Collection of soil spectral data in MIR range
Air-dried, crushed and 0.5 mm ground samples, mentioned earlier, were
filled in the cups meant of the Bruker alpha Fourier Transformed MIR
Spectrometer for recording spectral signatures. The FT-MIR was
stabilized for two hours to increase the amplitude count to more that
9000 and corrections were made for the background during instrument
calibration, before recording spectra in the near and middle MIR range.
For the Vis-NIR range, a spectroradiometer (Model: FieldSpec3 FR;
Analytical Spectral Devices Inc., USA). The raw spectra collected from
the Vis-NIR range ASD FieldSpec® and OPUS file from and MIR-spectrometer
respectively were subsequently pre-processed using R software (version
3.3.3, The R Foundation for Statistical Computing Platform). The most
widely used pre-processing techniques is divided into two categories:
scatter-correction methods and spectral derivatives. Since many workers
are of the openion that the soil properties can be related to absorbance
and reflectance and their first and second derivatives and it has been
reported that the absorption features in reflectance spectra were
enhanced by derivative spectroscopy (Tsai, 1998), the reflectance data
was transformed to absorbance through the expression,
absorbance=log10 (\(\frac{1}{\text{Reflectance}}\)).
First and second derivatives were obtained from reflectance and
absorbance data. The spectral derivative method consists of first
derivatives (FD) and second derivatives (SD) of the reflectance spectrum
using the equation (1) and (2) respectively.
\begin{equation}
\frac{\mathbf{\text{dR}}}{\mathbf{\text{dλ}}}\mathbf{=R}\left(\mathbf{\text{λi}}\right)\mathbf{-}\frac{\mathbf{R}\left(\mathbf{\lambda i-1}\right)}{\mathbf{\lambda i-\lambda i-1}}\mathbf{\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ldots\ldots\ldots(1)\ }\nonumber \\
\end{equation}\begin{equation}
\frac{\mathbf{d}^{\mathbf{2}}\mathbf{R}}{\mathbf{d}\mathbf{\lambda}^{\mathbf{2}}}\mathbf{=}\frac{\mathbf{d}}{\mathbf{d}\mathbf{\lambda}}\left(\frac{\mathbf{\text{dR}}}{\mathbf{d}\mathbf{\lambda}}\right)\mathbf{\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ldots\ldots..(2)\ }\nonumber \\
\end{equation}Where R, spectral reflection/absorbance; ʎi, ithwavelength / band. Further among scatter-correction methods,
multiplicative scatter correction (MSC), standard normal variate (SNV)
and other smoothing methods include averaging spectra, and median
filters, first order derivative, second order derivative and the
Savitzky–Golay transformation were used to reduce noise in spectral
signals.