Technical ArticleMultivariate Analysis of Laser-Induced Breakdown Spectroscopy Spectra of Soil SamplesYang, Ningfang1; Eash, Neal S.1; Lee, Jaehoon1; Martin, Madhavi Z.2; Zhang, Yong-Seon3; Walker, Forbes R.1; Yang, Jae E.4Author Information 1Biosystems Engineering and Soil Science, University of Tennessee, 2506 E.J. Chapman Dr, Knoxville, TN 37996. Dr. Neal S. Eash is corresponding author. E-mail: [email protected]2Environmental Sciences Division Oak Ridge National Laboratory, Oak Ridge, TN 37831. 3Soil and Fertilizer Management Division, National Academy of Agricultural Science, Suwon, South Korea. 4Department of Biological Environment, Kangwon National University, Chunchon, South Korea. Received March 23, 2010. Accepted for publication August 2, 2010. Soil Science: September 2010 - Volume 175 - Issue 9 - p 447-452 doi: 10.1097/SS.0b013e3181f516ea Buy Metrics Abstract Laser-induced breakdown spectroscopy (LIBS) is a rapid quantitative analytical technique that can be used to determine the elemental composition of numerous sample matrices, and it has been successfully applied in many types of samples. However, for chemically and physically complex soil samples, its quantitative analytical ability is controversial. Multivariate analytical techniques have great potential for analyzing the complex LIBS spectra. To demonstrate the feasibility of LIBS as an alternative technique to quantitatively analyze soil samples, the univariate and the partial least square (PLS) techniques are used to analyze the LIBS spectra of 12 soil samples and to build calibration models predicting Cu and Zn concentrations. The results show that PLS can significantly improve the analytical results compared with the univariate technique. The normalized root mean square error (NRMSE) and r2 of the univariate models are 16.60% and 0.71 in calibration and 18.80% and 0.62 in prediction for Cu and 18.97% and 0.62 in calibration and 22.81% and 0.45 in prediction for Zn. For the PLS models using the spectral range 300 to 350 nm, the NRMSE and r2 are 1.94% and 0.99 for both Cu and Zn in calibration and 7.90% and 0.94 for Cu and 8.14% and 0.94 for Zn in prediction, respectively. Compared with the univariate technique, PLS improves the NRMSE 87.53% and 87.78% in calibration and 44.47% and 53.44% in prediction for Cu and Zn, respectively. The results indicate that PLS can improve the quantitative analytical ability of LIBS for soil sample analysis. © 2010 Lippincott Williams & Wilkins, Inc.