A preliminary study on LEP-AES full-wavelength prediction of soil properties

Rapid soil diagnosis is essential for data-driven and sustainable agricultural management. However, conventional laboratory analyses are time-consuming and costly. This study evaluates liquid electrode plasma atomic emission spectrometry (LEP-AES) as a labor-saving, cost-effective approach for predicting multiple soil properties and compares its performance with the results reported for inductively coupled plasma atomic emission spectroscopy (ICP-AES). Owing to its portability and simple analytical setup, LEP-AES is particularly relevant in resource-limited regions.Full-spectrum LEP-AES data from soils extracted using 0.5 M HCl were combined with machine-learning models, including random forest and extreme gradient boosting, to predict multiple soil properties, while reference values were determined using standard analytical methods. Parameters associated with elements exhibiting distinct LEP emission lines, particularly exchangeable calcium and potassium, were predicted more accurately than exchangeable magnesium and sodium. In contrast, exchangeable aluminum and available phosphorus were consistently difficult to predict owing to their complex emission behaviors, weak far-UV signals, and LEP-specific limitations.Despite the lack of characteristic emission wavelengths, several properties have been indirectly predicted with test R² > 0.50, including the total carbon, total nitrogen, pH (KCl), cation exchange capacity, effective cation exchange capacity, and clay content. Although LEP-based predictions were generally less accurate than ICP-AES models trained on larger datasets, consistent mechanistic hierarchies were preserved (Ca ≈ K > Mg > Al > Na; pH (KCl) > pH (H₂O); effective cation exchange capacity > cation exchange capacity). Overall, LEP-AES combined with machine learning shows potential as a portable, affordable tool for rapid soil diagnosis.