Abstract
This study evaluated whether the accuracy of soil organic carbon
measurement by laboratory hyperspectral imaging can match that of standard
point spectroscopy operating in the visible–near infrared.
Hyperspectral imaging allows a greater amount of spectral information to
be collected from the soil sample compared to standard spectroscopy,
accounting for greater sample representation. A total of 375
representative Irish soils were scanned by two-point spectrometers (a Foss
NIR Systems 6500 labelled S-1 and a Varian FT-IR 3100 labelled S-2) and
two laboratory hyperspectral imaging systems (two push broom line-scanning
hyperspectral imaging systems manufactured by DV optics and Spectral
Imaging Ltd, respectively, labelled S-3 and S-4). The objectives were (a)
to compare the predictive ability of spectral datasets for soil organic
carbon prediction for each instrument evaluated and (b) to assess the
impact of imposing a common wavelength range and spectral resolution on
soil organic carbon model accuracy. These objectives examined the
predictive ability of spectral datasets for soil organic carbon prediction
based on optimal settings of each instrument in (a) and introduced a
constraint in wavelength range and spectral resolution to achieve common
settings for instruments in (b). Based on optimal settings for each
instrument, the deviation (root-mean square error of prediction) from the
best fit line between laboratory measured and predicted soil organic
carbon, ranked the instruments as S-1 (26.3 g kg−1)
< S-2 (29.4 g kg−1) < S-3 (34.3 g
kg−1) < S-4 (41.1 g kg−1). The
S-1 model outperformed in all partial least squares regression performance
indicators, and across all spectral ranges, and produced the most
favourable outcomes in means testing, variance testing and identification
of significant variables. It is assumed that a larger wavelength range
produced more accurate soil organic carbon predictions for S-1 and S-2.
Under common instrument settings, the prediction accuracy for S-3 that was
almost equal to S-1. It is concluded that under standard operating
procedures, greater soil sample representation captured by hyperspectral
imaging can equal the quality of the spectra from point spectroscopy. This
result is important for the development of laboratory hyperspectral
imaging for soil image analysis.
© 2018 The Author(s)
PDF Article
More Like This
Cited By
You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.
Contact your librarian or system administrator
or
Login to access Optica Member Subscription