Opportunities for using the EOS Imaging Spectrometers and Synthetic Aperture Radar in Ecological Models

Susan L. Ustin1, Carol A. Wessman2, Brian Curtiss2, Eric Kasischke3,
JoBea Way4, and Vern C. Vanderbilt5
1Department of Land, Air, and Water Resources
University of California, Davis, California 95616 USA
 
2CSES/CIRES
University of Colorado, Boulder, Colorado 80309-0449 USA
 
3Radar Science Laboratory
Environmental Research Institution of Michigan
P.O. Box 8618,  Ann Arbor, Michigan 48107 USA
 
4Jet Propulsion Laboratory
Mail Stop 300233, Pasadena, California 91109 USA
 
5NASA Ames Research Center
Mail Stop 2424, Moffett Field, California 94035 USA
  
 
Author for Correspondence:
Susan L. Ustin
Department of Land, Air, and Water Resources
University of California
Davis, CA 95616
Phone:  (530) 752-0621
FAX:  (530) 752-5262
email:  slustin@ucdavis.edu

Abstract

In order to predict ecological responses to environmental change we must understand, measure, and model the functioning of biotic, atmospheric, and hydrospheric processes and their interactions at many temporal and spatial scales. Until recently it has been impossible to define the role of ecological processes at regional and global scales or to determine whether processes have been altered. We think that new approaches in the use of remotely sensed measurements, especially as inputs to ecosystem models, hold a key to this problem. Starting in 1997, the Earth Observing System (EOS) will offer the first opportunity to obtain frequent (daily to 1 6d repeat) remote observations in microwave, optical, and thermal bands over most of the Earth's surface. These data can be used to infer biophysical and geophysical attributes of the surface of the Earth using algorithms that convert sensor units (e.g., reflectance, phase, and backscatter) into physical units (e.g., absorbed photosynthetically active radiation (APAR), evapotranspiration, soil moisture, biomass) that are needed to drive ecosystem and climate models.

While it is impossible to forecast exactly how the information from the EOS spaceborne sensors will be used, prototype airborne systems have provided considerable insight into the capabilities of the EOS systems. The High Resolution Imaging Spectrometer (HIRIS), the MODerate Resolution Imaging Spectrometer (MODIS), and the Synthetic Aperture Radar (SAR) represent innovative technologies that will provide new types of spectral and spatial information for studies of physiological processes and structural organization of ecosystems (Figs. 1-3). The specific capabilities of these sensors are discussed more fully in Wickland (1991, this Special Feature). The high spatial resolution and spectroscopic capability of HIRIS will complement the more frequent passes and global mapping capability of the lower resolution MODIS. Similarly, SAR may be operated in either a multifrequency, polarimetric high resolution mode or a complementary global mapping mode that provides more frequent lower resolution microwave coverage. All three sensors will view the earth from multiple angles in addition to directly down. MODIS and HIRIS will be able to estimate the bidirectional reflectance distribution function (BRDF), a key property describing the radiative transfer at the land surface. Both HIRIS and SAR can be programmed to provide data at varying resolutions and scales of measurement, thus functioning at an important interface between the experimental scale of field measurements and the global scale of the MODIS. Moreover, these sensors offer promise for developing spaceborne assessments of biogeochemical processes.

To utilize this new information effectively, ecologists need to understand the capabilities and limitations of the EOS sensors and develop models that use the remote sensing data more directly. In this paper we present some ways in which ecological and remote sensing models can utilize the new remote sensing information to characterize ecological properties at coarse scales as well as to estimate within ecosystem properties.
 

Development of ecological models that use remote sensing inputs

Remote sensing images display landscape characteristics at a range of spatial scales, but most ecological models function at only one scale (Woodcock and Strahler 1987, O'Neill et al. 1989, Turner 1989). The scales of ecological models need to be compatible with scales of remote sensing data acquisition. In the cases of HIRIS and SAR these scales are constrained by pixel size (smallest spatial unit resolved) to tens of metres and in the case of MODIS to kilometres. Pixel sizes can always be enlarged by averaging but cannot shrink to accommodate ecosystem models, thus limiting the types of ecological models that can be applied.

As noted earlier, the EOS sensors will provide new combinations of data at greater spectral and spatial resolution than currently available. In order to use this data to its fullest potential, a number of different types of models need to be developed and employed. These include biophysical models, which attempt to interpret the remote sensing data in the context of ecological and physical variables, and ecological models, which will then use those variables to drive simulations of ecological and physical processes. Some of the ecological variables (Table 1 ) that are expected to be derived from EOS data include estimates of land classification, biome extent and distribution, stand structure and density, biomass, and physiological attributes like APAR and water content. Many of these variables are now estimated from current satellites but EOS should improve accuracy and resolution.

For many of the biophysical models, existing approaches utilize only a fraction of the spectral and spatial information that will be available from the new EOS sensors. For example, a widely used but simple model, the normalized difference vegetation index (NDVI), uses only 2 reflectance bands (out of 197 available for HIRIS and 5065 for MODIS) to identify green vegetation and estimate biomass accumulation (Tucker 1977, 1979). Much has been learned using NDVI, which is based on reflectance differences between red and near-infrared bands. It is relatively well correlated with light interception (Asrar et al. 1984), and its annual integral is correlated with annual light interception at the leaf scale (Sellers 1985, 1987a). This canopy property has important ecological consequences, since absorbed photosynthetically active radiation (APAR) has been found to be a good indicator of biomass accumulation and net primary productivity (Monteith 1981, Waring 1983, Linder 1986). Spatial patterns of NDVI appear reasonable representations of vegetative growth at continental scales (Goward et al. 1985, Running et al. 1989), and global scales (e.g., Fung et al. 1987). However, NDVI predictions are not generally reliable at the pixel scale due to local environmental and atmospheric variability (Huete 1986, Huete and Jackson 1987, Smith et al. 1990a). A way to improve reliability of estimates of these and other canopy properties will be to utilize information effectively from all regions of the electromagnetic spectrum measured by the EOS sensors. Plant canopies have constituents with absorptions in visible, near and shortwave infrared, and radar wavelengths, and emissions in the thermal infrared. Assessing this new information should improve and expand our ability to quantify the biophysical and biochemical properties of ecosystems. In addition to much greater spectral range, EOS will provide data at a wider range of spatial scales. In order to make full use of these data, we believe new approaches in image processing and ecological modeling are required to extract detailed physical parameters. One such approach involves developing a linked hierarchical series of remote sensing models (Kasischke et al. 1990, Ustin et al. 1990, 1991, Adams et al. 1991). First, image variation is stratified by modeling the large spatial scale properties of landscapes and geomorphology, such as topography, land cover, biome boundaries (e.g., forests and grasslands), and geological and/or soil units. Interpretations would be based on spectral classification methods as well as measures of the structure, organization, and roughness obtained from one or more EOS sensors using variables like those listed in Table 1. Second, the smaller sources of image variation, such as those associated with canopy processes, e.g., photosynthesis or transpiration, and states, e.g., biochemical characteristics, are analyzed using specific submodels. Analyses derived from images obtained at different times or from different EOS instruments may be combined in tertiary image products using an approach analogous to overlaying data in a Geographic Information System (GIS). These images also might be linked with more traditional ecological or biometeorological measurements to derive an integrated analysis of the landscape. Finally, the combined parameters could be used to drive various types of ecosystem process and/or dynamics models, such as carbon, water, and nutrient cycling (e.g., Pastor and Post 1986, Parton et al. 1987), or successional models (e.g., Shugart 1984, Wight and Skiles 1987).
 

Characterization of ecological properties at landscape scales

Ecologists have become increasingly concerned about the long-term consequences of massive changes in land use and the fragmentation and loss of critical habitats. MODIS will provide the frequently repeated routine data necessary to monitor changes from regional to global spatial scales at seasonal to decadal temporal scales, substantially expanding the capabilities of present satellites (e.g., Sellers 1987b). SAR and HIRIS data will make it possible to address issues related to the detailed structural integrity of ecosystems and successional dynamics. Because crown shape, size, and spacing create spatial and spectral variance in remotely sensed data, it should be possible to detect these properties. A number of approaches have been used to quantify canopy and landscape geometry using optical (Li and Strahler 1985, 1986, 1988, Smith et al. 1990a, b, Wessman et al. 1990) and microwave models (Richards et al. 1987, Sun and Simonett 1988, Dobson et al. 1990, McDonald et al. 1990, Ulaby et al. 1991). These models assess the large-scale geometry of the landscape (topography, texture, community distributions) and the fine scale geometry of the canopy (height and width of tree crowns, frequency and spacing of gaps).

Sensors capable of detecting changes in the structure, distribution, and biochemistry of the landscape could fundamentally alter our perception of ecological processes. One instrument with the potential to make such measurements is the Airborne Visible Infrared Imaging Spectrometer (AVIRIS), a prototype HIRIS sensor producing a complete spectrum over the visible and reflected infrared region for each pixel. Fig. 1 illustrates the dimensionality of a typical data cube from AVIRIS. Conceptually, the face of the cube depicts the spatial aspect of the landscape with an added third dimension of spectral information. The cube is equivalent to a stack of 224 images of the surface, each representing a single 10 nm wide spectral band. The face of the cube in Fig. 1 shows a forested scene created from a three-band false color composite. The sides of the cube depict a spectrum using colors to indicate the intensity of each wavelength for each of the edge pixels. Areas of forest, shrubs, and dry meadows differ in their spectral characteristics as seen by the patterns on the face and sides of the cube. The face of the cube is predominantly red because of the very high reflectance of vegetation in the near-infrared. Vegetation has low reflectance in the red region of the visible spectrum (shown in the image as green) and slightly higher reflectance in the green region of the visible spectrum (shown in the image as blue). Numerical analysis of the digital data could include identification of absorption features in specific regions of the spectrum, utilization of the entire 224 band continuous spectrum (Fig. 2), or combination of image data from various sources. Fig. 2 shows a canopy spectrum extracted from the image to illustrate the resolution possible with sensors of this type.

Radar also depicts canopy structure. By using both the polarization phase and wavelength information of the SAR, considerable detail can be derived concerning the vertical structure of tree and shrub canopies and the structure and organization of the ecosystem on a horizontal scale (Durden et al. 1989). In general, the longer the radar wavelength, the farther the radar waves penetrate into the canopy, while the sizes and orientations of the leaves and stems strongly affect the amount of backscatter in different wavelengths and polarizations. Distinct polarimetric signatures of vegetation canopies have been demonstrated in a few cases (Ulaby et al. 1987, 0tt et al. 1990, Sheenetal. l990); however, these examples have been limited to canopies with a distinct row structure (e.g., crops or orchards).

Fig. 3 presents a false color SAR image of a forested area, generated from three radar frequencies at a single polarization. Different successional stages of loblolly pine and mixed pine hardwood forest can be distinguished. This figure illustrates that the SAR imagery can be used to classify land units in much the same way as current Landsat imagery. However, because SAR is more sensitive to canopy or surface water content, while optical sensors are more sensitive to green leaf biomass, combined analyses can characterize vegetation more completely. For example, recent studies have shown that longer wavelength radars respond to variation in aboveground biomass, especially in early stages of forest succession (Kasischke et al. 1990). We expect that this type of data will be used in quantifying terrestrial carbon storage and in identifying patterns of land use change. Fig. 4 shows the effect of changes in forest biomass on radar backscatter. Shorter wavelengths provide better assessment of aboveground biomass in low stature ecosystems, such as wetlands and grasslands.
 

Characterization of ecological properties at the ecosystem or canopy scale

While the EOS sensors should improve our ability to characterize the diversity of ecosystems on the landscape, they should also contribute to our ability to understand within ecosystem characteristics and processes. As noted earlier, the structure of leaves and canopies affects spectral reflectance, but so do biochemical constituents and water. Because the general shape of reflectance curves for green leaves is similar for all species (Fig. 5), it is the relative proportions of the spectral features that define functionally significant differences. Some of these features may be useful for diagnosing the physiological state of the ecosystem (Wessman 1990). For example, the visible spectral region is dominated by the presence of photosynthetic pigments. Wavelength specific absorption differences among chlorophylls, xanthophylls, and carotenes may permit quantification of their concentrations, and these may be related to photosynthetic activity and primary productivity. In one case, a light induced reversible change in a xanthophyll pigment that is closely linked to changes in photosynthetic activity was monitored by changes in green reflectance (Gamon et al. 1990). It should be possible to extend models of nutrient cycling, gas exchange, and energy flow processes to landscapes and regions using information about the biochemical and biophysical states of the canopy (Fung et al. 1987, Reinhert et al. 1989, Matson and Vitousek 1990; J. Way et al., personal communication).

Information in the infrared areas of the spectrum may allow direct assessment of canopy biochemical properties, potentially leading to improved modeling of spatial patterns in biogeochemical processes. For example, changes in the proportion of lignin and nitrogen are related to microbial decomposition rates and to carbon allocation. Minor spectral features associated with lignin, cellulose, starch, and proteins occur in the shortwave infrared (shown in Fig. 5; Weyer 1985). The recent development of routine laboratory spectroscopic assays for identifying these and other organic constituents in foliar material (Marten et al. 1989) raises the possibility that some biochemicals will be identified from high-resolution spaceborne sensor data. The cellulose and lignin features that spectrally characterize wood and letter may provide a measure of the non-photosynthetic biomass in the canopy, and, through seasonal monitoring of proportions of foliar and non-photosynthetic biomass, information about phenological conditions.

It has long been known that infrared reflectance changes with the water content of leaves (e.g., Thomas et al. 1971, Tucker 1980; C. E. Olson, Jr., personal communication). Since then, some investigators have suggested that canopy moisture content may be detected using reflectance at one or more of the water absorption features in the infrared region (Tucker 1980, Ripple 1986, Hunt et al. 1987). Other investigators have found that reflectance changes within a biologically meaningful range were too insignificant to be remotely detectable (Hunt and Rock 1989, Bowman 1990). However, Goetz et al. (1990) and Gao and Goetz (1989, 1990) reported a new approach using a spectral curve-fitting procedure applicable to a contiguous spectrum (e.g., from a AVIRIS- or HIRIS-type sensor) which offers renewed promise for detecting canopy water (Fig. 6a-c). The residual spectrum (difference between the leaf and water spectra) shows lignin and cellulose features that correspond with constituent absorptions seen in dry leaves. Such features have previously been detectable only after the leaf had dried. This type of spectral matching could potentially be used to map the distribution of major biochemical constituents of canopies based on imaging spectrometer data.

An alternative approach to measuring the water status of canopies is to measure the thermal infrared energy emitted by plant canopies and the land surface. The water status of plant communities may potentially be monitored using the thermal infrared channels measured by MODIS and ASTER (a multichannel thermal EOS sensor). A number of agricultural studies have used apparent surface temperatures, measured by satellite, to estimate latent heat flux and energy balance (Jackson 1985, Kustas et al. 1989, Price 1989).

Radar backscatter is also sensitive to the water content of canopies, due to the high dielectric constant of liquid water and the large range in water and the large range in water contents of soils and plants. Similarly, the state of water in a canopy is important. The dielectric constant is 80 for water, <5 for dry leaves, and <3 for frozen plant material. The effects of changes between liquid and solid forms are illustrated in Fig. 7, from images collected when air temperatures were ~2oC and ~-15oC. Dielectric constants measured on trees in the forest were dramatically different during these two days (Dobson it al, 1990, Way et al. 1990). Because the magnitude of differences also depended on the tree species and stand density, detection of such changes would probably concentrate on temporal analyses. Radar can detect diurnal changes in canopies that are well correlated with diurnal water potential and with dielectric constants (McDonald et al. 1990, Way et al. 1991). These observation demonstrate that some aspects of water transport (movement of water through the xylem and the associated resistance and capacitance in stems and leaves) are observable in the radar signature and imply that short-term climatic events and even diurnal physiological changes in canopy water status could be observed with the SAR. This application of SAR is important for long-term EOS global monitoring, since environmental and phenological states of forests could be monitored at all latitudes and under all cloud conditions.

Variations in the soil/litter moisture content can be measured with SAR imagery. Longer wavelength SARs allow direct sensing of soil moisture in areas dominated by crops and nonwoody vegetation (Ulaby et al. 1987). Even in communities dominated by woody plants, recent studies have shown strong correlations between radar intensity and soil or canopy moisture contents (McDonald et al. 1990). By changing the phase angle of the backscattered signature (incidence and measurement angles of the SAR), trunk moisture (which controls the phase signature) may be separated from soil moisture (which has no effect on this component of backscatter). The presence of standing water under a deciduous canopy acts like a mirror and causes increased radar image intensity at all wavelengths when leaves are absent, but only at longer wavelengths when leaves are present, due to attenuation (Fig. 8). Measuring the seasonal changes in intensity could provide estimates of LAI. Because of the SAR's sensitivity to canopy geometry, much research has focused on utilizing differences between vertical and horizontal polarizations combined with the wavelength and SAR viewing angle information to estimate the leaf and stem angular distributions (LAD) and vertical stratification of canopy constituents. It appears that both the SAR and HIRIS may provide independent estimates of canopy architecture and LAI based on interactions between plant canopies and energy at several wavelengths.

These three EOS instruments: MODIS, HIRIS, and SAR, will make repeated measurements of the Earth's surface from different view directions. The HIRIS and MODIS measurements will permit estimates of the bidirectional reflectance distribution function (BRDF), a measure of the hemispherical albedo of the surface, from which landscape and global scales of this property can be estimated (Kimes 1991). The spatial distribution of the surface albedo and its changes over time are critical measures necessary for improved global climate modeling and for better understanding of the structure of the biosphere (Sellers 1987b). Current satellite sensors provide insufficient data for accurate spatial estimation of albedo at the global scale.

CONCLUSIONS

We have outlined several promising approaches for assessing biochemical and architectural properties of landscapes and strategies for using the new EOS sensors in ecological models. SAR and HIRIS will measure components related to the structure and biomass of communities that will improve global ecosystem classifications and follow the distribution of photosynthetic and nonphotosynthetic biomass over short intervals of time and successional processes over longer periods. SAR and MODIS will provide near daily global estimates of green vegetation, ecosystem distribution, and changes over time. The data from these EOS sensors should provide an opportunity to improve understanding of environmental processes at unprecedented scales and provide the tools with which to address pressing global and regional environmental concerns, such as the effects of greenhouse gases and pollutants on terrestrial vegetation. However, connecting the spatial, spectral, and temporal components of the EOS data to environmental variables that are useful to ecologists still presents a formidable task.

We are at an exciting juncture in ecological research due to the simultaneous emergence of several new technologies. High-powered microcomputer and workstation capabilities are now available at modest cost for image processing, new mathematical and statistical techniques for complex data and spatial analysis are in development, and new environmental models extrapolating processes to large regions using both top down (primarily climatological) and bottom up (primarily physiological and population dynamics) approaches are being developed. The convergence of technologies and models are promising a new era in remote sensing. How well we use these tools to address many of the important ecological issues of the next decades will be determined less by the capabilities of the EOS sensors than by our intelligence in developing, testing, and validating ecological models that effectively utilize the information. For the potential of the EOS to be realized, there is a strong need for active involvement of a broad segment of the ecological community.
 

ACKNOWLEDGMENTS

We thank P. A. Matson, J. A. Doyle, and two anonymous reviewers for comments on the earlier drafts of this manuscript and to NASA Land Processes and EOS programs for research support.

REFERENCES

1998, Center for Spatial Technologies and Remote Sensing (CSTARS)
University of California, Davis