|
|
|
One challenge in applying satellite or aircraft images to ecological studies lies in relating spectral and spatial information in an image to vegetation patterns and processes on the ground. The difficulties of coregistering ground data with remote sensing imagery are often enhanced by inaccessibility of field sites, the inherent complexities of vegetation structure, and the necessity of comparing measurements gathered at multiple spatial and temporal scales. Because of these challenges, remote sensing studies typically follow a "top-down" approach, trying to infer biological information from a spectral image with little knowledge of the actual "biological content" of the image. For example, vegetation indices calculated from images are often used in conjunction with ecosystem process models to estimate patterns of CO2 flux or productivity without ground confirmation of vegetation status or flux rates during the time of the overflight (e.g., Running and Nemani, 1988). As a result, applications of remote sensing to ecological questions often involve simplifying assumptions about the biological content that could be significantly in error, particularly when multiple vegetation types are involved.
Greater confidence in our understanding of vegetation function from remote sensing can be obtained by a systematic program of ground sampling during a satellite or aircraft overflight. Ideally, ground sampling would involve simultaneous measurements of canopy biochemistry, physiology, structure, and productivity. Many of these indicators of vegetation function can now be measured and related to spectral properties from the ground with new, portable, commercially available instruments.
A primary goal of this study was to obtain ground data that would allow
biological interpretation of AVIRIS images. We compared AVIRIS data from
the 15 May 1991 overflight of the Jasper Ridge Biological Preserve to available
maps of vegetation and topography, and conducted a detailed ground sampling
campaign on a 9-ha portion of the preserve. This region was dominated by
a relatively uniform annual grassland with scattered "islands" of trees
and shrubs, providing a simple system for evaluating links between remote
sensing data and vegetation function. These data allowed us to test the
hypothesis that there is a "functional convergence" between canopy biochemistry,
physiology, and structure; to the extent that these variables scale together
in time and space, it may be possible to infer physiological function from
estimates of canopy structure obtained by remote sensing (Field, 1991).
Optical estimates of leaf area index (LAI) and fractional intercepted photosynthetically active radiation (IPAR) were conducted at all 225 flagged locations with a"plant canopy analyzer" (Model LAI-2000, LI-COR, Inc., Lincoln, Nebraska) and a"sunfleck ceptometer" (0.8-m Model, Decagon Devices, Pullman, Washington), respectively. Twenty-six of the 225 locations (primarily in 1 ha at the north-central portion of the 9-ha region), containing both soil types and exhibiting a wide range of vegetation productivity, were chosen for intensive study, including estimates of canopy chemistry (chlorophyll and nitrogen), biomass, surface temperature, and spectral reflectance (Table 2). Trees and shrubs were not examined in this intensive study.
Preliminary tests with the plant canopy analyzer indicated a 50% midday decline in LAI estimates due to increased canopy illumination associated with increased solar elevation. This effect results from the increasing percentage of canopy receiving direct illumination as solar elevation increases (Welles and Norman, 1991). Consequently, the LAI measurements reported here were conducted only in the early morning, with canopy shade provided by a 3-m2 black cloth stretched across a PVC frame. This precaution virtually eliminated artifacts associated with changing sun angle, and yielded good agreement with harvested LAI (Fig. 2). LAI values were confirmed with optical measurements on a subset of harvest samples using a light table and video camera attached to a leaf area meter (Model AMS, Delta T Devices, Cambridge, U.K.).
Fractional intercepted photosynthetically active radiation (IPAR) was
estimated with the "sunfleck ceptometer" from one above-canopy and one
below-canopy measurement at each sampling location. Fractional intercepted
photosynthetically active radiation (IPAR) was calculated as
|
(1)
|
On the ground, canopy spectral reflectance was obtained with a portable spectroradiometer mounted with a 15° FOV fore-optic (Model SE590 with Detector Model CE390WB-R, Spectron Engineering, Inc., Denver, Colorado), using pressed halon as a reflectance standard (Weidner and Hsia, 1981). This spectroradiometer had a 10-nm bandwidth at half-maximum response, with an average band-to-band distance of 2.97 nm and a spectral range from 368.4 nm to 1113.7 nm. The detector head was attached to a self-levelling mount on a large tripod and boom, approximately 4 m above the ground, yielding a single scan with a field-of-view approximately 1-m in diameter at each sampling location.
The Normalized Difference Vegetation Index (NDVI) was calculated from
spectral reflectance according to the following general equation:
|
(2)
|
For comparative purposes, a broad-band NDVI was also calculated from Spectron spectral reflectance by simulating AVHRR Bands 1 (red) and 2 (NIR) and integrating over these bands. Comparative statistics using both narrow- and broad-band NDVI are provided (Table 3), and the implications of the two approaches are also discussed.
At each of the 26 flagged locations chosen for intensive study, a 0.5-m diameter canopy circle was harvested for determination of fresh weight, dry weight, leaf area, foliar nitrogen content, and foliar chlorophyll content. Harvested plant material from eight of these locations was manually separated into visibly green and nongreen components to estimate "greenness," which was the ratio of green area (or biomass) to total area (or biomass). Fractional "greenness factors" were then used as empirical coefficients to correct LAI, biomass, and fractional IPAR for greenness (Table 2). Chlorophyll was estimated spectrophotometrically on 80:20 (v:v) acetone: water extracts (Porra et al., 1989), and total nitrogen was estimated using a micro-Keldahl technique (Isaac and Johnson, 1976). Unless otherwise noted, biomass, leaf area, chlorophyll and nitrogen content were expressed on a ground-area basis for comparison with spectroradiometric measurements. Chlorophyll and nitrogen concentrations were also expressed on a leaf mass and leaf area basis for comparison with ground-based estimates (Table 3).
Surface temperature was measured between thirty and one hundred minutes
after solar noon on 15 May with a 2° FOV infrared thermometer (Model
110C Everest Interscience Inc., Tustin, California) held in a nadir orientation
from the top of a stepladder (2-3 m above the surface). An average surface
temperature was computed from four readings at each of 25 locations.
The AVIRIS data were converted to reflectance using an empirical line
calibration (Roberts et al., 1985; Conel, 1990; Elvidge and Portigal, 1990),
which was modified to remove the requirement of a priori knowledge
of the calibration targets (see steps 1-3 in Roberts, 1991). Three calibration
targets were selected from the image: water from Searsville Lake, a medium
reflectance soil, and a high reflectance soil. Average encoded radiance
(DN) for each target was determined using 4-9 pixels extracted from the
image. Encoded target radiance was then regressed against a library consisting
of 464 reflectance spectra. From this library, a subset of three spectra
(one water and two soils) were selected based on an analysis of R2
values as a function of wavelength and on the shapes of the residual spectra.
The water spectrum had been measured in the field at Searsville Lake using
a portable spectroradiometer (Model SE590, Spectron Engineering) with a
Spectralon standard (Labsphere, Inc., North Sutton, New Hampshire). The
two soil spectra had been measured in the laboratory using a spectrophotometer
(Model DK2A, Beckman, Inc., Fullerton, California) and halon as a standard.
All three reflectance spectra were converted to absolute reflectance using
a National Bureau of Standards halon spectrum convolved to the AVIRIS wavelengths
of the May scene. The final correction of AVIRIS data to reflectance from
uncalibrated radiance (DN) employed the following equations determined
from the modified empirical line calibration:
|
(3)
|
|
(4)
|
Spectral mixture analysis can be used to illustrate biologically significant patterns in multispectral images (Smith et al., 1990b). The basic techniques of spectral mixture analysis involve two steps: 1) the selection of spectra from the image, providing "image endmembers" that represent the most significant sources of spectral variability in the image, and 2) the "alignment" of these image endmembers to laboratory or field-measured spectra, providing "reference endmembers" that represent more fundamental components. Image endmembers typically are mixtures, although they most likely represent the "purest" spectra in the image and correspond to dominant components in the field (e.g., green vegetation, nonphotosynthetic vegetation, bare rock or soils, and shade). Reference endmembers corresponding to the image endmembers typically include green leaves, nonphotosynthetic plant materials (litter, stems, etc), rock or soil, and shade. The number of image endmembers selected is the minimum number required to adequately describe the scene variability, and characteristically accounts for 97-99% of the total spectral variation. Endmembers must be spectrally distinct and cannot be reproduced by linear combinations of the other endmembers in the scene. The fractional endmember combination is determined for each pixel in the image by least squares regression. Endmember fractions can be expressed in terms of image endmembers or reference endmembers. In this article, they are expressed as reference endmember fractions. The theory, methods, and application of spectral mixture analysis has been described extensively elsewhere (Adams et al., 1986; Huete, 1986; Roberts et al., 1990, 1993; Smith et al., 1990a,b; Ustin et al., 1993).
In our study, spectral unmixing was accomplished on a subset of 178 bands from an AVIRIS scene (910522). A three-endmember model was found to represent 98.3% of the image's spectral variation, and identified the following unique endmembers: green vegetation (GV, library ID Sasp0007), nonphotosynthetic vegetation (NPV, library ID Quag0016), and photometric shade. Reference endmembers were selected from a library consisting of 464 laboratory-measured spectra of plant and soil materials (most collected at Jasper Ridge), categorized as nonphotosynthetic vegetation (58), green leaves (141), or soils/rock (265). The shade reference endmember was obtained from the same field-measured water spectrum used in the calibration. The selected reference endmembers are shown in Figure 1. Additional details regarding the calibration and selection of endmembers can be found in Roberts et al. (1993).
To examine spatial patterns, provide data compression, and facilitate
comparison of ground sampling and AVIRIS imagery, a narrow-band normalized
difference vegetation index (NDVI) was produced from AVIRIS Band 29 at
677 nm (red) and Band 49 at 833 nm (NIR) [see Eq. (2)].
|
(5)
|
Statistics
Variograms illustrating sample variance as a function of spatial scale
(Curran, 1988; Rossi et al., 1992) in both ground and AVIRIS data were
used to evaluate the spatial resolution of the AVIRIS pixel. Semivariance
was calculated according to the following equation:
 |
(6)
|
is the estimate of semivariance,
h is the lag (between-sample distance), z is the measured
parameter, x is the pixel or sample location (1 to n), and
m is the number of pairs of pixels or samples locations (Curran,
1988). Relationships between ground samples and NDVI (Table
3) were examined with Pearson correlation matrices (Systat v. 5.1 for
MacIntosh Computers, Systat, Inc., Evanston, Illinois).
Scatter in the NDVI-chlorophyll and the NDVI-nitrogen relationships was partly attributable to two factors: 1) the different spatial scales of wet chemical assays and reflectance measurements (Table 2), and 2) the difficulties of obtaining accurate, area-based biochemical estimates using wet chemical assays subsampled from stored field samples. Methods for comparing remote sensing imagery to canopy biochemical content deserve further study. Sample size and speed could be improved, and sampling error reduced, by substituting properly calibrated optical assays for wet chemistry. Ideally, these optical measurements would be field-portable, carefully calibrated against lab measurements for each vegetation type, and of a spectral resolution comparable to AVIRIS. In this way, extensive ground sampling of narrow-band chemical features could be more directly compared to possible chemical signals in AVIRIS images.
The scatter in the comparison of NDVI to canopy structure estimates can largely be explained by the varying proportions of green and nongreen vegetation components. Correcting LAI, biomass, and fractional IPAR estimates by an empirical canopy "greenness" factor led to significant improvements in these relationships (Fig. 2C, D, E, Table 3). Expressing chlorophyll or nitrogen concentration on a ground area basis provided alternate estimates of canopy greenness (as viewed from above). However, when expressed on a leaf area or leaf mass basis, chlorophyll and nitrogen concentration yielded poor correlations with NDVI (Table 3). Clearly, leaf-level concentrations of chlorophyll or nitrogen are not the primary factors driving variation in NDVI. However, canopy-level greenness (per unit ground area, as viewed from above) appeared to be a primary variable driving NDVI. In this study, canopy greenness was accurately expressed as green biomass, green LAI, total canopy chlorophyll or total canopy nitrogen, and our results indicate that NDVI would be a useful predictor of any of these expressions of canopy greenness in this grassland (see Table 4 for regression statistics). Spectral mixture analysis also provided estimates of green and non-green canopy components and presented an elegant way of remotely detecting green and nongreen canopy fractions (see below). The relationship between NDVI and surface temperature can easily be explained by the effect of green canopy structure (NDVI) on latent heat exchange (Nemani and Running, 1989; Running, 1990).
A close relationship between canopy structure and physiology in the
grassland was illustrated by the results of the concurrent eddy correlation
study. In the grassland, ground-based NDVI measurements yielded a linear
relationship with maximum daily CO2 flux measured by eddy correlation
at different times of the growing season on two grassland types (Fig.
3A). This relationship differed sharply with that from adjacent oak
woodland and chaparral communities, which were largely composed of evergreen
species. For example, in Quercus agrifolia, a dominant, evergreen
species of the adjacent woodland, NDVI completely failed to capture seasonal
changes in photosynthetic rates (Fig. 3B).
Similar results have been found for several other evergreen species at
this site (Gamon et al., in preparation).
Relative to the evergreen vegetation, the grasslands had a higher fraction
of senescing, nonphotosynthetic canopy material as indicated by the predominance
of red shades. This is consistent with the early spring senescence of herbaceous
species (supported by the high percentage of nongreen foliage sampled on
the ground during the overflight; see Fig. 2).
The low-productivity serpentine grassland is clearly visible as an anomalous
purple band due to combined effects of early vegetation senescence, high
soil visibility, and relatively dark soil. In our model, soil was not included
as a separate endmember. Thus, when bare soil was abundant, it partitioned
primarily into the most spectrally similar endmember, nonphotosynthetic
vegetation. Furthermore, a decrease in soil reflectance could not be distinguished
from shading of a higher reflectance soil. Thus, the abundance of bare,
dark soil in the serpentine grassland yielded a high "nonphotosynthetic
vegetation" fraction (red) with an artificially high "shade" (blue) component.
If desired, soil and nonphotosynthetic vegetation could be more clearly
distinguished with AVIRIS data through the use of narrow-band lignin and
cellulose features present in the nonphotosynthetic vegetation fraction
(Roberts et al., 1993). The addition of a second, darker soil endmember
to the mixture analysis would reduce the "shade" contribution in the serpentine
grassland. However, based on the relative proportions of three endmembers
alone, spectral mixture analysis clearly distinguished the different vegetation
types.
Coregistration of the 9-ha images was complicated by slightly different pixel sizes, different azimuths, and the different sampling schemes employed by AVIRIS and ground measurements. Adjacent AVIRIS pixels are nominally 17 m from center to center, and have approximately 3 m of overlap, leading to a 20-m instantaneous FOV (Vane, 1987; Robert Green, personal communication). Evaluation of known distances in the 15 May AVIRIS image indicated an actual pixel size of approximately 15.8 m. Ground sampling more closely approximated "point" samples without any overlap, spaced 20 m apart. In the LAI image derived from ground measurements (Fig. 6C), the positions of large landscape features (e.g., trees) could be nearly 20-m off their true location due to the point-sampling technique. In heterogeneous regions, unusual scene components (e.g., the road or shrubs) were often over- or underestimated by the point-intercept sampling. Thus, areas with patchy "islands" of woodland or chaparral vegetation, such as the eastern part or the northwestern corner of the 9-ha region, produced local areas of relatively poor correspondence between AVIRIS and ground images. In two locations (marked by two red spots in the right half of Fig. 6C), ground sampling points coincided with the road, resulting in local underestimates of average LAI. These factors made exact image coregistration difficult.
Although the images in Figure 6 are probably mislocated by one or two pixels, the spatial resolution of the AVIRIS sensor appeared to capture the major patterns of vegetation productivity in this grassland as measured on the ground.
Geostatistical techniques incorporating both AVIRIS and ground data
provided a more objective evaluation of the AVIRIS spatial resolution and
allowed statistical comparison of spatial patterns in AVIRIS and ground
data. At a 20-m between-sample distance ("lag"), the semivariance of LAI
(measured on the ground) and NDVI (calculated from AVIRIS data) was essentially
at the minimal or "nugget" value (Fig. 7).
At larger distances, semivariance increased, suggesting that larger sampling
lags (analogous to larger pixel sizes characteristic of TM or AVHRR data)
would miss the fine-scale patterns in vegetation productivity illustrated
in Figure 6. Thus, the approximately 20-m
AVIRIS pixel size appeared to be appropriate for studies of productivity
in this particular grassland ecosystem, supporting the more subjective
evaluations based on visual comparison of AVIRIS and ground images.
In simple systems such as the Jasper Ridge grassland, structurally based indices (NDVI or GVF) appear to be sufficient to capture seasonal and spatial patterns of productivity. These results appear to support the widespread application of vegetation indices in large scale estimates of photosynthetic fluxes (e.g., Tucker et al., 1986; Fung et al., 1987). However, the poor sensitivity of NDVI to seasonally changing photosynthetic fluxes in evergreens (Fig. 3B) indicates that NDVI may be an insufficient indicator of vegetation-atmosphere fluxes in many cases, and that different ecosystems may exhibit different degrees of functional convergence. Clearly, grassland and evergreen ecosystems follow different sets of "ecophysiological rules," even in immediately adjacent locations of identical soil type and similar microclimate.
In conjunction with carefully planned ground measurements, AVIRIS imagery
could be applied to studies of functional differences between these contrasting
vegetation types. Further development of spectral mixture analysis and
new narrow-band indices using high spectral resolution data could provide
several advantages over NDVI. In addition to GVF, spectral mixture analysis
provides estimates of other endmember fractions such as nonphotosynthetic
vegetation and shade, which can add new dimensions to studies of vegetation
function. For example, shade can indicate canopy architectural features,
and nonphotosynthetic vegetation can be used to evaluate nongreen structural
components or senescing vegetation. Furthermore, important biological processes
might be detectable as changes in specific narrow-band features present
in residual spectra. For example, lignin and cellulose signatures are detectable
in AVIRIS spectra (Roberts et al., 1993), and might yield ecologically
important indicators of nutrient turnover (Aber et al., 1989). Subtle changes
in the content of plant pigments, including chlorophylls, carotenes, xanthophylls,
and flavonoids, might be detectable in residual spectra or in new narrow-band
indices. Many of these pigments have important functions in PAR absorption
or photoprotection under stress (Goodwin, 1988; Young and Britton, 1990),
and thus provide optical indicators of CO2 flux. Diurnal changes
in xanthophyll pigment content are detectable in intact canopies with narrow-band
reflectance and have been correlated with CO2 flux (Gamon et
al., 1992). Mixture models, residual spectral analysis, and new narrow-band
indices might be particularly fruitful in evergreens and other woody communities
that are relatively inaccessible by conventional broadband techniques,
and the spatial and spectral resolution of AVIRIS might allow new applications
of these techniques. We are currently evaluating the relative merits of
spectral mixture analysis, NDVI, and new narrow-band indices in more detail.
We wish to thank N. Chiariello and the Jasper Ridge Biological Preserve
for the loan of vegetation maps, topographic maps, aerial photographs and
the Spectron spectroradiometer, G. Joel, B. Mortimer, R. Rousseau, M Holbrook,
and M Lerdau for field assistance, and J. Penuelas for helpful comments
on the manuscript. This work was supported by a grant from the Mellon Foundation
to the Carnegie Institution of Washington, the CNR (Italian Research Council,
RV), the NASA EOS program including an IDS grant and Contracts NAS5-31359,
NAS53076, and NAS53174 (SLU). Research at the University of Washington
(DAR) was supported by the W. M. Keck Foundation and by NASA Grants NAGW
1319 and EOSRAM NAGW-2652.
Adams, J. B., Smith, M. O., and Johnson, P. E. (1986), Spectral mixture modeling: a new analysis of rock and soil types at the Viking Lander I site, J. Geophys. Res. 91(B8):8098-8112.
Baldocchi, D. D., Hicks, B. B., and Meyers, T. P. (1988), Measuring biosphere-atmosphere exchanges of biologically related gases with micrometeorological methods, Ecology 69:1331-1340.
Bloom, A. J., Chapin, F. S., III, and Mooney, H. A. (1985), Resource limitation in plants - an economic analogy, Annul Rev. Ecol. Syst. 16:363-392.
Conel, J. E. (1990), Determination of surface reflectance and estimates of atmospheric optical depth and single scattering albedo from Landsat Thematic Mapper data, Int. J. Remote Sens. 11(5):783-828.
Curran, P. J. (1988), The semivariogram in remote sensing: an introduction, Remote Sens. Environ. 24:493-507.
Elvidge, C. D., and Portigal, F. P. (1990), Change detection in vegetation using 1989 AVIRIS data, in Proc. SPlE, imaging Spectroscopy of the Terrestrial Environment (G. Vane, Ed.), Orlando, FL, 16-17 April, pp. 178-189.
Field, C. B. (1991), Ecological sealing of carbon gain to stress and resource availability, in Response of Plants to Multiple Stresses (H. A. Mooney, W. E. Winner, and E. J. Pell, Eds.), Academic, London, pp. 35-65.
Fung, I. Y., Tucker, C. J., and Prentice, C. K. (1987), Application of advanced very high resolution radiometer vegetation index to study atmosphere-biosphere exchange of CO2, J. Geophys. Res. 92:2999-3015.
Gamon, J. A., Penuelas, J., and Field, C. B. (1992), A narrow-band spectral index that tracks diurnal changes in photosynthetic efficiency, Remote Sens. Environ. 41:35-44.
Goodwin, T. W., Ed. (1988), Plant Pigments Academic, London.
Goulden, M. C. (1991), Nutrient and water utilization by evergreen oaks that differ in rooting depth, Ph.D. dissertation, Stanford University.
Green, R. O., Conel, J. E., Carrere, V., et al. (1990), Determination of the in-flight spectral and radiometric characteristics of the airborne visible / infrared imaging spectrometer (AVIRIS), in Proc. 2nd Airborne Visible/lnfrared imaging Spectrometer Workshop, 4-5 June, JPL, Pasadena, CA, pp. 15-34.
Hobbs, R. J., and Mooney, H. A., Eds. (1990), Remote Sensing of Biosphere Functioning, Springer-Verlag, New York.
Huete, A. R. (1986), Separation of soil-plant spectral mixtures by factor analysis, Remote Sens. Environ. 19:237-251.
Isaac, R. A., and Johnson, W. C. (1976), Determination of total nitrogen in plant tissue, using a block digestor, J. Assoc. Off. Anal. Chem. 59:98-100.
Nemani, R. R., and Running, S. W. (1989), Estimation of regional surface resistance to evapotranspiration from NDVI and thermal-IA AVHRR data, J. Appl. Meteorol. 28(4):276-284.
Page, B. M., and Tabor, L. L. (1967), Chaotic structure and decollement in Cenozoic rocks near Stanford University of California, Geol. Soc. Am. Bull. 78:1-12.
Porra, R. J., Thompson, W. A., and Kriedemann, P. E. (1989), Determination of accurate extinction coefficients and simultaneous equations for assaying chlorophylls a and b extracted with four different solvents: verification of the concentration of chlorophyll standards by atomic absorption spectroscopy, Biochim. Biophys. Acta 975:384-394.
Roberts, D. A. (1991), Separating spectral mixtures of vegetation and soils, Ph.D. dissertation, University of Washington, 180 pp.
Roberts, D. A., Yamaguchi, Y., and Lyon, R. J. P. (1985), Calibration of airborne imaging spectrometer data to percent reflectance using field spectral measurements, in Proc. 19th Int. Symp. On Remote Sens. Environ., Ann Arbor, MI, 21-25 October.
Roberts, D. A., Smith, M. O., Adams, J. B., Sabol, D. E., Gillespie, A. R., and Willis, S. C. (1990), Isolating woody plant material and senescent vegetation from green vegetation in AVIRIS data, in Proc. of the Second Airborne Visible/lnfrared imaging Spectrometer (AVlRlS) Workshop (R. O. Green, Ed.), Pasadena, CA, 4-5 June, pp. 42-57.
Roberts, D. A., Smith, M. O., and Adams, J. B. (1993), Green vegetation, nonphotosynthetic vegetation, and soils in AVIRIS data, Remote Sens. Environ. 44:255-269.
Rossi, R. E., Mulla, D. J., Journel, A. G., and Franz, E. H. (1992), Geostatistical tools for modeling and interpreting ecological spatial dependence, Ecol. Monogr. 62(2):277314.
Running, S. W. (1990), Estimating terrestrial primary productivity by combining remote sensing and ecosystem simulation, in Remote Sensing of Biosphere Functioning (R. J. Hobbs and H. A. Mooney, Eds.), Springer-Verlag, New York, pp. 65-86.
Running, S. W., and Nemani, R. R. (1988), Relating seasonal patterns of the AVHRR vegetation index to simulated photosynthesis and transpiration of forests in different climates. Remote Sens. Environ. 24:347-367.
Smith, M. O., Ustin, S. L., Adams, J. B., and Gillespie, A. R. (1990), Vegetation in deserts I. a regional measure of abundance from multispectral images, Remote Sens. Environ. 31(1):1-26.
Smith, M. O., Ustin, S. L., Adams, J. B., and Gillespie, A. R. (1990), Vegetation in deserts: II. Environmental influences on regional abundance, Remote Sens. Environ. 31(1):27-52.
Tucker, C. J., Fung, I. Y., Keeling, C. D., and Gammon, R. H. (1986), Relationship between atmospheric CO2 variations and a satellite-derived vegetation index, Nature 319:195-199.
Ustin, S. L., Smith, M. O., and Adams, J. B. (1993), Remote sensing of ecological processes: a strategy for developing and testing ecological models using spectral mixture analysis, in Scaling Physiological Processes: Leaf to Globe a Ehleringer and C. B. Field, Eds.), Academic, New York, pp. 339-357.
Vane, G., Ed. (1987), Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). A description of the sensor, ground data processing facility, laboratory calibration, and first results, NASA JPL Publication 87-38, Pasadena, CA.
Vane, G., Green, R. O., Chrien, T. G., Enmark, H. T., Hansen, E. G., and Porter, W. M. (1993), The airborne visible / infrared imaging spectrometer (AVIRIS), Remote Sens. Environ., 44:127-143.
Verma, S. B. (1990), Micrometeorological methods for measuring surface fluxes of mass and energy, Remote Sens. Rev. 5(1):99-115.
Weidner, V. R., and Hsia, J. J. (1981), Reflectance properties of pressed polytetrafluoroethylene powder, J. Opt. Soc. Am. 71(7):856-861.
Welles, J. M., and Norman, J. M. (1991), Instrument for indirect measurement of canopy architecture, Agron. J. 84:818-825.
Young, A., and Britton, G. (1990), Carotenoids and stress, in Stress Responses in Plants: Adaptation and Acclimation Mechanisms (R. G. Alscher, and J. R. Cumming, Eds.), Wiley-Liss, New York, pp. 87-112.