A Spectroscopic Approach to Assess Heavy Metals Contents of the Mine Waste of Jalta and Bougrine in the North of Tunisia ()
1. Introduction
In the abondoned waste mines, acid water drainage can adversely impact the quality of drinking water and the health of riparian ecosystems. The Acidic water formed by sulfide oxidation can relatively generate soluble and insoluble secondary Fe-related minerals which are controlled by pH, level of oxidation, moisture content, and solution composition ([1] FRITZE et al., 1994; [2] DAVIES, 2001). The secondary Fe-related minerals are of special interest to this study since they have specific spectral features. In extremely acidic condition (pH 2.5 - 3.5) due to the high sulfur content of waste mine, the dissolution and precipitation of the secondary minerals genrate heavy metals which make possible to spectrally identify them on the basis of their spectral reflectance. The position of the specific absorption bands corresponding to seconddary Fe-related minerals are caused by the transition elements and crystal fields ([3] (CLARK, 1999).
Near-InfraRed Spectroscopy (NIRS) reflectance is a quantitative and qualitative technique that gives good resuts for the determination of various parameters in food, agriculture, textile, petrochemical, and pharmaceutical industries. NIRS has given good outcomes for the determination of physical, chemical, and biological soil properties such as moisture; total C, N, and P contents; mineral N; quality of organic matter; and biological parameters of soil such as respiration and microbial biomass ([1] FRITZE et al., 1994; [4] MALLEY,1998; [5] CONFALONIERI et al., 2001; [6] SLAUGHTER et al., 2001; [7] SMITH et al., 2001; [8] CHODAK et al., 2002; [9] MCCARTY et al., 2002). NIRS has been also used to predict trace metal content in sediments or soils ([10] MALLEY et al., 2000; [11] KEMPER ANDSOMMER, 2002).
([12] Malley and Williams, 1997) obtained accurate predictions of metals such as Fe, Mn, Zn, Cu, Pb, and Ni. In our study for the first time, we will use the NIRS reflectance Spectroscopy approach to investigate the potential acidic drainage effect within waste mine sites in the North of Tunisia. In contrast to previous studies, we focused more on modeling the relations between reflectance spectroscopy and geochemical variables of the waste mines by the use of Partial Least Square Regression (PLSR).
2. Materials and Methods
2.1. Study Area
The mining village of Jalta is located at 85 km from Tunis. It is located at the western North of Tunisia. In the mining sector of Jalta exploitation begun in 1898 and stopped in 1986. The total quantity exploited was about 270.000 tons. 70% to 80% of Pb was obtained by treating 400 tons/day ([13] SAINFELD, 1952; [14] ROUSSEV et al., 1976; [15] BOUSSEN, 2004). The dump is situated at 500 m in the west of the mine. It covers an area of 700 m length and 500 m width (Figure 1).
The mining village of Bougrine is located at 175 km from Tunis. It is located at the western North of Tunisia. In the mining sector of Bougrine, exploitation begun in 1901 and stopped in 2005 ([16] MASSIN, 1972; [17] GUEDRIA, 1981; [18] SEBAI, 2007). The mine waste is situated at 500 m in the west of the mine. It covers an area of 500 m length and 270 m width (Figure 2).
The Jalta and bougrine mining district are located at the western north of Tunisia. The waste mine which are rich in pyrite and other sulfides are now disposed in a large dump. The areas of Jalta and Bougrine mines waste are approximately 350,000 m2 and 67,500 m2, respectively. Oxidation of sulfides in these wastes releases trace metals into the rivers Gzenni (Jalta) and El Malah (Bougrine).
Figure 1. Location of the study area: mine of Jalta and waste samples.
Figure 2. Location of the study area: mine of Bougrine and waste samples.
2.2. Soil Sampling
A total of 22 waste samples were collected from the dump from Jalta and Bougrine at a depth <10 cm in April 2005 (see Figures 1 and 2). The selection of sample areas was based on a visual interpretation of a false color composite of an Aster image. The collected soil samples were air-dried and passed through a 2 mm sieve to remove large debris, stones, and stubbles. Each sample was split into two subsamples. One subsample was used for spectroscopic measurements and the other for physical and chemical analysis to characterize the heavy metals content.
2.3. Chemical Analysis
The laboratory analytical data are extracted from threeleachate phases: water, partial and total ([19] THOMPSON, 1983; [20] SIEGEL, 2002). Eight metals are selected based on their: Toxicity/Environmental context, Relation to local geology and mining activities, Spectral absorption property (transition elements) and Data completeness and variability. The selected elements are lead (Pb), zinc (Zn), manganese (Mn), iron (Fe), cadmium (Cd), copper (Cu), Chrome (Cr) and Nickel (Ni). The acidity (pH) and conductivity (EC) are also to be considered with the aforementioned parameters since they are important environmental indicators. The ICP-AES analyticcal instrumentsis used. Some elements were analyzed from the three extractions by the use of water, hydroxylamine for the partial extraction and aqua regia for the total extraction as digestion solutions (reagents).
2.4. Spectral Measurements
The waste samples were spectrally measured in laboratory by a spectrophotometer (Cary 500 version 8.01) with wavelengths varying from 400 to 2500 nm and a spectral sampling interval of 1 nm. In the laboratory, waste samples were sieved (<2 mm) for bi-directional reflectance measurements to circumvent micro-shadow effects from macro-aggregates. Waste samples were illuminated with two tungsten quartz halogen filament lamps in housing with aluminum reflectors. The lamps were placed on each side of the sample, with the light beam 30˚ from the vertical, to give a distance of 50 cm between the lamps and the sample. Reflected light was collected with a 25˚ field-ofview, perpendicular to the sample at a distance of 10 cm. Absolute bidirectional reflectance spectra were obtained by multiplying the raw reflectance spectra by the certified reflectivity of the Spectralon standard.
It is not always possible to measure all the waste properties in a single measurement (called “site” here). The measurements by spectrophotometer are three times repeated for the 22 samples. As a result, 66 spectra are obtained. Some properties of the waste spatially vary even for short distance and this is due to its structure in the plowed horizon. The difference in measurement condition generates an additional variability which is considered as noise, masking a part of the bands existing between the measured properties. The most obvious solution consists in reducing the number of the characterized sites in order to increase the number of repetitions of each measurement for the same site and thus to compensate the number of measurements being invariant. The decrease of the number of the characterized sites generates in its turn a reduction of the power of the statistical methods which are particularly used when the number of characterized sites is small. However, if the number of characterized sites is irreductible and it is necessary to increase the number of repetitions, it is still possible to use simple models, which require fewer measurements to be gauged. And that was the strategy which was retained for this study.
2.5. Modeling Approaches
Partial Least Square Regression (PLSR) is a method of relating Y-variables (dependants) with X-variables (predictors), which preferably have much collinearity so that the method gives better results ([21] WOLD et al., 2001). PLSR is thus a method where the information is commonly explained by the original X variables that is compressed into few variables known as PLS components (latent variables). The resulting smallest possible number of components would then simplify the interpretation of the relationship between X-data and Y-data since this relationship is concentrated on them ([2] DAVIES, 2001; [22] WANG et al., 2007). The PLSR in this research is based on the Beer Lambert law (or Beer’s law) which describes the linear relationship between absorbance of energy and concentration of an absorbing species ([23] BEN-DOR et al., and BENIN, 1995; [24] BEN-DOR et al., 2002; [25] BEN-DOR et al., 2003; [26] ESTIFANOS, 2006; [27] GOMEZ et al., 2005; [28] KOOISTRA et al., 2001; [29] KOOISTRA, 2004; [12] MALLEY and WILLIAMS, 1997; [10] MALLEY et al., 2000; [30] MILA, 2007; [31] THOMAS and STEFAN, 2002; [22] WANG et al., 2007; [32] YONGLING et al., 2008; [33] WU et al., 2005a; [34] WU et al., 2005b; [35] WU et al., 2007). The effect of the pollutants on the spectral property of the sediments stream is supposed to vary depending on their content in the mine waste. The statisticcal modeling of the geochemical variables was done using four sets of spectral data: the 32 spectral parameters and the original spectral reflectance values at 2151 wavelength positions. The spectral parameters are numerical values which are derived from some wavelength positions in the original reflectance or calculated as hull quotient and 1st derivative values. The original spectral reflectance values continuously vary for each interval of 1 nm in the whole wavelength range (400 - 2500 nm). The best predicting spectral parameter are compared with the best predicting spectral wavelength regions for confirmation. The magnitude with which the geochemical variables are predicted is compared in both cases. The software package, The Tanagra, employs a leave-one-out type validation method known as full cross-validation which uses the same data for both calibration and validation of a model. The regression can repeatedly be run each time avoiding insignificant X-variables till the best explaining significant predictors remain and thus make the model more robust. The significance level of the predictors is determined based on p-value <0.005 i.e. there shouldn’t be more than 5% probability of risk to consider so that the relationship between the dependant (Y-variables) and predictors (X-variables) is not due to chance. The Prediction Influence Value (PIV) >1 is used to evaluate the importance of the properties. The values and the hierarchy of the properties PIV that are proved to be enough dependent on the data initially used. In a first stage, the cross validation error “Leave-One-Out” which was retained to carry out a selection of the properties in a semi-independent way and the PIV is used only in a second stage to evaluate the importance of the properties finally retained. The cross-validation error equation is as follows:
where, RMSEcv is the root mean square error. Ymi is the value measured for the sample i from the chemical analysis, Ypi is the value predicted for the sample i by a model adjustment without the sample i, and nc is the number of samples used in the calibration set.
3. Results and Discussion
3.1. Chemical Analysis
The magnitude of the potential toxic metals impact on the ecosystem depends on their solubility in solutions of varying pH conditions and on the stages of formation of the metals. A toxic metal has the highest impact on the ecosystem when it exists in a mineral form that easily dissolves in water, otherwise, it has the lowest impact when this mineral needs rigorous acidic conditions to be dissolved, a rare case in nature such as AMD. Tables 1, 2 summarize the geochemical analysis results. For eight cations, pH and EC were measured within water solution, which was used for analyzing the water extractable cations.
The minimum pH value is 3 and maximum is 8.58. The EC ranges from 90 to 963 microS. Pb, Zn, Mn, Cu and Fe could be detected in all the three leachate phases of extraction. The maximum values in the total extractions are about 8000 ppm, 3000 ppm, 5300 ppm, 285 ppm and
Table 1. Statistical summary for the geochemical parameters in three extraction datasets (dump of mine of Bougrine).
Table 2. Statistical summary for the geochemical parameters in three extraction datasets (dump of mine of Jalta).
5.7% respectively. Ni, Cr, and Cd are detected in total and/or partial extraction. The maximum values in total extraction are about 45 ppm, 20 ppm, and 18 ppm respecttively.
The partial geochemical extraction was conducted using 20 ml of 0.25 mol hydroxylamine and 0.1 mol of HCl solution. It was aimed to extract the pollutants that might have been attached to amorphous phases of minerals. Unlike, sequential extraction which can selectively dissolve particular phase/mineral types, partial extraction also includes the water soluble mineral phases. The water extractable values, the difference between partial and water extraction (P-E) as well as between total and partial (T-P) extraction were considered to determine the pollutant concentration which are related to amorphous and crystalline phases. The following pie diagrams (Figure 3) illustrate the approximate and relative proportion of heavy metals in the three phases. Base metals have different modes of formation in waste, sediments and soils. The pie diagrams show that Pb-Zn-Cu which are attached to amorphous phases are stronger than Ni-Cr-Cd. The Cd was detected only in the total extraction and the Cr has very small proportion in the partial extractable phases (Figure 3). The high proportion of Mn in the (P-E) phase indicates the presence of Mn related to the amorphous phase (Figure 4).
Figure 3. Pie diagrams showing relative proportions of attached concentration: Pb, Zn, Ni, Cr, Cu, Mn and Fe. E: the water extraction, T-P: the difference between total and partial extraction and P-E: the difference between partial and water extraction.
Figure 4. A spectrum with wavelength positions of the waste mine sample.
3.2. Spectral Analysis
Analysis of spectra measured by the spectrophotometer was carried out for qualitative and quantitative interprettation of minerals. Spectra of laboratory were utilized to connect absorption bands to the chemical composition of the samples which were collected from the waste mine. The qualitative analysis of the spectral signatures of waste mine consists of investigating the shape of the curve and simple comparison with the curves of pure minerals. In spectroscopy, the absorption bands within 400 - 2500 nm range are very important for the identification of the minerals which have the absorbing physicochemical structure of photon. We can clearly observe two spectral areas in this range: the electronic area, Visible Near InfraRed (VNIR) (400 - 1300 nm) where absorptions are caused by transition metals for example, Fe, Pb, Mn, Ti, Cu, Co, and Cr and the area of vibration, Short Wavelength InfraRed (SWIR) (1300 - 2500 nm) where absorptions are caused by OH–, H2O, NH4+, 
and 
([36] SWAZE and CLARK, 1995; [37] SWAZE et al., 1998; [38] SWAZE et al., 2000). A first comparison of the samples spectra colledted in the two areas of study (Figure 4), show the existence of several minerals such as the iron which is detected by absorption bands in visible near infra-red (400 - 1300 nm). The presence of quartz, carbonates (calcite and dolomite), clays, water, organic matter is detected by absorption bands in infra-red (1300 - 2500 nm) respectively. During iron speciation, the oxidation of the waste of the mines causes the formation of the Fe– related secondary and release of heavy metals which can be precipitated or absorbed. Thus, the application of the spectroscopy to evaluate the mineralogical composition of pollutants is studied indirectly by the Fe-related secondary minerals which have specific spectral signatures in the wavelengths interval 400 - 2500 nm. In this spectral analysis, the iron minerals are mainly interpreted within VNIR area. The interpretation of the spectra of the iron minerals in this study is based on the combination of the absorption bands of load transfer and crystalline field which includes: The band nearby 438 nm, position of the minimum of absorption around 900 nm and the shoulder slope between 600 and 750 nm. Three endmembers were defined to represent the relatively dominant types of the iron mineral: “probable jarosite”, “probable goethite”, “and probable hematite”.
A spectrum is a rich source of information that can be extracted to determine the mineralogical composition. The quantification is based on the depth, the width and the asymmetry of an absorption band, the position of minimum reflectance and the shoulder slope of reflectance. Various algorithms were developed to transform the spectrum of reflectance to other forms to obtain better spectrum information. In this research, the curve of original reflectance, the first derivative and “hull quotient” are mainly considered. The original spectra of reflectance are adapted to calculate the difference of shoulder and the inflections of the slopes and the wavelength position of the minimum absorption. The depth and the bandwidth of absorption can be better estimated by “hull quotient” of spectrum. The first derivative determines the slope and length of the whole range of the spectrum. Consequently, 32 spectral parameters were chosen by the spectrum of reflectance (Table 3). Hull quotient and first derivative frorm of the spectra allowed us to observe the variability of absorption bands between samples. TSG software (www.thespectralgeologist.com), was used for the visualization and also the quantification of this variability. These spectral parameters were applied when modeling the physicochemical properties of the mine waste.
3.3. Coefficients of Regression Model and Spectral Properties of Heavy Metal Content
Figure 5 shows different plots resulting from one step of the regression process in the software Tanagra (http://chirouble.univ-lyon2.fr/~ricco/tanagra). The Principal Components Analysis (PCA) is used to determine the most significant parameters. The score plots display the results of PCA (Figure 5(a)). The closer the samples the more similar with respect to X and Y they are. Outlier samples could be removed during the next regression. In the regression coefficients plot (Figure 5(b)), each predictor