Open Access
Add to BibSonomy
Abstract
Cadmium (Cd) can migrate in the soil and is readily absorbed by crops. High Cd accumulated in grains poses a huge threat to human health by inhibiting the function of the kidney system. Thus, it is crucial to reveal the content of soil Cd in vertical-depth series using a fast, real-time, and reliable method. For this purpose, laser-induced breakdown spectroscopy (LIBS) combined with multivariate chemometrics was developed to analyze Cd content in the soil with vertical-depth series. Soil samples spiked with different levels of Cd were prepared, and LIBS spectra were obtained by single-pulse LIBS (SP-LIBS) and collinear double-pulse LIBS (CDP-LIBS) with wavelengths of 532 nm and 1064 nm. With appropriate parameters, CDP-LIBS showed better performance in detecting Cd than SP-LIBS. Partial least squares regression (PLSR), genetic algorithm (GA)-optimized back propagation artificial neural network (BP-ANN), and particle swarm optimization (PSO)-optimized least squares-support vector machine (LS-SVM) were tested for quantitative analysis of the spectra after median absolute deviation (MAD), multiple scattering correction (MSC), wavelet transform (WT), spectral averaging, and normalization. PSO-optimized LS-SVM yielded an ideal result, with a coefficient of determination (R2, 0.999) and root mean square error (RMSE, 0.359 mg/Kg) in the prediction dataset. Finally, CDP-LIBS coupled with PSO-optimized LS-SVM was employed to analyze soil Cd content in vertical-depth series to reveal the migration pattern of Cd. Our results indicated that soil Cd had a significant positive relationship with the inverse of soil depth. However, Cd was mainly concentrated in 0-20 cm and rarely leached below 45 cm in the soil. This study suggests that LIBS and its enhancement techniques provide a reliable method for revealing the content of soil Cd in vertical-depth series.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Cadmium (Cd) is a priority environmental pollutant in soils during industrialization and urbanization [1,2]. Plant growth is supported by soil, hence, heavy metal-contaminated soil may pose a considerable threat to crop production and human health via the food chain. For example, Cd exposure to humans is known to cause prostate and kidney cancers, Itai-Itai disease, etc [3–5]. Heavy metals are usually not distributed homogeneously in soil but tend to show a vertical-depth distribution pattern and can migrate with time and moisture. Recently, the potential public health risks associated with heavy metals, especially Cd, have become a major concern for government and regulatory authorities [6]. Therefore, developing an efficient and accurate method to detect the content and distribution of Cd in soil has been urgently needed.
Many techniques can detect soil Cd, such as atomic absorption spectroscopy (AAS) [7,8], atomic fluorescence spectrometry (AFS) [9,10], inductively coupled plasma mass spectrometry (ICP-MS) [11,12], and inductively coupled plasma atomic emission spectroscopy (ICP-AES) [13,14]. However, they need concentrated acid digestion before the analysis. By contrast, laser-induced breakdown spectroscopy (LIBS) is a cutting-edge technique, owing to the non-cumbersome sample pre-processing, high efficiency, and simplicity [15,16]. As a result, LIBS has been widely applied in many fields, such as aerospace [17,18], environmental monitoring [19,20], archaeology [21], biomedical [22,23], coal mining, and metallurgical [24,25], and agriculture [26,27], etc. Meanwhile, LIBS shows potential in nuclear and isotopic analytical applications. Such as Hull, et al. [28] reported the development of a system comprising laser ablation-tuneable diode laser absorption spectroscopy and LIBS, which maintains the benefits of conventional LIBS and enables isotopic analysis of lithium and uranium. Doucet, et al. [29] described the determination of isotope ratios using LIBS for partially resolved uranium-235/uranium-238 and hydrogen/deuterium isotope shift lines. Notably, Menegatti, et al. [30] achieved LODs of 1 mg/kg for Cd evaluation using a single-pulse LIBS (SP-LIBS) under a controlled atmosphere. However, under natural atmospheric conditions without specific control, SP-LIBS shows a deficiency in reproducibility, sensitivity, and accuracy when monitoring heavy metals in soil [16–32], due to its unstable laser energy and single laser wavelength.
Compared with SP-LIBS, collinear double-pulse LIBS (CDP-LIBS) shows greater flexibility in wavelength selection, pulse spacing, and pulse width [33–36]. Guanyu Chen, et al. [33] found that CDP-LIBS exhibited better analytical performance than SP-LIBS, when detecting Cu, Ni, and Pb in soil. Similarly, Gustavo Nicolodelli, et al. [37] reported that the CDP-LIBS system improved the detection accuracy by increasing the intensity of elemental emission lines about 5-fold. Spectral interferences and matrix effects may significantly affect the quantitative analysis of heavy metals, particularly for the “trace” metals in soil [38]. Fortunately, multivariate chemometrics could be employed to reduce spectral interferences and matrix effects in the soil, because it can perform nonlinear regression [39]. Xiang, et al. [40] found that the least squares-support vector machine (LS-SVM) and back propagation-artificial neural network (BP-ANN) models had much better performance than the multiple linear regression (MLR) and partial least squares regression (PLSR) models when predicting lead (Pb) and Cd contents in soil. Similarly, Mao-Gang Li, et al. [41] developed a random forest (RF) model for rapid quantitative analysis of cuprum (Cu), chromium (Cr), nickel (Ni), and Pb in soil. Tao Wang, et al. [42] also showed that appropriate multivariate chemometrics for LIBS data was essential when quantifying heavy metals in soil. However, the selection of weights and thresholds usually relies on practical experience, which may diminish the accuracy of the model. Therefore, CDP-LIBS combined with weight and threshold optimization of the multivariate chemometrics may provide a reliable and robust way to determine the heavy metal content in the soil.
However, detecting heavy metals in the soil at vertical depth series using CDP-LIBS coupled with multivariate chemometrics was rarely reported. This study aims to reveal the vertical-depth distribution of soil Cd by CDP-LIBS combined with PLSR, genetic algorithm (GA)-optimized BP-ANN, and particle swarm optimization (PSO)-optimized LS-SVM. This study is expected to provide a novel and reliable method to monitor heavy metals in soil and offer new insights into remediation and management of heavy metal contamination in soil.
2. Materials and methods
2.1 Sample preparation
The soil was acquired from a wheat-growing experiment site of Northwest A & F University, Yangling District, Shaanxi Province, China (108°074'E-102°076'E, 34°291’N-25°293’N). The soil is typical sandy clay loam (soil description: USDA). CdCl2 of 99% purity was purchased from Penny Chemical Reagent Co., Ltd. (Zhengzhou, China). According to the U.S. Environmental Protection Agency (EPA) general soil screening levels, the industrial screening levels for soil Cd contamination is 800 mg/kg (calculated based on non-carcinogenic hazard entropy). Thus, the CdCl2 solutions in the range of 60 - 800 mg/kg were added to the collected soil and well stirred to make an anthropogenic Cd-contaminated soil for simulating high levels of Cd soil contamination that may be present at sites of industrial activity, such as mining, electroplating, pigment production, plastic stabilizer manufacturing, and electronic waste dumps. The soil was dissolved, air-dried, ground, and sieved through a 100-mesh sieve (Sieve hole size: 0.150 mm). Finally, Round cake shaped soil samples with a mass of 4 g were prepared using a soil compactor at a pressure of 25 MPa. Figure 1 shows the prepared soil samples.
Fig. 1. The prepared soil samples contaminated with Cd. (a) side elevation of the soil sample; (b) front elevation of the soil sample.
Highly concentrated CdCl2 solutions in the range of 60 - 800 mg/kg were added to simulate high levels of Cd soil contamination that may be present at sites of industrial activity closely associated with hazardous chemical places. 60 soil samples with 15 concentration gradients were provided for the quantitative analysis in Table 1. The laser hits the sample in 25 different positions. 3 LIBS spectra were acquired for each position. Finally, 4500 spectra were extracted for the quantitative analysis.
Table 1. The Cd concentrations in the soil samples.
The topsoil (0-30 cm) layer, composed of various fine-grained minerals, is believed to be the vehicle for crop growth [43,44]. The 30-50 cm soil layer plays a minor role in supporting crop growth but hosts much fertilizer and water leached downward from the topsoil [45]. As a result, the 0-50 cm soil layer receives the most tillage, irrigation, and fertilization [46]. In this study, 3 polyvinyl chloride (PVC) pipes, with a depth × inner diameter × pipe-wall thickness of 55 cm × 55 mm × 4 mm, was designed and employed in our vertical-depth migration experiments.
Three PVC pipes without closure at both ends were inserted into the wheat field at 50 cm depth to acquire a natural soil column 50 cm deep. To simulate the vertical-depth migration of soil Cd in smelters and e-waste piles, the CdCl2 solution of 623.0 mg/Kg was supplied to all soil columns through the reserved 5-cm PVC pipe. The soil columns were incubated under 20 ± 5 °C and 26 ± 4% relative humidity, and ∼40 mL purified water was sprayed every 2 days to simulate sufficient rainfall and irrigation. 3 PVC pipes were total added with purified water ∼200 mL (10 days), ∼400 mL (20 days), and ∼600 mL (30 days) respectively. In addition, the bottom end of the PVC pipes were wrapped with sealing films to avoid water leaching. The vertical-depth migration pattern was studied by detecting the Cd content in soil columns with vertical-depth series. 4 samples were collected from each 5 cm depth soil layer on days 10, 20, and 30 i.e., 4 × 10 (from each tube) × 3 (tubes) = 120 soil samples. The laser hits the sample in 25 different positions. 3 LIBS spectra were acquired for each position, so 120 (number of soil samples) × 25 (number of positions hit for each sample) × 3 (number of spectra acquired for each position) = 9000 (total number of spectra) were acquired to analyze the distribution of Cd content at vertical depth series. The experimental design for studying Cd migration in the soil is depicted in Fig. 2.
2.2 Experimental procedures
The LIBS system used in this study is shown in Fig. 3. The laser was generated by two Nd: YAG lasers with different emission wavelengths. TINY-200L operating at 1064 nm and Nasor-900 operating at 532 nm served as the laser 1 and laser 2, respectively. The spectrometer system consisted of a medium-echo spectrometer ARYELLE 200 and an intensified charge-coupled device (ICCD) camera. The two laser pulses were combined or output separately through dichroic mirrors (1064 nm and 532 nm) in the laser optical path, and further focused on the soil sample through the lens in the optical acquisition path. The lens with an anti-laser reflection coating was attached to the end of the fiber, enabling the efficient collection of the emitted plasma. The soil sample was placed on an X-Y-Z stage controlled by a miniature calculator, which allowed the laser to focus quickly and accurately on the targeted sample surface. “LTB Sophi” software (Lasertechnik Berlin GmbH, Germany) was used to analyze LIBS spectra.
The information of the instrument is detailed in Table 2. The LIBS experiments were performed under standard atmospheric pressure, a room temperature of 20 °C, and relative humidity of 26%.
Table 2. The specific information of the experimental instrument.
Several parameters of the system were fixed to investigate the influence of delay time (DT) and laser energy (LE) on SP-LIBS, as well as interval time (IT) on CDP-LIBS. Table 3 shows those parameters.
Table 3. The Specific information of the experimental parameters.
2.3 Analytical methods and procedures
During the spectral acquisition process, fluctuation of the soil environment, such as the soil matrix effect, instrument performance, etc., could generate noises and affect the stability of the signal. As a result, median absolute deviation (MAD), multiple scattering correction (MSC), wavelet transform (WT), spectral averaging, and normalization were applied to reduce the noise effects. Those attempts were expected to improve the sensitivity and reliability of the spectra and ensure the accuracy of the quantitative Cd detection model.
MAD was used to remove anomalous spectra with large variations, which are caused by the randomness of plasma jumps [47,48]. Equation (1) and a restraint condition were applied to judge anomalous spectra based on MAD.
Restraint condition: $|{{X_i} - median(X)} |> k \times MAD$
where, ${X_i}$ : the spectral value of the i th spectrum; X : all spectral values of the sample; a : a constant in the data distribution, and a = 1.4826 when the spectrum conforms to a normal distribution; k : represents the coefficient of the degree of rejection, and in general, k = 2.5.
MSC was employed to correct the baseline translation and offset the spectral data with ideal spectra to effectively promote the correlation between the spectra and the data by eliminating the spectral differences due to different scattering levels [49,50]. WT was adopted to remove random noises by eliminating noise signals of high-frequency and retaining the low-frequency signals [51,52]. Spectral averaging and normalization, frequently used in spectral data analysis, could minimize the matrix effects in samples and spectral line fluctuations [53]. The average spectrum of 3 spectra was approximated to be the actual at the position.
PLSR, BP-ANN, and LS-SVM models were tested to quantify Cd because of their advantages in accuracy and efficiency. PLSR, a typical linear regression method, was proven to accurately quantify elemental concentrations in soil combined with LIBS [40,54]. Since BP-ANN can better solve the matrix effect and spectral interference, thus it is widely utilized in soil heavy metal determination [40,55,56]. However, the weights and thresholds of BP-ANN rely on a manual empirical selection. GA is believed to be able to solve the problems posed by traditional search and algorithm optimization, by mimicking the process of natural selection and reproduction [57]. Therefore, it was introduced in this study to optimize the weights and thresholds of BP-ANN. During the optimization, LS-SVM utilized equation constraints instead of inequality constraints in the support vector machine (SVM), which enabled a highly efficient and accurate calculation [58]. However, the kernel function parameter “sig2” and the regularization parameter “gam” of LS-SVM depend on manual experience, which may reduce the calculation efficiency. Moreover, PSO was adopted to optimize the parameter of LS-SVM, since PSO has a significant advantage in fast convergence. The model establishment and optimization were performed in MATLAB R 2019 b (The MathWorks, USA) software.
2.4 Assessment of indicators
Relative standard deviation (RSD) was employed to evaluate the stability of the signal. Equation (2) shows the calculation of RSD. High SBR and RI values mean high signal strength, while low RSD indicates good signal stability.
where, $\sigma$: the standard deviation of the elemental Cd signal when measuring different positions of the soil press sample; $\overline I$: the average of the Cd element signal when measuring different positions of the soil press sample.The coefficient of determination (R2) and root mean square error (RMSE) were utilized to evaluate the performance of the model in our quantitative analyses. Equation (3) shows the calculation of R2, which reflects the accuracy of the quantitative analysis of the model. The reliability of the model is high when R2 is close to 1.
RMSE is the standard deviation of the predicted and actual values, which reflects the prediction error of the model. The model is thought to have good prediction performance when RMSE is close to 0. Equation (4) shows the calculation of RMSE.
where, ${\hat{y}_i}$: the prediction of the i th sample; ${y_i}$: the actual of the i th sample; M : the number of samples.The signal-to-back ratio (SBR) and relative intensity (RI) were adopted to evaluate the response of Cd concentrations to LIBS signals. An optimized calculation of SBR and RI was proposed in this study. Figure 4, Eq. (5) and (6) presents the calculation of SBR and RI.
where, $Are{a_{(signal)}}$: The area of the region enclosed by the black dotted line; $Are{a_{(back)}}$: The area of the region enclosed by the black dotted line; ${l_{back}}$: The intensity of the black line; $AI$: The intensity of the green line.2.5 Systematic research procedures
Figure 5 shows the systematic research process of this study. The details of the research were divided into four steps. First, 60 samples with 15 different concentrations were quantitatively analyzed to construct the model. The characteristic spectral lines of Cd were identified from the atomic spectral database of NIST, which were used for further analysis of the LIBS spectrum. Second, the effects of DT and LE on SP-LIBS and pulse IT on CDP-LIBS were examined. Third, GA-optimized BP-ANN, PSO-optimized LS-SVM, and PLSR were introduced to model the relationship between Cd contents and Cd characteristic spectral lines. Finally, the vertical-depth migration pattern of Cd in soil was revealed by analyzing the soil Cd content in the vertical-depth series.
3. Results and discussion
3.1 LIBS curve and characteristical emission lines of Cd in soil
The LIBS spectra (270-750 nm wavelength) of Cd-contaminated soil are shown in Fig. 6. The various elements in the soil caused multiple spectral lines and heterogeneous distribution of spectral lines in soil samples. As shown in Fig. 6, the emission spectral lines of most metallic elements were mainly distributed in the wavelength range of 270-450 nm, with few elements’ spectral lines in the wavelength range of 450-750 nm.
Fig. 6. LIBS curves of the Cd-contaminated soil samples. I : the atom emits new spectral lines; II : the ion emit new spectral lines.
The spectral lines of Cd in the atomic spectral database from NIST were carefully reviewed, and the main specific spectral line information is presented in Table 4. Combining Table 4 and Fig. 6, it can be found that if the Cd spectral line <350 nm is used for quantification, the accuracy of the analysis will be reduced due to the low intensity of the spectral line; if the Cd spectral line >400 nm is selected for quantification, the accuracy of the analysis will also be reduced due to the interference of the spectral line of large and medium nutrient elements. Therefore, Cd I 361.05 nm is a typical characteristic wavelength with a distinct Cd peak. The accurate wavelength of Cd was 360.82 nm, but the spectrometer’s accuracy and external interferences could be affected. Indeed, the clear Cd peak at 359.7-361.4 nm (including 145 spectral features arising from various elements) was also found in Fig. 6.
Table 4. The main specific spectral line information of Cd.
Where, Aki : the spontaneous transition probability; Ei: the excitation energy of the upper energy level; Ek : the excitation energy of the lower energy level; gi and gk: the upper and lower energy level degeneracy, respectively; I: the atom spectral lines.
3.2 Analysis of the influence parameters
One soil sample was selected to explore the influence of various parameters on LIBS for reducing the impact of differences between samples. In each parameter case, 30 spectra were obtained from 10 random points on the selected sample. Researchers found that DT and LE for SP-LIBS, and pulse IT for DP-LIBS, had significant effects on detecting the elemental content in soil [36,59]. Therefore, it is high important to study the influence of DT and LE on SP-LIBS, and IT on DP-LIBS before the downstream quantitative analysis.
Nasor-900 laser (532 nm) and TINY-200L (1064 nm) were applied to investigate the influence of LE and DT on SP-LIBS, respectively. LE was set to 10-120 mJ (with an increment of 10 mJ), and DT was set to 0.5, 1-11 µs (with an increment of 1 µs). The variations of Cd spectra, SBR, and RI for SP-LIBS are shown in Fig. 7.
Fig. 7. The variations of Cd characteristic spectral lines, SBR and RI for SP-LIBS.(a) Variation of RI and SBR for LE of 532-LIBS; (a1) Variation of Cd characteristic spectral lines for LE of 532-LIBS;(b) Variation of RI and SBR for DT of 532-LIBS; (b1) Variation of Cd characteristic spectral lines for DT of 532-LIBS.(c) Variation of RI and SBR for LE of 1064-LIBS; (c1) Variation of Cd characteristic spectral lines for LE of 1064-LIBS;(d) Variation of RI and SBR for DT of 1064-LIBS; (d1) Variation of Cd characteristic spectral lines for DT of 1064-LIBS.
Lower SBR and RI were observed when LE was < 90 mJ (Figs. 7 a & c), which may be attributed to the under-excitation of Cd plasma with the small laser energy [60]. Our results clearly showed that a suitable LE value is essential for the accurate determination of soil Cd with LIBS. Accordingly, a large LE (>110 mJ) tends to decrease SBR and RI (Figs. 7 a & c), possibly due to the instability of excitation. In addition, when DT was 2 µs, a decrease in SBR and RI could also be caused by the incomplete acquisition of the Cd signal with the spectrometer (Fig. 7 b & d). When DT was > 5 µs (Figs. 7 b & d), RI was relatively constant but SBR showed an obvious decreasing trend. This may be because the spectrometer was not able to efficiently collect Cd signals at high DTs, while the change in SBR may be mainly attributed to dark noises. Nunes, et al. [61] also captured a similar phenomenon when utilizing SP-LIBS to detect Cd in soil.
Nasor-900 laser (532 nm) and TINY-200L (1064 nm) were combined to compose the CDP-LIBS system. The energy of both lasers was 100 mJ, and the DT of 1 µs was utilized to study the influence of IT on DP-LIBS. The variations of Cd spectra, SBR, and RI for DP-LIBS and the comparison between SP-LIBS and DP-LIBS are shown in Fig. 8.
Fig. 8. The variations of Cd characteristic spectral lines, SBR, and RI for DP-LIBS and comparison of DP-LIBS with SP-LIBS. (a) Variation of RI and SBR for IT; (b) Variation of Cd characteristic spectral lines for IT; (c) Comparison of RI, SBR for different LIBS systems; (d) Comparison of Cd characteristic spectral lines for different LIBS systems.
Both RI and SBR were low (RI< 2.05 ×104 264 counts; SBR< 2.07) and nearly constant when IT ≤ 20 µs (Figs. 8(a) & (b)). This occurred because of the incomplete secondary heating of the Cd plasma [62]. When 20 µs < IT ≤ 24 µs, a rapid increase in RI and SBR was observed. This may be due to the heating effect of the Cd plasma by DP, thus the Cd signal is enhanced. It should be noted that >24 µs was too high for IT. Under the high IT conditions, the heated Cd plasma tended to cool quickly before the arrival of the second laser beam. This would lead to insufficient heating of the Cd plasma during the second laser scan. Ultimately, RI raised slowly while SBR appeared to decrease.
As shown in Fig. 8(c) and (d), 532-LIBS was more responsive than 1064-LIBS, owing to the good penetration and high stability of lower wavelength lasers. However, since DP-LIBS allowed a sufficient secondary heating and complete combustion of the Cd plasma, it may be more suitable for soil Cd detection than the SP-LIBS system, particularly when signal stability (RSD) and intensity (RI, SBR) were considered.
3.3 Quantitative analysis of soil Cd contents
The accuracy of vertical-depth migration patterns depends on reliable quantitative analysis models. Therefore, CDP-LIBS was used in the quantitative analysis because it was more reasonable than SP-LIBS. After comparison, the optimal LE, IT, and DT were 100 mJ, 24 µs, and 1 µs, respectively.
MAD, MSC, WT, spectral averaging, and normalization were introduced to reduce the noise in advance. Based on the principle of median deviation, MAD greatly improved the accuracy of the spectral analysis by eliminating 382 anomalous spectra out of 1500 spectra. With MSC, the average spectrum for each Cd concentration was taken as its ideal spectrum. With WT, “Symlets5” was chosen as the wavelet function, and 6-layer wavelet decomposition and reconstruction of the signal were performed, generating a better SBR. The spectra at 357-363 nm after MSC and WD are presented in Fig. 9. The instability and noise could be significantly reduced by MSC and WD.
After spectral pre-processing i.e. MAD, MSC, WD, spectral averaging and normalization, PLSR, BP-ANN, GA, LS-SVM, and PSO were applied to quantitatively analyze soil Cd. Before modeling, the training and prediction datasets were randomly divided into a ratio of 7:3. In the process of building the PLSR, the potential variable was set to 10 considering to extract as much as possible the Cd concentration of the sample and the 145 principal components of the spectral features while ensuring that the correlation between the two principal components was maximized. The University of North Carolina Genetic Algorithm Toolbox and the BP-ANN toolbox in MATLAB were employed to construct GA-optimized BP-ANN models. The 145 spectral features extracted from all the spectra acquired from the samples were used as input to the input layer in the BP-ANN, and the Cd concentrations of the samples used for quantification were used as the output of the output layer in the BP-ANN. Considering the training speed and generalization ability of BP-ANN The neurons in the BP-ANN hidden layer were set to 10. The “tan-sigmoid” function and the “traingd” function were set as the transfer function and the training function of BP-ANN, respectively. The maximum number of iterations, training accuracy, and learning rate of BP-ANN were set to 200, 0.0001, and 0.1, respectively. In constructing the BP-ANN model, the optimal weights and thresholds are searched for using GA to improve the accuracy of the BP-ANN. The genetic algorithm went through initialization, individual evaluation, selection, crossover, and variation operations, and finally found the best weights and thresholds for BP-ANN from the initial population and applied them to BP-ANN. The value of R2 divided by RMSE was used as the fitness function to evaluate the individual merit in the population.The population size of GA was set to 100, and the selection function, crossover function, and variation function table were set to “normGeomSelect”, “arithXover”, and “nonUnifMutation”, respectively. “gam” and “sig2” are the main parameters that affect the accuracy of LS-SVM models, thus PSO was introduced to optimize the LS-SVM parameters. We randomly set the initialized positions and velocities of the particles in the PSO, using the value of R2 divided by RMSE as the objective function to find the optimal parameters. The values of iterations, inertia factor, and population size of PSO were set to 200, 0.9, and 20, respectively. The most reasonable “gam” and “sig2” were retrieved by PSO with 100 and 3.0062, respectively. LS-SVM lab 1.8 toolbox was utilized to construct the LS-SVM model. Likewise, the Cd concentration is used as its output and the extracted spectral features as the input.
The predicted results of the Cd concentration by PSR, GA-optimized BP-ANN, and PSO-optimized LS-SVM are shown in Fig. 10. The PSO-optimized LS-SVM had a satisfactory performance with an RMSE of 0.359 mg/Kg and R2 of 0.999 in the prediction dataset. The prediction performance of the nonlinear model was superior to the linear prediction method, probably because the nonlinear analysis method is more efficient to reduce the impacts of soil matrix effects and spectral interferences.
Fig. 10. The predicted results of Cd concentration by PSR, GA-optimized BP-ANN, and PSO-optimized LS-SVM. (a) The predicted results of Cd concentration by PSR. (b) The predicted results of Cd concentration by GA-optimized BP-ANN. (c) The predicted results of Cd concentration by PSO-optimized LS-SVM.
3.4 Distribution of Cd content in vertical-depth series
Our quantitative analyses illustrated that the PSO-optimized LS-SVM combined with CDP-LIBS provided reliable and robust detection of soil Cd contents. Therefore, PSO-optimized LS-SVM combined with CDP-LIBS was further applied to explore the vertical-depth migration patterns of soil Cd in the vertical-depth series. The vertical-depth distribution of soil Cd in the vertical-depth series is given in Fig. 11. Figure 11 clearly shows that the soil Cd content had a significantly positive relationship with the inverse of the soil depth on Day 30. Cd was found to concentrate in the 0-20 cm soil layer. Although a relatively high Cd content (109.14 mg/Kg < Cd content < 353.24 mg/Kg) was found at 20-40 cm soil depth, the Cd content of the 20-40 cm soil layer was much less than that of the 0-20 cm soil layer. In addition, considering in relation to the one-dimensional convective-dispersion model [63], the soil at 20-40 cm could prevent the leaching of Cd to groundwater. Therefore, detecting a relatively high Cd in the 40-50 cm soil layer is usually hard. These results could explain the sharp decreasing trend of Cd content from < 20 cm to > 40 cm, and the moderate change in the middle 20-40 cm soil layer. Furthermore, along with incubation and irrigation, soil Cd began to gradually migrate downward. Cd contents showed an obvious decreasing trend in the 0-15 cm soil layer.
Fig. 11. The vertical-depth distribution of soil Cd. (a) The migration distribution of Cd on the 10th day; (b) The migration distribution of Cd on the 20th day; (c) The migration distribution of Cd on the 30th day; (d) The change curves of Cd migration on the 10th, 20th, and 30th day.
In contrast, Cd contents gradually increased in the 15-50 cm soil layer. Although soil Cd can downward migrate, the soil at 20-40 cm still hosts much downward-migrated Cd. As a result, it is reasonable to assume that Cd is mainly distributed in the plow layer of the soil, (i.e., 0-20 cm). The results of this study are comparable to other researchers who have studied the heavy metals vertical-depth distribution in soil using alternative traditional analytical tools [64–67]. This result presents adequate evidence of LIBS’s effectiveness in revealing the vertical-depth content of soil heavy metals. Although this study provides the vertical-depth distribution of Cd content in well-controlled soil columns, the vertical-depth distribution of Cd content in the natural environment needs further investigation.
4. Conclusions
In this study, we developed a rapid, reliable, and accurate method for detecting the vertical-depth distribution of Cd in soil. We employed LIBS combined with multivariate chemometrics to detect the vertical-depth distribution pattern of Cd in soil. CDP-LIBS showed good performance at an IT of 24 µs, which significantly enhanced the spectral signal of Cd. Compared with SP-LIBS, CDP-LIBS presented a much higher SBR and RI. Moreover, the PSO-optimized LS-SVM showed much better performance than the GA-optimized BP-ANN and PLSR, with an R2 of 0.999 and an RMSE of 0.359 mg/Kg in the prediction dataset. Hence, the combination of CDP-LIBS and PSO-optimized LS-SVM was adopted to detect soil Cd in vertical-depth series, which was helpful in revealing the vertical-depth migration pattern of Cd in soil. These results illustrate the vertical-depth migration pattern of Cd in soil using CDP-LIBS & chemometrics methods: (i), Cd content had a significant positive relationship with the inverse of soil depth; (ii), along incubation time, Cd slowly migrated downward in the soil. However, Cd was mainly concentrated in the 0-20 cm topsoil. This study indicates that CDP-LIBS technology provides a valuable tool for vertical depth detection of heavy metal elements in soils in areas with serious heavy metal contamination such as smelters and landfills.
Funding
Natural Science Foundation of Shaanxi Province (2022JM-100); National Natural Science Foundation of China (61705188).
Acknowledgments
This work was supported by the Key Laboratory of Agricultural Internet of Things, Ministry of Agriculture and Rural Affairs, P. R. China.
Disclosures
The authors declare that there are no conflicts of interest related to this article.
Data Availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
References
1. C. W. Gray, R. G. McLaren, A. H. C. Roberts, and L. M. Condron, “Solubility, sorption and desorption of native and added cadmium in relation to properties of soils in New Zealand,” European Journal of Soil Science 50(1), 127–137 (1999). [CrossRef]
2. J. Zhang, L.-L. Yu, S.-Z. Xin, and D.-C. Su, “Phytoavailability and chemical speciation of cadmium in different Cd-contaminated soils with crop root return,” Huan jing ke xue = Huanjing kexue 34, 685–691 (2013).
3. Y. T. Jiang, Y. J. Sun, L. P. Zhang, and L. Wang, “Influence factor analysis of soil heavy metal Cd based on the GeoDetector,” Stochastic Environmental Research and Risk Assessment 34(6), 921–930 (2020). [CrossRef]
4. S.-Z. Cao, Y. Mu, J.-X. Cui, A.-D. Liu, X. Cheng, Y.-H. Fu, S.-Q. Wei, and J.-Z. Zhang, “Cd Pollution and Safe Planting Zoning in Paddy Soils: A Case Study in a District of Chongqing,” Huan jing ke xue = Huanjing kexue 42, 5535–5544 (2021). [CrossRef]
5. Z. U. Rehman, S. Khan, M. L. Brusseau, and M. T. Shah, “Lead and cadmium contamination and exposure risk assessment via consumption of vegetables grown in agricultural soils of five-selected regions of Pakistan,” Chemosphere 168, 1589–1596 (2017). [CrossRef]
6. L. Zeng, Y. H. Wang, L. H. Jing, and Q. M. Cheng, “Quantitative determination of auxiliary information for mapping soil heavy metals and soil contamination risk assessment,” Appl. Geochem. 130, 104964 (2021). [CrossRef]
7. P. Andrews, R. M. Town, M. J. Hedley, and P. Loganathan, “Measurement of plant-available cadmium in New Zealand soils,” Soil Res. 34(3), 441–452 (1996). [CrossRef]
8. M. Yaman and Y. Dilgin, “AAS determination of cadmium in fruits and soils,” At. Spectrosc. 23, 59–64 (2002).
9. T. Frentiu, E. Darvasi, M. Senila, M. Ponta, and E. A. Cordos, “Preliminary investigation of a medium power argon radiofrequency capacitively coupled plasma as atomization cell in atomic fluorescence spectrometry of cadmium,” Talanta 76(5), 1170–1176 (2008). [CrossRef]
10. H. Zhang, S. Xie, Z. Bao, H. Tian, E. J. M. Carranza, W. Xiang, L. Yao, and H. Zhang, “Underlying dynamics and effects of humic acid on selenium and cadmium uptake in rice seedlings,” J Soils Sediments 20(1), 109–121 (2020). [CrossRef]
11. C. W. Gray, R. G. McLaren, and J. Shiowatana, “The determination of labile cadmium in some biosolids-amended soils by isotope dilution plasma mass spectrometry,” Soil Res. 41(3), 589–597 (2003). [CrossRef]
12. H. Tsukada, A. Takeda, H. Hasegawa, S. Ueda, and T. Iyogi, “Comparison of NAA and ICP-MS for the determination of major and trace elements in environmental sample,” J. Radioanal. Nucl. Chem. 263(3), 773–778 (2005). [CrossRef]
13. M. Rouillon and M. P. Taylor, “Can field portable X-ray fluorescence (pXRF) produce high quality data for application in environmental contamination research?” Environmental Pollution 214, 255–264 (2016). [CrossRef]
14. X.-l. Wang, J.-Y. Cui, A.-H. Tang, W.-X. Han, and R.-F. Jiang, “Optimal Operating Operating Condition of ICP-AES for Determination of Soil Nutrients Extracted by Mehlich 3 Through Solution Simulation,” Spectroscopy and Spectral Analysis 30, 2564–2566 (2010). [CrossRef]
15. G. Nicolodelli, J. Cabral, C. R. Menegatti, B. Marangoni, and G. S. Senesi, “Recent advances and future trends in LIBS applications to agricultural materials and their food derivatives: An overview of developments in the last decade (2010-2019). Part I. Soils and fertilizers,” Trac-Trends in Analytical Chemistry 115, 70–82 (2019). [CrossRef]
16. P. R. Villas-Boas, M. A. Franco, L. Martin-Neto, H. T. Gollany, and D. Milori, “Applications of laser-induced breakdown spectroscopy for soil characterization, part II: Review of elemental analysis and soil classification,” Eur J Soil Sci 71, 805–818 (2020). [CrossRef]
17. S. Kubitza, S. Schroeder, E. Dietz, S. Frohmann, P. B. Hansen, K. Rammelkamp, D. S. Vogt, M. Gensch, and H.-W. Huebers, “Detecting sulfur on the Moon: The potential of vacuum ultraviolet laser-induced breakdown spectroscopy,” Spectrochimica Acta Part B-Atomic Spectroscopy 174, 105990 (2020). [CrossRef]
18. J. Moros, M. M. ElFaham, and J. Javier Laserna, “Dual-Spectroscopy Platform for the Surveillance of Mars Mineralogy Using a Decisions Fusion Architecture on Simultaneous LIBS-Raman Data,” Anal. Chem. 90(3), 2079–2087 (2018). [CrossRef]
19. M. Wu, X. Wang, G. Niu, Z. Zhao, R. Zheng, Z. Liu, Z. Zhao, and Y. Duan, “Ultrasensitive and Simultaneous Detection of Multielements in Aqueous Samples Based on Biomimetic Array Combined with Laser-Induced Breakdown Spectroscopy,” Anal. Chem. 93(29), 10196–10203 (2021). [CrossRef]
20. D. Chen, T. Wang, Y. Ma, G. Wang, Q. Kong, P. Zhang, and R. Li, “Rapid characterization of heavy metals in single microplastics by laser induced breakdown spectroscopy,” Sci. Total Environ. 743, 140850 (2020). [CrossRef]
21. A. Botto, B. Campanella, S. Legnaioli, M. Lezzerini, G. Lorenzetti, S. Pagnotta, F. Poggialini, and V. Palleschi, “Applications of laser-induced breakdown spectroscopy in cultural heritage and archaeology: a critical review,” J. Anal. At. Spectrom. 34(1), 81–103 (2019). [CrossRef]
22. S. J. Rehse, H. Salimnia, and A. W. Miziolek, “Laser-induced breakdown spectroscopy (LIBS): an overview of recent progress and future potential for biomedical applications,” J. Med. Eng. Technol. 36(2), 77–89 (2012). [CrossRef]
23. V. K. Singh and A. K. Rai, “Prospects for laser-induced breakdown spectroscopy for biomedical applications: a review,” Lasers Med. Sci. 26(5), 673–687 (2011). [CrossRef]
24. M. Dong, L. Wei, J. J. Gonzalez, D. Oropeza, J. Chirinos, X. Mao, J. Lu, and R. E. Russo, “Coal Discrimination Analysis Using Tandem Laser-Induced Breakdown Spectroscopy and Laser Ablation Inductively Coupled Plasma Time-of-Flight Mass Spectrometry,” Anal. Chem. 92(10), 7003–7010 (2020). [CrossRef]
25. J. Li, Z. Zhu, R. Zhou, N. Zhao, R. Yi, X. Yang, X. Li, L. Guo, X. Zeng, and Y. Luo, “Determination of Carbon Content in Steels Using Laser-Induced Breakdown Spectroscopy Assisted with Laser-Induced Radical Fluorescence,” Anal. Chem. 89(15), 8134–8139 (2017). [CrossRef]
26. P. Modlitbova, P. Porizka, and J. Kaiser, “Laser-induced breakdown spectroscopy as a promising tool in the elemental bioimaging of plant tissues,” Trac-Trends in Analytical Chemistry 122, 115729 (2020). [CrossRef]
27. P. Yang, Z. Nie, and M. Yao, “Diagnosis of HLB-asymptomatic citrus fruits by element migration and transformation using laser-induced breakdown spectroscopy,” Opt. Express 30(11), 18108–18118 (2022). [CrossRef]
28. G. Hull, E. D. McNaghten, C. A. Sharrad, and P. A. Martin, “Combined laser ablation-tuneable diode laser absorption spectroscopy and laser-induced breakdown spectroscopy for rapid isotopic analysis of uranium,” Spectrochimica Acta Part B-Atomic Spectroscopy 190, 106378 (2022). [CrossRef]
29. F. R. Doucet, G. Lithgow, R. Kosierb, P. Bouchard, and M. Sabsabi, “Determination of isotope ratios using Laser-Induced Breakdown Spectroscopy in ambient air at atmospheric pressure for nuclear forensics,” J. Anal. At. Spectrom. 26(3), 536–541 (2011). [CrossRef]
30. C. R. Menegatti, G. Nicolodelli, G. S. Senesi, O. A. da Silva, H. J. I. Filho, P. R. Villas-Boas, B. S. Marangoni, and D. M. B. P. Milori, “Evaluation of LIBS under controlled atmosphere to quantify cadmium at low concentration in landfill leachates,” Appl. Phys. B 125(5), 74 (2019). [CrossRef]
31. J. Wang, N. Zhang, K. Hou, and H. Li, “Application of LIBS technology to the rapid measure of heavy metal contamination in soils,” Progress in Chemistry 20, 1165–1171 (2008).
32. K.-Q. Yu, Y.-R. Zhao, F. Liu, J.-Y. Peng, and Y. He, “Study on Soil Elements Detection with Laser-Induced Breakdown Spectroscopy: A Review,” Spectroscopy and Spectral Analysis 36, 827–833 (2016).
33. G. Chen, G. Yang, Z. Ling, Y. Yang, Y. Zhan, and X. Jin, “The parameter optimization of lasers’ energy ratio of the double-pulse laser induced breakdown spectrometry for heavy metal elements in the soil,” Anal. Methods 13(12), 1502–1510 (2021). [CrossRef]
34. B. S. Marangoni, G. Nicolodelli, G. S. Senesi, N. Fonseca, H. J. Izario Filho, A. A. Xavier, P. R. Villas-Boas, D. M. B. P. Milori, and C. R. Menegatti, “Multi-elemental analysis of landfill leachates by single and double pulse laser-induced breakdown spectroscopy,” Microchem. J. 165, 106125 (2021). [CrossRef]
35. G. Nicolodelli, G. S. Senesi, A. C. Ranulfi, B. S. Marangoni, A. Watanabe, V. d, M. Benites, P. P. A. de Oliveira, P. Villas-Boas, and D. M. B. P. Milori, “Double-pulse laser induced breakdown spectroscopy in orthogonal beam geometry to enhance line emission intensity from agricultural samples,” Microchem. J. 133, 272–278 (2017). [CrossRef]
36. S. Zhao, C. Song, X. Gao, and J. Lin, “Quantitative analysis of Pb in soil by femtosecond-nanosecond double-pulse laser-induced breakdown spectroscopy,” Results Phys. 15, 102736 (2019). [CrossRef]
37. G. Nicolodelli, G. S. Senesi, R. A. Romano, I. L. de Oliveira Perazzoli, and D. M. Bastos Pereira Milori, “Signal enhancement in collinear double-pulse laser-induced breakdown spectroscopy applied to different soils,” Spectrochimica Acta Part B-Atomic Spectroscopy 111, 23–29 (2015). [CrossRef]
38. R. Yi, J. Li, X. Yang, R. Zhou, H. Yu, Z. Hao, L. Guo, X. Li, X. Zeng, and Y. Lu, “Spectral Interference Elimination in Soil Analysis Using Laser-Induced Breakdown Spectroscopy Assisted by Laser-Induced Fluorescence,” Anal. Chem. 89(4), 2334–2337 (2017). [CrossRef]
39. W.-Y. Zhao, H. Min, S. Liu, Y.-R. An, and J. Yu, “Application Progress of Artificial Neural Network in Laser-Induced Breakdown Spectral Data Analysis,” Spectroscopy and Spectral Analysis 41, 1998–2004 (2021).
40. L.-R. Xiang, Z.-H. Ma, X.-Y. Zhao, F. Liu, Y. He, and L. Feng, “Comparative Analysis of Chemometrics Method on Heavy Metal Detection in Soil with Laser-Induced Breakdown Spectroscopy,” Spectroscopy and Spectral Analysis 37, 3871–3876 (2017). [CrossRef]
41. M.-G. Li, J. Liang, C.-H. Yan, H.-S. Tang, T.-L. Zhang, and H. Li, “Rapid Quantitative Analysis of Heavy Metals in Soil by Laser Induced Breakdown Spectroscopy Combined with Random Forest Algorithm,” Chinese Journal of Analytical Chemistry 49, 1410–1418 (2021). [CrossRef]
42. T. Wang, M. He, T. Shen, F. Liu, Y. He, X. Liu, and Z. Qiu, “Multi-element analysis of heavy metal content in soils using laser-induced breakdown spectroscopy: A case study in eastern China,” Spectrochimica Acta Part B-Atomic Spectroscopy 149, 300–312 (2018). [CrossRef]
43. G. W. McCarty, N. N. Lyssenko, and J. L. Starr, “Short-term changes in soil carbon and nitrogen pools during tillage management transition,” Soil Science Society of America Journal 62(6), 1564–1571 (1998). [CrossRef]
44. D. Yan, J. Li, J. Pei, J. Cui, M. Nie, and C. Fang, “The temperature sensitivity of soil organic carbon decomposition is greater in subsoil than in topsoil during laboratory incubation,” Sci. Rep. 7(1), 5181 (2017). [CrossRef]
45. K. Makkonen and H. S. Helmisaari, “Assessing fine-root biomass and production in a Scots pine stand - comparison of soil core and root ingrowth core methods,” Plant Soil 210(1), 43–50 (1999). [CrossRef]
46. A. J. Franzluebbers, “Soil organic matter stratification ratio as an indicator of soil quality,” Soil & Tillage Research 66(2), 95–106 (2002). [CrossRef]
47. W. Wang, W. Kong, T. Shen, Z. Man, W. Zhu, Y. He, and F. Liu, “Quantitative analysis of cadmium in rice roots based on LIBS and chemometrics methods,” Environ Sci Eur 33(1), 37 (2021). [CrossRef]
48. W. Wang, W. Kong, T. Shen, Z. Man, W. Zhu, Y. He, F. Liu, and Y. Liu, “Application of Laser-Induced Breakdown Spectroscopy in Detection of Cadmium Content in Rice Stems,” Front. Plant Sci. 11, 599616 (2020). [CrossRef]
49. G. Rao, L. Huang, M. Liu, T. Chen, J. Chen, Z. Luo, F. Xu, X. Xu, and M. Yao, “Identification of Huanglongbing-infected nave oranges based on laser-induced breakdown spectroscopy combined with different chemometric methods,” Appl. Opt. 57(29), 8738–8742 (2018). [CrossRef]
50. G.-F. Rao, L. Huang, M.-H. Liu, T.-B. Chen, J.-Y. Chen, Z.-Y. Luo, F.-H. Xu, H. Yang, X.-W. He, H.-M. Zhou, J.-L. Lin, and M.-Y. Yao, “Discrimination of Microbe Species by Laser Induced Breakdown Spectroscopy,” Chinese Journal of Analytical Chemistry 46, 1122–1128 (2018). [CrossRef]
51. C. Yan, T. Zhang, Y. Sun, H. Tang, and H. Li, “A hybrid variable selection method based on wavelet transform and mean impact value for calorific value determination of coal using laser-induced breakdown spectroscopy and kernel extreme learning machine,” Spectrochimica Acta Part B-Atomic Spectroscopy 154, 75–81 (2019). [CrossRef]
52. R. X. Yi, L. B. Guo, X. H. Zou, J. M. Li, Z. Q. Hao, X. Y. Yang, X. Y. Li, X. Y. Zeng, and Y. F. Lu, “Background removal in soil analysis using laser-induced breakdown spectroscopy combined with standard addition method,” Opt. Express 24(3), 2607–2618 (2016). [CrossRef]
53. N. B. Zorov, A. A. Gorbatenko, T. A. Labutin, and A. M. Popov, “A review of normalization techniques in analytical atomic spectrometry with laser sampling: From single to multivariate correction,” Spectrochimica Acta Part B-Atomic Spectroscopy 65(8), 642–657 (2010). [CrossRef]
54. A. Erler, D. Riebe, T. Beitz, H.-G. Loehmannsroeben, and R. Gebbers, “Soil Nutrient Detection for Precision Agriculture Using Handheld Laser-Induced Breakdown Spectroscopy (LIBS) and Multivariate Regression Methods (PLSR, Lasso and GPR),” Sensors 20(2), 418 (2020). [CrossRef]
55. M. De-shuo, Z. Nan-jing, M. Ming-jun, G. Yan-hong, Y. Yang, F. Li, W. Yuan-yuan, J. Yao, L. Wen-qing, and L. Jian-guo, “Rapid Soil Classification with Laser Induced Breakdown Spectroscopy,” Spectroscopy and Spectral Analysis 37, 241–246 (2017).
56. H. Duan, L. Han, and G. Huang, “Quantitative Analysis of Major Metals in Agricultural Biochar Using Laser-Induced Breakdown Spectroscopy with an Adaboost Artificial Neural Network Algorithm,” Molecules 24(20), 3753 (2019). [CrossRef]
57. X. Feng, Y. Liu, X. Li, and H. Liu, “RSM, ANN-GA and ANN-PSO modeling of SDBS removal from greywater in rural areas via Fe2O3-coated volcanic rocks,” RSC Adv. 12(10), 6265–6278 (2022). [CrossRef]
58. K.-Q. Yu, Y.-R. Zhao, F. Liu, and Y. He, “Laser-Induced Breakdown Spectroscopy Coupled with Multivariate Chemometrics for Variety Discrimination of Soil,” Sci. Rep. 6(1), 27574 (2016). [CrossRef]
59. K. Yu, Y. Zhao, Y. He, and D. He, “Response surface methodology for optimizing LIBS testing parameters: A case to conduct the elemental contents analysis in soil,” Chemom. Intell. Lab. Syst. 195, 103891 (2019). [CrossRef]
60. S. Ma, Y. Tang, Y. Ma, D. Dong, L. Guo, H. Zhu, J. Liu, and Y. Lu, “The pH effect on the detection of heavy metals in wastewater by laser-induced breakdown spectroscopy coupled with a phase transformation method,” J. Anal. At. Spectrom. 35(1), 198–203 (2020). [CrossRef]
61. L. C. Nunes, G. G. Arantes de Carvalho, D. Santos Junior, and F. J. Krug, “Determination of Cd, Cr and Pb in phosphate fertilizers by laser-induced breakdown spectroscopy,” Spectrochimica Acta Part B-Atomic Spectroscopy 97, 42–48 (2014). [CrossRef]
62. Z. Wang, Y. Deguchi, R. Liu, A. Ikutomo, Z. Zhang, D. Chong, J. Yan, J. Liu, and F.-J. Shiou, “Emission Characteristics of Laser-Induced Plasma Using Collinear Long and Short Dual-Pulse Laser-Induced Breakdown Spectroscopy (LIBS),” Applied Spectroscopy 71, 2187–2198 (2017). [CrossRef]
63. C. L. Liu, T. W. Chang, M. K. Wang, and C. H. Huang, “Transport of cadmium, nickel, and zinc in Taoyuan red soil using one-dimensional convective-dispersive model,” Geoderma 131(1-2), 181–189 (2006). [CrossRef]
64. Peter S. Hooda, Trace Elements in Soils, (A john Wiley and Sons: Chichester, 2010), Chap. 4.
65. A. Kabata-Pendias, Trace Elements in Soils and Plants, (Crc Press2000), Chap. 4.
66. Irena Sherameti and Ajit Varma, Soil Heavy Metals, (Springer, 2010), Chap. 3.
67. B. J. Alloway, Heavy Metals in Soils, (Springer, 2013), Chap. 2.
