Multi-criteria screening of acidic soils by energy-dispersive X-ray fluorescence and random forest-based pattern recognition

• Published in: Microchemical journal 2024-12, Vol.207, p.111932, Article 111932
• Main Authors: Lucena, Pedro G.C., Aquino, Ramon V.S., Sousa, Juliet E.S., Souza Júnior, Valdomiro S., Pacheco, José G.A., Pereira, Claudete F.
• Format: Article
• Language: English
• Subjects: Classification; Clustering; Elemental characterization; Machine learning; Soil acidity
• ISSN: 0026-265X
• DOI: 10.1016/j.microc.2024.111932

[Display omitted] •Elemental characterization of soils from humid and semi-arid tropical regions.•Classification based on pH, CEC, and BSP thresholds for soil analysis.•Comparison of metaheuristic algorithms for selecting relevant variables.•Simulated annealing models outperformed genetic algorithm models.•Random Forest with simulated annealing achieves over 94 % classification accuracy. Soil acidity is an issue often found in humid tropical areas, characterized by pedogenetically evolved soils with low pH, base saturation percentage (BSP), and cation exchange capacity (CEC), impacting its productivity. Assessing these soil parameters (pH, CEC, and BSP) across large areas is expensive and time-consuming. Energy-dispersive X-ray Fluorescence (EDXRF) together with chemometrics show promise for acidity analysis but requires expertise to interpret numeric results. This study proposes a screening method for analysis of acidic soils using EDXRF data and Random Forest-based pattern recognition. Distinct horizons from 15 soil groups from humid and semi-arid tropical environments in Pernambuco state, Northeast Brazil, were analyzed using a benchtop EDXRF instrument to acquire the elemental information. Chemometric analyses were performed employing R software and autoscaled elemental emission line peak intensities. Classification models were built in Weighted Random Forest (WRF) to classify samples based on pH (