Hyperspectral Reflectance Characteristics Of Main Soil Types In Shaanxi Province And Organic Matter Estimation Model

Hyperspectral data has high spectral resolution and strong continuity;it can detect feature attribute information well.It has a wide application prospect in aspect of soil nutrient information monitoring because of having many advantage,such as saving time and no pollution.The spectral reflectance of soil is actually the result of comprehensive effects of various factors such as organic matter content,soil texture,soil moisture content,iron oxide,soil salinity,and clay mineral composition.How to extract and mining from soil spectral information in hyperspectral data has become a vital part of rapid and accurate monitoring of soil nutrient.At present,the prediction accuracy of soil organic matter using hyperspectral data in many studies is very high,but most of the research objects based on a single soil type or surface soil.Since the response bands of organic matter in distinct types of soils are different,the universality of the application needs further verification.Shaanxi Province has large climate span,diversity geological structure,complex soil forming environment,As a result,forming numerous soil types and abundant soil resources.Exploring the common response band of organic matter in different types of soil and improving the accuracy of organic matter hyperspectral prediction has become the emphasis and difficulty in the research of soil hyperspectral prediction.Based on the soil profile reflecting the process of soil formation as the starting point,this paper collect 216 soil samples of 51 soil profiles and 9 soil types in Shaanxi Province,including of Lou soil,Paddy soil,Loessial soil,Dark loessial soil,Aeolian sandy soil and so on.Determine soil organic matter and spectral reflectance curve in the laboratory,then take five mathematical transformation of the spectral reflectance including the original mean spectrum,first-order derivative,logarithm of reciprocal,derivative of the logarithm of the reciprocal,continuum removal transformation.Compare soil reflectance spectra at different levels in different soil types.Extract the sensitive bands based on correlation analysis with soil organic matter,establish three soil organic matter spectral prediction model,including Simple linear regression model Based on Spectral Feature Index,Multivariate Linear Model Based on Partial Least-Squares(PLSR),nonlinear support vector machine(SVM)based on kernel theory,apply independent samples to verify and evaluation the model results accuracy.The results show as follow:(1)The overall trend of spectral reflectance curves of different soil types are the same,but the spectral reflectance,absorption peak position and depth are different.In general,the Lou soil spectral reflectance is the lowest,Aeolian sandy soil and Loessial soil spectral reflectance is the highest,other types of soil spectral reflectance are in the middle.The content of organic matter is the main factor affecting the spectral reflectance of each level of the soil profile,demonstrating a significant negative correlation.Ferric oxide,calcium carbonate can also affect the soil spectral reflectance;the influence degree depends on its content.(2)Due to the interaction of organic matter and calcium carbonate,the spectral reflectance of each layer of loessial soil profile showing a change regulation of increased,reduced and then increased from top to bottom,1400 nm,1900nm band has a deep moisture absorption trough.Due to ferric oxide content,Paddy soil resulting in high reflectivity,absorption trough of ferric oxide exists near 900 nm.The spectral reflectance of Loessial soil and Aeolian sandy soil are the highest,the spectral reflectance of the soil horizon is almost the same,but except for the surface soil,organic matter content is higher than the average caused by a lower reflectivity.The spectral reflectance difference of each layer of Dark loessial soil profile is very small,However,the first-order derivative spectrum changes more widely,shows a wavy.The response spectrum of organic matter of different soil types are not the same,but the high correlation with soil organic matter appeared in 540nm~1050nm,1390~1500nm,1530~1560nm,1650~1790nm,1860~2190nm,2260~2340nm,and the sensitive band gathered around 540~630nm,750~930nm,2080 ~2130nm,2180~2190 nm.(3)Nonlinear organic matter content prediction model established by the SVR method has the highest accuracy,the PLSR method accuracy are lower and the linear regression model based on spectral indices are the worst.The estimate coefficients of calibration model and validation model established by the SVR method based on derivative of the logarithm of the reciprocal Spectral curve are respectively 0.9210,0.8874,the RMSE is 2.1843 g/kg,RPD is 2.8751,and it is the optimal prediction model of soil organic matter.