Table of contents
Previous Section
Next Section
Imaging Radar Home Page
The past five years have seen significant growth in research focused on developing approaches for using synthetic aperture radar (SAR) to study ecological processes. During this time, we have seen the development of a number of advanced airborne SAR systems, and the deployment of three spaceborne SAR systems: ERS-1, JERS- 1, and SIR-C/X-SAR. Additional spaceborne SARs will be deployed through the remainder of this decade and into the next, including ERS-2, RADARSAT, and ASAR (see Appendix B).
The deployment of the multi-frequency, polarimetric SIR-C/X-SAR instruments on two Space Shuttle missions in April and October 1994 represents a milestone in the history of imaging radars. The data collected during these missions provide an important opportunity to further evaluate the utility of imaging radar data for examining surface characteristics important in a wide range of ecological processes.
Ecologists are generally aware of the utility of remote sensing data for studying processes at landscape scales. However, imaging radar data, such as that collected by synthetic aperture radars, have received less attention than optical data. Hindrances to use of SAR systems have included: (1) difficulty of understanding the information content of the complex phase and amplitude information recorded in multifrequency, polarimetric SAR data; (2) the lack of available, calibrated data over sites of interest; (3) the lack of accessible computer software to exploit the information present in the data; and (4) certain characteristics of SAR data, including relief displacement and image speckle. Computer software and hardware have developed to the point where most of the technological constraints of using SAR data have dissipated. In addition, a great deal of calibrated SAR imagery exists and is being collected by several airborne and satellite systems. This chapter focuses on our current understanding of the information content of SAR imagery with respect to ecological applications, based on recent research results.
In November 1994, a working group of scientists met on the campus of the University of California (Santa Barbara) to perform a critical review of the use of imaging radars to estimate surface characteristics important for the study of terrestrial ecosystem processes. The consensus of this group was that the ability of imaging radars to detect ecologically important characteristics of vegetated landscapes is well founded in both theory and observation. The working group felt the demonstrated capabilities of imaging radars for investigating terrestrial ecosystems could best be organized into four broad categories: (1) classification and detection of change in land cover; (2) estimation of woody plant biomass; (3) monitoring the extent and timing of inundation; and (4) monitoring other temporally dynamic processes, such as freeze/thaw status and soil moisture in fire-disturbed boreal forests.
This review is organized in the following manner: First, we present a brief review of the origins of the signatures recorded in a SAR image collected over a vegetated terrain. Then, we discuss each of the four topical areas, including an overview of the scientific importance or application of the topic area and a review of the demonstrated capabilities of SAR to provide information necessary to provide specific inputs. This is followed by a discussion of the optimal system configurations for specific ecological applications and an assessment of the capabilities of existing SAR systems (including SIR-C/X-SAR) and those scheduled for launch in the near future to provide information required for the application areas. Finally, we present recommendations for future SAR programs within the Mission to Planet Earth (MTPE) and the U.S. Global Change Research Program.
Radar Sensing of Vegetated Landscapes: Physical Basis
Microwave backscatter is highly dependent on the orientation and size distribution of the scattering elements present within the region being imaged. Because of their high moisture content, individual components of forest canopies and other vegetative covers (e.g., leaves, branches, trunks) represent discrete scattering and absorbing elements to the microwave power transmitted by imaging radars. Variations in the microwave dielectric constant of vegetation elements or ground surface play a central role in determining the magnitude and phase of the microwave energy which is scattered from a vegetated surface and recorded and processed into a SAR image. Factors influencing the dielectric constant of vegetated surfaces include temperature of the scattering medium, relative moisture content of vegetation, soil, and snow cover, and the presence of water on vegetation.
Microwave scattering from land surfaces is strongly dependent on the size and orientation of the different elements comprising the vegetation. At longer radar wavelengths (P- and L-bands, 67 and 24 cm wavelengths), microwave scattering and absorption results from interactions with the tree boles and larger branches found within forests, as well as the ground surface. At these wavelengths, the smaller woody stems and the foliage act mainly as attenuators. At shorter radar wavelengths, (C- and X-bands, 6 and 3 cm wavelengths), microwave scattering and absorption results from interactions from smaller branches and leaves and needles in the canopy. The presence of a water-saturated or flooded surface leads to increased double-bounce scattering that enhances the strength of the ground-vegetation interaction term. Finally, the polarization combination of the received backscatter is dependent on the polarization of the transmitted microwave power and on the horizontal and/or vertical orientation of the scattering elements present in the vegetation.
Modeling clearly shows the differential dependence of microwave backscatter on the overall structure of vegetation canopies and on the variations in the characteristics of the ground layer. These models treat a forest stand either as a set of continuous horizontal layers (Richards et al., 1987; Durden et al., 1989; Sun and Simonett, 1990; Ulaby et al., 1990; Chauhan et al., 1991) or as a discontinuous layer with individual trees acting as distinct scattering centers (Sun et al., 1991; McDonald and Ulaby, 1993). Both model classes are similar in that they calculate the same major scattering terms: (1) volume scattering from the tree canopy (the branches and leaves/needles); (2) direct ground scattering; (3) ground-to-trunk scattering; (4) ground-to-crown scattering; and (5) ground-to-crown-to-ground scattering. (Figure 2-1 presents examples of outputs from the theoretical model of Ulaby et al., 1990). Most models use formulations which assume the tree trunks and branches can be modeled as lossy dielectric cylinders, and the leaves or needles as dielectric discs or cylinders, respectively.
A three-dimensional microwave backscatter model for forest canopies, which allows explicit spatial arrangement of scatterers, has been published (Sun and Ranson, 1995). Scattering models have been exercised and validated using SAR and scatterometer data collected over a wide range of vegetation canopies (Sun and Simonett, 1988; Chauhan et al., 1991; Durden et al., 1989; Lang et al., 1994; McDonald et al., 1990, 1991; Moghaddan et al., 1994; Way et al., 1994; Ranson and Sun, 1994; Wang et al., 1993a, 1993b, 1994a). Because of their complexity, however, these models have not proved invertible to allow estimation of surface and canopy characteristics needed to study specific ecological features or processes. The value of these models lies in their utility in understanding the dependence of microwave backscatter on system and imaging parameters (frequency, polarization, and viewing geometry of the transmitted microwave radiation) and the basic geometric characteristics of the vegetated surface being studied (Figure 2-2). In addition, these models have also been useful in developing an understanding of the effects of temporally varying factors which influence microwave backscatter, including soil moisture (Wang et al., 1994b), air temperature (Rignot et al., 1994a), and flooding (Wang et al., 1995) (Figure 2-3). This understanding has proven critical in developing approaches to use SAR data in algorithms to estimate specific surface characteristics (see, e.g., Dobson et al., 1995c; Wang et al., 1994a; Kasischke et al., 1994a; and Rignot and Way, 1994).
Ecologists use remote sensing technologies for two distinct purposes. The first addresses the fundamental ecological goal to understand relationships between organisms and their environments. To this end, remote sensing is used for interpretation of landscape patterns, examination of correlations among physical and biotic parameters, and extrapolation of known relationships to larger spatial scales. The second application involves using information derived from remote sensing systems in the study of specific ecosystem processes.
Vegetation classifications make possible studies of successional rates; landscape change; vegetation productivity and biomass; and effects of disturbances such as flooding, fires, disease, and harvesting or logging. These provide inputs for modeling of a variety of ecosystem processes, such as forest succession; vegetation/atmospheric exchanges of energy and water; and local, regional, and global-scale biogeochemical cycling.
Further terrestrial applications for remotely sensed data sets include inventory of forest resources, monitoring agricultural crops, locating vegetation containing particular species of interest, and monitoring land use and land-use change. Users of land cover classifications include plant and animal ecologists, modelers including those operating general circulation models (GCMs), land managers, government agencies, and economic forecasters. All applications require the classification of vegetation into types and the delineation of the structural and compositional boundaries of biotic communities.
For many ecological studies, there is a need for current information on the distribution and amount of vegetation. This need has not been fully addressed by a quarter century of spaceborne remote sensing systems operating in the visible and near-infrared region of the electromagnetic spectrum. Collection of visible/near-infrared imagery over ecologically important regions on a continuous basis is often limited by cloud cover, particularly in tropical and boreal biomes.
Image classification algorithms discriminate based on features extracted from the spectral, spatial, or temporal domains. Two general image classification approaches have been applied to SAR data: (1) maximum-likelihood classifiers including supervised and unsupervised cluster analysis and (2) knowledge- based techniques such as hierarchical decision trees and those based on determination of dominant scattering mechanisms from electromagnetic theory. A key issue for any of these approaches is how consistent or stable the classifier is when applied to new regions or the same region at different times.
A summary of the demonstrated capabilities of SAR-derived classification is given in Table 2-1 and Figure 2-4. This material is not intended to be all-inclusive, but draws examples from a number of ecosystems as recently reported in the literature. Manual classification of airborne SAR imagery provided the first comprehensive mapping of many tropical areas in the 1970's. The most notable of these was project RADAM in Brazil. Unsupervised classification of digital SAR imagery is a useful tool for characterizing landscapes without adequate vegetation maps (e.g., Pope et al., 1994). Most recent classification efforts have used supervised maximum likelihood approaches; these often lead to high classification accuracies for a given scene (de Grandi, et al., 1994; Lemoine, et al., 1994; Ranson and Sun, 1994; Rignot and Chellappa, 1992; Rignot et al., 1993). When applied to temporal sequences of images, this technique implicitly incorporates ancillary knowledge such as phenologic development or cropping calendars. The extendibility of supervised maximum likelihood techniques to regional or global scales is impaired by the need for local training. The knowledge-based techniques may overcome this limitation by first classifying on the basis of explicit relationships between radar backscatter and structural attributes (Dobson et al., 1995b; Pierce et al., 1994; van Zyl, 1989). These structural classes can then be relabelled locally on the basis of known linkages between structure and floristic community. At present, such classifiers have been successfully tested locally.
SAR can clearly (1) detect major hydrologic changes such as inundation (Figure 2-3, Ormsby et al., 1985; Morrissey et al., 1994, 1995; Hess et al., 1995), the presence of intercepted precipitation (Ulaby et al., 1983), and freeze/thaw status of vegetation (Rignot and Way, 1994; Way et al., 1994); (2) differentiate major structural differences in land-cover such as forest versus clear cuts or marshes versus flooded forests (Beaudoin et al., 1994; de Grandi et al., 1994; Dobson et al., 1995b; Drieman, 1994; Hess and Melack, 1994, Lopes et al., 1993; Lozano-Garcia and Hoffer, 1993; Pierce et al., 1994; Ranson and Sun, 1994), and (3) discriminate crop cover on the basis of structural attributes (Foody et al., 1994; Lemoine et al., 1994; van Zyl and Burnette, 1992). In addition, major height classes within a given vegetation type can be detected. Compositional variations within major vegetation types (physiognomic types) are not as readily distinguished, but can be separated with use of multitemporal, multifrequency and/or multipolarization imagery (Ahern et al., 1993; Drieman, 1994).
In regions characterized by persistent cloud-cover (e.g., boreal regions, tropical forest and much of Northern Europe), SAR may be the only viable alternative for classification of actual land-cover. In other regions, SAR is very useful as it provides information on structure and moisture status that is complementary to the information provided by optical sensing techniques (Lozano-Garcia and Hoffer, 1993). Moreover, SAR has a generic advantage over optical sensors because atmospheric correction is not needed. In addition, a number of studies have shown the power of multidate imagery for enhanced classification results (Ranson and Sun, 1994). Orbital SAR has proven itself to be very reliable for provision of multidate data because it is practically insensitive to local weather conditions. The calibration stability of existing satellite SARs makes it possible to incorporate time-dependent ancillary information, such as phenological development and cropping calendars, into classification.
A complication to SAR-derived land-cover classification is that imposed by topographic relief. In severe cases, the layover and shadowing produced by mountainous terrain makes classification inappropriate for these regions unless azimuth and viewing geometry have been carefully considered in the SAR sampling strategy. In less severe cases, the ancillary digital elevation data have been used to generate approximate corrections for terrain effects prior to classification. A second limitation is the spatial resolution of a given SAR. Landscape patches cannot be unambiguously discriminated and classified unless they are much greater than the spatial resolution (for a single-look image). In general, classification of patch sizes <~2000 m2 (i.e., ~45m x 45 m) is not practical from spaceborne SAR.
Measuring Above-Ground Woody Plant Biomass
The amount and distribution of biomass over the Earth's land surface is one of the major uncertainties in our ability to understand the global carbon cycle and how it may change in the future (Post, 1993). The living and dead biomass in both above and below-ground storage pools constitutes a major terrestrial store of carbon. Our knowledge of the biomass density within the Earth's terrestrial biomes is quite limited due to the difficulty of obtaining sufficient high quality observations that are representative of a region or ecosystem type (Smith et al., 1993; Dixon et al., 1994). Measurements on the ground are very time-consuming, labor-intensive and often constrained by lack of access. The physiological activity of living biomass and the fate of dead biomass determine the fluxes of carbon from the terrestrial biosphere to the atmosphere, and, thus, the accumulation or removal of important greenhouse gases (primarily carbon dioxide and methane) in the atmosphere. These processes are fairly dynamic and subject to change in response to a variety of environmental factors (e.g., temperature, moisture, nutrient availability) and patterns of disturbance, both natural (e.g., fire, windthrow, insect-induced mortality) and anthropogenic (deforestation, land degradation) (e.g., Solomon and Cramer, 1993).
Numerous studies have demonstrated that approaches using optical remotely sensed data do not work for most terrestrial biomass densities, because there is a saturation effect at very low levels of biomass. Currently, radar remote sensing appears to offer the greatest promise for obtaining estimates of biomass via remote sensing techniques.
The dependence of microwave backscatter on total above-ground biomass has been documented in monospecific pine forests found in the southeastern U.S. (Figure 2-5) and France (Dobson et al., 1992; Kasischke et al., 1994a; LeToan et al., 1992), mixed deciduous and coniferous forests of Maine, northern Michigan, and Alaska (Ranson et al., 1994; Dobson et al., 1994; Harrell et al., 1995; Rignot et al., 1994), and coniferous forests of the Pacific Northwest (Moghaddam et al., 1994). These studies all show the same results: (1) the sensitivity of microwave backscatter to biomass variations saturates after a certain level is reached; and (2) the biomass dependence of microwave backscatter varies as a function of radar wavelength and polarization (Figure 2-6). In summary, the saturation point is higher for longer wavelengths, and the HV polarization is most sensitive and VV the least.
A conclusion drawn by some scientists is these single-frequency saturation levels represent the upper limit of SAR's ability to monitor changes or differences in aboveground biomass in forests (Waring et al., 1994). However, this conclusion overlooks several important considerations. Microwave backscatter is correlated with total biomass and various components of biomass (e.g., branch biomass, needle biomass, bole biomass) or other physical characteristics (e.g., tree height, basal area) (Dobson et al., 1995c; Hussin et al., 1991; Kasischke et al., 1994a). This should not be surprising, since we know that different biomass components of trees are closely correlated. Since different radar frequencies and polarization combinations are sensitive to different layers of a forest canopy, it should be possible to use multiple channels of SAR data to estimate total above-ground biomass. Recent research supports this hypothesis.
Kasischke et al., (1994a) used a two-stage approach to estimate biomass of southern pine forests using JPL AIRSAR data. In step one, total branch biomass was estimated as a function of several different radar frequencies/polarizations. Total biomass was then estimated from branch biomass based upon allometric equations, and resulted in a relative error on the order of 20% for biomass levels up to 400 t ha-1. Ranson et al., (1994) used a ratio of P-band HV (PHV) and C-band HV (CHV) to estimate total biomass (up to 250 t ha-1) in mixed coniferous/deciduous forests in Maine. This technique was applied to SIR-C/X-SAR LHV and CHV data to estimate boreal forest biomass up to 200 tons/ha within +20 tons/ha (Ranson and Sun, 1995). Finally, Dobson et al., (1995c) used a multi-step, semi-empirical approach to estimate aboveground biomass from a combination of channels from SIR-C data collected over a mixed coniferous/deciduous forest in northern Michigan. In this approach, different SAR frequency/polarization combinations were used to estimate canopy-layer biomass, total height and total basal area, which were then used to estimate total biomass. Biomass estimates up to 250 t ha-1 with an uncertainty on the order of 16 t ha-1 were achieved (Figure 2-7).
Delineation of Wetland Inundation and Vegetative Cover
The availability of SAR data from airborne and satellite platforms has provided a unique opportunity to study dynamic wetland processes, information critical for the study of many ecosystem processes and applications. In this section, we discuss using SAR for several wetlands issues, including biogenic trace gas exchanges, monitoring the effects of rises in the average sea levels, and monitoring disease vectors.
In studies of biogenic trace gas exchange, remotely sensed data can provide unique information on the type and distribution of wetlands and on temporal distribution of inundation. Uncertainties in the spatial and seasonal extent of methane source and sink areas remain one of the greatest unknowns in the global methane budget (Bartlett and Harriss, 1993). Natural wetlands comprise the largest natural source of atmospheric methane (Aselmann and Crutzen, 1989; Fung et al., 1991). Hence, characterization of the areal and temporal extent of global wetlands would greatly extend our understanding of trace gas exchange from these ecosystems and of their significance to global processes.
Recent results using data from the TOPEX/POSEIDON radar altimeter has provided clear evidence that average sea levels are rising, at a rate of several millimeters per year (Nerem). These observations support evidence derived from tidal gages that sea levels have been rising over the past half century. The distribution of different vegetation species in coastal wetlands, especially estuaries, is highly sensitive to levels of tidal inundation. Monitoring changes in inundation and vegetation patterns in coastal wetlands will provide a key means to monitor the progression and effects of rises in sea level.
Finally, insect-borne organisms cause disease outbreaks throughout much of the developing world, particularly in the southern hemisphere. Many of these diseases are carried and transmitted by various species of mosquitoes. In regions with distinct wet and dry seasons, mosquito populations increase dramatically when breeding sites flood during prolonged episodes of high rainfall. Accurate maps of vegetation inundation are critical in identifying breeding grounds for disease vectors and in predicting and monitoring outbreaks of a variety of diseases.
For most scientific questions involving wetlands, it is necessary to distinguish not only flooded from non-flooded areas, but herbaceous from woody vegetation. For example, in floodplains of the central Amazon, methane generation rates from floating meadows are much higher than those from flooded forest. Delineation of both flooding status and vegetation, with accuracies greater than 90% for all categories, has been demonstrated using multi-frequency, polarimetric SAR data sets for wetlands in the southeastern United States (JPL AIRSAR) and the central Amazon (SIR-C) (Melack et al., 1994; Hess et al., 1995).
The ability to penetrate the extensive cloud cover of northern and equatorial latitudes and to detect standing water beneath vegetation canopies is unique to SAR (Hess et al., 1990). In northern Alaska, for example, cloud cover and infrequent repeat cycles have limited the acquisition of optical satellite data (Landsat and SPOT) to a total of two scenes for the last 20 years. By comparison, ~50 scenes of ERS-1 SAR data have been acquired for Barrow over a two-year period, and RADARSAT will provide daily access. Similarly, in the equatorial regions of Manaus, Brazil, between 0 and 2 Landsat scenes are available each year.
SAR has proven useful in delineating inundation, a key indicator of the anaerobic conditions necessary for methane production. Backscatter from ERS-1 SAR acquired over Barrow, Alaska in 1991 is strongly related to the position of the local water table and thus to methane exchange rates (Figure 2-8; Morrissey et al., 1994). Backscatter from non-inundated sites was low, that from herbaceous inundated sites was high, and that from sites with the water table at the surface was intermediate, mirroring methane exchange rates for the region.
The capability to differentiate wetland source areas and non-wetlands with SAR is further enhanced by the availability of time series ERS-1 SAR data (Kasischke et al., 1995a). Seasonal changes in microwave backscatter for northern wetlands and non-wetlands are shown in Figure 2-9. Under an extended period of freezing temperatures in winter of 1992, radar returns for wetland and non-wetland did not differ significantly. Following snowmelt in the spring of 1992, backscatter for wetlands was consistently higher than that from non-wetlands. With the onset of colder temperatures and decreasing daylight in late summer, backscatter for both wetlands and non-wetlands decreased dramatically with freezing (Way and Rignot, 1994). These time series SAR data provide the only basis for an ongoing effort to map tundra wetlands on a global basis because cloud cover precludes using visible/near-IR channel imagery to monitor this biome on a continuing basis.
The sensitivity of microwave backscatter to vegetation and surface make SARs an attractive tool for characterizing wetland communities. While species composition per se usually cannot be detected with single-channel SAR, plant communities often can be, due to differences in vegetation height, density, or architecture. For example, a recent study using airborne SAR data collected over wetlands in Belize showed that a high degree of separability between the different vegetation communities could be achieved using a variety of indices based upon different radar frequencies and polarizations (Pope et al., 1994). Field studies of the same Belizean marsh communities demonstrate that vegetation species, composition, and biomass correlate well with salinity and nutrient gradients (Rejmankova et al., 1995). Given these relationships, the spatial distribution of wetland types and inundation patterns can be used to infer much about the chemistry and flow of surface and near-surface waters.
Studies by Tanis et al. (pri. comm., paper in review with Rem. Sens. Environ.) have shown that ERS-1 SAR imagery can discriminate between flooded and non-flooded areas in a coastal estuarine ecosystem along the Gulf Coast of Florida. Examination of tidal gage data revealed that the flooding detected on the ERS-1 imagery was due to variations in tidal stage. Techniques were developed to map the ranges of tidal flooding based on comparison of ERS-1 SAR imagery collected at high and low tides.
Another important issue in applied wetland ecology is the role of wetlands as breeding sites for mosquitoes that transmit a variety of diseases. Pope et al., (1992) used airborne X-, C- and L-band SAR data to examine flooding in sedge and grass-covered mosquito larval habitats in Kenya. LHH data provided the best flood detection in both wetland types, but CHH and LVV also provided limited detection capability. The radar data (backscatter magnitude only) were not adequate for habitat mapping, which was accomplished with TM data. In contrast, the multifrequency, polarimetric airborne SAR data collected over the wetlands in Belize (Pope et al., 1994) were capable of mapping Eleocharis sp. dominated marshes, which have been shown to be important breeding habitats of the malaria vector Anopheles albimanus (Rejmankova et al., 1993). In summary, SAR data hold great promise for malaria risk assessment efforts in Belize and adjacent regions by providing spatial and temporal information on the distribution and flooding status of anopheline breeding sites.
Monitoring of Dynamic Processes in High-Latitude Ecosystems
The need to better quantify factors influencing the carbon cycle in northern biomes stems from several factors. First, based on current estimates of the present-day rates of increases in atmospheric carbon concentrations balanced against oceanic carbon uptake, Tans et al., (1990) concluded that a Northern Hemisphere terrestrial carbon sink on the order of 2.0 to 3.4 Gt C yr-1 is required to balance the global carbon budget. While Tans et al., (1990) associate this sink with temperate latitude forests, Bonan (1991a,b) suggests this sink may actually be a consequence of an imbalance in production and decomposition in high northern latitude ecosystems (the terrestrial biomes in high northern latitudes account for >40% of all carbon sequestered in living and dead biomass). Second, it is the consensus of most atmospheric general circulation models that significant warming of the northern-hemisphere, high latitudes will result from a doubling of atmospheric CO2 concentrations (e.g., McFarlane et al., 1992). Given that the storage capacity of high northern latitudes is driven by the low rates of plant decomposition because of low annual temperatures, the projected temperature increases in this region will have profound influences on carbon cycling in its terrestrial biomes.
Uptake and release of CO2 by the boreal forest may account for approximately 50% of the annual seasonal amplitude in atmospheric CO2 at Point Barrow, Alaska, and ~30% of the seasonal amplitude at Mauna Loa (D'Arrigo et al., 1987). Data have shown that the seasonal amplitude of atmospheric CO2 concentrations in northern latitudes has increased with time. This may reflect increased metabolic activity of ecosystems in northern latitudes due to warmer air temperatures and "CO2 fertilization" (Bacastow et al., 1985; Houghton, 1987). Change to a warmer, drier climate may release more than 1 to 2 Gt C yr-1 to the atmosphere from boreal ecosystems due to a variety of feedbacks (Kasischke et al., 1995b). In addition to increasing metabolic activity, increased high latitude temperatures may also extend the growing season resulting in increased annual productivity, as well as periods of frost drought which may reduce annual productivity. Measurements of the length of the growing season may significantly improve current estimates of net annual CO2 flux in the boreal regions (Way et al., 1994). For coniferous species, the summer frost-free period bounds the growing seasonal length.
Fires are an extremely important factor to a variety of processes which affect forest succession and biogeochemical cycles in boreal forests (Bonan and Shugart, 1989; Kasischke et al., 1995b; Viereck, 1983). Recent studies show that global warming may increase the frequency and intensity of fires in boreal forests and result in a significant release of carbon to the atmosphere (Kasischke et al., 1995b). Much of the carbon released will come from increased rates of decomposition in the ground layer. While global data sets are not available, records from North America show that an average of 2.5 million hectares have been affected annually by fire over the past decade. Fires in boreal forests tend to cover large areas, typically greater than 1,000 ha in size. Fires covering 50,000 to >1,000,000 ha are not uncommon; thus, satellite remote sensors are an ideal tool for monitoring the locations and effects of fires in this region.
The ERS-1 SAR data collected at the Alaskan SAR Facility and Canadian and European receiving stations have provided a unique opportunity to study high-latitude terrestrial ecosystems found in North America. In addition to monitoring tundra biomes, ERS-1 SAR imagery was used to study boreal forests, focusing on issues related to the terrestrial carbon cycle. Specifically, this research studied the effects of changes in the length of growing season on net seasonal fluxes of CO2 and the effects of fire on carbon cycling.
A number of remote sensing instruments may be used to estimate growing season length, and it is likely that a combination of sensors will provide the most accurate information (Way et al., 1994). The Advanced Very High Resolution Radiometer (AVHRR), for example, may provide good estimates of leaf-on period, thus bounding the growing season length for deciduous species. For coniferous species, growing season may be halted when air temperatures drop below -2 degrees C. For closed canopy forests, canopy temperatures are within a few degrees of air temperatures (Luvall and Holbo, 1989), and can be estimated using thermal infrared emissions gathered by AVHRR. Access to these data is, however, limited by cloud cover. A third technique is to measure the length of the growing season by monitoring freeze/thaw transitions using imaging radar data (Figure 2-10) (Way et al., 1990; Way et al., 1994; Rignot et al., 1994; Rignot and Way, 1994). At microwave frequencies, freezing results in a large decrease of the dielectric constant of the dielectric constant of soil and vegetation because the crystal structure prevents the rotation of the polar water molecules contained within the soil and vegetation. This phase change results in a drop in radar backscatter of several dB.
Over the past several years, scientists have developed a greater understanding of the role of biomass burning in boreal forest in the global carbon cycle and the need to monitor these fires on a continuing basis. Because of their large sizes and remote locations, satellite sensing systems are now recognized as the only reliable means to annually locate and estimate the areal extent of fires in boreal forests (Kasischke and French, 1994). In 1990 and 1991, over 2 million ha of land surface were affected by fire in Alaska, with most (1.85 million ha) occurring in forested regions. Studies have shown that these recent fires resulted in characteristic signatures on ERS-1 SAR imagery (Kasischke et al., 1992; 1994c). Field research (Kasischke et al., 1995a) has shown that the spatial and temporal signatures present on the ERS-1 SAR imagery (Figure 2-11) are highly correlated with variations in the moisture found in the top 5 to 10 cm of the soil (Figure 2-12).
Schlenter and VanCleve (1985) have shown that rates of aerobic decomposition in black spruce forests are directly proportional to both temperature and soil moisture. Since soil temperatures increase significantly after fires in boreal forests, the rates of soil respiration in these fire-affected forests should also increase. Recent field measurements (Kasischke, pers. comm.) have shown this to be true in fire-disturbed black spruce forests in Alaska, where fluxes of CH4 and CO2 were greater within the burned sites than unburned sites. Areas which had high soil moistures also had significantly higher rates (by over an order of magnitude) of CH4 and CO2 fluxes than areas of low soil moisture. The ability of SAR data to monitor the spatial and temporal patterns of soil moisture therefore provides a means to more accurately quantify greenhouse gas emissions from fire-disturbed forests.
The wide range of applications discussed in this paper illustrates that there is probably no one ideal SAR system for ecological applications. For some applications, existing or planned single-frequency/polarization systems may provide an adequate data set. For other applications these systems are inadequate.
To assess SAR system considerations for ecological applications we take the approach of first defining the system parameters for a specific list of applications. We then discuss the potential of existing or planned SAR systems for these applications.
The SAR parameters which define the utility of a specific system for ecological applications are its frequency, polarization, angle, resolution and sampling frequency. The common frequencies used today include P-band, L-band, S- band, C-band and X-band. (See Appendix B for details on SAR frequency bands and instruments used.) Radars typically transmit horizontally or vertically-polarized microwave energy and can receive either polarization, resulting in four linear polarizations - HH, HV, VH, and VV. Today's spaceborne SAR systems usually have a fixed center angle between 20 degrees and 50 degrees with images covering a few degrees from near edge to far edge. Future systems will operate in a SCANSAR mode, with image swaths covering a 20 degrees to 50 degrees range in angles. Today's spaceborne SARs have fairly fine resolution (20 to 40 m), narrow swath widths (60 to 100 km), and long sampling frequencies (20 to 40 days). SCANSAR systems will have the ability to cover wide areas at lower resolutions (up to 500-km swaths with 100 to 200 m resolution) and higher sampling frequencies (every 2 to 4 days).
Table 2-2 lists eight (8) ecological or land surface applications for imaging radar systems and summarizes the optimal SAR parameters for each application.
Utility of Existing/Planned SAR Systems
Three spaceborne imaging radar systems are now in operation or were deployed during the last year: ERS-1, JERS-1 and SIR-C/X-SAR. The ERS-1 SAR is a C- band VV (vertical transmit polarization/vertical receive polarization) launched in the summer of 1991. This system has a 25 m resolution and 100 km swath. The orbit of this system is tailored such that it has a 35-day exact repeat orbit during the northern hemisphere summer and fall (which means it can image the same ground location every 18 days or so), and a 3- day exact repeat orbit during the winter and spring in order to obtain frequent coverage of the polar ice cap. Coverage is limited to locations where ground receiving stations are installed. The JERS-1 SAR is an L-band HH system launched during the summer of 1992. It has a 25-m resolution and a 75-km swath. The exact repeat orbit on this system is 44 days. Onboard recording provides global access. The SIR-C/X-SAR system was flown onboard NASA's Space Shuttle on two ten day missions in April and October of 1994. This system consisted of a C- and L-band SAR system that was fully polarimetric (i.e., it collected HH, HV, VH, and VV imagery) and an X-band SAR which collected VV data. The resolution of this system ranged between 10 and 40 m, and it collected image swaths between 15 and 90 km wide. The ground coverage of this system was limited in order to image specific test sites during its two missions.
Three spaceborne SAR systems are planned for the future: ERS-2, RADARSAT, and ASAR. The ERS-2 SAR will be an exact duplicate of ERS-1, and is scheduled for launch in early 1995. The Canadian RADARSAT will consist of a C-band SAR with HH-polarization and has a variety of modes for resolution/swath width. The SCANSAR mode can yield swath widths up to 500 km with a spatial resolution of 100 m. This wide swath mode will allow imaging of the same geographic location once every 2 to 3 days. The ERS-2 follow-on, known as Advanced SAR or ASAR, will be deployed on the planned European ENVISAT and consists of a dual-polarized C-band SAR. It will have both co-polarized channels (HH and VV), but not cross-polarized (HV). The planned "wide swath" mode has a 400- km swath width and 100-m resolution.
Table 2-3 summarizes the potential or demonstrated capabilities of these systems relative to the applications of ecological interest.
It is the consensus of the ecological working group that SAR data contain unique information that can be exploited by the terrestrial science community to study a wide range of physical, chemical and biological processes important to NASA's Mission to Planet Earth (MTPE), as well as the U.S. Global Change Research Program (GCRP). Over the past decade, NASA has taken the U.S. lead in development of the technological as well as the scientific research infrastructure related to airborne and spaceborne SAR systems. NASA initiatives in this area include: (1) development and operation of the JPL AIRSAR system; (2) development and deployment (with the German and Italian Space Agencies) of the SIR-C/X-SAR system; (3) establishment of the Alaska SAR Facility to receive SAR data from satellite systems such as ERS-1, JERS-1 and RADARSAT; and (4) sponsoring of extensive research programs to develop techniques to use information derived from these systems to study a wide range of oceanic and terrestrial processes. The size of NASA's investment in this area over the past decade has been significant.
These NASA-sponsored programs, as well as parallel programs in other countries, resulted in most of the research discussed in this paper. Based on these results, other U.S. government agencies are now beginning to recognize the utility of SAR data for monitoring terrestrial surfaces and processes, and are initiating research and development programs to exploit existing or future SAR systems. For example, the Environmental Protection Agency is sponsoring programs which will utilize SAR data as one component of a system to map and monitor variations in forest and vegetation cover in order to derive better estimates of the global terrestrial carbon budget. In another program, the U.S. Fish and Wildlife Service sponsored research to develop techniques to monitor tidal periods and vegetation cover in coastal estuaries.
It is the consensus of the ecology working group that the full utility of SAR for ecological applications is now only emerging and that this potential will not be fully realized within the MTPE and U.S. GCRP without a well-focused SAR R&D program within NASA. Fully exploiting the potential of imaging radars systems requires careful consideration of a number of sometimes conflicting factors, including:
(1) The need to conduct additional remote sensing science studies to understand the source of the different types of signatures present in radar imagery. This is particularly true in the multi-temporal data sets now being collected by existing satellite systems such as ERS-1 and JERS-1.
(2) Based on the results of recent studies using existing satellite SAR systems as well as recently collected SIR-C/X-SAR data, the need to conduct efforts to develop and validate algorithms which use radar data to estimate specific surface characteristics.
(3) The need to continue to develop new approaches to exploit the spatial and temporal information derived from radar imagery in ecosystem and process studies.
(4) The need to exploit the full breadth of the SIR-C/X-SAR data sets beyond the ~50 experiments presently being conducted. This need includes not only providing existing experiments with adequate resources to conclude their work, but also identification of additional research which could be carried out using this unique data set. We have not identified a compelling reason for a third flight that is an exact repeat of the first two flights, especially in light of the scale of effort required to analyze the existing data set and the fact that it is unlikely that sufficient resources will be provided to analyze these additional data.
(5) The need to develop and implement a coherent research plan within NASA's Mission to Planet Earth to fully exploit the SAR data sets being provided by existing and planned
international SAR missions.
(6) To fully exploit the potential of SAR to meet the objectives of Mission to Planet Earth in the area of terrestrial ecosystem science, there is a need to develop a multi-frequency (with a minimum of two frequencies, C- and L- bands), polarimetric SAR system. In the development of such a system, the tradeoffs between resolution, swath width, and sampling frequency in terms of scientific requirements versus cost considerations must carefully be considered. However, if the cost of obtaining a multi-frequency, polarimetric spaceborne SAR is a reduction of spatial resolution, such a lower- resolution system would be quite suitable for many ecological applications.
The above needs, in essence, are the primary recommendations of the ecology working group. The working group realizes that it would be very difficult to encompass these recommendations in the context of existing programs within NASA, and that they are unlikely to be implemented unless a new program is started. Thus, we make the following summary recommendations:
(1) NASA should consider development of a new satellite remote sensing system which would contain a multi-frequency, polarimetric SAR.
(2) Even without the development of an advanced SAR system by NASA, the existing SAR data sets as well as those which will be collected by satellite SAR systems in the future do provide unique information for a variety of ecological processes. NASA should take a more proactive role in ensuring that these data are used in an effective manner within its Mission to Planet Earth.
(3) To achieve recommendation 2, NASA should create a multi-disciplinary SAR Facility Instrument Team. In addition to carrying out the recommendations outlined above, this team would also serve as the advisory board for development of the new SAR satellite system recommended above.
Table of contents
Previous Section
Next Section
Imaging Radar Home Page
