Prediction of soil attributes via pXRF spectrometry, magnetic susceptibility, and terrain attributes in a highly heterogeneous tropical area

Abstract

Digital elevation models (DEM) and their derived variables, terrain attributes (TA), are commonly used in soil mapping. The use of proximal sensors, such as portable X-ray fluorescence spectrometer (pXRF) and susceptibilimeter, which determines magnetic susceptibility (MS), provides additional information that has improved the results obtained using only TAs. This work is composed of two chapters, whose studies were conducted at the Palmital Experimental Farm, belonging to the Federal University of Lavras (UFLA). The chapters are related to the use of proximal sensors in conjunction with TA for the prediction of physical and chemical attributes of soils. The first chapter contemplates the use of two proximal sensors, pXRF and MS, together with TA for the prediction of clay, silt, and sand contents through the random forest algorithm. The second chapter discusses the use of pXRF and MS in conjunction with TA in predicting available contents of B, Cu, Fe, Mn, and Zn. The maps were generated for the Palmital farm and validated for each predicted attribute, comparing the efficiency of each model. For the prediction of clay, silt, and sand, all models used the information acquired by pXRF in the final models. On the other hand, for the prediction of B and Zn, only the TA information was sufficient to achieve satisfactory R2 values. Clay and sand showed moderate accuracy, while silt showed low accuracy. For the prediction of chemical attributes, Cu, Fe, Mn, and Zn presented high to moderate accuracy. However, B reached low accuracy. This shows that pXRF is a powerful tool to assist in the accurate prediction of some soil attributes in a punctual and spatial way, contributing to the digital soil mapping.