A new approach to predict organic matters in soils by using near infrared spectroscopy

Near infrared reflectance (NIR) spectroscopy is as a rapid, convenient and simple nondestructive technique useful for quantifying several soil properties. This method was used to estimate organic matters (OM) in soils of Shanxi Taigu. A total of 42 soil samples obtained from 0-20cm depths were collected and analyzed their spectra features using a spectrophotometer. 34(set (I)) randomly chosen samples were used during the calibration and validation stage. The calibration equations were developed using partial least squares (PLS) analysis, and PLS-artificial neural network (ANN) techniques. In the PLS-ANN analysis, PLS method was used to find some spectra actives to OM, where eight wavelengths were obtained, and then regarded them as the input neurons of ANN. The expected results were obtained when the training time was 4710. The calibration equation developed from set (I) was used to predict the constituent values for the independent spectra in set (II) (8 samples). The results indicated that the observed results using PLS-ANN were better than those obtained by PLS. The r2 for PLS-ANN prediction is 0.864, and SEP is 0.327, however, the r2 for PLS prediction is 0.73 1, and SEP is 0.385. It showed that it is feasible to predict OM content in this soil using near infrared spectroscopy with data treatment of PLS-ANN.