Remote Sensing Letters
Vol. 3, No. 4, July 2012, 285–294
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
Spectroscopic detection of nitrogen concentrations in sagebrush
JESSICA J. MITCHELL*†, NANCY F. GLENN‡, TEMUULEN T. SANKEY‡,
DEWAYNE R. DERRYBERRY§, MATTHEW O. ANDERSON¶
and RYAN C. HRUSKA¶
†Department of Geosciences, Idaho State University, Idaho Falls, ID 83402, USA
‡Department of Geosciences, Idaho State University, Boise, ID 83702, USA
§Department of Mathematics, Idaho State University, Pocatello, ID 83209, USA
¶Idaho National Laboratory, Idaho Falls, ID 83402, USA
(Received 10 February 2011; in final form 7 April 2011)
The ability to estimate foliar nitrogen in semi-arid landscapes can yield information on nutritional status and improve our limited understanding of controls on
canopy photosynthesis. We examined two spectroscopic methods for estimating
sagebrush dried leaf and live shrub nitrogen content: first derivative reflectance
(FDR) and continuum removal. Both methods used partial least squares (PLS)
regression to select wavebands most significantly correlated with nitrogen concentrations in the samples. Sagebrush dried leaf spectra produced PLS models (R2
= 0.76–0.86) that could predict nitrogen concentrations within the data set more
accurately than PLS models generated from live shrub spectra (R2 = 0.41–0.63).
Inclusion of wavelengths associated with leaf water in the FDR transformations
appeared to improve regression results. These findings are encouraging and warrant further exploration into sagebrush reflectance spectra to characterize nitrogen
concentrations.
1.
Introduction
Spectroscopy laboratory experiments and empirical studies link foliar nitrogen concentrations to narrow absorption regions in the visible and near-infrared portions
of the spectrum (e.g. Bolster et al. 1996, Kokaly and Clark 1999, White et al. 2000,
Kokaly 2001). Absorption regions in the visible range are associated with photosynthetic pigments (centred near 490 and 680 nm), while absorption regions in the
near-infrared are associated with protein bonds and can be variable due to absorption overlap (Curran 1989, Barton et al. 1992). Recent hyperspectral remote sensing
studies have successfully mapped canopy nitrogen at landscape and regional scales for
forested ecosystems using derivative transformation (e.g. Card et al. 1988, Johnson
et al. 1994, Townsend et al. 2003, Ollinger and Smith 2005) and continuum removal
(Huang et al. 2004, Mutanga et al. 2004, Huber et al. 2008) methods. Both approaches
are designed to enhance the signal of narrow absorption band features. The extent to
which similar methods can be applied to sparsely vegetated systems has been largely
unexplored.
*Corresponding author. Email: [email protected]
Remote Sensing Letters
ISSN 2150-704X print/ISSN 2150-7058 online © 2012 Taylor & Francis
http://www.tandf.co.uk/journals
DOI: 10.1080/01431161.2011.580017
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
286
J. J. Mitchell et al.
Leaf water is a major challenge to extending near-infrared spectroscopy (NIRS)
techniques from the dried leaf scale to live shrub (field) and canopy (remote sensing)
scales because strong water absorption bands tend to mask subtle absorption features
associated with constituents such as nitrogen, lignin and cellulose (e.g. Curran 1989,
Kumar et al. 2001). Laboratory studies traditionally apply derivative transformations
to smoothed reflectance data (i.e. logarithm of the inverse of reflectance) to minimize
noise and background signal variation and resolve overlapping absorption features
(Hruschka 1987). Kokaly and Clark (1999) illustrated the use of continuum removed
reflectance (Clark and Roush 1984) and band normalization to reduce the influence
of leaf water, as well as other factors such as soil and sensor noise on biochemical
estimations at field and remote sensing scales. Spectral transformation techniques
with relatively low sensitivity to partial canopy coverage hold particular promise in
semi-arid landscapes, where changes in leaf chemistry are not thought to translate to
landscape level detection unless cover exceeds 70% (Asner et al. 2000).
The objective of this study is to determine if signals from individual shrub canopies
are sufficient for estimating foliar nitrogen concentrations. In so doing, the use of continuum removal to estimate sagebrush foliar nitrogen concentrations is also evaluated,
and changes in signal response when scaling from dried leaf to live shrub are examined.
Findings are strategic in determining the feasibility of implementing similar endeavours from airborne platforms across larger spatial scales in semi-arid landscapes.
Remote estimation of sagebrush foliar nitrogen would allow for direct estimates
of nutritional status and contribute to assessments of habitat quality, productivity,
plant/soil water dynamics and controls on canopy photosynthesis.
2. Materials and methods
2.1 Study area
The study area is located in a cold desert sagebrush-steppe environment on the
Department of Energy, Idaho National Laboratory (INL), in eastern Idaho. The sampling area consists of all land within an 805 m radius of an unmanned aerial vehicle
test runway (43◦ 35 N; 112◦ 54 W). Wyoming big sagebrush (Artemisia tridentata ssp.
wyomingensis) is the dominant shrub species, while basin big sagebrush (A. tridentata
ssp. tridentata) occurs in association with depressional areas and drainage channels. Other species common to the area include yellow rabbit brush (Chrysothamnus
viscidiflorus), prickly pear cactus (Opuntia spp.) and crested wheatgrass (Agropyron
cristatum). Plant water stress in this ecosystem is minimal in the spring and early summer, increases during the mid and late summer and is greatest in August (DePuit and
Caldwell 1973). Accordingly, sagebrush crude protein content is usually highest in
the spring and gradually decreases into winter. We inferred from local precipitation
records that leaf water content may have been higher than usual in early and midsummer 2009, when this study was conducted. Total precipitation recorded by the
Atomic City NOAA mesonet station for the months of May, June and July were relatively wet (142.24 mm) in 2009, compared to corresponding precipitation amounts for
2008 (37.85 mm), 2007 (37.78 mm) and 2006 (80.26 mm).
2.2 Sampling design
A total of 36 spatially isolated sagebrush (Wyoming and basin big sage) were sampled
for leaf- and shrub-level nitrogen content. Five of the shrubs were sampled on 21 May
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
Sagebrush nitrogen spectroscopy
287
2009 and the remaining shrubs were sampled from 6 to 9 July 2009. We identified
individual sagebrush sample locations by generating random points, locating these
points in the field, and then selecting the nearest ‘suitable’ sagebrush. For the purpose
of this study, an individual sagebrush was considered ‘suitable’ if it was short enough
to be measured with a field spectroradiometer held by a scientist (<1 m) and spatially
isolated (unclustered), to avoid confounding measurements. The centre location of
each shrub was recorded using a wide-area augmentation system (WAAS)-enabled
Trimble GeoXT (Trimble Navigation Ltd., Sunnyvale, CA, USA) model GPS receiver
and locations were differentially corrected using Trimble Pathfinder software (Trimble
Navigation Ltd.).
2.3 Dried leaf measurements
Sagebrush green-leaf samples were collected from each individual shrub. Two longstem specimens (∼40 cm in length) containing representative leaf forms (i.e. ephemeral
and deciduous) were clipped from the top of each sagebrush canopy, stored in
polyethylene bags and refrigerated until laboratory analysis. Single-sided leaf mass
per unit area measurements were made for each fresh leaf specimen using a Li-COR
Biosciences area metre (model LI-3100; Lincoln, NE, USA). The samples were then
oven dried (60◦ C for 72 hours), ground in a Wiley mill, passed through a 1 mm mesh
screen, and weighed to the nearest 0.1 mg. Leaf level nitrogen concentrations (g N/100
g sample) of oven dried ground foliage were determined using a Leco TrueSpec CN
analyser (St. Joseph, MI, USA). Reflectance spectra (350–2500 nm) were collected for
optically dense dried and ground leaf samples using a FieldSpec® Pro spectroradiometer and a high intensity contact probe attachment (Analytical Spectral Devices (ASD)
Inc., Boulder, CO, USA). The spectral resolutions of the instrument are 3 and 10 nm
and the sampling intervals are 1.4 and 2.0 nm for the regions 350–1000 and 1000–
2500 nm, respectively. For each sample, a series of 20 replicates were obtained and the
probe was then re-calibrated using a white reflectance cap accessory available from the
manufacturer.
2.4 Live shrub measurements
Sagebrush top-of-canopy absolute reflectance measurements (350–2500 nm) were collected in the field using a FieldSpec® Pro spectroradiometer (see Section 2.3). For
each shrub, a series of four measurements (15 replicates per measurement) were made
holding a bare fibre tip (25◦ ) 25–40 cm above the shrub, which equates to a field of
view approximately 11–18 cm in diameter. The height at which the foreoptic was held
was optimized in the field based on the quality of signature. Reflectance was calibrated between samples using a white spectralon panel (Labsphere Inc., North Sutton,
NH, USA) and collection was limited to within 2 hours of solar noon under clear-sky
conditions.
2.5 Spectral data processing
Leaf- and shrub-level reflectance data (see figure 1) were transformed using standard derivative analysis and continuum removal methodologies (Kokaly and Clark
1999). Both transformations were applied directly to the spectrometer data and/or to
a smoothed version of the spectrometer data. We opted to smooth the spectrometer
data by resampling to coarser HyMap sensor channel configurations (472–2488 nm;
288
J. J. Mitchell et al.
90
80
Reflectance (%)
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
70
60
50
40
30
20
10
0
350
550
750
950
1150 1350 1550 1750 1950 2150 2350
Wavelength (nm)
Figure 1. Average reflectance spectra for sagebrush dried, ground leaf samples (thin line) and
live shrub canopy samples (bold line). Strong atmospheric water absorption window (1800–2000
nm) in the live spectra is not shown.
∼15–20 µm bandpass and sampling interval), because the authors are familiar with
the sensor and HyMap imagery has been successfully used in recent remote sensing
of foliar biochemistry studies (Huang et al. 2004, Huber et al. 2008, Skidmore et al.
2010). Standard derivative analysis was implemented by calculating first derivative
reflectance (FDR) values for log-transformed reflectance data (log10 (1/R), where R
is reflectance) using ViewSpec Pro software version 6.0 (ASD). The FDR values are
equivalent to the slope and were calculated every 1 nm as the difference between
wavelengths located three sample points away on either side of the wavelength being
calculated. Three is the smallest gap setting in ViewSpec Pro and was selected to avoid
smoothing narrow absorption features that may be related to nitrogen concentration.
The second type of transformation combines continuum removal with normalized band depth. Continuum removal estimates and removes absorptions that are not
related to the feature of interest (Clark and Roush 1984), while band depth normalized
to the band centre compares the shape of absorption features. Changes in shape are
correlated to differences in foliar chemistry. A broadening of the absorption feature,
for example, at 2100 nm occurs with an increase in nitrogen concentration (Kokaly
2001). Continuum removal was applied to select wavelength intervals with known
nitrogen associations (Curran 1989) using Spectral Library Builder in Environment
for Visualizing Images version 4.7 (ITT Visual Information Solutions, Boulder, CO,
USA). Interval endpoints were chosen independently for the set of dried leaf samples and for the set of live shrub canopy samples by identifying local spectral maxima
locations (see table 1). To minimize the masking effect of leaf and atmospheric water
absorption, nitrogen absorption features at 1020, 1510 and 1940 nm were excluded
from continuum removal analysis. NBD were then calculated following the methods
described by Kokaly and Clark (1999).
Sagebrush nitrogen spectroscopy
289
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
2.6 Relating foliar nitrogen concentrations to spectral measurements
To begin relating foliar nitrogen concentrations to spectral measurements, we generated correlograms and examined general patterns between nitrogen concentrations and dried leaf and live shrub spectral measurements (reflectance, coarsened
reflectance, continuum removed reflectance and FDR). Partial least square (PLS)
regression was then used to select wavebands most significantly correlated with nitrogen concentrations in the samples (Kokaly et al. 2009). Wide-kernel PLS regression
models were developed between six different sets of reflectance measurements and the
foliar nitrogen samples analysed in the laboratory using the ‘plsdof’ (PLS degrees of
freedom) package (Krämer and Sugiyama 2010) in version 2.11.1 of the statistical
software program R (The R Foundation for Statistical Computing, Vienna, Austria).
Four of the spectral measurement sets we tested consisted of NBD values calculated
for pre-defined absorption features (see table 1) – two sets using high-resolution leaf
and shrub spectra and two sets using coarsened leaf and shrub spectra. The two
remaining sets of spectral measurements were full-spectrum FDR values calculated
using the coarsened leaf and shrub spectra. PLS regression models were generated
using up to 11 components (m) for each of the six data sets, and models were selected
using Akaike’s Information Criterion (AICc) values adjusted for small sample size
(Burnham and Anderson 1998, Krämer et al. 2009). Wavelength selection was based
on regression coefficients generated in the best PLS models. Selection performance
was evaluated in the context of association with known nitrogen foliar chemistry (electron transitions in chlorophyll and nitrogen bonds in proteins). We also experimented
Table 1. The location of absorption features selected for continuum removal.
Dried leaf spectra (n = 36)
Short-wavelength
endpoint (nm)
Live shrub spectra (n = 36)
Long-wavelength
endpoint (nm)
Short-wavelength
endpoint (nm)
Long-wavelength
endpoint (nm)
455
597
1650
2010
525
742
1778
2207
465
558
1680
2058
540
742
1742
2180
2217
2383
2243
2286
HyMap-simulated bandwidths
Known
nitrogen
absorption
features (nm)
430, 460
640, 660
1690
2060, 2130,
2180
2240, 2300∗
HyMap-simulated bandwidths
472
590
1650
2005
531
736
1784
2205
476
560
1675
2043
546
736
1737
2187
2223
2391
2223
2391
430, 460
640, 660
1690
2060, 2130,
2180
2240, 2300∗
Notes: ∗ Due to excessive noise beyond 2300 nm in the live shrub spectra, the 2300 nm nitrogen
absorption feature was not included. Selected for continuum removal along with absorption
features of known association with nitrogen in ground and dried leaves, as reported by Curran
(1989), are shown in the table.
290
J. J. Mitchell et al.
with calculating FDR values from the high-resolution spectra, using normalized difference nitrogen indices (Serrano et al. 2002) and removing the five May samples from
the analysis. Results either were too poor to report or did not improve the regression
models.
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
3. Results and discussion
Nitrogen levels in the dried, ground Wyoming big sagebrush samples (n = 36) did not
vary widely. Concentrations ranged from 1.54% to 2.78%, with an average of 2.15%
(see table 2). Differences in the spectra of dried, ground leaf material and live shrub
canopy (see figure 1) were mostly attributable to leaf water. The spectral data sets
highlight an increased reflectance from leaf drying and a drastic change in the shape
of the red-edge shoulder. These findings are comparable to a similar illustration of
oak leaves in Kokaly et al. (2003), where absorption features (980 and 1200 nm) in the
near-infrared plateau (750–1300 nm) are more pronounced in the fresh plant spectra
and absorption features in the dried leaf spectra near 1730, 2100 and 2130 nm are not
masked by leaf water features (1400 and 1900 nm).
In general, the dried leaf and live shrub wavelengths most relatable to nitrogen
concentration occurred in association with the absorption spectra of plant pigments
(β-carotene (500 nm), phycoerythrin (550 nm), phycocyanin (620 nm) and chlorophylla (660 nm)), the red-edge position (670–780 nm), water absorption features (1400
and 1940 nm) and biochemical features (1020, 2210 and 2310 nm). These correlogram results (data not shown) are consistent with the sagebrush morphology and
pigmentation characteristics.
Table 2. Average foliar nitrogen concentration (% dried matter) of 36 Wyoming big sagebrush
shrubs.
Site
Nitrogen (%)
SD of replicates∗
Site
Nitrogen (%)
SD of
replicates∗
1
2
3
4
5
5b∗∗
6
7
8
9
10
11
12
13
14
15
16
17
2.78
2.59
2.19
2.78
2.53
2.01
2.34
2.07
2.16
1.85
2.09
1.61
2.12
2.33
2.08
1.67
2.23
2.04
0.167 (3)
0.186 (3)
0.040 (3)
0.174 (3)
0.040 (3)
0.055 (2)
0.002 (2)
0.001 (2)
0.028 (2)
– (1)
0.062 (2)
0.024 (2)
0.022 (2)
0.029 (2)
0.049 (2)
0.009 (2)
0.003 (2)
0.005 (2)
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
1.95
2.28
2.35
1.71
2.41
2.34
1.54
1.89
1.77
2.25
2.33
2.15
1.90
1.92
2.04
2.52
1.88
2.74
– (1)
– (1)
0.003 (2)
0.018 (2)
0.037 (2)
0.035 (2)
0.006 (2)
0.015 (2)
0.007 (2)
0.001 (2)
0.033 (2)
0.043 (2)
0.018 (2)
0.022 (2)
0.025 (2)
0.008 (2)
0.004 (2)
0.003 (2)
Notes: SD, standard deviation.
∗
Values in parentheses are the number of replicates; ∗∗ shrub 5b was sampled in May and July.
Sagebrush nitrogen spectroscopy
291
Table 3. Partial least square regression results for relating foliar nitrogen concentrations to
sagebrush dried leaf and live canopy reflectance measurements.
No. of PLS
components (m)
R2
AICc
Wavelength selection
NBD – dried leaf spectra
4
0.86
−101.18
NBD – live shrub spectra
1
0.41
−84.02
NBD, coarsened to HyMap
bandwidths – dried leaf
spectra
NBD, coarsened to HyMap
bandwidths – live shrub
spectra
First derivative of (log10 (1/R)),
coarsened to HyMap
bandwidths – dried leaf
spectra
First derivative of (log10 (1/R)),
coarsened to HyMap
bandwidths – live shrub
spectra
2
0.80
−120.98
1651∗ , 1725∗ , 1775,
2010∗∗ , 2220∗∗
600, 1692∗∗ , 2152∗∗ ,
1750, 456
2098∗ , 1662, 1727∗ ,
2080∗ , 605
1
0.54
−99.99
2
0.76
−166.621
2
0.63
−94.10
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
Method
2098∗ , 664, 693, 578,
722
1272, 1185, 2201∗∗ ,
1043∗∗ , 2168∗∗
1065, 1200, 1263,
1232, 2194∗∗
Notes: R, reflectance; NBD, normalized band depth.
Wavelength selection is based on best models, which have the lowest Akaike information
criterion, adjusted for small sample size (AICc).
∗
Within 30 nm of a known nitrogen absorption feature; ∗∗ within 40 nm of a known nitrogen
absorption feature.
Sagebrush dried leaf spectra produced PLS models that could predict nitrogen concentrations within the data set more accurately than the PLS models generated from
live shrub spectra (see table 3). At the dried leaf scale, high-resolution spectrometer
data in combination with an NBD transformation produced optimal results, with
a coefficient of determination equal to 0.86 (R2 = 0.86). At the live shrub scale,
spectrally coarsened data (∼15 nm bandwidths) in combination with the FDR transformation produced optimal results (R2 = 0.63). Spectrally averaging reflectance data
to coarser bandwidths did not improve results for the dried leaf data set but did
improve results for the live shrub data set. For the leaf data set, correlation strength
decreased when dried leaf spectra were coarsened in combination with NBD (R2 =
0.80) and FDR (R2 = 0.76) transformations. For the live shrub data set, correlation
strength was greatest when FDR was used in combination with coarsened reflectance
data (R2 = 0.63) and also increased when NBD values were used in combination with
coarsened spectra (R2 = 0.54) rather than the high-resolution spectra (R2 = 0.41).
A previous study (Huang et al. 2004) found that the coefficient of determination
increased from 0.65 using standard derivative analysis to 0.85 when using continuum
removal to estimate foliar nitrogen concentration of eucalyptus tree canopies in open
woodland from an airborne platform.
A single, disproportionately high coefficient occurred at 2098 nm in both the dried
leaf and live shrub regressions using NBD values calculated from the coarsened spectral data. This wavelength selection likely translates into a cellulose feature at 2100
nm since the cellulose in sagebrush leaf hair is a dominant spectral feature (Kokaly
292
J. J. Mitchell et al.
2001). Overall, the NBD regressions tended to select wavelengths in the visible and
near-infrared associated with plant pigments and the chlorophyll red-edge. The FDR
regressions included wavelength selections in the mid-infrared region associated with
leaf water (1185, 1200, 1232, 1263 and 1272 nm). Wavelength selection patterns, as
indicated by PLS regression coefficients, were similar to selection patterns observed in
the correlograms.
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
4. Conclusions
The results of this study are encouraging and warrant further exploration into
sagebrush reflectance spectra to characterize nitrogen concentrations. As expected,
sagebrush dried leaf spectra produced PLS models that could predict nitrogen concentrations within the data set more accurately than the PLS models generated from
live shrub spectra because noise (e.g. soil, atmosphere, sampling error, leaf water) was
minimized. At the live shrub scale, spectrally resampling the data to coarser bandwidths improved the results (from R2 = 0.41 to R2 = 0.54), which is an indication
that noise is being averaged out. Also, inclusion of wavelengths associated with leaf
water in the FDR transformations appeared to improve regression results compared
to NBD calculations, which did not include wavelengths associated with leaf water
absorption. Leaf water plays an obvious role at estimating nitrogen at the shrub scale;
depending on the relationship between leaf water and available nitrogen, August may
be an optimal time for acquiring additional sagebrush reflectance spectra because leaf
water is lowest in late summer. Spectral transformation techniques should be further
tested and refined to improve the selection of wavelengths related to known nitrogen
absorption features. Building a data set that contains a larger number of field samples
is ideal and would facilitate the establishment of calibration equations. While continuum removal did not improve results at the individual live shrub scale, it may be more
useful when scaling from the individual plant to the open canopy scale. A hyperspectral sensor with precise and sensitive instrumentation should be selected if sagebrush
foliar nitrogen research is extended to the pixel level.
Acknowledgements
This study was made possible by the data and a grant provided by the INL, grants
from the Idaho Space Grant Consortium and NOAAOAR ESRL/Physical Sciences
Division Grant # NA06OAR4600124. Many thanks to Matt Germino, for use of his
plant ecology laboratory; to Roger Blew of the S. M. Stoller Corporation, who facilitated field data collection access; and to Chris Forsgren, whose field assistance was
essential.
References
ASNER, G.P., WESSMAN, C.A., BATESON, C.A. and PRIVETTE, J.L., 2000, Impact of tissue, canopy, and landscape factors on the hyperspectral reflectance variability of arid
environments. Remote Sensing of Environment, 30, pp. 271–278.
BARTON, F.E., HIMMELSBACH, D.S., DUCKWORTH, J.H. and SMITH, M.J., 1992, Twodimensional vibration spectroscopy: correlation of mid- and near-infrared regions.
Applied Spectroscopy, 46, pp. 420–429.
BOLSTER, K.L., MARTIN, M.E. and ABER, J.D., 1996, Determination of carbon fraction and
nitrogen concentration in tree foliage by near infrared reflectance: a comparison of
statistical methods. Canadian Journal of Forest Research, 26, pp. 590–600.
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
Sagebrush nitrogen spectroscopy
293
BURNHAM, K.P. and ANDERSON, D.R., 1998, Model Selection and Multimodel Inference: A
Practical Information-Theoretic Approach, 2nd ed., pp. 66–445 (New York: Springer).
CARD, D.H., PETERSON, D.L., MATSON, P.A. and ABER, J.D., 1988, Prediction of leaf chemistry by the use of visible and near infrared reflectance spectroscopy. Remote Sensing of
Environment, 26, pp. 123−147.
CLARK, R.N. and ROUSH, T.L., 1984, Reflectance spectroscopy: quantitative analysis
techniques for remote sensing applications. Journal of Geophysical Research, 89,
pp. 6329−6340.
CURRAN, P.J., 1989, Remote sensing of foliar chemistry. Remote Sensing of Environment, 30,
pp. 271–278.
DEPUIT, E.J. and CALDWELL, M.M., 1973, Seasonal pattern of net photosynthesis of Artemisia
tridentata. American Journal of Botany, 60, pp. 426–435.
HRUSCHKA, W.R., 1987, Data analysis: wavelength selection methods. In Near-Infrared
Technology in the Agricultural and Food Industries, P. Williams and K. Norris (Eds.),
pp. 35–53 (Saint Paul: American Association of Cereal Chemists).
HUANG, Z., TURNER, B.J., DURY, S.J., WALLIS, I.R. and FOLEY, W.J., 2004, Estimating foliage
nitrogen concentration from HyMap data using continuum removal analysis. Remote
Sensing of Environment, 93, pp. 18–29.
HUBER, S., KNEUBUHLER, M., PSOMAS, A., ITTEN, K. and ZIMMERMAN, N.E., 2008,
Estimating foliar biochemistry from hyperspectral data in mixed forest canopy. Forest
Ecology and Management, 256, pp. 491–501.
JOHNSON, L.F., HLAVKA, C.A. and PETERSON, D.L., 1994, Multivariate analysis of AVIRIS
data for canopy biochemical estimation along the Oregon transect. Remote Sensing of
Environment, 47, pp. 216–230.
KOKALY, R.F., 2001, Investigating a physical basis for spectroscopic estimates of leaf nitrogen
concentration. Remote Sensing of Environment, 75, pp. 153–161.
KOKALY, R.F., ASNER, G.P., OLLINGER, S.V., MARTIN, M.E. and WESSMAN, C.A., 2009,
Characterizing canopy biochemistry from imaging spectroscopy and its application to
ecosystem studies. Remote Sensing of Environment, 113, pp. S78–S91.
KOKALY, R.F. and 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, pp. 267–287.
KOKALY, R.F., DESPAIN, D.G., CLARK, R.N. and LIVO, K.E., 2003, Mapping vegetation in
Yellowstone National park using spectral feature analysis of AVIRIS data. Remote
Sensing of Environment, 84, pp. 437–456.
KRÄMER, N. and SUGIYAMA, M., 2010, The degrees of freedom of partial least square
regression. Available online at: http://arxiv.org/abs/1002.4112 (accessed 6 December
2010).
KRÄMER, N., SUGIYAMA, M. and BRAUN, M., 2009, Lanczos approximations for the speedup
of kernel partial least squares regression. In Proceedings of Twelfth International
Conference on Artificial Intelligence and Statistics (AISTATS2009), 16–18 April 2009,
Clearwater Beach, FL (Brooklyn, MA: Microtome Publishing), pp. 288–295.
KUMAR, L., SCHMIDT, K.S., DURY, S. and SKIDMORE, A.K., 2001, Imaging spectrometry and
vegetation science. In Imaging Spectrometry, F. van der Meer and S.M. de Jong (Eds.),
pp. 111–155 (Dordrecht: Kluwer Academic Publishing).
MUTANGA, O., SKIDMORE, A.K. and PRINS, H.H.T., 2004, Predicting in situ pasture quality in
the Kruger National Park, South Africa, using continuum-removed absorption features.
Remote Sensing of Environment, 89, pp. 393−408.
OLLINGER, S.V. and SMITH, M.L., 2005, Net primary production and canopy nitrogen in a
temperate forest landscape: an analysis using imaging spectroscopy, modeling and field
data. Ecosystems, 8, pp. 760–778.
Downloaded by [Battelle Energy Alliance], [Jessica J. Mitchell] at 15:12 30 August 2011
294
J. J. Mitchell et al.
SERRANO, L., PENUELAS, J. and USTIN, S., 2002, Remote sensing of nitrogen and lignin in
Mediterranean vegetation from AVIRIS data: decomposing biochemical from structural signals. Remote Sensing of Environment, 81, pp. 355–364.
SKIDMORE, A.K., FERWERDA, J.G., MUTANGA, O., VAN WIEREN, S.E., PEEL, M., GRANT, R.C.,
PRINS, H.T., BALCIK, F.B. and VENUS, V., 2010, Forage quality of savannas – simultaneously mapping foliar protein and polyphenols for trees and grass using hyperspectral
imagery. Remote Sensing of Environment, 114, pp. 64–72.
TOWNSEND, P.A., FOSTER, J.R., CHASTAIN JR., R.A. and CURRIE, W.S., 2003, Application of
imaging spectroscopy to mapping canopy nitrogen in the forests of central Appalachian
mountains using Hyperion and AVIRIS. IEEE Transactions on Geoscience and Remote
Sensing, 41, pp. 1347–1354.
WHITE, J.D., TROTTER, C.M., BROWN, L.J. and SCOTT, N., 2000, Nitrogen concentration in New
Zealand vegetation foliage derived from laboratory and field spectroscopy. International
Journal of Remote Sensing, 21, pp. 2525–2531.