Groundwater vulnerability assessment in degraded coal mining areas using the AHP–Modified DRASTIC model

1 INTRODUCTION

Coal is the largest source of primary energy and is a major economic contributor especially in developing countries, as other industrial setup such as steel plants and thermal power plants depend on it. But the process of coal mining critically damages the regional environment and is the precursor to land degradation (Mukhopadhyay, George, & Masto, 2016). Despite existence of numerous strategies to control and extenuate the impacts of land degradation, successful implementation of these strategies is challenging (Lechner, Baumgartl, Matthew, & Glenn, 2014). Surface and underground coal mining activities degrade the land permanently and severely affect the ecosystem including the hydrological processes at scales ranging from small plots to large watersheds (Eshleman, 2004; Machowski, Rzetala, Rzetala, & Solarski, 2016). Removal of vegetation and topsoil enhances the process of withering and contaminant transport through the degraded areas. The risk of coal mining is not only limited to surface environment, but it also alters the regional water quality by the action of contaminant movement through induced and natural fractures. Groundwater is a major source of irrigation and drinking water and the largest store of unfrozen freshwater. The dependence of humans on groundwater is more because of its accessibility and less vulnerability to quality degradation (Aeschbach-Hertig & Gleeson, 2012). It is reported that about 660 million people lack access to safe drinking water globally (Human Development Report, 2015); and another 1.6 billion people face economic water shortage, especially in developing countries because of the absence of necessary infrastructure for water purification and distribution (United Nations-Water, 2007). Thus, protection of water resources in these regions requires open-ended retrospection of water policies at all levels.

Groundwater vulnerability assessment (GVA) mapping is critical in developing strategies for effective protection and management of groundwater quality. GVA is based on the aggregation of hydrogeological factors controlling the transport of contaminants to the groundwater (Sener & Davraz, 2013). The most widely used and accepted model for GVA is DRASTIC (Aller, Lehr, Petty, & Bennett, 1987), which is a standardised system to evaluate groundwater pollution potential. Since its inception, DRASTIC has effectively been modified and used in several mapping studies (Hamza et al., 2015; Huan, Wang, & Teng, 2012). DRASTIC is based on the concept of hydrogeological settings, which includes all the major geologic and hydrogeological factors ([D] depth of water, [R] net recharge, [A] aquifer media, [S] soil media, [T] topography, [I] impact of vadose zone, and [C] hydraulic conductivity of aquifer) affecting the groundwater movement into, through, and out of an area. It also includes a scheme for relative ranking of these parameters to forecast the comparative impact of groundwater contamination potential of any of the hydro-geologic settings (Aller et al., 1987). Although, DRASTIC has been implemented successfully in many studies, but the method is widely criticised for its subjectivity in assigning numerical ratings to the parameters (Babiker, Mohamed, Hiyama, & Kato, 2005). Apart from this, the seven hydrogeological parameters of DRASTIC model neglect the regional characteristics (Neshat, Pradhan, Pirasteh, & Shafri, 2014). On the other hand, the advantage of using DRASTIC model is that, it is flexible to the changes in parameters as per the region specific requirements of different studies. However, availability of data for a sector specific implementation of DRASTIC model can be a major limitation in multicriteria GVA studies. Furthermore, addition of factors such as land use may further improve the model in predicting groundwater contamination as a result of anthropogenic disturbances (Pacheco & Sanches Fernandes, 2013). Thus, identification and optimisation of parameters, which could contribute contaminants towards groundwater, become critical for GVA. Several studies have reported the efficacy of modified DRASTIC scheme for GVA. A detailed comparison of recent literature on GVA with advantages and critical observations has been presented in Table 1. DRASTIC model has been suitably modified to assess the groundwater vulnerability using several different techniques for multiple studies globally. However, for coal mining areas, only a few studies have been reported, especially on the use of DRASTIC model for GVA. A. K. Tiwari, Singh, and DeMaio (2016) used the DRASTIC model in West Bokaro coalfield of India. They reported that a large percentage of their study area lied in moderate to high vulnerability zone, which was mostly because of geogenic and anthropogenic activities. Bukowski, Bromek, and Augustyniak (2006) used the DRASTIC system in the Upper Silesian Coal Basin of Poland to assess the vulnerability of groundwater. They reported that the impact of mining and its influence on the individual hydrogeological components must be considered for development of an accurate vulnerability map. They concluded that several DRASTIC parameters need to be modified to consider mine specific factors in order to accurately map the vulnerability in mining areas. Globally, in large coalfields, opencast mines dominate the landscapes, as a result, the aquifers get broken permanently, and once the mining operations cease, there is a huge pit at the bottom of the mines in locations where backfilling is not done. This leads to the mixing of surface water and groundwater, and the contaminants released from mining activities may get transported into the groundwater. The geographic information system (GIS)-based GVA models highlighted in literature do not consider these characteristics for a site specific study in coal mining areas and hence may not be suitable for assessment of vulnerability in these areas. Thus, there is a requirement for developing a suitable model for improving the forecasting accuracy by optimising the existing GVA practices.

In this study, the DRASTIC model was optimised for efficient estimation of GVA by parameter moderation and analytic hierarchy process (AHP) techniques. Initially, the existing DRASTIC model was used to evaluate the vulnerability of groundwater using original parameter weights and rates. Then, study specific parameters were added to the existing DRASTIC model to evaluate the change in GVA. Finally, AHP was used to optimise the weights and rates of all the parameters including the additional ones to remove redundancy and to improve the forecasting accuracy. Subsurface forecasting and visualisation are challenging tasks, especially when it comes to modelling of contaminant transport into groundwater. Additionally, governments have mandated continuous environmental monitoring in environmentally sensitive zones such as coal mining areas. Groundwater sampling and analysis for monitoring quality can be costly, because the number of samples required to establish the background concentration is very high. This study will help in feasible groundwater sampling design by reducing the number of samples required for establishing the background concentrations of different contaminants. Especially for coal mining areas, there is a need to develop a simplified model for vulnerability assessment by categorising each of the specific group of activities together for overall improvement of interpretation and clarity of the models. The flowchart of the methodology for the present study is presented in Figure 1.

2 MATERIAL AND METHODS

2.1 Study area

Jharia Coalfield (JCF) is situated in Dhanbad District of Jharkhand State, India (Figure 2). Dhanbad district supports 2.6 million inhabitants and has the highest population density of 1,316 persons per km² out of all the districts in Jharkhand (Registrar General-India, 2011). JCF is known for being the exclusive storehouse of prime coking coal in the country, and it lies in the heart of the Damodar River valley. JCF lies between the latitudes 23° 39′ N to 23° 50 N and longitudes 86° 05′ E and 86° 30′ E. Coal mining and increased population near the mines have affected the water resources badly in the region, with inadequate supply of potable water especially during lean season. Thus, majority of the population depends on groundwater. The present study area covers about 355 km² and features undulating profile with very low rolling slope in the eastern part.

2.2 Data collection

2.2.1 Depth of groundwater (D)

Groundwater water table data were collected in the month of August 2017 from 51 dug wells spread across the study area (Figure 2) using a sensor-based water level recorder. The groundwater table data were reclassified in the subparameter ranges according to Neshat, Pradhan, and Dadras (2014; Table 2). The depth of groundwater in the study area varied from 0.5 to 9.3 m below ground surface. For the present study, D can be considered as a static layer, which may only act as a destination layer of the proposed model. The highest rates were assigned to the lower ranges, as these areas may be more prone to contamination considering the minimum time required to reach the groundwater table. The interpolation of the data was carried out using Kriging interpolation technique; the data were reclassified using the reclassification tool of ArcGIS 10.2.2 (Figure 3a).

Table 2. Parameters, subparameters, ratings, and weights

		Modified DRASTIC technique		AHP–Modified DRASTIC
Parameter	Subparameter	Weights	Ratings	Weights	Ratings
Depth of groundwater, (m) D	0–1.5	5	10	0.201	0.567
	1.5–4.6		9		0.279
	4.6–9.1		7		0.104
	>9.1		3		0.050
Net recharge, (mm/year) R	0–50	4	1	0.110	0.048
	50–100		3		0.115
	100–177		6		0.266
	>177		8		0.571
Aquifer media, A	Metamorphic gneiss and schist	3	5	0.069	0.071
	Coarse and medium sandstone and grit shale		6		0.141
	Fine grain sandstone		8		0.418
	Igneous aquifers, granite gneiss		7		0.306
	Gneiss shale and sandstone		3		0.064
Soil media, S	Poorly drained	2	6	0.037	0.633
	Moderately drained		5		0.260
	Well drained		4		0.106
Topography, (Slope°) T	0–2	1	10	0.031	0.469
	2–6		9		0.290
	6–12		5		0.134
	12–18		3		0.074
	>18		1		0.033
Impact of vadose zone, I	8–10	5	10	0.187	0.490
	6–8		8		0.285
	4–6		5		0.114
	3–4		3		0.070
	2–3		1		0.042
Hydraulic conductivity, (m/s) C	<10−7	3	2	0.073	0.043
	10−6–10−7		4		0.071
	10−5–10−6		6		0.124
	10−4–10−5		8		0.278
	10−3–10−4		10		0.484
Land use, Lu	Barren land/dense vegetation/middense vegetation	5	2	0.202	0.069
	Built up		1		0.044
	Coal quarry/OB dump		10		0.595
	Sand		3		0.104
	Surface water		5		0.188
Distance from lineament, L (m)	0–250	3	9	0.089	0.491
	250–500		7		0.279
	500–750		5		0.114
	750–1,000		3		0.071
	>1,000		1		0.045

• Note. AHP = analytic hierarchy process; DRASTIC = depth of groundwater, net recharge, aquifer media, soil media, topography, impact of vadose zone, and hydraulic conductivity; OB = overburden.

2.2.2 Net recharge (R)

The net recharge (R) map (Figure 3b) was generated from the method presented in Piscopo (2001); Equation 1 was used to calculate R, which is the ability of an area to act as a natural recharge zone relative to another area. The factors used to generate the recharge map include slope of the study area (extracted from Digital Elevation Model [DEM]), average annual rainfall (monthly data were collected from India Meteorological Department), and soil permeability (obtained from State Agriculture Management & Extension Training Institute, Jharkhand, Department of Agriculture and Cane Development, Government of Jharkhand). The R values in the study area varied from a minimum of 5 mm/year to a maximum of 204 mm/year. The final recharge map was created by adding the selected factors using the raster calculator tool of ArcGIS 10.2.2. The map was reclassified in the subparameter ranges according to Neshat, Pradhan, and Dadras (2014; Table 2).

(1)

2.2.3 Aquifer media (A)

The geological map and other geological data of the study area were procured from Geological Survey of India, Kolkata. The study area is divided into five major geological formations with several cracks and deformations in between. The five geologic formations are Barren Measures (fine grain sandstone), Barakar Beds (coarse and medium sandstone and grit shale), Talchir Formation (gneiss shale and sandstone), Archean Gneiss (metamorphic gneiss and schist), and Raniganj Stage (igneous aquifers and granite gneiss). The geological map was hand digitised in ArcGIS environment, converted into raster format, and reclassified according to their relative rates (consistent with Ghosh, Tiwari, & Das, 2015; Table 2; Figure 3c).

2.2.4 Soil media (S)

Soil in the study area is distributed in three major classes: (a) well drained, was assigned the highest rate, as areas with these soils will be more prone to infiltration; (b) moderately drained soil with slightly lesser infiltration capacity; and (c) poorly drained soil with very low infiltration capacity (Karan & Samadder, 2016; Table 2). The soil map was prepared from the State Agriculture Management & Extension Training Institute dataset, which was converted into raster format and reclassified as per their relative rates using ArcGIS 10.2.2 (Figure 3d).

2.2.5 Topography (T)

The topographical features of the study area were extracted from Cartosat-1 DEM of 30-m resolution. The DEM was standardised to Universal Transverse Mercator zone 45 N Datum WGS 1984 projection using ArcGIS 10.2.2; percentage slope was derived from the DEM using ERDAS Imagine 2014 (Figure 3e). The slope in the study area varied from near-level slope (0°) to steep slope (27°). The slope in the study area was divided into five categories as per Aller et al. (1987) and was reclassified using ArcGIS 10.2.2 (Table 2). The nearly plain areas were assigned maximum rate as those areas are expected to be more prone to contaminant infiltration because of less surface runoff.

2.2.6 Impact of vadose zone (I)

The factor I is one of the important parameters affecting groundwater contamination, as it can act as a potential barrier to reduce contamination by the process of absorption and adsorption depending upon the media. I was computed as per the method presented by Piscopo (2001). Impact of vadose zone was calculated using Equation 2.

(2)

The impact of vadose zone map (Figure 3f) was prepared by adding the two layers in a GIS module. The generated map was reclassified in the subparameter ranges as per Piscopo (2001; Table 2).

2.2.7 Hydraulic conductivity (C)

Hydraulic conductivity of an aquifer is its ability to transmit water, hence, C can be an important factor in transporting dissolved contaminants towards groundwater. The data of C were collected from environmental impact assessment and environmental management plans of different cluster of mines available at www.environmentclearance.nic.in. The values of C in the study area varied from a maximum of 0.17 × 10⁻⁴ m/s to a minimum of 0.06 × 10⁻⁷ m/s. Areas having higher values of C are capable of contributing substantially more contaminants quickly than areas having lesser C values. The data of C were geocoded and imported in ArcMap and were projected to Universal Transverse Mercator WGS 1984 datum, Zone 45 N. The imported data were then interpolated using inverse distance weighted method. The interpolated map (Figure 3g) was reclassified according to their relative risk of contaminating groundwater following Sener and Davraz (2013; Table 2).

2.2.8 Land use (Lu)

Support vector machine-based supervised land-use classification was performed on Sentinel-2 data of June 2017 using ENVI 5.2 by selecting appropriate number of training samples. Eight land-use classes were demarcated (Table 2), and a relative rate was assigned to each of the classes with respect to their virtual influences as a source of groundwater contamination. Assigning weights for Lu is consistent with Sener and Davraz (2013). Confusion matrix-based pseudo accuracy assessment was done using the same Sentinel-2 image by taking random test data. The overall classification accuracy was observed as 93.40% with a kappa coefficient of 0.91. The classified image (Figure 3h) was imported into ArcGIS 10.2.2, and each land-use class was reclassified according to their respective rates. As per the ranking scheme of the present study, mining areas (which includes OB Dump and Coal Quarry) were assigned the highest rate due to their high relative impact on groundwater vulnerability.

2.2.9 Distance from lineament (L)

According to Lee (2003), lineaments are the numerous linear features on the ground surface, related to the geologic development, and they are closely related to groundwater flow and contaminant transport. They may be geological structures such as faults, joints, and line weaknesses and geomorphological features such as cliffs, terraces, or linear valleys. In order to extract the lineaments, the first principal component (PC1) of the Landsat 8 pan-sharpened reflected band was used, as the PC1 carries most of the information. ENVI 5.2 was used to extract the PC1 of the Landsat 8 image. The PC1 was then subjected to lineament extraction algorithm available in the trial version of PCI Geomatica 2016 (PCI Geomatics, 2016). The extracted lineaments were saved as vector files, and ring buffer was created around them using the multiple ring buffer tool available in ArcGIS 10.2.2. The buffer file was converted into raster format and was reclassified as per their respective rates (Figure 3i; consistent with Sener & Davraz, 2013; Table 2).

2.3 GIS-based GVA

2.3.1 GVA using DRASTIC

In this step, the existing GIS-based DRASTIC technique was used to estimate the groundwater vulnerability of the study area using Equation 3.

(3)

where DVI is the DRASTIC Vulnerability Index, n is the number of parameters; r and w in subscripts of the parameters represent their rates and weights, respectively. After expanding Equation 3, it becomes

(4)

In Equation 4, D is the depth of groundwater, R is the net recharge, A is the aquifer media, S is the soil media, T is topography (characterised by slope), I is impact of vadose zone, and C is the hydraulic conductivity. The typical weights and rates of these parameters are presented in Table 2. The reclassified parameters were scaled to a resolution of 30 m to maintain consistency in the overlay technique, and then the parameters were added to form a composite DVI map of the study area using raster calculator available in map algebra toolkit of ArcGIS 10.2.2.

2.3.2 GVA–Modified DRASTIC model

A Source ➝ Vector ➝ Destination model based on the concept proposed by Ansell and Wharton (1992) was developed in conjunction with the existing DRASTIC technique to study the groundwater vulnerability due to coal mining. For the present study, coal mining and its related activities such as coal quarry and overburden dump handling were considered as the source of groundwater contamination. The vector or pathway for the contaminants released from these activities can be the existing DRASTIC parameters, and the destination or receptor will be the groundwater. Therefore, the existing DRASTIC model was modified as Equation 5;

(5)

where Lu is land use, and L is the distance from lineaments, and rest of the parameters are same as the existing DRASTIC model.

Single parameter sensitivity analysis

The rationale of sensitivity study is to evaluate the contribution of individual values and of the selected parameters on the output of the proposed analytical model (Napolitano & Fabbri, 1996). It also provides a valuable insight on the influence of rates and weights assigned to each of the parameters that further assists the analyst in arbitrating the importance of subjectivity elements (Huan et al., 2012; Pacheco, Pires, Santos, & Fernandes, 2015). The effects of the additional parameters on the model results were determined using single parameter sensitivity analysis as per Equation 6, which was used to check their spatial significance. This is performed by comparing the original weights with the effective weights by setting the original weights to higher or lower values ignoring the spatial distribution of corresponding rates (Sener & Davraz, 2013).

(6)

where W is the effective weight of the parameter, P_r is the rate of the parameter, P_w is the weight of the parameter, and DVI is the DRASTIC vulnerability index.

2.3.3 AHP–Modified DRASTIC technique for estimation of groundwater contamination

Several studies have ignored the insignificant parameters and added new ones as per the requirement (Babiker et al., 2005). Parameter selection and addition may be important to sector specific studies, but the selection of optimal weights and rates of the selected parameters is often ignored. This becomes even more vital when the selected parameters are related with each other, such as in the case of depth of groundwater and impact of vadose zone or slope and net recharge. Hence, removal of redundancy among factor weights and rates becomes critical in efficient model forecasting. For the present study, AHP was used to determine the rating coefficients of each parameter for the modified DRASTIC model. AHP serves as a powerful decision-making tool involving complex interrelated multicriteria decision problem (Chuang, 2001). Typically, in AHP, a 9-point scale ranging from 1 (equal importance) to 9 (absolute importance) is used to evaluate the off-diagonal relationship between components (Sener & Davraz, 2013). Rezaei-Moghaddam and Karami (2008) reported that the pairwise comparison helps to evaluate the contribution of each component of the problem more independently. For pairwise comparison of the parameters, appropriate weights were assigned based on the relative importance of the selected parameters (Table 3). The diagonal elements were kept 1. The consistency of the AHP matrix of the Order 9 including all the existing DRASTIC parameters and the two additional ones were evaluated by computing a consistency index (CI) as per Equation 7.

(7)

Table 3. Analytic hierarchy process matrix for the specific groundwater vulnerability assessment for calculating weights of each parameter (consistency ratio: 0.0303)

	Groundwater depth	Net recharge	Aquifer media	Soil media	Topography	Impact of vadose zone	Hydraulic conductivity	Land use	Lineament
Groundwater depth	1	3	4	4	5	1	3	1	2
Net recharge	1/3	1	2	3	3	1/2	2	1/2	2
Aquifer media	1/4	1/2	1	3	3	1/4	1	1/4	1
Soil media	1/4	1/3	1/3	1	2	1/5	1/3	1/5	1/3
Topography	1/5	1/3	1/3	1/2	1	1/5	1/3	1/5	1/3
Impact of vadose zone	1	2	4	5	5	1	2	1	2
Hydraulic conductivity	1/3	1/2	1	3	3	1/2	1	1/3	1/2
Land use	1	2	4	5	5	1	3	1	3
Lineament	1/2	1/2	1	3	3	1/2	2	1/3	1

where λ_max is the largest principal eigenvalue of the matrix, and n is the number of parameters. As per Sener and Davraz (2013), the CI can be linked to a stochastic matrix SI, such that the ratio CI/SI is the consistency ratio (CR). As a generic guideline, CR should be less than or equal to 0.1 (CR ≤ 0.1) to maintain the consistency of the AHP matrix (Saaty, 2003). Afterwards, the local rates in concurrence with each criterion were determined by aggregating the computed results of weights of each subcriterion multiplied by the rate of each alternative. Finally, the global rates were compiled by multiplying weights of criteria with the local rates (Bhushan & Rai, 2007). The CR for the AHP matrix in the present study was calculated as 0.03; thus, the matrix was consistent.

2.4 Data normalisation

Because of the modification of the existing DRASTIC process using the two different techniques, the observed minimum and maximum values of the proposed DVI are expected to be different than the original scale. Hence, the generated DVI maps were normalised using Equation 8 (Sener & Davraz, 2013).

(8)

where X_norm is the normalised DVI, X_min is the minimum DVI value, and X_max is the maximum DVI value.

2.5 Validation and analytical methods

Fifty-one representative groundwater samples were collected from dug wells located across the study area in the month of August 2017. Same dug wells were used for assessing the groundwater water table data (Figure 2). High-density polypropylene bottles (1 L capacity) were used for collection of groundwater samples. pH and total dissolved solids (TDS) values were measured in the field using a portable pH and TDS meter (Hanna, Mauritius). The samples were preserved by reducing the pH to less than 2 using nitric acid and stored at 4 °C in a freezer for further investigation. Whatman No. 42 filter paper was used to filter the samples, and later, they were analysed for different water quality parameters. Standard methods (American Public Health Association [APHA], 2012) were followed for the analysis of various parameters. Triplicate analysis was carried out, and mean values of the parameters (pH, TDS, total hardness, Mg, Cl, NO₃, SO₄, Fe, Na, and K) were considered. Measurement of total hardness was done using the standard ethylenediaminetetraacetic acid titrimetric method (APHA 2340C) with suitable precaution to conduct the titration at normal room temperature to ensure subtle colour change. Mg in groundwater was measured by volumetric method using ethylenediaminetetraacetic acid (IS 3025: Part 46). Chloride concentration was measured using the argentometric method (IS 3025: Part 32). Turbidimetric method (APHA 4500-SO₄²⁻ E) was used for estimating SO₄ concentrations. The concentrations of Na and K were determined using flame photometric method, whereas the concentration of NO₃⁻ N was analysed using ultraviolet–visible spectrophotometer (Shimadzu UV-1800) (APHA 4500-NO₃⁻). The values of Na and K concentrations were not used in calculating the Water Quality Index (WQI), as these parameters are generally not toxic at low concentrations. There are no guidelines for permissible limits of Na and K in drinking water. Iron (Fe) concentration was measured using atomic absorption spectrophotometer (FAAS-GBC Avanta PM, Australia). Analytical grade reagents and calibrated glassware were used during the analysis of different water quality parameters. For quality assurance and quality control (QA/QC), the whole process of sample collection from field to analysis in the laboratory was done conscientiously to avoid contamination and to obtain precise data.

2.5.1 Water Quality Index

WQI is an indexing technique of rating, which provides the composite influence of individual water quality parameter on the overall quality of water. Calculation of WQI involves three major steps: (a) selection, (b) standardisation, and (c) aggregation of the parameters (Stigter, Ribeiro, & Carvalho Dill, 2006). In the first step, the selected water quality parameters are assigned a weight (ω_i) according to its relative importance in the overall water quality for drinking purposes. In the second step, the relative weight of each water quality parameters is computed using Equation 9.

(9)

where W_i is the relative weight of each parameter, ω_i is the weight of each parameter, and n is the total number of parameters. The third step involves assigning a quality scale (q_i; Ramakrishnaiah, Sadashivaiah, & Ranganna, 2009) for each parameter by dividing its observed concentration in each water sample by its respective standard value and then multiply the resultant by 100 as presented in Equation 10. For the present study, Bureau of Indian Standards, IS: 10500 (Indian Standards-Drinking Water Specification; IS-10500, 2012) was considered.

(10)

where q_i is the quality rate of each parameter, C_i is the concentration of each parameter in its respective unit, and S_i is the Indian drinking water standards for each parameter in its respective unit. Finally, for computing the WQI, the sub index (SI_i) for each parameter is computed using Equation 11, which is then used to determine the WQI using Equation 12 (Ramakrishnaiah et al., 2009).

(11)

(12)

where SI_i is the sub index of ith parameter; q_i is the rate based on concentration of the ith parameter, and n is the number of parameters. WQI values were divided into five classes, which are ‘Excellent’ (<50), ‘Good Water’ (50–100), ‘Poor Water’ (100–200), ‘Very Poor Water’ (200–300), and ‘Water unsuitable for drinking’ (>300).

2.5.2 Statistical validation

Simple linear regression analysis (SLRA) was done between the DVI results and the validation parameters WQI, SO₄ concentrations, and Fe concentrations. Pearson correlation coefficients were also determined to evaluate the degree of association between the predictor models and the validation parameters. The purpose of performing the statistical analysis is to measure and interpret the relative impact of a predictor variable on a particular outcome (Sener & Davraz, 2013). This will help to identify the ranges of vulnerability zone where the proposed model is expected to perform better than the existing model. SPSS 21.0 was used for performing the automatic linear regression modelling and the Pearson correlation, where validation parameters were selected as the observed values and the DVI points were selected as the predictors. The DVI values were extracted to the WQI points using the “extract multi value to points” tool in ArcGIS 10.2.2.

3 RESULTS AND DISCUSSIONS

To evaluate the groundwater vulnerability using the original DRASTIC model, the seven hydrogeological factors (depth of groundwater, net recharge, aquifer media, soil media, topography, impact of vadose zone, and hydraulic conductivity) were combined using overlay and index techniques available in ArcGIS 10.2.2. As per DRASTIC technique, the factors were subcategorised into different ranges, and higher rates were assigned to the classes which were more likely to assist in contaminant transport. The results of the original DRASTIC model were normalised to a scale of 100 (Figure 4a), and its performance was evaluated by correlating with WQI, SO₄, and Fe concentrations. DVI was subcategorised into five spatial zones (Very Low, Low, Moderate, High, and Very High) based on equal distribution of the vulnerability values. Maximum distance was noticed between the observed and the predicted values for high DVI ranges (Figure 4d), which indicates the inability of the original DRASTIC model to predict high vulnerable zones. SLRA revealed that original DRASTIC model had a determination coefficient (R²) of .04 with WQI, .09 with SO₄ concentrations (Figure 5a), and .01 with Fe concentrations (Figure 5d). The results revealed that the north-eastern part of the study area along with south-eastern part are the most vulnerable to groundwater contamination. DVI values were subcategorised into five zones (Very Low, Low, Moderate, High, and Very High) as per Neshat, Pradhan, and Dadras (2014). It was observed from the spatial distribution of the study area that almost 29% (10,329 ha) of the entire area was lying in moderate vulnerability zone and about 28% (9,959 ha) of the total area was under high vulnerability zone (Table 4). Only 8% (2,913 ha) of the entire area was lying under very high vulnerability zone.

In order to improve the groundwater vulnerability prediction in the study area, site-specific parameters land use and distance from lineaments were added to the existing DRASTIC model, and the DVI was re-evaluated. Single parameter sensitivity analysis was performed to study the contribution of individual variable of each parameter on the DVI, but no significant change was observed in the overall DVI range. Single parameter sensitivity analysis gives importance to the factors, which were consistently high across the entire region as was reported by Pacheco et al. (2015) for another study area. The results of the modified DVI were normalised to a scale of 100 to compare it with other models (Figure 4b). SLRA revealed that Modified DRASTIC model had a determination coefficient (R²) of .62 with WQI (Figure 4e), .72 with SO₄ concentrations (Figure 5b), and .52 with Fe concentrations (Figure 5e). The results of the modified DRASTIC model showed improvement in the prediction. The entire active mining area of the coalfield was under high vulnerability zone spread throughout the northern part of the study area along with some region in the south-eastern part. Furthermore, it was observed from the spatial distribution of the study area that almost 33% (11,738 ha) of the entire area was lying in the moderate vulnerability zone and about 26% (9,258 ha) of the total area was under low vulnerability zone (Table 4).

To further strengthen the prediction of groundwater vulnerability considering coal mining as source of contamination in the present study area, the modified DRASTIC model was coupled with AHP. In AHP, pairwise comparison was done to evaluate the relative impact of each of the parameters on the vulnerability index. This also helps in reducing the redundancy by optimising the weights of the factors. The optimised AHP rates and weights (Table 2) were selected for each parameter and their subparameters. Again, the selected parameters were subjected to the overlay and index techniques in a GIS environment, and then the index was calculated. The computed index was normalised to a scale of 100 (Figure 4c). SLRA revealed that AHP–Modified DRASTIC model had a determination coefficient (R²) of .88 with WQI (Figure 4f), .86 with SO₄ concentrations (Figure 5c), and .70 with Fe concentrations (Figure 5f). Moreover, it was observed from the spatial distribution of the study area that almost 30% (10,743 ha) of the entire area was lying in low vulnerability zone and about 18% (6,552 ha) of the total area was under moderate vulnerability zone (Table 4). Whereas, the areal coverage of high vulnerability zone was observed as 17% (6,042 ha), and the areal coverage of very high vulnerability zone was observed as 6% (2,429 ha). WQI values in the very low to low DVI zone varied from 26.4 to 63.3, representing excellent quality of water in this zone. The ranges of all the analytes were within the permissible limits as per drinking water quality standards (IS-10500, 2012). WQI values in the moderate DVI zone varied from 68.3 to 112.0, indicating good water quality, but at some locations, the value of the total hardness, NO₃, and Fe exceeded the standards for drinking water quality. WQI values in the high DVI zone varied from 113.7 to 162.9, implying poor water quality. In the very high DVI zone, WQI values varied from 188.9 to 270.4, implying some of the areas were having very poor quality of water, and nearly all the samples exceeded the drinking water standards in this zone. Most of the highly vulnerable zones of groundwater contamination lied in the vicinity of active mining areas. The prominence of each parameter is represented by its weight, which varies considerably amongst the selected techniques. Therefore, critical care should be taken for deciding the weighting technique for a specific sector.

A correlation analysis was done with each of the GVA models and the validation parameters WQI, SO₄ concentrations, and Fe concentrations. WQI showed statistically positive significant correlation (p < .01) with Modified DRASTIC model (r = .79) and AHP–Modified DRASTIC model (r = .94). SO₄ concentrations showed weak positive (significant at p < .01) correlation with results of the original DRASTIC model (r = .31), and it showed strong positive significant relation with Modified DRASTIC model (r = .85) and AHP–Modified DRASTIC model (r = .93). These r values signify that the SO₄ concentrations showed the best correlation with AHP–Modified DRASTIC model followed by Modified DRASTIC model. Fe concentrations showed no significant correlation with the original DRASTIC model. Modified DRASTIC and AHP–Modified DRASTIC models showed strong positive statistically significant (p < .01) correlation with Fe concentrations with r values of .72 and .84, respectively. Fe concentrations were found positively correlated to SO₄ concentrations. WQI had a strong positive statistically significant (p < .01) correlation with Fe concentrations with r value of .93.

The WQI of the selected samples ranged from excellent (<50) to very poor water (200–300). The results of analysis of water samples along with calculated WQI are presented in Table 5. The pH of the analysed samples varied from 6.23 to 7.39 (slightly acidic to slightly alkaline nature) with an average value of 6.88. Unlike several coalfields, Jharia coalfield does not have evidences of acid mine drainage, hence, the pH of all of the samples was below the threshold value. TDS concentrations varied from 176 to 1,721 x 10⁻³ kg/m³ with an average value of 660 x 10⁻³ kg/m³. Groundwater contamination is a direct repercussion of land degradation due to mining operations. Absence of topsoil accelerates the process of infiltration of contaminants towards groundwater. Furthermore, mining processes such as blasting, excavation, and dumping of overburden expose the rocks (overburden) for weathering and accelerate the dissolution process and the rate of release of ions in water; all these ultimately cause increase in concentration of TDS and other dissolved ions (Lin et al., 2005). Total hardness of the water in the study area varied from 45 to 1,270 x 10⁻³ kg/m³ with an average value of 390 x 10⁻³ kg/m³. Concentrations of nitrate ranged from 4 to 87.5 x 10⁻³ kg/m³ with an average value of 40 x 10⁻³ kg/m³. Chloride concentrations varied from 49 to 394 x 10⁻³ kg/m³ the average concentration of chloride was observed as 145.4 x 10⁻³ kg/m³. The concentration of nitrate was high in some of the groundwater samples near mining areas. Elevated nitrate concentrations in groundwater may be directly attributed to the explosives used for blasting purposes, which contain nitrate compound as a parent material. In addition to higher values of nitrate concentrations in mining areas, the results also revealed high values of SO₄ concentrations in the groundwater samples of the mining areas; these high values are attributed to the oxidative weathering of pyrites in mining areas (Rivas-Pérez et al., 2016). SO₄ concentrations in the study area varied from 8 to 379 x 10⁻³ kg/m³, with a mean concentration of 172 x 10⁻³ kg/m³. The concentrations of SO₄ in groundwater exceeded the drinking water quality standards (200 x 10⁻³ kg/m³) in 37% of the collected groundwater samples, most of which are located in high and very high DVI zones. High sulphate concentrations in groundwater around mining areas were also observed in other studies (Singh, Mondal, Kumar, & Singh, 2008; R. K. Tiwari, 2001). Fe concentrations in the study area varied from 0.04 to 0.83 x 10⁻³ kg/m³ with an average concentration of 0.29 x 10⁻³ kg/m³. Nearly all of the samples in high and very high DVI zones had iron concentrations higher than the drinking water quality standards (>0.3 x 10⁻³ kg/m³).

4 CONCLUSIONS

Mining activities severely degrade the land in majority of these areas. These areas act as a direct source of groundwater contamination. The objective of the present study was to evaluate the effect of coal mining and its related activities on groundwater quality taking Jharia coalfield as the study area. The existing DRASTIC GVA model was modified by incorporating two additional parameters (land use and distance from lineaments) in a GIS environment. Single parameter sensitivity analysis was performed on the modified DRASTIC model for the present work to study the effect of individual variables on the overall vulnerability index. Furthermore, AHP was used to optimise the weights and rates of the parameters for better prediction of the groundwater vulnerability. The results of the GIS-based models were validated by analysing groundwater samples from the study area for different water quality parameters and then by computing the overall WQI for every sampling location. The original DRASTIC model had no significant correlation with WQI, which renders the original DRASTIC model unsuitable for GVA studies in similar areas globally. The AHP–Modified DRASTIC model was observed as the best amongst the studied techniques with a strong positive significant correlation with WQI (r = .94). The area around the mines is densely populated, and the economic conditions of people inhabiting these regions are usually poor. During the lean season, people rely on groundwater as the only feasible source of drinking water. The extracted groundwater should be properly treated before it is consumed, or else, it may have adverse health effects. This study can be useful in identifying regions that have contaminated groundwater near mining areas for framing better water management policies.