Partial Least Square Regression Approach for the Extraction Phosphorus Concentration from Soil VNIR Spectra

Abstract

The present study used partial least square regression (PLSR) method to establish the relationship between soil reflectance spectra measured under laboratory condition and soil total phosphorus. The spectral range of soil reflectance was divided into two wavelength ranges (320nm~1100nm, and 1000nm~2480nm). Besides original reflectance (R), several spectral parameters were also calculated: derivative reflectance spectra (FDR), inverse-log spectra (Log (1/R)) and Depth. All of the 69 samples were randomly divided into training and test set. Based on the PLSR method, statistical relationships between the four spectral parameters and total phosphorus were established with training set and were validated using the test set. In all models, the prediction values for high phosphorus concentration were lower than the measured values. The precision of models built using near infrared (NIR) range is better than that derived with visible bands except for the depth parameter. Good results indicate that VNIR spectra, especially NIR range, are potentially useful for rapid estimation of phosphorus concentration in soils.