Estimating Soil Particle Density using Visible Near‐infrared Spectroscopy and a Simple, Two‐compartment Pedotransfer Function

Abstract

Core Ideas The two‐compartment pedotransfer function successfully predicted soil particle density. Vis–NIR showed slightly poorer performance than the two‐compartment function for predicting soil ρd. Spectroscopy or OM based pedotransfer models gave better estimates of ρd when a wide range in soil OM data was used. The average particle density (ρd) is a fundamental soil property, used for calculating the total porosity. Traditional ρd measurement by pycnometer method is tedious and time‐consuming. In this study, visible–near‐infrared (vis–NIR) spectroscopy and a simple two‐compartment linear and curvilinear pedotransfer function only requiring knowledge of soil organic matter content (OM) were tested and compared as alternative, indirect, rapid, and cost‐effective methods. Soil ρd was measured by water pycnometer on 179 soils representing a wide range of OM (0.002–0.767 kg kg−1), whereas soil spectra were measured on air‐dry samples by vis–NIR spectroscopy. The ρd models were developed using partial least squares regression with leave‐one‐out‐cross‐validation using vis–NIR spectral data, and a simple two‐compartment pedotransfer function, ρd = A(OM) + B(1 − OM) using the OM content. Predictive abilities of these two methods were tested using three different datasets: (i) minerals soils (OM < 0.1 kg kg−1), (ii) organic soils (OM > 0.1 kg kg−1), and (iii) all soils. Calibrating the two‐compartment pedotransfer function for the entire dataset gave expected values for the individual particle densities of OM (A = 1.244 g cm−3) and mineral particles (B = 2.615 g cm−3). The vis–NIR spectroscopy model successfully predicted soil ρd for the entire dataset (R2 = 0.87, RMSECV = 0.10 g cm−3), with a poorer performance than the two‐compartment linear model (R2 = 0.96, RMSE = 0.06 g cm−3). Using only the mineral soils data did not suffice to obtain realistic and accurate vis–NIR spectroscopy (R2 = 0.62, RMSECV = 0.02 g cm−3) or OM based (R2 = 0.80, RMSE = 0.02 g cm−3) models for ρd, illustrating the importance of the wide range of OM content considered in the present study.