Application Of Near Infrared Spectroscopy(NIRs) In Determination Of Soil Nitrogen,Phosphorus And Nitrogen In Forest Soil

Soil nutrient is an important indicator of soil fertility and provides nutrient sources for plant growth.The determination of soil carbon(C),nitrogen(N),phosphorus(P)and other nutrient contents are mainly analyzed by chemical method,this method is heavy cumbersome and costly.It is shown that near-infrared spectroscopy(NIRs)has the advantages of fast test speed,high efficiency and less pollution,and can be used to determine soil nutrient content.However,the NIRs method requires a large number of samples,as well as measured nutrient and spectral data.It is also necessary to pre-treat the spectral data and then establish a model to test and optimize the model before it can be used to determine the soil nutrient content.Therefore,it is of great value to study the content of C,N and P in forest soil by NIRs method.In this study,a 100 m x 100 m permanent plot is established in the Choerospondias axillaris deciduous broadleaved forest and the Lithocarpus glaber-Cylobalanopsis glauca evergreen broadleaved forest,respectively.A total of 100 samples of 10m × 10m are parted in the plot by 10cm,10-20cm and 20-30cm soil samples.The soil organic carbon,total nitrogen and total phosphorus are measured by near infrared spectroscopy and stoichiometric method.The soil composition model was determined by near infrared spectroscopy.The use of near-infrared spectroscopy technology is high efficient with low cost,fast test,and green pollution-free advantages.(1)The near-infrared spectral information of each soil sample ranges from 4000 nm to 11000 nm.Six methods,including linear reduction method,vector normalization method,minimum maximum normalization method,multiple scattering correction,first derivative and second derivative,are used as pretreatment,most of the information of the spectrum is concentrated in the range of 4000 nm-7000 nm,the obvious absorption peak ranges from 4400 nm to 5400 nm,the characteristic spectrum is 4400 nm-5400 nm,the pre-treated spectrum basically retains the absorbing characteristics of original spectrum,which can better determine its spectral characteristics.(2)The higher the number of soil samples is,the higher the model fitting degree achieves.If the number of samples reaches a certain number(400 soil samples),the model of the decision coefficient(R1)is 0.9 or more,and stabilizes.The prediction model of soil nutrient content in the evergreen broadleaved forest of Lithocarpus glaber-Cylobalanopsis glauca forest is lower than that of the prediction model of soil nutrient content in the Choerospondias axillaris forest deciduous broad-leaved forest.After the two soil samples were integrated into the soil,the fitting degree of the model is lower than that of the Choerospondias axillaris forest,but higher than that of the Lithocarpus glaber-Cylobalanopsis glauca forest.It is probably because the soil of Lithocarpzts glaber-Clobalanopsis/gluca forest is more complex,and the nutrient content is relatively difficult to be fitted by linear method.It is also indicated that the same prediction model will have different fitting degree in different forest soils.(3)The prediction degree of soil organic C and total N content is higher than that of soil total P prediction model,and the prediction value of soil organic C and total N is closer to the true value.The prediction coefficient(R2)(RMS)ranges from 0.8 to 0.9,the relative error(RPD)is in the available range(>2.0),indicating that the soil containing bonds C-H,N-H and O-H are more suitable for NIRs technology to determine its content,and the established prediction model is reproducible and can be widely used in nutrient determination of forest soil.(4)In the OPUS software,according to the effects of the six pretreatment methods,the dimension and the coefficient of determination(R2)of the prediction model are improved after the pretreatment with the multiple scattering correction method.The cross-test RMS error(RMSECV)is relatively small,relative analysis error(RPD)is relatively high.A variety of pretreatment methods will be used to further improve the fitting degree and accuracy of the model.It is suggested that multiple scattering correction and vector normalization should be used simultaneously for pretreatment.