An integrated GIS based fuzzy pattern recognition model to compute groundwater vulnerability index for decision making

Available online at www.sciencedirect.com

Journal of Hydro-environment Research 5 (2011) 63e77 www.elsevier.com/locate/jher

Research paper

An integrated GIS based fuzzy pattern recognition model to compute groundwater vulnerability index for decision making Dhundi Raj Pathak*, Akira Hiratsuka Graduate School of Engineering, Osaka Sangyo University, 3-1-1 Nakagaito, Daito, Osaka 574-8530, Japan Received 3 July 2009; revised 24 September 2009; accepted 30 October 2009

Abstract This study highlights the computational technique of groundwater vulnerability index to identify the aquifer’s inherent capacity to become contaminated benefiting from fuzzy logic employing various hydrogeological parameters in the framework of Geographic Information Systems (GIS). This is usually carried out by using GIS based overlay index method. DRASTIC is one of the widely used popular overlay index method to compute groundwater vulnerability index over the large geographical areas involving a variety of hydrogeological settings. DRASTIC method uses linear model to calculate vulnerability index and factors that pertinent to the groundwater vulnerability should be divided into ranges to employ rating value to each range. This system is unable to demonstrate a continuous output of vulnerability index from the easiest to be polluted to the most difficult to be polluted that is fuzzy nature of the groundwater vulnerability to contamination. In this paper, integrated GIS based fuzzy pattern recognition model is developed to generate the continuous vulnerability function benefiting from the same input parameters of DRASTIC method. Moreover, vulnerability variation resulting from fuzzy and DRASTIC model with respect to any single input variable, making other parameters constant, is computed taking the characteristics of selected hydrogeological settings to compare the output of fuzzy model with DRASTIC index. The ability of GIS based fuzzy pattern recognition model to generate continuous output of vulnerability index may be considered as a pronounced advantage over DRASTIC method. Groundwater vulnerability map has been developed utilizing its output in shallow groundwater aquifer of Kathmandu, Nepal as a case study. Finally, output of vulnerability models are tested by nitrate data which were measured from ninety sources from shallow groundwater systems of study area. In large geographical areas with limited data, the groundwater vulnerability maps provide important preliminary information to decision makers for many aspects of the regional and local groundwater resources management and protection. Ó 2010 International Association of Hydro-environment Engineering and Research, Asia Pacific Division. Published by Elsevier B.V. All rights reserved. Keywords: Groundwater vulnerability map; GIS; Decision making; Fuzzy pattern recognition model; DRASTIC method; Nepal

1. Introduction Groundwater is a globally important, valuable and renewable natural resource of water supply due to its relatively low susceptibility to contamination in comparison to surface water and its large storage capacity; however, it is under threat of degradation both by inappropriate use and by contamination. For e.g. the quality of groundwater in urban areas of developing countries, like Nepal has been deteriorating in recent * Corresponding author. E-mail address: [email protected] (D.R. Pathak).

years mainly due to the high growth of population, unplanned growth of cities, excessive use of fertilizers and pesticides in agriculture land, no proper sewage system and poor disposal of the wastewater both from household as well as industrial activities. Therefore, contamination of groundwater has become a major anxiety of planners, decision makers and water managers involved with managing the quantity and quality of water in relation to human health in recent years. The contamination of groundwater however is a widespread problem and requires huge investments for remediation. Therefore, it is important to identify which aquifer systems and hydrogeological settings are most

1570-6443/$ - see front matter Ó 2010 International Association of Hydro-environment Engineering and Research, Asia Pacific Division. Published by Elsevier B.V. All rights reserved. doi:10.1016/j.jher.2009.10.015

64

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

vulnerable to contamination prior to implementing groundwater monitoring program in large geographical areas. In recognition of the need for effective and efficient methods for protecting groundwater resources from future contamination, scientists and resource managers have sought to develop techniques for predicting which areas are more likely than others to become contaminated as a result of activities at or near the land surface (NRC, 1993). This concept has been widely termed to groundwater vulnerability to contamination. It is the sensitivity of groundwater quality to an imposed contaminant load, which is determined by the intrinsic characteristics of the aquifer. The groundwater vulnerability map based on aquifer vulnerability index is the subdivision of the area into several hydrogeological units with different levels of vulnerability which shows the distribution of highly vulnerable areas, in which pollution is very common because contaminants can reach the groundwater within a very short time. In general, numerical groundwater modeling is an important predictive tool for managing water resources in aquifers (El Yaouti et al., 2008), nevertheless such models employ sets of extensive field measured data, which is in fact, very costly and inefficient in large geographical areas for preliminary groundwater resources management and protection program. Due to the difficulty in mathematical formalism and lack of sufficient hydrogeologic and geochemical database at regional scale, the quite conceptual method like generalized knowledge base (GKB) approach may be appropriate. In GKB approach, decision is made on the basis of general hydrogeological knowledge of the contaminant transport in aquifer media (Afshar et al., 2007). In order to tackle the groundwater pollution and to protect its quality in a more scientific and efficient way, many different methods based on GKB approach have been developed to evaluate the groundwater vulnerability to pollution such as GOD method (Foster, 1987), DRASTIC (Aller et al., 1987), SINTACS (Vrba and Zaporozec, 1994), EPIK technique (Doerfliger and Zwahlen, 1997). The most typical and popular method based on GKB approach is the DRASTIC method, developed by the United States Environmental Protection Agency (EPA) (Aller et al., 1987). 2. DRASTIC method The DRASTIC acronym stands for the seven hydrogeological parameters; depth to water, recharge, aquifer media, soil type, topography (slope), impact on the vadose zone media and hydraulic conductivity of the aquifer. Overlay and index methods, such as DRASTIC, are relatively easy to implement and require little data, but the result can be questioned because these methods rely more on the judgment of the analyst than on the actual hydrogeological processes (Frind et al., 2006). Gogu et al. (2003) reported that different overlay and index methods like DRASTIC applied to same hydrogeological system can generate dramatically dissimilar results. Despite these criticisms due to the lack of proper validation, this method has been adopted in the various part of world (Barber et al., 1994; Lynch et al., 1997; Babiker et al., 2005; Rahman, 2008). This method is often modified to better address local issues or better represent

a local hydrogeologic setting (Merchant, 1994). Further, different researchers modified this methodology for mapping the intrinsic vulnerability of aquifers to better represent a local hydrogeological setting (Zhang et al., 1996; Thirumalaivasan et al., 2003; Denny et al., 2007; Gomezdelcampo and Dickerson, 2008). Thirumalaivasan et al. (2003) developed AHP-DRASTIC model to derive ratings and weights of modified DRASTIC model parameters for use in specific aquifer vulnerability assessment studies. They applied Analytic Hierarchy Process (AHP) to compute the ratings and weights of the criteria and sub-criteria of all parameters used in the DRASTIC model. Recently, this popular GIS based overlay index method was introduced in Nepal to estimate the vulnerability index of shallow groundwater aquifer of Kathmandu (Pathak et al., 2009). Further, sensitivity analysis was utilized to evaluate the relative importance of model parameters and then revised their weights, what is different from original DRASTIC method, to better address local hydrogeological settings and terrain characteristics of Kathmandu (Pathak et al., 2009). It was the first attempt to develop intrinsic groundwater vulnerability map of Kathmandu Valley that provided only a preliminary relative evaluation tool because of the some limitations of adopted approach (DRASTIC method) and insufficient input parameters. Therefore, it is noted that the method and input parameters to produce the groundwater vulnerability index should be improved to make reliable tool for groundwater quality protection and decision making in this region. In this context, this study focuses to develop more reliable groundwater vulnerability map improving computational technique and input parameters. Moreover, output of vulnerability model was validated by nitrate data measured from ninety sources of shallow groundwater systems in Kathmandu. In general, DRASTIC index uses linear model to calculate the final vulnerability index cumulating the product of rating value (r) with its corresponding weight (w) of seven hydrogeological parameters given by following relation: Vi ¼

7 X

wi ri

ð1Þ

i¼1

DRASTIC elected to use weight for each parameter based on its relative significance contributing to the pollution potential (depth to groundwater ¼ 5; net recharge ¼ 4; aquifer media ¼ 3; soil type ¼ 2; topography ¼ 1; impact on vadose zone ¼ 5; and hydraulic conductivity ¼ 3). Either each factor has been divided into ranges or media types and assigned a rating from 1 to 10 based on their significance to pollution potential as shown as Tables 1 and 2. Although, DRASTIC is one of the most widely used standard groundwater vulnerability method, it is unable to describe a continuous transition from the easiest to be polluted to the most difficult to be polluted that is fuzzy nature of the groundwater vulnerability to contamination. In this method, the nature of the vulnerability is taken into account by dividing the values of each affecting factor into ranges and then giving to rating value to each range. However, it should be noted that if a factor value can

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

65

Table 1 DRASTIC standard ranges and ratings for DRASTIC factors that can be measured directly. Depth to water table(D)

Recharge (R)

Topography (T)

Range (m)

Rating

Range (mm)

Rating

Range (%)

Rating

Range (m/d)

Rating

0e1.5 1.5e4.6 4.6e9.1 9.1e15.2 15.2e22.5 22.5e30 >30

10 9 7 5 3 2 1

0e51 51e102 102e178 178e254 >254

1 3 6 8 9

0e2 2e6 6e12 12e18 >18

10 9 5 3 1

0e4.1 4.1e12.2 12.2e28.5 28.5e40.7 40.7e81.5 >81.5

1 2 4 6 8 10

be measured numerically, unlike the function of DRASTIC index, the fuzzy system generates a continuous vulnerability function. Hence, fuzzy approach can be used to assess the groundwater vulnerability to contamination. 3. Fuzzy approach Basically, this contribution aims to identify the aquifer’s inherent capacity to become contaminated based on DRASTIC system benefiting from fuzzy concept in the frameworks of GIS. As an example of vulnerability linguistic evaluation of vulnerability, the more shallow the water table, the higher the groundwater pollution potential and less the recharge rate, the smaller the groundwater pollution potential. First introduced by Zadeh (1965), fuzzy logic and fuzzy set theory have been extensively used in ambiguity and uncertainty modeling in decision making. The basic concept in fuzzy logic is quite simple; statements are not only “true” or “false” but also represents the degree of truth or degree of falseness for each input. Fuzzy sets are characterized by membership functions. Several approaches have been used to apply fuzzy set theory to groundwater contamination problems, including fuzzy pattern recognition and optimization technique (Zhou et al., 1999; Chen and Fu, 2003), fuzzy rule-based systems (Uricchio et al., 2004; Dixon, 2005; Gemitzi et al., 2006; Afshar et al., 2007), fuzzy hierarchy model (Nobre et al., 2007). Zhou et al. (1999) used a multi objective fuzzy pattern recognition

Hydraulic Conductivity (C)

model and further, Chen and Fu (2003) developed the generalized fuzzy pattern recognition model to evaluate groundwater vulnerability, taking only the standard value matrix of five samples of the study area. However, the previous studies lack to incorporate the continuous input parameters benefiting from fuzzy concept to generate continuous vulnerability index for mapping of actual aquifer systems in watershed scale utilizing the powerful spatial and visual capability of GIS. Hence, in this study, GIS based two level fuzzy pattern recognition model is developed to evaluate the degree of vulnerability by means of natural language; the “easiest to be polluted” to “most difficult to be polluted”. The flowchart presented in Fig. 1 clearly illustrates how GIS and vulnerability models integrated to develop groundwater vulnerability map using DRASTIC parameters. 3.1. Fuzzy pattern recognition model Groundwater vulnerability assessment can be regarded as pattern recognition problem in which, identification of the vulnerability level to which a sample belongs according to the seven factor values of the sample when compared with the standard values obtained from DRASTIC method. Standard values of two levels with regard to each factor are presented on the basis of data in the DRASTIC system as shown as Table 3. According to Table 3, standard value matrix of the factors is given by

Table 2 DRASTIC standard ratings value for parameters that cannot be measured directly. Aquifer media (A) Range Massive shale Metamorphic/Igneous Weathered Metamorphic/Igneous Glacial Till Bedded Sandstone, Limestone and Shale Sequences Massive Sandstone Massive Limestone Sand and Gravel Basalt Karst Limestone

Soil type (S) Rating 2 3 4 5 6 6 6 8 9 10

Impact of Vadose Zone (I)

Range

Rating

Range

Thin or absent Gravel Sand Peat Shrinking and/or Aggregated Clay Sandy loam Loam

10 10 9 8 7

Confining Layer Silt/Clay Shale Limestone Sandstone

1 3 3 6 6

Bedded Limestone, Sandstone, Shale Sand and Gravel with significant Silt and Clay Metamorphic/Igneous Sand and Gravel Basalt Karst Limestone

6 6

Silty Loam Clay Loam Muck Nonshrinking/ Nonaggregated Clay

6 5 4 3 2 1

Rating

4 8 9 10

66

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

Fig. 1. Flow chart of methodology adopted to develop groundwater contamination potential map using DRASTIC and fuzzy pattern recognition model in framework of GIS.



0 25:4 10 10 0 10 81:5 F¼ 30:5 0 2 1 18 1 0

T

  ¼ fi;h

ð2Þ

where fi,h is the standard value of level h with regard to factor i; i ¼ 1, 2.7 and h ¼ 1, 2. The level 1 and level 2 correspond to easiest to be polluted and most difficult to be polluted in term of linguistic variables respectively. According to Table 3, higher the standard value, higher the level h for parameters; D

Table 3 Standard values of two levels with regard to each factor based on DRASTIC system. Factors

D (m)

R (mm)

A

S

T (%)

I

C (m/d)

Level 1 Level 2

0 30.5

254 0

10 2

10 1

0 18

10 1

81.5 0

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

and T, while higher the standard value, lower the level h for parameters; R, A, S, I and C. The membership degree of first level standard value with regard to linguistic concept “easiest to be polluted” is supposed to be 1 and the membership degree of the second level standard value i.e. “most difficult to be polluted” in term of fuzzy concept supposed to be 0. The membership degree of other levels varies from 0 to 1. The membership degree, si,h of fi,h with respect to “easiest to be polluted” is computed by: 0 fi;h  fi;2 si;h ¼ fi;1  fi;2 1

fi;h ¼ fi;2 fi;1 > fi;h > fi;2 ; fi;1 < fi;h < fi;2 fi;h ¼ fi;1



1 1 1 1 1 1 1 0 0 0 0 0 0 0

ð4Þ

where xij is the value of sample j with regard to factor i; i ¼ 1,2,.,7; j ¼ 1,2,.nand n is total number of samples to be evaluated. The factors in DRASTIC system can be classified into two groups: A and B. In group A, the groundwater vulnerability increases with increasing the value of factors, whereas it is reverse in group B, the groundwater vulnerability reduces when factor value increases. For the group A and B, the membership degree of factors i.e. ri,j, can be calculated by using the following Eqs. (6) and (7) respectively: xij  xminj ; xminj < xij < xmaxj rij ¼ x : maxj  xminj xij  xmaxj 1

rij ¼

8 < 0xmaxj  xij

xij  xmaxj ; xminj < xij < xmaxj : xmaxj  xminj xij  xminj 1

Range

Rating

Weight

Normalized weight

D R A S T I C

4.6e9.1 m 178e254 mm Sand & Gravel Sand 0e2% e e

7 8 8 8 10 8 8

5 4 3 2 1 5 3

0.217 0.174 0.130 0.087 0.043 0.217 0.130

T

Considering the factor values of the samples in study area from following factor value matrix:   ð5Þ X ¼ xij 7xn

8 < 0xij  xminj

Factors

w ¼ ðw1 ; w2 ; .; w7 Þ

T ¼ ðsi;h Þ:

Table 4 Characteristics of selected hydrogeological unit for comparison of fuzzy pattern recognition and DRASTIC model.

ð3Þ

where fi,1 and fi,2 are the standard values of the “easiest to be polluted” and “most difficult to be polluted”, respectively. By using Eq. (3), Eq. (2) can be transformed into membership degree matrix of standard values, which given by: S¼

67

ð9Þ

The distance of sample j to the level h can be described as: vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u 7 u X  p p ð10Þ dhj ¼ t wi rij  si;h i¼1

where p is a distance parameter, when p ¼ 1, and p ¼ 2, the distances are called Hamming and Euclidean distances respectively, which are commonly used. Euclidean distance is used in our case i.e. p ¼ 2. In the view of fuzzy sets theory, uh,j can be considered as a weight for distance dhj, so synthetically weighted distance can better express difference between factor value and level h in each sample, which is given as follow: vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi u 7 u X  p p ð11Þ wi rij  si;h Dhj ¼ uh;j t i¼1

In order to acquire the optimized solutionPof uh,j, the objective function is established with constraint 2h¼1 uh;j ¼ 1, as follows: ) ( 2   X 2 Dhj ð12Þ min F uh;j ¼ h¼1

ð6Þ

ð7Þ

where xmaxj and xminj equal the maximum and minimum value, respectively of factor i assigned in DRASTIC system. By using Eqs. (6) and (7), Eq. (5) can be transformed into membership degree matrix of factors:   ð8Þ R ¼ rij 7xn Each factor is of different importance in relation to vulnerability, hence different weights are attributed to different factors as shown as Table 4, which are usually normalized to sum to one in the evaluation process. The weight vector is denoted by:

Then, the Lagrange function can be derived as follows:

X  X 2 2  uh;j  1 ð13Þ uh;j dhj  lj L uh;j ; lj ¼ where lj is a Lagrange multiplier. Taking the partial derivatives to the Lagrange function with respect to uh,j and lj and setting the simultaneous equation equivalent to zero, i.e.     ð14Þ vL uh;j ; lj vuh;j ¼ 0; vL uh;j ; lj vlj ¼ 0 Solving Eq. (14), we get the formula for calculating the membership degree of sample j that belongs to level h is: uh;j ¼

dhj2

2 X

!1 dkj2

ð15Þ

k¼1

when dhj ¼ 0, i.e. ri,j ¼ si,h, which shows that sample j completely belongs to level h, such that uh,j ¼ 1. Further, two level fuzzy pattern recognition model can be dealt as a fuzzy optimization model, where, h ¼ 1, 2. Then, the

68

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

membership degree of sample j corresponding to level 1 that represent the “easiest to be polluted” in term of linguistic variables is computed by using Eqs. (4), (11), and (15), which is given by: 8 911 0 > > > > =C 7 BX  2 < 1 1 C B wi rij wi þ u1;j ¼ @ A 7  7    P P > > 2 2 > > i¼1 : wi rij wi wi rij ; i¼1

4. Case study 4.1. Study area

i¼1

ð16Þ Ultimately, we get, 31 2 87 2 9 P > > > wi rij  wi > =7 < 6 7 6 i¼1 ui;j ¼ 61 þ 7 7   P > > 5 4 2 > > ; : wi rij

represents the fuzzy concept “easiest to be polluted”) in each sample of study area in the framework of GIS. According to this model, higher the u1,j, the “easiest to be polluted” the sample j.

ð17Þ

i¼1

Eq. (17) is 2-level fuzzy pattern recognition model, which is used to evaluate the degree of groundwater vulnerability (that

The groundwater vulnerability map of Kathamndu Valley was prepared which includes three major cities: Kathmandu, Bhaktapur and Lalitpur. The total area of valley for the study is about 350 square kilometers as shown as Fig. 2. The valley consists of gentle hills and flat lands at elevations of 1300e1400 m. The surrounding hills rise to more than 2000 m in elevation Phulchoki to the south of the Valley has the highest elevation at 2762 m. Average annual precipitation in the Kathmandu Valley is around 1400 mm, about 80% of which falls in the monsoon period during June and July.

Fig. 2. Location map of Kathmandu Valley (study area).

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

Within the valley, municipal and other water supplies depend on monsoon rains and the stream and groundwater systems fed by this precipitation. Surface runoff is high during the monsoon and recharge to the shallow aquifers occurs mostly along the basin margins, directly from precipitation and by supply from a number of small rivers. However, recharge to the deeper aquifers is considered to be limited, due to the presence of clay beds that significantly restrict downward percolation. Because the Kathmandu Valley is a closed basin with gentle slopes toward the center, groundwater flow is assumed to be slow, particularly in the deeper aquifers. The surface of the Kathmandu Valley is almost flat but it has buried bedrock surface with irregular shapes and high relief. The depth of the Precambrian bedrock from the ground surface ranges from tens of meters to more than 500 m. The thick quaternary deposits consist of lacustrine and fluvial deposits, which have been eroded, however the original thickness of the deposits is unknown. The basin fill sediments of Kathmandu Valley are mainly divided into two formations; Quaternary and Plio-pleistocene formation, each with different lithologic, geotechnical properties (Shrestha et al., 1999). Based on the engineering and environmental geological map of Kathmandu Valley (Shrestha et al., 1998), the geological setting of

69

Kathmandu Valley with different formations is shown in Fig. 3. The Quaternary formation, mainly formed by unconsolidated materials/sediments, which consists of different four units; recent alluvial soil, residual soil, colluvial soil and alluvial fan deposit while the Plio-pleistocene formation consists of slightly consolidated sediments and has different seven units; Tokha formation, Gokarna formation, Chapagaon formation, Kalimati formation, Kobgaon formation, Lukundol formation and Basal boulder bed. The Quaternary formation, mainly formed by unconsolidated materials/sediments, which consists of different four units; recent alluvial soil, residual soil, colluvial soil and alluvial fan deposit. The brief description of each formation has been presented in previous work (Pathak et al., 2009). By convention, the aquifers in the Kathmandu Valley can be divided into shallow and deep systems. A shallow unconfined aquifer occurs at around 0e10 m depth and a deep confined aquifer occurs at around 310e370 m (Khadka, 1993). Other isolated groundwater storeys are situated at significantly deeper levels (Gautam and Rao, 1991). Groundwater from the shallow aquifers is drawn from hand-dug wells, hand pumps or roar pumps, whereas the deeper aquifers are exploited from deep wells. Traditional stone spouts (locally known as dhunge dhara) are also common, drawing water from shallow aquifers.

Fig. 3. Geological map of Kathmandu Valley.

70

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

Groundwater from both shallow and deeper aquifers has been used extensively for drinking and industrial purposes. About 50% of the water used in the city of Kathmandu is derived from groundwater (Jha et al., 1997; Khatiwada et al., 2002). Exploitation of these aquifers, especially the shallow aquifer, has been increased rapidly in recent years. The quality of water extracted from such sources is under threat of degradation by contaminants because of the different anthropogenic activities, resulting from rapid unplanned and haphazard urbanization of entire valley. The urban growth detection was 10.86 square kilometers from 1988 to 1997 (ICIMOD, 2000), which has been further increased since then. 4.2. Model input parameters and groundwater vulnerability index All seven input data layers used in DRASTIC system were generated and/or obtained from its original source as a point, line, or polygon layer (Fig. 4). Then, all parameters contributing to groundwater vulnerability were converted from vector (point, line, or polygon) to raster (grid) using the GIS. In raster layer, space is subdivided into discrete cells with required resolution. In this work, all input parameters for the DRASTIC

and fuzzy pattern recognition model were generated in seven separate raster layers of 30 m  30 m grid resolution (Fig. 5). GIS techniques were utilized with the help of Eqs. (6) and (7) to generate the continuous input layer of each DRASTIC parameter. Depth to water table was collected from borehole log information, direct measurement of existing groundwater wells and other secondary information. Both inverse distance moving average interpolation technique and kriging were tested on the measured depth to groundwater point data to generate raster surface. However, the kriging technique was found to be suitable to generate smooth surface. The membership degree value map was computed using Eq. (7) and rating map was prepared by assigning sensitivity rating values as 10 for depth (<1.5 m), 9 for depth (1.5e4.6 m), 7 for depth (4.6e9.1 m), 5 for depth (9.1e15.2), 3 for depth (15.2e22.5 m), 2 for depth (22.5e30 m) and 1 for depth (>30 m). The shallow aquifer of the valley is recharged mainly by direct infiltration from precipitation therefore net recharge was estimated by using following formula: Net recharge ¼ rainfall  evaporation  runoff

Fig. 4. Example of developing model input parameters.

ð18Þ

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

71

Fig. 5. Seven input raster layers to compute vulnerability index.

where rainfall map was prepared by interpolation mean of annual precipitation (mm/year) from the 21 representative rainfall stations in the Kathmandu Valley (DHM, 2006). Evaporation data was used from only one station of the valley recorded in international airport of Kathmandu (DHM, 2006). Runoff was calculated on each pixel based on empirical relation in which the runoff coefficients assumed to be 0.8 for built up/urban area, 0.27 for forest, 0.25 for open field/lawn, 0.4 for agricultural field with clay, 0.3 for agricultural field with sand and 0.15 for water body and highly permeable recent flood plain. Thus obtained recharge value from Eq. (18) was used to calculate the membership degree value as well as rating map to evaluate the degree of vulnerability.

The aquifer media map was developed based on various sources regarding groundwater basin and geological formation map of Kathmandu (Shrestha et al., 1998; JICA, 1990; Jha et al., 1997). Rating value was assigned based on DRASTIC method. The grid layer of soil media was generated from soil map from Department of Survey, Nepal (NGIIP, 1994). The major soil types available in study area are loamy, loamy skeletal and loamy/bouldery. Hence, rating was assigned as according to DRASTIC method based on the soil type. The topographic contours map of 1:25000 scale (NGIIP, 1994) was digitized to construct slope map using 3D analyst and spatial analyst in ArcGIS9.2. The slope was converted into membership value using Eq. (7) and rating map also prepared

72

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

Fig. 6. Illustration for how integrated GIS based fuzzy model compute different vulnerability value within same range of input parameters despite of DRASTIC method.

with assigning sensitivity rating as 10 for plain (<2%), 9 for gentle (2e6%), 5 for moderate (6e12%), 3 for steep (12e18%) and 1 for very steep (>18%). The parameter, impact of vadose zone represents the influence of unsaturated zone above the water table, which controls the passage and attenuation of the contaminated material to the aquifer. The impact of vadose zone map layer was prepared using geological formation and soil map of Kathmandu Valley. This map was also verified using some borehole log information from different part of valley. Rating value was assigned according to DRASTIC method to compute vulnerability index.

Aquifer hydraulic conductivity is the ability of the aquifer formation to transmit water. It depends on the intrinsic permeability of the material and on the degree of saturation. Generally, the hydraulic conductivity is measured from the field pumping tests data. In this study, hydraulic conductivity values were obtained from pumping test data (Metcalf and Eddy, 2000) and have been interpolated to generate hydraulic conductivity map of required resolution. According to Metcalf and Eddy (2000), hydraulic conductivity of the study area is lower than 10 m/d that suggests hydraulic conductivity has less contribution to groundwater vulnerability to contamination in designed study area. Hydraulic conductivity data was converted to fuzzy

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

73

membership value using Eq. (6) and rating value was assigned according to DRASTIC method. Finally, relative degree of vulnerability (i.e. vulnerability index) of the each sample of the study area was calculated using Eq. (17). The output of this model was utilized to generate groundwater vulnerability map, where index value is ranged from “most difficult to be polluted” to “easiest to be polluted” in term of linguistic variables. Further, DRASTIC index was computed using the Eq. (1) to compare the output of fuzzy pattern recognition model. 5. Results and discussion 5.1. Vulnerability variation in fuzzy pattern recognition model In DRASTIC method, all seven input raster layers are divided into certain ranges to employ rating value to each range then final vulnerability index is computed. However, it should be noted that if a factor value can be measured numerically, unlike the function of DRASTIC index, the fuzzy system generates a continuous vulnerability function. The input parameters, which are very important to groundwater vulnerability to contamination, such as depth to water table, recharge, hydraulic conductivity and slope could be measured numerically. Therefore, it is not necessarily to divide in certain range to compute vulnerability index by assigning rating value what is usually done in DRASTIC method. For e.g., rating value 7 is assigned for depth 4.6e9.1 m range that implies DRASTIC index is equal at any place at this range of water depth however, vulnerability value may differ at water depth of 4.6 m and 9.1 m. Fig. 6 illustrates an example how integrated GIS based fuzzy pattern recognition model compute different vulnerability index value within same range of input parameters at particular hydrogeological setting despite of DRASTIC method. The decision making model benefiting from fuzzy logic, which is also based on the knowledge of the DRASTIC system hence its verification seems quite vital. Hence, this study validates the model’s performance by comparing the results with those of normalized DRASTIC index. Since maximum and minimum values of the DRASTIC index are 226 and 23 respectively, then the normalized DRASTIC index may be obtained as: In ¼ ðId  23Þ=203

ð19Þ

where In and Id are normalized and computed DRASTIC indices, respectively. To compare the output of the fuzzy model with DRASTIC index, vulnerability variation resulting from fuzzy and DRASTIC model with respect to any single input variable, making other parameters constant, is computed taking the characteristics of selected hydrogeological settings. The characteristics of the seven factors of the selected hydrogeological setting used for comparison study are shown in Table 4. The comparison between outputs of two models indicate that the fuzzy system has continuous nature with respect to input

Fig. 7. (a) Vulnerability variation of water depth in fuzzy pattern recognition model and DRASTIC method. (b) Vulnerability variation of recharge in fuzzy pattern recognition model and DRASTIC method. (c)Vulnerability variation of topography in fuzzy pattern recognition model and DRASTIC method.

factors and unveils two upper and lower bound whereas the output of DRASTIC has a discrete nature (Fig. 7aec). The fuzzy index is higher than DRASTIC index however, both models follow same trend. It is shown that by assigning ratings for related factors falling into certain range, DRASTIC method will ignore the difference of factor values within the same range and is unable to reflect to the influence the variation of hydrogeological factors on the groundwater vulnerability. Since, the model parameters are derived from the DRASTIC system, similar results with that of DRASTIC are expected, however, the main difference is that fuzzy pattern recognition model can generates a continuous vulnerability function unlike step DRASTIC output function, which may be considered as a distinct advantage over DRASTIC method.

74

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

Fig. 8. (a) Groundwater contamination potential map based on vulnerability index computed from fuzzy pattern recognition model. (b) Groundwater contamination potential map based on vulnerability index computed from DRASTIC method.

5.2. Vulnerability map Groundwater vulnerability index was computed using Eq. (17) utilizing all developed input parameters in GIS frameworks. Fig. 8a shows the relative degree of groundwater vulnerability to contamination which was obtained from the fuzzy pattern recognition model based on DRASTIC system. The values extend from 0.24 to 0.87 i.e. “most difficult to be polluted” to “easiest to be polluted” in term of linguistic

variables. Fig. 8b also shows the degree of groundwater vulnerability to contamination in term of normalized DRASTIC index, where the values vary from 0.29 to 0.79 with the lowest possible rating being 0.29 and the highest rating being 0.79. We categorized groundwater vulnerability map into five classes: very low, low, medium, high and very high by introducing the higher the index, the greater the relative pollution potential. Fig. 9 illustrates the number of samples (pixels) corresponding to vulnerability output of fuzzy and

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

75

while the dominant part of the study area had more than 254 mm/year. The results also indicate the rich groundwater resources area; northern part of Kathmandu like, Gokarna formation and highly permeable alluvial deposits are highly susceptible for vulnerability in which, if pollution is common, contaminants can reach the groundwater within a very short time. 5.3. Validation of output of vulnerability model by field measured nitrate data

Fig. 9. Frequency distribution of vulnerability index from fuzzy pattern and DRASTIC model.

DRASTIC model in study area. A high index that corresponds to “easiest to be polluted” in linguistic term, indicates the capacity of the hydrogeologic environment and the landscape factors to readily move waterborne contaminants into the groundwater and consequently need to be managed more closely. Low index i.e. “most difficult to be polluted” represents groundwater that is better protected from contaminant leaching by natural environment. The output of fuzzy model reveals especially northern part of valley and recent alluvial deposits falls under very high vulnerable that is about 28% of the total area. About 47, 21 and 4% of the valley was classified as high, medium and low vulnerable area respectively. No area was found in the category very low vulnerable. While in DRASTIC method, no area was categorized as very high vulnerable zone nevertheless 58 and 38% of the area categorized as high and medium vulnerable. Similarly, 4% of the total area is classified as low vulnerable zone. The combination of the model parameters that pertinent to groundwater vulnerability like very shallow depth to water table (<10 m) in the most part of the study area with almost flat area (<2% slope) and high recharge rate in study area led to this high pollution potential index from both method. The lowest recharge was associated with urban land

The modeled results should validate from field observations which is however quite difficult and expensive in large geographical area. In this study, the output of vulnerability models were tested using measured nitrate data from the shallow aquifer of Kathmandu. Out of 90 water sources, 15 sources have higher concentrations, which exceeded United State Environmental Protection Agency (USEPA) guidelines of 10 mg/L as nitrateenitrogen (USEPA, 2009); however, another 30 wells had impacted levels of nitrate between 2 and 10 mg/L. The sampled groundwater wells were overlaid on the groundwater vulnerability map using GIS in order to see how many wells with high concentration of nitrate are found in different vulnerable zones. Groundwater vulnerability map developed in this study based on aquifer characteristics do not consider contaminants sources, loading and transport mechanism into groundwater systems nevertheless it was encouraging to find that majority of sampled sources which violated USEPA limit value of nitrate are located in the high to very high vulnerable zones for both models (Fig. 8a and b). Although, few sampling wells were highly contaminated by nitrate in some areas, DRASTIC predicted no area that was categorized as very high vulnerable zone, which indicates it could not predict vulnerability level more accurately (Fig. 8b). However, fuzzy based method predicted sampling wells that were highly contaminated by nitrate in very high vulnerable zone. Fuzzy pattern recognition method predicted three and six wells out of 15 contaminated wells as very high vulnerable and high vulnerable respectively (Fig. 8a). From this result, it

Fig. 10. Relationship between groundwater depth and nitrateeN in shallow aquifer of Kathmandu.

76

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77

can be concluded that vulnerability predicted by fuzzy pattern recognition method is more reliable than DRASTIC method. If a watershed manager uses this result to conduct the groundwater sampling strategy for potential contaminants by human activities in the study area, this will be far more useful, compared to results generated by conventional overlay index method like DRASTIC. In addition, the relationship between nitrate and groundwater depth (one of the important parameters that pertinent to groundwater vulnerability to contamination) indicates nitrate concentration was, high it existed within the top 10 m and non detectable amounts were found in deeper groundwater (Fig. 10). As the groundwater wells are contaminated by nitrate due to the anthropogenic activities from or nearby ground surface, concentrations of nitrate should be higher at wells of low water depth. However, significant number of wells which have a high concentration of nitrate is located in the low vulnerable zones, especially areas that belongs old urban setting of valley. Possible reasons that nitrate concentrations observed high values in old urban areas even categorized as low vulnerable zones based on intrinsic vulnerability index are due to the inadequate disposal of human and animal waste and leach from septic tanks for a long time. In those areas, many households use septic tanks and the proximity of the septic tanks and the groundwater wells are not maintained, where the nitrate could infiltrate into the shallow aquifers.

however, both models follow same trend. The study shows that 75% and 58% of the valley’s shallow groundwater aquifer is under high to very high vulnerability to contamination from fuzzy and DRASTIC method respectively which is the main cause of concern for more than 2 million people living in Kathmandu. Moreover, the accuracy of the DRASTIC and fuzzy results was evaluated by comparing the results with nitrate data sampled from shallow groundwater aquifer of Kathmandu. Fuzzy pattern recognition model predicted three and six wells out of 15 contaminated wells as very high vulnerable and high vulnerable respectively while DRASTIC predicted no area that was categorized as very high vulnerable zone. From this result, it can be concluded that vulnerability predicted by fuzzy pattern recognition method is more reliable than DRASTIC method. This result affirms the validation and reliability of an integrated GIS based fuzzy pattern recognition model to some extent, which reflect an aquifer’s inherent capacity to become contaminated. However, special emphasis should be given to update model input parameters, loadings and fate of contaminants transport into groundwater systems to get the reliable output for policy and decision making in groundwater management in watershed scale. The groundwater vulnerability maps developed in this study are significant screening tools in policy and decision making for many aspects of the regional and local groundwater resources management and protection.

6. Summary and conclusions The overall goal of this study is twofold. First, it aims to improve the methodology for the computation of groundwater vulnerability index to generate contamination potential map by incorporating the continuous nature of vulnerability to contamination using DRASTIC parameters in large geographical area. Second, it brings up to date the input parameters of vulnerability model and geochemical data to validate vulnerability map of shallow groundwater aquifer of Kathmandu, where more than half of the population depend on groundwater sources to fulfill their water demand. Generally speaking, there is a continuous transition from easiest to the most difficult aquifer to be polluted, which is in fact fuzzy nature of groundwater vulnerability to contamination. In this regard, integrated GIS based fuzzy pattern recognition model generates the continuous vulnerability function unlike step DRASTIC index, which is in fact the pronounced advantage over DRASTIC method. This approach could take fuzziness nature of groundwater vulnerability (i.e. continuous transition from easiest to the most difficult aquifer to be polluted) more efficiently than DRASTIC method. An integrated GIS based fuzzy pattern recognition model based on DRASTIC system can be applied to any aquifer systems to predict groundwater vulnerability more efficiently. This approach has been applied to develop the groundwater vulnerability map to shallow groundwater systems of Kathmandu Valley as case study. A comparison between the output of fuzzy pattern recognition model and the DRASTIC was accomplished. The fuzzy index is higher than DRASTIC index

References Afshar, A., Marino, M.A., Ebtehaj, M., Moosavi, J., 2007. Rule-based fuzzy system for assessing groundwater vulnerability. Journal of Environmental Engineering ASCE 133 (5), 532e540. Aller, L., Bennet, T., Lehr, H.J., Petty, J.R., Hackett, G., 1987. DRASTIC; a standardized system for evaluating groundwater pollution potential using hydrogeologic settings. USEPA-600/2-87-035, 622 pp. Babiker, I.S., Mohammed, A.A.M., Hiyama, T., Kato, K., 2005. A GIS-based DRASTIC model for assessing aquifer vulnerability in Kakamigahara Heights, Gifu Prefecture, central Japan. Science of the Total Environment 345 (1e3), 127e140. Barber, C., Bates, L.E., Barron, R., Allison, H., 1994. Comparison of standardized and Region-specific methods assessment of the vulnerability of Groundwater to pollution: A case study in an Agricultural catchment. In: Proceedings of 25th IAH Congress water Down under, Melbourne, Australia. Chen, S.Y., Fu, G., 2003. A DRASTIC fuzzy pattern recognition methodology for groundwater vulnerability evaluation. Hydrological Science Journal 48 (2), 211e220. Denny, S.C., Allen, D.M., Journeay, J.M., 2007. DRASTIC-Fm: a modified vulnerability mapping method for structurally controlled aquifers in the southern Gulf Islands, British Columbia, Canada. Hydrogeology Journal 15 (3), 483e493. Department of Hydrology and Metrology (DHM), 2006. Precipitation Records of the Bagmati Zone (1999e2005). Ministry of Environment, Science and Technology, Government of Nepal. Dixon, B., 2005. Groundwater vulnerability mapping: a GIS and fuzzy rule based integrated tool. Applied Geography 25 (4), 327e347. Doerfliger, N., Zwahlen, F., 1997. EPIK: a new method for outling of protection areas in karstic environment. International Symposium on Karst Waters and Environmental Impacts, pp. 117e123. El Yaouti, F., El Mandour, A., Khattach, D., Kaufmann, O., 2008. Modelling groundwater flow and advective contaminant transport in the Bou-Areg

D.R. Pathak, A. Hiratsuka / Journal of Hydro-environment Research 5 (2011) 63e77 unconfined aquifer (NE Morocco). Journal of Hydro-environment Research 2 (2008), 192e209. Foster, S.S.D., 1987. Fundamental concepts in aquifer vulnerability, pollution risk and protection strategy. Proceedings and Information/TNO Committee on Hydrological Research 38, 36e86. Frind, E.O., Molson, J.W., Rudolph, D.L., 2006. Well vulnerability: a quantitative approach for source water protection. Ground Water 44 (5), 732e742. Gautam, R., Rao, G.K., 1991. Groundwater resources evaluation of the Kathmandu Valley. Journal of Nepal Geological Society 7, 39e48. Gemitzi, A., Petalas, C., Tsihrintzis, V.A., Pisinaras, V., 2006. Assessment of groundwater vulnerability to pollution: a combination of GIS, fuzzy logic and decision making techniques. Environmental Geology 49 (5), 653e673. Gogu, R.C., Hallet, V., Dassargues, A., 2003. Comparison of aquifer vulnerability assessment techniques, application to the Neblon river basin (Belgium). Environmental Geology 44 (8), 881e892. Gomezdelcampo, E., Dickerson, J.R., 2008. A modified DRASTIC model for sitting confined animal Feeding Operations in Williams County, Ohio, USA. Environmental Geology 55 (8), 1821e1832. Jha, M.G., Khadka, M.S., Shrestha, M.P., Regmi, S., Bauld, J., Jacobson, G., 1997. The Assessment of Groundwater Pollution in the Kathmandu Valley, Nepal. Report on Joint Nepal-Australia project 1995e96. Australian Geological Survey Organization, Canberra, pp. 64. JICA, 1990. Groundwater management project in Kathmandu Valley, Final report, main report and supporting reports, November 1990. Kathmandu Valley GIS database. Towards bridging the data gap, 2000. MENRIS/ICIMOD. Khadka, M.S., 1993. Groundwater quality situation in alluvial aquifers of the Kathmandu Valley. AGSO. Journal of Australian Geology & Geophysics 14, 207e211. Khatiwada, N.R., Takizawa, S., Tran, T.V.N., Inoue, M., 2002. Groundwater contamination assessment for sustainable water supply in Kathmandu Valley, Nepal. Water Science and Technology 46 (9), 147e154. Lynch, S.D., Reynders, A.G., Schulze, R.E., 1997. A DRASTIC approach to groundwater vulnerability in South Africa. South African Journal of Science 93 (2), 59e65. Merchant, J.W., 1994. GIS-based groundwater pollution hazard assessment: a critical review of the DRASTIC model. Photogrammetry Engineering & Remote Sensing 60 (9), 1117e1127. Metcalf and Eddy, 18 February 2000. Urban Water Supply Reforms in the Kathmandu Valley (ADB TA Number 2998-NEP), Completion Report. In: Executive summary, main report and Annex 1 to 7, vols. I, II. Metcalf & Eddy, Inc. with CEMAT Consultants Ltd.. National Geographic Information Infrastructure Project (NGIIP), 1994. Department of Survey, Government of Nepal, Kathmandu, Nepal.

77

Nobre, R.C.M., Filho, O.C.R., Mansur, W.J., Nobre, M.M.M., Cosenza, C.A. N., 2007. Groundwater vulnerability and risk mapping using GIS, modeling and a fuzzy logic tool. Journal of Contaminant Hydrology 94, 277e292. National Research Council (NRC), 1993. Ground Water Vulnerability Assessment: Contamination Potential under Conditions of Uncertainty. National Academy Press, Washington, DC. Pathak, D.R., Hiratsuka, A., Awata, I., Chen, L., 2009. Groundwater vulnerability assessment in shallow aquifer of Kathmandu Valley using GISbased DRASTIC model. Environmental Geology 57 (7), 1569e1578. doi: 10.1007/s00254-008-1432-8. Rahman, A., 2008. A GIS based DRASTIC model for assessing groundwater vulnerability in shallow aquifer in Aligarh, India. Applied Geography 28 (1), 32e53. Shrestha, O.M., Koirala, A., Karmacharya, S.L., Pradhananga, U.B., Pradhan, P.M., Karmacharya, R., Hanisch, J., Kerntke, M., Joshi, P.R., Steiner, L., Busch, K., Jnwali, B.M., Maske, N.D., Tuladhar, G.B., Kaphle, K.P., 1998. Engineering and Environmental Geological Map of Kathmandu Valley. Scale 1:50000. Department of Mines and Geology, Nepal. Shrestha, O.M., Koirala, A., Hanisch, J., Busch, K., Kerntke, M., Jager, S., 1999. A geo-environmental map for the sustainable development of the Kathmandu Valley, Nepal. Geo-journal 49 (2), 165e172. Thirumalaivasan, D., Karmegam, M., Venugopal, K., 2003. AHP-DRASTIC: software for specific aquifer vulnerability assessment using DRASTIC model and GIS. Environmental Modeling and Software 18 (7), 645e656. Uricchio, V.F., Giordano, R., Lopez, N., 2004. A fuzzy knowledge e based decision support system for groundwater pollution risk evaluation. Journal of Environmental Management 73 (3), 189e197. United State Environmental Protection Agency (USEPA), 2009. 2009 Edition of the Drinking Water Standards and Health Advisories, EPA 822-R-09011, Office of Water U.S. Environmental Protection Agency, Washington, DC, pp 18. Vrba, J., Zaporozec, A., 1994. Guidebook on Mapping Groundwater Vulnerability. In: International Contributions to Hydrology, vol. 16. Heinz Heise, Hannover, 131 pp. Zadeh, L.A., 1965. Fuzzy sets. Journal of Information and Control 8 (3), 338e353. Zhang, R., Hamerlinck, J.D., Gloss, S.P., Munn, L., 1996. Determination of nonpoint-source pollution using GIS and numerical models. Journal of Environmental Quality 25 (3), 411e418. Zhou, H.C., Wang, G.L., Yang, Q., 1999. A multiobjective fuzzy pattern recognition model for assessing groundwater vulnerability based on DRASTIC system. Hydrological Science Journal 44 (4), 611e618.