| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
a USDA-Agricultural Research Service, Crop Science Research Laboratory, P.O. Box 5367, Mississippi State, MS 39762
b Texas A&M Research and Extension Center, 1509 Aggie Dr., Beaumont, TX 77713
c Plant and Soil Science Dep., Box 9555, Mississippi State, MS 39762
* Corresponding author (jjread{at}msa-msstate.ars.usda.gov)
Received for publication January 17, 2001.
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSNOTESRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Abbreviations: Chl, chlorophyll • NIR, near infrared range of the spectrum • R, leaf or canopy reflectance
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSNOTESRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Remote sensing in the visible and near-infrared (NIR) wavelengths (between about 400 and 1000 nm) is based on the principle that changes in plant light interception and utilization influence the reflectance properties of vegetation (Carter, 1998; Jensen, 2000). Nitrogen deficiency in cotton decreases chlorophyll (Chl) and soluble protein content (Longstreth and Nobel, 1980), as well as the rate of leaf expansion and canopy development (Reddy et al., 1997; Gerik et al., 1998). Remote sensing of Chl has potential for rapidly estimating cotton N status, and hence crop productivity. A critical need in PCM is the unambiguous detection of plant N status, preferably at an early stage of crop development, which requires identification of spectral wavebands or indices in which vegetation reflectance is most responsive to unfavorable growth conditions (Carter, 1994).
The dominant biochemical factor affecting canopy reflectance in the visible wavelengths is the concentration and total amount of leaf chlorophylls a and b (Chappelle et al., 1992; Filella et al., 1995). Using a priori knowledge of leaf physiology, some authors have related reflectance at 675 nm (R675), where Chl a absorption is maximal, to the concentration of total Chl (Jacquemoud and Baret, 1990; Filella et al., 1995). Chappelle et al. (1992) found R675/R700 in ratio to the value obtained from well-fertilized control plots yielded a precise relationship with Chl content in N-deficient soybean [Glycine max (L.) Merr.]. Others have proposed using R550 (the green peak where Chl a and b absorption is minimal), because relatively low leaf reflectance at 675 nm decreases sensitivity of the 675-nm band to changes in Chl a. Thomas and Gausman (1977) examined single leaf reflectance in three principle wavebands, and reported 550 nm was superior to either 450 or 675 nm for relating leaf reflectance to pigment concentration in cotton. Reflectance at 550 nm or its ratio with a waveband in the near-infrared was associated with corn (Zea mays L.) grain yield, an estimator of N status (Blackmer et al., 1996), and with chlorophyll loss during leaf senescence in horse chestnut (Aesculus hippocastanum L.) and Norway maple (Acer platanoides L.) (Gitelson and Merzlyak, 1996).
The position of the red edge, which is the sharp increase in spectral reflectance between 690 and 730 nm, may shift to shorter wavelengths as the concentration of Chl a decreases (Horler et al., 1983; Carter, 1994). This stress-sensitive region in digital imagery is often referred to as the "blue shift" (Jensen, 2000). Because the reflectance red edge in cotton leaves is known to increase under N stress (Tarpley et al., 2000), this spectral feature has potential as an index for the remote sensing of crop canopies deficient in N. Increased reflectance of approximately R700 results from strong absorption by Chl in the visible region, near 670 nm, and relatively high reflectance by leaf structural features in the near-infrared region, between 720 and 920 nm. Several studies have related the red edge, or single-waveband ratios involving the red edge, to changes in Chl concentration (Gitelson and Merzlyak, 1996), photosynthetic activity (Carter et al., 1996; Carter, 1998), or micronutrient content (Adams et al., 2000). The use of simple reflectance ratios may correct for variations in irradiance, leaf orientation, irradiance angles, and shading (Carter, 1994). In leaves of cotton, single-waveband ratios that combined a low reflectance, red-edge measure with a waveband of high reflectance in the NIR region (755 to 920 and 1000 nm) were very sensitive indicators of N concentration (Tarpley et al., 2000).
Leaf N may also be associated with the concentration of other pigments. Carotenoids are a group of yellow to orange pigments involved in light harvesting for photosynthesis, and may also serve to protect chlorophyll from O2 damage when irradiance levels or other stressors are excessively high. The ratio of carotenoids to Chl a typically increases under stress (Thomas and Gausman, 1977) or in senescing leaves (Gitelson et al., 1996). Normalized difference pigment index [NPCI, (R680 - R430)/(R680 + R430)] and pigment simple ratio (PSR, R430/R680) appear to be sensitive reflectance indices of environmental and phenological changes in the carotenoid to Chl a ratio (Filella et al., 1995).
Besides leaf constituent absorption, vegetation reflectance is affected by the contribution of stems and leaf orientation to canopy reflectance (Carter and Miller, 1994; Jensen, 2000). Therefore, it is generally believed that in situ measurements with a hand-held instrument may provide more direct and accurate assessments of leaf Chl. Moreover, leaf spectra are useful in explaining how canopy spectra are influenced by changes in leaf reflectance as compared with changes in canopy structure. Radiative transfer models, which are beyond the scope of this paper, can be used to simulate the spectral contribution of leaf biochemical constituents to the canopy reflectance (e.g., Jacquemoud and Baret, 1990).
The present study determined optical properties in full canopies of cotton (i.e., without soil background) that best correlate with leaf Chl and N concentration. Because N deficiency alters cotton growth and development, reflectance ratios were compared between measurements obtained at the level of single leaves and full canopies. Measurements were conducted under natural illumination conditions to directly relate vegetation reflectance with remote (e.g., airborne) observations. Simple waveband ratios were evaluated as indicators of Chl content, and hence the activity of green vegetation. Linear response functions were sought across a wide range of vegetation and phenological conditions, so the reflectance ratios would be more generally applicable, and to facilitate calibration and validation of the index. Plant N status, and hence N stress, was ascertained from total leaf N values, because they provide a reliable estimate of N accumulated by the leaf prior to sampling (Gerik et al., 1998). We expect reflectance-based algorithms that are sensitive to N deficiency symptoms would provide functionality in image processing of a cotton plantation, watershed, or an entire county, which along with ground truth data on soil chemical and physical properties can be used to generate a site-specific soil fertility map. The intended outcome is to apply the necessary amounts of fertilizer N when and where it is needed to optimize cotton N management and yield.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSNOTESRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
All control and border rows were fed a favorable supply of water and nutrients using half-strength Hoagland's nutrient solution via one Netafim dripper per pot rated at 1 L of liquid per hour. Plants were irrigated three times per day. If most plants were visibly wilted at midday, the duration of subsequent daily irrigations was increased to maintain soil water above the wilting point. A weather station adjacent to the field site continuously monitored radiation, precipitation, air temperature, and relative humidity, which aided in determining plant water requirements. The different N solutions for each treatment were contained in large mixing tanks and pumped under pressure by chemical metering pumps through plastic lines to the three-row treatment plots. All timing and duration of flows to individual plots was under computer-controlled switches and solenoid valves. At various times (listed below), the plants in some three-row plots were supplied nutrients with an osmotically balanced Hoagland's solution containing either 0 or 20% of control N. The five N treatments were: (i) half-strength Hoagland's supplied either up to and beyond the time N treatment was imposed (control), (ii) 20% of control N commenced 31 d after emergence at first floral bud (square) stage, (iii) 20% of control N commenced 52 d after emergence at flowering stage, (iv) 0% of control N commenced 52 d after emergence at flowering stage, and (v) 0% of control N commenced 89 d after emergence at fruit-filling stage.
Spectroradiometric and Physiological Measurements
Hyperspectral reflectance data from 350 to 950 nm with a resolution of about 1.5 nm were obtained from single leaves and full canopies (i.e., no soil background) with a GER 1500 spectroradiometer (Geophysical and Environmental Research Corp., Millbrook, NY). Measurements were taken at midday (1100 to 1300 h) under incident solar radiation from an angle perpendicular to the target. Four scans were averaged in every measurement. Immediately (2 min) prior to each set of reflectance measurements, a reference spectra was collected from a high reflectance, white reference panel (Spectralon; Labsphere, North Sutton, NH). Percent reflectance was computed by dividing observed radiance data by radiance measured from a white reference panel under approximately the same sun conditions, and multiplying by 100.
Leaf reflectance was measured from the uppermost, fully expanded leaves of five plants selected at random from each three-row treatment. Measurements were conducted with a 4° field of view foreoptic from a height of 0.3 m above the leaf. Thus, spectral reflectance was detected from an approximately 9.5-cm2 area (3.45 cm long x 2.75 cm wide) of the middle lobe. Because a fully absorptive (black) background was not placed behind a leaf during the measurement, we expect near-infrared reflectance is artificially enhanced, as compared with measurements obtained using an integrating sphere. This additive reflectance is due to the fact that healthy green leaves in the vicinity of our field of view may reflect approximately 80% of the incident near-infrared energy at 900 nm (Jensen, 2000). We minimized certain limitations to measuring reflectance under natural illumination conditions, such as specular reflectance (e.g., glare) and variability in irradiance angle and intensity by (i) positioning the radiometer between the sun and the target of the upper leaf (adaxial) surface, (ii) selecting the best angle with minimal specular reflectance, (iii) avoiding shadows, and (iv) completing measurements in <1 h.
Canopy reflectance was measured with a 23° field of view foreoptic from a height of 1 m above the central portion of a canopy (40-cm width). Thus, reflectance typically involved fully developed plant crowns arising from three consecutive pots, and virtually no soil background. A fiber optic cable was run to the radiometer along an extensible boom that attached to the top of a tripod, which also had a platform for attaching the spectroradiometer. Three separate rows were measured in each treatment by quickly positioning the boom as needed above each row. A nadir view (single viewing geometry) may be insufficient to fully characterize N stress in the cotton canopy. This is because measurements do not reveal different aspect of vertical structure, leaf area index, leaf orientation, and shading that have a strong influence on overall characteristics of the reflectance spectrum (Jensen, 2000). We expect these effects may largely cancel out as the number of measurements is increased, as in the present study. Further, our main intent was to compare reflectance indices between single leaves and canopies that correspond most strongly with Chl concentration. Further studies may require that the comparison of leaf and canopy reflectance be evaluated with radiative transfer models (Jacquemoud and Baret, 1990).
Immediately following leaf reflectance measurements, five uppermost, fully expanded leaves were harvested from each treatment, placed in a cooler over ice, and transported to the laboratory. A cork borer was used to remove five discs from each leaf, avoiding areas with large veins. Leaf discs were placed into vials containing 4.0 mL dimethyl sulfoxide (DMSO), and kept at room temperature overnight in the dark to extract pigments. Absorbances of the extract at 470, 648, and 664 nm were recorded and concentrations of carotenoid pigments and chlorophylls a, b and a + b were computed following the formulae of Chappelle et al. (1992). Chlorophyll concentration of the extract and the total disk surface area of 0.98 cm2 were used to compute total leaf Chl concentrations per unit projected leaf area. Fully expanded uppermost leaves were sampled for Chl because interaction with the incident light and proximity to the spectroradiometer would strongly affect canopy reflectance, and because these leaves express early symptoms of N stress, such as chlorosis. The five leaves used for pigment analysis were pooled and dried at 70°C for 72 h. Leaf tissue N concentration was subsequently determined on dried, ground samples according to standard micro-Kjeldahl methods (Nelson and Sommers, 1972), and expressed as either g kg-1 dry weight or g cm-2 leaf area.
Single-leaf gas exchange was determined during the boll filling period on 30 July and 5 August with a LI-COR (Lincoln, NE) Model LI-6400 photosynthesis system. Uppermost, fully expanded leaves were measured from five plants selected at random within each treatment, except 0% N at boll-filling onward, giving 20 observations per sampling date. Irradiance from a red–blue LED was set at 1500 µmol m-2 s-1 and ambient CO2 concentration was controlled at 350 µmol mol-1. Environmental conditions inside the leaf chamber were typically 27°C air temperature, 55% relative humidity, with a leaf-to-air vapor pressure deficit of about 1.4 kPa.
Data Analysis
A total of 73 observations for leaf pigments and 72 observations for leaf N concentration were generated from samples collected at approximately biweekly intervals. These data were analyzed for significant treatment effects (P < 0.05) across sampling dates with the proc MIXED procedure in SAS (SAS Institute, 1999). A balanced set of leaf constituent data (n = 69) is presented in Fig. 1
. This figure also shows the sampling dates when both leaf and canopy reflectance were measured, which represent leaf Chl and N data used in the correlation and regression analyses (described below).
|
|
|
|
|
We also used multiple linear regression to determine the combination of spectrally narrow canopy reflectance wavebands (1.5 nm resolution) that best explains variation in leaf N concentration. Because of the potential for inflation of r2 values (over-fitting data) when independent variables are autocorrelated, as is the case for spectral reflectance data (Thenkabail et al., 2000; Grossman et al., 1996), we used a partial least squares (PLS) procedure in SAS as a selection methodology (SAS Institute, 1999). Because PLS provides a list of all independent variables in the model, the first step involved ranking the 360 single wavebands based on two SAS outputs, the standardized regression coefficients and the Variable Importance for Projection (VIP) (Lindgren et al., 1994). Whereas the regression coefficients represent the importance each predictor has in the prediction of just the response, the VIP represents the value of each predictor in fitting the PLS model for both predictors and response. Sixty of the highest-ranked wavebands were then selected as inputs in the SAS-MAXR procedure, so the model would have fewer predictor variables (single wavebands) than dependent variables (66 observations for leaf N), and thus minimize the potential of over-fitting. We then selected the best 20-term model based on MAXR regression analysis. Then, the next 40 wavebands identified in the PLS ranking were included in the MAXR model statement, and the regression analysis was repeated, keeping the best 20-term model each time. This step was repeated until the PLS ranking list was exhausted. The best combination of wavebands was obtained from the final MAXR iterative analysis.
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSNOTESRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Consistent with other N-deficit studies in cotton (Reddy et al., 1997), leaf N reached a minimum of about 18.7 g kg-1 dry weight on 9 July, and then remained level in plants grown at 20% control N from first square stage onward. Averaged across sampling dates, leaf N values were 41.2, 35.3, 31.3, 26.2, and 26.3 g kg-1 in controls, 0% N at boll-filling, 20% N at first flower, 20% N at first square, and 0% N at first flower, respectively (Fig. 1c). Gerik et al. (1998) reported that cotton leaf N levels > 45 g kg-1, 30 to 45 g kg-1, 25 to 30 g kg-1, and <25 g kg-1, respectively, are considered excessive, sufficient, low, and deficient. Deficient levels of N were recorded on most sampling dates among plants supplied 20% of control N at first square stage onward; whereas leaves from the 0% N at first flower treatment ranged from low to deficient in N across sampling dates (Fig. 1c).
Leaf N was weakly associated with either Chl a (r2 = 0.37, df = 70) or total Chl (r2 = 0.32); however, the relationship with Chl improved somewhat when N was expressed as g m-2 leaf area (r2 = 0.42; df = 49; some samples were ground prior to weighing). Luxury consumption of N may weaken the relationship between Chl and N, but was of little consequence in the present study because relatively few values for leaf N exceeded 45 g kg-1 (Gerik et al., 1998). Further, leaf N values ranged from 18.7 to 56.0 g kg-1, which is very similar to the range reported by Tarpley et al. (2000) for either growth-chamber plants with restricted N supply or field-grown plants with favorable nutrient conditions. Earlier studies indicate cotton photosynthesis would change about 50% across this range of deficient to excessive leaf N values, with both linear and quadratic responses being observed (Reddy et al., 1997; Gerik et al., 1998).
Leaf photosynthetic rates measured during the boll-filling period increased as Chl concentration increased across the four N treatments (Fig. 2) . Photosynthesis was lower on 5 August than 30 July in the present study, and was associated with a 50% reduction in leaf conductance to water vapor due in part to somewhat greater leaf-to-air vapor pressure deficit (1.3 vs. 1.6 kPa). Longstreth and Nobel (1980) demonstrated that photosynthetic potential in cotton relates strongly to Chl concentration, which was directly correlated with plant mineral status, notably N and potassium, and to a lesser extent phosphorus.
|
Spectral Reflectance Properties in Single Leaves and Full Canopies
When plot means were analyzed across all N treatments and sampling dates, changes in Chl were best correlated with single-leaf reflectance values centered at 755 nm (r2 = 0.15), and to a lesser extent 695 nm (Fig. 3a)
. Because R755 is near a sun-induced fluorescence peak at approximately 760 nm, stress-induced changes in R755 may be influenced by factors other than Chl content. Nevertheless, Theisen (2000) reported that ratios utilizing far-red fluorescence at 755.56 nm consistently yielded strong correlation with leaf Chl in N-stressed bean (Phaseolus vulgaris L.). Sensitivity of a red edge feature at 695 nm is less tenuous, and represents the transition (or degree of slope) between maximal Chl absorption at approximately R675 and minimal absorption of light at approximately R700.
Because no further absorption occurs beyond approximately 750 nm, NIR reflectance becomes primarily a function of leaf structure and scatter. This suggests N-deficit treatments in the present study led to changes in both the pigmentation and internal structure of leaves. However, caution must be used when interpreting results for any single-waveband reflectance feature measured under naturally sunlit conditions. This is because reflectance at one determined wavelength is sensitive to the geometric arrangement between the leaf and canopy surface and sensor (the viewing geometry), variable irradiance, background effects, and green biomass. Carter (1994) proposed computing single-waveband reflectance ratios between a "stress sensitive" and "stress-insensitive" band as a way to correct for these variable effects, and subsequent studies have confirmed the validity of this method (Carter and Spiering, 2000, and references therein). Single-waveband ratios are mathematically similar to commonly used vegetation indices (Jensen, 2000), but can be optimized for detection of leaf N status (Carter and Spiering, 2000).
The association between Chl and leaf reflectance improved substantially when reflectance at 755 nm, a relatively Chl-insensitive waveband, was divided by reflectance values centered at either 705 or 555 nm (Fig. 3b). Reflectance at 705 nm is inversely related to changes in Chl a content, whereas R750 serves as an internal reference with respect to spectral reflectance due to leaf internal structure. In previous studies, R700/R750 was linearly related to Chl content in tobacco (Nicotiana tabacum L.) genotypes (Lichtenthaler et al., 1996), and in senescing leaves of two deciduous tress species (Gitelson et al., 1996). Sensitivity at R700 is a red-edge measure that typically occurs in the 680 to 760 nm range, and appears to be more strongly dependent on Chl content than reflectance near 550 nm (Gitelson and Merzlyak, 1996; Carter et al., 1996; Barnes et al., 2000; Carter and Spiering, 2000). Reflectance at about 550 nm or its ratio with a NIR waveband was closely associated with plant N status in corn (Blackmer et al., 1996), as well as two deciduous tree species with relatively low N content (Gitelson and Merzlyak, 1996). Overall, the best correlation with Chl in the present study (approximate r2 = 0.56) was obtained for leaf reflectance ratios R705/R930 and R705/R715 (Fig. 3c).
Similar to changes in Chl in our study, leaf N was best correlated with leaf reflectance values centered at 755 nm (r2 = 0.13), and to a much lesser extent at 695 nm (Fig. 4a) . The association with N improved substantially when R755 was divided by reflectance at either 700 or 590 nm (Fig. 4b), but the best correlation was obtained for reflectance ratios R700/R710 (r2 = 0.38) and R700/R755 (r2 = 0.34) (Fig. 4c). Because R700/R710 is essentially a feature of the red edge, we expect both wavebands are sensitive to Chl a, and sensitivity of the ratio is a result of variability in the reflectance response to early loss of Chl (chlorosis) in young, green leaves (see Fig. 2 in Gitelson et al., 1996). The proximity of these two wavebands would present difficulties in radiometer construction. Moreover, the commonly reported ratio of red edge to NIR reflectance, either R705/R930 for total Chl (Fig. 3c) or R700/R755 for N (Fig. 4b), was consistently identified as a sensitive spectral feature of N status.
Our success in identifying spectral indices sensitive to leaf N in cotton follows from the strong association between leaf Chl and N content observed in this and other nutrient stress studies (Longstreth and Nobel, 1980). For instance, the predominance of spectral features along the red edge in the present study is due to a combination of strong absorption by Chl near 670 nm, and high reflectance by leaf structural features between 720 and 920 nm (Jensen, 2000; Gitelson et al., 1996; Chappelle et al., 1992). In previous studies, single-leaf reflectance near 700 nm or in ratio with wavebands in the near-infrared region were closely associated with stress-induced changes in leaf N (Tarpley et al., 2000). They found that about half of the single-band reflectance ratios related to N concentration had r2 values of 0.64 or greater, and ranged up to 0.92. The best leaf reflectance ratios identified in the present study explained only about 35 to 40% of the variation in leaf N (Fig. 4).
Unexpectedly, treatment and seasonal changes in leaf Chl and N concentration were more closely associated with spectral reflectance from canopies (Fig. 5 and 6) than from leaves (Fig. 3 and 4). Further, leaf-level and canopy-level reflectance differed in the spectral bands with maximal sensitivity to plant N status. Presumably, these differences are due to additional factors, such as the influence of shadows, leaf density and orientation, and stems and/or flowers on canopy optical properties under N-limiting conditions (Carter et al., 1996; Barnes et al., 2000). In regard to single wavebands, the strong correlation observed between leaf N status and near-infrared reflectance (approximately 755 nm) in leaves was absent in canopies (Fig. 5 and 6). Multiple leaf layers reflect strongly in the near-infrared region due to additive reflectance (Jensen, 2000), but our results indicate N deficiency had little affect on either the density of leaves or number of leaf layers in cotton. The association in leaves between about 560 and 590 nm apparently shifted to higher wavebands between about 605 and 625 nm in canopies (Fig. 3a, 4a, 5a, and 6a), a region closely related to Chl a absorption (Chappelle et al., 1992). As with leaf reflectance data, canopy reflectance was best correlated with leaf Chl concentration at 700 nm (r2 = 0.52), but also at 625 nm (r2 = 0.43) (Fig. 5).
The canopy spectral association with Chl improved somewhat when reflectance at 700 nm was divided by reflectance at 415 nm (r2 = 0.67) or 675 nm (r2 = 0.50) (Fig. 5b). Because Chl a and carotenoid pigments absorb strongly at 420 nm (Chappelle et al., 1992), a relatively large amount of Chl must be lost before reflectance increases significantly at this wavelength. This suggests canopy ratios involving R415 are ratioing to more stable regions of the spectrum, as compared with more pronounced changes along the red edge under N stress. At 605 nm, and particularly at 695 and 710 nm, the absorptivity of Chl a and b is relatively weak (Lichtenthaler et al., 1996), approaching zero at wavelengths near 720 nm when Chl remains associated with chloroplast membranes. Thus, as Chl begins to decrease under N stress, leaf reflectance increases first and then most dramatically in the red edge due to further narrowing of the major Chl a absorption peak at 680 nm (Carter et al., 1996). Because canopy reflectance ratios (or indices) include the effect of both biomass and pigments, the significance of R675 is likely a function of strong absorption by leaves between 660 and 690 nm due to changes in Chl (Chappelle et al., 1992). Overall, the best coefficient of determination was obtained from the reflectance ratio R415/R695 (r2 = 0.72) (Fig. 5c). This resulted because reflectance typically decreased at 415 nm and increased at 695 nm as Chl decreased under N stress. Stress-induced loss of Chl also was associated with changes in R415/R605 (r2 = 0.68) (Fig. 5c), due mostly to increased reflectance at 605 nm by N-deficient leaves. Due in part to the close association between Chl and carotenoids in the present study, canopy reflectance was best correlated with carotenoid pigments at R415/R685 (r2 = 0.79), R415/R625 (r2 = 0.78), and R415/R505 (r2 = 0.56) (data not presented).
Unlike leaf reflectance data, leaf N was best correlated with canopy reflectance wavebands centered in the blue region of the spectrum (maximum r2 = 0.31 at 410 nm), and to a lesser extent in the far-red region, at about 700 nm (Fig. 6a). Strong correlation was obtained between leaf N and canopy reflectance when R415 was divided by either R710 (r2 = 0.70) or R585 (r2 = 0.67) (Fig. 6b). A slightly lower coefficient of determination was obtained for R710/R405 (r2 = 0.63), which is nearly the inverse of R415/R710 (Fig. 6c).
Among the various canopy reflectance ratios identified from correlation analysis, the most sensitive indicator of N stress in cotton appeared to be R360/R710, R415/R585, R415/R710, and R415/R695 (Fig. 7) . Somewhat better relationships with leaf N were obtained with quadratic rather than simple linear functions for each of these single waveband ratios. A weak correlation was obtained between changes in leaf N and R755/R700 in canopies, but this relationship was significant (r2 = 0.82) when N values > 40 g kg-1 were excluded (Fig. 7f, dashed line). Although close associations were obtained between R755/R700 and Chl concentration in the present study (canopy: r2 = 0.56, Fig. 5; leaf: r2 = 0.51, Fig. 3), variability in R755/R700 at high Chl values suggests this canopy reflectance ratio may not be a sensitive indicator of N stress in cotton (Fig. 7f).
|
Canopy Reflectance Optimization with Multiple Linear Regression
The best three-waveband model for predicting leaf N from canopy spectral reflectance had an r2 value of 0.80, as compared with r2 of 0.92 for the best seven-waveband model (Table 1). The three-waveband model identified R437, R610, and R689 spectral regions within 25 nm of the canopy reflectance ratios closely correlated with leaf Chl (R605, Fig. 5) and N (either R415, R585, or approximately R700; Fig. 6). Most regression models shared R437 and R610. The best six- and seven-waveband models identified R755±5 (Table 1), a reflectance region closely associated with N status in plants under N-limiting conditions (Fig. 3, 4, and 7). None of the models identified R700±10, which was a significant spectral region in the present study and a measure commonly associated with changes in leaf Chl and photosynthesis in other hyperspectral reflectance studies (Chappelle et al., 1992; Carter et al., 1996). Strong colinearity of this red-edge reflectance feature with the other spectral variables may have decreased the contribution of R700 to the model (Gitelson and Merzlyak, 1996). Models with more than four variables indicated that R912, R890, and R852 were important in explaining observed variation in leaf N of cotton. Although NIR reflectance in canopies (R710
755) was insensitive to changes in N status, its association improved when expressed in ratio to 415 nm (Fig. 7). In contrast, NIR measures in single leaves were very sensitive to changes in Chl and N under restricted N supply (Fig. 3 and 4). Unfortunately, these single-waveband regression models may not be easily transferred to other situations or crops in the field, because as mentioned previously, viewing geometry and canopy structure have a strong effect on overall characteristics of the reflectance spectrum (Jensen, 2000). Such effects on single waveband reflectance are typically modeled using radiative transfer approaches (Jacquemoud and Baret, 1990).
|
|
CONCLUSIONS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSNOTESRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
Our analysis to discern spectral reflectance ratios sensitive to leaf N status in cotton support results of Tarpley et al. (2000). They found that leaf reflectance ratios between wavebands in the red edge (700–716 nm) and a waveband in the very near infrared region (755–920 nm) provided good prediction of leaf N concentration in cotton. A noticeable difference in the present study is the prevalence of changes in the 400- to 450-nm spectral range in canopies. The best correlation with Chl was obtained with R415/R695 in canopies. Multiple linear regression found that R437, R610, and R689 in canopies were important in explaining the observed N-induced and seasonal variation in leaf N concentration.
Somewhat larger correlation was obtained between plant N status and canopy reflectance than leaf reflectance. Further, the spectral signature(s) differed between the two sightings, with reflectance measurements at either 410 ± 5 or 700 ± 10 nm in canopies and in the 695- to 755-nm range in leaves being the most sensitive indicators of plant N status.
|
ACKNOWLEDGMENTS |
|---|
|
NOTES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSNOTESRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSNOTESRESULTS AND DISCUSSIONCONCLUSIONSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  D. R. Tilley, M. Ahmed, J. H. Son, and H. Badrinarayanan Hyperspectral Reflectance Response of Freshwater Macrophytes to Salinity in a Brackish Subtropical Marsh J. Environ. Qual., April 5, 2007; 36(3): 780 - 789. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. A. Poss, W. B. Russell, and C. M. Grieve Estimating Yields of Salt- and Water-Stressed Forages with Remote Sensing in the Visible and Near Infrared J. Environ. Qual., May 31, 2006; 35(4): 1060 - 1071. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 P. J. Starks, D. Zhao, W. A. Phillips, and S. W. Coleman Development of Canopy Reflectance Algorithms for Real-Time Prediction of Bermudagrass Pasture Biomass and Nutritive Values Crop Sci., February 24, 2006; 46(2): 927 - 934. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M-C. Belanger, A. A. Viau, G. Samson, and M. Chamberland Determination of a Multivariate Indicator of Nitrogen Imbalance (MINI) in Potato Using Reflectance and Fluorescence Spectroscopy Agron. J., October 19, 2005; 97(6): 1515 - 1523. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
D. Zhao, K. R. Reddy, V. G. Kakani, J. J. Read, and S. Koti Selection of Optimum Reflectance Ratios for Estimating Leaf Nitrogen and Chlorophyll Concentrations of Field-Grown Cotton Agron. J., January 1, 2005; 97(1): 89 - 98. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||
