Modeling Island Soil Elements With Visible Near Infrared Reflectance Spectroscopy In Pingtan

As the precious natural resource,Soil in island has ecological multi-functions,because the soil in island is easily disturbed by wind and water erosion with high vulnerability,it is liable to be degraded by overexploitation,therefore,investigating the soil and recording the soil background quality is necessary for assessment the impact of island exploitation on local soil environment.Traditionally,the cost of monitoring soil chemical elements of soil samples is expensive,nowadays,predicting soil properties with visible-near-infrared reflectance spectroscopy is fast developed because of the low cost and time-saving.The study selected 75 soil samples in Pingtan island,analyzed the hyperspectal reflectance(350-2500 nm)with ASD in the laboratory,soil organic matter,total nitrogen,iron,cobalt element,transfomed the primitive reflectance with many kinds of mathematical methods,and build up the models for predicting soil chemical elements with the hyperspectral reflectance data.Mathematic transformation on the original spectral reflectance includes 11 algorithms such as logarithmic,absorbance,reciprocal,logarithmic and differential etc.The regression models include multiple stepwise regression analysis(MLSR),principal component analysis(PCA),partial least squares regression(PLSR)and Back-Propagation Artificial Neural Network(BP-ANN).Comparing the four modeling algorithms with 11 spectral transformation,the paper selected the optimal inversion models for soil organic matter,total nitrogen,iron and cobalt according to the model accuracy.Research results are listed as following:(1)Sensitive brands of reflectance spectroscopy to soil properties are picked out respectively.For soil organic matter,sensitive bands are 611-776nm,878-938nm,1121-1241nm,1321-1351nm,1571-1601nm.For soil total nitrogen,sensitive bands are 437-524nm,626-758nm,920-932nm,1121nm,1501-1531nm,1961-2041 nm,2241-2261nm;For soil iron,sensitive bands are 450-623nm,806-869nm,1351-1501nm,1901nm,2101-2391nm;For soil cobalt,sensitive bands are:350-518nm,540-614nm,737-884nm,1041-1101nm,1271-1291nm,1361-1501nm,1911-1991n=,2101-2301nm.(2)The paper improved the accuracy assessment indicators for spectral inversion models.Normally,the coefficient of determination(R2)or Root Mean Square Error(RMSE)are the indicators for model judgement.The paper applied the comprehensive evaluation system made of R2,verify R2 and the RPD to assess the accuracy of the models.(3)The paper choose the optimal hyperspectral inversion models for predicting soil SOM,TN,Fe,Co.In the light of the accuracy,stability and convenience,models of MLSR and BP-ANN for predicting soil elements with hyper-spectral data are superior to PLSR and PCA,and the better spectral transformation algorithm are the first order differential on the original reflection and the first order differential on the absorption.(4)Comparing the inversion results by use of the model developed in this paper for island soil elements with that by use of the model developed from soil samples in Fuzhou,the paper found that the inversion modeling technique can be universal,while the models may be different,and the inversion results may be satisfied with a limited accuracy.