Better soil structure promotes extension of plant roots thereby improving plant growth and yield.Differences in soil structure can be determined by changes in the three phases of soil,which in turn affect soil functio...Better soil structure promotes extension of plant roots thereby improving plant growth and yield.Differences in soil structure can be determined by changes in the three phases of soil,which in turn affect soil function and fertility levels.To compare the quality of soil structure under different conditions,we used Generalized Soil Structure Index(GSSI)as an indicator to determine the relationship between the“input”of soil three phases and the“output”of soil structure.To achieve optimum monitoring of comprehensive indicators,we used Successive Projections Algorithm(SPA)for differential processing based on 0.0–2.0 fractional orders and 3.0–10.0 integer orders and select important wavelengths to process soil spectral data.In addition,we also applied multivariate regression learning models including Gaussian Process Regression(GPR)and Artificial Neural Network(ANN),exploring potential capabilities of hyperspectral in predicting GSSI.The results showed that spectral reflection,mainly contributed by long-wave near-infrared radiation had an inverse relationship with GSSI values.The wavelengths between 404-418 nm and 2193–2400 nm were important GSSI wavelengths in fractional differential spectroscopy data,while those ranging from 543 to 999 nm were important GSSI wavelengths in integer differential spectroscopy data.Also,non-linear models were more accurate than linear models.In addition,wide neural networks were best suited for establishing fractional-order differentiation and second-order differentiation models,while fine Gaussian support vector machines were best suited for establishing first-order differentiation models.In terms of preprocessing,a differential order of 0.9 was found as the best choice.From the results,we propose that when constructing optimal prediction models,it is necessary to consider indicators,differential orders,and model adaptability.Above all,this study provided a new method for an in-depth analyses of generalized soil structure.This also fills the gap limiting the detection of soil three phases structural characteristics and their dynamic changes and provides a technical references for quantitative and rapid evaluation of soil structure,function,and quality.展开更多
基金funded by the National Natural Science Foundation of China(31871571,31371572)the earmarked fund for Shanxi Province Graduate Education Innovation Project(2022Y312)+3 种基金supported by Modern Agro-industry Technology Research System(2023CYJSTX02-23)Scientific and Technological Innovation Fund of Shanxi Agricultural University(2018YJ17,2020BQ32)Key Technologies R&D Program of Shanxi Province(201903D211002,201603D3111005)National Key R&D Program of China(2019YFC1710800)。
文摘Better soil structure promotes extension of plant roots thereby improving plant growth and yield.Differences in soil structure can be determined by changes in the three phases of soil,which in turn affect soil function and fertility levels.To compare the quality of soil structure under different conditions,we used Generalized Soil Structure Index(GSSI)as an indicator to determine the relationship between the“input”of soil three phases and the“output”of soil structure.To achieve optimum monitoring of comprehensive indicators,we used Successive Projections Algorithm(SPA)for differential processing based on 0.0–2.0 fractional orders and 3.0–10.0 integer orders and select important wavelengths to process soil spectral data.In addition,we also applied multivariate regression learning models including Gaussian Process Regression(GPR)and Artificial Neural Network(ANN),exploring potential capabilities of hyperspectral in predicting GSSI.The results showed that spectral reflection,mainly contributed by long-wave near-infrared radiation had an inverse relationship with GSSI values.The wavelengths between 404-418 nm and 2193–2400 nm were important GSSI wavelengths in fractional differential spectroscopy data,while those ranging from 543 to 999 nm were important GSSI wavelengths in integer differential spectroscopy data.Also,non-linear models were more accurate than linear models.In addition,wide neural networks were best suited for establishing fractional-order differentiation and second-order differentiation models,while fine Gaussian support vector machines were best suited for establishing first-order differentiation models.In terms of preprocessing,a differential order of 0.9 was found as the best choice.From the results,we propose that when constructing optimal prediction models,it is necessary to consider indicators,differential orders,and model adaptability.Above all,this study provided a new method for an in-depth analyses of generalized soil structure.This also fills the gap limiting the detection of soil three phases structural characteristics and their dynamic changes and provides a technical references for quantitative and rapid evaluation of soil structure,function,and quality.
