| Soil organic matter is an important indicator of soil fertility, and it is of greatsignificance for improving the structure of soil nutrients and the quality of arable land. Thetraditional method of measuring soil organic matter is complicated and not suitable forlarge-scale agricultural production. Near-Infrared (NIR) spectroscopy occupies theadvantages of rapidness and simpliness, and some research indicates that NIR spectroscopycan be applied in non-destructive measurement of soil organic matter. Since soil is complexsystem with multi-component, establishing models based on the type of soil is of greatsignificant to improve the practicability and analytic precision of NIR technique.In this paper, the NIR model for organic matter analysis was established and comparedaccording to two kinds of typical soil samples from different areas. Based on randomnessand stability, the reliable analyzing models can be obtained after the processes ofcalibration, prediction and validation. Three kinds of wavelength models, which includesthe continuous-mode moving window partial least squares (MW-PLS), thequasi-continuous-mode equidistant combination multiple linear regression (EC-MLR) andthe discrete-mode successive projections algorithm multiple linear regression (SPA-MLR),were adopted and discussed here. The corresponding wavelength models, modeling effectand validation effect of the two kinds of soil were different. However, two wavelengthmodels contained public information wavelengths which located in first and secondharmonic areas. While using one kind of soil model to predict another kind of soil, theprediction effect was remarkably decreased. The result indicated that different organicmatter NIR analyzing models from different types of soil shared some common features,however, due to differences in composition and structure, establishing calibration andmodeling process according to the type of soil was necessary. |
PDF Full Text Request