Introduction

Water is one of the essential constituents for sustaining all forms of life among all renewable resources on earth, which exists in three forms, i.e., solid, liquid and gaseous states. It is a unique gift to mortality from nature, and it is tough to de-pollute, expensive to transport, and impossible to substitute its most uses (Kumar et al. 2005). Three fourth area of the earth's surface is covered with water, out of which the ocean holds about 96.54% that, importantly saline in nature, only 2.8% is available as freshwater (USGS). Over time, these available water sources are also decreasing at an alarming rate due to the rapid increase of water demand resulting in an increasing number of water-scarce regions. It derives the most significant challenge of supplying adequate water to meet social needs (Ayob and Rahmat 2017). According to (McDonald and Mitchell 2019), worldwide remarkable dropping of freshwater availability put several countries to suffer from water scarcity. Water scarcity is now a global issue; significant numbers of people live in severe water-scarce regions. (Seckler et al. 1999) was estimated that several regions of developing countries, having one-third of the total population, will experience severe water scarcity in the near future. Out of 2.8% freshwater, only 0.76% is available as fresh groundwater (GW), stored beneath the earth's surface through GW recharge (USGS). Water infiltration from unsaturated zone to saturated zone is identified as GW recharge (Yeh et al. 2016); usually formed by entry of rain and snow melts into underlying rock structure through soil (Nampak et al. 2014).It is a prime source of fresh water in many parts of the world and is essential for human well beings.

As a dynamic and essential renewable natural resource, GW has a crucial role in global climate change and satisfying human needs (Chatterjee et al. 2010; Duan et al. 2016). Seckler et al. (1999) exposed that about 30% of freshwater resources are regarded as GW resources, the second most abundant freshwater resource of the globe. It is the only available water free from pollution due to no direct contact with the earth's surface, contributing around 34% of the total annual water supply (Magesh et al. 2012). The consumption of GW has been increasing rapidly day by day due to the failure of surface water in meeting necessary demands in the daily life of humankind (de Vries and Simmers 2002; Das and Pal 2019). In recent years, the world has been facing threatening conditions regarding the availability of groundwater due to its excessive use. The large dependency of several regions in India and fast consumption of this valuable resource in comparison to its natural recharge leads to a significant decline in the water table beneath the surface (Rodell et al. 2009). In India, 90% of rural domestic water usage is based upon GW. In comparison, 70% of the freshwater used for agricultural purposes is extracted from aquifers, and about 50% of urban water usage is based upon GW (Das et al. 2019). GW is replenishable in nature but finite in distribution; therefore, we should be more concerned about sustainable use (Chatterjee and Purohit 2009). At present, the resiliency of the aquifer system is under a threatening condition due to unsustainable use of GW and evokes a dubitation about the ability to provide a long-term water source (Richey et al. 2015). In the recent era, the problem of rural water supply, drought problem, irrigation project and low cost of development draws attention towards GW management (Kamila et al. 2018); therefore, identification of GW potential zone as well as GW recharge area is critical to maintaining water quality and better management of GW system (Waikar et al. 2014). The GW occurrence and movements of a particular region are affected by several topographical (geology, surface slope, geomorphology, lineament density, NDVI, soil texture) and hydrological factors (drainage density, rainfall, NDWI, GW level); considering all above controlling factors groundwater potential zone (GWPZ) can be identified of an area (Jha and Peiffer 2006; Chowdhury et al. 2009; Singh et al. 2019). Until now, several methods have been adopted to identify the GWPZ of a particular area; traditionally, test drilling and stratigraphy analysis (Sander et al. 1996), soil moisture model and hydrogeological field investigation model (Thorpe and Scott 1999; White et al. 2003). However, all those in-situ investigation methods are cost-effective and notably time-consuming. Remote sensing (RS) and GIS integrated platforms have become a quicker and cost-efficient alternative to identify GWPZ (Saraf et al. 2004; Solomon and Quiel 2006; Gupta and Srivastava 2010; Hatti 2011; Ghosh et al. 2015; Thapa et al. 2018). Several statistical techniques (MCDA, AHP and Fuzzy logic) were also applied by different researchers (Machiwal et al. 2011; Jhariya et al. 2016; Arabameri et al. 2020; Pal et al. 2020; Mallick et al. 2021) by using expert knowledge or literature values. In groundwater data analysis, strong and flexible analytical methods are required (Rahmati et al. 2015). In such a case, advance data-driven model can aid to delineate groundwater potential zone. Recently various machine learning (ML) algorithms and artificial intelligence such as logistic regression (LR), decision tree, ANN (artificial neural network) random forest, SVM (support vector machine) get huge importance due to large data handling capacities and remarkable accuracy level. Proper approach and modeling are essential to finding out groundwater potentiality and planning for the water-scarce region.

This study has applied a comprehensive GIS technique with the three most significant ML algorithms like LR, random forest, and SVM to delineate GWPZ of Purulia district, India. Stage of groundwater development is an important index to explain groundwater availability in a particular area; based on it, several Safe Category blocks are 215, Semi-Critical blocks are 53, and Critical block is 1 in West Bengal and same way Purulia has a good position in the stage of groundwater development which is 13.78% (GEC 1997). Despite that, Purulia is notoriously famous as a water-scarce region in every summer season. The village women of this area walk a very long distance along village streets in search of water. Various factors are responsible for water scarcity in Purulia. The rapid growth of the population in this area is highly responsible for widening the gap between demand and supply of water in recent times. Very few studies have been done regarding the GW issues of Purulia (Das et al. 2019). Our study intends to close the research gap in the existing literature. The primary objective of our study is to find out the groundwater potential zone (GWPZ) of the Purulia district through LR, Random forest and SVM with the help of the RS-GIS platform. The final zonation map will help predict potential groundwater areas and will also be helpful in proper planning to manage this valuable resource for the future.

Study area

The Purulia district is situated in the south-western part of West Bengal, between 22°43′ and 23°42′ North latitude and between 85°49′ and 86°54′ East longitudes (Fig. 1). It occupies 5th position in the state according to its area, which is 6259. The district is surrounded by several districts of Jharkhand and West Bengal. Hazaribagh–Dhanbad, Singbhum, and Ranchi districts of Jharkhand are situated in the North, South, and West of Purulia, respectively. On the eastern side, it is bounded by the Bankura, Burdwan, and Midnapore districts of West Bengal. Physio graphically Purulia is located at a transition zone between the plateau of Chotanagpur and the young alluvial plains of West Bengal (Mahato and Gupta 2016). As a part of the Chotanagpur plateau, undulating rugged topography, hills, high ridges, and low valleys are the main characteristics of this study area. The eastern part consists of a rolling upland which changes ultimately into a rugged and dissected topography in the west. The study area has numerous streams namely Kangsabati, Damodar, Darakeswar, Kumari, and Subarnarekha. All the rivers are east and south east-flowing, except for Subarnarekha which flows in the south direction. There are some alluvial areas found in very narrow strips along the rivers. Kangsabati is the master stream of the area. Though many rivers flow through the region, more than 50% of it runs off due to undulating topography, resulting in low infiltration. Porosity and permeability are very low due to hard rocky terrain surfaces; that is mainly dominated by metamorphic rocks. Shallow low fractures and weathered mantle are suitable for GW storage which remains in unconfined or semi-confined condition. Purulia district mainly falls under sub-tropical climate and is characterized by low precipitation and high evaporation, making it a drought-prone area of West Bengal. The district represents extreme climatic conditions with a very low temperature of 5 °C and maximum temperature of 46 °C in winter and summer months, respectively. Average annual rainfall of 1000–1600 mm occurs in the month of mid-June to September by southwest monsoons. There are three types of soil namely residual types derived from weathering of igneous rocks and metamorphic rocks, lateritic soils in the upland areas, clayey loam to clay in the valleys. The soil pH ranges between 5.5 and 7.2 (Lok Kalyan Parishad, Purulia); 75.05 thousand hectares of area under forest (District Industrial Profile Purulia, 2018–19). The study area is endowed with natural forest in the North-Western part and some areas of Ajodhya Hills and Panchet hills in the Northeast. The forest is mainly composed of Sal trees with different other species like Kusum, Palash, Mahua, Sirish, Neem, Simul, etc.

Fig. 1
figure 1

Location map of study area

Full size image

Materials and methodology

Figure 2 represents the methodological framework used in this study.

Fig. 2
figure 2

Methodological framework adopted for this study

Full size image

Materials used

A truthful GWPZ map and its accuracy level depend on the size and availability of the data sets as well as used modeling approaches to produce the maps. Some factors which affect the groundwater potentiality have been considered in this study through literature review. They are Geology, Slope, Lineament Density, Normalized Difference Vegetation Index (NDVI), Geomorphology, Drainage Density, Normalized Difference Water Index (NDWI), Rainfall, GW level, and Soil texture. The maps of all these factors are prepared by the collection of data from different sources. Geology and geomorphology map of the study area has been collected from the Geological Survey of India (GSI). Slope, lineament density, and drainage density maps are prepared from Shuttle Radar Topography Mission 1 arc-second for global coverage (USGS, 30 m resolution). NDVI and NDWI maps were prepared from Landsat 8 (USGS) data. Rainfall and groundwater level maps are prepared by Inverse Distance Weighting (IDW) method with the help of Indian Meteorological Department rainfall data and Central Groundwater Control Board (CGWB) respectively. Soil texture data have been collected from the National Bureau of Soil Survey and Land Use Planning (NBSS and LUP), Government of India. ArcGIS 10.4 and ERDAS IMAGINE 2014 software have been used for the representation of the data. R software and SPSS statistical software were used for adopted ML algorithms to delineate GWPZ with the help of available data sets.

Methodology

Multi-collinearity analysis

In a regression model to ensure the accuracy, testing of independence of each factor is very important (Chen et al. 2011). Multi-collinearity occurs when more than one variable is significantly correlated to each other in a regression model; a big problem can be resulted through a small percentage of multicollinearity (Daoud 2017). For this study, several causative factors were used, therefore it is essential to identify the linear relationship among the variables through this test which helps to exclude associated variables to eliminate the error chances and increase the accuracy in adopted models. In this study, variance inflation factors (VIF) and Tolerance (TOL) techniques play a significant role in defining errors. The TOL and VIF were calculated as follows:

$${\text{TOL}} = 1 - r^{2}$$
(1)
$${\text{VIF}} = \frac{1}{{{\text{TOL}}}}$$
(2)

where the VIF value > 10 and TOL value < 0.1 shows multicollinearity issue (Khosravi et al. 2019). For present study, the threshold value for GWPZ were considered to be less than 5 VIF value.

Logistic regression

Logistic regression is the most widely used multivariate statistical techniques in diverse field of science, significantly in environmental science (Nguyen et al. 2020; Chowdhuri et al. 2020). In a generalized linear method this empirical model is based on logit link function; by considering several conditioning factors it enhances the prediction accuracy (Yariyan et al. 2020). In LR, categorical outcomes can be predicted i.e., occurrence and non-occurrence; possibility of occurrence of an event is assessed within the range of 0–1. The presence and absence of attribute can be predicted by considering several independent variables in binomial LR when the dependent variables are in binomial nominal level whereas multivariate LR is used to measure the relationship between various independent variables with one dependent variable. Generally, LR model expressed in following equations (Ozdemir 2011; Park et al. 2017):

$$P = \frac{{e^{z} }}{{\left( {1 + e^{z} } \right)}}$$
(3)

where the probability of occurrence is indicated by P and the linear combination of the independent variable by Z. Z is expressed as follows:

$$Z = \alpha + \beta_{1} x_{1} + \beta_{2} x_{2} \ldots \, \ldots \, \ldots + \beta_{n} x_{n}$$
(4)

where \(\alpha\) defines the intercept, n is the number of independent variables whereas \({\beta }_{n}\) and \({x}_{n}\) represent the regression coefficients and independents variables, respectively. Z value ranges from \(-\infty\, \mathrm{to}+ \infty\); the positive regression coefficient defines the positive correlation of dependent variable with independent variables whereas the negative regression indicates the negative relation in between dependent and independent variables.

Support vector machines

Support vector machine learning is one of the most significant supervised algorithms used as a classifier besides regression problems; it is performed based on statistical learning theory and structural risk minimization (SRM) principle (Tien Bui et al. 2012; Tehrany et al. 2015). The SRM principle is an important factor in SVR in identification of the relationship between input and output variables (Saha et al. 2021). However, these learning algorithms are mostly used for classification purposes by developing hyperplane which is decision boundary to segregate n-dimensional space into classes. (Vapnik 1998) proposed that by maximizing the margin between the classes, hyperplane classify the linear data. The epsilon tolerance margin is considered, that makes the difference between SVM and SVR which is the regression version of SVM (Smola and Schölkopf 2004).Generally, the SVR modelling approach applied in case of very complex dataset; several curved margins are developed to solve this dataset (Kalantar et al. 2018).

Following mathematical formula can describe the SVM model:

$$\frac{1}{2}\left| {\left| W \right|} \right|^{2}$$
(5)

The following constraints of subject is

$$y_{i} = \left( {w \cdot x_{i} } \right) + b \ge 1$$
(6)

where, \(|\left|W\right||\) is the hyperplane, \(b\) is the scalar base, (∙) indicates the scalar product. The cost function of SVM can be defined by using Lagrangian multiplier:

$$L = \frac{1}{2}\left| {\left| W \right|} \right|^{2} - \mathop \sum \limits_{i = 1}^{n} \lambda_{i} \left( {y_{i} \left( {\left( {w \cdot x_{i} } \right) + b} \right) - 1 } \right)$$
(7)

where,\(\lambda_{i}\) indicates Lagrangian multiplier. In the case of non-separable function the constraints can be modified by introducing slack variables

$$y_{i} \left( {w \cdot x_{i} } \right) + b \ge 1 - \zeta_{i}$$
(8)

And finally the equation becomes as follow

$$L = \frac{1}{2}\left| {\left| W \right|} \right|^{2} - \frac{1}{vn}\mathop \sum \limits_{i = 1}^{n} \zeta_{i}$$
(9)

where, \(v\) (0, 1) represent the generated to account for misclassification. In addition to this, kernel function \(K\left( {x_{i} ,x_{j} } \right)\) was introduced by Vapnik in the year of 1995 to explanation for non-linear decision boundary.

Random forest

Random forest (RF) decision tree algorithm was first developed by Breiman (2001); as an advanced version of bagging, randomness added to it consisted of tree predictors and each tree depends on the values of a random vector sampled autonomously with the similar distribution for all trees in the forest (Breiman 2001). Two types of trees such as classification trees and regression trees are separated from decision trees (Rodriguez-Galiano et al. 2014). RF as a flexible ensemble learning algorithm constructs a set of classifiers rather than one classifier based on multiple decision tree helps to produce repeated predictions of similar phenomenon (Ok et al. 2012; Micheletti et al. 2014; Zabihi et al. 2016; Islam et al. 2021). RF classification algorithm helps to develop new training sites by choosing sample training sites. In this study, RF machine learning approach was used to identify key variables for mapping GWPZ. Random forest requires two components (Goel and Abhilasha 2017):

$${\text{GE}} = P_{x,y} \left( {mg\left( {x,y} \right) < 0} \right)$$
(10)
$$mg\left( {x,y} \right) = a_{vk} I\left( {h_{k} \left( x \right) = y} \right) - \max_{j \ne k} av_{k} I\left( {h_{k} \left( x \right) = j} \right)$$
(11)

where x and y are groundwater potential parameters reflecting the probability of x and y, the edge part of the indicator is mg and I( ∗).

Model validation

Model validation is a prominent part to evaluate a model that can prove the scientific reliability of the result. Therefore, the widely applied Receiver Operating Characteristics (ROC) and Area Under Curve (AUC) curve are used to determine the accuracy of the models(Chen et al. 2018; Gayen et al. 2019). The key focus of this spatial modelling approach is to quantify the successful and unsuccessful events on the basis of model testing data; it works independently to assess the model’s predictive ability of a specific probability threshold. According to Mandrekar (2010) this curve is regarded as the average value of the specificity of a test over all possible values of a sensitivity or vice versa. AUC values range between 0.5 and 1, where values closer 1 denote excellent performance and nearer to 0.5 indicate the very poor predicting accuracy (Sameen et al. 2020). In this study, three models were adopted to delineate GWPZ, therefore comparative review of the adopted models executed by this validation technique. Dou et al. (2019) asserts that the trapezoidal rule integral calculus helps to calculate AUC value:

$${\text{AUC}} = \mathop \sum \limits_{k = 1}^{n} \left( {X_{k + 1} - X_{k} } \right)(S_{k + 1} - S_{k} - S_{k} /2)$$
(12)

where area under curve represented by AUC and in case of specificity and sensitivity is represented by \({X}_{k}\) and \({S}_{k}\), respectively.

Result and discussion

Multi-collinearity analysis

Extensive literature review helps to identify several groundwater conditionings factors. In the present study, 10 groundwater affecting factors were identified after positive multi collinearity analysis. The results of multi collinearity analysis were shown in Table 1. The results display that the lowest TOL is 0.27 and the highest value of VIF is 3.71; that indicate there are no multi-collinearity issues among adopted 10 groundwater conditioning factors.

Table 1 Multi-collinearity test of groundwater potential factors
Full size table

Geology

Geology is a major controlling factor in the groundwater potentiality of a particular area that generally affects groundwater recharge by supervising water flow percolation(Elewa and Qaddah 2011) and subsequently it is associated with water permeability (Oikonomidis et al. 2015). The occurrence, movement, and qualities of groundwater are significantly affected by the geological units of an area (Rajaveni et al. 2017). According to the Geological Survey of India, the Purulia district consists of seven separable major lithostratigraphic units which are Chotanagpur gneissic complex, unclassified metamorphic, Singhbhum GP, Dalma volcano, Manbhum granite, Kuilapal granite series, and some places covered with sediments (Fig. 3). The study area is mainly characterized by numerous rocks and minerals such as granite, gneiss, schist, phyllite, quartzite, sandstone, shale, mica, feldspar, china clay which are abundant in nature. Chotanagpur gneissic complex is present throughout the whole district with an area of 3465 (55.38%). The southern portion of the district is mostly dominated by the Singhbhum GP series which covers 1159 (18.53%) area. Some part of this district is covered with Dalma volcanic series which have an area of 338 (5.41%). Some part of the northern and central region is dominated by manbhum granite and unclassified metamorphic which are distributed in a scattered way with 552 (8.82%) and 531 (8.5%) area respectively. Sedimentary rocks are also present in a very small part of the central region of the district with an area of 135 (2.16%). According to Dolui et al. (2016), Proterozoic hard granite gneiss including soft phyllite and mica-schist of the Singhbhum group are the dominant rocks in the Purulia district as a part of the Chotanagpur plateau. Hardy rocks like granite, gneiss, quartzite, feldspar, etc. are mostly found in this area; that makes the barrier to infiltrate the surface water and increase the surface runoff which is not suitable for groundwater potentiality. But there is some strip area covered with sandstone which is very preferable for groundwater potentiality in this district.

Fig. 3
figure 3

Geology map

Full size image

Geomorphology

Geomorphology of a particular area plays a dominant role in estimating the storage and movement of groundwater (Sinha et al. 1990; Dinesh Kumar et al. 2007). Many researchers (Waters et al. 1990; Krishnamurthy et al. 1996; Sahu and Sahoo 2006) have successfully adopted geomorphology as a principal parameter in delineating groundwater potentiality. The identification of the geomorphologic features is very important to delineate groundwater potential zone. Pediment pediplain, dissected hills, valleys are the main geomorphic features of this region but water bodies, active flood plain, rivers are situated in a very small part of this region (Fig. 4), (Geological Survey of India). Most of the areas of the district are dominated by high hills and sloppy land. Pediment and Pediplain complex are present in almost the entire district covering an area of 5458 (85.97%). Apart from it, 9% of the district is covered with high hills and valleys which are not suitable for groundwater potentiality. Ramaiah et al. (2012) stated that structural hill, residual hill, and linear ridge landforms are not suitable for GWP due to its un-fractured rock characteristics having a low infiltration rate. Only a few areas of this district are covered with water bodies, dams and reservoirs, rivers, and ponds consisting of only around 7% of the study area. The water bodies play a very significant role in the groundwater potentiality of this rugged and hardy terrain. As a result, only a few areas have high potentiality.

Fig. 4
figure 4

Geomorphology

Full size image

Lineament density

Generally geologic structures such as fractures, faults and discontinuous surfaces are defined as lineaments that can be identified by RS-GIS techniques (O’Leary et al. 1976); that is architecture of the rock basement (Hobbs 1904) and significantly important in hydrological studies which works as pathways of groundwater movement (Abdalla 2012) and as a curvilinear feature of earth having a significant role in GW exploration (Pothiraj and Baskaran 2013). Lineament is basically a permeable zone and has positive relation with high GWP (Haridas et al. 1998). According to Krishnamurthy et al. (2000), the buffer area of 300 m around lineament and faults is considered as suitable groundwater recharge and potential zone. Lineament density (\(L_{{\text{d}}}\)) is expressed by following equation:

$$L_{{\text{d}}} = \frac{{\mathop \sum \nolimits_{i = 1}^{i = n} L_{i} }}{A}$$
(13)

where \(\sum\nolimits_{i = 1}^{i = n} {L_{i} }\) defines the total length of lineaments and A is the total area \((L^{2} )\).

The distribution of lineament concentration is not uniform throughout the district. This has been classified into five groups namely very low, low, medium, high and very high-density area. Among them maximum portion of the study area has very low lineament density zone with around 5273 km2 (83%). Some part of the northern region, southern region and western hilly region are fall under low 475 km2 (7.48%), moderate 300 km2 (4.73%), high 205 km2 (3.23%) and very high density 93 km2 (1.43%) area (Fig. 5). Very small region of western hilly area of the district at Bagmundi, have greater lineament density which is suitable for groundwater recharge and increases groundwater potentiality. Similarly, the entire study area having very low lineament density is mostly responsible for least groundwater recharge.

Fig. 5
figure 5

Lineament density map

Full size image

Slope

Slope gradient is one of the most important parameters in zone identification of groundwater potentiality of an area; it has a direct influence on the infiltration of surface water (Selvam et al. 2015). Generally, the slope is the ratio of change in elevation between two points and has an inverse relation with GWP. On a steep slope, water flows downward quickly having inadequate time to infiltrate that resulting in very low GW recharge (Machiwal et al. 2011; Magesh et al. 2012). Whereas, gentle slopes provide an environment for GW recharge due to widespread high retention of surface water. According to Saraf and Choudhury (1998), a break in the slope accelerates infiltration leading to more groundwater recharge. Here, the study area is categorized into five zones viz, very low, low, moderate, high, and very high. Among them, the greater part of the study area is characterized by undulating topography which falls under very low (0–2 degrees) and low (2–5 degrees) categories covering 3408 (53%) and 2275 (35%) areas, respectively. This region with a very low degree of slope is very suitable for groundwater recharge. Western, Southern and NE parts of the district are under moderate (5–12 degree), high (12–22 degree) and very high (22–65 degree) category with an area of 461 (7.27%), 142 (2.25%) and 59 (0.93%) respectively (Fig. 6). These three regions are not suitable for groundwater recharge and can be designated as low groundwater potentiality areas.

Fig. 6
figure 6

Slope map

Full size image

Soil texture

Punmia and Jain (2005) asserts that to delineate GWPZ, the soil cover is an essential factor; Porosity and permeability of soil plays a significant role in the infiltration of surface water which is based on soil characteristics such as texture, depth, and water transmission capacity of an area that control the magnitude of groundwater recharge (Terence and Viessman 1977). Generally, soil texture means the composition of soil that refers to small (clays), medium (silts), and large (sands) size particles. Groundwater recharge is significantly controlled by soil texture because different soil texture has different infiltration rate; fine-grained soil has very low GWP compared to Coarse-grain soil because of its low level of porosity and permeability. This study area is characterized by very thin soil cover, containing sandy and reddish laterite material resulting from granite and gneiss through weathering process (Das et al. 2019). Different parts of this district are characterized with different soil textures such as gravelly loam–loam soil constitutes 1704 (27%), fine loam soil covered 1989 (31%), fine soil fall under the area of 1418 (22.45%), gravelly loam and fine loamy- coarse loamy soil present in the 378 (5.98%) and 772 (12.23%) area, respectively (Fig. 7). This gravelly loam to loam, gravelly loam, and coarse loam soil has moderate porosity and permeability which is preferable to groundwater potentiality and fine and fine loam soil have the least porosity and permeability which leads to the least potential zone.

Fig. 7
figure 7

Soil texture map

Full size image

Normalized difference vegetation index (NDVI)

NDVI is a very common and broadly used remote sensing index (Bhandari et al. 2012); generally, it helps to measure live green vegetation and also analyzes the characteristics of vegetation cover of a certain area. The groundwater enrichment can be directly reflected by surface vegetation (Huajie et al. 2016). The quantity of green vegetation in a particular area is an essential factor of groundwater recharge; if a large area is covered with vegetation, it will be suitable for groundwater recharge because the area will gain water holding capacity and get sufficient time to infiltrate the surface water in the groundwater layer. As a result, the groundwater availability of that area increases. This spatial index derived through following equation:

$${\text{NDVI}} = \frac{{{\text{NIR}} - {\text{RED}}}}{{{\text{NIR}} + {\text{RED}}}}$$
(14)

where NIR represent the near infrared band and RED is red band in satellite image.

The largest part of the study area is covered with light vegetation and few parts have very high vegetation cover in a scattered way. The entire area is classified into five NDVI zones: very low, low, moderate, high and very high region. Among them maximum part of Purulia district (Fig. 8) fall under moderate, high and very high vegetation type covering 2594 km2 (41%), 2097 km2 (33%) and 755 km2 (11%) area respectively. These areas are responsible for moderate to high groundwater potentiality. The study area also has low to very low vegetation which covers 778 km2 (12%) and 122 km2 (2%) area respectively but they are not suitable and results low ground water potentiality.

Fig. 8
figure 8

NDVI map

Full size image

Drainage density

Avtar et al. (2011) defined drainage density as the total length of the entire stream in a particular basin divided by its total area; that is a quantitative measurement of the length of stream channels (Singh et al. 2014) and also as spacing of stream channels in a specific area. It helps to measure the drainage concentration in a specific area which denotes how well and poorly drained the area is by stream channels. A lot of work (Bagyaraj et al. 2013; Jenifer and Jha 2017; Thomas and Duraisamy 2018; Andualem and Demeke 2019) established the inverse relation between drainage density and permeability of aquifers; which plays an important role in the runoff distribution and level of infiltration. There is a close relationship between drainage density and infiltration rate of the surface water in a specific area. It has a direct relationship with slope but is inversely proportional to permeability. Based on the result, this study area is categorized into five drainage density zones. They are very low, low, moderate, high and very high zones covering 1736 (27.3%), 1529 (24.9%), 1515 (23.86%), 1071 (16.87%), and 495 (7.81%) of the study area (Fig. 9) respectively. The result depicts that average drainage densities are found throughout the district. High drainage density is the result of more surface runoff which is not suitable for the groundwater potential zone but indirectly it helps in groundwater recharge in those areas where the slope is very low. The western part of this district, as well as some areas of the northern and southern parts have steep slopes causing a high concentration of drainage which is not suitable for GWP. But most of the area is characterized by a gentle slope which is very suitable for the groundwater potentiality of the Purulia district.

Fig. 9
figure 9

Drainage density

Full size image

Groundwater level

The groundwater depth of an area fully depends on the recharge and discharge of the subsurface water and it shows the spatial differentiation in the depth of water level of a specified area. This is an important factor for GWPZ (Jhariya et al. 2021); the connection between the geological formation and average water depth carries the ability of its water transmission, storage, and discharge capacity (Indhulekha et al. 2019). Therefore, where the shallow water depth indicates high water availability and deeper water levels define water scarcity instead of its abundance. Based on groundwater depth data, this area is categorized into five groups such as very low, low, medium, high, and very high groundwater level zone; which ranges from 1.65 to 8.15 m (Fig. 10). Higher water depth area denotes lower groundwater potentials and lower depth denotes higher groundwater potentials in this area.

Fig. 10
figure 10

GW level map in the study area

Full size image

Rainfall

Rainfall plays a vital role in groundwater recharge of a particular region; without rainfall, recharge of groundwater is impossible. Purulia district is a part of the Western Plateau region as a result it falls under the dry tropical climatic regions. The study area experiences maximum rainfall only in the monsoon period (Jun–Sep) (Fig. 11). According to the Govt. of West Bengal (1985), the average annual rainfall of the study area is around 1200–1400 mm. The key source of rainfall in this region is the southwest monsoon. The study area experienced a drought situation throughout the year except for monsoonal months. Eastern and South-Eastern parts of this district receive a high amount of rainfall compared to other regions. It indicates high groundwater potentiality in these particular areas. The western part of this district receives the least amount of rainfall and is not suitable for groundwater potentiality. This little variation of rainfall does not make any greater variability of groundwater in this district because here the major dominating factor is the geology and geomorphology which control the surface runoff of rainwater due to hardy rock terrain.

Fig. 11
figure 11

Rainfall map

Full size image

Normalized difference water index (NDWI)

NDWI is mainly used to delineate the different water bodies as surface water viz, ponds, reservoirs, rivers, lakes etc. which are more responsible for sub surface water in a region with the location of permeable rocks. There is a strong relationship between surface and sub-surface water; the reasonable number of surface water source makes good groundwater resource by normal hydrologic cycle. The communication between surface water and groundwater is mostly controlled by an intricate relationship between hydrological, geological, geomorphological, climate and landscape factor. This index is calculated as follows:

$${\text{NDWI}} = \frac{{{\text{GREEN}} - {\text{NIR}}}}{{{\text{GREEN}} + {\text{NIR}}}}$$
(15)

NDWI calculated through green band and NIR band of satellite image.

In this study, Purulia district is divided into five areas on the basis of the result which can help to understand the amount of surface water situation in this region. Those five categories are very low (9%), low (42%), moderate (41%), high (6%) and very high (2%) which cover the entire area (Fig. 12). This is clear that very few areas are under the high and very high surface water condition but it is least influence to groundwater due to its small areal extent throughout the district. Majority of the study area have less amount of NDWI index which implies least amount of surface water present in this area which have negative impact on GWPZ.

Fig. 12
figure 12

NDWI map

Full size image

Groundwater potential zone and validation

Groundwater potential zone

Groundwater potential zonation mapping is an essential task for sustainable use and planning of groundwater management. As Purulia district is a drought-prone area, it is necessary to know about the exact location and zone of groundwater storage which in turn helps to make proper water management. In this study, three machine learning methods such as LR, RF, and SVM methods have been used to delineate the GWPZ of the Purulia district. In the current study, the stepwise LR model was used to analyze the data and the model coefficient was calculated by using SPSS software; therefore, that was used to produce GWPZ (Fig. 13a). According to the LR model, the Positive coefficient plays a significant role to make groundwater recharge whereas negative values indicate insignificant roles in groundwater potentiality (Table 2). Dominating groundwater affective factors are as follows Singbhum GP (9.632), Highly dissected structural hills and valleys (5.023) and Gravelly Loam–Loam (3.025) whereas, Dalma volcanics (−1.547), fine soil (−0.623), and rainfall (−0.01) have an insignificant role in GW recharge in this area. SVM method helps to reclassify all the data layers and make a reliable result in GWPZ mapping (Fig. 13b). RF modeling approach also gave the most suitable and acceptable result in this study area due to its small spatial coverage (Fig. 13c); by calculating relative importance values (Fig. 14). The results illustrate that rainfall (77.83), GW level (67.44), slope (63.98), geology (61.32), soil (59.14) and lineament density (55.17) have great impact whereas, NDVI (39.58), NDWI (24.58), drainage density (30.96) and geomorphology (45.36) have the least impact on groundwater potentiality. All the adopted models in this study construct more or less similar results for groundwater potentialities in this region. Based on this result, the Purulia district has been categorized into five potential zones, those are very low, low, moderate, high, and very high area; spatial coverage all the models are shown in (Table 3). Very low, low potentiality zones are scattered in the Northern, North Eastern, Western, Southern, and Central parts of the district. The potentiality areas under high and very high zones are also found throughout the district in a scattered way which is more or less half of the study area. This assessment indicates that groundwater potentiality is restricted throughout the majority portion of this district while it is relatively high in a small segment of the north-eastern and north-western part of the study area; water scarcity is a common thing here because of rugged terrain and hardy rock surface. Though sufficient water potentiality is present in some places of the district, we have to manage it appropriately.

Fig. 13
figure 13

Delineation of groundwater potentiality by using—a LR, b SVM, c RF

Full size image
Table 2 Groundwater potential factors, their classes and respective logistic regression coefficients
Full size table
Fig. 14
figure 14

Determining the relative importance of causative factors using RF

Full size image
Table 3 Area coverage (%) of different zones by different methods
Full size table

Validation of groundwater potentiality

The Receiver Operating Characteristics (ROC) curve has been used in this study for reviewing the result of groundwater potentiality zone with the help of the current water depth data of groundwater which has been taken from CGWB as a substitute for groundwater storage data and 484 data points also used to validate the produced result (Fig. 15). Assessment of the models reveals that AUC values are 0.801, 0.841, and 0.878 for the LR, SVM, and RF, respectively (Fig. 16). Narkhede (2021) claimed that where the AUC value is 0.7, there is a 70% chance to differentiate among positive and negative classes conversely where the value is greater than 0.7 that means the reliability of this study is more than 70%. This evaluation established that RF modeling approach is most effective to delineate GWPZ in the present area followed by SVM and LR.

Fig. 15
figure 15

Validation point of groundwater potentiality

Full size image
Fig. 16
figure 16

Validation of the models, using ROC–AUC curve, a LR, b SVM, c RF

Full size image

Conclusion

A precise assessment of the GWPZ of a particular area especially in water-scare hard rock region plays a significant role in proper water resource management, land use planning, and also in environmental protection. In recent times, RS-GIS technique with data-driven algorithm has been emerged and increasingly implemented as a very powerful and cost-effective technique. It serves as a useful tool for delineating the groundwater potential zone of Purulia district which is one of the water-scarce regions of West Bengal. The assessment was performed by applying different kinds of ML algorithms such as LR, SVM, and RF; which gives more or less similar results. These coupled algorithms help to identify GWPZ based on different causative factors; multi-collinearity analysis was also used to eliminate any collinearity issues in this study. GW level, rainfall, slope and geology are the most influencing factors for prospective GWPZ in the study area. The study indicates that some parts of this district are not suitable for groundwater recharge and results in low groundwater potentialities because of rugged terrain; very low potentiality area is minimal in comparison to other categories and moderate potentiality is more common in the entire Purulia district. It is evident that porous soil texture, low slope, permeable geological strata, high rainfall, the highest frequency of water bodies, sufficient vegetation cover, and higher lineament concentration are the favorable conditions for the development of GWPZ. The success rate of the ROC curve also depicts the efficiency of all adopted models and the resulting output is significantly reliable. So, this assessment will be helpful in the successful detection of suitable locations for extraction of further GW and to make planning for better management of this valuable groundwater resource in the Purulia district.