Quantitative Prediction of Soil Organic Matter of Mined-Land by Hyperspectral Remote Sensing and Mathematical Statistics

Remote sensing provides an effective approach to extract soil information, whereas to construct appropriate hyperspectral model is the key for soil property estimation. In this study, the ASD portable spectrometer was used to measure hyperspectral reflectance of the soil of subsided mined-land in Jining city, China. Spectral reflectance was transformed into its reciprocal, logarithmic, the reciprocal of logarithmic and their respective corresponding first-order derivative for sensitive spectral variable extraction with respect to soil organic matter (SOM) estimation. By correlation analysis between mathematically transformed spectral reflectance and SOM, the absolute values of correlation coefficients of the first derivative at 930 nm achieved the maximum correlation coefficient. Through model performance comparison analysis, it turned out that the multiple linear stepwise regression model outperformed the simple linear regression model and the model with (1 / log (R))' as spectral variables performed best. Its variables obtained high R2 value as 0.965 and low RMSE as 0.415.