Estimation Of Soil Organic Carbon Content In Farmland In Inner Mongolia Based On Mid-infrared Spectroscopy

As an important indicator of soil fertility,soil organic carbon(SOC)plays a long-term key role in plant growth and development,and the quantitative prediction model of soil organic carbon based on mid-infrared spectroscopy can effectively estimate soil organic carbon.In order to explore the quantitative prediction model of soil organic carbon suitable for farmland in Inner Mongolia,864 farmland soil samples were collected in Inner Mongolia and the soil organic carbon content and mid-infrared spectral information were measured,and the mid-infrared spectral information was pretreated with various combinations,and the soil organic carbon quantitative prediction models of partial least squares regression(PLSR),principal component regression(PCR)and support vector machine regression(SVR)were established and optimized by combining the soil organic carbon content.The model validation results are used to filter out the optimal modeling methods for this area.The findings are as follows:(1)The partial least squares regression,principal component regression and support vector machine regression models of soil organic carbon were established on the data processed by 90 different preprocessing methods,and the corresponding cross-validation root mean square errors(RMSECV)were analyzed.value found:Compared with the original spectral modeling,the prediction accuracy of the three methods has been improved after orthogonal signal correction(SNV)combined with centralization(Centralization),SG smoothing and Baseline processing,indicating that orthogonal signal correction can increase the The signal-to-noise ratio of the mid-infrared spectrum in this region.(2)Comparing the three modeling methods,it is found that:the obtained modeling results all reflect that the partial least squares regression is better than the principal component regression than the support vector machine regression.Outlier removal and parameter adjustment for SNV+SG smoothing-PLSR and Normalization-PCR using leverage values respectively R~2=0.7852,RMSEP=1.3418 g/kg,RPD=2.1406 for PLSR after parameter optimization for Baseline+MSC-SVR;PCR R~2=0.6447,RMSEP=1.7173g/kg,RPD=1.6726;SVR R~2=0.7101,RMSEP=1.5464 g/kg,RPD=1.8574,among which,the PLSR prediction model is a very good quantitative prediction model(2.0<RPD<2.5),which can make a good prediction of soil organic carbon in farmland in this area.(3)Comparing the data of the validation set by banners and counties,it is found that the prediction effect of Arong Banner(black soil)and Horqin Left Wing Middle Banner(chestnut soil)is better,while the prediction effect of Hetao Irrigation Area(alkaline soil)is poor,indicating that alkaline soil is not suitable for mixed soil Model establishment.And when the sub-ban counties are verified,the prediction accuracy of the model established by PLSR is better than the other two methods.To sum up,the optimal modeling method for establishing a prediction model for farmland soil organic carbon content in Inner Mongolia based on mid-infrared spectroscopy is partial least squares regression,and the corresponding optimal preprocessing method is SNV+SG smoothing,so this method(SNV+SG-PLSR method)can be used as the estimation method of soil organic carbon content in this area.