Reflectance Spectroscopy for Routine Agronomic Soil Analyses

Near-infrared reflectance spectroscopy is a demonstrated tool for quantitative analysis of numerous soil properties. Reported advantages include analytical precision, predictive accuracy, and reduced costs and processing times. A library (N = 1933) representing all major soil orders in Florida was assembled from samples submitted to the University of Florida Extension Soil Testing Laboratory for routine testing during 2004-2005. High-resolution diffuse reflectance spectra from each sample in the visible/near infrared were used to predict observations made using standard laboratory analytical procedures for soil pH, Mehlich-1-extractable P, K, Ca, Mg, Cu, Mn, and Zn, percentage of organic matter, and saturated hydraulic conductivity (Ksat). Calibrations were immediately applicable for organic matter and Al, based on relative performance determinant values (RPD = standard deviation/standard error of validation > 2.0). Models for pH, P, Ca, and Ksat showed moderate accuracy (1.5 < RPD < 2.0), whereas those for K, Cu, Mg, Mn, Zn, and Fe exhibited low efficiency (RPD < 1.5), indicating a need for further refinement before near-infrared reflectance spectroscopy is a viable alternate method to standard laboratory procedures. Prediction of soil fertility and productivity based on published Florida soil diagnostic categories showed effective discrimination for pH, P Mg, Mn, Cu (phytotoxicity), and Ksat. Our study showed that prediction efficiency is a strong function of mean nutrient/analyte concentration in the soil. We further demonstrated that near-infrared reflectance spectroscopy error rates were comparable to, and in some cases smaller than, laboratory analytical error rates, suggesting that the observed low spectral prediction efficiency may be substantially because of uncertainty in the laboratory data.