Comparison of chemometrics method on K detection in soil using laser induced breakdown spectroscopy

Abstract

Potassium detection in the soil is of significant importance for agricultural industry. In this paper, chemometrics methods of artificial neural networks (ANN) and partial least squares (PLS) were comparatively used to detect K in the soil with laser induced breakdown spectroscopy (LIBS). In total, 12 certified reference soils and 17 simulated soil samples with the K concentration of 0.1~3.3% were prepared. LIBS spectra at the wavelength of 723.62~808.24 nm were collected, and then analyzed with ANN and PLS method. The PLS model presented the result of R2 val=0.92 and RMSEV=0.26, the ANN model presented the result of R2 val=0.82 and RMSEV=0.40. ANN model showed under-fitting and the PLS model performed a better RPD than that of ANN. This demonstrated that the linear PLS model is capable to determinate K concentration in the soil using LIBS.