Relationships Between Canopy Chemistry and Reflectance for Plant Species from Jasper Ridge Biological Preserve, California

Yaffa L. Grossman, Eric W. Sanderson, and Susan L. Ustin
Department of Land Air and Natural Resources,
University of California, Davis, CA 95616 USA
Address for Correspondence:
Dr. Yaffa L. Grossman
Department of Pomology
University of California
Davis, CA 95616
phone: (530) 752-1843
Fax: (530) 752-8502
email: ylgrossman@ucdavis.edu

Abstract

Our study examined the statistical relationships among a number of foliar biochemicals including carbon, cellulose, lignin, nitrogen, and water on a leaf area (content) and dry weight basis (concentration). Foliar reflectance measurements of 484 samples taken from 17 naturally occurring species from Stanford University's Jasper Ridge Biological Preserve were made over five dates during 1992-1993. Chemical determinations of total carbon, nitrogen, lignin, cellulose, and water were made on a subset of the data (26-38 samples depending on chemical). The contents and concentrations of some of these biochemicals were highly correlated with one another, but the degree of correlation varied with the weight or area basis used. Stepwise multiple linear regression between 800 and 2498 nm with a maximum of six regressors explained at least 89% of the variance when the second derivative of log(1/R) was used as the independent variable, where R is reflectance at a given wavelength. However, the bands selected by the regressions did not correspond to known absorption features and were completely different if the chemical data were expressed on a dry weight or an area basis. In general, more of the variance was explained using an area basis than a weight basis. Principal components analysis separated individual species from one another. Stepwise multiple linear regression using the first five principal component scores explained lower proportions of the variance than did regression using log(1/R) and its derivatives. However, this approach may be preferred because it retains information from all spectral bands.

INTRODUCTION

The development of airborne, high resolution, narrow band infrared sensors may make detection of biologically important compounds in plant canopies on regional scales possible (Curran, 1989). Several studies have correlated narrow band reflectance values with measured biochemical concentrations (e.g. Wessman et al. 1988, Card et al. 1988), however, the bands selected by multiple linear regression are not consistent between datasets and often do not correspond to known absorption bands for the biochemicals of interest.

Empirical techniques using laboratory-based Near Infrared Spectroscopy (NIRS) have been developed to determine protein, carbohydrate, and other biochemical contents from dried plant materials (Marten et al., 1989; Barton et al., 1992). However, because the spectrum of water dominates fresh leaf reflectance (Elvidge, 1990), it remains to be determined whether these biochemicals can be detected in living plant material at the leaf level and at the canopy level.

This study investigates whether the carbon, cellulose, nitrogen, lignin, water, and dry weight composition of fresh leaves can be detected using the infrared reflectance at the leaf level.

METHODS

Seventeen plant species were selected from several California coastal plant communities at Jasper Ridge Biological Preserve, Stanford University, Stanford, California (Table 1). Leaf samples (approximately 10 replicate samples per plant, 82 plants over entire sampling, 484 reflectance spectra) were collected for reflectance measurements and biochemical analysis on five dates between June 1992 - May 1993. In most cases, leaves were collected from the same plant on each sampling date.

Immediately after sampling, reflectance measurements were made with an NIRS spectrometer. Measurements were made between 400-2498 nm at 2 nm intervals with a full width-half maximum slit width of 10 nm, although this study only reports the results from the infrared portion of the spectrum. One standard sized leaf disk was punched from each leaf, weighed, and backed with a black background for reflectance measurements. Following reflectance measurements, leaf disks were weighed, wrapped in aluminum foil, frozen in liquid nitrogen, lyophilized, weighed and electronically scanned for area measurement. The difference in weight between fresh and dried sample was attributed to water content.

An additional 15-20 leaves (bulk samples) of each plant species were wrapped in aluminum foil and frozen in liquid nitrogen for later biochemical analysis. The samples were stored on dry ice during transport, lyophilized, and ground through a 1 µm screen using a ball mill grinder. Leaf powders were stored at -70oC, desiccated.

A subset of the lyophilized bulk samples were sent to the laboratory of Dr. John Aber, University of New Hampshire for biochemical analysis. Total carbon and nitrogen were determined with a Perkin-Elmer 2400 CHN Elemental Analyzer. Cellulose and lignin were determined by proximate analysis, a technique of sequential extractions in dichloromethane, water and sulfuric acid. The sulfuric acid fraction represents cellulose and the residue represents lignin (Effland, 1977).

RESULTS AND DISCUSSION

Chemistry data

The chemistry data represents information from 38 plants and spans a broad range of values. The relationships between the chemicals were analyzed on fresh weight, dry weight, and area bases, and summarized using cluster analysis (Figure 1, Table 2). Total carbon, lignin, cellulose, and water clustered together, with cluster coefficients greater than 0.49. Water was not included in the cluster when it was expressed on an area basis. Total carbon and water contents were inversely correlated, and total nitrogen was uncorrelated with the other chemicals on all bases.
 

Spectral data

The average log(1/R) spectrum of all the leaves in the dataset shows two primary absorption features, near the water absorption bands at 1450 and 1940 nm, with the largest deviations from the average curve occurring near these features. Principal component analysis of log(1/R)-spectra between 800 and 2498 nm separated leaves into species-specific clusters, suggesting that spectral differences are species specific. The first five principal components explained 99.9% of the variance in the spectral dataset.
 

Stepwise multiple linear regression

Stepwise multiple linear regression between 800 and 2498 nm with a maximum of six regressors was performed using log(1/R)-transformed reflectance values and their first and second derivatives as the independent variables (Table 3). Multiple spectra taken from one plant were averaged and matched with the chemical data for the same plant. The second derivative regressions explained at least 89% of the variance in the chemistry data, and were the best fit regressions for all data except leaf dry weight per unit area for which the first derivative explained an additional 1% of the variance. The log(1/R) and first derivative regressions explained lower proportions of the variance all other cases, but of at least 73% of the variance.

In all cases, more of the variance was explained when the chemistry data were expressed on a content (g/unit area) basis than when they were expressed on a concentration (g/g) basis. This probably resulted from the fact that the spectrometer beam probes the leaf on an area basis.

The bands selected by the regressions differed depending upon whether log(1/R), first derivative or second derivative were used. Similar to the findings of Curran et al. (1992), the selected bands rarely corresponded to known features for the chemical being examined. In addition, band selection was heavily dependent upon the basis on which the chemistry was expressed. The bands selected for regressions of each chemical on dry weight and area bases coincided less than 10% of the time.

Wessman et al. (1988) found that the nitrogen concentration of dry leaves was best predicted using the first derivative of log(1/R) and that lignin concentration was best predicted using the second derivative of log(1/R). Using their coefficients on our dataset explained 64% of the variance in nitrogen concentration but none of the variance in lignin concentration. Thus, at least for lignin, the coefficients selected in that study were not useful with the present dataset.

Several lines of evidence suggest that correlations obtained from multiple linear regressions lack robustness for estimates of the biochemical contents or concentrations of fresh leaves. These are the lack of relationship between the bands selected and known absorption features, the inconsistency with which bands are selected, and the lack of a significant regression using lignin coefficients from another study. In addition, the majority of the spectral information is lost when using multiple linear regression with a small number of regressors.

Principal components analysis reduces the dimensionality of the reflectance data while preserving the information represented by reflectance in each of the 850 spectral bands between 800 and 2498 nm. Stepwise multiple linear regression using the first five principal component scores for each reflectance spectrum and the biochemistry dataset explained 49-91% of the variance in the chemistry content (area basis) and 23-69% of the variance in the chemistry concentration (dry weight basis), except that no significant regression was found for cellulose concentration. For all chemicals, more of the variance was explained by regressions using principal component scores for content rather than concentration (Table 4).

Principal components analysis was also used to remove the effects of water spectrum by summing the product of the water content (g/m2) of each sample and log(1/T) (Norris and Williams, 1987) for water for each band and partialling this sum out of the analysis. The first five principal components explained 99.8% of the variance in the spectral dataset. Stepwise multiple linear regression using the first five principal component scores explained 40-82% of the variance in the chemistry content and 40-62% of the variance in the chemistry concentration, except that no significant regression was found for cellulose concentration. Removing the water spectrum improved the fits for carbon, nitrogen, and lignin concentration by 24, 15 and 5% respectively, and improved the fit for nitrogen content by 12%. Removing the water spectrum degraded the fits for carbon, cellulose, lignin, and leaf content by 2, 9, 3 and 4% respectively.

CONCLUSIONS

Almost all of the variance in carbon, cellulose, nitrogen, and lignin content and concentration, and water and leaf dry weight content, were explained by second derivative log(1/R). However, the bands selected did not correspond to known absorption features and were different from those reported for dry leaves. Principal components analysis explained lower proportions of the variance than were explained by the regressions, but may prove to be more robust because they retain the full spectral information, rather than focusing on only a few bands. Analysis of a larger reflectance/biochemistry database using both stepwise multiple linear regression and principal components analysis is currently being planned.

Acknowledgments

This research was performed under NASA NAS5-31359 and NAS5-31714. The computer analyses were performed on Digital Equipment Corporation DEC 5000 and DEC Alpha computers provided under the Sequoia 2000 research program.

REFERENCES

Barton, F.E., Himmelsbach, D.S., Duckworth, J.H., Smith, M.J. "Two-dimensional vibration spectroscopy: Correlation of mid- and nir-infrared regions." Applied Spectroscopy 46 (1992): 420-429.

Card, D.H., Peterson, D.L., Matson, P.A., Aber, J.D. "Prediction of leaf chemistry by the use of visible and near infrared reflectance spectroscopy." Remote Sensing of the Environment 26 (1988):123-147.

Curran, P.J. "Remote sensing of foliar chemistry." Remote Sensing of the Environment 30 (1989):271-278.

Curran, P.J., Dungan, J.L., Macler, B.A., Plummer, S.E., Peterson, D.L . "Reflectance spectroscopy of fresh whole leaves for the estimating of chemical concentration." Remote Sensing of Environment 39 (1992): 153-166.

Effland, M. "Modified procedure to determine acid-insoluble lignin in wood and pulp." TAPPI 6 (1977):10.

Elvidge, C.D. "Visible and near infrared reflectance characteristics of dry plant materials." International Journal of Remote Sensing (1990) 11:1775-1795.

Marten, G.C, Shenk, J.S, Barton II, FE (eds.) Near Infrared Reflectance Spectroscopy (NIRS): Analysis of Forage quality. U.S.D.A. Handbook 643. (1985).

Norris, K.H., Williams, P. Near-infrared technology in the agricultural and food industries. St. Paul, Minn.: American Association of Cereal Chemists, 1987, 293-313.

Wessman, C.A, Aber, J.D, Peterson, D.L., Melillo, J.M. "Foliar analysis using near infrared reflectance spectroscopy." Canadian Journal of Forestry 18 (1988):6-11.

1998, Center for Spatial Technologies and Remote Sensing (CSTARS)
University of California, Davis