Synthesized use of VisNIR DRS and PXRF for soil characterization: Total carbon and total nitrogen

• PXRF elemental concentrations and VisNIR spectra were correlated to soil C and N.
• Synthesized (PXRF + VisNIR) models were more effective than singular models.
• Random forest and penalized spline regression were effective statistical approaches.

Soil total carbon (TC) and total nitrogen (TN) both play critical roles in soil health and ecosystem dynamics. The former is involved in soil structural formation, atmospheric carbon sequestration, and improved soil tilth while the latter is a plant essential element which is often deficient in agronomic production systems. Traditionally, both TC and TN were limited to laboratory determination using techniques such as wet or dry combustion, ion sensing electrodes, loss on ignition, or via chemical assays. These techniques, while generally accurate, require extensive soil sampling, laboratory analysis, and are inherently destructive to the sample analyzed. An approach which could quantify both TC and TN in situ would result in considerable time and cost savings and provide the analyst with the ability to capture more data for a given field of interest. Portable x-ray fluorescence (PXRF) and visible near infrared (VisNIR) diffuse reflectance spectrometry were used to scan 675 soil samples in a laboratory with diverse physicochemical properties from three states of the USA, then compared via random forest (RF) regression and penalized spline regression (PSR) to TC and TN data obtained through traditional laboratory analysis (Dumas method high temperature combustion). Results clearly demonstrated that merging the PXRF and VisNIR datasets improved the power of predictive models by improving the residual prediction deviation (RPD) and R2 statistics. Using synthesized (PXRF + VisNIR) models, independent validation data produced quality predictive statistics for soil TC (RPD = 2.90; R2 = 0.88 via PSR) and TN (RPD = 2.99; R2 = 0.89 via RF). Both proximal sensing techniques were also used to independently predict TC and TN, with results less robust than the synthesized approach. The general order of optimal prediction can be summarized as PXRF + VisNIR > VisNIR > PXRF. In conclusion, the use of synthesized proximal data from PXRF and VisNIR was shown to be a solid, stable predictor of soil TC and TN with widespread agronomic and environmental science applications.