Abstract:
Rapid and reliable estimation of soil moisture constants namely, field
capacity (FC) and wilting point (WP) is significant for scientific irrigation
scheduling. The conventional methods for their estimation are cumbersome, time
consuming and not suitable for their estimation at different space and time domains.
An alternative would be the use of diffuse reflectance spectroscopy (DRS) for
which the development of calibration functions that link the soil attributes with
spectral signature is a major pre-requisite. In this study, the utility of spectral index,
feature projection of full-spectrum and variable selection approaches namely,
normalized difference reflectance index (NDRI), partial least squares regression
(PLSR) and ordered predictor selection (OPS), respectively to build accurate and
less complex calibration functions was evaluated. The performance of calibration
functions were judged in terms residual prediction deviation (RPD) criteria. The
NDRI based calibration functions developed in this study do not comply with the
minimum accuracy level (RPD<1.4) expected from DRS analysis. In contrast, both
full-spectrum based PLSR and OPS approaches yielded calibration functions which
were capable for accurate (RPD>2.0) and moderate (1.4<RPD>2.0) estimation of
FC and WP, respectively. Specifically, the full-spectrum based calibration function
developed using second derivative of reflectance was found to be the best for both
FC (RPD=2.01) and WP (RPD=1.74). The OPS approach in conjunction with
variable indicators namely, combination of regression & correlation coefficient (β-
r) and combination of adjacency values of mutual information & signal-to-noise
vector (AMI-StN) yielded best calibration functions in case of FC and WP,
respectively. The calibration functions so developed consisted of only 19.09% (FC)
and 34.39% (WP) of total number of spectral variables as that in full-spectrum.
Thus, the result of the study advocate the use of OPS approach to develop simple
and parsimonious calibration functions to estimate FC and WP from spectral
signature of soil.
