IRSAC-corrected CF-LIBS validated by ICP-OES for soil classification using machine learning

Abstract

Laser-Induced Breakdown Spectroscopy (LIBS) combined with plasma diagnostics and machine learning was investigated as a physics-guided laboratory framework for crop-specific soil characterization. Plasma excitation conditions were evaluated using Boltzmann and Stark analyses, yielding excitation temperatures in the range of 7476–12 163 K and electron number densities on the order of 10¹⁷ cm⁻³, satisfying the McWhirter criterion for local thermodynamic equilibrium. To address optical-thickness effects associated with strong resonance transitions, internal-reference self-absorption correction (IRSAC) was systematically incorporated into the LIBS workflow, restoring physically consistent line-intensity ratios and stabilizing plasma parameters. The IRSAC-corrected spectra were subsequently employed for calibration-free LIBS (CF-LIBS) quantification. Elemental concentrations of Ca, Fe, K, and Mg derived from CF-LIBS showed strong agreement with ICP-OES measurements, with most deviations confined to within 10%, confirming the quantitative reliability of the corrected spectra. Chemometric analysis revealed that Ca-, Mg-, Fe-, and K-related features dominated soil discrimination, while supervised machine-learning models achieved classification accuracy exceeding 97% using IRSAC corrected inputs. Overall, the results demonstrate that IRSAC-assisted stabilization of LIBS spectra is essential for reliable quantitative analysis and data-driven soil classification, providing a robust proof-of-concept for precision agriculture-oriented soil screening under controlled laboratory conditions.