Spectral Soil Signatures and sensing Phosphorus

Developing a phosphorus (P) sensing system using ultraviolet (UV), visible (VIS) and near infrared (NIR) provides a non-contact, cost and time effective, and less laborious measurement of phosphorus concentration in a soil sample. The determination of soil spectral signatures improves the prediction accuracy of calibration models for the determination of soil P concentration. This study aimed to obtain spectral signatures of soil samples and develop calibration models for sensing P concentrations in a soil sample. A total of 345 soil samples were collected from the drainage basin of the Lake Okeechobee to obtain a broad range of P concentrations from very low to very high (1-2709 mg/kg). Thirty-eight soil samples from different fields were sub-sampled and leached in order to remove their existing nutrient contents. The reflectance spectra of leached and unleached soil samples were measured using UV, VIS and NIR spectroscopy. Phosphorus concentrations of the samples were correlated with the absorbance of the same sample. Leached soil signatures were subtracted from those of unleached soil samples to obtain only constituent spectrum. Data analysis for P concentration determination was conducted with and without soil signature correction. For each set, two-thirds of the data was used for calibration while one-thirds of the data was for validation. Partial least squares analysis (PLS) was applied to the data set. Values for the coefficient of determination (R2) increased from 0·87 to 0·93 when original soil and constituent spectra were used, respectively. Also, using constituent spectra rather than original soil absorbance spectra reduced the root mean square errors (RMSEs) from 222 to 172 mg/kg.