Determination Of Soil Organic And Total Nitrogen By Modeling With Different Spectral Pretreatment Methods

The realization of precision agriculture requires real-time,rapid and accurate detection of soil nutrient status,rational fertilization,and increase crop yield.Soil organic matter and total nitrogen are essential nutrients for healthy growth of crops and key indicators to measure soil fertility.Therefore,the determination of soil organic matter and total nitrogen content is significant for precision agriculture.The traditional laboratory method for determining the content of organic matter and total nitrogen in soil is relatively slow and costly.Because it is a chemical experimental method,it easily pollutes the environment.With the development of modern optical and computer data processing technologies,near-infrared spectroscopy combined with chemometrics has become the fastest-growing data analysis technology since the 1990s.It completely overcomes the shortcomings of traditional laboratory methods for measuring components.Being non-destructive,low cost,convenient test,fast speed,no pollution,etc.,near-infrared spectroscopy is very suitable for determining the organic matter and total nitrogen content in soil.In this study,189 soil samples from Shandong Province were obtained.The soil organic matter and total nitrogen content were predicted by collecting soil near-infrared spectroscopy,abnormal sample rejection,spectral preprocessing,and establishing PLS quantitative regression model.Then,we select the spectral pretreatment method which is most suitable for the determination of soil organic matter and total nitrogen by correlation coefficient and root mean square error of the model.Firstly,the near-infrared spectrum of the soil sample was collected and analyzed for the spectral characteristics.The near-infiared spectrum of the soil has three absorption peaks at 1400,1900 and 2200 nm,and the absorbance gradually decreases with wavelength increasing.PCA-MD was used to eliminate abnormal samples of soil.Principal component analysis is to reduce the amount of data,eliminate line overlap and spectral matrix collinearity,and reduce data interference.Calculate data after principal component analysis.The Mahalanobis distance,the threshold was set,the samples within the threshold range were retained,and samples outside the threshold were rejected.Then,different spectral pretreatment methods were used to treat the near-infiared spectrum of the soil to eliminate spectral noise interference.In this paper,the characteristics of near-infrared spectroscopy after different spectral pretreatments were analyzed.However,the optimal spectral preprocessing method can’t be compared by the characteristics of near-infrared spectroscopy.It is also necessary to establish a PLS quantitative regression model and compare them according to the corresponding model indicators.The optimal spectral pretreatment method was selected.Finally,the PLS quantification was established by using the normalized,smoothed,first derivative,second derivative,normalization and smooth combination,normalization and derivative combination,smoothing and derivative combination,and normalized smooth derivative.The smoothing method uses 3-point smoothing,5-point smoothing,and 7-point smoothing.The establishment of the PLS quantitative regression model,the determination of the optimal number of pr-incipal components is a key step.This paper used the K-fold Cross Validation(K-CV)method to determime the optimal number of principal components.The modeling results show that the combination of 3-point smoothing and first derivative is most suitable for the determination of organic matter and total nitrogen in soil.The soil organic matter PLS model,the root mean square error RMSECV and the correlation coefficient Rt of the training set are 0.1385 and 0.96496,respectively;the root mean square error RMSEP and correlation coefficient RP of the prediction set are 0.1819 and 0.9276.The PLS model of soil total nitrogen,the RMSECV and correlation coefficient Rt of the training set are 0.0841 and 0.99435,respectively;the RMSEP and correlation coefficient RP of the prediction set are 0.1157 and 0.9821.The way in which the method combinations were demonstrated was feasible in determining soil organic matter and total nitrogen content.