Remote sensing of nitrogen and lignin in Mediterranean vegetation from AVIRIS data

Remote sensing of nitrogen and lignin in Mediterranean vegetation from AVIRIS data

Remote Sensing of Environment 81 (2002) 355 – 364 Remote sensing of nitrogen and lignin in Mediterranean vegetation from ...

160KB Sizes 0 Downloads 3 Views

Remote Sensing of Environment 81 (2002) 355 – 364

Remote sensing of nitrogen and lignin in Mediterranean vegetation from AVIRIS data: Decomposing biochemical from structural signals Lydia Serranoa,*, Josep Pen˜uelasa, Susan L. Ustinb a

Unitat Ecofisiologia CSIC-CREAF, Centre de Recerca Ecolo`gica i Aplicacions Forestals (CREAF), Universitat Auto`noma de Barcelona, Edifici C, 08193 Bellatera, Barcelona, Spain b Department of Land, Air, and Water Resources, University of California, Davis, CA, USA Received 10 July 2001; received in revised form 28 November 2001; accepted 15 December 2001

Abstract Remote sensing estimates of vegetation nitrogen (N) and lignin concentration are central to assess ecosystem processes such as growth and decomposition. Although remote sensing techniques have been proven useful to assess N and lignin contents in continuous green canopies, more studies are needed to address their capabilities, particularly in low and sparsely vegetated ecosystems. We investigated the possibility of estimating canopy N and lignin concentrations in chaparral vegetation using Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) reflectance acquired over an area around Point Dume in the Santa Monica Mountains (Los Angeles, CA, USA). Two approaches were tested: multiple stepwise regression based on first difference reflectance (FDR) and reflectance (R) indices. Multiple stepwise regressions (of three or fewer wavelengths) accounted for a large variance in canopy biochemical concentration (r2  0.9, P < 0.01). Log transformed R indices [log (1/R)] formulated on the basis of previously known N and lignin absorption wavelengths also showed significant correlations ( P < 0.01) with canopy biochemical concentration (r2 ranging from 0.39 to 0.48). In addition, the contribution of structural and biochemical signals and background effects on the performance of these indices was evaluated. These indices accounted for a increased variance when adding information on canopy structural attributes (e.g., relative contribution of each species and biomass amount) to foliar biochemical concentration. The relative contributions of foliar biochemical concentration and canopy structure (biomass amount) on the spectral signal were further evaluated by analyzing the residuals from linear regressions: foliar N concentration accounted for 42% of the variance for a normalized difference index based on the 1510-nm N absorption feature, while the foliar lignin concentration accounted for 44% of the variance for a normalized difference index based on the 1754 nm lignin absorption feature. These percentages increased to 58% when stands with senescing vegetation were disregarded. We propose the two indices, Normalized Difference Nitrogen Index (NDNI = [log (1/R1510)  log (1/R1680)]/[log (1/R1510) + log (1/R1680)]) and Normalized Difference Lignin Index (NDLI = [log (1/R1754)  log (1/R1680)]/ [log (1/R1754) + log (1/R1680)]) as indices to assess N and lignin in native shrub vegetation. D 2002 Elsevier Science Inc. All rights reserved.

1. Introduction Several important ecosystem processes have been linked to plant biochemical composition, specifically to nitrogen (N) and lignin concentrations (Aber & Federer, 1992). N concentration in foliage is related to maximum photosynthetic rate and photosynthetic capacity (Field & Mooney, 1986) as well as to aboveground net primary production (Birk & Vitousek, 1986). Foliar lignin concentration is related to litter decomposition rates (Melillo, Aber, &

* Corresponding author. Tel.: +34-3-581-1877; fax: +34-3-581-1312. E-mail address: [email protected] (L. Serrano).

Muratore, 1982). Thus, plant biochemical composition is related to factors controlling growth and decomposition and it provides information on plant and ecosystem processes such as carbon (C) and N cycling (Aber & Federer, 1992). Remote sensing is a key tool for assessing vegetation condition over large areas, offering the possibility to analyzing ecological issues at a wide range of spatial scales (Hope, 1995; Ustin et al., 1991). Furthermore, remotely sensed data can be used as inputs in ecosystem models that are used to assess functional changes brought by climate variability and land use change (Scholes & Archer, 1997). Estimation of plant biochemical composition through high spectral resolution remote sensing has its foundation in laboratory near-infrared spectroscopy (NIRS) (Martin,

0034-4257/02/$ – see front matter D 2002 Elsevier Science Inc. All rights reserved. PII: S 0 0 3 4 - 4 2 5 7 ( 0 2 ) 0 0 0 11 - 1


L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364

1992; Williams & Norris, 1987). At the leaf level, NIRS has been successfully used to predict N, cellulose, and lignin concentrations on ground dried foliage (Joffre, Gillon, Dardenne, Agneessens, & Biston, 1992). Several studies have also demonstrated the utility of hyperspectral data to estimate canopy N and lignin content at increasing spatial scales. Improvements in intrinsic characteristics of spectrometers such the signal-to-noise ratio, as well as in methods to correct for undesirable atmospheric effects have led to the accuracy necessary for the estimation of the biochemical concentration of forest canopies (Curran, Kupiek, & Smith, 1997). Wessman, Aber, and Peterson (1989) reported large correlations between forest canopy biochemical composition and first-difference-at-sensor radiance measured by the Airborne Imaging Spectrometer (AIS). Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) spectra have been used to estimate lignin and N concentrations in mixed-species forest canopies (Johnson, Hlavka, & Peterson, 1994; Martin, 1994; Martin & Aber, 1997; Matson, Johnson, Billow, Miller, & Pu, 1994) and single-species canopies (Curran et al., 1997) using first difference reflectance (FDR) data and stepwise regression techniques. Although hyperspectral imagery has been used to detect changes in canopy biochemical concentration at the pixel level in dense vegetation (crops and forests), it has proven difficult on arid and semiarid ecosystems characterized by sparse vegetation and low soil cover (Asner, Wessman, Bateson, & Privette, 2000; Asner, Wessman, Schimel, & Archer, 1998). Moreover, equations formulated using multiple stepwise regression techniques have been reported to give inconsistent results when applied across different vegetation types and to yield inaccurate estimates when applied to data sets other than those where the relationships were developed (Grossman et al., 1996; Martin & Aber, 1997). In addition, the selected wavelengths in stepwise regression are not always related to the biochemical of interest, but to wavelengths that relate to biomass amount (Johnson et al., 1994), or to wavelengths associated to absorption by other biochemicals (Curran, Dungan, & Peterson, 2001). To overcome these drawbacks, several analytical techniques have been developed. Based on band depth analysis of continuum-removed reflectance spectra coupled with stepwise regression, Kokaly and Clark (1999) obtained good predictions of N, lignin, and cellulose contents on dried and ground leaves. Curran et al. (2001) using the latter methodology estimated the concentration of 12 foliar biochemicals with high accuracy. This methodology is not directly applicable to fresh whole leaves and canopies due to the presence of water (Kokaly & Clark, 1999), and, thus, has not yet been tested using field and airborne spectra. However, other techniques such as spectral matching techniques (Gao & Goetz, 1995) and radiative transfer models (Dawson, Curran, North, & Plummer, 1999) have been successfully applied to predict water and lignin concentration from AVIRIS data in forest canopies. We aimed to test the feasibility of reflectance

indices to estimate N and lignin contents at the landscape level. Band ratioing and normalization provide a simple and straightforward means to enhance the biochemical absorption signal in vegetation while minimizing background effects (Jackson & Huete, 1991). We were interested in studying to what extent foliar and canopy biochemical concentration could be estimated using hyperspectral reflectance data in Mediterranean ecosystems characterized by a relatively high soil cover combined with a low Leaf Area Index (LAI). We aimed (1) to compare the reliability of indices formulated with wavelengths related to known N and lignin absorption features and stepwise procedures to assess canopy N and lignin concentration and (2) to study the relative contributions of foliar biochemical concentration and canopy structure (standing biomass) on the proposed indices. Since, in contrast with previous AVIRIS studies, our study focuses on a complex mosaic of vegetation communities, species composition, as well as distinct functional types and vegetation structure, we finally aimed (3) to assess the dependence of reflectance-based biochemical estimates on confounding factors, such as exposed soil and nonphotosynthetic vegetation, across sites with different vegetation types in Mediterranean ecosystems.

2. Material and methods 2.1. Field site and vegetation description The study area was located in the Santa Monica Mountains (Los Angeles County, CA, USA), an east – west trending range along the Pacific Coast (3450N – 118400W). This region is characterized by a Mediterranean climate. In this study, three distinct chaparral communities were considered: ceanothus chaparral (dominated by Ceanothus spp.), chamise chaparral (dominated by Adenostoma fasciculatum), and coastal sage scrub (dominated by Salvia and Eriogonum spp. and Artemisia californica) (Hanes, 1988). Two sites with the abovementioned vegetation communities were selected for canopy measurements: Zuma ridge and Castro Crest. Zuma ridge is a coastal site near the Pacific coast with coastal sage scrub, while Castro Crest has characteristic ceanothus and chamise chaparral vegetation. Thirteen plots of ceanothus and chamise chaparral and 10 plots with characteristic coastal sage scrub (which exhibits pronounced seasonal leaf senescence) were studied. Stands of homogeneous vegetation — in correspondence to a plant community series as described in Sawyer and Keeler-Wolf (1995), aspect, and slope were chosen on the ground as the basic unit of study and their boundaries were traced in a map (Tom Harrison Trail Map of the Santa Monica Mountains Central, 1996, Tom Harrison Cartography, San Rafael, CA, USA), thus defining a series of polygons. Areas of individual polygons ranged from 4300 to over 29,000 m2.

L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364

Transects, ranging from 22 to 55 m in length, were located within these polygons. The point –intercept transect method with point measurements every 1 m (Sawyer & Keeler-Wolf, 1995) was used to assess vegetation composition and structure. At each point intercept, plant species was recorded and structure measured as height. When no vascular plant was hit, it was recorded as soil. Species absolute cover was determined as the percentage area of total polygon area it occupied, including non-vegetated areas, and species relative cover was calculated as the percentage of polygon vegetated area it occupied (Gordon & White, 1994). Measurements of vegetation composition and structure were taken within weeks before the AVIRIS flight.


2.2. Canopy biochemical concentration Samples to determine foliar N and lignin concentration were taken within 4 days of the AVIRIS flight. Within each polygon, N and lignin were determined by sampling five branches from the upper third of the canopy crown for each species. Five disks of 0.45 cm2 area each were punched using a cork borer on different leaves of each branch (each branch was considered a replicate). For needle-leafed species and for small leaves, the entire leaf was sampled. Due to time-labor constraints, only those species having an absolute cover > 10% were considered for each polygon. See Table 1 for a detailed list of the species sampled at each polygon.

Table 1 Vegetation community, cover type, species composition, and average foliar N and lignin concentration (% dry matter) of each species Polygon

Vegetation community

Functional type

Species composition

N (%)

Lignin (%)










4 5 6

ceanothus chamise ceanothus

green green + soil green

7 8

chamise ceanothus

green + soil green



green + soil




11 12

chamise chamise

green + soil green + soil


coastal sage



coastal sage



coastal sage



coastal sage


17 18 19 20

coastal sage coastal sage chamise coastal sage

senescing senescing green senescing


coastal sage


22 23

coastal sage coastal sage

senescing senescing

Ceanothus megacarpus Malosma laurina Adenostoma fasciculatum Ceanothus megacarpus Adenostoma fasciculatum Ceanothus megacarpus Ceanothus megacarpus Adenostoma fasciculatum Ceanothus megacarpus Ceanothus spinosus Quercus dumosa Adenostoma fasciculatum Ceanothus spinosus Cercocarpus betuloides Quercus dumosa Adenostoma fasciculatum Arctostaphyllos glandulosa Cercocarpus betuloides Heteromeles arbutifolia Adenostoma fasciculatum Cercocarpus betuloides Adenostoma fasciculatum Arctostaphyllos glandulosa Adenostoma fasciculatum Malosma laurina Salvia leucophylla Eriogonum cinereum Salvia leucophylla Artemisia californica Salvia leucophylla Artemisia californica Malosma laurina Salvia leucophylla Artemisia californica Salvia leucophylla Adenostoma fasciculatum Artemisia californica Salvia leucophylla Artemisia californica Salvia leucophylla Salvia mellifera Artemisia californica Malosma laurina Salvia leucophylla

1.65 1.67 1.26 1.70 1.10 1.60 1.72 1.04 1.38 1.98 1.76 1.03 1.64 1.68 2.04 1.13 0.91 1.50 1.42 1.25 1.76 0.85 0.97 1.01 1.71 1.82 – 1.31 2.16 5.63 2.07 1.64 1.41 2.03 1.65 1.05 1.71 1.49 1.96 1.63 1.40 1.93 1.61 1.48

21.8 4.5 14.1 30.7 11.5 19.2 20.8 23.5 21.6 7.1 9.6 7.1 17.5 12.9 11.9 10.3 13.7 11.4 12.0 10.5 10.1 11.4 12.8 14.3 7.7 25.2 21.0 5.7 12.2 5.2 16.2 6.0 9.4 12.9 8.1 11.8 13.9 6.1 9.6 7.4 7.4 8.8 3.5 6.7

(0.075) (0.053) (0.050) (0.035) (0.064) (0.008) (0.099) (0.055) (0.041) (0.060) (0.013) (0.066) (0.015) (0.090) (0.053) (0.013) (0.014) (0.091) (0.056) (0.018) (0.030) (0.016) (0.018) (0.030) (0.025) (0.168) (0.069) (0.146) (3.738) (0.177) (0.020) (0.040) (0.084) (0.098) (0.036) (0.150) (0.082) (0.047) (0.061) (0.020) (0.168) (0.045) (0.041)

Only those species with vegetation cover > 10% are listed. Cover types are indicated as following: green = green vegetation cover > 70% (green); green + soil = green vegetation cover < 70% (soil); senescing = nongreen vegetation. Values in parentheses are S.E.M. (n = 3).


L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364

C and N were determined by the CHN combustion method (Perkin-Elmer 2400) (Page, Miller, & Keeney, 1982) on three dried and ground replicates. Lignin was determined gravimetrically by using a sequential extraction/digest method (Van Soest & Wine, 1991) on the two remaining replicate samples (that were pooled to allow sufficient sample amount). Assessment of canopy biochemical content was conducted by weighing each species N (or lignin) concentration by its absolute cover (C) and height (as a surrogate of LAI) and summing the resulting values for all the species considered in a polygon. Since this is a rough approach to the actual canopy N (lignin) content, it will be called bulk canopy N (lignin) herein. Thus, Bulk canopy N ¼

i¼n X

ðNi Ci Hi Þ


where Ni is the foliar N concentration for the species i, Ci is the fraction of total area that the species i occupies in a polygon, and Hi is the mean height (in meters) of the species i. Bulk canopy lignin was equally assessed by using foliar lignin concentration instead of foliar N concentration. 2.3. AVIRIS data AVIRIS data were acquired over the Santa Monica Mountains on 17 October 1996 from an ER-2 aircraft at an elevation of approximately 20 km. Data presented in this study were obtained from Flight 961017B, Run 4, Scene 5, centered over Point Dume, Los Angeles County, CA, USA. A typical AVIRIS scene consists of 614 samples and 512 lines covering an 11  9 km2 area and 224 spectral bands from 370 to 2500 nm with a sampling interval of 10 nm (Green et al., 1998; Vane et al., 1993). AVIRIS data were radiometrically corrected and processed to retrieve apparent surface reflectance by using a modified version of the MODTRAN radiative transfer code (Green, Conel, & Roberts, 1993). AVIRIS-retrieved surface reflectance was registered by selecting ground control points. Image registration was performed with ENVI (Research Systems, Boulder, CO, USA) using linear equations, and resampling was based on the nearest neighbor method (to preserve the spectral values). The root mean square registration error was less than 1 pixel size (20  20 m). Polygons traced in the base map were reconciled to the registered scenes. Average reflectance spectra were extracted for each polygon. Regions comprised between 1400 –1500 and 1800– 1900 nm were disregarded because of strong atmospheric water absorption. 2.4. Data analysis We used reflectance (R), FDR, and log transformed reflectance [log (1/R)] spectra. Log (1/R) (also called pseudoabsorbance) is often used because it provides a curve

comparable to an absorption curve, with peaks occurring at the corresponding absorption wavelengths. Similarly, FDR approximates the slope of reflectance (first derivative), which is more closely related to absorption features than reflectance magnitudes per se (Dixit & Ram, 1985). A number of indices were derived from reflectance and log (1/R) data. These indices were formulated as ratios and normalized differences using known N and lignin absorption wavelengths (as reported in Curran, 1989) and a nearby reference wavelength. Because of the two strong water absorption bands located within the SWIR, we considered two subregions (from 1500 to 1800 nm and beyond 1900 nm) and a reference wavelength was selected on each spectral region. These reference wavelengths were located at 1680 and 2100 nm, respectively. The indices tested were formulated as follows: Ratio indices ¼ Rlabs =Rlref Normalized indices ¼ ðRlref  Rlabs Þ=ðRlref þ Rlabs Þ where Rl denotes reflectance (or other reflectance derived values) at the l wavelength and the subindices abs and ref denote the absorption and reference wavelengths, respectively. See Table 2 for a list of the precise absorption wavelengths used for the indices’ formulation in this study. The N and lignin absorption wavelengths were those reported in Fourty, Baret, Jacquemoud, Schmuck, and Verdebout (1996) based on previous information after Curran (1989) and Himmelsbach, Boer, Akin, and Barton (1988). We chose to test those wavelengths that unequivocally corresponded to either N or lignin (e.g., the 1200-nm wavelength was not used because it responds to lignin and water absorption features). Data were analyzed using correlation and regression routines in SPSS 10.0. Stepwise regression equations to predict foliar N and lignin concentration and bulk canopy N and lignin were developed from FDR full spectrum AVIRIS data. The regression model was run on the forward mode (the first step selected the FDR at a wavelength that was the best statistical predictor). The number of terms in the regression equation was set to a maximum of three to avoid overfitting (Wessman, 1990). Goodness of fit was evaluated on the basis of r values (r2 indicates the variance accounted for) and the standard error of calibration. The selected wavelengths were also examined to identify the source signals. In addition, linear regressions of the log (1/R)-based

Table 2 Absorption peaks of several biochemicals as reported in Fourty et al. (1996) based on previous works by Curran (1989) and Himmelsbach et al. (1988) Absorbing biochemical

Wavelength (nm)

Water Nitrogen Lignin

970, 1200, 1400, 1450, 1940 1020, 1510, 1730, 1980, 2060, 2130, 2180, 2240, 2300 1120, 1200, 1420, 1450, 1690, 1754, 1940, 2262, 2380

Absorption peaks used to derive ratio and normalized indices in this study are shown in bold characters.

L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364


3.1. Foliar N and lignin concentration

r2 than at the leaf level (Table 3). In particular, the 1748-nm wavelength corresponds to an N absorption peak, while 480 nm (Table 3) wavelength corresponds to absorption by carotenoids and chlorophyll, and chlorophyll content is highly correlated with total protein and N content (Field & Mooney, 1986). On the other hand, none of the selected wavelengths to predict bulk canopy lignin was related to a lignin absorption feature. Despite these inconsistencies in wavelength selection, particularly for lignin estimates, there was a close agreement between measured and predicted bulk canopy N and lignin (Fig. 1).

Foliar N concentration widely varied among species, according to the contrasting phenological stages and functional types. N concentrations ranged from 0.85 ± 0.02% SEM (Standard Error of the Mean, SEM) in Adenostoma fasciculatum to 2.16 ± 0.02% SEM in Artemisia californica. These were the lowest and highest values determined. By species and on average, the lowest and highest N concentrations were found in Arctostaphyllos glandulosa (0.94 ± 0.04% SEM) and Artemisia californica (1.98 ± 0.07% SEM), respectively. In evergreen species, we measured particularly low and high concentrations of N while in drought-deciduous species we found intermediate values. Foliar lignin concentration ranged from 4% in Malosma laurina to 31% in Ceanothus megacarpus. By species, average lignin concentration ranged from 8 ± 1.26% in Salvia mellifera to 18 ± 2.8% in Cenothus megacarpus.

3.3.3. Vegetation type Partitioning the data set according to vegetation type (i.e., ceanothus, chamise chaparral, and coastal sage scrub communities) yielded the highest r2 values (r2 > 0.97, P < 0.001), with varying wavelength selection among subsets. Bulk canopy N in ceanothus and chamise communities was predicted by near-infrared region wavelengths and by wavelengths associated to N absorption peaks (Table 4). For coastal sage scrub (which coincided with senescing vegetation), the wavelengths selected were related to N and to carotenoid and chlorophyll pigments. On the other hand, bulk canopy lignin was predicted by wavelengths related to N and chlorophyll absorption features, except in coastal sage scrub where one of the selected wavelengths was related to aromatic structures present in lignin (  2340 nm) (Table 5) (Osborne & Fearn, 1986).

3.2. Bulk canopy N and lignin

3.3.4. Vegetation cover Further categorization of polygons into green vegetation cover above and below 70% also provided a different wavelength selection. In polygons with green vegetation cover > 70%, bulk canopy N was predicted by wavelengths located at 1241 nm (a water absorption peak), 1748 nm (an N-protein absorption peak), and 761 nm (red edge region) (r2 = 0.995, P < 0.001). In polygons with green vegetation cover < 70%, wavelengths selected to predict bulk canopy N

normalized indices against Normalized Difference Vegetation Index (NDVI) were performed and the resulting residuals were, in turn, regressed against foliar biochemical concentration to evaluate the relative contribution of foliar biochemical concentration and canopy structure (biomass amount) to the reflectance signal.

3. Results and discussion

Characteristic attributes of the studied polygons are described in Serrano, Ustin, Roberts, Gamon, and Pen˜uelas (2000). As an indication, vegetation cover ranged from 67% to 100%, while NDVI varied from 0.33 ± 0.04 to 0.69 ± 0.07. As a result of large differences in canopy height, bulk canopy N varied by a factor of ca. 8 among polygons, while foliar N concentration varied by a factor of ca. 2. Similarly, bulk canopy lignin varied by a factor of ca. 18 while foliar lignin concentrations varied by a factor of ca. 5. 3.3. Stepwise regressions from FDR 3.3.1. Foliar estimates Wavelength selection to predict foliar N and lignin concentration from FDR data was consistent in the sense that the selected wavelengths were mostly associated to characteristic absorption peaks. One out of two wavelengths selected for foliar N concentration was related to an N absorption feature. Similarly, in predicting foliar lignin concentration a unique wavelength was selected that was attributable to a lignin absorption peak (data not shown). 3.3.2. Canopy estimates FDR data predicted bulk canopy N from three wavelengths located in the SWIR and visible regions with higher

Table 3 Regression terms and selected wavelengths by stepwise regression using first difference AVIRIS reflectance data to predict bulk canopy N and lignin for the whole data set (n = 23) Term

Wavelength (nm)

Bulk canopy N Intercept l1 1609 1748 l2 l3 480 Bulk canopy lignin Intercept l1 2048 l2 1619 l3 867





 9.00 23.70 18.44  14.21

0.34 0.65 0.75

0.003 0.001 0.013

0.837 0.636 0.551

8.13 105.22 175.96  127.01

0.48 0.75 0.81

< 0.001 < 0.001 0.026

8.537 6.004 5.392

The coefficients of determination (r 2), level of significance ( P ), and the root mean square error (RMSE) are also provided.


L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364

pounds as well as to absorption features associated to biomass amount (Johnson et al., 1994) or absorption by other compounds correlated to the biochemical of interest (Curran, Dungan, Macler, Plummer, & Peterson, 1992) and varied among data sets (Grossman et al., 1996). The stepwise multiple regressions accounted for a large variance, and their predictive power improved when considering separately vegetation types (Martin & Aber, 1997). 3.4. Reflectance-based spectral indices Reflectance-based indices formulated using wavelengths associated with known absorption features showed a slightly weaker relation with either bulk canopy N and or lignin than equations derived from stepwise regressions. We here report those that provided consistent and significant results across vegetation communities and functional types. The best predictor of bulk canopy N was derived from reflectance at 1510 nm (herein called Normalized Difference Nitrogen Index, NDNI), while for bulk canopy lignin the best indicator was derived from reflectance at 1754 nm (herein called Normalized Difference Lignin Index, NDLI). Bulk canopy N and lignin derived from the above mentioned wavelengths were similarly estimated with ratio indices than with normalized indices (data not shown). However, the latter are preferred over ratio indices due to their relative range of variation. Wavelengths similar to those used in the formulation of NDNI and NDLI have been proven useful to assess N and Fig. 1. Relationship between bulk canopy N and lignin predicted from stepwise regressions using first difference AVIRIS reflectance and measured bulk canopy N (A) and lignin (B), respectively (n = 23). Predicted and measured bulk canopy N and lignin have relative units (see text for a detailed explanation).The coefficients of determination are given on each panel.

were located at 2157 (close to an N absorption peak), 1252 (a water absorption peak), and 876 nm (a red edge region wavelength) (r2 = 1, P < 0.001). Although distinct from those wavelengths selected in polygons with green vegetation cover > 70%, they were still related to N and water absorption peaks and to wavelengths located in the red edge region. Selection of red edge wavelengths is consistent with previous studies (Johnson et al., 1994; Matson et al., 1994; Rock, Hoshizaki, & Miller, 1988), as reflectance in this region varies with varying chlorophyll concentration and biomass amount (Pen˜uelas & Filella, 1998). In polygons with green vegetation cover > 70%, selected wavelengths to predict bulk canopy lignin were located at 723, 1183, and 460 nm (r2 = 0.978, P < 0.001), which correspond to the red edge region and to a lignin and carotenoid absorption peaks, respectively. No wavelength was selected for predicting bulk canopy lignin when green vegetation cover was < 70%. Results for senescent vegetation coincide with the above reported results for coastal sage scrub. Thus, as reported in previous studies, the wavelengths selected were related to absorption by biochemical com-

Table 4 Regression terms and selected wavelengths by stepwise regression using first difference AVIRIS reflectance data to predict bulk canopy N Coefficient




10.54  36.20 3.03  10  6 1.34

0.95 1 1

0.005 0.001 < 0.001

0.225 0.029 0.001

 1.19 12.46 3.20 3.12

0.79 0.96 0.99

0.003 < 0.001 < 0.001

0.292 0.146 0.061

Ceanothus + chamise (green) (n = 13) Intercept 3.84 l1 2048 17.53 780  20.70 l2 l3 1649 20.45

0.57 0.86 0.91

0.003 < 0.001 < 0.001

0.875 0.529 0.433

Coastal sage (senescent) (n = 10) Intercept 4.22 l1 2346  8.22 l2 470 5.47 l3 1669  3.54

0.67 0.93 0.99

0.004 < 0.001 < 0.001

0.304 0.146 0.065


Wavelength (nm)

Ceanothus Intercept l1 l2 l3

(n = 5) 991 2008 867

Chamise (n = 7) Intercept l1 1599 l2 2256 l3 2346

The coefficients of determination (r2), level of significance ( P ), and the root mean square error (RMSE) are also provided. Data were grouped according to vegetation community and functional type (see text for more details).

L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364 Table 5 Regression terms and selected wavelengths by stepwise regression using first difference AVIRIS reflectance data to predict bulk canopy lignin Terms

Wavelength (nm)





Ceanothus (n = 5) Intercept l1 972 l2 668 905 l3

21.35 333.85 224.95  2.21

0.94 1 1

0.007 < 0.001 < 0.001

3.38 0.067 0.001

Chamise (n = 7) Intercept l1 1599 l2 2048 l3 668

 21.74 139.01 96.32 35.14

0.83 0.98 1

0.002 < 0.001 < 0.001

3.331 1.231 0.657

0.57 0.85

0.003 < 0.001

9.029 5.518

0.67 0.93 0.99

0.003 < 0.001 < 0.001

0.304 0.146 0.065

Ceanothus + chamise (green) (n = 12) Intercept  20.39 l1 2048 161.85 1609 208.16 l2 l3 Coastal sage (senescent) (n = 10) Intercept 5.35 l1 678  85.04 l2 1540 14.61 2346  11.30 l3

The coefficients of determination (r2), level of significance ( P ), and the root mean square error (RMSE) are also provided. Data were grouped according to vegetation community and functional type (see text for more details).

lignin at various scales and with different techniques. Lignin content was assessed by measuring the depth of the 1720-nm residue (that results from fitting AVIRIS data and apparent pure water reflectance spectra) (Gao & Goetz, 1995). A close wavelength (1700 or 1740 nm depending on the data set) was selected using stepwise regression methods to predict lignin from FDR in forests (Martin & Aber, 1997). Likewise, through stepwise regression procedures, a wavelength close to 1510 nm was selected to predict N concentration on a deciduous forest using AIS first and second difference radiance data (Wessman et al., 1989). When comparing indices derived from reflectance (R) from those derived from log (1/R), there was an increase in significance, particularly in N predicting indices. The correlation coefficient between NDNI and bulk canopy N increased from 0.54 ( P = 0.003) to 0.63 ( P < 0.001), while the increase in the correlation coefficient between NDLI and bulk canopy lignin was rather small (from 0.653, P = 0.001, to 0.696, P < 0.001) when using log (1/R) instead of R. Jacquemoud, Verdebout, Schmuck, Andreoli, and Hosgood (1995) at the leaf level and Gastellu-Etchegorry, Zagolsky, Mougin, Marty, and Giordano (1995) at the landscape level also reported slightly better results when using log (1/R) instead of R data. Although log (1/R) does not take into account the multiple scattering that takes place at the leaf and the canopy scales (Dawson et al., 1999), it roughly linearizes the absorption effects, which might explain the improved correlation when compared to R-derived indices. The low LAI values in this Mediterranean vegetation imply


little multiple scattering, and, thus, minor distortion of the biochemical signal, which might explain the rather small differences between R- and log (1/R)-derived data. 3.4.1. The relative contribution of foliar biochemical concentration and canopy structure For the sake of clarity, we only present herein the results derived from log (1/R) data. Correlation coefficients between NDNI and NDLI and biochemical concentration increased as additional information on canopy structural characteristics was added. NDNI was significantly correlated to foliar N concentration (r = 0.582, P = 0.004) (Fig. 2A). The significance slightly increased when the relative contribution of each dominant species was taken into account (r = 0.594, P = 0.003) and significantly improved when accounting for canopy biomass amount (r = 0.762, P < 0.001) (Fig. 2B). Similarly, NDLI was significantly correlated with foliar lignin concentration (r = 0.451, P = 0.031) (Fig. 3A). The correlation significantly increased when the relative contribution of each dominant species was taken into account (r = 0.545, P = 0.007), and increased again when accounting for canopy biomass amount (r = 0.696, P < 0.001) (Fig. 3B). Thus, NDNI and NDLI responded to both biochemical and structural features. We further evaluated the relative contribution of these chemical and structural signals by analyzing the residuals from linear regressions. On a first step, NDNI and NDLI were regressed against NDVI, a known indicator of green

Fig. 2. Relationships between the reflectance-based NDNI and (A) foliar N concentration (%) and (B) bulk canopy N. Bulk canopy N has relative units (see text for a detailed explanation). The coefficients of determination are given on each panel (n = 23).


L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364

ever, NDLI was significantly correlated with the species weighted foliar lignin concentration (r = 0.623, P = 0.023) and the correlation significantly increased when accounting for biomass amount (r = 0.764, P = 0.002). Moreover, when we removed the structural effects by regressing NDNI and NDLI vs. NDVI and analyzed the residuals vs. bulk canopy N and lignin relationships in these polygons, the N signal accounted for 59% of the NDNI variance (42% when considering the whole data set), while lignin accounted for 57% of NDLI variance (instead of 44%). Further categorization allowing to distinguish green vegetation cover below and above 70% did not improve the significance of the correlation between either NDNI or NDLI and bulk canopy N and lignin. Nonetheless, in polygons with green vegetation cover >70% (n = 8), NDNI and NDLI estimates benefited from additional canopy structural information (data not shown). Conversely, in polygons with green vegetation cover < 70% (n = 5), the correlations between NDNI and NDLI and bulk canopy N and lignin, respectively, were not significant (data not shown). However, both NDNI and NDLI were closely correlated with foliar N concentration (r = 0.972, P = 0.006, and r = 0.95, P = 0.013, Fig. 3. Relationships between the reflectance-based NDLI and (A) foliar lignin concentration (%) and (B) bulk canopy lignin. Bulk canopy lignin has relative units (see text for a detailed explanation). The coefficients of determination are given on each panel (n = 23).

canopy structure (Gamon et al., 1995; Pen˜uelas & Filella, 1998; Serrano, Gamon, & Pen˜uelas, 2000) to remove the structural effects. Afterwards, these residuals were regressed against bulk canopy N and lignin, respectively. The N signal accounted for 42% variance in NDNI, while lignin accounted for 44% of the NDLI variance. 3.4.2. Background effects When scaling from leaf to the canopy and landscape levels, changes in canopy structure and background contribution make it difficult to adequately solve for foliar biochemical components. We evaluated the effects of structural properties and the relative contribution of chemical and structural signals on the performance of NDNI and NDLI by considering green or senescing vegetation. In addition, in order to evaluate the effects of exposed soil, we considered polygons with green vegetation cover above and below 70%. In polygons with senescing vegetation (n = 10), the relationships between either NDNI or NDLI and foliar biochemical concentration or either bulk canopy N or lignin were not significant (data not shown). In contrast, in polygons with green vegetation (n = 13), NDNI was significantly correlated with foliar N concentration (r = 0.723, P = 0.005). The degree of correlation slightly increased when the relative contribution of each dominant species was taken into account (r = 0.724, P = 0.005) and it increased significantly when accounting for canopy biomass amount (r = 0.760, P = 0.003) (Fig. 4). In contrast, in polygons with green vegetation, the correlation between NDLI and foliar lignin concentration was not significant. How-

Fig. 4. Relationship between bulk canopy N and the reflectance-based NDNI (A) and relationship between bulk canopy lignin and the reflectancebased NDLI (B) in stands with green (filled symbols, n = 13) and senescing vegetation (open symbols, n = 10). Bulk canopy N and lignin have relative units (see text for a detailed explanation). The coefficients of determination are given on each panel.

L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364

respectively) and the correlations with foliar lignin concentration were not significant (data not shown). The lack of correlation between NDLI and the foliar and bulk canopy lignin in polygons with senescing vegetation or low vegetation cover agrees with the results reported by Dawson et al. (1999) who found that the lignin absorption feature located around 1720 nm was not able to track changes in canopy lignin at low levels of LAI and canopy cover. Moreover, it has been reported that in areas of low vegetation (LAI < 2.0), such as in our senescing vegetation plots, or sparse vegetation (LAI > 2 and vegetation cover < 70%), the tissue optical properties are not propagated to the pixel level (Asner et al., 2000). Thus, in agreement with previous studies, biochemical content at the canopy and landscape level was found to be determined by biochemical and structural signals (Johnson et al., 1994). In addition, the relative impact of foliar biochemical concentration, canopy, and landscape factors on pixel-level reflectance was found to vary with plant composition and structure (Asner, 1998). The results obtained are encouraging, since, to our knowledge, none of the previous studies conducted using airbone imaging spectrometers accounted for such a wide range of species composition, foliar anatomy and moisture, and canopy structure. More research is needed in order to test the predictive power of the proposed indices. New methodologies (such as the one reported by Kokaly & Clark, 1999) could also provide enhanced estimates of biochemical content. However, in contrast with this latter methodology, reflectance indices are simple and straightforward to obtain, and, thus, readily applicable at the landscape level.

4. Conclusions The spectral indices NDNI and NDLI were affected by canopy closure (soil exposed) and were unable to assess foliar or bulk canopy N and lignin in senescing vegetation. In contrast, in green continuous canopies, NDNI and NDLI provided good estimates of the bulk canopy N and lignin. The relative contribution of the biochemical signal to the reflectance signal was significant, up to 58% of the residual variance, after taking into account the structure of the green vegetation. NDNI and NDLI showed promising results for the remote sensing of N and lignin at the landscape level, particularly in low vegetation continuous green canopies, which makes them good candidates for a standard estimation of those two biochemicals of high relevancy for ecological processes.

Acknowledgments This research was supported by a National Aeronautics and Space Administration grant. The manuscript was completed under support by Spanish CICYT grants REN


2000-0278/CLI and CLI99-0479. A fellowship to L.S. from Ministerio de Ciencia y Tecnologia (Spain) is greatly appreciated.

References Aber, J. D., & Federer, C. A. (1992). A generalized, lumped parameter model of photosynthesis, evapotranspiration, and net primary production in temperate and boreal forest ecosystems. Oecologia, 92, 463 – 474. Asner, G. A. (1998). Biophysical and biochemical sources of variability in canopy reflectance. Remote Sensing of Environment, 64, 234 – 253. Asner, G. P., Wessman, C. A., Bateson, C. A., & Privette, J. L. (2000). Impact of tissue, canopy and landscape factors on the hyperspectral reflectance variability of arid ecosystems. Remote Sensing of Environment, 74, 69 – 84. Asner, G. P., Wessman, C. A., Schimel, D. S., & Archer, S. (1998). Variability in leaf and litter optical properties: implications for canopy BRDF model inversions using AVHRR MODIS, and MISR. Remote Sensing of Environment, 63, 200 – 215. Birk, E. M., & Vitousek, P. M. (1986). Nitrogen availability and nitrogen use efficiency in loblolly pine stands. Ecology, 67, 69 – 79. Curran, P. J. (1989). Remote sensing of foliar chemistry: review. Remote Sensing of Environment, 30, 271 – 278. Curran, P. J., Dungan, J. L., Macler, B. A., Plummer, S. E., & Peterson, D. L. (1992). Reflectance spectroscopy of fresh whole leaves for the estimation of biochemical composition. Remote Sensing of Environment, 39, 153 – 166. Curran, P. J., Dungan, J. L., & Peterson, D. L. (2001). Estimating the foliar biochemical concentration of leaves with reflectance spectrometry. Testing the Kokaly and Clark methodologies. Remote Sensing of Environment, 76, 349 – 359. Curran, P. J., Kupiek, J. A., & Smith, G. M. (1997). Remote sensing the biochemical composition of a slash pine canopy. IEEE Transactions on Geoscience and Remote Sensing, 35 (2), 415 – 420. Dawson, T. P., Curran, J. P., North, P. R. J., & Plummer, S. E. (1999). The propagation of foliar biochemical absorption features in forest canopy reflectance: a theoretical analysis. Remote Sensing of Environment, 67, 147 – 159. Dixit, L., & Ram, S. (1985). Quantitative analysis by derivative electronic spectroscopy. Applied Spectroscopy Reviews, 21, 311 – 418. Field, C., & Mooney, H. A. (1986). The photosynthesis – nitrogen relationship in wild plants. In: T. J. Givnish (Ed.), On the economy of plant form and function ( pp. 25 – 55). Cambridge, UK: Cambridge University Press. Fourty, Th., Baret, F., Jacquemoud, S., Schmuck, G., & Verdebout, J. (1996). Leaf optical properties with explicit description of its biochemical composition: direct and inverse problems. Remote Sensing of Environment, 56, 104 – 117. Gamon, J. A., Field, C. B., Goulden, M. L., Griffin, K. L., Hartley, A. E., Joel, G., Pen˜uelas, J., & Valentini, R. (1995). Relationships between NDVI, canopy structure, and photosynthesis in three Californian vegetation types. Ecological Applications, 5, 28 – 41. Gao, B.-C., & Goetz, A. F. H. (1995). Retrieval of equivalent water thickness and information related to biochemical components of vegetation canopies from AVIRIS data. Remote Sensing of Environment, 52, 155 – 162. Gastellu-Etchegorry, J. P., Zagolsky, F., Mougin, E., Marty, G., & Giordano, G. (1995). An assessment of canopy chemistry with AVIRIS — a case study in the Landes forest, south-west France. International Journal of Remote Sensing, 16 (3), 487 – 501. Gordon, H. J., & White, T. C. (1994). Ecological guide to Southern California chaparral plant series (Technical publication R5-ECOLTP-005). San Francisco, CA: USDA Forest Service, Pacific Southwest Region.


L. Serrano et al. / Remote Sensing of Environment 81 (2002) 355–364

Green, R. O., Conel, J. E., & Roberts, D. A. (1993). Estimation of aerosol optical depth and additional atmospheric parameters for the calculation of apparent reflectance from radiance measured by the Airborne Visible/Infrared Imaging Spectrometer. In: Summaries of the 4th Annual JPL Airborne Geoscience Workshop, vol. 1. AVIRIS Workshop, Washington, DC, October 25 – 29 ( pp. 73 – 76). Pasadena, CA: Jet Propulsion Laboratory. Green, R. O., Eastwood, M. L., Sarture, C. M., Chrien, T. G., Aronsson, M., Chippendale, B. J., Faust, J. A., Pavri, B. E., Chovit, C. J., Solis, M., Olah, M. R., & Williams, O. (1998). Imaging spectroscopy and the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Remote Sensing of Environment, 65, 227 – 248. Grossman, Y. L., Ustin, S. L., Jacquemoud, S., Sanderson, E. W., Schmuck, G., & Verdebout, J. (1996). Critique of stepwise multiple linear regression for the extraction of leaf biochemistry information from leaf reflectance data. Remote Sensing of Environment, 56, 182 – 193. Hanes, T. L. (1988). California chaparral. In: M. G. Barbour, & J. Major (Eds.), Terrestrial vegetation of California, Special Publication, 9 ( pp. 417 – 470). Sacramento, CA: California Native Plant Society. Himmelsbach, D. S., Boer, J. D., Akin, D. E., & Barton, F. E. (1988). Solid state carbon-13 NMR, FTIR and NIR spectroscopic studies of ruminant silage digestion. In: C. S. Creaser, & A. M. C. Davies (Eds.), Analytical applications of spectroscopy ( pp. 410 – 413). London: Royal Society of Chemistry. Hope, A. S. (1995). Monitoring ecosystem response to global change: high spatial resolution digital imagery. In: J. M. Moreno, & W. C. Oechel (Eds.), Global change and Mediterranean-type ecosystems, Ecological Studies, 117 (pp. 239 – 253). New York: Springer-Verlag. Jackson, R. D., & Huete, A. R. (1991). Interpreting vegetation indices. Preventive Veterinary Medicine, 11, 185 – 200. Jacquemoud, S., Verdebout, J., Schmuck, G., Andreoli, G., & Hosgood, B. (1995). Investigation of leaf biochemistry by statistics. Remote Sensing of Environment, 54, 180 – 188. Joffre, R., Gillon, D., Dardenne, P., Agneessens, R., & Biston, R. (1992). The use of near-infrared spectroscopy in litter decomposition studies. Annales des Sciences Forestieres, 49, 481 – 488. Johnson, L. F., Hlavka, C. A., & Peterson, D. L. (1994). Multivariate analysis of AVIRIS data for canopy biochemical estimation along the Oregon transect. Remote Sensing of Environment, 47, 216 – 230. Kokaly, R. F., & Clark, R. N. (1999). Spectroscopic determination of leaf biochemistry using band-depth analysis of absorption features and stepwise multiple linear regression. Remote Sensing of Environment, 67, 267 – 287. Martin, K. (1992). Recent advances in near-infrared reflectance spectroscopy. Applied Spectroscopy Review, 27, 325 – 383. Martin, M. E. (1994). Measurement of foliar chemistry using laboratory and airborne high spectral resolution visible and infrared data. PhD thesis, University of New Hampshire, USA.

Martin, M. E., & Aber, J. D. (1997). High spectral resolution remote sensing of forest canopy lignin, nitrogen and ecosystem processes. Ecological Applications, 7 (2), 431 – 443. Matson, P., Johnson, L., Billow, C., Miller, J., & Pu, R. (1994). Seasonal patterns and remote spectral estimation of canopy chemistry across the Oregon transect. Ecological Applications, 4, 280 – 298. Melillo, J. M., Aber, J. D., & Muratore, J. F. (1982). Nitrogen and lignin control of hardwood leaf litter decomposition dynamics. Ecology, 63, 621 – 626. Osborne, B. G., & Fearn, T. (1986). Near-infrared spectroscopy in food analysis. Harlow, UK: Longman Scientific & Technical. Page, A. L., Miller, R. H., & Keeney, D. (1982). Methods of soil analysis, 2. Madison, WI: American Society of Agronomy. Pen˜uelas, J., & Filella, I. (1998). Visible and near-infrared reflectance techniques for diagnosing plant physiological status. Trends in Plant Science, 3, 151 – 156. Rock, B. N., Hoshizaki, T., & Miller, J. R. (1988). Comparison of in situ and airborne spectral measurements of the blue shift associated with forest decline. Remote Sensing of Environment, 24, 109 – 127. Sawyer, J. O., & Keeler-Wolf, T. (1995). A manual of California vegetation. Sacramento, CA: California Native Plant Society. Scholes, R. J., & Archer, S. R. (1997). Tree – grass interactions in savannas. Annual Review of Ecological Systems, 28, 514 – 544. Serrano, L., Gamon, J. A., & Pen˜uelas, J. (2000). Estimation of canopy photosynthetic and non-photosynthetic components from spectral transmittance. Ecology, 81, 3149 – 3162. Serrano, L., Ustin, S. L., Roberts, D. A., Gamon, J. A., & Pen˜uelas, J. (2000). Deriving water content of chaparral vegetation from AVIRIS data. Remote Sensing of Environment, 74, 570 – 581. Ustin, S. L., Wessman, C. A., Curtiss, B., Kasischke, E., Way, J. B., & Vanderbilt, V. C. (1991). Opportunities for using the EOS imaging spectrometers and synthetic aperture radar in ecological models. Ecology, 72, 1934 – 1945. Vane, G., Green, R. O., Chrien, T. G., Enmark, H. T., Hansen, E. G., & Poter, W. M. (1993). The Airborne Visible Infrared Imaging Spectrometer. Remote Sensing of Environment, 44, 127 – 143. Van Soest, P. J., & Wine, R. H. (1991). Use of detergent in the analysis of fibrous feeds: IV. Determination of plant cell wall constituents. Journal Association of Official Agricultural Chemistry, 50, 50 – 55. Wessman, C. A. (1990). Evaluation of canopy biochemistry. In: R. J. Mooney, & H. A. Mooney (Eds.), Remote sensing of biosphere functioning ( pp. 135 – 156). New York: Springer-Verlag. Wessman, C. A., Aber, J. D., & Peterson, D. L. (1989). An evaluation of imaging spectrometry for estimating forest canopy chemistry. International Journal of Remote Sensing, 10 (8), 1293 – 1316. Williams, P., Norris K. (Eds.) (1987). Near-infrared technology in the agricultural and food industries. St. Paul, MN: American Association of Cereal Chemists.