SSSAJ Journal of Natural Resources and Life Sciences Education
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Published online 16 May 2007
Published in Soil Sci Soc Am J 71:918-926 (2007)
DOI: 10.2136/sssaj2006.0285
© 2007 Soil Science Society of America
677 S. Segoe Rd., Madison, WI 53711 USA
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Wu, Y.
Right arrow Articles by Ma, H.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Wu, Y.
Right arrow Articles by Ma, H.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Wu, Y.
Right arrow Articles by Ma, H.
Related Collections
Right arrow Heavy Metals
Right arrow Remote Sensing
Right arrow Soil Pollution

SOIL CHEMISTRY

A Mechanism Study of Reflectance Spectroscopy for Investigating Heavy Metals in Soils

Yunzhao Wua,b,*, Jun Chenb, Junfeng Jib, Peng Gonga, Qilin Liaoc, Qingjiu Tiand and Hongrui Mab

a State Key Lab. of Remote Sensing Science jointly sponsored by the Inst. of Remote Sensing Applications of Chinese Academy of Sciences and Beijing Normal Univ.
b State Key Lab. of Mineral Deposit Research Inst. of Surficial Geochemistry, Dep. of Earth Sciences, Nanjing Univ., Nanjing 210093, China
c Geological Survey of Jiangsu Province, Nanjing 210093, China
d International Inst. for Earth System Science, Nanjing Univ., Nanjing 210093, China

* Corresponding author (njuessi{at}yahoo.com.cn).


   ABSTRACT
TOP
 NOTES
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Conventional methods for investigating heavy metal contamination in soil are time consuming and expensive. In this study, we (i) explored reflectance spectroscopy as an alternative method for assessing heavy metals, and (ii) further explored the physicochemical mechanism that allows estimation of heavy metals with the reflectance spectroscopy method. We first investigated the spectral response of changing concentrations of heavy metals in soils. The results indicated that only at very high concentration can transition elements exhibit their inherent absorption features. In spite of this observation, we successfully predicted low levels of heavy metals in agricultural soils. The best prediction accuracies were obtained for the siderophile elements Ni, Cr, and Co. The poorest prediction was for Cd. The order of prediction accuracy for metals was approximately the same as the order of their correlation coefficients with Fe. Complementary to some previous studies that found that the intercorrelation between heavy metals and active soil components (such as Fe oxides, organic matter, and clay) is the major predictive mechanism, in the present study we concluded that the correlation with total Fe (including active and residual Fe) is the major mechanism. This conclusion was further supported by both correlation analysis and chemical sequential extraction. Correlation analysis showed that all metals are negatively correlated with reflectance while positively correlated with the absorption depth at about 500 nm, a feature resulting from goethite. The chemical forms of heavy metals, which showed that besides the crystalline Fe oxide and organic matter fractions, heavy metals have significant amounts in the residual fraction, also strengthened the conclusion.

Abbreviations: CSE, chemical sequential extraction • ICP–AES, inductively coupled plasma–atomic emission spectroscopy • ICP–MS, inductively coupled plasma–mass spectroscopy • PI, pollution index • PLSR, partial least-squares regression • RMSEP, root-mean-square error of prediction • RPD, ratio of the standard deviation of the population to the root-mean-square error of prediction • VNIRS, visible and near-infrared reflectance spectroscopy


   INTRODUCTION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Heavy metals, trace elements in natural media at concentrations of <0.1%, are prevalent in soils. It is well known that excessive levels of heavy metals can be detrimental to organisms and human health. Conventional methods for investigating soil heavy metal contamination are based mostly on field sampling and wet digestion in concentrated acids and either atomic absorption spectrometry or inductively coupled plasma spectroscopy measurements, which are time consuming and relatively expensive. The expense is a serious problem because heavy metal contamination in soils is particularly prevalent in developing countries, where money is invested in economic development while environmental protection is relatively ignored. Therefore, the development of rapid and inexpensive techniques for investigating heavy metal soil contamination could be of great value.

Reflectance spectroscopy within the visible and near-infrared reflectance spectroscopy (VNIRS) region (380–2500 nm) has been widely used as an inexpensive tool for rapid analysis of a wide range of soil properties (Deaton and Balsam, 1991; Ben-Dor and Banin, 1995; Malley and Williams, 1997; Chang et al., 2001). In addition to rapidity and relatively low cost, it has another advantage in that reflectance spectra can be measured in the field using portable spectroradiometers. A successful prediction of various soil characteristics based on reflectance spectra can open the possibilities for a rapid and dynamic mapping using remote sensing technology. Recently there has been growing interest in the use of reflectance spectroscopy for soil characterization (Ji et al., 2002; Siebielec et al., 2004; Wu et al., 2005a, 2005b; Brown et al., 2006).

Early work on the spectral reflectance signatures of soil concentrated on qualitative identification of soil components (Condit, 1970; Stoner and Baumgardner, 1981). Since about the 1990s, with the advent of various mathematic methods such as partial least-squares regression, artificial neural networks, and multivariate adaptive regression splines, research has concentrated more on the quantitative prediction with reflectance spectroscopy of various soil constituents, most of which have absorptions within the VNIRS region, such as water (Liu et al., 2002; Whiting et al., 2004), Fe oxides (Scheinost et al., 1998; Ji et al., 2002; Grygar et al., 2003), carbonates (Ben-Dor and Banin, 1990), and organic matter (Dalal and Henry, 1986; Fidencio et al., 2002; Reeves et al., 2002). In addition, because of the intercorrelations to these spectral constituents, even spectrally featureless soil properties such as cation exchange capacity, base saturation, pH, exchangeable bases, and extractable P can been estimated, but only indirectly (Ben-Dor and Banin, 1995; Dunn et al., 2002; Shepherd and Walsh, 2002; Brown et al., 2006).

Some heavy metals, such as Ni, Cr, and Co, are transition elements. These elements have an unfilled d shell. The energy levels of d orbits will split when the atom of a transition element is located in a crystal field. Electromagnetic energy is absorbed when an electron moves from a lower level into a higher one (Burns, 1993; Clark, 1999). Malley and Williams (1997) first predicted heavy metals in sediments using reflectance spectroscopy. After several years, the quantitative prediction of heavy metals was made again for sediments (Kooistra et al., 2001), soils polluted by a mining accident (Kemper and Sommer, 2002), soils from a metal mining region (Siebielec et al., 2004), and agricultural soils (Malley et al., 2000; Wu et al., 2005a, 2005b).

These studies made significant improvements in the application of reflectance spectroscopy from soil composition to the soil microelements. There are some deficiencies, however. First, to our knowledge, there are no publications that report how great a concentration of a heavy metal is needed for it to directly express its intrinsic absorption features on the soil spectral curves. Second, most of the predictions have been based largely on mathematical relationships between spectra and chemical data, but did not show the physical relationship between them; i.e., the prediction mechanism by which trace metals, whose spectra are featureless, can be predicted, was seldom investigated. Even though, in some studies, the prediction mechanism was investigated, it was concluded indirectly using statistical methods such as regression coefficients or correlation analysis, and the conclusion was that heavy metal bonded to active soil components such as organic matter, Fe oxides, or clay is the predictive mechanism (Malley and Williams, 1997; Kooistra et al., 2001; Kemper and Sommer, 2002). Besides these fractions, heavy metals in soils can also exist in other chemical fractions, such as in the residue. In this study we used chemical sequential extraction (CSE) procedures to investigate the chemical forms of heavy metals (Shuman, 1985; Hall et al., 1996; Sahuquillo et al., 1999), and thereby hope to contribute to the revelation of the predictive mechanism.

Based on previous research, the objectives of this study were (i) to investigate the concentration at which heavy metals can exhibit their intrinsic absorption features, (ii) to assess the predictability of heavy metals in agricultural soils using reflectance spectroscopy, and (iii) to further explore the prediction mechanism by combining statistical methods and chemical sequential extraction procedures.


   MATERIALS AND METHODS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
This study was performed in two phases. First, we investigated the spectral response of heavy metals at various concentrations by artificially modeling contamination in the laboratory. Second, heavy metal concentrations in natural soil samples were predicted.

Study Area and Soil Samples
The study area, Baguazhou Island, is located at the northeast outskirts of Nanjing City, the capital of Jiangsu Province (Fig. 1). The climate of the Nanjing area is northern subtropical, with an average annual temperature of 16.7°C and mean annual precipitation 1239 mm. The topography of Baguazhou Island is flat with a maximum elevation of 5 m, which is below the water level of the Changjiang (Yangtze) River during the flood season. As a result, a dike was built around the entire island to prevent flooding. Baguazhou Island is the third largest island in the Changjiang River. It has a long history of cultivation. This alluvial island is dominated by paddy soil (USDA taxonomic classification of Agrudalf), with a small quantity of fluvo-aquic soil in the northwest. The industrial area of Nanjing is located off the north bank of Baguazhou Island and contains large industrial plants, such as iron and steel, chemical, and petrochemical works.


Figure 1
View larger version (25K):
[in this window]
[in a new window]

  		Fig. 1. The map of Baguazhou Island with sample locations.

 
A total of 61 topsoil samples (0–20 cm) were collected across the entire island, with a sampling density of one sample/km2. At each sampling point, three subsamples were taken and mixed to obtain a representative bulk sample that was used to estimate the heavy metal concentration at that sample locality. The soil samples were air dried at 20°C for 4 d and sieved through a 2-mm polyethylene sieve to remove large debris, stones, and pebbles. The samples were then ground and passed through a standard 0.15-mm (100-mesh) polyethylene screen. Each sample was divided into two subsamples. One was used for spectral measurements and the other analyzed for heavy metal concentrations.

Reflectance Measurements and Spectral Pretreatments
Reflectance spectra of the soil samples were recorded with a PerkinElmer Lambda 900 spectrophotometer (PerkinElmer, Waltham, MA). For the current measurements, we used a slit width of 2 nm in the 380- to 860-nm region and 20 nm in the 862- to 2500-nm region, at 2-nm increments, which produced 1061 spectral points between 380 and 2500 nm. Ground samples were made into a slurry on a glass microslide with distilled water, smoothed, and dried slowly at low temperature (<40°C). Data, reflectance intensity relative to a white spectralon standard, were written directly to a computer disk. Artifacts at detector changes, due to insensitivity at the limits of the detectors, were eliminated in the range of 842 to 898 nm. Because of low signal-to-noise levels near 2500 nm, only the 380- to 2300-nm wavelength range was used.

To diminish noise, the spectra were smoothed using a Savitzky–Golay smoothing algorithm. Because the finely ground soil sample was used for the spectral measurements, multiplicative spectral effects are very low. Therefore, in this study, multiplicative scatter correction was not used. To establish a robust prediction model, the first and second derivative spectra were calculated using Savitzky–Golay smoothing with a second-order polynomial fit on the smoothed reflectance data. The absorption features of metals are shallow and hence difficult to describe. One more transformation method—continuum removal—was used to enhance the weak absorption features. Continuum removal was accomplished using a program that ratios the spectrum with the lowest convex curve lying above the spectrum (Clark and Roush, 1984). In that case, the absorption band depth, D, was defined as follows:

Formula 1[1]
where Rb is the reflectance at the band center and Rc is the reflectance of the continuum at the same wavelength as Rb.

Modeling Contamination Experiments
One sample selected from the 61 soil samples was used for the modeling contamination experiments. In accordance with crystal field theory, three spectrally active heavy metals, Ni, Cr, and Cu were chosen to investigate the level at which they can exhibit their inherent absorption features. To provide a comparison, Hg was also modeled, although its spectra are featureless. Table 1 shows the sequential concentrations of heavy metals added artificially and the chemical reagents used in the modeling process. The electron configuration and the crustal abundance of the metal cation are also shown in Table 1. All the solutions used in this study were made with deionized water (Milli-Q, 18.2 M{Omega}/cm).


View this table:
[in this window]
[in a new window]

  		Table 1. The sequential concentrations of metals and reagents for modeling contamination.

 
For the samples in which the metal concentration was ≤1000 mg kg–1, 2-g soil samples were placed in 40-mL glass beakers and wetted with deionized water. The volumes of solution calculated according to the concentration shown in Table 1 were added to the slurries to obtain the desired concentrations, and then immediately placed in an ultrasonic bath for 15 min. For the samples in which the metal concentration was >1000 mg kg–1, the solid reagents instead of solution were added to 2-g soil samples and ground in an agate grinder to homogenize the samples. The above distinction in sample preparation was made to avoid overadsorption at the interface due to high concentration while pouring the solution into the soils. Finally, additional water was added so that all the samples were 40 mL. The beakers were covered with a watch glass in a forced-air oven at 40°C. When there was no visible water, another 40 mL of deionized water was added to the sample. The above process was repeated four times (22 d) before reflectance spectra were measured on these artificially contaminated samples.

Prediction Experiments
Chemical Analysis
The 61 soil samples were analyzed for Ni, Cr, Co, Cu, As, Zn, and Pb using an acid digestion method (Li and Thornton, 1993) and with inductively coupled plasma–atomic emission spectroscopy (ICP–AES, PerkinElmer 3300DV). Cadmium was analyzed by atomic absorption spectrometry. Although Fe is not a toxic metal, it was also analyzed with ICP–AES because it strongly affects soil spectra and Fe oxides can adsorb heavy metals. Soil organic carbon (SOC) has many of the same effects described for Fe. The SOC content was determined by wet digestion with an acid solution of potassium dichromate with external heat and reflux condensers. This was followed by a titration of the excess potassium dichromate using a solution of ferrous ammonium sulfate following modifications of Cambardella and Elliott (1992). The pH of the soils in water (1:2) was measured using an Orion Model 810 pH meter (Orion Research, Boston, MA).

Model Construction and Validation
The regression method used to establish relations between the spectral responses (reflectance and their derivatives) and soil heavy metal concentration was partial least squares regression (PLSR) performed using the Unscrambler software package version 8.0.5 (CAMO ASA, Trondheim, Norway). Unscrambler has two types of regression, PLS1 for calibrating one variable at a time and PLS2, which includes all the variables in the calibration. In our research, the PLS1 regression was used to build the model. A leave-one-out cross-validation procedure was used to estimate the optimum number of terms to avoid overfitting. With leave-one-out cross-validation, one sample was left out of the global data set and the model was calculated on the remaining data points. The value for the left-out sample was then predicted, and the prediction residual computed. The process was repeated with another sample of the data set, and so on, until every sample had been left out once; then all prediction residuals were combined to compute the root-mean-square error of prediction (RMSEP). The optimum number of terms was taken as the number resulting in the minimum RMSEP. For validation, only the cross-validation was performed while not using test set validation due to a limited number of samples. Moreover, the ratio of the standard deviation of the population to RMSEP, also know as RPD, was also calculated. The RPD demonstrates how well the calibration models could predict chemical data (Fearn, 2002; Moron and Cozzolino, 2004).

Chemical Sequential Extraction
Eight samples, the Fe concentrations of which exhibited significant variation, were selected from the 61 samples for the analysis of chemical fractions of soil metals. Selected properties of the eight soil samples are given in Table 2. The measurement of the specific surface area for the eight samples was obtained from gas-adsorption isotherms. Nitrogen adsorption isotherm measurements were performed on an ASAP2010 volumetric adsorption analyzer from Micromeritics (Norcross, GA) at liquid N2 temperature (77K) in a relative pressure (p/p0) range from about 0.01 to 0.995 using about 1.0000 g of soil samples. Before isotherm measurements, each sample was outgassed to the pressure of <0.02 Pa at 423K for about 10 h. The specific surface area (SBET) was calculated using the standard Brunauer–Emmett–Teller (BET) method (Brunauer et al., 1938) with a p/p0 range between 0.05 and 0.20. The primary aim of the CSE in this study was to explore the mechanism that allows reflectance spectroscopy to predict heavy metal concentration. Iron oxides and organic matter in soils are important because they are spectrally active and both of them can adsorb heavy metals. Therefore, with emphasis on the species bound to the two soil components and based on examining and comparing numerous published results (Shuman, 1985; Hall et al., 1996; Sahuquillo et al., 1999; Ahnstrom and Parker, 1999), the CSE used in our study is shown in Table 3.


View this table:
[in this window]
[in a new window]

  		Table 2. Selected characteristics of the eight samples used for the chemical sequential extraction.

 

View this table:
[in this window]
[in a new window]

  		Table 3. Sequential extraction procedure.

 
An aliquot of the supernatant was filtered into clean glass test tubes. To ensure one uniform matrix for inductively coupled plasma–mass spectroscopy (ICP–MS) analysis, the various extraction matrices were digested with concentrated HNO3. To the 5-mL aliquots of extractant solutions, 5 mL of HNO3 (concentrated, 70% v/v) was added, and then the solutions were heated to dryness at 90°C. The residue in the tubes was then leached and 2% HNO3 added to produce a final volume of 10 mL. The concentration of trace elements in the extracts was determined by ICP–MS, and by ICP–AES for Fe. Two independent replicates were performed for each sample and blanks were measured in parallel for each set of analyses using the extraction reagents described above. For each step, reflectance spectra were measured for the residue.


   RESULTS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 CONCLUSIONS
 REFERENCES
 
Spectral Response of Heavy Metals in Soils
Figure 2 shows the mean reflectance spectra of the eight soil samples after each sequential extraction step and the continuum-removed spectra. In the near-infrared region, the absorption features are narrow and sharp, while in the visible region the features tend to be wider and weaker. The three most distinct absorption bands at 1400 and 1900 nm are attributable to vibrational frequencies of OH groups in the adsorbed water, and the absorption feature at 2200 nm is related to OH groups in the crystal lattice water of illite (x-ray diffraction not shown data). The presence of Fe results in absorption at wavelengths in the 380- to 1300-nm region and rapid fall-off of reflectance toward the blue wavelengths, while the absorption features are weak (Hunt et al., 1971; Burns, 1993). After removing carbonates, Mn oxides, and organic matter, the absorption features in the visible region became obvious (Fig. 2a). Through continuum removal, a doublet absorption feature near 500 nm, which is caused by goethite, can be exhibited (Fig. 2b). Although the absorption bands of goethite at 950 nm are also strong, it cannot be exhibited on either the original or the continuum-removed curve, even with the soil organic matter removed. Therefore, the diagnostic spectral feature for identifying goethite in soils is 500 nm but not 950 nm, and this has been validated by previous research (Balsam and Deaton, 1991; Wu et al., 2005a).


Figure 2
View larger version (22K):
[in this window]
[in a new window]

  		Fig. 2. The mean reflectance spectra of (a) the eight soil samples after each sequential extraction step and (b) the continuum-removed spectra (original = original soil; F1 = carbonates removed; F2 = carbonates and Mn oxides removed; F3 = carbonates, Mn oxides, and organic matter removed; F4 = carbonates, Mn oxides, organic matter, and amorphous Fe oxides removed; F5 = carbonates, Mn oxides, organic matter, amorphous Fe oxides, and crystalline Fe oxides removed).

 
Similar to Fe, the absorption bands of the studied transition metals are shorter than 1300 nm (Fig. 3). Soil spectral features within this region are weak, and thus it is difficult to identify the spectral difference of the soils with different heavy metal concentrations from the original reflectance curve. Making a ratio of the spectra against that of the untreated sample can enhance the spectral difference, thereby contributing to the identification of spectral features. Moreover, continuum removal was used on the ratio spectra to further enhance the weak absorption features of transition elements. From Fig. 3 it can be seen that, regardless of the absorption features, heavy metal elements cannot be detected with reflectance spectroscopy at concentrations ≤1000 mg kg–1 When the concentration for Cr and Cu is as high as 4000 mg kg–1, their absorption features around 610 and 830 nm can be distinguished and the depth increases as concentration increases.


Figure 3
View larger version (30K):
[in this window]
[in a new window]

  		Fig. 3. Continuum-removed reflectance spectra of soils with varying concentrations of Cr and Cu. The "continuum-removed" spectrum was calculated as the ratio of the reflectance of the artificially contaminated samples to that of the untreated samples.

 
The modeling contamination experiment answered the question about the concentration at which heavy metals can exhibit their inherent absorption features. For agricultural soils, however, few are contaminated to such high levels. Therefore, under real-world conditions, heavy metals cannot be identified directly by their absorption features with reflectance spectroscopy. New questions then emerge: (i) can heavy metals in agricultural soils be predicted with reflectance spectroscopy; and (ii) if the answer is positive, what is the predictive mechanism?

Prediction for Agricultural Soils
Descriptive data statistics for the 61 soil samples, including mean, SD, minimum and maximum concentrations, CV, etc., are shown in Table 4. The CV, which is SD/mean, indicates the degree of discrete distribution of metal concentrations and, indirectly, the activity of the element. To approximate the intensity of soil contamination, a pollution index (PI) was defined. The PI was calculated as the ratio between the mean concentration and the background value specifically for this element. The background value comes from the research results of the Geological Survey of China (Wang, 2003).


View this table:
[in this window]
[in a new window]

  		Table 4. Statistical information for the 61 soil samples.

 
Principal component analysis (PCA) can be used to clarify the relationships between the heavy metals (Li et al., 2004; Han et al., 2006). A component loading plot of Factors 1 and 2 obtained from the results of PCA for all analyzed elements of the 61 soil samples shows that the four siderophile elements Fe, Co, Ni, and Cr cluster together (Fig. 4). Correlation analysis also indicates that the correlation coefficients between Fe and the other three siderophile elements is very high (R > 0.97). The two chalcophile elements Cd and Pb are separated from the other elements. Their correlation coefficients with Fe are lower than those of other heavy metals (Table 4). From Table 4 it can be seen that Cd has the highest PI and CV values. The three siderophile elements, Ni, Cr, and Co, have PI values near 1.0, suggesting that they are closely related, which is not surprising considering their affinity during geochemical cycles. Cadmium, however, behaves differently than the other metals.


Figure 4
View larger version (9K):
[in this window]
[in a new window]

  		Fig. 4. Loading plot of heavy metals and Fe by principal component analysis of 61 samples.

 
The RMSEP is known as a good criterion to judge the prediction performance of PLSR models (Shenk and Westerhaus, 1993). Here, the smallest RMSEP value was used to determine the optimal calibration model. Table 5 shows the optimal validation results of the PLSR models for the heavy metals of the 61 agricultural soil samples. The best prediction results for Cr and As were acquired with the first derivative spectra. For other metals, the best predictions were acquired with the reflectance percentage data. The results of the second derivative spectral data were all unsatisfactory.


View this table:
[in this window]
[in a new window]

  		Table 5. Cross validation statistics for heavy metals of the 61 soil samples.

 
From Table 5 it can be seen that the highest prediction accuracies were acquired for the three siderophile elements—the R2 for these elements was about 0.8, and RPDs > 2.0. The validation results for Cu and As, the two chalcophile elements, gave R2 near 0.7, and RPDs > 1.7. The poorest prediction was for Cd. Figure 5 shows the scatterplots of predicted vs. measured concentrations in the validation stage for the 61 samples. The points for all the metals except Cd fall in the vicinity of the 1:1 line. For Cd, the points at the higher end of the measured range show an underestimation of the predicted value. In this study, the prediction accuracies for the siderophile elements Ni, Cr, and Co are the highest, Cu, As, Zn, and Pb are moderate, while Cd is the worst.


Figure 5
View larger version (35K):
[in this window]
[in a new window]

  		Fig. 5. Plots of measured vs. predicted concentrations of eight heavy metals. The 1:1 line is indicated on each figure.

 

   DISCUSSION
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
CONCLUSIONS
 REFERENCES
 
To further understand the mechanism that allows estimation of heavy metals with reflectance spectroscopy, some statistical analysis was first made. Figure 6 shows the correlation coefficients between reflectance and heavy metals as well as Fe across the spectral range. For all eight trace elements, the general trend of the curves of their correlation coefficients as a function of wavelength was virtually similar, and similar to the coefficients of Fe. In general, metals are all negatively correlated with reflectance. Regions around 480 to 510 nm, the feature resulting from goethite, had particularly high correlation coefficients. Conversely, metals are positively correlated with the absorption depth at 500 nm (Fig. 7). The highest correlation coefficients were acquired for the three siderophile elements Ni, Cr, and Co, while the correlation coefficient for Cd was the lowest among the eight elements.


Figure 6
View larger version (25K):
[in this window]
[in a new window]

  		Fig. 6. Correlation coefficients between reflectance and metals.

 

Figure 7
View larger version (21K):
[in this window]
[in a new window]

  		Fig. 7. Correlation coefficients between absorption depth around 500 nm and metal concentrations (D_500), and crystalline Fe oxide content and crystalline Fe oxide fractions for all the metals.

 
The results shown for the relationships between spectral parameters (absorption depth and reflectance values) and heavy metal concentration indicate that Fe oxides play an important role in the prediction of soil heavy metals using reflectance spectra. The crystalline Fe oxides in soils of the research area are almost entirely goethite (Y. Wu, unpublished data, 2005). The yellowish color of soil is attributable to the Fe3+ absorption near the 500 nm of goethite. Specimens of goethite have large surface areas and can adsorb heavy metals. As goethite content in soils increases, the heavy metal concentrations increase accordingly. On the other hand, the absorption depth at 500 nm becomes deeper with increased goethite content (Wu et al., 2005a). Therefore, heavy metals are positively correlated with the absorption depth at 500 nm. In contrast, the reflectance values of soils decrease because of the absorption of goethite. Therefore, soil reflectance is negatively correlated with heavy metals.

Besides being bonded to Fe oxides, heavy metals in soils can also exist in several other chemical fractions as well as Fe. The results from sequential extraction show that Fe was primarily in the crystalline Fe oxide, residue, and amorphous Fe oxide fractions (Fig. 8). Other forms of Fe were almost negligible. The distributions of Ni and Co among different fractions were visually almost identical, with the highest in the crystalline Fe oxide fraction and lower in residual fraction. Chromium was highest in the residue and low in the crystalline Fe oxide fraction, showing abnormity with Ni and Co despite being a siderophile element. Although the fractionation scheme is diverse for all the metals, the results of chemical sequential extraction indicate that, besides the crystalline Fe oxide fraction, heavy metals also have significant amounts in the residual fraction. Although only crystalline Fe oxides have distinct absorption peaks, other forms of Fe such as residual and amorphous Fe oxide can also contribute to soil reflectance. For example, they may reduce the values of reflectance within the whole spectral region. Note that in this study, the prediction of heavy metals also uses whole wavebands instead of 500-nm absorption peaks. Therefore, complementary to the previous study in which the correlation with active soil components (such as Fe oxides, organic matter, and clay) allowed estimation of heavy metals with reflectance spectroscopy (Malley and Williams, 1997; Kooistra et al., 2001; Kemper and Sommer, 2002), in this study it was concluded that the intercorrelation between heavy metals and total Fe (including active and residual Fe) is the major mechanism.


Figure 8
View larger version (29K):
[in this window]
[in a new window]

  		Fig. 8. Average percentage distribution of heavy metals and Fe between various fractions determined by the chemical sequential extraction procedure (F1 = exchangeable; F2 = Mn oxides; F3 = organic matter; F4 = amorphous Fe oxides; F5 = crystalline Fe oxides; F6 = residual).

 
This conclusion is further supported by the relationships between heavy metals, Fe and reflectance. Comparing the prediction results and the correlation coefficients between Fe and heavy metals (Tables 4 and 5), it is interesting to note that all the heavy metals that have a high correlation with Fe also have high cross-validation statistics. The three siderophile elements Ni, Cr, and Co have the highest correlation coefficients with Fe. Their predictive accuracy is also the highest among the eight elements. The most poorly predicted element, Cd, shows the lowest correlation with Fe.

In addition to Fe content, there are many other properties affecting soil reflectance, such as moisture, particle size, organic matter, and roughness. In this study, moisture, particle size, and roughness were similar for all the samples because the samples had been dried and ground before the spectral measurements. Another important property of soil, organic matter, can absorb electromagnetic waves and adsorb heavy metals; however, the correlation between organic matter and heavy metals is very poor in the study area (Table 4). The crystalline Fe oxide content (i.e., goethite) was also positively correlated with the crystalline Fe oxide fractions for all the metals (Fig. 7), while the correlation coefficients between organically bound heavy metals and organic matter content were not significant (Y. Wu, unpublished data, 2005). Moreover, as seen from Fig. 8, the sum of the fractions related to Fe, such as that bonded to amorphous Fe oxide, crystalline Fe oxide, and residue (i.e., F4 + F5 + F6) is significantly larger than the fraction bonded to organic matter. Therefore, it is thought that Fe plays a more important role in predicting heavy metals than organic matter. The reason is perhaps because the content of inorganic component fractions in dry soils is much larger than the organic component. Here, it should be clarified that it does not mean that organic matter does not have any effect on the relationship between soil spectra and heavy metals. Soil organic matter also contributes to the relationship, although its contribution is not so much as that of total Fe. This is different from food and forage crop research by reflectance spectroscopy. In that case, the content of inorganic elements is low relative to the organic matrix. Thus, successful prediction for inorganic components in forage crops could be based on the correlations between the inorganic components and the organic constituents (Siebielec et al., 2004).

The poorer predictive results for Cd suggest that it behaves differently than the other metals. Figure 8 shows that Cd is the element with the highest exchangeable fraction, whereas the residual fraction is low in comparison with other elements. It is not like the siderophile elements, which are closely correlated with each other. Cadmium is separated from the other chalcophile elements Zn and Cu (Fig. 4). Moreover, it has the highest PI value (PI = 1.65) and CV (0.18) among the eight elements.


   CONCLUSIONS
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
REFERENCES
 
Reflectance spectroscopy is a simple and nondestructive analytical method that can be used to predict not only spectral active constituents but also trace elements, which are spectrally featureless. This study shows that only transition elements, which have an unfilled d shell and are present at very high concentration, can exhibit their inherent absorption features. The results of modeling experiments indicate that when the concentration of Cr or Cu is increased to 4000 mg kg–1, their absorption features can be exhibited, and the intensity increases with increasing concentration. Although few agricultural soils have heavy metal contaminants that can be identified directly by reflectance spectroscopy, it still can be used to estimate heavy metal concentrations. For the research area, the highest prediction accuracies were for the siderophile elements Ni, Cr, and Co, and the poorest prediction was for Cd. The order of prediction accuracy for metals is approximately the same as the order of their correlation coefficients with Fe. Complementary to the previous study in which the intercorrelation between heavy metals and active soil components (such as Fe oxides, organic matter, and clay) was elucidated as the major mechanism by which trace metals can be predicted, in the present study it was concluded that the correlation with total Fe is the major mechanism. This conclusion was further supported by both statistical analysis and the chemical sequential extraction procedures. Correlation analysis showed that all metals are negatively correlated with reflectance while positively correlated with the absorption depth at about 500 nm, a feature resulting from goethite in soils. The chemical forms of heavy metals, which showed that besides the crystalline Fe oxide and organic matter fractions, heavy metals have significant amounts in the residual fraction, also strengthened the relationship.


   ACKNOWLEDGMENTS
 
This research was supported by the Geological Survey of Jiangsu Province and the Hundred Talent Project of the Chinese Academy of Sciences. We are grateful to Emeritus Professor Joseph Giles of North Dakota State University, Professor Arthur P. Cracknell of the University of Dundee, and Professor William L. Balsam of University of Texas at Arlington for technical assistance and English language support. We thank professor Xiancai Lu of Nanjing University for specific surface area measurements.


   NOTES
TOP
 NOTES
ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 
All rights reserved. No part of this periodical may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Permission for printing and for reprinting the material contained herein has been obtained by the publisher.

Received for publication August 17, 2006.


   REFERENCES
TOP
 NOTES
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 CONCLUSIONS
 REFERENCES
 





This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF) Free
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Similar articles in this journal
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Wu, Y.
Right arrow Articles by Ma, H.
Right arrow Search for Related Content
PubMed
Right arrow Articles by Wu, Y.
Right arrow Articles by Ma, H.
GeoRef
Right arrow GeoRef Citation
Agricola
Right arrow Articles by Wu, Y.
Right arrow Articles by Ma, H.
Related Collections
Right arrow Heavy Metals
Right arrow Remote Sensing
Right arrow Soil Pollution

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
The SCI Journals Agronomy Journal Crop Science
Journal of Natural Resources
and Life Sciences Education		 Vadose Zone Journal
Journal of Plant Registrations Journal of
Environmental Quality		 The Plant Genome