idmore,
igeningen, The
concentration of
), willow (Salix
sation technique
ive power of the
tion R? between
or assessing the
fter returning to
n additional 24
ns sampled, an
Netherlands. 10
aris L.) and 10
pling technique
measurements,
sported to the
t 70 degrees.
) degrees until
ructed with a
ozamsky et al.,
neasured with a
were measured
mic Absorption
ed tannin were
germann(1988).
g the improved
lyphenols were
assay for total
lyphenol levels
in solution of
nd recalculated
Concentrations
ral reflectances,
averaged. This
cach vegetation
uum removal
nuum-removed
'entres differed
e to the use of
each containing
set served as a
Using stepwise
the continuum
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring", Hyderabad, India,2002
removed reflectance data to predict the concentration of the
various chemicals. The number of wavebands used in each
regression model was limited to a maximum of 3 to avoid over
fitting of the model (Table 2). The regression model was used
to predict the chemical composition of the samples in the test
dataset. To test the effectiveness of the regression model the
correlation between measured and predicted concentrations
were calculated for each chemical (Table 2).
Training dataset (N=24) Test dataset (N=24)
Mean Min Max St.Dev Mean Min Max St.Dev
Component (%) (%) (%) (%) (%) (%) (%) (%)
Nitrogen 1.43 0.88 2.59 0.44 1.45 0.69 2.53 0.46
Phosphorous 0.100 0.044 0.198 0.040 0.104 0.049 0.197 0.046
Sodium 0.014 0.007 0.021 0.004 0.013 0.008 0.022 0.004
Potassium 0.606 0.37 0.98 0.14 0.620 0.38 0.98 0.17
Calcium 1.37 0.36 2.30 0.72 1:35 0.37 2.80 0.68
Magnesium 0.217 0.145 0.310 0.053 0.210 0.124 0.341 0.060
C. Tannin* 15.85 0.04 54.91 21.63 16.23 0.00 49.02 19.72
Polyphenol* 8.14 2,33 13:32 3.30 9.10 3:36 17.42 3.73
Table 1. Overview of the concentration range of N, P, Na, K, Ca, Mg, Tannin and Polyphenol for the training and the test dataset.
Tannin and Polyphenols are in 96 quebracho tannin equivalents.
3. RESULTS
The concentrations predicted by the regression models for the
different components showed a strong correlation with the
measured chemical composition. The best prediction ocurred
for nitrogen (R? = 0.72), sodium (R? 2 0.79), calcium (R2 =
0.81) magnesium (R? 2 0.65) and tannin (R? = 0.70). Less
succesfull was prediction of phosphorous (R* = 0.40),
potassium (R? = 0.42) and polyphenol (R* = 0.41). Correlations
between observed and predicted concentrations were significant
at the 0.01 level, except for the predicted concentration of
phosphorous (Table 2).
Component Waveband 1 Waveband 2 Waveband 3 Pred. R°
N 741 1665 479 0.72
P 1022 957 1762 0.40
Na 1127° 1208 1733 0.79°
K 2053 821 968 0.42"
Ca 698 964 510 0.81”
Mg 482 1172 741 0.65"
Tannin 694 1013 673 0.70
Polyphenol 812 804 823 0.41°
Table 2. Wavebands used for prediction of chemical composition of a test dataset, based on stepwise regression of a training dataset.
(*-p «0.05; **: p « 0.001)