Analysis On Characteristics Of Soil Organic Matter In Opencast Using Ground Hyperspectral Data

With the formation and the continuous development of remote sensingtechnology, since the1980s, hyperspectral remote sensing is gradually matured,which is a large area simultaneous observations,obtained the data timeliness,continuous, non-destructive and other characteristics, then make remote sensingtechnology is widely used in the soil environment monitoring,and gradually become anew way to acquire data information for precision agriculture, mineral compositionand soil nutriention monitoring rapidly. Soil organic matter content is an importantindicator to measure the level of soil fertility. Soil organic matter is not only thesupporters who required various nutrients for plants growth.Simultaneouslydetermines the formation of soil structure, improve soil physical properties. Usinghyperspectral remote sensing to predict the organic matter content provides the basisfor land reclamation and ecological restoration of Opencast mining area.In this study,60soil samples which collected from Zhundong opencast mineswere measured the spectroscopy indoor and physic-chemical analyzed the organicmatter content,then made seven math processing of the original spectralreflectance:the original reflectance(REF),first derivate of spectral reflectance (FDR),second derivate of spectral reflectance(SDR), the reciprocal logarithm first derivativeof spectral reflectance(lg(1/R)′),the reciprocal logarithm second derivative ofspectral reflectance(lg(1/R)″), the normalized difference index(NDI),the continuumremoved(CR).Compared the original spectral reflectance curves with the othermathematical transformation curves, through its correlation with soil organic mattercontent, we found the normalized difference index is the optimal index.Established the predictive models with full bands based on the stepwise multiplelinear regression(SMLR) and the partial least squares regression(PLSR), and test thestability and accuracy of two models.In this study,chose the normalized differenceindex to compare prediction value with observation value,results demonstrated thattwo models’ inversion modeling group determination coefficients were high, reaching at0.80,the root mean square error were less than0.60%; judged from the results ofvalidation group, stepwise multiple linear regression and partial least squaresregression’s determination coefficient were0.71and0.75, respectively, root meansquare error of4.33%and44.32%.From the view of rate of prediction to deviation(RPD), the partial least squares regression model were more stable and higherprecision than the stepwise multiple linear regression model.Partial least squares regression model was the optimal model to predict the soilorganic matter content, and the results are more consistent with the results of previousstudies, and thus proved the universality of the method,which is a good attempthelped to analyze soil characteristics and revealed spectral features bands.