Prediction Of Total Iron And Free Iron Contents In Soil Based On Visible-near Infrared Spectroscopy In Guangxi

Iron is not only a necessary nutrient element for plant growth and development,but also an important element affecting soil formation process and soil properties.Iron is one of the main colorants in the soil of South China.The content of different forms of iron is an important indicator to characterize the degree of soil weathering development in this region.Laboratory chemical method is the main method to obtain free and total iron content in soil.Although the precision is high,it is time-consuming and laborious.Spectral reflectance method has the characteristics of real-time and high efficiency.The simple and rapid determination of free iron and total iron in soil by spectral technique is of great significance for understanding the biogeochemical cycle of elements,the process of soil genesis and guiding Chinese Soil Taxonomy.In this paper,162 soil profiles were collected from Guangxi Zhuang Autonomous Region,with a total of 666 soil samples.After natural air drying and grinding in the laboratory,the spectral data of 0.25 mm and 2 mm soils in the range of350～2500 nm were measured.The application potential of visible and near infrared reflectance spectroscopy in the prediction of iron oxide content in soils with different particle sizes was studied,and the effects of organic matter content on the spectral characteristics and quantitative inversion of soil iron oxide were discussed.The main results are as follows:（1）The spectral reflectance of soils with different particle sizes was negatively correlated,and the spectral reflectance of soils with<0.25 mm particle size was higher than that of soils with<2 mm.The absorption valleys and reflectance peaks of free iron and total iron soils with different particle sizes are the same,but the reflectance is different,and the influence on reflectance in near infrared band is higher than that in visible band.There is no obvious relationship between the spectral reflectance and soil with different free iron and total iron content of<0.25 mm,but there is a negative correlation between the spectral reflectance and soil with<2 mm particle size.That is,the higher the content of free iron or total iron,the lower the spectral reflectance.The correlation between spectral reflectance and soil iron oxide content is improved to some extent by different spectral transformation treatments,such as continuum removal,reciprocal logarithm,reciprocal first-order differential,square root first-order differential and logarithmic first-order differential.From different particle sizes,the correlation between free iron and total iron content in soil with<2 mm particle size treated by original and reciprocal logarithmic spectral transformation is higher than<0.25 mm,and the correlation between them is continuum removal,reciprocal first-order differential,square root first-order differential and logarithmic first-order differential.（2）The prediction models of free iron and total iron content in soil were established by linear model stepwise multiple linear regression（SMLR）,partial least squares regression（PLSR）and non-linear model back propagation neural network（BPNN）.The prediction performance of the models was tested by the determination coefficient R² of three evaluation indexes,root mean square error RMSE and relative analysis error RPD.The results showed that the best model for predicting the content of free iron in soils with<0.25 mm diameter was better than that of soils with 2 mm diameter.The best prediction model of soil free iron content is<0.25 mm particle size BP neural network combined with square root first order differential transformation.The validation set R_P²=0.87,RMSE_P=8.24 g·kg^-1,RPD=2.63.Soil total iron content model cannot be successfully predicted when<2 mm diameter soil is used.The best prediction model of soil total iron content established by<0.25 mm diameter soil is BP neural network combined with continuum removal spectral transformation,the validation set R_P²=0.80,RMSE_P=11.67 g·kg^-1,RPD=2.22.Ability.In conclusion,the optimum soil particle size used to establish the prediction model of iron oxide content in soil is<0.25 mm.（3）There was a significant negative correlation between soil organic matter and iron oxide content in the study area,and the correlation between soil organic matter and free iron was higher than that of total iron.There is a typical negative correlation between soil organic matter content and spectral reflectance,that is,the increase of soil organic matter content,the decrease of spectral reflectance,and the spectral characteristic band of soil organic matter is580～1200 nm.Organic matter content has an effect on the spectral characteristics and quantitative inversion of soil iron oxides.The results showed that when the content of soil organic matter was>20 g·kg^-1,the reflectance peak near 800 nm and the absorption valley near 900 nm of iron oxide could be masked.With the increase of soil organic matter content or the ratio of organic matter to iron oxide content,the stability and prediction ability of the inversion model of soil iron oxide content decreased,but when the content of soil organic matter was less than 20 g·kg^-1 and the ratio of organic matter to iron oxide was less than 0.40,organic matter had no effect on it.