Optimization of pXRF instrumentation conditions and multivariate modeling in soil fertility attributes determination

Abstract

The increasing global demand for agricultural products, food, and energy requires a sustainable and secure approach to agricultural production. Soil nutrition is vital for enhancing agricultural productivity, and understanding the spatial variability of fertility attributes is crucial in soil fertility prevention. Traditional soil analysis methods are time-consuming, costly, and generate waste. Spectral-analytical techniques, such as energy dispersive X-ray fluorescence (EDXRF) combined with multivariate analysis, offer a faster, cost-effective, and environmentally friendly alternative. This study aimed to optimize measurement conditions for predicting soil fertility attributes using a commercial routine from a portable EDXRF (pXRF) equipment and compare its performance with benchtop EDXRF using Partial Least Squares Regression (PLS) modeling. The data fusion from different measurement conditions using low- and mid-level fusion approaches was explored to evaluate the models' synergy. 364 soil samples at three different depths were used. The study area covers 12 ha of Red Latosol and Red Nitosol. Ten fertility attributes were evaluated: exchangeable macronutrients (Ca2+, Mg2+, and K+), available phosphorus (PM), pH, potential acidity (H + Al), soil organic carbon (SOC), cation exchange capacity (CEC), sum of exchange bases (SB) and base saturation percentage (BSP). The results indicated that the measurement of the three experimental conditions tested is unnecessary, and the data fusion does not significantly improve the model's performance. The use of 15 kV and 12.15 μA without filters provided the best performance results for all attributes simultaneously. By the established RPD ranges, the SOC, CEC, SB, and Ca2+ models were considered very good for quantitative prediction (RPD >2.0), while the K+ and Mg2+ models were considered good for quantitative predictions (1.8 < RPD < 2.0). The PM, pH, BSP, and H + Al models provided fair predictions, which may be used for evaluation and correlation (RPD <1.8). Furthermore, the pXRF optimized condition provided results equivalent to those modeled with benchtop EDXRF data. These results highlight the potential of pXRF as an economical, rapid, and efficient alternative approach to the conventional method, and may provide valuable insight for decision-making related to the use of fertilizers and other soil correction products.