Water source identification is important for water hazard controlling in coal mines. In this study, major ion concentrations of the groundwater collected from four representative aquifer systems in the Baishan coal mine, northern Anhui Province, China, have been analysed by a series of statistical methods. The results indicate that the major ion concentrations of the groundwater from different aquifer system are different with each other, and provided the possibility of water source identification based on hydrochemistry. Factor analysis indicates that these differences are controlled by different types of water rock interactions. The analysis based on US Environmental Protection Agency (EPA) Unmix model identified three sources (weathering of silicate minerals, dissolution of carbonate and evaporate minerals) responsible for the hydrochemical variations of the groundwater. Also, it shows that their contributions for the groundwater in different aquifer systems vary considerably. Based on these variations and on step by step analysis, the source aquifer system for the groundwater samples with unknown source has been determined and, similar to the result obtained by the cluster and discriminant analysis. Therefore, EPA Unmix model can be applied for water source identification in coal mine, as it can provide information about water rock interaction and water source identification simultaneously.

INTRODUCTION

Among the five most dangerous disasters (including water, fire, gas, dust and roof) in coal mine, water inrush has brought to human with highest loss, not only the loss of property, but most importantly, the death of people. In recent years, more than 100 accidents related to water inrush have occurred per year (Wu et al. 2013). During the last twenty years, more than 220 coal mines have been flooded, more than 8,000 persons died and more than 30 billion Yuan (¥RMB) have been lost in the accidents. Previous studies revealed that a large number of factors can affect the occurrence of water inrush: the geographical characteristics of the coal mines, the conditions of structure and hydrogeology, as well as the status of coal mining (Xu & Bu 2015).

In recent years, two kinds of methods have been applied for the controlling of water inrush: one is passive prevention, which focused on finding out the source of water after the water inrush; another one is positive prevention, which focused on forecasting the possibility of water inrush based on the comprehensive understanding of the hydrogeological conditions of the coal mine (Wu et al. 2013). Before or after water inrush, some of the information is contained in the groundwater characteristics, including the quantity, temperature and hydrochemistry etc. They have been used for identifying the source of inrush water, which is important for either management after water inrush or, prevention before water inrush. Based on the information, a large number of studies have been brought to light, and a series of methods have been reported, including quality-level-temperature method, hydrochemical method, mathematical method, GIS method and others (Jiang & Liang 2006; Sun et al. 2007; Chen et al. 2009; Zhang et al. 2009; Zhou et al. 2010; Huang & Chen 2011).

However, there are still some problems that have not been well documented, which are considered to be important in the future. Especially, the mechanism of water-rock interaction in the groundwater system has not been well understood, which is considered to be the basis for applying of the hydrochemistry of groundwater for water source identification (Sun & Gui 2012). If the information about water rock interaction (source of chemical constituents) and statistical analysis can be used together, it might be a better choice for solving the problem related to the source identification of inrush water in coal mine.

Therefore, in this study, a total of twenty-five groundwater samples have been collected from four representative aquifer systems in the Baishan coal mine, northern Anhui Province, China. Their major ion concentrations have been analyzed by the EPA Unmix model for identifying the source of water, which has previously been used for identifying the source of major ions in groundwater (Sun & Gui 2015a) and the wall rock compositions of the aquifer system (Sun 2015). The study can provide information for understanding the water rock interactions in the groundwater system as well as water source identification for water hazard controlling in the coal mine.

MATERIALS AND METHODS

Hydrogeological background

The coal field in northern Anhui Province is an important energy base in China. There are more than thirty coal mines located in the area and more than 100 million tons of coals have been exploited per year. The Baishan coal mine is located northwest to the Suzhou city and southwest to the Huaibei city, the distances are 35 and 15 km, respectively. Previous investigations revealed that the aquifer in the coal mine can be subdivided into four systems from shallow to deep: loose layer (LA), coal bearing sandstone (CA), Taiyuan Formation limestone (TA) and Ordovician limestone (OA). Their characteristics are as follows:

The LA is mainly composed of orange-deep yellow mudstone, sandstone and conglomerate (mostly composed of calcite), with thickness between 20 and 57 m. The depth of the aquifer system is 220–280 m below the surface. The CA is mainly composed of mudstone, siltstone, sandstone and coal seams, and the plagioclase is the main mineral. The thickness of the aquifer system varies significantly, from 37 to 128 m, and the depth is 280–400 m below the surface. In what concern TA and OA, they are mainly composed of thick layer of light gray limestone, the minerals are mainly calcite and dolomite, the thickness of these two aquifer systems are small than 136 m and depth between 530 and 680 m.

Previous investigations suggested that the surface water and the water in the CA are the direct threaten for the safety of coal mining in the study area. However, the water from the LA and TA can recharge the CA through the fault system or other channels. Therefore, during the management of water inrush in the coal mine, the water source identification related to the LA, CA and TA should be paid attention to.

Sampling and analysis

A total of twenty-five groundwater samples have been collected from four representative aquifer systems in the Baishan coal mine (Figure 1), five from the LA, CA and OA, respectively, and four from the TA through drainage holes with known source in the shafts. There are other six samples, which were collected from the drilling hole in the coal seam floor and their source aquifer have not been determined (maybe TA or OA according to their depth). Concentrations of eight major ions (Na+, K+, Ca2+, Mg2+, Cl, SO42−, HCO3 and CO32−) have been analyzed, and because of the low concentrations of K+, Na+ and K+ were merged and presented as (Na+ + K+). The analytical methods are as follows: Na+ + K+, Ca2+, Mg2+, Cl and SO42− were analyzed by ion chromatography, whereas HCO3 and CO32− were analyzed by acid–base titration in the Engineering and Technological Research Center of Coal Exploration, Anhui Province, China.
Figure 1

Location of the Baishan coal mine.

Figure 1

Location of the Baishan coal mine.

All of the data were firstly processed by Microsoft Excel for calculating the mean concentrations of the groundwater samples from each aquifer systems. Then, all of the data were analyzed by factor analysis (Mystat 12) for identifying the main sources of the major ions, method often used in environmental studies (Maiz et al. 2000; Liu et al. 2003; Shukla et al. 2006). Then, Q-mode cluster and discriminant analyses (Mystat 12) have been applied in order to allow the identification of the relationships between samples, and then for determination of the source aquifer of the unknown samples based on the relationships between samples (Panda et al. 2006; Shi & Xu 2010). Finally, the Unmix model has been applied for quantifying the source of major ions and to obtain the contributions of each source. Then the source aquifer of the unknown samples was determined according to the contribution ratios, because samples from similar source aquifer should have similar contributions from same host rocks.

RESULTS AND DISCUSSIONS

Concentrations of major ions

Range and mean concentrations of major ions in the groundwater samples from different aquifer systems are synthesized in Table 1. As can be seen from the table, the chemical compositions of the groundwater from the four aquifer systems vary considerably: the groundwater samples from the CA possess the highest mean concentrations of Na+ + K+ and HCO3 relative to other aquifer systems, whereas the groundwater samples from the TA and OA have relatively higher Ca2+, Mg2+, Cl, and SO42− concentrations, especially the OA. These differences suggest that the types and degrees of water rock interactions in these four aquifer systems are different. Moreover, the water samples with unknown sources are similar to the groundwater samples from the TA and OA, but different with those from the LA and CA.

Table 1

Range and mean concentrations (mg/l) of major ions in the groundwater samples from different aquifer systems

Aquifer Na+ + K+ Ca2+ Mg2+ Cl SO42− HCO3 CO32− 
LA (n = 5) 99.7–245 (208) 73.6–105 (96.4) 51.4–64.8 (60.2) 97.0–155 (141) 204–368 (329) 524–639 (559) 0–0 (0) 
CA (n = 5) 480–504 (494) 31.0–44.3 (37.6) 26.8–30.7 (28.7) 144–154 (147) 495–513 (505) 565–701 (642) 0–41.1 (15.5) 
TA (n = 4) 171–245 (197) 124–162 (137) 41.8–144 (74.9) 90.0–210 (136) 407–805 (569) 325–388 (359) 0–0 (0) 
OA (n = 5) 123–181 (158) 137–178 (168) 128–156 (147) 141–178 (168) 691–895 (843) 247–349 (300) 0–29.5 (5.90) 
UN (n = 6) 177–276 (212) 113–189 (166) 74.9–141 (120) 163–206 (192) 672–785 (760) 321–450 (397) 0–0 (0) 
Aquifer Na+ + K+ Ca2+ Mg2+ Cl SO42− HCO3 CO32− 
LA (n = 5) 99.7–245 (208) 73.6–105 (96.4) 51.4–64.8 (60.2) 97.0–155 (141) 204–368 (329) 524–639 (559) 0–0 (0) 
CA (n = 5) 480–504 (494) 31.0–44.3 (37.6) 26.8–30.7 (28.7) 144–154 (147) 495–513 (505) 565–701 (642) 0–41.1 (15.5) 
TA (n = 4) 171–245 (197) 124–162 (137) 41.8–144 (74.9) 90.0–210 (136) 407–805 (569) 325–388 (359) 0–0 (0) 
OA (n = 5) 123–181 (158) 137–178 (168) 128–156 (147) 141–178 (168) 691–895 (843) 247–349 (300) 0–29.5 (5.90) 
UN (n = 6) 177–276 (212) 113–189 (166) 74.9–141 (120) 163–206 (192) 672–785 (760) 321–450 (397) 0–0 (0) 

Note: LA, CA, TA and OA are the loose layer, coal bearing sandstone, Taiyuan limestone and Ordovician limestone aquifer systems, respectively.

Factor analysis

Two factors have been obtained based on the factor analysis with Eigen value higher than one after varimax rotation (Table 2). As can be seen from the table, the first factor, which accounts for 72.1% of the total variance explanation, has high positive loadings of Na+ + K+ and HCO3, whereas factor with 18.6% of the total variance explanation, has high positive loadings of Ca2+, Mg2+, Cl and SO42−. Based on the investigations and previous studies (Sun & Gui 2012; Wu et al. 2013), these two factors can be explained to be the weathering of silicate minerals (Factor 1) and dissolution of carbonates, sulfates and chlorides (Factor 2). This consideration is consistent with the characteristics of major ion concentrations of the groundwater from each aquifer (Table 1). The groundwater samples from the CA (sandstone dominant) show high Na+ + K+ and HCO3 concentrations, whereas the groundwater samples from the TA and OA with high SO42− and Cl might be an indication of high contents of evaporate minerals.

Table 2

Results of factor analysis

Species Factor 1 Factor 2 
Na+ + K+ 0.98 −0.05 
Ca2+ −0.75 0.46 
Mg2+ −0.57 0.65 
Cl 0.10 0.97 
SO42− −0.15 0.69 
HCO3 0.60 −0.12 
Eigen value 4.32 1.12 
Variance explained 72.1% 18.6% 
Species Factor 1 Factor 2 
Na+ + K+ 0.98 −0.05 
Ca2+ −0.75 0.46 
Mg2+ −0.57 0.65 
Cl 0.10 0.97 
SO42− −0.15 0.69 
HCO3 0.60 −0.12 
Eigen value 4.32 1.12 
Variance explained 72.1% 18.6% 

Cluster analysis

Hierarchical Q-mode cluster analysis has been applied to the data, and the ‘Ward’ linkage and the ‘Pearson’ distance have been chosen for calculation. The results are shown in Figure 2 as a dendrogram. As can be seen from the figure, two main groups and five subgroups can be identified: the Group 1 can be subdivided into two subgroups, including samples 1–5 (Group 1-1) and 6–10 (Group 1-2), which are consistent with the samples collected from the LA and CA, respectively. The Group 2 can be subdivided into three subgroups, including Group 2-1 (samples 11–13, 20), Group 2-2 (samples 14, 21–25) and Group 2-3 (samples 15–19). It can be found that all of the samples collected from the OA are classified to be Group 2-3, and the samples collected from the TA and unknown source aquifer system are classified to be Group 2-1 and 2-2. Such a classification indicates that the samples with unknown source are chemically similar to the samples collected from the OA and TA, especially the TA, as already observed based on descriptive statistics.
Figure 2

Results of Q-mode cluster analysis. Samples 1–5, 6–10, 11–14 and 15–19 are collected from the LA, CA, TA and OA, whereas samples 20–25 are with unknown sources.

Figure 2

Results of Q-mode cluster analysis. Samples 1–5, 6–10, 11–14 and 15–19 are collected from the LA, CA, TA and OA, whereas samples 20–25 are with unknown sources.

Discriminant analysis

Discriminant analysis provides linear or quadratic functions of the variables that ‘best’ separate cases into two or more predefined groups. In this study, the groundwater samples collected from the LA, CA, TA and OA are signed to be 1, 2, 3, and 4, respectively, whereas the samples with unknown source are signed to be ‘blank’. The classical method of Mystat (version 12) has been chosen for calculation and the results are shown in Table 3.

Table 3

Results of discriminant analysis

ID Source Predicted Distance 1 Distance 2 Distance 3 Distance 4 
S1 LA 10 309 368 1,014 
S2 LA 11 336 380 1,050 
S3 LA 310 403 1,103 
S4 LA 324 424 1,138 
S5 LA 293 374 1,050 
S6 CA 301 288 757 
S7 CA 295 290 768 
S8 CA 347 336 803 
S9 CA 311 358 866 
S10 CA 303 299 776 
S11 TA 458 362 153 
S12 TA 404 244 194 
S13 TA 329 331 236 
S14 TA 379 344 11 181 
S15 OA 978 724 156 11 
S16 OA 1,095 828 195 
S17 OA 1,178 881 230 
S18 OA 1,065 786 182 
S19 OA 1,031 759 170 
S20 UN 548 253 45 169 
S21 UN 389 324 184 
S22 UN 343 311 209 
S23 UN 308 383 31 256 
S24 UN 207 353 69 400 
S25 UN 328 396 30 249 
ID Source Predicted Distance 1 Distance 2 Distance 3 Distance 4 
S1 LA 10 309 368 1,014 
S2 LA 11 336 380 1,050 
S3 LA 310 403 1,103 
S4 LA 324 424 1,138 
S5 LA 293 374 1,050 
S6 CA 301 288 757 
S7 CA 295 290 768 
S8 CA 347 336 803 
S9 CA 311 358 866 
S10 CA 303 299 776 
S11 TA 458 362 153 
S12 TA 404 244 194 
S13 TA 329 331 236 
S14 TA 379 344 11 181 
S15 OA 978 724 156 11 
S16 OA 1,095 828 195 
S17 OA 1,178 881 230 
S18 OA 1,065 786 182 
S19 OA 1,031 759 170 
S20 UN 548 253 45 169 
S21 UN 389 324 184 
S22 UN 343 311 209 
S23 UN 308 383 31 256 
S24 UN 207 353 69 400 
S25 UN 328 396 30 249 

Note: Distance 1–4 are distances of each samples to each cluster.

As can be seen from Table 3, the discrimination of all of the samples collected from the known aquifer systems (LA, CA, TA and OA) is corrected to be 100%: samples 1–5, 6–10, 11–14 and 15–19 have the lowest distance value to group 1 (LA), 2 (CA), 3 (TA) and 4 (OA), respectively, which suggests that the efficiency of the discrimination is valid. Based on the discrimination model established by the known samples, all of the groundwater samples with unknown source are classified to be 3 (TA), implying that the TA is the source aquifer system for them, similar to the result obtained by cluster analysis that the unknown source samples are clustered together with the TA samples.

Unmix model analysis

Based on the calculation of UNMIX model, three sources have been identified and the results are listed in Tables 4 and 5, and shown in Figures 3 and 4. These three sources have Min Rsq = 0.99 and Min Sig/Noise = 2.53, higher than the minimum requirement of the model (Min Rsq > 0.8 and Min Sig/Noise > 2). It can also be obtained from the relationships between predicted and observed values of the major ions, that the r2 values for Na+ + K+, Ca2+, Mg2+, Cl, SO42− and HCO3 are 1.00, 0.95, 0.97, 0.83, 0.98 and 0.98, respectively (Figure 3), suggesting that the modeling is efficient (Ai et al. 2014).
Table 4

Source compositions (Sn in mg/l)

Species Source 1 Source 2 Source 3 
Na+ + K+ 168 30.3 56.2 
Ca2+ 12.4 33.3 76.1 
Mg2+ 8.64 18.1 61.7 
Cl 49.1 35.4 75.2 
SO42− 169 83.1 355 
HCO3 209 137 108 
Species Source 1 Source 2 Source 3 
Na+ + K+ 168 30.3 56.2 
Ca2+ 12.4 33.3 76.1 
Mg2+ 8.64 18.1 61.7 
Cl 49.1 35.4 75.2 
SO42− 169 83.1 355 
HCO3 209 137 108 
Table 5

Portions of the sources (Xn)

ID Source aquifer Source 1 Source 2 Source 3 
S1 LA 0.831 3.609 −0.605 
S2 LA 0.104 3.707 −0.058 
S3 LA 0.907 2.508 0.103 
S4 LA 0.928 2.611 0.087 
S5 LA 0.909 2.610 0.067 
S6 CA 2.955 0.089 0.039 
S7 CA 2.836 −0.229 0.169 
S8 CA 2.971 0.498 −0.170 
S9 CA 2.957 0.606 −0.221 
S10 CA 2.951 −0.047 0.047 
S11 TA 0.520 1.326 0.616 
S12 TA 0.990 0.576 1.001 
S13 TA 0.447 2.067 0.299 
S14 TA 0.443 0.302 2.090 
S15 OA 0.106 0.109 1.878 
S16 OA 0.131 0.046 2.345 
S17 OA 0.172 −0.558 2.570 
S18 OA 0.267 0.062 2.275 
S19 OA 0.300 0.061 2.333 
S20 UN 1.283 −0.484 1.360 
S21 UN 0.577 0.619 1.757 
S22 UN 0.505 0.636 1.802 
S23 UN 0.219 1.481 1.751 
S24 UN 0.398 1.511 1.678 
S25 UN 0.293 1.284 1.788 
ID Source aquifer Source 1 Source 2 Source 3 
S1 LA 0.831 3.609 −0.605 
S2 LA 0.104 3.707 −0.058 
S3 LA 0.907 2.508 0.103 
S4 LA 0.928 2.611 0.087 
S5 LA 0.909 2.610 0.067 
S6 CA 2.955 0.089 0.039 
S7 CA 2.836 −0.229 0.169 
S8 CA 2.971 0.498 −0.170 
S9 CA 2.957 0.606 −0.221 
S10 CA 2.951 −0.047 0.047 
S11 TA 0.520 1.326 0.616 
S12 TA 0.990 0.576 1.001 
S13 TA 0.447 2.067 0.299 
S14 TA 0.443 0.302 2.090 
S15 OA 0.106 0.109 1.878 
S16 OA 0.131 0.046 2.345 
S17 OA 0.172 −0.558 2.570 
S18 OA 0.267 0.062 2.275 
S19 OA 0.300 0.061 2.333 
S20 UN 1.283 −0.484 1.360 
S21 UN 0.577 0.619 1.757 
S22 UN 0.505 0.636 1.802 
S23 UN 0.219 1.481 1.751 
S24 UN 0.398 1.511 1.678 
S25 UN 0.293 1.284 1.788 

Note: Xn (in this table) × Sn (in Tables 4) is the contribution of source n in mg/l.

Figure 3

Predicted versus measured concentrations (mg/l).

Figure 3

Predicted versus measured concentrations (mg/l).

Figure 4

Variations of source contributions ((Xn × Sn)/∑(Xn × Sn)). Samples 1–5, 6–10, 11–14 and 15–19 are collected from the LA, CA, TA and OA, whereas samples 20–25 are with unknown sources. Take sample 1 as an example, its chemical composition is contributed by 20% of source 1 and 80% of source 2.

Figure 4

Variations of source contributions ((Xn × Sn)/∑(Xn × Sn)). Samples 1–5, 6–10, 11–14 and 15–19 are collected from the LA, CA, TA and OA, whereas samples 20–25 are with unknown sources. Take sample 1 as an example, its chemical composition is contributed by 20% of source 1 and 80% of source 2.

The relationship between the hydrochemistry of the groundwater and the sources are as follows: 
formula
Cm is the measured concentration of groundwater; Sn is the source concentration (Table 4); Xn is the portion of source n (Table 5), and (Xn × Sn)/∑(Xn × Sn) is the contribution of source n for the Cm (Figure 4).

Weathering of silicate minerals, dissolution of carbonate and evaporate minerals are the main processes controlling the groundwater chemistry in the coal mining related strata (Sun 2015; Sun & Gui 2015a). Therefore, the Source 1 dominated by Na+ + K+, and HCO3 should be explained to be the weathering of silicate minerals, and the Source 3 dominated by Ca2+, Mg2+, Cl, and SO42− can be explained to be the dissolution of evaporate minerals (sulfates and chlorides).

Such considerations can be demonstrated by the contributions of these sources (Figure 4), as the samples from the CA have highest contributions from the Source 1, consistent with the truth that this aquifer system is rich in plagioclase sandstone. Samples from the OA have highest contributions from the Source 3, which is also consistent with the previous investigations that the rocks in the Ordovician strata are mainly composed of calcite, dolomite and gypsum.

The Source 2 should be considered to be the carbonate source, because it cannot be explained to be silicate and /or evaporate sources relative to the Source 1 and 3, since the weathering of silicate will generate high Na+ + K+ and HCO3 simultaneously (Source 1), whereas the high Cl and SO42− concentrations can be generated by dissolution of evaporate minerals (Source 3) rather than carbonate minerals.

However, this consideration is also ambiguous because: (1) the Source 2 shows only medium contributions of Ca2+, Mg2+, and HCO3, which is not consistent with the characteristics generated by the dissolution of carbonates; (2) it is not consistent with the truth that the TA and OA have more carbonate minerals than the other two aquifer systems (LA and CA), whereas the groundwater samples collected from the LA possess the highest contributions from this source.

In consideration with the hydrogeological conditions of the study area, the high contributions of the Source 2 (dissolution of carbonate minerals) for the LA samples may be related to the high contents of carbonate conglomerate in the strata, whereas the low contributions of this source for the TA and OA samples might be replaced by the dissolution of evaporate minerals, because evaporate minerals have much higher solubility than the carbonates in similar conditions.

Implications for water source identification

As mentioned in the introduction, several methods have been applied for water source identification in coal mines (Jiang & Liang 2006; Sun et al. 2007; Chen et al. 2009; Zhang et al. 2009; Zhou et al. 2010; Huang & Chen 2011). However, hydrochemical analysis is the most popular and reliable method, because water flow over the rocks can inherit the chemical signature of the rocks, and the rocks in the different aquifer systems are different. In this study, the main rocks in the LA, CA, TA and OA are different with each other, which therefore, provided the possibility of water source identification based on water hydrochemistry.

The concentrations of major ions in the groundwater samples firstly revealed that they have undergone different types and degrees of water rock interactions in their aquifer systems. Factor and cluster analysis confirmed that the groundwater samples collected from the LA, CA, TA and OA have undergone different types and degrees of water rock interactions. The source aquifer system of the unknown samples has been identified to be the TA based on discriminant analysis with high extent of probability. However, can similar result be obtained from the EPA Unmix model?

Here the analysis is focused on the source contributions (Figure 4), and the analysis is step by step (see also in Figure 5(a) and 5(b)):
  • Step 1: according to the contributions of the Source 1 (silicate weathering), the CA samples have high contributions from this source, which is different with those of the samples from the LA, TA, OA and unknown source. Therefore, CA is not considered to be the source for the unknown samples, and the probability sources are LA, TA and OA.

  • Step 2: according to the contributions of the Source 2 (carbonate dissolution), the OA samples have low contributions from this source, different with those of the samples from the LA, TA and unknown source. Therefore, OA is not considered to be the source for the unknown samples, and the probability sources are LA and TA.

  • Step 3: according to the contributions of the Source 3 (evaporite dissolution), the LA samples have low contributions from this source, different with those of the samples from the TA and unknown source. Therefore, LA can be ruled out, and TA is considered to be source for the unknown source samples.

Figure 5

Step by step water source identification. UN–unknown source.

Figure 5

Step by step water source identification. UN–unknown source.

No matter which source considered firstly, the results are similar to each other (see in Figure 5(a) and 5(b)). Moreover, the results are also consistent with the results obtained by statistical analyses (including cluster and discriminant analyses), which suggests that the method is applicable and believable.

Moreover, although the statistical methods have long been used for water source identification in coal mines (Sun 2014; Ma et al. 2014; Sun & Gui 2015b), which have played important roles for water hazards controlling and management. However, the water source discrimination model established for one coal mine based on statistical differences cannot be migrated to another coal mine easily, because of the lack of the understanding about the water rock interactions. The method with EPA Unmix model has advantages relatively to previous statistical methods, because it considers not only the statistical similarity, but also the wall rock compositions as revealed by hydrochemistry.

CONCLUSIONS

Based on the analyses of major ion concentrations of groundwater samples collected from four representative aquifer systems in the Baishan coal mine, northern Anhui Province, China, the following conclusions have been obtained:

  1. The major ion concentrations of the groundwater samples are different with each other, and then provided the possibility of water source identification.

  2. Factor analysis suggests that weathering of silicate minerals and dissolution of carbonates, sulfates and chlorides are the main factors controlling the groundwater chemistry.

  3. Cluster analysis and discriminant analysis indicate that the TA is the source aquifer system for the unknown source samples.

  4. The analysis based on EPA Unmix model identified three sources (weathering of silicate minerals, dissolution of carbonate and evaporate minerals) responsible for the hydrochemistry of the groundwater, and show that their contributions for the groundwater in different aquifer systems vary considerably.

  5. Based on the variations of contributions of each source and step by step analysis, the TA is determined to be the source aquifer system for the groundwater samples with unknown source.

  6. EPA Unmix model can be applied for water source identification in coal mine, as it can provide information about water rock interaction and water source identification simultaneously.

ACKNOWLEDGEMENTS

This work was financially supported by National Natural Science Foundation of China (41302274) and the Foundation of Scholarship Leaders in Suzhou University (2014XJXS05).

REFERENCES

REFERENCES
Ai
J. C.
Wang
N.
Yang
J.
2014
Source apportionment of soil heavy metals in Jiapigou goldmine based on the UNMIX model
.
Environmental Science
35
(
9
),
3530
3536
.
Chen
H. J.
Li
X. B.
Liu
A. H.
2009
Studies of water source determination method of mine water inrush based on Bayes’ multi-group stepwise discriminant analysis theory
.
Rock and Soil Mechanics
30
(
12
),
3655
3659
.
Huang
P. H.
Chen
J. S.
2011
Fisher identify and mixing model based on multivariate statistical analysis of mine water inrush sources
.
Journal of China Coal Society
36
(
Sup 1
),
131
136
.
Jiang
A. N.
Liang
B.
2006
The particle swarm optimization support vectors machine method of identifying standard components of ions of groundwater
.
Journal of China Coal Society
31
(
3
),
310
313
.
Ma
L.
Liu
Y.
Zhou
X. P.
2014
Fuzzy comprehensive evaluation method of F statistics weighting in identifying mine water inrush source
.
International Journal of Engineering, Science and Technology
2
(
7
),
123
128
.
Shi
L.
Xu
L. Y.
2010
Prediction of mine water inrush sources based on cluster analysis of hydrogeochemical features
.
Coal Science and Technology
38
(
3
),
97
100
.
Shukla
M. K.
Lal
R.
Ebinger
M.
2006
Determining soil quality indicators by factor analysis
.
Soil and Tillage Research
87
(
2
),
194
204
.
Sun
L. H.
Gui
H. R.
2015a
Source Quantification of major ions in groundwater and hydrological implications: Liuqiao case study
.
Electronic Journal of Geotechnical Engineering
20
(
5
),
1791
1800
.
Sun
Y. J.
Yang
G. Y.
Zheng
L.
2007
Distinguishing system study on resource of mine water inrush based on GIS
.
Coal Geology & Exploration
35
(
2
),
34
37
.
Wu
Q.
Zhao
S. Q.
Sun
W. J.
Cui
F. P.
Wu
C.
2013
Classification of the hydrogeological type of coal mine and analysis of its characteristics in China
.
Journal of China Coal Society
38
(
6
),
901
905
.
Xu
H.
Bu
W.
2015
Analysis of water-inrush for deep coal floor in a coal mine
.
Electronic Journal of Geotechnical Engineering
20
(
10
),
4189
4196
.
Zhang
R. G.
Qian
J. Z.
Ma
L.
Qin
H.
2009
Application of extension identification method in mine water-bursting source discrimination
.
Journal of China Coal Society
34
(
1
),
33
38
.
Zhou
J.
Shi
X. Z.
Wang
H. Y.
2010
Water-bursting source determination of mine based on distance discriminant analysis model
.
Journal of China Coal Society
35
(
2
),
278
282
.