A correlation-based LIBS strategy for quantitative analysis of soil nutrients in field soil samples

• 214 field soil samples were analyzed by lab wet chemistry and LIBS for ten indicators • Wavelet denoising improved spectral stability and predictive performance • Variable selection performance depended on specific soil indicators • VIP analysis elucidated band contribution differences across indicators • Seven of ten indicators achieved reliable predictive performance To improve the accuracy and applicability of laser-induced breakdown spectroscopy (LIBS) for rapid quantitative analysis of soil nutrients under complex matrix effects and spectral redundancy, 214 field soil samples were analyzed. Ten soil indicators were determined using standard wet chemistry methods, including representative direct indicators (Ca, Na, Mg, and total nitrogen, TN) and indirect indicators (soil organic matter, SOM; available potassium, AK; and pH). Pearson correlation analysis was employed to investigate relationships among soil indicators and evaluate the feasibility of predicting indirect indicators. LIBS spectra were acquired from pelletized samples, and the effects of spectral preprocessing, variable selection, and multivariate modeling on prediction performance were systematically assessed. Direct indicators associated with characteristic emission lines (e.g., TN, Ca, and Na) generally achieved good predictive performance (RPD > 1.5). In contrast, the prediction of indirect indicators depended strongly on their correlations with direct indicators. For example, SOM showed a strong correlation with TN (r = 0.89) and achieved satisfactory prediction accuracy (RPD = 1.56), whereas pH and available phosphorus (AP), with weaker correlations, exhibited limited predictive performance (RPD < 1.4). VIP analysis further revealed the contributions of spectral bands to different indicators. A comparative analysis of LIBS-based soil quantitative studies was also conducted with respect to sample sources, spectral acquisition conditions, and modeling strategies. Importantly, this study demonstrates that the predictability of indirect soil indicators in LIBS is largely governed by their correlations with direct indicators rather than independent spectral responses, providing additional insight into LIBS-based soil analysis.