Agriculture Reference
In-Depth Information
and tobacco. Archaeologists have used NIRS to identify dyes, pigments,
oils and resins (Goffer, 1980).
Using NIRS to test our preliminary classification of ADE, we employed
relatively simple methods of sample preparation, processing and analysis.
We dried the samples in a forced air oven, removed large inclusions by hand
picking and then ground the soil matrix and screened it through a 2.00-mm
mesh. The screened material was placed in a simple ring sample cup and
automatically processed through the scanning monochrometer. The wave-
length interval between data points for our analysis was 4 nm. Projected
over the measured spectral range of 1108-2492 nm, this interval results in a
total of 346 data points for each sample. Sample spectra interpretation
was facilitated by calibration and comparison with the results from other
chemical and physical determinations. We chose a chemometric statistical
package called WINISI (Infrasoft International, LLC) from among the
various statistical packages designed for NIRS analysis because its use
of modified partial least squares regression was best suited for our
investigation (Shenk and Westerhaus, 1991a,b).
Results
For the initial phase of our test, we scanned 65 samples from the three
distinct cultural and non-cultural soil contexts - TP, TM and the
background sands and clays (SC) - as identified by field and laboratory
methods described above. Distinctive NIR spectra of the three groups
were identified and a calibration statistic developed (Table 3.10.1). To test
the validity of this calibration, the individual group placement of an
additional 143 samples from other known contexts in the Rio Tapajós
region was then queried based upon their individual spectral signatures.
Projected placement based on field observations, Munsell colour and
chemical characteristics was correct for > 90% of these samples. In other
words, there is discrimination between the three groups, but with variation
within each data set and a small number of samples with a spectral
signature that falls outside of the expected category. This is exactly what we
would expect from a relatively continuous, real world, data set. Many of our
samples were collected along transects initiated outside the dark earth zone
and traversing both TP and TM zones, and the transitional areas between
them.
In this ongoing study, we presently are testing the distinctiveness of
these NIRS groupings and identifying the band ranges that account for the
variation between groups. For our data set, the shapes of the TP, TM and
SC spectra are very similar throughout most of the spectral range. Both the
greatest activity and the greatest differences in the spectral patterns
occurred at the high end of the spectrum, between 2149 and 2492 nm. As
Search WWH ::
Custom Search