**3. Method**

## **3.1 Site description and sampling**

This study was carried out at the Xuzhou-Datun coal mine district, located at the northwest of Jiangsu province, eastern China (**Figure 1**). The area of Xuzhou city is in the plain of Huanghuai, South part of northern China. Sediment stratum covering the Archean system are Simian, Cambrian, middle-lower Ordovician, middle-upper Carboniferous, Permian, Jurassic, Cretaceous, Tertiary, and Quaternary system, from bottom to top. The hydrogeology cell selected for this study is isolated by a series of faults. This includes Sanhejian, Yaoqiao, and Longdong coal mines shown in **Figure 1**. In this area, groundwater flows from northeast to southwest.

The coal seams that are being mined are located in the Carboniferous and Permian systems, the former include Benxi and Taiyuan formations, and the latter include

**Figure 1.** *Location of the study area.*

#### *Leaching Mechanisms of Trace Elements from Coal and Host Rock Using Method of Data Mining DOI: http://dx.doi.org/10.5772/intechopen.100498*

Shanxi and Lower-Shihezi formations, listed from the bottom to top in both systems. In Permian strata, there are mostly low sulfur content Gas coal and fat coal. The lower formation in Carboniferous has a higher content of sulfur than the upper layers. Mass percentage of sulfur in Permian Shanxi formation coal seams is around 0.83% in coal seam No.7 and 1.09% in coal seam No.9. In coal seam No.17 and No.19 in the Taiyuan formation, the average sulfur content was tested to be 1.87 and 3.49%, respectively. The two mining coal seams (No.2 and No.7) in the Permian system were included in this study; these are located in the middle Lower-Shihezi formations (No.2) and Shanxi formations (No.7). The two formations give thickness of 187–302.95 m and 81.67–136.13, respectively. White feldspar, quartz granule-sandstone, and siliconmudstone cementation are the main minerals in the lower Shanxi formation. In addition, siltstone, siderite, carbon-mudstone, and plant-fossil clast can also be found. Gray mudstone, sand-mudstone, and sandstone are the major rocks in the middle Shanxi formation with some silicon-mudstone and siderite also present.

There are six aquifers in the sediment stratum of the hydrogeology cell. A grit aquifer in the Quaternary, a conglomerate rock aquifer in the Jurassic, two sandstone aquifers—one in the lower-Shihezi formation, and one above the coal seam in the Shanxi formation; and two limestone aquifers—one is located in the Carboniferous Taiyuan formation (thickness of 180–200 m) and the other in the Ordovician (thickness of 600 m). These last two aquifers are the main water sources of the coal seam.

#### **3.2 Leaching experiments and sample test**

A total of 16 water samples and 28 rock/coal samples were collected from the study area. Water samples were collected in 1000 mL Nalgene bottles previously acidcleaned and rinsed twice using the water to be collected. pe and pH of water samples were taken in the field by using a JENCO 6010 pH/ORP meter. Coal and rock samples were collected from the working area at the mine and put into plastic bags that were immediately sealed.

Major ions and physical parameters of water samples were determined according to Chinese standard protocols in Jiangsu Provincial Coal Geology Research Institute. Solid samples were acid digested to determine the concentration of trace elements. The concentration of trace elements in water/coal/rock samples was determined by ICP-MS and the ICP-AES. The ICP-MS analysis was carried out in the China University of Mining and Technology using the X-Series ICP-MS—Thermo Electron Co. An internal standard of Rh was used to determine the limit of detection (0.5 pg/mL) and analytical deviation (less than 2%). The ICP-AES analysis was carried out in the Nanjing University using a JY38S ICP-AES model. The limit of detection and deviation for the analysis carried out by such equipment are 0.01 μg/mL and less than 2%, respectively.

Leaching experiments were conducted using the batch mode to simulate conditions in a coal seam where water movement is slow and dissolution reactions tend to achieve equilibrium, with regard to the previous studies [44, 45]. To simulate a "closed environment" (with low pO2; see Stumm and Morgan [46] for details), bottles were closed with a rubber stopper; samples were taken out using syringes. The pe of the solution during experiments was determined by a JENCO 6010 pH/ORP meter.

Three subsamples were used for each sample: one per 1000 mL aliquot of deionized water at the following pHs: 2, 5.6, 7, and 12. Flasks were sealed and shaken every 2 h for up to 10 days. The temperature was controlled using a water bath at about 40°C. Leachate solutions were collected using syringes at 2, 6, 24, and 48 h. A total of 0.5 mol/L HNO3 was added into all the samples. Leachate aliquots were titrated with

HCl or NaOH, depending on the pH conditions, to compare the behavior of leaching elements in acid, neutral, and alkali environments. In addition to leaching experiments, water samples including those collected from the Zhaoyang Lake and Yunlong lakes, shown in **Figure 1**, were shaken every 2 h for up to 10 days at a constant temperature of 40°C.

#### **3.3 Multivariate analysis**

While univariate statistical analysis of a large scale of data could be cumbersome and cause misunderstanding and error in the interpretation, multivariate statistical techniques are more robust. Therefore, it becomes a more useful tool for environmental data treatment and identification of anomalous patterns. During the immigration process of the trace elements from coal bedding seam to groundwater and surface water, in the complex matrix system, solid and liquid bodies are involved. In each system, the elements show different or similar coexisting patterns, and immigration behavior, including dissolution, transport, adsorption. Therefore, the multivariate analysis can be used to find out different and similar components, which suggest similar and dissimilar occurrences in solids, and immigration mechanisms during the process of water-rock interaction.

In the area of hydrochemical studies, the PCA method has been widely used to reduce dimensions and analyze the relations among the variates and samples [32–34, 47–51]. The PCA is a typical nonsupervised analytical method. To calculate the PCA result, data are first standardized by mean centering each column within the original data matrix and then dividing each of the values within each column by the column standard deviation. With PCA, the large data matrix is reduced to smaller ones that consist of PC loadings and scores. PC loadings are the eigenvectors of the correlation matrix depending on PC scores. Therefore, it contains information on all of the variables combined into a single number, with the loadings indicating the relative contribution that each variable makes to that score. PCs are calculated so that they take into account the correlations present in the original data but are uncorrelated with others. Typically, the data can be reduced to two or three dimensions representing the majority of the variance within the original data. Sometimes, more dimensions may have to be included to present more variance of the original data [33]. Based on the PCA analytical result, the loadings and scores of the data frame were then clustered in the dimensions that PCA has reduced. As the axis of coordinates was rotated to achieve maximum loadings of elements, the rotated axis of coordinates was marked as RCs.

The bi-plot of the PCA result is usually drawn to show patterns of parameters and samples. However, the loading and score of the PCA axis show different aspects of the result. In our study, the loadings of every drawn show coexisting pattern of elements, and scores of every drawn show the coexisting pattern of samples. What we focus on is the coexisting pattern of elements to disclose their migration mechanism. The clustering result of loadings shows similar and different patterns among elements and parameters. Therefore, the coexisting behavior of elements and parameters can be summarized. The clustering result of scores shows similar and different patterns among samples. Therefore, the coexisting behavior of samples, which means types of solid and liquid samples, can be summarized. The clustering method was based on the Gaussian mixture model. The GM model can cluster target reasonably. Comparing with K-means algorithm, the GM model does not divide the different group by stiff border but allows some mixture of different groups. So, the classifying probability for each group can be calculated.

*Leaching Mechanisms of Trace Elements from Coal and Host Rock Using Method of Data Mining DOI: http://dx.doi.org/10.5772/intechopen.100498*

We have applied software R as a tool, the packages psych and mclust were used to calculate PCA and GM model clustering results.
