TITLE:
Effect of the Continuum Removal in Predicting Soil Organic Carbon with Near Infrared Spectroscopy (NIRS) in the Senegal Sahelian Soils
AUTHORS:
Macoumba Loum, Mateugue Diack, Ndeye Yacine Badiane Ndour, Dominique Masse
KEYWORDS:
NIRS, Soil Proprieties, Continuum Removal, PLSR Model, Senegal River Delta
JOURNAL NAME:
Open Journal of Soil Science,
Vol.6 No.9,
September
23,
2016
ABSTRACT: Spectroscopy
plays a major role in the access of the analytical parameters of the soil. It
tends to substitute the conventional laboratory analysis because hyperspectral
data were least expensive and easier to obtain. The objective of this study was
to evaluate the effect of the continuum removal (CR) in the validation of the
accurate prediction model of the soil properties with Vis-NIR spectroscopy data. Few studies
using Vis-NIR reflectance spectroscopy have well focused the calculation of the
CR method; its effect in the calibration of the accurate models was also not
well emphasized. In this study, we used the remote sensing software ENVI 4.7 to compute the CR function where the value of the continuum for
each sample and for each spectral wavelength was obtained by dividing the
reflectance values of the full spectrum (FS) with those of the continuum curve
(CC). The partial least square regression (PLSR) model was applied in the
spectral data from the soil of the Senegal Sahelian region. It was calibrated with
both data from the full spectrum (FS) and those obtained after the application
of the continuum removal. With the application of the CR, ultraviolet
wavelengths (350 - 429 nm) and those of near infrared (2491 - 2500 nm) were
removed from the explanatory variables of PLSR model. With the FS, all
wavelengths between 350 and 2500 nm were taken into account in predicting soil
properties. Our findings show a positive effect of the application of CR in the
estimation of soil organic carbon. In calibration, the R2 increased up to 10%
with the continuum removal in the model of 12 components (CP). In terms of
validation, it’s the 15-component model which is the most accurate with the
same range in calibration between the FS and the CR. The lowest RMSE ranged from
0.04 with the FS to 0.03 with the application of the CR in calibration and
validation. These results show that the interest of this study as soil organic
carbon is recognized as a key indicator of fertility of the soil in Sahelian-African
regions. For future studies, it’s important to apply the model of neural
networks to better evaluate the effect of continuum removal in predicting soil
properties from the spectral data and other methods of preprocessing like the
multiplicative scatter correction (msc).