Elemental composition and moisture prediction in manure by portable X-ray fluorescence spectroscopy using random forest regression

Abstract

Manure elemental composition determination is essential to develop farm nutrient budgets and assess environmental risk. Portable X-ray fluorescence (PXRF) spectrometers could facilitate hazardous waste-free, rapid, and cost-effective elemental concentration determinations. However, sample moisture is a problem for elemental concentration determination by X-ray methods. The objective of this study was to quantify the effect of sample moisture content, predict moisture content, and correct for moisture effect on elemental concentration determinations in livestock manure. Oven-dried manure samples (n = 40) were ground and adjusted to five moisture ranges of (w/w moisture) <10%, 10-20%, 20-30%, 40-50%, and 60-70%. Samples were scanned by PXRF for 180 s using a vacuum (<1,333 Pa) and without a filter. The presence of moisture negatively affected elemental determination in manure samples. Calibrations (n = 200) were prepared using random forest regression with detector channel counts as independent variables. A three-step validation was performed using all the data, random cross-validation and external validation. The back end of the spectrum (14-15 keV) had strong predictive power (r² = .98) for moisture content. The random forest approach increased r² between PXRF and wet chemical methods from <.66 to >.90 for P, K, and Mg and from .78 to .98 for Fe, compared with linear, nonlinear, and Lucas-Tooth and Price equations. These results indicated that elemental concentration can accurately be measured in dried and moist manure samples using PXRF and expands the potential applications of PXRF to in situ elemental determinations for agricultural and environmental samples.