Research Information System
Research Journal of Chemistry and Environment, vol.17, no.6, pp.5-11, 2013 (Scopus)
The use of Attenuated Total Reflectance (ATR) is an alternative method in determining carbon (C), nitrogen (N) and other elemental contents of organic and inorganic soils for which diffuse reflectance infrared Fourier transform (DRIFT) spectroscopy has been mostly utilized. In this study, the combined use of ATR-Fourier transform infrared (FTIR) spectroscopy and partial least square regression (PLSR) or artificial neural network (ANN) models in estimating total soil C and N have been explored which provide direct, rapid, economical and multiple in situ measurements. Total soil C and N data obtained from 153 soil samples across agricultural lands and analyzed using CNH elemental analyzer were used to build PLSR and ANN models as a function of ATRFTIR spectrum ranges based on a training dataset with leave-one-out cross validation (LCV) and independent validation (IV) dataset that randomly constitute 67% and 33% of the entire dataset respectively. Wavenumber ranges of 650-2365 cm-1 and 773-1726 cm-1 in ATR- FTIR data were selected as predictors for PLSR and ANN models of soil C respectively. PLSR model of soil C led to r2 = 0.86 for training and r2 = 0.68 for validation, with PLSR model of soil N as a result of wavenumber range of 1300-1400 cm-1 leading to r2 = 0.81 for training and r2 < 0.1 for validation. Multilayer perceptron model appeared to be the best-performing ANN for the emulations of both total soil C and N and outperformed PLSR model of total soil N.