Several factors can result in errors in NDVI based estimates of APAR. In ecosystems with incomplete canopy closure, the spectral signature received by the sensor has components from both the vegetation and from the substrate (ea. soil, rock, plant litter). Depending on the viewing and illumination geometries, canopy architecture, and on the albedo of the substrate in the region of chlorophyll absorption, errors in NDVI estimated APAR can be as high as 50 percent. The presence of a large proportion of shadow can also produce errors in the observed reflectance of a forest canopy (Smith et al. 1983; Curtiss and Ustin 1988), that, if they are not corrected for, can result in an underestimation of NDVI. Thus, calculated NDVI can vary within a scene due to changes in viewing geometry, and can change seasonally because of illumination geometry. These changes will occur independent of any real change in ecosystems photosynthetic capacity.
Ecosystems tend to adjust their maximum photosynthetic capacity in relation to available resources through changes in leaf area index and other architectural changes, chlorophyll content, and species composition. When the rate change in a stress is greater than the rate at which the ecosystem can adjust its photosynthetic rate, the assumption of constant photosynthetic efficiency is not valid. Examples of this are. short term stresses, such as diurnal water stress; longer term (36 months) stress for ecosystems with slow rates of photosynthetic capacity adjustment, such as coniferous forests.
In some instances, these assumptions produce errors that cancel each other, and, therefore, give the appearance that the model is working. Northern latitude boreal forests are an example of such an ecosystem. During the winter months, calculated NDVI for these ecosystems goes down as would be expected if NDVI is related to photosynthetic productivity. In reality, NDVI goes down not because of a reduction in APAR, but because of the increase importance of shadowing associated with the seasonal change in solar zenith angle. This underestimate of APAR is offset by a lowering of photosynthetic efficiency resulting from the lower winter temperatures. Because the predictions of ecosystem productivity are linked to solar zenith angle, it is likely that they will be insensitive to true changes in boreal forest productivity such as might be associated with longer, colder winters or more frequent spring freezes.
Many of the errors associated with NDVI based estimates of APAR can be overcome by using contiguous, high spectral resolution measurements made by sensors such as the High Resolution Imaging Spectrometer planed for the Eos polar platform, (Goetz and Herring 1989). These new sensors measure the reflected spectrum in many, contiguous, spectral bands in the visible and reflected infrared wavelengths; this data can be used for the direct identification and quantification of spectral features diagnostic of leaf constituents and, will thus permit a more physiologically based understanding of vegetation spectra. Spectral patterns within the visible and shortwave infrared regions are diagnostic for the presence of vegetation (Asrar et al., 1984). They may permit the identification of fundamental physical canopy properties (Goel and Thompson, 1984a,b; Sellers, 1985, 1986), and potentially define spectral patterns related to the assessment of environmental stress (Chang and Collins, 1983; Bauer, 1985). While minerals can be identified based on the presence of diagnostic spectral features (Goetz et al. 1985), vegetation spectra are generally very similar to one another due to their small base of shared constituents, e.g. pigments, cellulose, lignin, protein, starch. Variations in shape of the reflectance spectra of vegetation will be largely controlled by differences in leaf and canopy structure and foliar optical properties. Additionally, because of the high dimensionality of these datasets, spectral un-mixing models (Adams and Adams 1984; Smith et al. 1985; Adams et al. 1986) can be used to separate the spectral signatures present in a mixed pixel (ea. vegetation, substrate, shade, see Ustin et al. 1986).
To date, there have been few experimental studies concerning the remote detection of photosynthetic properties. Most experimental studies investigating spectral characteristics associated with physiological processes have examined environmental stress(es), such as drought or mineral excess, on the reflectance characteristics of individual leaves (Chang and Collins 1983; Ripple 1986). Stress-induced shifts in the red-edge have been reported, both toward longer (i.e. red) or shorter (blue) wavelengths, and have been attributed to changes in chlorophyll content. This red-shift of the chlorophyll absorption edge was first reported by Gates et al. (1965) and later by Collins (1978).
In a pilot study of ozone treated conifers, Ustin and Curtiss (1987) found a marked increase in reflectance in the chlorophyll absorbance bands of ozone treated conifers (Figure 1). Additionally, they report a small blue shift of 3-9 nm in the derivative of the inflection point wavelength of the 'red edge'. Such changes are consistent with reports of chlorophyll loss as a characteristic consequence of ozone injury. Further, Ustin and Curtiss (1987) suggest that subtle changes in the 590-690 nm spectral region indicate that chlorophyll a and b concentrations may be quantified in addition to total chlorophyll.
This paper presents a first step towards a better understanding of the
parameters that affect that portion of the reflected light spectrum dominated
by chlorophyll pigment absorption. It is hoped that this understanding
will provide a basis for the development of more accurate techniques for
the assessment of APAR and photosynthetic efficiency.
The Ponderosa Pine seedlings were exposed to a single seasons duration of realistic fumigation profiles having stochastic daily doses and episodic peaks. Dose levels were chosen to be representative of summer air quality currently experienced in many parts of the western United States. Reflectance measurements of the seedlings were made on 9 different areas on composite canopies formed from arranging about 30 seedlings in a tray. The elevation of the trays holding the seedlings were adjusted to maintain a constant distance between the canopy top and the PIDAS foreoptics. FiberFrax, a uniformly reflective ceramic fiber wool, was used as the reflectance standard. This standard was placed at the same distance from the foreoptics as the Ponderosa Pine canopies and was measured before and after each set of canopy measurements.
The Southern Sierra Nevada is an area of regional ozone exposure. Reflectance
spectra of three branches from each of 20 Ponderosa Pine trees at four
sites experiencing a range of ozone exposure levels were measured using
a setup similar the one used for the seedlings. Chlorophyll analysis and
reflectance measurements where made separately on each years growth. For
these measurements, Halon was used as a reflectance standard. A more complete
description of the site selection and sample analysis is presented in Ustin
et al. (1988).
The primary spectral changes observed in the seedings is that associated with chlorophyll loss, see Figure 1. Chlorophyll loss results in an increase in reflectance in the region of chlorophyll pigment absorption (450 - 700 nm). Additionally, this chlorophyll loss results in a shift in the red edge inflection to shorter wavelengths (a blue shift). The red edge inflection is defined as the center at half maximum of the first derivative of the spectrum, see Figure 2. The use of the wavelength of the red edge inflection (REI) is useful because it is much more insensitive to mixed pixel effects as compared to NDVI or other band ratio techniques for the quantification of light absorption by chlorophyll. The reflectance spectra in Figure 3 show that the reflectance spectrum of an ozone exposed seedling could be confused with the mixed signature of a control (no ozone exposure) seedling and granitic soil when using a broad band sensor such as TM or AVHRR. However, the wavelength of the REI of the mixture does not match the REI of the ozone exposed seedling.
While the relationship between chlorophyll concentration and maximum photosynthetic capacity is good within a single species, architectural variation between species and environmentally produced variation in photosynthetic efficiency result in a breakdown of this relationship. Thus, chlorophyll concentration can be used to asses the ability of parameters derived from reflectance data to predict maximum photosynthetic capacity only when the comparison is made within a species and when all samples have equivalent photosynthetic efficiency. Figure 4 show the relationship between REI wavelength and needle chlorophyll content for two site with similar ozone exposures. Each years growth (up to four) for 20 trees at 2 sites is plotted as a separate data point.
Some of the variance about the regression line observed in Figure
4 can be explained in terms of variance in photosynthetic efficiency.
Within a single site there is considerable variation in the observed visual
symptoms of ozone injury, such a chlorosis, low needle retention and fewer
years of needles retained. The greater chloroplast damage in these trees
produces a broadening of the chlorophyll absorption bands that results
in a shift to longer wavelengths (a red shift) of the REI (see Fig.
5) that is indicative of a lower overall photosynthetic efficiency.
Because of the complex interaction of chlorophyll absorption band broadening
and band depth the ability to determine the components of the photosynthetic
process using the remote sensing techniques discussed here is untried and
problematic. However, current techniques relying on measures like NDVI
have been demonstrated to be quite sensitive to canopy structure, variability
of the substrate, and viewing and illumination geometry. Thus, more accurate
means of predicting photosynthetic capacity and efficiency are required
before remotely sensed techniques can be applied for quantitative scene-independent
predictions of ecosystem productivity. If the specific chlorophyll absorption
feature parameters identified in this paper can alone, or in combination,
produce greater reliability of prediction, then the results of this study
will have broad significance.
2. Adams, J. B., M. O. Smith and P. E. Johnson. 1986. Spectral mixture modeling: a new analysis of rock and soil types at the Viking Lander I site. J. Geophys. Res. V80.
3. Asrar, G., E. T. Kanemasu, and M. Yoshida. 1985. Remote Sens. Environ. 17:1
4. Bauer, M.E. 1985. Spectral Inputs to Crop Identification and Condition Assessment. Proc. IEEE 73:1071-1085.
5. Chang, S.H. and W. Collins. 1983. Confirmation of the airborne biogeophysical mineral exploration technique using laboratory methods. Econ. Geol. 78: 723-736.
6. Collins, W. 1978. Remote sensing of crop type and maturity. Photogram. Eng. and Remote Sens. 44: 43-55.
7. Curtiss B. and S. L. Ustin. 1988. Characterization of sources of illumination in Ponderosa Pine forest communities using the Portable Instantaneous Display and Analysis Spectrometer. SPIE technical Symposium on Optic, Electro-Optics and Sensors in Orlando, Florida. April 4-8.
8. Gates, D. M., H. J. Keegan, J. C. Schleter, and V. R. Weidner. 1965. Spectral properties of plants. Appl. Optics 4: 11-20.
9. Goel, N. S. and R. L. Thompson. 1984a. Inversion of vegetation canopy reflectance models for estimating agronomic variables. 111. Estimation using only canopy reflectance data as illustrated by the Suits model. Remote Sens. Environ. 15: 223-236.
10. Goel, N. S. and R. L. Thompson. 1984b. Inversion of vegetation canopy reflectance models for estimating agronomic variables IV Total inversion of the SAIL model. Remote Sens. Environ. 15: 237 253.
11. Goetz, A. F. H., G. Vane, J. E. Solomon and B. N. Rock. 1985 Imaging spectrometry for earth remote sensing. Science 228:1147-1153.
12. Goetz, A. F. H. 1987. The portable instantaneous display and analysis spectrometer (PIDAS). Proc. 3rd. Airborne Imaging Spectrometer Data Analysis Workshop, June 2-4, JPL Publication #87-30.
13. Goetz, A. F. H. and M. Herring. 1989. The High Resolution Imaging Spectrometer (HIRIS) for Eos. IEEE Trans. Geosci. Remote Sensing 27: 1136-144.
14. McLauglin, S. B. 1985. Effects of air pollution on forests: A critical review. J. APCA 35:512-534.
15 Ripple, W. 1986. Spectral reflectance relationships to leaf water stress. Phonogram. Eng. and Remote Sens. 52: 1699-1675.
16. Sellers, P. J. .985. Canopy reflectance, photosynthesis and Transpiration. Int. J. Remote Sens. 6:1335-1372.
17. Sellers, P. J. 1986. Canopy reflectance, photosynthesis and transpiration. 11. The linearity of their response. Rem. Sensing Environ. 21:143-183.
18. Smith M., Cymerman C., Curtiss B. 1983. Application of reflectance models to mapping forest resources using LANDSAT images. Proc. RNRF Symp. Appl. Remote Sensing to Resource Manag., 137-146.
19. Smith, M. O., P. E. Johnson and J. B. Adams. 1985. Quantitative determination of mineral types and abundances from reflectance spectra using principal components analysis. Proc. 15th Lunar and Planetary Sci. Conf. II. J. Geophy. Rsrch. 90:C797-C804.
20. Schutt, J. B., R. R. Rowland, and W. H. Heartly. 1984. a laboratory investigation of a physical mechanism for the extended infrared absorption ('red shift') in wheat. Int. J. Remote Sens. 5: 92-102.
21. Thomson, W. W., W. M. Dugger and R. L. Palmer. 1966. Effects of ozone on the fine structure of the palisade parenchyma cells of bean leaves. Can. J. Bot. 44:1677-1682.
22. Ustin S. L., J. B. Adams, C. D. Elvidge, M. Rejmanek, B. N, Rock, M. O. Smith, R. W. Thomas and R. A. Woodward. 1986. Thematic Mapper studies of semiarid shrub communities. BioSci. 36(7):446-452.
23. Ustin S. L. and B. Curtiss. 1987. Spectral Characteristics of Ozone Treated Conifers. Final Report, EPA contract #7B0008NTEX.
24. Ustin S. L., B. Curtiss, S. Martens, V. C. Vanderbilt . 1988. Use of high spectral resolution sensors to detect air pollution injury in conifer forests in Proc. EPA/ASPRS Symp. on the Remote Sensing of Air Poll. St. Louis, MO April 2-8.