived. These two components might provide additional potential in applications of polarimetric SAR backscatter data.
By studying the decomposed scattering powers from ponderosa pine forest, we have noted that at P-band, as the stand density changed from sparse to dense, the increase of the SP-E return was about 9 dB, but the increase of the total power was Decomposed SAR scattering components in forests around 5 dB. This enlarged dynamic range of the SAR data could provide a better potential in applications (e.g., the retrieving of the biophysical parameters of forests) because the decomposed data may be more sensitive to the change of the parameters.
This is a task we will pursue in the near future.
In the analysis, we also noticed that the noise of SAR data could a ect the decomposition. There may be falsely decomposed scattering components with an even number of re¯ ections because of the noise e ect, whereas the correct ones may be the scattering power with an odd number of re¯ ections. This false decomposition was likely to occur when the backscatter from targets (e.g., river channels) was small or close to the noise level. Because most natural targets with low backscatter were ¯ at and smooth (to slightly rough) surfaces (such as open ® elds, and water and river surfaces), scattering from these surfaces should be the scattering power with an odd number of re¯ ections. To prevent or eliminate this false decomposition, one method used in our decomposition was to set a threshold for the noise level. If the total power of targets within an image pixel was less than the threshold, the larger value of decomposed scattering powers (SP-O and SP-E) would be assigned to the SP-O return, and the smaller one to the SP-E return. We tested thresholds ranging from Õ 50 dB to Õ 25 dB. For the Altamaha SAR data, as the thresholds increased from Õ 50 dB to Õ 25 dB, the SP-E ( green) from the river channels became less and less, and ® nally disappeared. The threshold used for ® gure 7 was Õ 40 dB; we left the scattered green areas within the river channels for showing the noise e ect on the decomposition. It should be noted that, normally, one might not know what was th noise level of SAR data or what threshold value should be used. Several iterations by using di erent threshold values might, empirically, lead to a satisfactory result.
Acknowledgments We thank Annie Richardson at the Radar Data Center ( JPL) for providing AIRSAR data of the Goldstone area (CA), and thank Laura Hess at the University of California, Santa Barbara ( UCSB) for o ering AIRSAR data of the Altamaha River ¯ oodplain (GA). The AIRSAR data of Mt. Shasta (CA) was obtained from the UCSB SIR-C/X-SAR project funded by NASA through JPL (contract # 958468), and this research was also funded by the SIR-C/X-SAR project.
Referen ces Cloude, S. R., 1992, Uniqueness of target decomposition theorems in radar polarimetry. In Direct and Inverse Methods in Radar Polarimetry, Part I, edited by W. M. Boerner et al. (Dordrecht: Kluwer Academic Publishers), pp. 267± 296.
Freeman, A., Shen, Y., and Werner, C. L., 1990, Polarimetric SAR calibration experiment using active radar calibrators. I.E.E.E. T ransactions on Geoscience and Remote Sensing, 28, 224± 240.
Hess, L. L., and Melack, J. M., 1994, Mapping wetland hydrology and vegetation with synthetic aperture radar. International Journal of Ecology Environment Science, 20, 74± 81.
Hess, L. L., Melack, J. M., Filoso, S., and Wang, Y., 1995, Realtime mapping of inundation on the Amazon ¯ oodplain with the SIR-C/X-SAR synthetic aperture radar. I.E.E.E.
T ransactions on Geoscience and Remote Sensing, 33, 896± 904.
Richards, J. A., Woodgate, P. W., and Skidmore, A. K., 1987, An explanation of enhanced radar backscattering from forest. International Journal of Remote Sensing, 8, 1093± 1100.
Sun, G., Simonett, D. S., and Strahler, A. H., 1991, A radar backscattering model for discontinuous coniferous forests. I.E.E.E. T ransactions on Geoscience and Remote Sensing, 29, 639± 650.
1332 Decomposed SAR scattering components in forests van Zyl, J. J., 1994, Application of Cloude’s target decomposition theorem to polarimetric imaging radar data. Proceedings of the T hird Spaceborne Imaging Radar Symposium, held at the Jet Propulsion L aboratory, Pasadena, California, 18± 21 January 1994, pp. 207± 216.
Wang, Y., Davis, F. W., and Melack, J. M., 1993, Simulated and observed backscatter at P-, L -, and C-bands from ponderosa pine stands. I.E.E.E. T ransactions on Geoscience