Introduction

In view of increasing demand for freshwater for various purposes such as agricultural, domestic and industrial, a greater emphasis is being laid for a planned and optimal utilization of groundwater resources. Unfortunately, the mismanagement of this valuable resource has created not only serious groundwater depletion and pollution problems, but also water supply problems for both present and future generations (Mays 2013). In addition, groundwater depletion (reduction in groundwater storage) problems has resulted in other allied problems like extensive land subsidence, reduction in stream/river flows and spring discharges, loss of wetlands and deterioration of groundwater quality. It is, therefore, imperative to make a quantitative estimation of available groundwater resources at a basin or sub-basin scale. In this regard, groundwater prospect mapping can play an important role in the identification of zones with probable occurrence of groundwater as it provides a rational picture of hidden freshwater resources. Nowadays, the delineation of groundwater potential zones is becoming increasingly important for successfully implementing groundwater protection and management programs.

Conventional approach of groundwater exploration using geological, hydrogeological and geophysical methods normally involve high budget, and hence are uneconomical and time consuming for large-scale investigations (Sander et al. 1996). In addition, these methods of investigation do not always account for the various factors that control the occurrence and movement of groundwater (Oh et al. 2011). Remote sensing (RS) and the geographical information system (GIS) with their advantages of spatial, spectral and temporal availability and manipulation of data covering large and inaccessible areas within a short time have become very handy tools in accessing, monitoring and conserving groundwater resources (Jha and Peiffer 2006; Jha et al. 2007; Meijerink 2007; Rodell et al. 2009; Moiwo et al. 2011; Scanlon et al. 2012; Voss et al. 2013). Moreover, remote sensing is one of the main sources of information about the surface (hydrological) features related to groundwater such as lineament, soil, topography, land use/land cover and landforms. Such spatial data and information can be easily input to a GIS environment for integration with other types of data/information followed by suitable data analyses and mapping (Hinton 1996; Jha et al. 2007).

The application of remote sensing and GIS with or without multicriteria decision analysis for the exploration/mapping of groundwater prospect zones has been reported by a number of researchers across the world (e.g., Krishnamurthy et al. 1996; Sander et al. 1996; Saraf and Choudhury 1998; Jaiswal et al. 2003; Sener et al. 2005; Solomon and Quiel 2006; Srivastava and Bhattacharya 2006; Tweed et al. 2007; Madrucci et al. 2008; Chowdhury et al. 2009; Jha et al. 2010; Machiwal et al. 2011; Jasrotia et al. 2013).

However, studies on groundwater prospect mapping using GIS and probabilistic approaches are very limited. Although the frequency ratio (FR) and weight of evidence (WOE) probabilistic models have been applied and compared to examine landslide susceptibility mapping (e.g., Lee and Choi 2004; Lee and Dan 2005; Oh et al. 2009, 2010; Pradhan and Lee 2010), ground subsidence hazard mapping (Kim et al. 2006; Lee et al. 2010), very few studies have been reported till date as far as the mapping of groundwater prospect using probabilistic approaches is concerned. Oh et al. (2011) used GIS and FR model to map regional groundwater potential in the area of Pohang City, Korea. The thematic layers used included specific capacity, transmissivity, topography, lineament, geology, forest and soil data for mapping groundwater productivity potential. Lee et al. (2012) produced the regional groundwater productivity potential map for the bedrock aquifer of Pohang City, Korea, using GIS and WOE model and the same thematic layers as used by Oh et al. (2011). On the other hand, Ozdemir (2011) performed a comparative evaluation of FR, WOE and logistic regression models for mapping spring potential in the area of Sultan Mountains of Konya, Turkey. Seventeen spring-related parameter layers of the study area, viz., geology, fault density, distance to fault, lithologies, elevation, slope aspect, slope steepness, curvature, plan curvature, profile curvature, topographic wetness index, stream power index, sediment transport capacity index, drainage density, distance to drainage, land use/cover and precipitation were used to generate spring potential map of the study area. The results indicated that the FR and WOE models were relatively good estimators of spring potential mapping in the study area as compared to logistic regression.

It is obvious from the review and literature that most of the studies on application of RS and GIS technologies in the delineation of groundwater prospect have evaluated single method/approach only: either multicriteria decision analysis (MCDA) method (simple or AHP-based) or probabilistic method. To the best of the authors’ knowledge, there has been no comprehensive study to date concerning application of probabilistic modeling and comparative evaluation of MCDA and involving the application, assessment and comparison of both MCDA and probabilistic modeling approaches for mapping groundwater prospect. Therefore, there is a need for such studies so as to identify suitable method(s) for groundwater prospecting that can provide a high predictive accuracy. The present study demonstrates the efficacy and usefulness of two GIS-based approaches, viz., MCDA approach and probabilistic modeling approach, for the identification and delineation of groundwater prospect zones. To achieve this goal, the Kushabhadra–Bhargavi groundwater basin of Mahanadi Delta, Odisha, eastern India, was considered as the study area. The first approach includes integration of a multicriteria decision analysis technique, namely analytic hierarchy process (AHP) with remote sensing and GIS to identify groundwater prospect zones in the study area. The second approach includes GIS and two probabilistic models (FR and WOE) for the identification of groundwater prospect zones in the study area.

Methodology

Study area

Location and hydrometeorology

The study area selected for the present study is Kushabhadra–Bhargavi groundwater basin, which is situated in the Mahanadi Delta Stage-II region, Odisha, eastern India (Fig. 1). It is spread mostly in the Puri and Khurdha (to some extent) districts of Odisha. The salient information about the study area is summarized in Table 1.

Fig. 1
figure 1

Location map of the study area

Full size image
Table 1 Description of the study area
Full size table

The average annual rainfall in the study area is about 1470 mm and mostly occurs during monsoon season from mid-June to the end of October. The rainfall during the months of June, July, August, September and October is about 15, 27, 26, 17 and 6 % of the annual rainfall, respectively. Three distinct seasons prevail in the area. The winter season starts from November and lasts up to the end of February; the summer season extends from March to the middle of June; and the monsoon (rainy) season from mid-June to the end of October. During the monsoon period, the area experiences heavy rainfall from the southwest monsoon. The spring, autumn and dew seasons are of very short duration and are telescoped into the three major seasons.

Sources of water supply

Among the various sources of surface water, canal is the most important source of irrigation in the study area. Canal irrigation accounts for about 280 km2 of the area, with the remaining supply from other sources such as river-lift, ponds or tube wells. However, the amount of area under canal irrigation has a decreasing trend with an increase in the size of land holding because of the fact that the more affluent farmers have access to relatively expensive, but dependable sources of water supply. Exploitation of groundwater is desirable to: (1) supply additional water during dry periods, (2) relieve stress on reservoirs for irrigation during low rainfall years and (3) protect wet period crops from long dry spells during the rainy season. The aquifer system of the Mahanadi delta offers an important additional source of water resource, which needs to be utilized efficiently so as to ensure sustainable water supply in the study area.

Collection of data and extraction of relevant themes

To investigate groundwater characteristics in the study area, the pre-monsoon and post-monsoon groundwater depth data of 24 observation wells (Fig. 1) for the 1997–2011 period were collected from the Central Groundwater Board (CGWB), Bhubaneswar, and Groundwater Survey and Investigation (GWS&I), Bhubaneswar, Government of Odisha. Groundwater potential is governed by a variety of surface and subsurface factors such as topography, land use/land cover, geology, soil, drainage density, slope, stream network, rainfall, climate, recharge, lineament and human activities, among others.

On the basis of availability of spatial and field-measured data in the study area, the themes (factors) influencing groundwater occurrences were selected for identifying groundwater prospect zones. The selected themes of the study area, viz., geology, soil, elevation, slope, drainage density and land use/land cover, were generated using both remote sensing and conventional data. The geology map, soil map and land use/land cover map of the study area at a scale of 1:250,000 were collected from Odisha Space Applications Centre (ORSAC), Department of Science and Technology, Government of Odisha. The DEM of the study area extracted from Shuttle Radar Topography Mission (SRTM) was used for generating topographic elevation, slope and drainage density maps of the study area. Moreover, the drainage network map was generated from the Survey of India toposheets (scale 1:50,000). Apart from the above spatial data, the discharge data of 77 pumping wells over the study area were collected from Orissa Lift Irrigation Corporation (OLIC), GWS&I, Bhubaneswar, Government of Odisha. The location of the pumping wells is shown in Fig. 1.

Identification and delineation of groundwater prospect zones

Thematic layers of geology, land use/land cover, soil, slope and drainage density were used for the identification and delineation of groundwater prospect zones in the study area. In this study, two approaches were used for identifying groundwater prospect zones. In Approach I, multicriteria analysis technique AHP and GIS technique were employed, whereas in Approach II two probabilistic models, namely FR and WOE along with the GIS technique, were used. The procedures of these two approaches are described in the subsequent sections.

Approach I: groundwater prospect zoning using AHP and GIS

The methodology adopted for the preparation of a groundwater prospect map using Approach I is presented in Fig. 2. After generating the above-mentioned five thematic layers of the study area, the importance of their features from groundwater viewpoint was taken into consideration and accordingly suitable weights were assigned to each thematic layer and the features of individual thematic layers. For the assignment of weights, the comparison judgment scale of Saaty (1980) as shown in Table 2 was used and the opinions of the experts were sought. One pairwise comparison matrix was developed for the comparison of the thematic layers and five pairwise comparison matrices were developed for the comparison of the features of individual thematic layers. Thereafter, normalized weights were calculated by the geometric mean method and the consistency ratio (CR) was calculated for all the thematic layers and their individual features to check the consistency of the weights assigned. Further, all the five raster thematic layers (geology, land use/land cover, soil, slope, drainage density) were introduced in ArcView GIS as a grid layer. All the thematic layers were resampled to a grid size of IRS-1D-LISS-III image resolution 23.5 × 23.5 m. Arithmetic model for spatial raster data was developed for the integration of thematic layers using ArcView 3.2 software. The total weights of each grid (polygon) of the final integrated raster layer were derived by the weighted linear combination method expressed as follows:

$$ {\text{GWPI }} = {\text{ GG}}_{w} {\text{GG}}_{\text{wi}} + {\text{ LU}}_{w} {\text{LU}}_{\text{wi}} + {\text{ ST}}_{w} {\text{ST}}_{\text{wi}} + {\text{ SL}}_{w} {\text{SL}}_{\text{wi}} + {\text{ DD}}_{w} {\text{DD}}_{\text{wi}}$$
(1)

where GWPI is the groundwater potential index for a polygon, GG the geology, LU the land use/land cover, ST the soil type, SL the slope, DD the drainage density, ‘w’ the normalized weight of a theme and ‘wi’ the normalized weight of the individual features of a theme.

Fig. 2
figure 2

Protocol of the methodology adopted for the preparation of groundwater prospect map using Approach I

Full size image
Table 2 AHP scale and its interpretation (Saaty 2001; Eastman 2003)
Full size table

Approach II: groundwater prospect zoning using probabilistic models and GIS

In this approach, groundwater prospect maps were prepared by the FR and WOE probabilistic models. These models require the data on location of pumping wells over the study area, together with the thematic layers having significant influence on groundwater occurrence. A well location map (Fig. 3) was prepared depicting 77 pumping wells over the study area, out of which 55 were used as training wells and the remaining 22 were used as testing wells. The training wells were used in the FR and WOE probabilistic models, and the results were then implemented over the entire study area, including the test area. The testing wells were used solely for the verification of the modeling results. The selected thematic layers were overlaid with the well location map. On the basis of these intersections, the FRs and WOE probability (P) values were calculated for each of the thematic layers. The step-by-step procedures for the application of these two probabilistic models are presented below.

Fig. 3
figure 3

Location of training and testing wells of the study area

Full size image
  1. 1.

    Frequency ratio modeling

Frequency ratio is the probability of occurrence of a certain attribute (Bonham-Carter 1994). For groundwater prospecting, FR indicates the quantitative relationship between well occurrence and different causative parameters (i.e., factors). For determining the FR of groundwater potential, the ‘area ratio’ and ‘well occurrence ratio’ were calculated for different classes of each factor (thematic layer). Thereafter, FR for different classes of each factor was calculated by dividing the ‘well occurrence ratio’ with the ‘area ratio’. These FR values were used for generating a groundwater prospect map of the study area using the overlay function of GIS. The step-by-step procedures for FR modeling to delineate groundwater prospect zones are as follows:

  • Step 1: selection of thematic layers and the preparation of thematic maps.

  • Step 2: preparation of well location map and the selection of training and testing wells.

  • Step 3: overlaying of the thematic maps with the training well map.

  • Step 4: identification of pixels under different classes of a given factor (thematic layer).

  • Step 5: computation of ‘area ratio’ for a particular class of a given factor by dividing the total number of pixels present in that class with the total number of pixels present in the study area.

  • Step 6: calculation of ‘well occurrence ratio’ for a particular class of a given factor by dividing the number of training wells present in that class with the total number of training wells present in the study area.

  • Step 7: calculation of FRs for each class of a given factor by dividing the ‘well occurrence ratio’ with the ‘area ratio’.

  • Step 8: overlaying of the thematic layers and the computation of groundwater potential index (GWPI) over the study area. The pixel-wise GWPI over the study area was computed as follows:

    $$ {\text{GWPI}} = \sum\limits_{i = 1}^{N} {({\text{FR}})_{i} }$$
    (2)

    where FR is the frequency ratio of the ith factor and N is the total number of factors.

  • Step 9: preparation of a groundwater prospect map of the study area in the GIS environment based on the range of GWPI values over the study area.

  • Step 10: validation of the prepared groundwater prospect map using testing wells.

  1. 2.

    Weight of evidence modeling

The weight of evidence model calculates the weights for groundwater prospecting based on the presence or absence of wells in each groundwater influencing factors. Like the FR model, this model also requires a set of training and testing wells. The selected groundwater-related thematic layers were overlapped with the map depicting training wells. On the basis of these intersections, weight and WOE probability values were calculated for the individual classes of different thematic layers. The step-by-step procedures for WOE modeling to delineate groundwater prospect zones are as follows:

  • Step 1: selection of thematic layers (factors) and the preparation of thematic maps.

  • Step 2: preparation of well location map and the selection of training and testing wells.

  • Step 3: overlaying of the thematic maps with the training well map.

  • Step 4: identification of the pixels under different classes of a given factor (theme).

  • Step 5: calculation of weights (W +) for each class of a theme. Weights for the individual classes of different themes were calculated as:

    $$ W^{ + } { = }\,{ \ln }\,\left[ {\frac{{\frac{{{\text{Number}}\,{\text{of}}\,{\text{Wells}}\,{\text{in}}\,{\text{the Class}}}}{{{\text{Total}}\,{\text{Number}}\,{\text{of}}\,{\text{Wells in the Study Area}}}}}}{{\frac{{{\text{Number of Pixels}}\,{\text{in}}\,{\text{the}}\,{\text{Class}}\, - \,{\text{Number of}}\,{\text{Wells}}\,{\text{in}}\,{\text{the}}\,{\text{Class}}}}{{{\text{Number of Pixels}}\,{\text{in}}\,{\text{the Study Area}}\, - \,{\text{Total}}\,{\text{Number}}\,{\text{of Wells in the Study Area}}}}}}} \right] $$
    (3)

  • Step 6: calculation of WOE probability (P) for each class of a theme. The WOE probabilities of individual classes of different themes were calculated as:

    $$ P\, = \exp \,\left\{ {\sum W^{ + } \, + \,\ln \,P_{p} } \right\}$$
    (4)

    where P p is the prior probability of groundwater occurrence which was computed as:

    $$ P_{p} = \frac{\text{Total Number of Wells in the Study Area}}{\text{Total Number of Pixels in the Study Area}}$$
    (5)

  • Step 7: preparation of a groundwater prospect map of the study area based on the range of WOE probability values over the study area.

  • Step 8: validation of the prepared groundwater prospect map using testing wells.

Evaluation of groundwater prospect mapping methods

Finally, a comparative evaluation of the two approaches used for groundwater prospect mapping was performed on the basis of prediction accuracy calculated by comparing the groundwater prospect maps generated by the two approaches with the map of well yields.

Results and discussion

Groundwater characteristics of the study area

The variation of post-monsoon groundwater depths (mean of 1997–2011) over the study area is shown in Fig. 4. It reveals that the mean post-monsoon groundwater depth in the area ranges from 0.95 to 3.65 m below the ground surface (bgs). Similarly, Fig. 5 depicts the variation of mean pre-monsoon groundwater depths over the study area. Obviously, the mean pre-monsoon groundwater depth in the study area varies from 1.75 to 5.6 m below the ground surface. Moreover, the mean seasonal groundwater fluctuation ranges from 0.78 to 2.15 m (Fig. 6), with a major portion of the study area having a mean groundwater fluctuation of 1.25–1.5 m. Thus, the significant seasonal groundwater fluctuations in the study area indicate appreciable recharge to the aquifer during the monsoon season (wet period). It is recommended that future study should focus on the determination of safe yield/sustainable yield of the aquifer(s) underlying the study area.

Fig. 4
figure 4

Mean post-monsoon groundwater depth map of the study area

Full size image
Fig. 5
figure 5

Mean pre-monsoon groundwater depth map of the study area

Full size image
Fig. 6
figure 6

Mean groundwater fluctuation map of the study area

Full size image

Characterization of the thematic layers

The thematic layers or themes are characterized by classifying their features into suitable groups that can help interpret their importance in groundwater occurrence. The features of the six selected layers are described in the subsequent sub-sections.

Features of topographic elevation layer

The topographic elevation map of the study area is shown in Fig. 7, which indicates almost flat topography over the study area, with the elevation varying from 0 to 26 m MSL. Thus, on the basis of topographic elevation, the study area can be divided into 11 classes as illustrated in Fig. 7.

Fig. 7
figure 7

Elevation map of the study area

Full size image

Features of slope layer

The prevailing slope in the study area varies from 0 to 4 % (Fig. 8). The slope statistics of the study area are presented in Table 3, which reveals that a major portion of the study area (nearly 79 %) falls under the 0–1 % slope category. This slope class can be considered as ‘very good’ for groundwater occurrence due to nearly flat terrain and hence relatively high infiltration potential. The area having 1–2 % slope can be considered as ‘good’ from groundwater viewpoint due to slightly undulating topography with some runoff. On the other hand, the area having a slope of 2–4 % is likely to produce relatively high runoff and low infiltration and hence can be categorized as ‘moderate’.

Fig. 8
figure 8

Slope map of the study area

Full size image
Table 3 Slope statistics of the study area
Full size table

Features of geology layer

The statistics and map of geology of the study area are presented in Table 4 and Fig. 9, respectively, which indicate two types of geology features, viz., alluvium and laterite. A major portion of the study area (434 km2) is covered by laterite formation which encompasses about 61 % of the total area and is found in the middle and southern portions of the study area. Laterite is somewhat a porous rock formation, which can form potential aquifers along topographic lows and moderate groundwater potential can be expected in this area. Further, about 39 % of the study area (i.e., 186 km2) is occupied by alluvium formation. Alluvium constitutes a very good water-bearing formation and is supposed to be an important source of groundwater in the deltaic regions.

Table 4 Geology statistics of the study area
Full size table
Fig. 9
figure 9

Geology map of the study area

Full size image

Features of land use/land cover layer

Land use/land cover plays an important role in deciding the extent of infiltration rate and recharge rate. In the study area, six major land use/land cover categories, namely agricultural land, dense forest, degraded forest, wasteland, settlements, and rivers and water bodies are found (Table 5). The spatial distribution of land use/land cover in the study area is illustrated in Fig. 10.

Table 5 Land use/land cover statistics of the study area
Full size table
Fig. 10
figure 10

Land use/land cover map of the study area

Full size image

Features of soil layer

The soil map showing the variation of soil types in the study area is shown in Fig. 11 and the soil statistics are presented in Table 6. The soil classes found in the study area are: silty loam, clayey loam, coarse sand, very fine sand and sandy loam. According to their hydraulic characteristics, these soil classes can be considered as ‘very good’, ‘good’, ‘moderate’ and ‘poor’ depending on their contribution to groundwater recharge.

Fig. 11
figure 11

Soil map of the study area

Full size image
Table 6 Variation of soil types in the study area
Full size table

Features of drainage density layer

The density of surface drainage indirectly indicates groundwater potential of an area due to its relation with surface runoff and permeability, and hence it is considered as one of the hydrological indicators of groundwater occurrence in the study area. Drainage density map of the study area was prepared on the micro-watershed basis and reveals that the drainage density ranges from 0 to 2.8 km/km2 (Fig. 12). Based on the drainage density of the micro-watersheds, the drainage density was grouped into five classes, the statistics of which are presented in Table 7. It is apparent that a major portion of the study area (nearly 88 %) falls under 0–0.50 km/km2 drainage density category, which can be considered as ‘very good’ for groundwater occurrence. Similarly, the remaining drainage density classes such as 0.50–0.75 km/km2 can be considered as ‘good’, 0.75–1 km/km2 as ‘moderate’ and >1 km/km2 can be considered as ‘poor’ from the viewpoint of groundwater potential.

Fig. 12
figure 12

Drainage density map of the study area

Full size image
Table 7 Drainage density statistics of the study area
Full size table

Groundwater prospect maps of the study area

Groundwater prospect map based on Approach I

In this approach, RS, GIS and AHP techniques are integrated for identifying and delineating groundwater prospect zones. The weights were assigned to all the thematic layers using Saaty’s AHP scale as shown in Table 8. Normalized weights derived from the pairwise comparison matrix are presented in Table 9. Similarly, the weights assigned to all the features of individual themes were normalized and are presented in Table 10. The CR of the pairwise comparison matrix for all the thematic layers was found to be 0.00324, i.e., less than the threshold limit of 0.10 which indicates that the weights assigned to the thematic layers are consistent. The weights assigned to different features of individual thematic layer were also found to be consistent (CR < 0.10).

Table 8 Weights of the five thematic layers for groundwater prospect zoning using Saaty’s AHP scale
Full size table
Table 9 Pairwise comparison matrix and normalized weights of the five thematic layers
Full size table
Table 10 Normalized weights of the five thematic layers and their features
Full size table

The normalized weights of the five thematic layers and their features were aggregated and GWPI was computed in the GIS environment. Finally, a groundwater prospect map of the study area was generated by dividing the study area into four distinct zones based on the GWPI values, i.e., ‘very good’ (GWPI = 0.33–0.40), ‘good’ (GWPI = 0.29–0.33), ‘moderate’ (GWPI = 0.25–0.29) and ‘poor’ (GWPI = 0.04–0.25) as shown in Fig. 13.

Fig. 13
figure 13

Groundwater prospect map of the study area based on Approach I

Full size image

It is apparent from Fig. 13 that the groundwater prospect in the northern and southern coastal parts of the study area can be categorized as ‘very good’ and ‘good’. In other words, these portions of the study area have quite favorable hydrological/hydrogeological conditions for groundwater storage and hence these areas are expected to be rich in groundwater.

Groundwater prospect statistics of the study area based on Approach I are presented in Table 11, which reveals that the area covered by ‘very good’ and ‘good’ groundwater prospect zones is about 240.74 km2 (≈39 %) and these zones are dominated in the northern and southern portions of the study area. The ‘moderate’ groundwater prospect zone is prevalent in the central and southern portions of the study area and encompasses an area of nearly 313.3 km2, which is about 51 % of the total study area. The hydrogeomorphic feature available in this portion of the study area is laterite type of geology, which also suggests moderate capacity of groundwater storage. ‘Poor’ groundwater prospect zone covers an area of 64.54 km2 (about 10 %) and is distributed in scattered patches in the central and southern portions of the study area.

Table 11 Groundwater potential statistics of the study area based on Approach I
Full size table

Groundwater prospect maps based on Approach II

The groundwater prospect maps prepared by using two probabilistic models, viz., FR and WOE, are presented below with the help of examples. The number of wells and the number of pixels present in individual features of a factor (thematic layer) are summarized in Table 12.

Table 12 Number of pixels and number of wells in the individual features of the factors affecting groundwater
Full size table
  1. 1.

    Frequency ratio modeling results

The calculation of FR for the alluvium feature of the ‘geology’ factor is presented below as an example:

  • number of wells present in the alluvium feature of the geology factor = 23,

  • total number of wells in the study area = 55,

  • number of pixels in the alluvium feature of the geology factor = 203,858,

  • total number of pixels in the geology factor = 685,829,

  • percentage of wells for the alluvium feature of the geology factor = (23/55) × 100 = 41.8 and

  • percentage of area for the alluvium feature of the geology factor = (203,858/685,829) × 100 = 29.7.

  • Therefore, FR for the alluvium feature of the geology factor is 41.8 ÷ 29.7 = 1.407.

Similarly, FR for each feature of the thematic layers was calculated as shown in Table 13. The GWPI obtained from the FR values provided a basis for identifying groundwater prospect zones. Based on the GWPI values, a groundwater prospect map of the study area was generated by dividing the study area into four distinct zones: ‘poor’ (GWPI = 0–3.92), ‘moderate’ (GWPI = 3.92–5.05), ‘good’ (GWPI = 5.05–6.47) and ‘very good’ (GWPI = 6.47–12.48) as shown in Fig. 14. Groundwater potential statistics of the study area as obtained from the FR model are presented in Table 14.

Table 13 Frequency ratio (FR), weight of evidence (W +) and WOE probability (P) for the five factors
Full size table
Fig. 14
figure 14

Groundwater prospect map of the study area based on the FR model

Full size image
Table 14 Groundwater potential statistics of the study area based on the FR model
Full size table

It is discernible from Fig. 14 that the groundwater potential zones ‘very good’ and ‘good’ occur mainly in the northern part and in some patches scattered over the study area. The area covered by ‘very good’ and ‘good’ groundwater potential zone is about 236 km2 (38 %), while the ‘moderate’ groundwater potential zone encompasses an area of 322 km2, which is 52 % of the total study area (Table 14). Thus, the ‘moderate’ groundwater potential zone is dominating in the study area. Further, the southern part and scattered small patches in the central portions fall in the ‘poor’ groundwater potential zone that covers an area of 63 km2 (10.2 %).

  1. 2.

    Weight of evidence modeling results

As an example, the calculation of WOE (W +) and WOE probability (P) for the alluvial feature of the geology factor is illustrated below:

  • number of wells present in the alluvium feature of the geology factor = 23,

  • total number of wells in the study area = 55,

  • number of pixels in the alluvium feature of the geology factor = 203,858, and

  • total number of pixels in the geology factor = 685,829. Therefore,

    $$ W^{ + } = { \ln } \frac{23/55}{(203858 - 23)/(685829 - 55)} = 0.3414, $$

  • P P  = 55 ÷ 685,829 = 0.00008019 and

  • P = exp {0.3414 + ln (0.00008019)} = 0.0001128.

Similarly, the WOE (W +) and WOE probability (P) for each feature of the five thematic layers (factors) were calculated as summarized in Table 13. Finally, a groundwater prospect map of the study area was generated by dividing the study area into four distinct zones based on the WOE probability values, i.e., ‘poor’ (GWPI = 0–0.00168), ‘moderate’ (GWPI = 0.00169–0.00275), ‘good’ (GWPI = 0.00276–0.00526) and ‘very good’ (GWPI = 0.00526–0.00786). The map showing various groundwater potential zones of the study area as obtained from the WOE model is shown in Fig. 15. It can be seen from Fig. 15 that the groundwater potential in the northern portion and some parts of the central portion of the study area fall under ‘good’ groundwater potential zone, but the ‘very good’ groundwater potential zones occur in a couple of patches in the northern, central and southern parts of the study area. The area covered by ‘very good’ and ‘good’ groundwater potential zone is about 44.4 % (Table 15). The central portion, some part of the southern portion and some patches in the northern portion of the study area fall under ‘moderate’ groundwater potential zone, which constitute 48 % of the total study area. The lower southern part and a few small patches/strips in the central and northern parts of the study area have ‘poor’ groundwater potential encompassing an area of 7.6 % (Table 15).

Fig. 15
figure 15

Groundwater prospect map of the study area based on the WOE model

Full size image
Table 15 Groundwater potential statistics of the study area based on the WOE model
Full size table

Efficacy of the multicriteria decision analysis and probabilistic modeling

Prediction accuracy of Approach I

The verification of Approach I using measured well yields of 77 tubewells reveals that 30 ‘high-discharge’ (>100 m3/h) wells out of 40 exist in the ‘good’ and ‘very good’ zones, and out of the remaining 10 wells, 7 exist in the ‘moderate’ zone and 3 in the ‘poor’ zone. Furthermore, 4 out of 34 ‘medium-discharge’ (60–100 m3/h) wells fall in the ‘good’ zone, 28 in the ‘moderate’ zone and the remaining 2 in the ‘poor’ zone. Based on these findings, the prediction accuracy of Approach I is 76.62 %; its calculation is given below:

  • total number of tube wells = 77,

  • number of tube wells where there is an agreement between the expected and the actual yield = 59 and

  • number of tube wells where there is a disagreement between the expected and the actual yield = 18.

  • Therefore, the prediction accuracy of Approach I = 59 ÷ 77 = 76.62 %.

Prediction accuracy of Approach II

The verification of Approach II was performed using measured well yields of 22 testing tube wells in the study area. The verification of the groundwater prospect map based on the FR model reveals that five out of eight ‘high-discharge’ wells exist in the ‘good’ zone, two in the ‘moderate’ zone and one in the ‘poor’ zone. However, 2 out of 11 ‘medium-discharge’ wells exist in the ‘good’ zone, 8 in the ‘moderate’ zone and 1 in the ‘poor’ zone. Based on these findings, the prediction accuracy of the FR model is 68.18 % as computed below:

  • total number of testing tube wells = 22,

  • number of tube wells where there is an agreement between the expected and the actual yield = 15 and

  • number of tube wells where there is a disagreement between the expected and the actual yield = 7.

  • Therefore, the prediction accuracy of the FR model = 15 ÷ 22 = 68.18 %.

On the other hand, the verification of the groundwater prospect map based on the WOE model reveals that six out of eight ‘high-discharge’ wells exist in the ‘good’ zone and two in the ‘moderate’ zone. However, 2 out of 11 ‘medium-discharge’ wells exist in the ‘good’ zone, eight in the ‘moderate’ zone and one in the ‘poor’ zone. Based on these findings, the prediction accuracy of the WOE model is 72.72 % as estimated below:

  • total number of testing tube wells = 22,

  • number of tube wells where there is an agreement between the expected and the actual yield = 16 and

  • number of tube wells where there is a disagreement between the expected and the actual yield = 6.

  • Therefore, the prediction accuracy of the WOE model = 16 ÷ 22 = 72.72 %.

It is clear from the above verification results that although the results of both the approaches are satisfactory, Approach I is much superior to Approach II (both WOE and FR models). Among the two probabilistic models used for groundwater prospect mapping, the prediction accuracy of the WOE model is higher than that of the FR model.

Spatial prediction of groundwater prospect: comparative evaluation

A comparison of the areas under the four identified zones ‘very good’, ‘good’, ‘moderate’ and ‘poor’ as predicted by the three methods (AHP, FR and WOE) is illustrated in Fig. 16. It is evident from this figure that the total areas under ‘moderate’ and ‘poor’ zones predicted by the FR model are more or less the same as those predicted by the AHP technique; differences in the areas vary from 2.78 to 2.39 %, respectively, with respect to the AHP technique. However, the total area under ‘very good’ zone is underpredicted (57.78 %) and that under ‘good’ zone is overpredicted (40.63 %) by the FR model as compared to the AHP technique (Fig. 16). In contrast, the spatial distribution of the zones over the study area as predicted by the FR model varies appreciably, with a high level of disagreement with the spatial distribution of the zones predicted by the AHP technique (Figs. 13, 14). The ‘moderate’ groundwater prospect zone as predicted by the FR model and AHP technique is dominant and located in the central and southern portions of the study area. However, the major discrepancy in the spatial prediction by these two techniques is found in the southernmost part of the study area. This is because of the absence of pumping wells in the southernmost portion of the study area (Fig. 3) which provides the basic criteria for FR modeling.

Fig. 16
figure 16

Area under different groundwater potential zones predicted by the AHP, FR and WOE techniques

Full size image

On the other hand, the spatial distribution of the zones predicted by the WOE model matches with that predicted by the AHP technique to a greater extent as revealed by Table 16, Figs. 13 and 15. A major difference in the spatial prediction is found only in the southernmost portion of the study area due to the absence of pumping wells in this region. It should be noted that the presence/absence of pumping wells provides the basic criteria for WOE modeling, whereas AHP technique is dependent on only hydrologic/hydrogeologic factors. As a result, the AHP technique predicts good/very good potential of groundwater because of favorable hydrologic/hydrogeologic factors in the southernmost portion, while the WOE model predicts moderate and poor potential of groundwater in this portion. Furthermore, for the WOE model, the total area under ‘poor’ zone differs by 27.18 % as compared to the AHP technique, while the total area under ‘moderate’ potential zone differs by 5.2 %. Similarly, the differences in the total areas under ‘very good’ and ‘good’ potential zones vary from 79.85 to 86.04 %, respectively, with respect to the AHP technique (Fig. 16).

Table 16 Comparison of spatial prediction of different groundwater prospect zones by the AHP, WOE and FR techniques
Full size table

Thus, the findings of the comparative evaluation of spatial distribution of groundwater prospect zones are in agreement with the results of prediction accuracy of the probabilistic models, i.e., the reliability of AHP technique is greater than the probabilistic models (FR and WOE). Among the probabilistic models, the performance of the WOE model is superior to that of the FR model. Therefore, the AHP technique is strongly recommended for better identification and delineation of groundwater prospect zones in the study area. The WOE model can serve as an alternative technique for the study area. However, considering the basic requirement of adequate data on number of pumping wells over a basin/sub-basin for the WOE model, the practical applicability of this model is limited, especially in data-scarce countries.

Conclusions

The present study deals with the evaluation of the efficacy of two approaches for spatial prediction of groundwater prospect through a case study. Five significant thematic layers related to groundwater occurrence in the study area were considered for both the approaches. The first approach includes integrated use of RS, GIS and multicriteria decision analysis (AHP), whereas the second approach includes GIS-based probabilistic modeling using FR and WOE models. The analysis of the spatial modeling results of the two approaches indicated that the groundwater prospect map obtained by Approach I (AHP technique) demarcated the study area with the ‘very good’ zone encompassing about 17 % of the study area, ‘good’ zone 22 %, ‘moderate’ zone 51 % and the ‘poor’ groundwater prospect zone covering about 10 % of the study area. On the other hand, the groundwater prospect map obtained by the FR model (Approach II) reveals that about 7 % of the study area falls in the ‘very good’ groundwater potential zone, 31 % in the ‘good’ zone, 52 % in the ‘moderate’ zone, and 10 % in the ‘poor’ groundwater potential zone. In addition, the groundwater prospect map obtained by the WOE model (Approach II) reveals that about 3 and 41 % of the study area fall in the ‘very good’ and ‘good’ groundwater potential zones, respectively, while 48 % falls in the ‘moderate’ zone and 8 % falls in the ‘poor’ groundwater potential zone. The validation results of the two approaches suggested that the prediction accuracy of the AHP technique (Approach I) is about 77 %, that of the WOE model about 73 % and that of the FR model about 68 %.

Based on the findings of this study, it can be inferred that the performance of the AHP technique (Approach I) is much superior to the probabilistic models (FR and WOE models), though the performance of the WOE model can be considered somewhat comparable with that of the AHP technique. Therefore, for more reliable results, the AHP technique is recommended for identifying groundwater prospect zones in a basin/sub-basin in general and in the study area in particular. If adequate data are available in an area or region for probabilistic modeling, the use of the WOE model should be preferred to the FR model. Groundwater prospect maps are very useful for the cost-effective selection of suitable sites for well drilling as well as for the effective planning and development of groundwater resources. The methodology demonstrated in this study is also applicable to other regions of the world, irrespective of hydrologic and hydrogeologic conditions.