Estimation of Soil Organic Matter Content Based on Regional Feature Bands

Abstract

To estimate soil organic matter (SOM) content using hyper-spectral data, regional feature bands and principle component regression (PCR) were built a model of SOM. The results showed that the coefficients of determination (R2) of PCR model based on the regional feature bands were 0.650 for calibration set and 0.628 for validation set, respectively. The Root Mean Square Error (RMSE) values were 2.641 g/kg and 2.852 g/kg, respectively. The PCR models based on the significant bands had better estimation accuracy, but its total correlation coefficient(R = 0.770) between predicted SOM and measured SOM was lower than of model based on the regional feature bands (R = 0.803). Therefore, the PCR model based on regional feature bands provides a better estimation result than model based on significant bands.