Diffuse reflectance Fourier transformed mid-infrared (DRIFT-MIR) spectroscopy can predict many soil properties but extractable nutrients are often predicted poorly. This paper tested the combined DRIFT-MIR and total X-ray fluorescence (TXRF) spectroscopy analysis for prediction of soil properties related to soil fertility. A total of 700 soil samples from 44 stratified randomly located 100-km2 sentinel sites distributed across Sub-Saharan Africa (SSA) were analyzed for physicochemical composition using conventional reference methods, and compared with MIR and TXRF spectra using Random Forests (RF) regression algorithm and an internal out-of-bag (OOB) validation. Mid-infrared spectra resulted in good prediction models (R2 > 0.80) for organic C and total N, Mehlich-3 Ca and Al, and pH. Moderately predicted (R2 > 0.60) were extractable Mg, P sorption index, sand, silt, and clay. Models were less satisfactory (R2 < 0.60) for Mehlich-3 extractable K, Mn, Fe, Cu, B, Zn, P, S, and Na, exchangeable acidity and electrical conductivity (Ecd). Inclusion of total element concentration data from TXRF analysis in the MIR RF models significantly reduced root mean square error of prediction by 70% for Ecd, 66% for Mehlich-3 Na, 61% for Mehlich-3 S, and 50% for Mehlich-3 B due to detection by TXRF of some saline soils that were not well predicted by MIR. Overall, both methods predicted soil properties that relate to nutrient buffering capacity, including some exchangeable bases, pH, P sorption capacity, clay content, and organic matter content, and fingerprint basic soil mineralogy. Thus, further research should test whether MIR and TXRF fingerprinting could better predict soil nutrient supply capacity, as determined by crop nutrient uptake potential, than conventional soil P, K and micronutrient tests.
Read full abstract