Fast measurement of soil organic matter of rice rehabilitation fields based on vis-near-infrared spectroscopy (350–2500nm)

Abstract

Fast measurement of soil organic matter is essential for site-specific application of fertilizer into farmlands. Total 100 soil samples were collected from several rice rehabilitation fields on the outskirts of Hangzhou City, China. Using a portable 350–2500nm spectrometer, absorbance spectra of the samples were recorded for the study. The spectra were separated into a calibration set (70%) and a prediction set (30%). The spectra were firstly transformed by several preprocessing methods for the purpose of removing noise and bias. Then, the original spectra and the various transformed spectra of calibration set were subjected to a partial least squares regression (PLSR) algorithm for obtaining PLSR calibration models. Finally, the PLSR models were applied to measure soil organic matter with the unknown samples in prediction set. The results show that the PLSR model developed for the original spectra can obtain good prediction accuracy with coefficient of determination (R2) of 0.83 and residual prediction deviation (RPD) of 2.49. The PLSR models developed by various transformed spectra achieved better prediction performance. The spectra transformed by the 1st derivative preprocessing method produced the best PLSR model which achieved highest prediction accuracy with R2 of 0.93 and RPD of 3.77. The study suggests that (1) soil organic matter of rice rehabilitation fields can be accurately measured by a spectrometer, and (2) prediction performance of a PLSR model could be improved if the original absorbance spectra are transformed by appropriate preprocessing methods such as the 1st derivative transformation used in the study.