Rapid Determination And Instrumentation Of Soil Organic Matter Based On Visible-Near-Infrared Spectroscopy

Soil organic matter (SOM) is an important composition of soil. Acquiring SOM information is of great help for people understanding soil fertility, fertilizing reasonably and improving harvest. Traditional measurement for SOM is chemical detection method in laboratory. It has some shortages, such as high quality requirement for tester, time-consuming, exhausting and high cost, which do not meet the needs of rapid growth for modern digitizing agriculture. Therefore, a fast, real-time and accurate testing method for SOM is much-needed. This study aims at spectral modeling and instrument development for SOM based on visible-near-infrared spectroscopy. A total of104soil samples, collected from Shangyu, Fuyang, Changkou and other cities in Zhejiang province, were prepared. Soil spectra were recorded by a FieldSpec-3spectrometer (ASD, USA) in diffuse reflectance mode from350to2500nm. Main contents and collusions are listed as follows:(1) SOM spectral characteristics in various regions are different. Intelligent algorithms can be used to establish "SOM-spectra" model for these feature information. Through spectral pretreatment, wavelength interval optimization, data compression and feature band extraction, we can reduce the data dimension, improve the interpreting performance of data and extract effective spectral information.(2) Applying de-resolve spectral pretreatment, the PLSR model got improved. When wavelength interval was set as10, the best prediction performance of PLSR model was obtained with R2=0.944and RPD=4.22. Using13PCs as the input of BPANN, the best prediction result of PC-BPANN was generated with R2=0.957and RPD=4.81Combined with wavelet coefficients, the best PLSR model was built with the result of R2=0.942and RPD=4.16. Adopting the feature band extracted by CARS, the PLSR model was produced with amazing result of R2=0.912and RPD=5.93. By means of GA, the required wavelengths for precision model, F-test model and global model were determined. Using these wavelengths, PLSR models were developed with R2of0.941,0.959and0.970, RPD of4.13,4.94and5.82, respectively.(3) The SOM detection instrument based on visible-near-infrared spectroscopy was developed. This device is compact, lightweight and convenient to carry. Generally, its measurement relative error for SOM is less than10%. All of these were conducted to accumulate the developing experience for the production of SOM detection instruments.