Estimation of the Fe and Cu Contents of the Surface Water in the Ebinur Lake Basin Based on LIBS and a Machine Learning Algorithm

Abstract

Traditional technology for detecting heavy metals in water is time consuming and difficult and thus is not suitable for quantitative detection of large samples. Laser-induced breakdown spectroscopy (LIBS) can identify multi-state (such as solid, liquid, and gas) substances simultaneously, rapidly and remotely. In this study, water samples were collected from the Ebinur Lake Basin. The water samples were subjected to LIBS to extract the characteristic peaks of iron (Fe) and copper (Cu). Most of the quantitative analysis of LIBS rarely models and estimates the heavy metal contents in natural environments and cannot quickly determine the heavy metals in field water samples. This study creatively uses the Fe and Cu contents in water samples and the characteristics of their spectral curves in LIBS for regression modelling analysis and estimates their contents in an unknown water body by using LIBS technology and a machine learning algorithm, thus improving the detection rate. The results are as follows: (1) The Cu content of the Ebinur Lake Basin is generally higher than the Fe content, the highest Fe and Cu contents found within the basin are in the Ebinur Lake watershed, and the lowest are in the Jing River. (2) A number of peaks from each sample were found of the LIBS curve. The characteristic analysis lines of Fe and Cu were finally determined according to the intensities of the Fe and Cu characteristic lines, transition probabilities and high signal-to-background ratio (S/B). Their wavelengths were 396.3 and 324.7 nm, respectively. (3) The relative percent deviation (RPD) of the Fe content back-propagation (BP) network estimation model is 0.23, and the prediction ability is poor, so it is impossible to accurately predict the Fe content of samples. In the estimation model of BP network of Cu, the coefficient of determination (R2) is 0.8, the root mean squared error (RMSE) is 0.1, and the RPD is 1.79. This result indicates that the BP estimation model of Cu content has good accuracy and strong predictive ability and can accurately predict the Cu content in a sample. In summary, estimation based on LIBS improved the accuracy and efficiency of Fe and Cu content detection in water and provided new ideas and methods for the accurate estimation of Fe and Cu contents in water.

Remotely Sensed Estimation of Total Iron Content in Soil With Harmonic Analysis and BP Neural Network

“…Moreover, compared with the organic matter, water content, and other constant elements, total iron content are obviously different. The main reason is that the low TICS results in a weak spectral signal, and the performance of the inversion models using variables obtained by conventional methods is poor [31].…”

Determination of phosphorus in water using iron hydroxide assisted laser-induced breakdown spectroscopy

“…Laser-induced breakdown spectroscopy (LIBS) is a technique for the rapid determination of multiple elements 13 and is widely used in agriculture, 14 industry 15 and space exploration. 16 LIBS is commonly employed for the direct detection of nutrients such as calcium, 17 copper 18 and aluminum 19 in water systems. However, effects related to liquid sputtering and plasma quenching result in poor sensitivity and accuracy in the case where LIBS is used to analyze water specimens directly.…”

Possibility of using machine learning methods to reconstruct solid body parameters during laser-induced desorption analysis

“…Currently, to predict the behavior of objects with nonlinear dynamics, machine learning methods are increasingly used, which allow flexible approximation of complex functional dependencies. With regard to the problem of the solid surfaces analysis by means of laser techniques, machine learning methods have recently been used to interpret LIBS diagnostic data arising in geochemical problems [37][38][39], medical applications [40,41] and in materials science research [42]. At the same time, there are presently no works, in which machine learning methods were employed to recover the parameters of a solid body by means of the LID-QMS diagnostic.…”