LAUSR

Abstract In recent years many efforts have been made to overcome the conventional soil fertility analysis limitations contributing to the improvement of precision agriculture. Energy dispersive X-ray fluorescence (EDXRF) is one of the methodologies proposed for the soil parameters analysis in a fast, cost effective and environmentally friendly way. However, most of contributions reported in the literature have been limited to the use of elemental data determined by EDXRF, which may cause loss of useful information contained in their spectra. Therefore, this study evaluates the use of EDXRF spectral data, under two measurement conditions, combined with partial least square regression (PLSR) in order to quantify cation exchange capacity (CEC), sum of exchangeable bases (SB) and base saturation percentage (BSP) in agricultural soils. Multivariate calibration models using full spectrum and selection of the most important EDXRF variables were developed. The variable selection achieved the best prediction power for all analyzed parameters. From randomization test at 95% of confidence level it was concluded that the accuracy of the PLSR models with Na-Sc and Ti-U spectral data (condition measurement) were equivalent. The analysis of the figures of merit (linearity, accuracy, systematic errors, sensitivity and limit of detections) indicated that proposed models are suitable for modeling soil fertility indicators with EDXRF spectral data. Furthermore, it was possible to identify the spectral regions that most contributed for the models. Thus, this study contributes toward a better understanding of the nature of EDXRF spectral data in the modeling of SB, CEC and BSP in soils by PLSR.