Hyperspectral Estimation Models Of Major Soil Nutrients And Subregion Application In Farmland Of North China Plain

Soil nutrient is very important for agricultural cultivation.It is worth studying how to obtain soil nutrient information efficiently,accurately and non-invasively.This study takes Jingxian County,Yucheng County and Shenxian County,three typical agricultural counties of North China Plain,as the study area and takes soil total nitrogen,available phosphorus and available potassium as research objects,and hyperspectral skill is used as research means.Pearson correlation analysis was used to screen characteristic wave bands of soil total nitrogen,available phosphorus and potassium.Multivariate linear regression,partial least squares regression,support vector machine and artificial neural network were used to establish the estimation models of nutrient contents.On account of the low precision of soil nutrient modeling,Soil nutrient hyperspectral estimation model was established on the basis of dividing soil samples into textures,which was detected the hyperspectral estimation of soil nutrients in farmland in North China Plain and its subregion application.The main research contents and results are as follows:1)Hyperspectral estimation models of major soil nutrientsBased on the spectral reflectance characteristics of different soil nutrients,the sensitive bands were used to estimate soil nutrients.The characteristic bands of total nitrogen nutrients in farmland of North China Plain were 1060 nm,1527 nm,1543 nm,1675 nm,1776 nm and1915 nm.The sensitive bands of available phosphorus in soil were 1060 nm,1131 nm,1301nm,1469 nm,1491 nm,1817 nm,1907 nm,1950 nm,1964 nm and 1994 nm.The sensitive bands of available potassium were 1131 nm,1539 nm,1059 nm,1760nm,1645 nm,1950nm,1909 nm.The Hyperspectral Estimation Models of soil nutrients were established by using four methods:multiple linear regression,partial least squares regression,support vector machine and artificial neural network.The best estimation model of soil nutrients was selected according to the magnitude of determination coefficient R~2and average relative error.The best estimation models of soil total nitrogen,available phosphorus and available potassium were partial least squares regression model?multiple linear regression model and partial least squares regression model.From the point of view of model accuracy among nutrients,the best model accuracy of available phosphorus>the best model accuracy of total nitrogen>the best model accuracy of available potassium.From the point of view of model method,partial least squares regression has the highest prediction accuracy,followed by multiple linear regression.2)Subregion indicators for spectral estimation of major soil nutrientsAccording to the natural climate conditions,soil parent material,soil texture,nutrient status and management level of farmland in North China Plain,referring to the evaluation factors of cultivated land fertility in this area and combined with the selected sample sites in this paper,the soil nutrient spectral evaluation index was selected from soil texture,irrigation and drainage capacity,climate conditions.The total nitrogen,available phosphorus and available potassium contents of soil samples were analyzed by statistical analysis of the three zoning characteristics to clarifying the status of soil nutrient content in the experimental area and the importance of sub region on soil nutrients.According to the mean and variability of soil nutrients in each zoning factor grade,soil texture was selected as the subregion index for nutrient spectral estimation.The effect of texture on soil nutrient content and variation is obvious.After texture division,soil nutrient variability is lower than the whole.Furthermore,the spectral characteristics of soils with different textures were analyzed.In the 600-1900 nm spectral region,the spectral reflectance of soils with three textures was significantly different.Therefore,it is feasible to establish the hyperspectral estimation subregion models of soil major nutrients based on soil texture.3)Subregion application of spectral estimation of soil major nutrientsSoil sample data were classified into loam,sandy and clay soil according to soil texture.Clustering operations were carried out to calculate the correlation coefficients between nutrient content values and spectral reflectance derivatives.Four modeling methods were used to model nutrient estimation.The optimal model is selected by determining coefficient R~2and average relative error.According to the modeling accuracy and validation results,The best models for estimating total nitrogen,available phosphorus and available potassium in loam are multivariate linear regression model,artificial neural network model and partial least squares regression model,respectively;the best models for estimating total nitrogen,available phosphorus and available potassium in sandy soil are multivariate linear regression model,support vector machine model and partial least squares regression model,respectively;the best models for estimating total nitrogen,available phosphorus and available potassium in clay are multivariate linear regression model,support vector machine model and artificial neural network model.Compared the results of nutrient models,except available phosphorus and available potassium in clay soil,the accuracy of soil nutrient models in different texture areas are higher than that of the whole soil nutrient models,which indicates that the prediction models of soil nutrient after texture division are better.The comparison of modeling methods shows that the partial least squares regression and artificial neural network methods have higher accuracy and better precision.This study explored a non-destructive,simple and rapid method for estimating soil nutrients in Farmland of North China Plain,and studied the hyperspectral estimation of soil nutrients and zoning application.It had important theoretical and practical significance for hyperspectral application in estimating soil nutrients in large areas.