Towards the Remote Sensing of Matorral Vegetation Physiology: Relationships between Spectral Reflectance, Pigment, and Biophysical Characteristics of Semiarid Bushland Canopies George Alan Blackburn* and Caitriana Margaret Steele*
T his study investigates the relationships between the spectral reflectance characteristics and the concentrations of photosynthetic pigments and biophysical attributes of a structurally complex, spatially heterogeneous vegetation canopy with varying background properties. A field experiment was performed in the Guadalentin basin, Spain using matorral vegetation canopies dominated by Rosmarinus officinalis, Cistus albidus, and Anthyllis cytosoides. A spectroradiometer was used to record the reflectance of a series of sites at which measurements were made of the concentrations per unit ground area and per unit leaf mass of chlorophyll a and b and the carotenoids, together with leaf area index and percent canopy cover. A range of spectral characteristics was examined which have been found previously to be related to pigment concentrations and biophysical properties of vegetation. For matorral vegetation many of these spectral characteristics were unrelated or only weakly related to canopy properties. However, it was found that pigment concentrations per unit ground area were related to ratios of reflectance in narrow spectral bands within the near-infrared region, ratios of bands within the red region, and characteristics of the amplitude of first derivative spectra in the red edge region. Pigment concentrations per unit leaf mass were correlated with ratios of bands around the nearinfrared “shoulder” and the amplitude of the first deriva-
* Department of Geography, King’s College London, London, United Kingdom Address correspondence to G. A. Blackburn, Dept. of Geography, King’s College London, Strand, London WC2R 2LS, UK. E-mail: alan. [email protected]
Received December 1998; revised 11 May 1999. REMOTE SENS. ENVIRON. 70:278–292 (1999) Elsevier Science Inc., 1999 655 Avenue of the Americas, New York, NY 10010
tive in certain visible wavelengths. LAI and percent cover were related to ratios of reflectance in narrow bands on the near-infrared plateau and red edge features of canopy reflectance spectra, as well as with the amplitude of the first derivative in the red edge and visible regions respectively. Elsevier Science Inc., 1999
INTRODUCTION AND AIMS The application of remote sensing for estimating the concentration of photosynthetic pigments within plants is increasingly seen as an important first step in providing detailed and accurate information on the condition and functioning of terrestrial vegetation (e.g., Pen˜uelas et al., 1995a; Lichtenthaler et al., 1996; Gitelson and Merzlyak, 1997). The concentrations of the main photosynthetic pigments chlorophyll a (Chl a), chlorophyll b (Chl b), and the carotenoids (Cars) relate strongly to the photosynthetic potential of a plant and therefore give an indication of its overall physiological status (Danks et al., 1983; Young and Britton, 1990). More specifically, remote estimates of pigment concentrations will provide an improved evaluation of the spatial and temporal dynamics of vegetation stress (Schepers et al., 1996; Filella et al., 1995), better estimates of productivity by measuring and interpreting absorbed photosynthetically active radiation more accurately (Chappelle et al., 1992; Kim et al., 1994), and potentially the improved discrimination of species by monitoring phenological dynamics (Demarez et al., 1999). The future generation of high spectral resolution space-borne sensors will be much more suitable for 0034-4257/99/$–see front matter PII S0034-4257(99)00044-9
Towards the Remote Sensing of Matorral Vegetation Physiology
quantifying vegetation pigment concentrations (and other canopy parameters) than previous instruments (Curran, 1994; Vane and Goetz, 1993). Empirical studies at both leaf and canopy scales have indicated that three broad types of spectral approaches may have some utility in predicting plant pigment concentrations and all rely upon measurements in narrow spectral bands over the optical wavelengths. The first approach has been to propose the use of reflectance in various individual narrow wavebands (e.g., Filella et al., 1995; Carter et al., 1996; Lichtenthaler et al., 1996; Mariotti et al., 1996; Blackburn 1998a,b). While there is little agreement on the optimal wavelengths, there is good evidence that, at wavelengths where the absorption coefficients of pigments are high, reflectance is more sensitive to low concentrations, while spectral regions with low absorption are more sensitive to higher pigment concentrations (Jacquemond and Baret, 1990; Yamada and Fujimara, 1991). The second approach, using ratios of reflectance in narrow bands, has been proposed as a method of solving the problems of the overlapping absorption spectra of different pigments (e.g., Chappelle et al., 1992) and the effects of leaf structure, leaf surface interactions, and canopy structure (e.g., Pen˜uelas et al., 1995a). Such factors may confound the relationships between reflectance in individual narrow bands and pigment concentration. Most workers propose pigment indices which employ ratios of narrow spectral bands in the visible and near-infrared (e.g., Aoki et al., 1981; Carter, 1994; Gitelson and Merzlyak, 1994; Gitelson et al., 1996a; Schepers et al., 1996; Carter et al., 1996), while some identify only visible wavelengths and others use narrow wavebands in the region of the reflectance red edge (e.g., Filella et al., 1995; Pen˜uelas et al., 1995b; Lichtenthaler et al., 1996; Gitelson and Merzlyak, 1996; 1997). Blackburn (1998a,b) found that, for indices incorporating narrow near-infrared and visible wavebands, simple ratios had more linear relationships with pigments than did normalized difference ratios at leaf and canopy scales, remaining sensitive over a very wide range of concentrations (e.g., up to around 2000 mg m22 for Chl a). The final approach has been to investigate the relationships between characteristics of the first and second derivatives of reflectance spectra and pigment concentrations. It has been suggested that spectral derivatives have important advantages over spectral reflectance, such as their ability to reduce variability due to changes in illumination or background reflectance (Curran et al., 1991; Elvidge and Chen, 1995). The majority of workers have used derivatives to define the wavelength position of the red edge (kRE) and have illustrated relationships between this parameter and total chlorophyll (Chl tot) concentration at a range of scales (e.g., Horler et al., 1983; Curran et al., 1990; Filella and Pen˜uelas, 1994; Gitelson et al., 1996b; Lichtenthaler et al., 1996; Mariotti et al., 1996; Munden et al., 1994; Pinar and Curran, 1996). The amplitude of the first and second derivatives of reflectance (dR and ddR) at particu-
lar wavelengths has also been found to be closely related to pigment concentrations as has the amplitude of the first derivative of pseudo absorbance (d(Log 1/R)) (e.g., Boochs et al., 1990; Yoder and Pettigrew-Crosby, 1995; Blackburn, 1998b; in press). One could conclude from this overview that no single spectral approach or selection of wavelengths is likely to have the strongest relationship with pigment concentrations under all circumstances and that there is a need to examine the potential of different approaches in different vegetation physiognomic and community types. Very few studies have examined this problem using vegetation canopies growing naturally in the field situation, with most opting for controlled experiments using individual leaves, collections of leaves or small canopies in the laboratory. Field studies that have been undertaken often fail to address the effects of factors such as variations in canopy architecture and background characteristics and have not examined the effectiveness of all types of spectral approach. The objective of the present study was to examine the robustness, with respect to these complicating factors, of the relationships between a wide range of spectral approaches and pigment concentrations at the canopy scale. The matorral vegetation investigated in this experiment is a bushland community which is a characteristic functional type found the semiarid regions bordering the Mediterranean (Di Castri and Mooney, 1973). This community displays a wide range of structural and therefore spectral complexity, with variations in the proportions of photosynthetic and nonphotosynthetic biomass, living and dead tissues, canopy coverage (and therefore exposed background and shadowing effects), substrate characteristics, canopy stratification, species composition, and canopy and leaf morphology. Moreover, this work on matorral vegetation communities also has an applied theme. In many parts of the Mediterranean region the abandonment of former agricultural land is enabling matorral vegetation to expand and eventually ameliorate some ecological and geomorphological problems. However, there is evidence that the onset of a drier climate is causing plant death, despite the recognized droughtresistance of the component species, leading to, for example, increased soil erosion and land degradation (Dell et al., 1986; Brandt and Thornes, 1996). Therefore, a means of remotely monitoring the physiological status of matorral vegetation over broad regions will enable us to further understand and predict the impacts of anthropogenic effects and environmental stresses on these important semi-natural communities. METHODOLOGY The sites used for this study were situated within the Guadalentin basin, Murcia, South East Spain. Half of the sites were located approximately 2 km south of the town of Velez Rubio within an area encompassing zones of
Blackburn and Steele
heavy to moderate grazing by herded goats and sheep. This provided sites with sparse, degraded matorral vegetation. More well-developed matorral vegetation was sampled using sites at Rambla Chortal, an area more remote from human habitation with much lower grazing pressure. Twelve sites measuring 2 m32 m were selected to provide a range of canopy pigment characteristics, biophysical/structural properties, species composition, and background characteristics, but with a reasonable degree of homogeneity within each site area. Four of the sites used were dominated by Rosmarinus officinalis, four by Cistus albidus and four by Anthyllis cytosoides. The main subdominants were Thymus zygis, Thymus beaticus, Helianthenum almierense, and Dorycnium pentaphyllum, found in varying amounts across the sites sampled. Background color varied between sites, from white through grey to red, as soil characteristics (regosols) responded to both underlying lithology which changed between quartzite and phyllite and as a result of variable surficial deposits of calcrete. While it was impractical in the field situation to implement a rigid multifactorial experimental design, the sites chosen were considered to be representative of the range of variation exhibited by seminatural matorral communities with respect to factors such as canopy structure, background condition, and species mixture. Consequently, the sites provided a good basis on which to evaluate the robustness of the relationships between different spectral approaches and pigment concentrations for this community type. Measurements were made at all sites during April 1996, using a systematic methodology at each site. First, the site was delineated at ground level using pegs and string, and the spectral reflectance properties of the site were measured using a Spectron Instruments SE590 Spectroradiometer loaned from the UK Natural Environment Research Council Equipment Pool for Field Spectroscopy (NERC-EPFS). The SE590 records a continuous spectrum between 400 nm and 1100 nm by simultaneous measurement in 256 contiguous channels, using a nominal sampling interval of 3 nm and a bandwidth of 10 nm (manufacturers advertised figure), with a typical integration time of 16/60 s. Markham et al. (1995) have evaluated the radiometric and spectral characteristics of this diode-array field spectroradiometer and found the bandwidth to be closer to 16nm. In accordance with the recommendations of Markham et al. (1995) and the NERC-EPFS, each spectrum recorded in the present experiment was an average of four scans acquired by the instrument, in order to increase the signal to noise ratio (SNR) and only data between 400 nm and 1000 nm was used. However, contrary to the suggestion by Markham et al. (1995), no tests were performed to determine the SNR for the particular experimental setup due to time limitations in the field, meaning that the results may have to be considered with some degree of caution. Nevertheless, the spectra obtained were smooth
and only began to display any notable increase in noise at wavelengths above approximately 900 nm. In the present study the target sensor head was fitted with a 158 field-of-view fore-optic, mounted on a mast and supported approximately 2 m above the canopy, viewing from a nadir position. Ten randomly located radiance spectra were obtained from the canopy, each spectrum being an average of four scans, and each being paired with an irradiance spectrum obtained immediately after the target spectrum using an upwards-pointing cosinecorrected sensor. Subsequently, each radiance/irradiance pair was used to calculate a percent reflectance spectrum, applying the necessary intercalibration for the two sensor heads used (Rollin, 1992). All spectra were collected under cloudless skies and all within 61 h of solar noon. The 10 reflectance spectra for each site were averaged, and the various spectral transformations (discussed below) were applied to the mean spectrum. Initially, the percentage of ground covered by living vegetation (% canopy cover) within each quadrat was estimated by eye (by one worker for all sites). This technique was considered acceptable given the time constraints of the fieldwork program and the evidence that visual assessment of cover provides a good approximation to superficially more accurate measures (Kent and Coker, 1994). Within each site, five 25 cm325 cm quadrats were randomly selected for destructive sampling. Plant shoots located within each quadrat were then harvested, leaves stripped from the stems, and the total fresh mass of leaves was measured for each species. Reflectance spectra of the bare substrate were obtained following the harvesting. Samples of leaves of each species were taken randomly from the total stripped and used to determine the specific leaf area, moisture content, and pigment composition. The specific leaf area was calculated by measuring the single-sided area of samples of fresh leaves using a digital scanner and image processing software and measuring the mass of these samples. The mean value of specific leaf area for each species was subsequently used to calculate the total leaf area for each species within each quadrat, using the measurement of total mass of stripped leaves. The LAI of the quadrat was then calculated by dividing the total area of leaves for all species by the area of the quadrat. A second set of leaf samples were weighed and placed in sealed plastic bags in a cool box after stripping and frozen within a few hours of harvesting. Subsequently, these samples were divided, with one portion being dried and used to determine moisture content and the other portion being used for pigment determination. The concentration per unit leaf mass of Chl a, Chl b, and the Cars (in mg/g dry plant material) were measured spectrophotometrically for each species using methanol as a solvent and employing the equations of Lichtenthaler (1987). The results from these sample determinations were then used together with the total fresh leaf mass of each species to calculate the concentration per unit ground area for each
Towards the Remote Sensing of Matorral Vegetation Physiology
Table 1. Summary of the Pigment and Biophysical Properties of the Study Sites, Showing the Mean Value and Standard Deviation for Each Parameter, with Sites Ranked According to the Concentration per Unit Ground Area of Chl a Site No.
120.03 11.94 63.60 8.04 183.63 20.24 46.62 8.31
121.52 9.66 62.14 7.24 183.66 23.91 47.98 7.11
166.52 21.56 98.37 14.41 264.89 21.24 57.61 5.66
208.35 29.06 110.70 12.89 319.05 47.92 81.61 10.47
244.30 14.53 122.16 13.01 366.46 44.05 102.89 8.05
246.03 22.02 125.64 14.63 371.67 59.54 97.43 7.62
294.31 23.39 147.17 20.08 441.48 66.31 123.95 22.09
387.98 34.72 224.58 23.91 612.56 36.87 136.35 22.94
600.02 83.69 306.00 60.12 321.98 22.60 199.00 29.50
895.75 44.33 529.06 61.61 1424.81 157.01 314.75 52.95
234.39 35.04 124.19 14.46 358.59 28.76 91.03 8.03
140.63 21.02 71.92 8.38 212.54 21.30 55.52 8.23
225.03 20.14 132.93 19.47 357.96 53.76 77.85 13.10
235.43 21.07 125.09 20.82 360.51 50.54 92.22 13.67
240.45 28.73 120.24 14.00 360.69 25.32 101.27 17.04
127.74 10.15 65.24 8.25 192.98 27.05 50.59 4.46
240.45 19.11 120.24 12.80 360.69 46.96 101.27 9.95
183.44 10.91 106.18 10.24 289.63 34.81 64.47 10.85
177.87 26.59 97.09 12.28 274.96 24.80 66.83 7.23
223.44 26.70 131.97 20.65 355.40 46.27 78.51 6.14
0.51 0.19 66
0.86 0.19 27
0.74 0.18 28
0.89 0.23 37
1.02 0.28 24
1.93 0.61 35
1.22 0.25 33
2.12 0.50 33
2.70 0.24 74
4.01 0.82 80
Concn per unit ground area (mg Chl a 79.03 87.50 Std 5.49 8.70 Chl b 40.33 50.06 Std 5.50 6.83 Chl tot 119.35 137.55 Std 10.77 9.66 Cars 31.42 33.14 Std 3.71 5.58 Concn per unit leaf mass Chl a 128.37 Std 6.35 Chl b 65.50 Std 10.90 Chl tot 193.87 Std 11.67 Cars 51.03 Std 5.52 Biophysical properties LAI 0.62 Std 0.16 % cover 59
(mg g21) 188.45 24.40 107.81 19.02 296.26 26.72 71.38 8.44 0.46 0.09 28
pigment in the quadrat. A mean value for the concentration of pigments per unit leaf mass in a quadrat was calculated by weighting this value for each species by the proportional abundance (based on dry mass) of that species in the quadrat. The pigment and biophysical data from the five quadrats at each site were then averaged to give values for the whole canopy that could be compared with the spectral reflectance for that area. RESULTS AND DISCUSSION Canopy Characteristics The sites sampled in this experiment provided a good range of values for all parameters of interest. For example, Chl a concentration per unit ground area varied between 79 mg m22 and 896 mg m22, LAI between 0.46 and 4.01, and percent canopy cover between 24% and 80%. Table 1 summarizes the pigment and biophysical characteristics of the study sites. Ideally, given the objectives of this study the experimental design should enable individual pigments, Chl a, Chl b, and Cars, to vary completely independently of each other and canopy biophysical properties. However, as noted by Blackburn (1998b), when we are using whole plants that are growing naturally, this can be difficult to achieve. For the matorral vegetation canopies used in this study the concentrations of individual pigments per unit ground area were all highly linearly correlated with each other [R250.99** for Chl a versus Chl b and Cars; 0.98*** for Chl b versus Cars (asterisks refer to significance level of coefficient of determination: * p,0.05; ** p,0.01; *** p,0.001)] and with LAI (R25 0.93*** for LAI versus Chl a; 0.92*** for
LAI versus Chl b and Cars). As observed by Blackburn (1998b) for Pteridium aquilinum, these strong relationships can be attributed to the overriding influence of the degree of canopy structural development on pigment concentration per unit ground area. The concentrations of pigments per unit leaf mass were slightly less well correlated with each other. Chl a and Chl b had a strong linear relationship (R250.92***) while an exponential model best described the relationship between Chl a and Cars (R250.95***). The linearity of the former relationship and curvilinearity the latter relationship conforms with that found elsewhere (e.g., Chappelle et al. 1992; Pen˜uelas et al., 1995a). The relationship between the concentrations per unit leaf mass of Chl b and Cars was considerably weaker (R25 0.77*) and no pigments were related to LAI. Percent canopy cover was unrelated to the concentration (per unit ground area or leaf mass) of any pigment or LAI. Moreover, there are no relationships between the concentrations of pigments per unit ground area and per unit leaf mass. Therefore, the canopies sampled in this study enable us to evaluate the relationships between different spectral approaches and these independent pigment variables that may provide complimentary physiological information. Relationships between Pigment Concentrations and Spectral Reflectance An initial step in the analysis was to investigate whether there were systematic differences between sites according to the dominant species. In the absence of sufficient/suitable data for a meaningful statistical test (such as analysis of variance), plots were constructed of spec-
Figure 1. Mean reflectance spectra with a 61 standard deviation envelope (same grey level as the mean but thinner lines) for the matorral vegetation canopies (a, b, c) and the two main soil types (d).
282 Blackburn and Steele
Towards the Remote Sensing of Matorral Vegetation Physiology
Figure 2. Variations with wavelength in the coefficient of determination (R2) obtained when regressing percent reflectance against the concentration per unit ground area of Chl a and percent canopy cover, using a exponential and power models respectively. For reference, the reflectance spectrum of a well-developed matorral vegetation canopy is also given.
tral data versus pigment concentrations (per unit ground area and leaf mass) and biophysical characteristics, in which the three species were distinguished by using different colored points. A visual examination of these plots revealed that in terms of the relationships between spectral data and vegetation parameters there were no systematic differences between the three dominant species, Rosmarinus officinalis, Cistus albidus, and Anthyllis cytosoides. While these species are of the same functional type, characteristic of semiarid bushland communities, they do vary in terms of leaf and canopy morphology and therefore differences in the above relationships could be conceived. Clearly, more work is required to further establish the species-independence of the relationships between spectral reflectance and vegetation parameters for matorral communities, as only a limited number of species have been considered here. However, within the context of this experiment, differences between species appear to be negligible, hence, the following results present data for all sites combined. Figure 1(a, b, and c) show the mean reflectance spectrum for each matorral vegetation site, with a 61 standard deviation envelope. Figure 1d shows a mean spectrum (with 61 standard deviation envelope) for each of the two main soil types found at the study sites, formed on quartzite and phyllite bedrocks. In order to identify those spectral regions which were most strongly related to the vegetation parameters of interest, correlograms were constructed by sequentially regressing the value of percent reflectance at each SE590 channel (Rk) against pigment and biophysical values and plotting the coefficient of determination (R2) against wavelength. Figure 2 shows a correlogram that
illustrates how the correlation between percent reflectance and Chl a concentration per unit ground area changes with wavelength. An exponential model was found to best describe the relationship between percent reflectance and Chl a concentration per unit ground area for the majority of wavelengths; therefore, this model was used to construct the correlogram. For reference, a reflectance spectrum of a well-developed matorral vegetation canopy has been included in the plot, using the same wavelength axis. As Figure 2 shows, the correlogram has a very similar form to that of the reflectance spectrum, with little correlation throughout the visible wavelengths and reasonably high correlation throughout the near-infrared. Indeed, this form of correlogram is found almost identically for the relationships between spectral reflectance and concentrations per unit ground area of Chl b, Chl tot, and Cars and LAI. As none of the photosynthetic pigments absorb radiation in the near-infrared wavelengths, the relationships between near-infrared reflectance and pigment concentrations are unlikely to be causal. Instead this correlation could be explained as being indirect, as a result of the strong relationships between pigment concentrations per unit ground area and LAI for the canopies sampled (as noted in the previous subsection). As canopy LAI increases, then the potential for the scattering of incident radiation increases. In the near-infrared wavelengths, where there is little absorption by the canopy, this increased scattering leads to increased reflectance; therefore, there is a positive relationship between LAI and near-infrared reflectance, with a maximum correlation occurring at 783.2 nm (R783.2) (R250.58* using the model y50.22e0.11x). As the concentrations per unit ground area of pigments had
Blackburn and Steele
strong linear relationships with LAI, then this may explain why they too had a peak in correlation with R783.2 (R250.54* for Chl a and 0.53* for Chl b, Chl tot, and Cars). The reasons why pigment concentrations per unit ground area were not related to reflectance in the visible wavelengths are less apparent. One possibility is that this results from the complex structure of the canopies sampled, where variations may occur between sites in the proportions and relative magnitude of sunlit and shaded portions of canopy and ground within the FOV of the spectroradiometer, depending on the degree of canopy development. A second reason may be the variation in background reflectance between sites, which was greater in some visible wavelengths than in the near-infrared. Both of these effects would induce additional variability in visible reflectance between sites and therefore confound the relationship between reflectance and canopy pigment concentrations and LAI. Several workers have found that for monospecific canopies with a more uniform structure and more consistent background reflectance properties than those used in the present experiment, reflectance in the visible wavelengths is well correlated with pigment concentrations per unit ground area. For example, Filella et al. (1995) found significant correlations between Chl a concentration per unit area and R550 and R680 for wheat (Triticum aestivum) crop canopies. Similarly, Mariotti et al. (1996) found strong relationships between Chl a concentration per unit area and R550 for crop canopies of Laurus corn (Zea mays) and Oleica sunflower (Helianthus annuus). For bracken (Pteridium aquilinum) canopies, Blackburn (1998b) found that concentrations per unit ground area of Chl a, Chl b, and Cars were highly correlated with reflectance in the blue and red spectral regions in addition to the near-infrared. Further differences are found between the bracken and matorral vegetation canopies when we examine pigment concentrations per unit leaf mass. With the matorral vegetation canopies sampled in the present experiment, there were no relationships at all between concentrations per unit leaf mass of any pigment and spectral reflectance over the range of wavelengths measured by the SE590. In contrast, for bracken reasonably strong relationships (e.g., R250.7) were found between pigment concentrations per unit leaf mass and reflectance particularly in the red region but also in the blue and near-infrared. Interestingly, Gitelson and Merzlyak (1996) found that in reflectance spectra of individual leaves of chestnut (Aesculus hippocastanum) and maple (Acer platanoides) containing a range of pigment concentrations, there was a strong linear relationship between R550 and R700. In the present experiment a strong linear relationship was also found between R550 and R700 (R250.97***; y51.2x10.11) in the reflectance spectra of the matorral vegetation canopies. Unfortunately, in contrast to Gitelson and Merzlyak (1996), neither R550 nor R700 were related to Chl a
concentration (per unit ground area or leaf mass). Overall, in comparing the findings of the present experiment with those of previous studies, we can suggest that the optimum spectral approaches for remotely estimating canopy pigment concentrations are likely to be different between vegetation communities or functional types. Clearly, further work is needed to both explain and fully characterise these differences between vegetation types. Figure 2 also shows that percent canopy cover has a very different correlogram to that of Chl a, almost having an exact inverse form. A power regression model was found to best describe the relationship between percent reflectance and percent canopy cover for the majority of wavelengths; therefore, this model was used to construct the correlogram. This shows that percent canopy cover is reasonably well correlated with reflectance throughout the visible wavelengths, having a weaker correlation within the green region but a peak in the red at R673.1 (R250.58*; y5194.5x20.91) at around the absorption maximum of the chlorophylls. Through the red-edge region, correlation between reflectance and percent canopy cover drops markedly, and no relationship exists within the near-infrared. Correlograms (not shown here) of the variation with wavelength in the relationships between vegetation parameters and pseudo absorbance (Log 1/R) were virtually identical in form to those for R, with slightly lower correlations over most wavelengths. Relationships between Pigment Concentrations and Narrow-Band Spectral Ratio Indices The first approach taken was to examine a range of narrow-band spectral indices that have been proposed in the recent literature for tracking variations in pigment concentrations. Chappelle et al. (1992) developed the “ratio analysis of reflectance spectra (RARS)” indices which used narrow bands in the reflectance spectra of leaves which corresponded to unconvoluted absorption bands of Chl a and b and the Cars. These bands were identified by dividing a reference spectrum by a reflectance spectrum with a higher overall reflectance, producing a ratio spectrum which amplified spectral differences at wavelengths related to the pigment absorption bands. Hence, they specified the following indices: RARSa5(R675/R700)/ (r675/r700); RARSb5(R675/R6503R700)3(r6503r700/r675); and RARSc5(R760/R500)/(r760/r500). Here, a, b, and c denote the pigments Chl a, Chl b, and Cars, respectively, at which each RARS index is aimed; R5percent reflectance at the specified wavelength (in nm) in a reflectance spectrum—in this study the values of percent reflectance in the individual SE590 channels which were centered on or closest to the specified wavelengths; r5percent reflectance at the specified wavelength in a reference reflectance spectrum (in the present study this was the mean spectrum of two well developed matorral vegetation sites). Chappelle et al. (1992) found strong linear rela-
Towards the Remote Sensing of Matorral Vegetation Physiology
tionships between the above indices and the concentrations per unit area of Chl a, Chl b, and the Cars respectively, for individual soybean (Glycine max. Merr.) leaves. The findings for the present experiment were less positive. A linear regression between RARSa and Chl a concentration per unit ground area produced a coefficient of determination of only 0.5, while that for RARSc and Cars was slightly better at 0.56. There was no relationship between RARSb and Chl b concentration per unit ground area. Blackburn (1998a,b) found that it was possible to generate a relationship between the RARS indices and pigment concentrations of individual leaves and bracken canopies by modifying the wavebands used in the index, for example substituting R680 for R635 and R800 for R700. However, in the present study such waveband modifications failed to improve the relationships between the RARS indices and pigment concentrations per unit ground area. A similar set of indices to RARS were developed by Blackburn (1998a), namely, the “pigment-specific simple ratio” indices: PSSRa5R800/R680; PSSRb5R800/R635; and PSSRc5R800/R470. These were found by Blackburn (1998b) to have very strong curvilinear relationships with the concentrations per unit ground area of Chl a, Chl b, and Cars, respectively, for bracken canopies throughout a growing season. For the matorral vegetation canopies measured in the present study, PSSRa has a reasonably strong linear relationship with Chl a concentration per unit ground area (R250.71**), but substantial scatter about the regression model exists at low concentrations, which represent the majority of sites sampled in this experiment. Weaker linear relationships were found between PSSRb and Chl b concentration per unit ground area (R250.68*) and between PSSRc and Cars concentration per unit ground area (R250.5*). The “pigment-specific normalized difference” indices were also proposed by Blackburn (1998a): PSNDa5 (R8002R680)/(R8001R680); PSNDb5(R8002R635)/(R8001R635); and PSNDc5(R8002R470)/(R8001R470). These indices were generally less well correlated with matorral vegetation canopy pigment concentrations per unit ground area. Linear regression models best described the relationships between the PSND indices and Chl a (R250.53), Chl b (R250.54) and Cars (R250.51). These findings are substantially different to those for bracken canopies (Blackburn, 1998b) where the PSND indices were all found to have very strong exponential relationships with pigment concentrations per unit ground area. While the structural/ spectral complexity and variable background conditions between sites compared to bracken may explain the relative weakness of the relationships for matorral vegetation, the lack of curvilinearity could be a result of the much smaller range of pigment concentrations per unit ground area exhibited by the matorral vegetation. For example, Chl a varied between 47 mg m22 and 2800 mg m22 for the bracken, while for the matorral vegetation
the range was much narrower, between 79 mg m22 and 896 mg m22. The curvilinearity of these relationships for bracken meant that the PSND indices became insensitive to pigment concentrations above approximately 500 mg m22. In the present experiment as there are few data points above 500 mg m22, and those which are do not extend much beyond this figure, this insensitivity does not become apparent. Two simple ratio indices, R750/R700 and R750/R550 were identified by Lichtenthaler et al. (1996), Gitelson and Merzlyak (1996), and Gitelson et al. (1996b) as having strong linear relationships with Chl tot concentration at the leaf scale. Blackburn (in press) found that for stacks of deciduous tree leaves at various stages of senescence, these indices had good relationships with Chl tot concentration per unit ground area but these relationships were curvilinear over the much larger range of concentrations experienced than in the previous studies. In the present experiment the ratio R750/R700 was found to have a reasonably strong linear relationship with Chl tot (R250.73*), with similar relationships for Chl a and b individually. Again, in this experiment the range of pigment concentrations was much more limited than that of Blackburn (in press), and this may explain why a curvilinear relationship was not evident. The ratio R750/R550 was less well correlated with Chl tot (R250.5, using a linear model). Aoki et al. (1981) also used a narrow green band in the ratio R550/R800 which they found to be a good nondestructive method for estimating leaf chlorophyll concentration. In the present experiment this ratio was unrelated to Chl tot concentration per unit ground area (or Chl a and b individually). Similarly, no relationships were found between matorral vegetation canopy chlorophyll concentrations per unit ground area and R850/R550 (Schepers et al., 1996) or the “green NDVI” (5(R7502Rgreen)/ (R7501Rgreen), where Rgreen5mean reflectance between 530 nm and 570 nm) proposed by Gitelson et al. (1996a). A number of workers have proposed narrow-band indices for estimating the Cars:Chl a ratio. In the present study the following indices were tested: the “structure insensitive pigment index,” SIPI5(R8002R445)/(R8002R680) (Pen˜uelas et al., 1995a); the “simple ratio pigment index,” SRPI5R430/R680 (Pen˜uelas et al., 1995b); and the “normalized pigment chlorophyll index,” NPCI5(R4302 R680)/(R6801R430) (Filella et al., 1995). For the matorral vegetation canopies in this experiment none of these indices were related to the Cars:Chl a ratio. A further interesting, if rather negative, finding was that there were no relationships between any of the existing narrow-band indices and the concentrations of pigments per unit leaf mass (Chl a, b, tot, or Cars) for the matorral vegetation canopies. In a similar fashion, the majority of the indices tested had no relationship with either LAI or percent canopy cover. The only existing narrow-band index which was related to LAI was R750/ R700 (R250.62* using a linear model), but this was unre-
Blackburn and Steele
Figure 3. Variation in the coefficient of determination (derived using an exponential regression model) for the relationship between Chl a concentration per unit ground area and the matrix of combinations of Ra and Rb used in a simple ratio index.
lated to percent canopy cover. Furthermore, neither LAI nor percent canopy cover were related to the commonly used broadband “normalized difference vegetation index” (NDVI5(NIR2RED)/(NIR1RED)) and “simple ratio vegetation index”” (SR5NIR/RED), calculated by applying filter functions to the SE590 spectra to simulate Bands 2 and 3 of the SPOT-HRV instrument. This finding disagrees with many previous investigations which have found close relationships between these indices and canopy biophysical properties, but it conforms with the results of Blackburn (in press) where no relationship was found between NDVI or SR and LAI for stacks of leaves with widely differing pigment concentrations. Steele (in press) provides a more extensive examination of the relationships between matorral vegetation canopy biophysical and physiological properties and broadband vegetation indices, including indices that have been designed to remove the effects of variable background reflectance properties. Overall, the existing spectral ratio indices tested here perform less well or do not have any relationships with pigment concentrations in the matorral vegetation canopies sampled in this experiment. A possible explanation for this lack of success is that all the indices tested above use wavebands in the visible region, and, as already noted in the previous subsection, reflectance in this region is unrelated to pigment concentrations for matorral vegetation. Given the limited applicability of existing indices, an empirical approach was undertaken in order to
identify the combination of narrow wavebands in a simple ratio or normalized difference transformation which were most closely related to the pigment and biophysical characteristics of the matorral vegetation canopies. This was achieved using an iterative process that systematically combined reflectance in every narrow band of the spectrum of a site with reflectance in every other narrow band to calculate a simple ratio (SR) and normalized difference (ND): SR5Ra/Rb, ND5(Ra2Rb)/Ra1Rb), where Ra5R400 to 1000 and Rb5R400 to 1000. For each combination of wavebands, the values of SR and ND across all sites were regressed against pigment and biophysical properties using both linear and exponential models. Figure 3 shows some results from this exercise. This plot illustrates the variation in the coefficient of determination (derived using an exponential regression model) for the relationship between Chl a concentration per unit ground area and the matrix of combinations of Ra and Rb used in an SR index. Figure 3 illustrates that there is almost symmetry in R2 values either side of the diagonal line that represents a5b, and that most combinations of wavebands are unrelated to Chl a concentration per unit ground area. However, two regions of the matrix plot represent strong relationships. The first is an area encompassing combinations of near-infrared wavebands between 800 nm and 900 nm. Indeed, this region produces
Towards the Remote Sensing of Matorral Vegetation Physiology
Figure 5. Relationship between matorral vegetation canopy LAI and R956.4/R756.1.
Figure 4. Relationship between with Chl a concentration per unit ground area and a) R836.1/ R816.8 and b) R652.5/R623.2.
a ratio, R836.1/R816.8, which has the strongest relationship with Chl a concentration per unit ground area using an exponential model (see Fig. 4a). The second area of Figure 3 that represents strong relationships with Chl a concentration per unit ground area comprises combinations of wavebands in the red region (approximately 600–700 nm). From all combinations which contain a waveband in the visible, the ratio R652.5/ R623.2 has the strongest relationship with Chl a concentration per unit ground area (see Fig. 4b). Clearly, it is possible for this relationship to be causal, as both of these bands are positioned on the shorter wavelength wing of the major chlorophyll absorption feature in the red region. For Chl a, the matrix plot of R2 derived by applying a linear regression model is very similar to that shown in Figure 3, but with slightly lower R2 values overall. The waveband combination which has the strongest linear relationship with Chl a concentration per unit ground area is R968.9/R931.2 (R250.89**). It is not impossible for this to
be a causal relationship, but these bands are located on the shorter wavelength wing and middle of the water absorption feature, centred on 970 nm (Curran, 1989). Therefore, the relationship between R968.9/R931.2 and Chl a concentration per unit ground area may arise because of the correlation between Chl a concentration and the water content of the matorral vegetation canopies sampled. Clearly, this possibility requires further investigation. Those waveband combinations that have the strongest relationships with Chl a concentration per unit ground area also have the strongest relationships with Chl b and Chl tot, for both exponential and linear regression models (namely, R836.1/R816.8 and R968.9/R931.2, respectively). In terms of Cars concentration per unit ground area, the matrix plots of R2 are very similar to those for Chl a and the ratio R836.1/R816.8 is most strongly related to Cars using an exponential model, R902.8/R893.3 using a linear model. LAI has strong relationships with SR indices derived from combinations of near-infrared bands which are different to those related to pigment concentrations. The strongest linear relationship for LAI was with R956.4/ R756.1 (Fig. 5). These bands are situated both on the plateau of high reflectance in canopy spectra and in the upper wavelength regions of the reflectance red edge (see Fig. 2). Similarly, percent canopy cover is most strongly related to combinations of bands on the reflectance red edge and on the near-infrared plateau. Figure 6 shows the reasonably strong linear relationship that exists between percent canopy cover and R780.1/R747.1. Matrix plots of R2 (not reproduced here) illustrate that narrow-band SR indices are generally less well correlated with pigment concentrations per unit leaf mass. However, certain band combinations have reasonably strong linear relationships with Chl and Cars concentrations per unit leaf mass (e.g., Fig. 7). In contrast with the findings for pigment concentrations per unit ground
Blackburn and Steele
Figure 6. Relationship between percent canopy cover and R780.1/R747.1.
area, the band combinations which are most strongly correlated with pigment concentrations per unit leaf mass are located in the near-infrared “shoulder” region of the canopy reflectance spectra, where the red edge feature reaches an asymptote. As with Chl a, the concentrations per unit leaf mass of Chl b and Chl tot were most strongly related to R798.3/R768.0 (R250.75** for both, using a linear model), while Cars concentration per unit leaf mass had a maximum correlation with R792.2/R765.0 (R250.71*, using a linear model). The functional equivalence of SR and ND ratios meant that the waveband combinations used in the ND which have the strongest correlation with pigment and biophysical properties are the same as those for the SR and produced the same coefficients of determination.
Figure 7. Relationship between Chl a concentration per unit leaf mass and R798.3/R768.
Relationships between Pigment Concentrations and Spectral Derivative Characteristics A number of theoretical and empirical studies have demonstrated that the feature within reflectance spectra of vegetation that is most strongly related to physiological dynamics generally and pigment concentrations specifically is the red edge. In particular, the utility of the wavelength position of the red edge (kRE) has been highlighted as an increase in pigment concentration causes deepening and widening of the chlorophyll absorption feature in the red region (e.g., Horler et al., 1983; Pinar and Curran, 1996). In the present experiment the position of the red edge in the mean reflectance spectrum from each site was determined by identifying the wavelength of the peak in the corresponding first derivative spectrum between 680 nm and 750 nm, that is, the point of maximum slope. Interestingly, it was found that for the matorral vegetation canopies, kRE was unrelated to the concentrations of Chl a, Chl b, Chl tot, or Cars per unit ground area or leaf mass and was also unrelated to LAI or percent canopy cover. Boochs et al. (1990) found that a number of features of the red edge (other than the wavelength position), quantified using first derivative spectra were related to pigment and biophysical characteristics of plants. The features identified were the maximum amplitude of the first derivative in the region of the reflectance red edge (dRmax), the amplitude of the first derivative at 703 nm (dR703) and the ratio dRmax/dR703. In the present experiment it was found that dRmax had reasonably strong linear relationships with the concentrations per unit ground area of Chl a, Chl b, Chl tot, and Cars (R250.83** for all pigments) and with canopy LAI (R250.78**). dR703 had a moderate relationship with pigment concentrations per unit ground area while the amplitude of the first derivative at other wavelengths were much more closely related to pigment concentrations per unit ground area (discussed later). dRmax/dR703 was unrelated to either pigment or biophysical parameters. In an experiment on wheat canopies, Filella et al. (1995) found that Chl a concentration per unit area had strong linear relationships with dRmax as well as a further feature of the red edge, the sum of the amplitudes of the first derivative between 680 nm and 780 nm, o dR680–780. In the present experiment o dR680–780 was also well correlated with the concentrations per unit ground area of Chl a, Chl b, Chl tot, and Cars (R250.80** for all pigments) and with canopy LAI (R250.76**). However, neither dRmax, dR703, dRmax/dR703 nor o dR680–780 was related to the concentrations of pigments per unit leaf mass or percent canopy cover. Figure 8 illustrates the variation with wavelength in the coefficient of determination for the relationship between the amplitude of the first derivative, dR, and Chl a concentration per unit ground area. The plot was con-
Towards the Remote Sensing of Matorral Vegetation Physiology
Figure 8. Variation with wavelength in the coefficient of determination (derived using a linear model) for the relationship between Chl a concentration per unit ground area and the amplitude of the first derivative of reflectance, dR.
structed using a linear regression model to derive the R2 values, the best-fit model for most wavelengths. The correlogram shows that there are several regions of the visible and near-infrared which have a good degree of correlation, but that correlation is highly variable with wavelength. There is a reasonably broad region of high correlation in the vicinity of the reflectance red edge. The peak correlation between dR and Chl a concentration per unit ground area is at 750.1 nm (R250.90***), as is illustrated in Figure 9. dR750.1 was also the band in the first derivative spectrum most strongly correlated with the concentrations per unit ground area of Chl b, Chl tot, and Cars (R250.90*** using a linear model). LAI
Figure 9. Relationship between and Chl a concentration per unit ground area and dR750.1.
was most strongly related to dR at a different wavelength to that of the pigments, namely, at 735.3 nm (R250.88**). Correlograms (not shown) reveal a great deal of variability with wavelength in the strength of relationship between dR and the concentrations of pigments per unit leaf mass. Again the pigments all have peaks in correlation with dR at the same wavelength, this time in the blue region, at 417.7 nm (R250.65* using a linear model). Percent canopy cover is the parameter least well correlated with dR with a maximum at 556.1 nm (R250.59*, using a linear model). As has been found for other vegetation types (e.g., Blackburn 1998b), the correlation between the amplitude of the second derivative of reflectance (ddR) and pigment concentrations per unit ground area is highly variable with wavelength and generally lower than that of dR. Therefore, while a number of peaks of high correlation occur, ddR in the SE590 wavebands either side of these peaks only have a weak correlation with pigments in many cases. ddR is most strongly correlated with the concentrations per unit ground area of Chl a, Chl b, Chl tot, and Cars at 753.1 nm (R250.87** for the chlorophylls, 0.83** for the Cars, using an exponential model). LAI has a reasonably strong linear relationship with ddR738.2 (R250.82**) while percent canopy cover has moderate exponential relationship with ddR804.3 (R250.61*). No relationships were found between ddR and pigment concentrations per unit leaf mass. Previous studies have identified the potential of the amplitudes of first and second derivatives of pseudoabsorbance (d(Log 1/R) and dd(Log 1/R)) for quantifying pigments at leaf and canopy scales (Yoder and PettigrewCrosby, 1995; Blackburn, 1998b). In the present experi-
R956.4/R756.1 0.94 R780.1/R747.1 0.77
R864.7/R807.4 0.94e R777.1/R747.1 0.72e
R836.1/R816.8 0.9e R836.1/R816.8 0.9e R836.1/R816.8 0.9e R836.1/R816.8 0.9e
dRmax 0.83 dRmax 0.83 dRmax 0.83 dRmax 0.83
ddR680-780 0.8 ddR680-780 0.8 ddR680-780 0.8 ddR680-780 0.8
dR735.3 0.88 dR556.1 0.59
dR417.7 0.65 dR417.7 0.65 dR417.7 0.65 dR417.7 0.65
dR750.1 0.9 dR750.1 0.9 dR750.1 0.9 dR750.1 0.9
ddR738.2 0.82 ddR804.3 0.61e
ddR753.1 0.87e ddR753.1 0.87e ddR753.1 0.87e ddR753.1 0.83e
a Figures shown in grey are the coefficient of determination with the subscripts e and p indicating that this is derived using exponential and power regression models respectively, while all other relationships are linear. Blank spaces indicate that no relationship was found between that category of spectral feature and canopy properties.
R783.2 0.58e R673.1 0.58p
R783.2 0.54e R673.1 0.55
R968.9/R931.2 0.89 R968.9/R931.2 0.89 R968.9/R931.2 0.89 R902.8/R893.3 0.87
Novel NarrowBand Ratios
Biophysical properties LAI Feature R2 % cover Feature R2
R800/R680 0.71 R800/R635 0.68
Other Existing Ratios
R798.3/R768.0 0.75 R798.3/R768.0 0.75 R798.3/R768.0 0.75 R792.2/R765.0 0.71
R783.2 0.50e R783.2 0.50e R783.2 0.50e R783.2 0.50e
Pseudo Absorbance (Log 1/R)
Concn per unit leaf mass (mg g21) Chl a Feature R2 Chl b Feature R2 Chl tot Feature R2 Cars Feature R2
Concn per unit ground area (mg m22) Chl a Feature R783.2 R2 0.54e Chl b Feature R783.2 R2 0.53e Chl tot Feature R783.2 R2 0.53e Cars Feature R783.2 R2 0.53e
Table 2. Summary of the Relationships Found between Spectral Features and Pigment and Biophysical Propertiesa
290 Blackburn and Steele
Towards the Remote Sensing of Matorral Vegetation Physiology
ment, these spectral transformations had weaker relationships with pigment and biophysical parameters than the corresponding first and second derivatives of reflectance. Nevertheless, some strong linear relationships were found, for example, between Chl a concentration per unit ground area and d(Log 1/R649.6) (R250.85**). CONCLUSIONS Table 2 is a summary of the relationships found between spectral features and pigment and biophysical properties of the matorral vegetation. Many of the relationships previously found between spectral reflectance characteristics and the pigment and biophysical properties of vegetation do not hold for the matorral vegetation canopies examined in this study. Within matorral vegetation, factors such as the heterogeneous structure, consequent shadowing effects, and background variability may be confounding such relationships, which have been found to exist at the individual leaf scale and for more uniform canopies where there is a lack of spectral complexity. This discrepancy between matorral and other vegetation types is apparent in the finding that the concentrations of pigments per unit ground area and LAI are related to percent reflectance in the near-infrared region but not in the visible. Moreover, for a range of existing spectral ratio indices, which were found to be of use in other vegetation types, there were either much weaker or no relationships with pigment concentrations and biophysical properties of matorral vegetation. Notably, broadband SR and NDVI were unrelated to either pigment concentrations, LAI or percent cover. However, a number of spectral characteristics/transformations were strongly related to the pigment and biophysical properties of matorral vegetation. A systematic investigation of narrow-band combinations employed in SR functions revealed that ratios of reflectance in nearinfrared bands were most closely related to pigment concentrations per unit ground area, while ratios of bands in the red region were also of value. LAI and percent cover were strongly related to ratios of bands on the near-infrared plateau and red edge features of canopy reflectance spectra, while pigment concentrations per unit leaf mass were correlated with ratios of bands around the near-infrared “shoulder.” While kRE was unrelated to any vegetation parameters, other characteristics of the red edge, quantified using derivative spectra were useful. Pigment concentrations per unit ground area and LAI were related to the maximum amplitude and the area of the first derivative in the region of the reflectance red edge, but most strongly correlated with the amplitude of the first derivative at specific wavelengths (dR750.1 and dR735.3, respectively). Furthermore, the amplitude of the first derivative at particular wavelengths in the visible was related to pigment concentrations per unit leaf mass and percent
canopy cover. Compared to these characteristics of the first derivative of reflectance, the amplitude of the second derivative of reflectance and the derivatives of pseudoabsorbance were less well correlated with vegetation parameters. Many thanks to the UK Natural Environment Research Council—Equipment Pool for Field Spectroscopy for the loan of the Spectron SE590.
REFERENCES Aoki, M., Yabuki, K., and Totsuka, T. (1981), An evaluation of chlorophyll content of leaves based on the spectral reflectivity in several plants. Res. Rep. Natl. Inst. Environ. Stud. Jpn. 66:125–130. Blackburn, G. A. (1998a), Spectral indices for estimating photosynthetic pigment concentrations: a test using senescent tree leaves. Int. J. Remote Sens. 19:657–675. Blackburn, G. A. (1998b), Quantifying chlorophylls and carotenoids from leaf to canopy scales: an evaluation of some hyper-spectral approaches. Remote Sens. Environ. 66:273–285. Blackburn, G. A. (in press), Relationships between spectral reflectance and pigment concentrations in stacks of deciduous broadleaves. Remote Sens. Environ. 70:226–239. Boochs, F., Dockter, K., Kupfer, G., and Kuhbauch, W. (1990), Shape of the red edge as a vitality indicator for plants. Int. J. Remote Sens. 11:1741–1754. Brandt, C. J., and Thornes, J. B. (1996) Mediterranean Desertification and Landuse, Wiley, Chichester. Carter, G. A. (1994), Ratios of leaf reflectances in narrow wavebands as indicators of plant stress. Int. J. Remote Sens. 15:697–703. Carter, G. A., Cibula, W. G., and Miller, R. L. (1996), Narrow band reflectance imagery compared with thermal imagery for early detection of plant stress. J. Plant Physiol. 148:515–522. Chappelle, E. W., Kim, M. S., and McMurtrey, J. E., III (1992), Ratio analysis of reflectance spectra (RARS): an algorithm for the remote estimation of the concentrations of chlorophyll A, chlorophyll B, and the carotenoids in soybean leaves. Remote Sens. Environ. 39:239–247. Curran, P. J. (1989), Remote sensing of foliar chemistry. Remote Sens. Environ. 30:271–278. Curran, P. J. (1994), Imaging spectrometry. Prog. Phys. Geogr. 18:247–266. Curran, P. J., Dungan, J. L. and Gholz, H.L. (1990), Exploring the relationship between reflectance, red edge and chlorophyll content in slash pine. Tree Physiol. 7:33–48. Curran, P. J., Dungan, J. L., Macler, B. A., and Plummer, S. E. (1991), The effect of a red leaf pigment on the relationship between red-edge and chlorophyll concentration. Remote Sens. Environ. 35:69–75. Danks, S. M., Evans, E. H., and Whittaker, P. A. (1983), Photosynthetic Systems. Structure Function and Assembly, Wiley, New York. Dell, B., Hopkins, A. J. M., and Lamont, B. B. (1986), Resilience in Mediterranean-Type Shrublands, Elsevier, Amsterdam.
Blackburn and Steele
Demarez, V., Gastellu-Etchegorry, P., Mougin, E., et al. (1999), Seasonal variation of leaf chlorophyll content of a temperate forest. Inversion of the PROSPECT model. Int. J. Remote Sens. 20:879–894. Di Castri, F., and Mooney, H. (1973), Mediterranean-Type Ecosystems, Chapman and Hall, London. Elvidge, C. D., and Chen, Z. (1995), Comparison of broadband and narrow-band red and near-infrared vegetation indices. Remote Sens. Environ. 54:38–48. Filella, I., and Pen˜uelas, J. (1994), The red edge position and shape as indicators of plant chlorophyll content, biomass and hydric status. Int. J. Remote Sens. 15:1459–1470. Filella, I., Serrano, L., Serra, J., and Pen˜uelas, J. (1995), Evaluating wheat nitrogen status with canopy reflectance indices and discriminant analysis. Crop Sci. 35:1400–1405. Gitelson, A. A. and Merzlyak, M. N. (1994), Quantitative estimation of chlorophyll-a using reflectance spectra: experiments with autumn chestnut and maple leaves. J. Photochem. Photobiol. B: Biol. 22:247–252. Gitelson, A. A., and Merzlyak, M. N. (1996), Signature analysis of leaf reflectance spectra: algorithm development for remote sensing of chlorophyll. J. Plant Physiol. 148:494–500. Gitelson, A. A., and Merzlyak, M. N. (1997), Remote estimation of chlorophyll content in higher plant leaves. Int. J. Remote Sens. 18:2691–2697. Gitelson, A. A., Kaufman, Y. J., and Merzlyak, M. N. (1996a), Use of a green channel in remote sensing of vegetation from EOS-MODIS. Remote Sens. Environ. 58:289–298. Gitelson, A. A., Merzlyak, M. N., and Lichtenthaler, H. K. (1996b), Detection of red edge position and chlorophyll content by reflectance measurements near 700nm. J. Plant Physiol. 148:501–508. Horler, D. N. H., Dockray, M., and Barber, J. (1983), The red edge of plant leaf reflectance. Int. J. Remote Sens. 4:273–288. Jacquemond, S., and Baret, F. (1990), PROSPECT: a model of leaf optical properties spectra. Remote Sens. Environ. 34:75–91. Kent, M., and Coker, P. (1994), Vegetation Description and Analysis, Wiley, Chichester. Kim, M. S., Daughtry, C. S. T., Chappelle, E. W., McMurtrey, J. E., and Walthall, C. L. (1994), The use of high spectral resolution bands for estimating absorbed photosynthetically active radiation. In ISPRS Sixth International Colloquium on Physical Measurements and Signatures in Remote Sensing, Val d’Ise`re, France, 17–21 January, European Space Agency, Paris, pp. 299–306. Lichtenthaler, H. K. (1987), Chlorophylls and carotenoids: pig-
ments of photosynthetic membranes. Meth. Enzymol. 148:350–382. Lichtenthaler, H. K., Gitelson, A. A., and Lang, M. (1996), Non-destructive determination of chlorophyll content of leaves of a green and an aurea mutant of tobacco by reflectance measurements. J. Plant Physiol. 148:483–493. Mariotti, M., Ercoli, L., and Masoni, A. (1996), Spectral properties of iron-deficient corn and sunflower leaves. Remote Sens. Environ. 58:282–288. Markham, B. L., Williams, D. L., Schafer, J. R., Wood, F., and Kim, M. S. (1995), Radiometric characterization of diodearray field spectroradiometers. Remote Sens. Environ. 51:317–330. Munden, R., Curran, P. J., and Catt, J. A. (1994), The relationship between red edge and chlorophyll concentration in the Broadbalk winter wheat experiment at Rothamsted. Int. J. Remote Sens. 15:705–709. Pen˜uelas, J., Baret, F., and Filella, I. (1995a), Semi-empirical indices to assess carotenoids/chlorophyll a ratio from leaf spectral reflectance. Photosynthetica 31:221–230. Pen˜uelas, J., Filella, I., Lloret, P., Mun˜oz, F., and Vilajeliu, M. (1995b), Reflectance assessment of mite effects on apple trees. Int. J. Remote Sens. 16:2727–2733. Pinar, A., and Curran, P. J. (1996), Grass chlorophyll and the reflectance red edge. Int. J. Remote Sens. 17:351–357. Rollin, E. M. (1992), Spectron SE590 Processing Software: SMENU and SBATCH, Primary Processing Routines, NERC-EPFS Software Manual 1 (Version 1.2), Department of Geography, University of Southampton, UK. Schepers, J. S., Blackmer, T. M., Wilhelm, W. W., and Resende, M. (1996), Transmittance and reflectance measurements of corn leaves from plants with different nitrogen and water supply. J. Plant Physiol. 148:523–529. Steele, C. M. (in press), The Remote Sensing of Vegetation Physiology, unpublished Ph.D. thesis, King’s College London. Vane, G., and Goetz, A. F. H. (1993), Terrestrial imaging spectrometry: current status, future trends. Remote Sens. Environ. 44:117–126. Yamada, N., and Fujimura, S. (1991), Nondestructive measurement of chlorophyll pigment content in plant leaves from three-color reflectance and transmittance. Appl. Opt. 30: 3964–3973. Yoder, B. J. and Pettigrew-Crosby, R. E. (1995), Predicting nitrogen and chlorophyll content and concentrations from reflectance spectra (400–2500 nm) at leaf and canopy scales. Remote Sens. Environ. 53:199–211. 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.