This raises
gth to be
view of the
paration of
ongst rock
1 conditions,
n measure-
or the separ-
ress, and
ive red/near
be most
ton 1983,
) . For
00, 1600 and
sirable
m and Sommer
md Lyon
leful spectral
.nfrared wave-
il) but these
spending on
> testing of
.cations is
;ential of
following
;ion of soils
igricultural
CFFERENTIATE
L variability
high chemical
cw soil type
st on the
soil types
. If the
siderably
in intensive
nt in the
creasing
icient
fertilization
e production
. It is pro-
spectral
To show how
of soil
a multi-
ite what soil
ipectral
:hs are most
iw this infor-
i multispectral
and soil
on spectral
.spectral data
îers (e.g.
1976, Cipra
vegetation
soil moisture
al 1985)
pectra. It is
;tral laser
Lnce it can
ness at the
led.
three different
are field which
e of organic
rface soil
analysis and
highly
variable in organic carbon content. The second site
was an agricultural field in the driest part of
Canada located on a recently cleared glacio-lacus-
trine deposit. 58 soil samples were collected at
that site and the samples had very low organic matter
content (0.5-2.0%), while the soil texture was ex
tremely variable. The third set, consisting of 30
samples collected from ten different metal mine tail
ings in British Columbia and represented a sample set
with virtually no organic matter content but highly
variable iron content.
The following soil properties were analysed in all
samples: Organic carbon content (Walkley-Black
method), total carbon (Leco-Method), extractable iron
(citrate-dithionite-bicarbonate method), total iron
(HF-teflone bomb method), total nitrogen (autoanalys
er method), available P (Bray P-1 method), extracta
ble K and Ca (neutral ammonium acetate method), and
particle size (using the hydrometer method) and soil
color (Munsell Color System). All methods are des
cribed in Black et al (1965).
The spectral reflection measurements of the soil
samples were carried out with a Barringer Multi
channel spectrometer using various spectral bands
covering the 500-2350 nm wavelength range. The
measurements were carried out in the laboratory
under artificial light and all samples were air-dried
and the greater than 2mm fraction was removed.
The data were analysed using single and multiple
correlation and regression techniques so as to deter
mine the best relationships between soil fertility
properties and spectral reflection measurements at
different wavelength bands.
3.2 Results
The three soil data sets were analysed separately and
a plot of the reflection curves over the 500-2350 nm
range showed that set one and set two had character
istic curves which were unique and significantly
different from one another. Sets one and three could
also be separated on the basis of the shape of the
spectral curve but there was considerable overlap
between set two (low organic content and high
particle size variability) and set three (no organic
carbon, little particle size variation and high iron
variability). The results suggest that sample prov
enance or overall composition has an important in
fluence on reflection.
When individual properties were related to the
reflection at specific wavelengths the following
picture emerged: Significant correlations were
found between spectral reflection vs organic carbon
content, spectral reflection vs iron content, and
spectral reflection vs % sand content in the samples.
In each case the correlation was only significant
within a restricted concentration range and when the
data sets were combined the significance of the re
lationships were dramatically reduced. A regression
analysis was carried out for the best relationships
and the results are provided in Figures 3-5. Organic
carbon could best be predicted from reflection in
the 550 nm wavelength range using the organic rich
samples of sample set 1. In this case all samples
had organic carbon values which were greater than
2.0% and the regression in Figure 3 explains 86% of
the total variance in carbon. When the data from
samples with less than 2% carbon were introduced to
the equation the relationship could no longer be
extended. This clearly suggests that above a certain
threshold value organic carbon has a dominant in
fluence on reflection and can thus be predicted with
fairly good accuracy. Below 2% other properties
such as texture and iron appear to influence reflec
tion significantly. Figure 4 shows the regression
between total iron content and spectral reflection
at 630 nm wavelength for the mine tailing samples
which were characterized as having no organic carbon
content and very little textural variability but
great total iron variability. In this regression
the reflection values explained 87% of the variation
in iron content. Finally the soil reflection data
from the second data were found to correlate signifi
cantly with soil texture (% sand content) at 1600 nm
wavelength.
Figure 3. Relationship Figure 4. Relationship
between reflection (550 nm) between reflection (630
and organic carbon content nm) and total iron con
fali samples 2% carbon). tent (all samples 0.01%
carbon) «.
As shown in Figure 5 this relationship explained
61% of the variance in % sand content. Curvilinear
relationships with reflection and appropriate log-
transformations were made prior to the analysis.
% Sand Content
Figure 5. Relationship between reflection (1600 nm)
and % sand content (all samples 2% carbon).
In addition to these relationships two other correla
tions were significant. Total nitrogen was related
to reflection at 550 nm (r = -0.78) for the set rich
in organic carbon, and exchangeable Ca was signifi
cantly related to reflection at 1600 nm (r - 0.56)
in the low carbon content soils. The former is of
interest because it is one of the main and most var
iable fertility components in the soil. It is
obvious that nitrogen has no direct effect on reflec
tion but it is most often highly correlated with
organic carbon content thus giving us a significant
relationship. Because of this secondary relationship
it might be possible to also get a crude assessment
of the nitrogen status in the field using the reflec
tion data and this has important implications for
improving the efficiency of fertilizer use since con
ventional nitrogen tests are invariably inaccurate
and unreliable. The relationship between Ca and
reflection is also suspect but a possible explanation
for it is the fact that Ca is easily absorbed by clay
minerals and this results in clay flocculation
(Hosterman and Whitlow 1980). This might prevent
organic matter from replacing Ca in the clay structure
