Prediction of Soil Organic Matter Content Using VIS/NIR Soil Sensor

Abstract

In present work, the potential of visible/near-infrared (VIS/NIR) soil sensor to predict organic matter (OM) contents in soils was studied using VIS/NIR soil sensor. Different preprocessed methods were studied to improve the correlations of spectra with organic matter contents in soil samples, and three regression methods, including direct linear regression, principal component regression (PCR) and back propagation-neutral network (BP-ANN), were employed to construct the prediction models in calibration stage, whose performances were evaluated in the validation stage. The prediction results indicated that PCR for the preprocessed spectra by MSC together with S-G polynomial filter, and BP-ANN for the first ten principal components of the spectra by MSC pretreatment could provide appropriate results for the estimation of OM contents in unknown soil samples.

Keywords

• Soil OM content
• Near-infrared spectra
• Principal component regression
• BP neutral network.