Soil organic horizon characterization via advanced proximal sensors

The organic fraction of soils is critically important to soil health and optimal ecosystem functioning. Traditional analysis of soil organic horizons (O horizons) has been dependent upon laboratory-based instrumentation. Simultaneously, the use of proximal sensors such as portable X-ray fluorescence (PXRF) spectrometry along with visible near infrared diffuse reflectance spectroscopy (VisNIR DRS) has gained popularity for providing rapidly acquired spectral and elemental data useful for soil physicochemical property quantification. However, PXRF and VisNIR DRS have mostly been applied to the assessment of mineral soils. This preliminary study evaluated 136 organic laden soil samples (most aptly described as upland, non-saturated O horizons) using both laboratory based instrumentation (CN analyzer) and proximal sensors to evaluate total carbon (TC) and total nitrogen (TN). Results revealed that combining model outcomes using model fusion improved TC and TN prediction accuracies relative to using an individual instrument (PXRF or VisNIR DRS) or model averaging with improvements in root mean square error (RMSE) on the order of 10–47% and 10–67% for TC and TN, respectively. Partial least squares+random forest (PLS+RF) approaches emerged as the best model for predicting both TC and TN in organic laden soil samples. These results suggest that the strong predictive applications of proximal sensors extensively documented on mineral soils, may show similar promise for determination of a wide number of physicochemical properties on organic soil matrices, yet further exploration with a larger and more diverse dataset is recommended.