Modeling Wetland Vegetation Using Polarimetric SAR
Center for Space Research, University of Texas at Austin
3925 W. Braker Ln., Suite 200, Austin, TX 78759-5321
Bureau of Economic Geology, University of Texas at Austin,
Campus Mail
Code: E0630, Austin, TX 78712
ABSTRACT
Airborne polarimetric Synthetic Aperture Radar (SAR) data are investigated for their potential in mapping herbaceous coastal wetlands. The subenvironments of coastal wetlands have very distinct vegetation cover and surface properties. Qualitative analysis of the SAR images reveals the relative importance of surface and vegetation scatter in these subenvironments. Furthermore, sampled SAR data distinctly separate the subenvironments, indicating that classification techniques could be used to discriminate among them. Although wetland environments are typically too vegetated to use empirical surface models to explain the SAR return, discrete scatterer models can be used to account for the scattering due to the vegetation. A discrete scatterer model fitted to a coastal wetland site on Bolivar Peninsula near Galveston, Texas provides insight into the dominant scattering mechanisms, and may aid in the accurate mapping of coastal wetlands.INTRODUCTION
Modeling the scatter of microwave radiation by natural surfaces and vegetation is important for assessing the ability of radar remote sensing to accurately map terrain and land cover. Coastal wetlands comprise a critical ecosystem for specialized vegetation and wildlife habitats, as well as for the natural production of methane. The Synthetic Aperture Radar (SAR) backscatter coefficient (sigma-0) is a complex function of local characteristics including topography, geological composition, soil moisture and salinity, and vegetation density and structure. Modeling the scatter from the vegetation is important for classifying land cover, monitoring change in dynamic environments, and discriminating among mechanisms of the backscattered return.The focus of this work is the analysis and modeling of the SAR return from coastal wetland vegetation. Both fully polarimetric SAR data in C, L, and P bands and fixed-baseline interferometric SAR (TOPSAR) data were acquired by the NASA/JPL AIRSAR system in April 1995. The data were acquired in support of a project to detect topographic change and relict geomorphic features on barrier islands for NASA's Topography and Surface Change Program. Imagery over a salt marsh is being used for a preliminary study of the effects of vegetation on the SAR return. Interest in SAR response to wetland environments has increased in recent years. Ormsby and Blanchard [1] studied the effect of inundation on sigma-0 . Pope et al. [2] developed wavelength and polarization dependent indices of sigma-0 which represent scattering mechanisms such as attenuation due to vegetation and depolarization due to vegetation multiple scattering. Such indices can be used as qualitative measures of the effect of vegetation cover on the total s o return.
Understanding the scatter due to vegetation also helps relate the sigma-0 return to surface properties such as soil moisture. Knowledge of the soil moisture distribution can then be used as an input for hydrological models. SAR's value in retrieving soil moisture estimates from barren and sparsely vegetated areas has been shown by Dubois et al., [3] and others. Typically, surface models that relate sigma-0 to soil moisture are empirically derived for specific data sets. These models have difficulty separating the return due to soil moisture from the scatter due to surface roughness and vegetation multiple scatter. If the scattering due to vegetation could be well characterized, its effects might be accounted for, allowing for the estimation of soil moisture over vegetated areas. Models that represent vegetation as a layer of discrete scattering elements have been developed to characterize the scatter due to vegetation. Most of this research has focused on forested areas, but some work has been done for herbaceous vegetation. Saatchi et al. [4] developed a model for grass canopies, and Durden et al. [5] modeled scatter from inundated rice fields. The model developed by Lang and Sidhu [6] was used to study the SAR return from a coastal wetland test site. Before such a model is fit to the data, the sigma-0 values are often plotted versus incidence angle. These plots can be used to fit the model to the data, and to determine how well SAR is able to separate different environments. In this study, a variety of methods were used to study the SAR response to coastal wetlands, including visual interpretation, empirical surface modeling, and discrete scatterer modeling. Some preliminary observations from each method are discussed.
SITE DESCRIPTION
The test site is located on Bolivar Peninsula, Texas, shown in Fig. 1, and consists of an herbaceous salt marsh, vegetated upland flats, and an intermediate transition zone, shown in Fig. 2. The SAR scene contains a typical transition from a salt marsh to vegetated upland flats. This transition involves four subenvironments: a low salt marsh with barren tidal flats that is flooded often; a low salt marsh with nearly continuous vegetation cover that is less frequently flooded; a transition zone with occasional seawaterflooding; and the vegetated upland flats. The entire peninsula is extremely low relief (<4 m), so even small changes in elevation can produce significant changes in the soil moisture and salinity.The peninsula is a sandy barrier spit formed during the last 4,000 years, primarily through spit accretion, and modified by washover and tidal inlet processes. Because the uplands are higher in elevation and contain sandy soils, they are well drained and non-saline. The tidal-flat low marsh contains muddy (silt + clay) soils with high concentrations of organic material. It is flooded almost daily with seawater. The continuous-cover low marsh soils are similar to those found in the tidal-flat low marsh, but contain less organic material. The transition zone corresponds to the mean high water mark. Because of evaporation between seawater inundations, this area is extremely saline. These variations in ground conditions give rise to variations in the vegetation cover. In the tidal-flat low marsh, tall grasses grow to heights of 2 m [7]. These grasses often occur in small groups intermixed with the barren tidal flats. The continuous-cover low marsh also contains tall grasses (< 1.5 m), but the plants are more consolidated creating a more uniform cover. Due to the high saline concentration in the transition zone, that area contains many barren salt flats, and supports mainly small succulent plants. The uplands have a drier, rougher surface, and support short (< 0.5 m) range grasses. The variations in vegetation , soil type, soil moisture, and soil salinity all involve variations in electrical and geometric properties to which SAR is sensitive. It was therefore expected that SAR would be useful for mapping herbaceous wetlands.
ANALYSIS
Initial analysis of the SAR images consisted of visual interpretation. C-band showed subtle variations in vegetation, but was so strongly scattered by all vegetation that the subenvironments were not well delineated. The longer wavelength L- and P-band separated the four subenvironments very well. Because of the extremely low relief, local slope changes were assumed to be unimportant in the s o return. Several sigma-0 indices were examined to qualitatively determine the relative importance of surface scatter and vegetation multiple scatter in the four subenvironments.The "canopy structure index" (csi) [2] was used as a measure of the vertical co-polarized return relative to the sum of the vertical and horizontal co-polarized returns. The csi reveals variations in the predominant orientation of vegetation structure. Because attenuation of sigma-0 vv can be much greater than attenuation of sigma-0 hh at P-band [8], the csi was also expected to reveal attenuation due to vegetation. As expected, the P-band csi image showed that most attenuation occurred in the dense vegetation of the continuous-cover low marsh. The tidal-flat low marsh produced less attenuation due to the lack of continuous vegetation cover. The transition zone showed the least attenuation. A "volume scattering index" (vsi) was adapted from [2] and used as a measure of the depolarization relative to the sum of the co-polarized and cross-polarized returns. At L-band, the uplands exhibited the most volume scattering, while the continuous-cover low marsh exhibited the most volume scattering at P-band. It was therefore concluded that P-band was achieving some penetration of the upland vegetation.
From the qualitative analysis of these indices, it appeared that there were scattering contributions from both the surface and vegetation in the uplands, the return from the transition zone was almost entirely due to surface scatter, the continuous-cover low marsh returns were mostly due to strong vegetation scatter, and the weaker returns from the tidal-flat low marsh were due to specular reflection off of the water surface. In an effort to retrieve soil moisture estimates, empirical surface models were studied. Before any of these models could be applied, it was necessary to assess how the presence of the vegetation cover would affect the soil moisture estimates. The criterion developed in [3] was used to determine whether the vegetation cover over the test site was thick enough to reduce the accuracy of such models. That criterion consisted of the ratio sigma-0 hv / sigma-0 vv at L-band. If this ratio is greater than -11 dB, the model is not recommended for estimating soil moisture. This criterion was exceeded over most of the Bolivar SAR image. Empirical surface models were therefore abandoned in favor of discrete scattering models. In addition to the techniques mentioned above, sigma-0 data were sampled from the four subenvironments and plotted versus radar incidence angle. Fig. 3 shows one such plot for L-band. These plots clearly demonstrated polarimetric SAR's ability to separate the subenvironments and its potential for herbaceous wetland mapping. Strong sigma-0 hh , due mostly to vegetation multiple scatter at L-band, occurs in the uplands. A marked decrease in sigma-0 hh over the transition zone is due to that region's lack of vegetation and extremely smooth surface. The continuous-cover low marsh possesses strong sigma-0 hh due to vegetation scatter. The decrease in sigma-0 hh over the tidal-flat low marsh is due to specular reflection off of the water surface and is evidence of inundation.
Using all available bands and polarizations should provide ample discriminators for the classification of these subenvironments. The discrete scatterer model developed by Lang and Sidhu [6] was fitted to these data to gain insight into the scattering mechanisms in each subenvironment. The model represents the vegetation cover as a layer of discrete scattering elements over a flat half space. The lower half space was assumed to be saline water for the inundated tidal-flat low marsh [5], and soil for the other subenvironments. A soil dielectric mixing model adapted from [9] was used to compute the soil dielectric constant. The total sigma-0 is computed as the sum of the sigma-0 due to (i) direct scatter from the vegetation, (ii) scatter from a single-reflection ground/vegetation interaction, and (iii) scatter from a double-reflection ground/vegetation interaction. Preliminary attempts to fit the scattering model to the data have been successful, but are difficult to interpret because there are insufficient ground data to properly constrain the model. Additional ground truth will be collected so that the model may be quantitatively applied to the data.
SUMMARY AND FUTURE WORK
It is clear that polarimetric multiband SAR has potential for mapping the major subenvironments associated with coastal herbaceous wetlands. A discrete scatterer model can be fitted to the data to gain insight into the scattering mechanisms that occur. Additional ground data, such as soil and plant dielectric constants, will be collected to constrain the scattering model. Also, the effects of direct surface scattering will be included in the model to account for sparsely vegetated areas.
REFERENCES
[1] Ormsby, J. P., and B. J. Blanchard, "Detection of Lowland Flooding Using Active Microwave Systems," Photogrammetric Engr. and Remote Sensing, vol. 51, no. 3, pp. 317-328, 1985.[2] Pope, K. O., J. M. Reybenayas, and J. F. Paris, "Radar Remote Sensing of Forest and Wetland Ecosystems in the Central American Tropics," Remote Sensing Environ., vol. 48, pp. 205-219, 1994.
[3] Dubois, P. C., J. J. van Zyl, and T. Engman, "Measuring Soil Moisture with Imaging Radars," IEEE Trans. Geosci. Remote Sensing, vol. 33, no. 4, pp. 915-926, 1995.
[4] Saatchi, S. S., D. M. Le Vine, R. H. Lang, "Microwave Backscattering and Emission Model for Grass Canopies," IEEE Trans. Geosci. Remote Sensing, vol. 32, no. 1, pp. 177-186, 1994.
[5] Durden, S. L., L. A. Morrissey, and G. P. Livingston, "Microwave Backscatter and Attenuation Dependence on Leaf Area Index for Flooded Rice Fields," IEEE Trans. Geosci. Remote Sensing, vol. 33, no. 3, pp. 807-810, 1995.
[6] Lang, R. H. and J. S. Sidhu, "Electromagnetic Backscattering From a Layer of Vegetation: A Discrete Approach," IEEE Trans. Geosci. Remote Sensing, vol. GE-21, no. 1, pp. 62-71, 1983.
[7] Webb, J. W., Typical Dune Vegetation of the Upper Texas Coast, non-published Texas A&M at Galveston document, pp. iii, 3.
[8] Le Vine, D. M., and M. A. Karam, "Dependence of Attenuation in a Vegetation Canopy on Frequency and Plant Water Content," in Proc. IGARSS'95, vol. I, pp. 607-609, 1995.
[9] Ulaby, F. T., A. K. Moore, A. K. Fung, Microwave Remote Sensing, Active and Passive, vol. 3, Artech House, pp. 2086-2103, 1986.
Last Modified: Tue Sept 14, 1999
CSR/TSGC Team Web