Tracing tropical soil parent material analysis via portable X-ray fluorescence (pXRF) spectrometry in Brazilian Cerrado.

Abstract Parent material (PM) type is crucial for understanding the distribution of soils across the landscape. However, such information is not available at a detailed scale in Brazil. Thus, portable X-ray fluorescence (pXRF) spectrometry can aid in PM characterization by measuring elemental concentrations. This work focused on mapping soil PM (specifically variations of phyllite) using pXRF data and evaluating which soil horizon (A, B, or C) provides optimal PM identification in the Brazilian Cerrado. A total of 120 soil samples were collected from A, B, and C horizons across the study area as well as associated PMs; all were subjected to pXRF analysis. Artificial neural network, support vector machine, and random forest were used to model and predict PMs through pXRF data to the entire area. The nine maps (3 soil horizons data × 3 algorithms) generated for PM prediction were validated through overall accuracy, Kappa coefficient, producer's, and user's accuracy. The most accurate PM maps were obtained by using C horizon information (overall accuracy of 0.87 and Kappa coefficient of 0.79) via support vector machine algorithm. Land use dramatically influenced the results. In sum, pXRF data can be successfully used to predict soil PMs by robust algorithms. Specifically, V, Ni, Sr, and Pb were optimal for predicting PM regardless of land use. Highlights • Soil parent material can be accurately predicted via pXRF analysis of soils. • First time detecting parent material variability in area with unique parent rock. • Land use affects PM predictions, with less effect when using soil C horizon data. • V, Ni, Sr, and Pb produced good PM predictions due to their high stability in soils. • SVM and RF algorithms outperformed ANN in most predictions. [ABSTRACT FROM AUTHOR]