Removing The Effect Of External Parameter From Vis-NIR Diffuse Reflectance Spectra For The Prediction Of Soil Organic Matter

The soil data survey in the conventional land resource survey requires numerous soil samples and laboratory physical and chemical analysis experiments.These methods are relatively complex,time-consuming and expensive,and can not describe the spatial and temporal dynamic information of soil property over large areas.In recent years,airborne remote sensing technology and Proximal Soil Sensing(PSS)have provided potential methods for land resource investigation,and can be used to rapidly monitor and update soil property across multiple scales.However,airborne remote sensing technology is challenged by removing the influence of vegetation cover from the spectra.The proximally non-imaging visible and near-infrared spectroscopy(Vis-NIR)has the advantages of rapid,nondestructive,reproducible,and cost-effective,and can reveal the spectral characteristics of specific soil property more objectively.Compared with laboratory-based Vis-NIR spectroscopic measurement,field-based Vis-NIR spectroscopic measurements can greatly improve the efficiency of spectral data scanning due to avoiding the collection and preparation of soil samples(e.g.,transporting,air drying,grinding,sieving,etc.).However,the field spectral data are more susceptible to interference of external environmental factors,such as variable soil moisture,temperature,natural aggregation,the condition of soil surface and so on,which can lead to that the accuracy of prediction models are less accurate than that the results of laboratory-based stable environment.Among these factors,the presence of soil moisture has a large influence on spectral reflectance,which would mask the spectral absorption characteristics of soil property and further reduce the estimation accuracy.Therefore,the removal of interference from external environmental parameters is critical to improving the accuracy of the estimation model.First of all,we recorded the soil spectra by proximal in situ and stationary Vis-NIR sensing and also sampled them for measurements under laboratory conditions(a total of 78 samples).And the soil organic matter(SOM)of each soil sample was analyzed based on potassium dichromate external heating method.Five kinds of data analysis methods(raw spectral reflectance,difference spectral reflectance,continuum removal reflectance,first order differential reflectance and synchronous 2-dimensional spectra)were used to analyze the difference of spectral curves between field-moist intact soil and dried ground soil.Secondly,ratio index,difference index and normalized difference index derived from Vis-NIR spectral reflectance based on field-moist intact soil and dried ground soil were calculated using all two-band combinations in the range of 400-2400 nm.Then,the two-dimensional determination coefficients(R2)by F significant test were conducted(P<0.01),which could be used to extract sensitive spectral index,and we used partial least squares regression(PLSR)and support vector machine regression(SVMR)to build quantitative inversion models for SOM,and compared them.Finally,three non-overlapping subsets were selected:the model calibration set(S0),the development set(S1),the validation set(S2).Three kinds of methods were used to remove external environmental parameters from field-moist intact soil spectra,including direct standardization(DS),external parameter orthogonalization(EPO)and the orthogonal signal correction(OSC)methods.The final objective was to investigate and compare the effectiveness of the aforementioned three approaches(the coefficient of determination(R2),root mean squared error(RMSE)and the ratio of prediction to deviation(RPD)between the predicted and measured SOM),and choose the optimal model.The main research results are as follows:1)The difference between the field-moist intact spectra and the dried ground spectra was quite obvious;The effect of soil moisture was clearly stronger in NIR than visible,and large spectral differences could be observed at water absorption bands,especially near 1450 and 1900 nm,which might mask the absorption features of SOM.2)The model accuracy of the dried ground soil was better than that of the field-moist intact soil;In terms of the dried ground soil,compared with the one-dimensional spectral data,the maximum R2 of two-dimensional spectral index increased by 0.18;In terms of the field-moist intact soil,compared with the one-dimensional spectral data,the maximum R2 of two-dimensional spectral index increased by 0.19.In the PLSR models,the RPD values for the dried ground soil ranged from 1.87 to 2.05,and the RPD values for the field-moist intact soil varied from 1.65 to 1.76.In the SVMR models,the RPD values for the dried ground soil ranged from 2.19 to 2.92,and the RPD values for the field-moist intact soil varied from 1.86 to 2.06.The model accuracy of the dried ground soil was better than that of the field-moist intact soil.The nonlinear PSO-SVMR model could improve the model accuracy of field-moist intact soil,and ensure the flexibility and fault-tolerance of the practical application.The maximum RPD value was 2.06.3)In the PLSR models established after the correction of variable moisture conditions and other environmental factors,the OSC-PLSR model performed the best.In practical application,the parameters that needed to be optimised for EPO-PLS and OSC-PLSR were g(the number of EPO dimensions)and c(the number of filter factors).These two parameters have a certain impact on the model accuracy.From the perspective of predictive performance after the correction of variable moisture conditions and other environmental factors,OSC-PLSR model>DS-PLSR model>EPO-PLSR model.After OSC,the field-moist intact soil spectra had nearly identical positions in the feature space to the corresponding dried ground soil spectra,the values of R2Pret RPD for the OSC-PLSR model were 0.71 and 1.89,respectively,which may be used for approximate quantitative estimation of SOM under feld conditions.In the future,the OSC-PLSR model can be applied to the Vis-NIR spectral dynamic monitoring SOM information in the field-moist intact environment.The field-moist intact soil Vis-NIR spectroscopy provides a quick and non-destructive method for data acquisition,which greatly enriches the means of soil survey and has important practical significance to improve the efficiency of dynamic monitoring of cultivated land quality.