| Selenium as a trace element in the human body is essential,but excessive or lack of it will affect health.The selenium content of selenium-rich foods is thought to be closely related to the selenium content of the soil that provides growth for them.Ziyang of Shaanxi province is a natural selenium-rich area in China.It is of positive significance for the local residents’health and the development of selenium-rich products to make clear the influencing factors and contents of soil selenium.In this paper,the northern part of Ziyang County,Ankang City,Shaanxi Province was selected as the study area to find out the factors that affect the soil selenium.In order to obtain the selenium content of soil,a prediction model was established based on the remote sensing hyperspectral data of soil.The conclusions are as follows:(1)The influencing factors of soil selenium content are as follows:Soil selenium in the study area had a positive correlation with elevation,and had no obvious correlation with slope and slope direction.There was a significant positive correlation between soil selenium content and parent selenium of soil formation,which indicated the inheritance of soil selenium to parent selenium.In terms of land use type,soil selenium content decreased successively in grassland,dry land and forest land.Among the two soil types,the selenium content of the yellow brown soil was higher than that of the brown soil.In terms of soil physical and chemical properties,soil pH had no significant effect on soil selenium content,but the soil where the sampling point was located was mostly acidic.There was a positive correlation between soil organic matter content and selenium content.(2)Combined with the significant effect of soil organic matter content on soil spectral curve,the idea of better searching the characteristic bands of soil selenium spectrum through the relationship between organic matter and selenium was put forward.In order to find the characteristic bands of soil selenium spectrum,the original soil spectrum was corrected by continuous removal,first-order differential,reciprocal logarithm,multiple scattering correction transformation and standard normal variable correction.It was found that the correlation of the five transformations was improved to different degrees,but the reciprocal logarithm effect was poor,and the characteristic band of soil selenium could not be found out.After four transformations,the characteristic bands of soil selenium spectrum were concentrated around 750nm,1250nm and 2200nm.Partial least square method was used to model the feature bands and selenium content selected after the continuum removal,first-order differentiation,multivariate scattering correction and standard normal variable correction,respectively.The modeling effect of first-order differential transformation was the best.The modeling R~2 was 0.48,RMSE was 0.56,and the verification model R~2 was 0.41 and RMSE was 0.68.Using random forest method for modeling,the modeling effect of continuum removal and first-order differential transformation was better.The modeling R~2 of continuum removal transformation was0.91,RMSE was 0.31,the verification model R~2 was 0.55,RMSE was 0.64,the modeling R~2 of first-order differential transformation was 0.93,RMSE was 0.27,the verification model R~2 was 0.55,RMSE was 0.60.The reverse neural network algorithm modeling,the first-order differential transformation modeling effect is the best,modeling R~2 was 0.64,RMSE was 0.47,validation model R~2 was 0.59,RMSE was 0.57.The ability of random forest and reverse neural network to fit soil selenium content model is better than partial least square algorithm. |
PDF Full Text Request