Ecological Research Needs from Multiangle Remote Sensing Data

Ecological Research Needs from Multiangle Remote Sensing Data Gregory P. Asner,* B. H. Braswell,† David S. Schimel,‡ and Carol A. Wessman*

Remotely sensed land surface reflectance depends upon

changing sun and sensor viewing geometry, and this dependence is governed by the bidirectional reflectance distribution function (BRDF). Because the reflectance distribution of vegetation is strongly anisotropic, multi-view angle (MVA) observations of terrestrial ecosystems contain additional and unique information beyond that acquired with nadir or single-angle spectral measurements alone. With the NASA EOS instruments MODIS and MISR and France’s POLDER, new capabilities in MVA remote sensing will become widely available for ecological, biogeochemical, and land-surface biophysical research. However, a communication gap exists between the remote sensing and ecological communities in terms of the capabilities of the former and the needs of the latter. In this article, we present a summary of ecological research needs for remotely sensed data. Based on these needs, we present a review of some of the most promising MVA remote sensing methods for fulfilling these requirements. With this article, we hope to facilitate increased communication between the remote sensing, ecological, and biogeochemical research communities. Elsevier Science Inc., 1998

INTRODUCTION The remote sensing community will soon enter the Earth Observing System (EOS) era, and technologies formerly inaccessible for mapping and quantification of vegetation * Department of Environmental, Population, and Organismic Biology and Cooperative Institute for Research in Environmental Sciences, University of Colorado, Boulder † Institute for the Study of Earth, Oceans, and Space, University of New Hampshire, Durham ‡ National Center for Atmospheric Research, Boulder Address correspondence to G. P. Asner, CIRES/CSES, Campus Box 216, University of Colorado, Boulder, CO 80309-0216. Received 28 April 1997; revised 22 August 1997. REMOTE SENS. ENVIRON. 63:155–165 (1998) Elsevier Science Inc., 1998 655 Avenue of the Americas, New York, NY 10010

characteristics at large scales will become widely available. The EOS AM-1 satellite will have two instruments with the potential to revolutionize regional- and global-scale measurements of key ecological variables. The Moderate Resolution Imaging Spectrometer (MODIS) will image the Earth’s surface in 20 optical channels. Due to its 2330-km swath width, many off-nadir measurements of the surface will be contained in each across-track scan line. Over a sufficient period of time (e.g., 10 days), ground locations will be observed from multiple view angles and with slightly different solar geometries. The Multi-angle Imaging Spectroradiometer (MISR) will simultaneously provide nine view angles and four optical channels for each pass over a surface target. Because of their off-nadir imaging capability, both MODIS and MISR will acquire samples of the angular distribution of photon energy reflected from the surface. While the NOAA Advanced Very High Resolution Radiometer (AVHRR) has a similar angular sampling capability as MODIS, the combination of MODIS and MISR, as well as the planned data processing algorithms for EOS, will advance this capability to an operational level with increased spectral and spatial resolution (Barnsley et al., 1994; Running et al., 1994a). In addition, France’s POLDER (Polarization and Directionality of the Earth’s Reflectances) instrument has already acquired samples of the angular reflectance distribution of land covers (Table 1). This dependence of observed reflectance on sun and sensor geometry is termed the bidirectional reflectance distribution function (BRDF). For both the atmosphere and vegetation, the BRDF is strongly anisotropic (Fig. 1). Variation in the BRDF of vegetation results primarily from differences in canopy- and landscape-level structural characteristics, along with leaf biochemical and soil textural attributes (e.g., Myneni et al., 1989; Jacquemoud et al., 1992). At the landscape level, both the relative ground coverage and spatial distribution of vegetation 0034-4257/98/$19.00 PII S0034-4257(97)00139-9

3 visible, 1 near-IR 275 m–1.1 km 9 fixed cameras 08626.18, 645.68, 660.08,670.58

Similar to AVHRR 6558 6/98

types, each with differing crown geometry, largely determine the shape of the BRDF (Fig. 2; Li and Strahler, 1985; Li et al., 1995). However, canopy-level structural characteristics such as leaf area index (LAI), leaf angle distribution (LAD), and foliage clumping also play a major role (Ross, 1981; Goel, 1988; Myneni and Asrar, 1993; Chen and Cihlar, 1995). Scattering at the leaf and stem level generally determines the overall strength of the reflected signal at a particular wavelength, while the geometrical dependence is primarily a function of canopy and landscape structure (Goel, 1988; Jacquemoud et al., 1995; Li et al., 1995). Improved knowledge of the canopy attributes controlling vegetation BRDFs, combined with the EOS and POLDER instruments’ capability in acquiring angular radiance samples, will likely lead to a significant improvement in the role of remote sensing in ecological research. While nadir multispectral measurements have provided a means to map and estimate certain characteristics of vegetation cover types, multi-view angle (MVA) measurements of the vegetation BRDF will allow improved access to canopy structural characteristics (e.g., LAI) and simultaneous retrieval of important biophysical variables such as the fraction of photosynthetically active radiation absorbed by plant canopies (fAPAR). In this article, we review some of the more pressing needs of the ecological and biogeochemical research communities with respect to ecosystem- to global-scale structural and biophysical data. The discussion will be couched in a context of structure and function at different ecological scales, since the potential products derived from remote sensing data have varying degrees of utility depending on their spatial and temporal resolution. We then describe the unique capabilities of MVA remote sensing as they apply to ecological and biogeochemical research. What products or ecological parameters might be derived or improved from multiangle remote sensing measurements in the EOS/POLDER era, and how will these products influence ecological monitoring and analysis efforts? We hope this article will lay the groundwork for further discussion and collaboration between the ecological and remote sensing communities.

Multi-angle Imaging Spectroradiometer (MISR)

ECOLOGICAL RESEARCH NEEDS FROM REMOTE SENSING Moderate Resolution Imaging Spectrometer (MODIS)

2nd launch in 1999

6/98

20 optical channels spanning the full shortwave spectrum 250 m (Channels 1–2), 500 m (Channels 3–7), 1.1 km (Channels 7–20) (all at nadir)

4 visible, 4 near-IR Combination of very wide fore/aft FOV and temporal compilation 8/96 Failed 6/97 Polarization and Directionality of the Earth’s Reflectances (POLDER)

NOAA-14 12/94

6428

6 3 7 km

1.1 km at nadir, .4 km at 558 view zenith angle Across-track FOV; temporal compilation leads to multiangle data sets Advanced Very High Resolution Radiometer (AVHRR)

NOAA-12 5/91

6558

Spatial Resolution BRDF Sampling Method Angular Resolution Launch Date Sensor

Table 1. Current and Forthcoming Satellite Instruments Capable of Multi-View Angle Sampling of the Surface BRDF

1 visible, 1 near-IR

Asner et al.

Optical Bands

156

Modern ecological research places an emphasis on the variability in structure and function at all spatial and temporal scales. However, changes in function may or may not manifest in measurable structural changes across scales of interest. This is one common interpretation of hierarchy theory in ecological research (Allen and Starr, 1982; O’Neill et al., 1986). Knowledge of these scaling issues has largely fueled the interest in both top-down and bottom-up approaches to studying ecological pro-

Multiangle Remote Sensing and Ecology

Figure 1. Sample angular reflectance distribution along the solar principal plane (the plane defined by solar inclination and normal angles) collected with the MMR ground radiometer during the NASA FIFE campaign in Kansas. Notice the strong backscattering effect in the retrosolar direction (often called the hotspot) at 258 from nadir. This and other characteristics of the bidirectional reflectance distribution function (BRDF) are largely determined by canopy- and landscape-level structural characteristics.

cesses (Wessman, 1992). Ecological research efforts tend to focus at specific spatial scales ranging from individual plants, plant populations, and multispecies communities to whole ecosystems, biomes, and the globe. For the most part, ecological studies at ecosystem to global scales have most benefited and will likely continue to benefit from remote sensing research. Ecological processes, and specifically biosphere– atmosphere interactions and biogeochemistry, are highly interactive between regional and global levels. An obvious example was the effect of stratospheric aerosols emitted by Mt. Pinatubo on global temperature, vegetation net primary productivity, and atmospheric CO2 lev-

Figure 2. Two levels of vegetation structural characteristics that affect the angular reflectance distribution or BRDF measured by aircraft or spacecraft optical sensors. In horizontally discontinuous canopies, the macrostructure largely influences the BRDF through crown shape and shadowing effects. In continuous canopies, microstructure is the dominant driver of the surface anisotropy. Whether the BRDF is primarily a function of macro- or microstructure depends on the vegetation types present, illumination conditions, and image pixel size.

157

els (Ciais et al., 1995; Schimel et al., 1996). A localized geological event caused measurable changes in atmospheric chemistry which altered climate and ecosystem function at both regional and global scales. Similarly, anthropogenic activities are altering both the structural and chemical characteristics of ecosystems at regional scales which impact global biospheric, atmospheric, and oceanic processes. For example, while CO2 and nitrogenous trace gas emissions from automobiles and industry are largely constrained to specific regions, the magnitude of their impact on the biosphere has achieved global-scale significance (e.g., Townsend et al., 1996; Asner et al., 1997a). Studies focused on these types of global change issues rely heavily on our understanding of ecological and biogeochemical processes from field experiments and modeling exercises (e.g., Rastetter et al., 1992; Schimel et al., 1994; 1996). However, most ecologists are well aware of the implications of a relatively limited sampling capability and the broad-scale extrapolation of ground measurements. Despite these limitations, ecosystem- to global-scale process studies must often be based upon, or evaluated against, plot-level analyses. Moreover, it is widely accepted that ecosystems respond differently to external perturbation depending on climatic, edaphic, and plant physiological and structural characteristics (e.g., Rastetter, et al., 1992; Schimel et al., 1994). The causes and implications of change in vegetation types, vegetation structure, and biogeochemistry at ecosystem and regional scales is thus critically needed for evaluating these ecosystem dependencies and their influence on global-scale atmospheric processes. Similar but spatially coarsened ecological variables are of interest at a global extent to assess the more general patterns in vegetation change and the feedbacks between the biosphere and atmosphere as a whole biogeochemical system.

158

Asner et al.

Table 2. Structural and Biophysical Variables of Ecological Interest That May Be Retrieved from Multi-View Angle Remote Sensing Measurementsa Ecological Scale

Variable/Parameter

Ecosystem–Biome S: 1m to ,1km T: Daily-Seasonal

1. Distribution of vegetation types and characterization of landscape structure

Nutrient cycling, hydrology, mesoscale climatology, ecosystem management, habitat studies

2. Canopy structure/function: plant LAI, SAI, fAPAR

Extrapolation of CO2 and trace gas exchange measurements, carbon and nutrient cycling, vegetation allocation and senescence Hydrology, biogeochemistry, land management

3. Soil surface texture: identification, compaction, moisture Continent–Globe S: 1-4 km T: MonthlyAnnual

1. Vegetation cover and landscape structure

2. Canopy structure/function: effective LAI and fAPAR 3. Surface albedo

Primary Research Use

Large-scale biogeochemical modeling and analysis, biophysical land surface modeling, climate modeling Biogeochemical modeling and analysis, biophysical land surface modeling Global climate modeling

BRDF-Related Methods Improved classification of individual landcovers using added multi-angular information; improved spectral mixture analysis; geometric-optical model inversion Canopy radiative transfer model inversion

Soil and canopy radiative transfer model inversion BRDF-corrected vegetation indices

Canopy radiative transfer model inversion Direct sampling and modeling of the surface BRDF

a The list of uses applies to both modeling and nonmodeling research efforts. Spatial resolution (S) and temporal (T) scales required for significant ecological progress are provided. LAI5leaf area index, SAI5stem area index, fAPAR5fraction of absorbed photosynthetically active radiation.

KEY ECOLOGICAL VARIABLES FROM MVA APPROACHES Because the utility of remotely sensed structural and biophysical data is scale-dependent, we will discuss the importance of this information to ecologists at two spatial scales: 1) ecosystems to biomes and 2) continents to the globe. While most of the potential BRDF-related methods may be similar between scales of interest, the retrieved variables and their ecological application tend to be more scale dependent. Table 2 lists the structural and biophysical variables and ecological uses that might be derived from MVA remote sensing measurements. At each set of ecological scales, there are three categories of variables listed according to application. Ecosystem to Biome Scales At ecosystem to biome scales, the categories of information needed for significant progress in ecological research are: 1) the vegetation macrostructure—the distribution of vegetation types (e.g., trees, grasses) and characterization of landscape structure (e.g., placement of overstory tree crowns) at high spatial and temporal resolution; 2) the vegetation microstructure—the within-canopy structural attributes such as leaf and stem area index (LAI, SAI) as well as biophysical variables such as the fraction of absorbed photosynthetically active radiation (fAPAR); and 3) soil surface moisture and textural characteristics such as compaction and roughness.

At regional scales, ecologists and biogeochemists require improved spatial resolution and quantification of surface constituents (e.g., live vegetation, standing litter, bare soils). This requirement stems from a need to detect relatively small changes in vegetation structure and function that occur in ecosystems through both natural and anthropogenic forcings. For example, human-caused atmospheric perturbations to ecosystems associated with industry and automobile use (e.g., atmospheric CO2, NO2, and NO enrichment), land-use change (e.g., alteration, removal, and replacement of vegetation functional groups), and many other factors result in field-measurable changes in vegetation composition (e.g., Archer et al., 1995). These changes, however, occur against a spatially and temporally complex backdrop of natural ecosystem variability. Thus, a wide variety of ecological research applications calls for detecting relatively small signals of vegetation change at high spatial and temporal resolution across ecosystems, regions, and biomes. Landscape macrostructure is characterized by an important group of variables (e.g., canopy dimensions and spacing) in ecological and biogeochemical research. For landscape ecologists, macrostructural information provides insight to the links between ecological pattern and process, especially in terms of plant–animal interactions. Because many structural vegetation types (e.g., trees, shrubs, and grasses) tend to be functionally different, the location and macrostructural attributes of these groups

Multiangle Remote Sensing and Ecology

are needed to extend the capabilities of ecosystem process modeling. Landscape macrostructure also plays an important role in mesoscale climatological research through surface roughness, evapotranspiration, and CO2 assimilation (e.g., Milhailovic et al., 1993). Finally, the distribution and crown structure of different vegetation types is required for improved land and habitat management (e.g., Cole and Lorimer, 1994; Sturtevant et al., 1996). At ecosystem to biome scales, structural and biophysical attributes of individual canopies (microstructure), such as LAI and fAPAR, are needed to extrapolate CO2 and trace gas exchange measurements, and for carbon and nutrient cycle modeling. If used to define or constrain plant aboveground carbon or allocation parameters in ecosystem process models, these remotely sensed variables could significantly increase the accuracy of the models at these ecological scales. As opposed to “effective LAI or fAPAR” or other similar measures of total live biomass within large image pixels (Table 2: Continent– Globe section), actual LAI and fAPAR are needed at high spatial and temporal resolution for applications ranging from mesoscale climate research to ecosystem management. Finally, soil surface textural properties can help in mapping and modeling hydrological and biogeochemical processes such as nitrogenous trace gas emissions. Continental to Global Scales At continental to global scales, research needs are driven primarily by studies of global biogeochemistry and ecosystem dynamics (Schimel, 1995a,b; Field et al., 1995), global and mesoscale climate modeling (Washington and Mehl, 1996), and studies of large scale biosphere–atmosphere energy, water and CO2 exchange (e.g., Pollard and Thompson, 1995; Sellers et al., 1996; 1997) (Table 2). Far from being independent, these areas are becoming increasingly interactive. For example, most climate models contain a land-surface biophysics submodel used to determine mass and energy fluxes to/from the atmosphere. Also, biogeochemical models are used to study global carbon, water, and nutrient cycling at the process level, but they also potentially interact with physical and chemical process models. Consequently, the information needed by these communities is somewhat similar (Table 2). The data that are required at these very large scales, and which can be retrieved from remote sensing, are 1) vegetation cover type, 2) structural parameters such as effective LAI, and 3) biophysical parameters such as effective fAPAR and albedo. Global vegetation distribution maps are needed for both climate and biogeochemical studies. Parameters describing the characteristics of the land surface (e.g., roughness, resistance to CO2 and water vapor exchange), or ecosystem components (e.g., carbon-to-nitrogen ratios) are keyed to global databases of land cover and land use. In the past, these maps have been created by compiling country-by-country tabular statistics (Matthews,

159

1983; Olson et al., 1983). Recently, multitemporal optical reflectance data have been used to classify the land surface based on the notion that the phenological character (e.g., seasonality) of vegetation is strongly indicative of ecosystem type (Loveland et al., 1991; DeFries et al., 1995). By “ecosystem type” or “plant functional type,” we are not referring to species, but to structurally and functionally similar landscape units. This type of aggregation makes sense operationally because global-scale models must be parameterized to simulate processes within “grid cells” that contain multispecies canopies, and mixtures of canopies (e.g., mixed coniferous/deciduous forest). It is often important, though quite difficult, to have information that explicitly describes relative abundance of mixtures of vegetation types within a large grid cell used for global process studies. At continental to global scales, vegetation structural and biophysical parameters are related to vegetation type. LAI and fAPAR are perhaps the most important variables for terrestrial modeling studies because they are powerful indicators of potential canopy photosynthesis and transpiration. Similarly, LAI and canopy height are useful for quantifying aerodynamic resistance, and thus momentum transfer, within climate models. These parameters are often explicitly linked to vegetation cover type because their spatial and temporal patterns are not readily obtainable from global-level remote sensing. The decreasing and vegetation-type dependent sensitivity of NDVI to LAI and fAPAR is a principal reason for this difficulty (e.g., Asrar et al., 1984; van Leeuwan and Huete, 1996). Global-level LAI estimates have been made, however, using empirical and theoretical radiative transfer models (Sellers, 1987; Running et al., 1994b). Canopy biophysical parameters are a function of vegetation structural and biochemical characteristics and the quantity and quality of incoming radiation. These integral parameters are important for predicting both the radiative and chemical balance of ecosystems. Albedo is one of the most important climate system modeling variables; surface sensible and latent heat exchange are very sensitive to this parameter. Indeed, relatively high accuracy is required for albedo in order to predict large-scale climate patterns correctly (Graetz, 1991). Similarly, the absorption of photosynthetically active radiation by plant canopies drives photosynthesis (Psn), and thus is important for understanding net primary production (NPP5 Psn2plant respiration) and net ecosystem production (NEP5NPP2heterotrophic respiration). Ecosystem modeling efforts have shown that continental- and global-scale patterns of NPP can be simulated without considering PAR absorption because of the correlation between climate variables, nitrogen limitation, and carbon production within vegetation types (e.g., Schimel et al., 1997). Therefore, the possibility exists for an independent evaluation of global terrestrial models using remote sensing data, as well as an integration of biophysical and biogeo-

160

Asner et al.

chemical models with radiative transfer models (e.g., Sellers et al., 1997). The latter may be accomplished either empirically using the concept of production efficiency (e.g., Kumar and Monteith, 1981; Hunt et al., 1996), or analytically with physically based radiative transfer methods (Bouman, 1992). In contrast to the situation for ecosystem–biome scales, the distinction between canopy microstructure and macrostructure is not as meaningful for large-scale, coarse resolution applications. For example, modeled or remotely retrieved LAI for a 0.5830.58 grid cell has little or no meaning in the ecological sense, as real LAI is often highly variable even at the stand level. In this case, we are dealing with an “effective” parameter that optimally produces realistic results within the modeling context. Albedo and fAPAR are similarly scale-dependent, but perhaps are more easily treated as weighted averages because they are more linearly related to reflectance and vegetation indices, respectively. MVA METHODS FOR VEGETATION REMOTE SENSING Remote sensing methods that incorporate MVA information will improve the feasibility and accuracy of ecological parameter estimates at local to global scales. In this section, we outline several areas of MVA research that hold most promise for refining estimates of vegetation structure and function from optical remote sensing observations. While every possible development cannot be addressed here, our intent is to discuss those advances which have received an early research focus and have demonstrated promising results, or which have received a considerable level of planning and effort, and are thus likely to become operational in the EOS era. Ecosystem to Biome Scales Ecosystem- and biome-level studies of vegetation dynamics have spawned an entire area of remote sensing research focused on improving the spatial resolution of satellite and aircraft data. Spectral mixture analysis is a method by which image pixels are separated into components or endmembers (Borel and Gerstl, 1994; Smith et al., 1994). Depending on the spectral resolution of the data, these endmembers may represent land-cover types (e.g., forest, pastures, bare soils), material types (fresh leaf material, litter, soils), or even land-use characteristics (e.g., fire frequency, grazing intensity) (Roberts et al., 1993; Smith et al., 1994; DeFries et al., 1997; Wessman et al., 1997). The number and types of endmembers extracted via spectral unmixing are largely dependent on the uniqueness of each “endmember signature,” and increasing the information or degrees of freedom in each endmember signature can have a significant effect on the unmixing results (Smith et al., 1994; Wessman et al.,

1997). These traditionally spectral-based tools can be adapted to take advantage of the unique information contained in MVA measurements. Recent progress in unmixing techniques shows promise in incorporating both the spectral and angular dependence of each surface signature into algorithms for separating the land covers or surface materials within image pixels (Asner et al., 1997b). However, these methods are currently constrained to multisensor approaches since most multispectral and hyperspectral sensors do not contain a MVA capability. While these techniques are somewhat cumbersome due to the requirement for image co-registration and geometric and radiometric co-calibration, when MODIS and MISR become available, their combined spectral and angular resolution and co-registration accuracy will sharply improve this situation. Geometric–optical model inversions are another approach that will improve the classification and mapping accuracy of vegetation types at ecosystem to biome scales using MVA data. Geometric–optical models simulate the reflectance distribution of landscapes based primarily on crown geometries and shadowing (Li and Strahler, 1985; Li et al., 1995). Inversion of these models has provided estimates of crown dimensions and spacing from MVA remote sensing data, which could lead to improved classification of forest types and conditions (Strahler and Liang, 1994; Wu and Strahler, 1994). Since canopy macrostructural characteristics tend to be relatively stable elements within a vegetation class compared to LAI or fAPAR, which vary considerably at the plot level, retrieved macrostructural information will provide additional degrees of freedom for classifying vegetation types within regions. Moreover, large-scale estimates of crown size and spacing have important implications for ecosystem or habitat management, forestry, and landscape ecology. Physically based canopy BRDF model inversions are a very promising approach for retrieval of vegetation microstructural characteristics from MVA data. Since vegetation radiative transfer models simulate canopy-level photon interactions between leaves, stems, and soils, numerical inversion of these models has provided reasonable estimates of canopy structural and biophysical attributes (e.g., LAI, fAPAR) from MVA data (Goel and Thompson, 1984; Goel et al., 1984; Privette et al., 1994; Roujean and Breon, 1995; Braswell et al., 1996; Privette et al., 1996). The general approach is based on fitting, via an optimization routine, a set of modeled angularlydependent reflectances to actual measurements acquired from aircraft or spacecraft instruments. The set of parameters that minimizes the squared difference between the modeled and actual reflectance values (e.g., that produces the best fit) are considered the retrieved canopy microstructural parameters (Fig. 3). A major limitation to this approach lies in acquiring an adequate number of geometrically unique observations; that is, the number of

Multiangle Remote Sensing and Ecology

161

parameters). Likewise, overly complex models result in restrictively slow inversion times since the optimization routines iterate on forward simulations until modeled and measured reflectances most closely match. Linked to these considerations are issues of RT model parameter sensitivity and techniques for inverting the models [see Goel (1988) for details]. The development, intercomparison, and invertibility of existing and forthcoming canopy BRDF models remains a critical area of research. Model inversion methods have also been used to estimate soil surface attributes (Privette et al., 1995), but it is still too early to assess the accuracy and applicability of these approaches. Most ecosystems contain a significant amount of vegetation cover, and the invertibility of soil BRDF models must be further explored in conjunction with the canopy models (Jacquemoud et al., 1992; Privette et al., 1995). In arid and semiarid ecosystems, a focus on soil and litter radiative transfer may prove very useful.

Figure 3. The canopy radiative transfer (RT) model inversion process. A figure-of-merit function (example given) lies at the center of the inversion method and is used to minimize the difference between measured samples of the canopy BRDF (from satellite or aircraft sensors) and modeled samples using the same optical wavelengths and sun-view geometries. During the iteration process, model parameters such as leaf and stem area indices and leaf optical properties are continually adjusted. When the minimum of the figure-of-merit is reached, the leaf and canopy biophysical characteristics producing the minimum difference are said to be retrieved. The retrieved variables can then be used in forward integrations to estimate the fraction of PAR absorbed (fAPAR) by the canopy.

observations must exceed the number of free model parameters. Recent work has indicated several ways to constrain canopy RT model parameters while still allowing for robust physically based retrievals, such as through links between leaf optical properties in different spectral bands (Privette et al., 1996; Braswell et al., 1996; Asner et al., 1998). These constraints decrease the number of angular observations needed for accurate canopy parameter retrievals since there are fewer free parameters in the inversion. Another constraint on such inversions lies in the complexity of the canopy RT model used. If the model is too complex, there may be an excess of parameters, leading to a mathematically underdetermined situation (e.g., too few satellite observations and too many model

Continental to Global Scales At continental to global scales, methods for exploiting MVA and spectral information are similar to those used for higher resolution applications, but are more generalized. The models employed are necessarily more physically simplistic for two reasons: 1) Highly resolved structural details can neither be validated nor utilized at large spatial scales, where “effective” parameters have the most operational significance; and 2) most coarse resolution remote sensing data are limited in terms of the density of spectral and MVA sampling; that is, there are few degrees of freedom available for parameter estimation. Thus, empirical approaches using vegetation indices (VI) and physically based retrievals using relatively simple models are most valuable. Furthermore, both methods can explicitly account for vegetation anisotropy, either by normalizing a VI to nadir, or by using the bidirectional profile as independent information. The most common vegetation index employed for global-scale studies continues to be NDVI, though alternative indices have been proposed (Huete, 1988; Verstraete and Pinty, 1996). While NDVI is not a fundamental ecophysiological variable, there is a theoretical and experimental basis for the relationship between NDVI and a number of ecosystem parameters, including canopy photosynthetic efficiency, absorbed photosynthetically active radiation, stomatal conductance, and leaf area index (e.g., Tucker and Sellers, 1986; Sellers, 1987; Myneni and Williams, 1994). In turn, these quantities are linked to large-scale ecosystem net primary productivity (Running et al., 1994b; Field et al., 1995). The fundamental variables controlling NDVI are foliage density, chlorophyll content of leaves, and the highly variable effects of nonphotosynthetic vegetation (stems and senescent foliage). A number of factors not associated with land surface

162

Asner et al.

state can contribute to the NDVI signal, primarily atmospheric (Kaufman, 1989) and bidirectional (Cihlar et al., 1994) effects. Correction of atmospheric effects, which are highly anisotropic, are greatly facilitated by MVA observations. This radiometric correction represents one of the principal objectives of the MISR instrument (Borel and Gerstl, 1996), but its value is obviously not limited to NDVI alone. Similarly, land surface bidirectional effects can be reduced or removed from NDVI by deriving empirical relationships whose parameters depend on land surface type (Wu et al., 1995). These MVA-based corrections, in addition to those already performed on, for example, global AVHRR products (James and Kalluri, 1994), will increase the accuracy of NDVI, perhaps enough to investigate interannual changes in ecosystem processes. Correction of vegetation indices for BRDF effects will improve VI relationships (such as NDVI–fAPAR) for broad-scale applications, but these empirical relationships will vary with respect to external factors such as background/soil color, sun–sensor geometry, and surface litter (e.g., Huete, 1988; Goward and Huemmrich, 1992; van Leeuwen and Huete, 1996). While we realize that relationships between vegetation indices and vegetation characteristics are often spatially, temporally, and vegetation-type dependent, their application is computationally efficient and readily employable at global scales. Given how widely used VI methods have become within the ecological modeling community, any significant improvement upon them, such as through BRDF correction, will enhance their utility and accuracy for ecological research. As mentioned, canopy radiative transfer (RT) model inversion methods exploit the bidirectional properties of vegetation in order to estimate canopy structural properties (Goel, 1988). The RT model parameters are then forward integrated, accounting for the light environment (a function of latitude, day of year, etc.), to yield bulk biophysical properties such as fAPAR and albedo (Fig. 3). The feasibility of RT model inversions has been demonstrated using field-observed reflectances, usually with a dense sampling of the BRDF. A few successful retrievals have been performed using satellite data (e.g., Braswell et al., 1996; Privette et al., 1996). Hence, while the utility of multi-view angle data has been demonstrated, it is not clear how many independent degrees of freedom (related to how many parameters can be retrieved in the absence of noise) are contained in a land surface BRDF at large spatial scales. Information content in multiangle/ multispectral remote sensing data depends on observation conditions, atmospheric state, and land-cover type. How to appropriately balance these considerations for global-scale applications, given the specific sampling properties of, for example, MISR and MODIS data, is an important research topic. Finally, direct sampling of the BRDF, globally, using MISR will enable more immediate albedo estimates. These products will complement albedo estimated using

RT model inversions because the latter may be limited to regional studies due to its higher computational demand. With the advent of the EOS and POLDER instruments, an improved global land surface albedo product will be one of the first successes of the MVA remote sensing community. Links between MVA-Derived Information and Models Based on the research to date, MVA remote sensing approaches hold promise for improving the link between biogeochemical/land-surface biophysical models and timevariant vegetation and soil characteristics. The degree of integration will depend upon the spatial and temporal resolution and spatial extent of the specific modeling application. In general, MVA methods provide a way to move these models from a large dependency on tabulated data (e.g., look-up tables derived from country statistics) to quantitative estimates of critical ecological variables. Many biogeochemical process models, such as the Century Ecosystem Model (Parton et al., 1987; Schimel et al., 1994), are spatially parameterized using information on the extent and attributes of major vegetation functional groups (e.g., boreal forest, shrublands, grasslands) gathered from literature sources. For example, vegetation cover fraction, disturbance frequency (e.g., fire, logging), and LAI are parameterized by biome or region from statistics, and thus do not provide a means to link specific changes in canopy structure (e.g., resulting from human activities) to biogeochemical processes (Table 3). However, MVA methods such as canopy geometricoptical and radiative transfer model inversions and BRDFcorrected vegetation index relationships will improve the link between biogeochemical models and changes in land use, land cover and phenology. Some biogeochemical and land-surface biophysical models (e.g., Running et al., 1994; Field et al., 1995; Hunt et al., 1996; Sellers et al., 1996; 1997) are currently parameterized using a combination of look-up tables and remotely sensed VI relationships, and these models will be similarly improved using MVA methods (Table 3). Increased integration of remotely sensed information should improve the representation of surface heterogeneity in these models, which is largely lacking in global-level simulations. CONCLUSIONS Because vegetation canopies and atmospheric constituents are not isotropic scatterers of photons, multi-view angle observations of the Earth will significantly improve our ability to quantify vegetation structural and functional change over spatial scales ranging from individual ecosystems to the globe. Multispectral and hyperspectral methods have provided many important steps toward this goal; however, MVA methods have received relatively lit-

Multiangle Remote Sensing and Ecology

163

Table 3. Currently Available and Potential MVA-Derived Structural and Biophysical Variables for the Century Ecosystem Process Model (Parton et al., 1987; Schimel et al., 1994) and the SiB2 Land Surface Biophysics Model (Sellers et al., 1996; 1997)a Current Model Parameters Vegetation type Vegetation cover fraction Vegetation phenology Mean disturbance frequency b Leaf area index Canopy extinction coefficient Canopy greenness fraction Live vs. senescent biomass Woody vs. leaf biomass Canopy allometry Canopy gap fraction Canopy roughness length Leaf optical properties Leaf angle distribution fAPAR Albedo Soil surface texture

Century

SiB2

MVA Methods

LUT LUT LUT LUT LUT — — LUT LUT — — — — — NP — LUT

LUT LUT A — B B B — LUT LUT LUT LUT LUT LUT B B LUT

LUT1113 1–5 2,3 1,3,5 2,4 2,4 2,4 2 2 1 1 1 2 2 2,4 6 2

a Key: LUT5look-up table, NP5not currently parameterized but planned, —5not applicable to model, A5vegetation index (e.g., NDVI), B5empirical VI:biophysical parameter relationship, 15canopy geometric–optical model inversion, 25canopy radiative transfer model inversion, 35BRDF-corrected VI, 45BRDF-corrected empirical VI:biophysical parameter relationship, 55improved spectral mixture analysis, 65direct observation with some forward model extrapolation. b e.g., fire frequency.

tle attention outside of the theoretical development and exploration of canopy and atmospheric photon transport models. The lack of multi-view angle instrumentation has largely impeded the development of MVA applications in ecological, biogeochemical, and atmospheric research, but with the EOS and POLDER instruments, many of these constraints will be lifted, allowing MVA methods to achieve an operational level of use. The remote sensing needs of the ecological community are diverse, and dependent upon the spatial and temporal scales of individual applications. We believe that land surface reflectance anisotropy (or the “bidirectional profile”) fundamentally represents an independent and potentially important source of information for increasing the accuracy of vegetation parameter estimates at all scales. There is a growing body of evidence for this conclusion, but it is also true that the degree of utility of many MVA algorithms may not be fully known until high quality data are available from the EOS and/or POLDER instruments. While MVA methods have shown early promise for elevating vegetation remote sensing to a more quantitative level due to their emphasis on physical as opposed to empirical modeling, the algorithms required to employ these methods are likely beyond the research scope of most ecologists. Most ecologists’ interests lie in the final product—the ecologically beneficial data that can be derived from these methods to improve biogeochemical models, better manage ecosystems, increase the value of

land surface parameterizations in climate models, and advance a multitude of other applications. It is therefore a critically important time to foster an increased level of communication and collaboration between remote sensing and ecological scientists as we move into the EOS era. We hope that this communication provides an initial framework from which to pursue this interaction. We thank Diane Wickland, Don Deering, Jeff Privette and attendees of the NASA Workshop on Multi-angular Remote Sensing for Environmental Applications, held at the University of Maryland, College Park in January 1997. The impetus to write this article was provided through their questions to the ecological research community. We also thank Jeff Privette for providing the BRDF principal plane figure and for his comments on the manuscript. This work was supported in part by NASA Innovative Research Grant NAGW-4689, NASA Interdisciplinary Science Grants NAGW-2662 and NAGW-2669, and the NASA Earth System Science Fellowship Program. The National Center for Atmospheric Research is sponsored by the National Science Foundation.

REFERENCES Allen, T. F. H., and Starr, T. B. (1982), Hierarchy: Perspectives for Ecological Complexity, University of Chicago Press, Chicago, 310 pp. Archer, S., Schimel, D. S., and Holland, E. A. (1995), Mechanisms of shrubland expansion: land use, climate or CO2. Clim. Change 29:91–99.

164

Asner et al.

Asner, G. P., Seastedt, T. R., and Townsend, A. R. (1997a), The decoupling of terrestrial carbon and nitrogen cycles. BioScience 47:226–234. Asner, G. P., Wessman, C. A., and Privette, J. L. (1997b), Unmixing the directional reflectances of AVHRR sub-pixel landcovers. IEEE Trans. Geosci. Remote Sens. 35:868–878. Asner, G. P., Wessman, C. A., Schimel, D. S., and Archer, S. (in press), Variability in leaf and litter optical properties: implications for BRDF model inversions using AVHRR, MODIS, and MISR. Remote Sens. Environ. Asrar, G., Fuchs, M., Kanemasu, E. T., and Yoshida, M. (1984), Estimating absorbed photosynthetic radiation and leaf area index from spectral reflectance in wheat. Agron. J. 76: 300–306. Barnsley, M. J., Strahler, A. H., Morris, K. P., and Muller, J.-P. (1994), Sampling the surface bidirectional reflectance distribution function: 1. Evaluation of current and future satellite sensors. Remote Sens. Rev. 8:271–311. Borel, C. C., and Gerstl, S. A. W. (1994), Nonlinear spectral mixing models for vegetative and soil surfaces. Remote Sens. Environ. 47:403–416. Borel, C. C., and Gerstl, S. A. W. (1996), Algorithms to compute the top-of-the-atmosphere albedo for clear-sky conditions from multi-angle MISR observations. In Proc. Int. Geosci. Remote Sens. Symp., Lincoln, NE, 27–31 May. Bouman, B. A. M. (1992), Linking physical remote sensing models with crop growth simulation models applied for sugar beet. Int. J. Remote Sens. 13:2565–2581. Braswell, B. H., Schimel, D. S., Privette, J. L., et al. (1996), Extracting ecological and biophysical information from AVHRR optical data: an integrated algorithm based on inverse modeling. J. Geophys. Res. 101:23,335–23,345. Chen, J. M., and Cihlar, J. (1995), Plant canopy gap-size analysis theory for improving optical measurements of leaf area index. Appl. Opt. 34:6211–6220. Ciais, P., Tans, P. P., Trolier, M., White, J. W. C., and Francey, R. J. (1995), A large Northern Hemisphere terrestrial CO2 sink indicated by the 13C/12C ratio of atmospheric CO2. Science 269:1098–1102. Cihlar, J., Manak, D., and Voisin, N. (1994), AVHRR bidirectional effects and compositing. Remote Sens. Environ. 48: 77–88. Cole, W. G., and Lorimer, C. G. (1994), Predicting tree growth from crown variables in managed northern hardwood stands. For. Ecol. Manage. 67:159–168. DeFries, R. S., Field, C. B., Fung, I., et al. (1995), Mapping the land surface for global atmosphere-biosphere models: toward continuous distributions of vegetation’s functional properties. J. Geophys. Res. 100:20,867–20,882. DeFries, R. S., Hansen, M., Steininger, M., Dubayah, R., Sohlberg, R., and Townshend, J. (1997), Subpixel forest cover in Central Africa from multisensor, multitemporal data. Remote Sens. Environ. 60:228–248. Field, C. B., Randerson, J. T., and Malmstrom, C. M. (1995), Global net primary production: combining ecology and remote sensing. Remote Sens. Environ. 51:74–88. Goel, N. S. (1988), Models of vegetation canopy reflectance and their use in estimation of biophysical parameters from reflectance data. Remote Sens. Rev. 4:1–212. Goel, N. S., and Thompson, R. L. (1984), Inversion of canopy reflectance models for estimating agronomic variables. V.

Estimation of leaf area index and average leaf angle using measured canopy reflectances. Remote Sens. Environ. 16: 69–85. Goel, N. S., Henderson, K. E., and Pitts, D. E. (1984), Estimation of leaf area index from bidirectional spectral reflectance data by inverting a canopy reflectance model. In Proc. Machine Process. Remote Sens. Data Symp., pp. 339–347. Goward, S. N., and Huemmrich, K. F. (1992), Vegetation canopy PAR absorptance and the normalized difference vegetation index: an assessment using the SAIL model. Remote Sens. Environ. 39:119–140. Graetz, R. D. (1991), The nature and significance of the feedback of changes in terrestrial vegetation on global atmospheric and climatic change. Clim. Change 18:147–173. Huete, A. R. (1988), A soil-adjusted vegetation index (SAVI). Remote Sens. Environ. 25:295–309. Hunt, E. R., Piper, S. C., Nemani, R., Keeling, C. D., Otto, R. D., and Running, S. W. (1996), Global net carbon exchange and intra-annual atmospheric CO2 concentrations predicted by an ecosystem process model and three-dimensional atmospheric transport model. Global Biogeochem. Cycles 10:431–456. Jacquemoud, S., Baret, F., and Hanocq, J. F. (1992), Modeling spectral and bidirectional soil reflectance. Remote Sens. Environ. 41:123–132. Jacquemoud, S., Baret, F., Andrieu, B., Danson, F. M., and Jaggard, K. (1995), Extraction of vegetation biophysical parameters by inversion of the PROSPECT1SAIL models on sugar beet canopy reflectance data. Application to TM and AVIRIS sensors. Remote Sens. Environ. 52:163–172. James, M. E., and Kalluri, S. N. V. (1994), The Pathfinder AVHRR land data set: an improved coarse resolution data set for terrestrial monitoring. Int. J. Remote Sens. 15:3347– 3363. Kaufman, Y. J. (1989), The atmospheric effect on remote sensing and its correction. In Theory and Applications of Optical Remote Sensing (G. Asrar, Ed.), Wiley, New York, pp. 336–428. Kumar, M., and Monteith, J. L. (1981), Remote sensing of crop growth. In Plants in the Daylight Spectrum (H. Smith, Ed.), Academic, New York, pp. 134–144. Li, X., and Strahler, A. H. (1985), Geometric-optical modeling of a coniferous forest canopy. IEEE Trans. Geosci. Remote Sens. 23:207–221. Li, X., Strahler, A. H., and Woodcock, C. E. (1995), A hybrid geometric optical-radiative transfer approach for modeling albedo and directional reflectance of discontinuous canopies. IEEE Trans. Geosci. Remote Sens. 33:466–480. Loveland, T. R., Merchant, J. W., and Ohlen, D. (1991), Development of a land-cover characteristics database for the conterminous U.S. Photogramm. Eng. Remote Sens. 57:1453– 1464. Matthews, E. (1983), Global vegetation and land-use: new high-resolution databases for climate studies. J. Appl. Meteorol. 22:474–487. Milhailovic, D. T., Pielke, R. A., and Rajkovic, B. (1993), A resistance representation of schemes for evaporation from bare and partly plant-covered surfaces for use in atmospheric models. J. Appl. Meteorol. 32:1038–1047. Myneni, R. B., and Asrar, G. (1993), Radiative transfer in

Multiangle Remote Sensing and Ecology

three-dimensional atmosphere–vegetation media. J. Quant. Spectrosc. Radiat. Transfer 49:585–598. Myneni, R. B., and Williams, D. L. (1994), On the relationship between fAPAR and NDVI. Remote Sens. Environ. 49: 200–209. Myneni, R. B., Ross, J., and Asrar, G. (1989), A review on the theory of photon transport in leaf canopies. Agric. For. Meteorol. 45:1–153. Olson, J. S., Watts, J. A., and Allison, L. J. (1983), Carbon in live vegetation of major world ecosystems, ORNL-5862, Environmental Sciences Division Publication No. 1997, Oak Ridge National Laboratory, Oak Ridge, TN. O’Neill, R. V., DeAngelis, D. L., Waide, J. B., and Allen, T. F. H. (1986), A Hierarchical Concept of Ecosystems, Princeton University Press, Princeton, NJ. Parton, W. J., Schimel, D. S., Cole, C. V., and Ojima, D. S. (1987), Analysis of factors controlling soil organic matter levels in Great Plains grasslands. Soil Sci. Soc. Am J. 51:1173–1179. Pollard, D., and Thompson, S. L. (1995), Use of a land-surfacetransfer scheme (LSX) in a global climate model: the response the doubling stomatal resistance. Global Planet. Change 10:129–162. Privette, J. L., Myneni, R. B., Tucker, C. J., and Emery, W. J. (1994), Invertibility of a 1-D discrete ordinates canopy reflectance model. Remote Sens. Environ. 48:89–105. Privette, J. L., Myneni, R. B., and Pinty, B. (1995), Inversion of a soil bidirectional reflectance model for use with vegetation reflectance models. J. Geophys. Res. 100:25,497–25,507. Privette, J. L., Emery, W. J., and Schimel, D. S. (1996), Inversion of a vegetation reflectance model with NOAA AVHRR data. Remote Sens. Environ. 58:187–200. Rastetter, E. B., McKane, R. B., Shaver, G. R., and Melillo, J. M. (1992), Changes in carbon storage by terrestrial ecosystems: how carbon-to-nitrogen interactions restrict responses to CO2 and temperature. Water Air Soil Pollut. 64:327–344. Roberts, D. A., Smith, M. O., and Adams, J. B. (1993), Green vegetation, non-photosynthetic vegetation, and soils in AVIRIS data. Remote Sens. Environ. 44:255–269. Ross, J. K. (1981), The Radiation Regime and Architecture of Plant Stands, Kluwer Boston, Hingham, MA. Roujean, J., and Breon, F. (1995), Estimating PAR absorbed by vegetation from bidirectional reflectance measurements. Remote Sens. Environ. 51:375–384. Running, S. W., Justice, C. O., Salomonson, V., et al. (1994a), Terrestrial remote sensing science and algorithms planned for EOS/MODIS. Int. J. Remote Sens. 15:3587–3620. Running, S. W., Loveland, T. R., and Pierce, L. L. (1994b), A vegetation classification logic based on remote sensing for use in global biogeochemical models. Ambio 23:77–89. Schimel, D. S. (1995a), Terrestrial biogeochemical cycles: global estimates with remote sensing. Remote Sens. Environ. 51:49–56. Schimel, D. S. (1995b), Terrestrial ecosystems and the carbon cycle. Global Change Biol. 1:77–91. Schimel, D. S., Braswell, B. H., Holland, E. A., et al. (1994), Climatic, edaphic, and biotic controls over storage and turnover of carbon in soils. Global Biogeochem. Cycles 8: 279–293. Schimel, D. S., Braswell, B. H., McKeown, R., et al. (1996),

165

Climate and nitrogen controls on the geography and time scales of terrestrial biogeochemical cycling. Global Biogeochem. Cycles 10:677–692. Schimel, D. S., VEMAP Participants, and Braswell, B. H. (1997), Continental scale variability in ecosystem processes: models, data, and the role of disturbance. Ecol. Monogr. 67:251–271. Sellers, P. J. (1987), Canopy reflectance, photosynthesis, and transpiration II: the role of biophysics in the linearity of their interdependence. Int. J. Remote Sens. 6:1335–1372. Sellers, P. J., Randall, D. A., Collatz, G. J., et al. (1996), A revised land surface parameterization (SiB2) for atmospheric GCMs part I: model formulation. J. Clim. 9: 676–705. Sellers, P. J., Dickinson, R. E., Randall, D. A., et al. (1997), Modeling the exchanges of energy, water, and carbon between continents and the atmosphere. Science 275:502–509. Smith, M. O., Adams, J. B., and Sabol, D. E. (1994), Spectral mixture analysis—new strategies for the analysis of multispectral data. In Imaging Spectrometry—A Tool for Environmental Observations (J. Hill and J. Megier, Eds.), Kluwer Academic, Dordrecht, The Netherlands, pp. 125– 144. Strahler, A. H., and Liang, S. (1994), Retrieval of surface BRDF from multiangle remotely sensed data. Remote Sens. Environ. 50:18–27. Sturtevant, B. R., Bissonette, J. A., and Long, J. N. (1996), Temporal and spatial dynamics of boreal forest structure in western Newfoundland: silvicultural implications for marten habitat management. Forest Ecol. Manage. 87:13–21. Townsend, A. R., Braswell, B. H., Holland, E. A., and Penner, J. E. (1996), Spatial and temporal patterns in terrestrial carbon storage due to deposition of anthropogenic nitrogen. Ecol. Appl. 6:806–814. Tucker, C. J., and Sellers, P. J. (1986), Satellite remote sensing of primary production. Int. J. Remote Sens. 7:1395–1416. van Leeuwen, W. J. D., and Huete, A. R. (1996), Effects of standing litter on the biophysical interpretation of plant canopies with spectral indices. Remote Sens. Environ. 55: 123–134. Verstraete, M. M., and Pinty, B. (1996), Designing optimal spectral indexes for remote sensing applications. IEEE Trans. Geosci. Remote Sens. 34:1254–1264. Washington, W. M., and Mehl, G. A. (1996), High-latitude climate change in a global coupled ocean-atmosphere-sea ice model with increased atmospheric CO2. J. Geophys. Res. 101:12,795–12,932. Wessman, C. A. (1992), Spatial scales and global change: bridging the gap from plots to GCM grid cells. Annu. Rev. Ecol. Systemat. 23:175–200. Wessman, C. A., Bateson, C. A., and Benning, T. (1997), Detecting fire and grazing patterns in the Konza tallgrass prairie using spectral mixture analysis. Ecol. Appl. 7:493–511. Wu, A., Li, Z., and Cihlar, J. (1995), Effects of land cover type on Advanced Very High Resolution Radiometer bidirectional reflectances: analysis and removal. J. Geophys. Res. 100:9179–9192. Wu, Y., and Strahler, S. H. (1994), Remote estimation of crown size, stand density, and biomass on the Oregon transect. Ecol. Appl. 4:299–310.