266
* Nitrogen: it is the chemical compound which is best correlated with reflectances. With exception of
X=1982nm, best correlations occur at wavelengths associated to absorption bands. The two reflectance ratios
(i.e. p/pi68o P/P 2230 ) produce different results: p/p 16S0 tends to be efficient for wavelengths between
1722nm and 1759nm, whereas the ratio p/p 2 23 o tends to be efficient for wavelengths between 1982nm and
2190nm. This dual behaviour was also noticed for the other chemicals (L, Cell. W, Cell. S).
* Lignin: correlations are relatively large. Best correlations were obtained for wavelengths between 1680nm
and 1778nm, and for 2310nm. However, most of these wavelengths cannot be attributed to lignin. The fact
that large correlations tend to occur at the same wavelengths for lignin and nitrogen can partly be explained
by the large correlation between these two compounds (Table 1).
♦Cellulose: cellulose concentrations, especially Cell.S, were poorly correlated with reflectances. Best
correlations were obtained with attributable wavelengths.
III. Laboratory-derived Predictive Equations
Stepwise regression analysis was applied to laboratory-derived reflectance spectra for determining those
spectral bands, and associated linear relations, that are predictive of chemical concentrations. Only reflectance
data associated to the 13 spectral bands outside atmospheric absorption bands were considered. In addition to
the raw reflectance p, regression analysis was also applied to several transformations of reflectances. The most
successful were: log(l/p), p/p 168 o> and p/p 2 23 o- Two criteria were used for selecting "predictive wavelengths",
and associated predictive linear relations: Fisher-Snedecor values larger than the tabulated F-value at a 1%
significance level, and larger correlations between actual and predicted concentrations. Several predictive
wavelengths were determined. Table 2 displays the most successful wavelengths and associated correlations
between the predicted and actual chemical concentrations. It should be noted that:
- two to three wavelengths were sufficient for predicting N, L and ADF concentrations with correlations larger
than 97%, 87% and 84%, respectively. Results for cellulose S are much poorer than for cellulose W.
- selected predictive wavelengths could not always be attributed to absorption bands of compounds. This could
partly be explained by some intercorrelation between chemicals. This was typically the case with NDF: most
predictive wavelengths that were selected correspond to absorption bands of nitrogen and protein.
P
log(l/p)
P/PtôSO
P / P2230
N
2180*-2230: 97
2180*-2230: 97
2180*-2270: 97
2180*-2270: 97
2190*-2270: 97
L
2190-2230: 88
2180-2230: 88
2180-2230: 88
2190-2270*: 87
1734-1722*-2270*: 89
2190-2270*: 87
ADF
2100*-2230: 85
1778*-1982: 84
1982-1759-1722*: 88
2230-2190: 88
2230-2100*: 85
1778*-1722*-1982: 88
2190-2270*: 82
2180-2270*: 82
1982-1778*: 80
1734-2270*-1722: 85
2190-2270: 83
2180-2270: 82
2100*-2270: 80
1982-1734-1722*: 84
CellS
2100**2230: 50
2100*-2230: 49
2270*-2100*:49
2100*-2270*: 49
CellW
1722*-2230: 82
1722*-2100*: 82
1722*-2270*: 82
1722*-2100*: 83
1722*-2270*: 82
1722**2230: 82
1734*-2100*: 76
1759-2100*: 73
1734*-1982-2100*: 86
1759-2100*-2270*:82
1734*-l778*: 70
2190-2270*: 67
1734*-1778*-l 982: 82
2270*-2190-2139: 77
Hemi
2139-2100*: 58
2100*-2139-2230: 68
2139-2100*: 58
2100*-2139-2230: 68
2190-2100*: 59
2139-2100*: 59
2139-2100*-2270*: 66
1734-2270*-2100*: 68
2139-2100*: 57
2190-2100*: 62
2139-2100*-2270*: 67
NDF
1734*-1759: 84
2180-2230: 85
2180-1982: 84
1734*-1759: 84
2180-1982: 84
2190-1982:84
2139-2270: 82
2270-2190: 82
1734*-l 722*-2100*:85
1734*-l722*-2190: 85
2180-1982: 82
1734*-1778-1722: 85
Table 2 :
Optimal wavelengths
in order of selection,
from stepwise
regression analyses on
pine needle laboratory
reflectance spectra
with associated
correlations (F>1%).
Attributable
wavelengths, i.e.
within lOnmofa
known absorption
wavelength for that
compound, are
indexed with
The relationships predictive of lignin, ADF and nitrogen concentrations mentioned below are represented in
percentage of dry matter. Wavelengths that can be attributed to chemicals are indexed with the symbol *.