Introduction

Groundwater is an essential subsurface natural wealth (Arulbalaji et al. 2019), that is crucial for household, agricultural and industrial use in both cities and rural areas (Kumar et al. 2020), while intrinsically being vital in the ecological perspective. Nearly 70% of the groundwater extracted in India is used in the agricultural sector (GEC 2015) and a majority of households also use it for domestic purposes (Murmu et al. 2018). Currently, groundwater contributes about one-third of the total water supply obtained annually (Shekhar and Pandey 2015) and about 75% of households face extreme water stress from its scarcity (NITI Aayog 2017–18), making this a most vital natural resource (Das 2017). Globally, overdrafting, human actions and unsuitable farming practices have significantly augmented groundwater pollution (Deepika et al. 2020). Therefore, gauging the groundwater availability, quality and recharge potential in an area is extremely important for its sustainable management (Kumar et al. 2020; Shekhar and Pandey 2015).

Conventionally, groundwater surveys that use geological, geophysical and hydro-geological tools are expensive and time-taking, especially for detailed surveys (Mondal et al. 2008; Ganapuram et al. 2009). However, advances in GIS and remote sensing-based methods have enabled delineation of prospective groundwater zones in a more feasible and possibly focused manner (Adji and Sejati 2014; Moghaddam et al. 2015; Senanayake et al. 2016). Researchers have often used diverse methods such as probabilistic models like certainty factor (CF) (Razandi et al. 2015), frequency ratio (FR) (Razandi et al. 2015; Das and Pardeshi, 2018), logistic regression (LR) (Pourtaghi and Pourghasemi 2014), decision trees (Chenini and Ben Mammou 2010; Duan et al. 2016; Chen et al. 2020), artificial neural network (ANN) (Naghibi et al. 2018), fuzzy logic (Rajasekhar et al. 2019; Ahmad et al. 2020), multi-criteria decision-making (MCDM) (Acharya et al. 2019) and also machine learning techniques like random forest model (Golkarian and Rahmati 2018) and maximal entropy (Rahmati et al. 2016), to delimit and map potential groundwater zones. Among these techniques, AHP has been repeatedly used in different climatic and physiographic settings for successfully predicting groundwater prospects.

A review of the recent literature in this domain reveals that several parameters have been used manifold times for potential groundwater zone identification [e.g. slope, lineament density, stream network density, rainfall, geomorphology and land use/land cover (LULC)] (see Table S1 in the Supplementary Information file), while others are quite unconventional, e.g. Topographic Wetness Index (TWI) (Mukherjee and Singh 2020; Bera et al. 2020; Qadir et al. 2020), Cumulative Sand Thickness (CST) (Kaur et al. 2020), DI (Nair et al. 2017), Roughness (Ro) (Arulbalaji et al. 2019), Distance from River (DR) (Allafta et al. 2021; Doke et al. 2021), Aquifer Thickness (AT) (Jena et al. 2020), Modified Normalized Difference Water Index (MNDWI) (Biswas et al. 2020), Soil Texture (STx) (Achu et al. 2020), Andualem and Demeke 2019; Das et al. 2019), Topographic Position Index (TPI) (Arulbalaji et al. 2019), depending upon the geographical setting of the study area. Table S1 and Fig. S1 (both provided in the Supplementary Information file) also show the usage trend of the various thematic parameters that are most often employed in demarcating groundwater potential zones (GWPZs). Therefore, the thematic layer selection is by nature quite arbitrary and diversified across different studies (Patra et al. 2018). Slightly different from the above, Shahinuzzaman et al. (2021) investigated GWPZs based on a cost-efficient parsimonious approach using catastrophe theory and AHP in a GIS environment. They used some novel parameters, based on concepts from hydrogeological studies, like distance from surface water bodies, soil permeability, top aquitard thickness, aquifer thickness, hydraulic conductivity, specific yield, recharge, top aquifer resistivity and aquifer resistivity. These parameters provide fresh insight and allow newer variable formulations/combinations in the field of delineating GWPZs.

Of all the different methods applied for mapping GWPZs, RS-GIS based techniques are the most cost-effective, and efficient at providing first-look insights that can be further verified using field surveys, which is crucial (Chakrabortty et al. 2018; Arabameri et al. 2020). Thus, it effectively helps to narrow the area to be investigated and prioritises locations where such explorations can be undertaken, saving time, effort and expense. Over the last two decades, combinations of RS-GIS with MCDM frameworks have become quite suitable ways of addressing spatial data management (Murmu et al. 2018). The AHP is one such MCDM method conceptualized for effectively choosing the best possible alternative under conflicting criteria with absolute and relative measurements (Saaty 1986) and a feedback system framework (Saaty and Takizawa 1986). It is a method wherein problems are divided into different variables, arranging them into relative hierarchical structures and thereafter harmonizing the results derived (Saaty 1986).

Here, an amalgamation of raster and vector-based datasets to guide GWPZ demarcation is undertaken using combined RS-GIS and AHP methods. The aim is to demarcate GWPZs along a part of the Western Ghats that has a unique interaction of disparate physiographic elements, consisting of both plateau and tidal coastal tracts. It is also essentially a potentially good site for tourism, which can further lead to socioeconomic development in the area. Towards this, identifying adequate groundwater reserves can enable expansion of tourism activities and augment local use and farming alongside suitable conservation measures for this vital resource. This study can therefore be used to devise a well-defined groundwater policy for this part of coastal Karnataka.

Study area

Located between the districts of Uttara Kannada and Shimoga in Karnataka, the Aghanashini or Tadri River (River Basin Atlas of India 2012; Tirodkar et al. 2014) originates from Manjuguni near Sirsi (Ramachandra et al. 2015). It flows westward for nearly 121 km (Balachandran et al. 2012) before debouching into the Arabian Sea near the coastal town of Tadri (Ramachandra et al. 2017). Its two major tributaries, the Bakurhole and Donihalla, originate from Sirsi Taluk and after flowing south-southwestward, meet near Mutthalli, about 16 km south of Sirsi (Ramachandra et al. 2015), forming the Tadri River. The Tadri flows through the Western Ghats in a southerly direction through various gorges. The basin lies between 74.32°–74.92° E and 14.26°–14.62° N coordinates (Fig. 1), with a total catchment size of 1440 km2 (Avinash and Ashamanjari 2011). The basin elevation ranges from being at sea level near the river’s estuary to > 750 m in the southwestern parts of the basin, with an average altitude of 396 m (Avinash and Ashamanjari 2011). The Tadri Basin consists of various geomorphological units such as lower floodplains, pediment complexes, hills and valleys that are constituted primarily of igneous and metamorphic rocks (granites to schists, shales, quartzites and phyllites). The area receives about 3980 mm of rainfall on average annually, mostly occurring between May and September, while the average minimum and maximum temperatures are around 21 °C and 31 °C, respectively (Sharannya et al. 2021). The overall groundwater quality is within the safe limits as per the Bureau of Indian Standards (BIS). There are some places which have been identified as having a higher concentration of nitrates, which may be caused by local use of fertilizers within the river basin (CGWB 2012b). The yearly discharge of the basin is nearly 966 million m3 (Bhat 2003). Three dominant land use and land cover (LULC) types can be observed within this basin, i.e., forests covering the slopes and plateaus in the central parts of the basin, agricultural lands over the small fluvial deposits along the rivers and in the eastern section of the basin and finally plantations and fallow lands (Ramachandra et al. 2015), while the rest of the area is covered mostly by water bodies, orchards, urban settlements and barren lands, respectively. The lower areas near the coast are suitable for farming activities. Paddy and horticulture plantations which include teak, coconut and acacia are the prime agricultural outputs (Koppad and Janagoudar 2017), with limited irrigation facilities available (Noori and Inamati 2021), mainly in form of tanks and wells (Nagaraj 2020; Raveesha 2020).

Fig. 1
figure 1

Location of the Tadri River Basin in western India and the elevation distribution across the basin

Full size image

Materials and methods

Preparation of thematic layers

The entire workflow has been depicted in Fig. 2. For assessing the existing GWPZs, eight distinct data layers (Fig. 3 and Fig. 4) were used (Table 1). These were chosen considering their significant effects and reliability in controlling the groundwater scenario, as gleaned from the literature (cf. Saravanan et al. 2020). These parameters were weighed depending upon their relationship with the GWPZ and their respective classes (as ascertained from previous studies and expert knowledge about the study area), so that criterion having a larger value were more important in determining the ambient groundwater situation. The Survey of India (SoI) topographical sheets numbered 48 J/6, 48 J/7, 48 J/10, 48 J/11, 48 J/7 48 J/14 and 48 J/15, each of scale 1:50,000, were used for depicting the study area and verifying the basin extracted from DEM analysis. The following vector layers at the mentioned scales were obtained from GSI databases- lithology (1:2,000,000), geomorphology (1:250,000) and lineament (1:2,000,000), while the soil map (1:2,000,000) was digitized from the National Atlas and Thematic Mapping Organization’s (NATMO) South Indian plate. Slope and drainage density maps was prepared using the 30 m spatial resolution (1 arc-second) SRTM DEM obtained from USGS Earth Explorer. From this DEM, we successively generated the flow direction (using the commonly preferred D-8 method), flow accumulation and basin outline rasters and subsequently these were converted into vector layers, with the Tadri Basin being extracted from this. Lineaments were obtained from GSI Quadrangle sheets. A density raster was then created in ArcGIS using the line density tool to obtain the lineament density per km2. Landsat-8 OLI scenes were classified to create the basin LULC layer, using the maximum likelihood function in ArcGIS 10.5. The CHIRPS 2.0 global annual precipitation data for the years 1981 to 2019 was used to derive the rainfall spatiality herein. SoI topographical maps were used to accurately locate and digitize all the groundwater wells. These wells were then selected at random to prepare the well density map, which was later used to validate the ROC function output.

Fig. 2
figure 2

Schematic flowchart showing the process of identification of GWPZs in the Tadri Basin and the sensitivity analysis

Full size image
Fig. 3
figure 3

Raster parameters considered for identifying GWPZs—a Drainage Density, b Lineament Density, c Rainfall and d Slope

Full size image
Fig. 4
figure 4

Vector parameters considered for identifying GWPZs—a Lithological units, b Geomorphological units, c Soil types and d LULC classes

Full size image
Table 1 Datasets used for GWPZ identification in the Tadri River Basin
Full size table

Assignment of weights

At the core of such prospective groundwater zone mapping techniques (as used here), lies the respective weights allotted to the different thematic layers and their sub-classes. This assignment of weights to each parameter is essential to their subsequent integration and analysis as the outcome is primarily dependent on these (Muralitharan and Palanivel 2015). Such weight allocations, in the MCDM-AHP method, are determined mainly by the authors’/users’ expertise and knowhow along with their understanding of the interaction/impact of these different parameters with the ambient groundwater conditions in the study area (Patra et al. 2018).

The AHP method was applied as it can clearly incorporate the users’ perceptions and judgments, based on expert opinion and observations (Liu et al. 2021) into a multi-level hierarchical format arranged as per their influences in the decision-making process (Saaty 1986). It simplifies the otherwise complex assemblage and linkages of multiple criteria and their judgement into a pecking order through a pairwise analogy of the relative influence of the different criteria on each other and their sub-criteria (Saaty 2005). Additionally, the AHP technique is well suited for assessing the analysis’ consistency of the work and induces comparatively less bias in the elicited results (Arulbalaji et al. 2019). The normalization process, through the pairwise judgement matrix, was used to reduce the subjectivity of each parameter and these were weighed on Saaty’s 1 to 9 point scale, where 1 denoted equal importance between two different parameters and 9 denoted excess influence of one parameter with respect to the other (Table S2 in the Supplementary Information file). The weights assigned to each criterion were carefully considered based on already published studies and their assumed influence on groundwater availability. The sub-criterions of each criterion were also classified, categorized and ranked between 1 and 9, depending on their impact on the groundwater development (Table S3). The steps to calculate the weights of the different parameters and to find the consistency were-

Step 1: Summation of values column-wise in the pairwise matrix (Table S4):

$${A}_{j}=\sum_{i,j}^{n}{G}_{ij}$$
(1)

where \({A}_{j}\) = total values of each column and \({G}_{ij}\) = the criteria-wise value given at the \({i\mathrm{th}}\) row and \({j\mathrm{th}}\) column.

Step 2: Dividing the matrix elements by the column sum to obtain the standardized (normalised) matrix (Table 2):

$${X}_{ij}=\frac{{G}_{ij}}{{A}_{j}}$$
(2)

where \({X}_{ij}\) = the value at \({i\mathrm{th}}\) row and \({j\mathrm{th}}\) column of standardized pairwise matrix.

Table 2 Normalised pair matrix
Full size table

Step 3: Dividing the standardized row total by the frequency of criteria (N) to obtain standard weights:

$${W}_{i}= \frac{\sum {\mathrm{X}}_{\mathrm{ij}}}{\mathrm{N}}$$
(3)

where \({W}_{i}\) is the standard weight.

Step 4: Consistency analysis: after calculating the standard weights, consistency analysis needs to be done. The AHP method tries to assess any ambiguity present through this consistency analysis. These were computed as:

$$\mathrm{CI}= \frac{\lambda -n}{n-1}$$
(4)

where, CI = Consistency Index and \(n\) = parameters used in the analysis, and-

$$\mathrm{CR}=\frac{CI}{RI}$$
(5)

where \(\mathrm{CR}\) = Consistency Ratio and \(\mathrm{RI}\) = random inconsistency value (Table 3) as given by Saaty and Vargas (1991). The CR value should be \(\le 0.1\) as proposed by Saaty (1990), to continue with further analysis. If it is greater than 0.1, then the inconsistency needs to be ascertained and the calculations revised.

Table 3 Random Index values for various N values that were used for computing the consistency ratio (following Saaty and Vargas 1991)
Full size table

Our analyses elicited acceptable CR values of less than 0.1 for all the considered parameters (Table 4), and thus further computations were done using these allotted weights for demarcating the GWPZs within the Tadri Basin.

Table 4 Consistency analysis results, for demarcating GWPZs within the Tadri River Basin
Full size table

Multicollinearity checks

The multiple regression model comes up with a major statistical issue called multicollinearity, where it is seen that at least a single input factor, which is either the dependent or independent variable in the model, has a high correlation with another factor or a combination of other input factors (independent or dependent variables). Thus, the model output can be significantly affected by such linearity among the input factors, e.g., slope and slope-derived aspects (Mukherjee and Singh 2020). This typically occurs once the values of tolerance are  < 0.10 or the Variance Inflation Factor (VIF) is ≥ 10 (Saha 2017; Mukherjee and Singh 2020). Then, these parameters have to be removed from the analysis to avoid multicollinearity induced miscalculations. The tolerance and VIF are computed as:

$$\mathrm{Tolerance\,of \,the }\,i\mathrm{th\,dependent\,variable }\,(\mathrm{T}_i) = 1- {R}_{i}^{2}$$
(6)
$$\mathrm{VIF\,of\,the }\,it\mathrm{h\,dependent\,variable }\,(\mathrm{VIF}_i) =\frac{1}{{T}_{i}}$$
(7)

The respective equations are repeated for each of the input layers to compute these values for each parameter. The assessment of multicollinearity among the chosen parameters for conducting the AHP analysis was done with creation of 56 random points placed across the Tadri Basin. After obtaining all the data for each of the eight layers, the test was performed in SPSS (v.23) software. The derived result for each of the eight input parameters (Table 5) shows that the VIF < 10 and tolerance value > 0.10 for every parameter at ρ < 0.01 and ρ < 0.05. Thus, there was non-existence of collinearity within the eight input layers and thereby no uncertainties in the model results due to multicollinearity issues.

Table 5 Multicollinearity statistics of the parameters used for the groundwater potential zonation
Full size table

Delineation of GWPZs

Demarcation of GWPZs was done following a dimensionless weighted linear combination method that predicts groundwater availability/occurrence in a region, using the formula given by Malczewski (1999), in ArcGIS 10.5 via its raster calculator tool, wherein:

$$\mathrm{GWPZ}=\left({\mathrm{LI}}_{w}*{\mathrm{LI}}_{wi}\right)+\left({\mathrm{LD}}_{w}*{\mathrm{LD}}_{wi}\right)+\left({\mathrm{GM}}_{w}*{\mathrm{GM}}_{wi}\right)+\left({\mathrm{SL}}_{w}*{\mathrm{SL}}_{wi}\right)+\left({\mathrm{SO}}_{w}*{\mathrm{SO}}_{wi}\right)+\left({\mathrm{R}}_{w}*{\mathrm{R}}_{wi}\right)+\left({\mathrm{DD}}_{w}*{\mathrm{DD}}_{wi}\right)+\left({\mathrm{LU}}_{w}*{\mathrm{LU}}_{wi}\right)$$
(8)

where, GWPZ is the potential groundwater zone in an area, LI = Lithology, GM = Geomorphology, LD = Lineament density, SL = Slope, R = rainfall, SO = Soil type, DD = Drainage density, LU = LULC. The layer’s weight \((w)\) and the weight of a specific layer’s subclass \(({w}_{i})\) were integrated on a pixel-wise basis.

ROC validation

The model performance was evaluated via the Receiver Operating Characteristics (ROC) curve, which provides a good visualisation of the classifier in a two-dimensional space (Egan and Egan 1975; Mogaji and San Lim 2017; Saha 2017). The ROC curve shows the relationship among the True Positive Rates or sensitivity with the False Positives rates or specificity, whereby the true positive rates show the magnitude of predicted observations that correctly relate to the actual data, while the false positive rates show the proportion of predicted observations that were incorrectly identified. The Area Under Curve (AUC) value sums up the information provided by the ROC and indicates the ability of the model to distinguish between the classes. The AUC refers to the ‘excerpt of area within the square unit’ and its value ranges between 0 and 1 (Zabihi et al. 2016). Ideally, the value of AUC should be above 0.5 (50%) in a realistic classifier, notwithstanding random guessing. The ROC-AUC has been widely used for its applicability in decision-making and algorithm comparison (Bradley 1997; Fawcett 2006; Ferri et al. 2003). We computed the ROC curve using RStudio software, based on the predicted AHP and average depth to water level from CGWB monitoring wells and observed well density data from SoI topographical maps. Wells were used as a proxy of the ambient groundwater discharge/availability, as theoretically more wells are likely to be found in areas of higher groundwater yield (Das and Pardeshi 2018).

Sensitivity analysis

Analysis of parameter sensitivity provides valuable information about the influence of each parameter in a combined multi-parameter index such as the one used for identifying GWPZs in this study. To understand the impact of each thematic layer in the delineated GWPZ, a map-removal sensitivity examination was performed (cf. Lodwick et al. 1990). Furthermore, the influence of weights and sub-weights of each thematic layer were explored using the single-parameter sensitivity method (Napolitano and Fabbri 1996).

Map-removal sensitivity

In this method, the sensitivity of the parameters is calculated by removing each layer one by one and creating a new GWPZ by overlaying the remaining layers. The new GWPZs obtained by overlaying the reduced combination of the thematic layers is then used to calculate the relative influence or sensitivity of the removed thematic layer as:

$$\mathrm{MRS}= \left|\frac{\left(\frac{\mathrm{GWPZ}}{N}\right)-\left(\frac{{\mathrm{GWPZ}}^{\mathrm{^{\prime}}}}{n}\right)}{\mathrm{GWPZ}}\right| \times 100$$
(9)

where, MRS = the map removal sensitivity index of a layer on its removal, GWPZ = the groundwater potential zone prepared using all layers (i.e. all eight layers in our analysis), GWPZʹ = the same made by omitting one layer in each step, N = the total number of layers and n = the reduced number of layers. Repeating this process for all the layers provides insight about the relative importance and sensitivity of the respective layers/parameters in identifying the GWPZs.

Single-parameter sensitivity

Here, the influence of the weights and sub-weights of the classes of each layer in the delineated GWPZ map was examined. The resulting value reveals the influence of the effective or real weights assigned to the sub-classes of each layer compared to the empirical weights of the same, as follows:

$$W= \frac{{P}_{w} \times {P}_{r}}{\mathrm{GWPZ}} \times 100$$
(10)

where, W = the effective weight of the thematic layer, \({P}_{w}\) and \({P}_{r}\)= the weight and sub-weights of each layer and \(\mathrm{GWPZ}\)= index for identifying groundwater prospective zones as calculated by integration of all the thematic layers.

Results and discussions

Parameters influencing GWPZ

Lithology

Lithological characterization is crucial in identifying prospective zones for groundwater recharge (Muralitharan and Palanivel 2015; Murmu et al. 2018), as it regulates percolation (El-Baz and Himida 1992–1995; Shaban et al. 2006). The type and porosity of the ambient rock formations also determine the permeability to the deeper aquifers (Rahmati et al. 2015) and its water retention capacity (Dar et al. 2020). Thus, lithology was provided the prime weightage while preparing the AHP matrix. The lithological formations dominant within the Tadri Basin were classified into six major units (Fig. 4a), among which the greatest coverage is of doleritic formations (66.23% of the basin area). This basin is part of the Western Ghats that is mainly composed of volcanic basalts and intrusive dolerites or diabase, which are fine to medium grained, dark grey to black and have the second-most permeability among the present lithological units. The joints/fractures, shear zones and faults superimposed on the basalts and dolerites further enhance permeability for groundwater recharge (Basavarajappa et al. 2014). The second most important lithological layer is of vanadiferous titanomagnetite, covering 16.54%, which is genetically related to basic magmatic activity and are differentiates of gabbroanorthosite complexes (Ramiengar et al. 1978). This unit is present in the coastal areas and in the uphill portions in the eastern part of the Tadri Basin and was accorded the highest weightage as it has the highest permeability among all the lithounits present. The other lithounits, along with their respective areal coverage, are the hornblende-actinolite-chlorite schist group (10.38%), laterite (6.53%), banded iron formation (0.3%) and chlorite schist (0.1%).

Lineament density

Joints, fractures and faults are classical lineament features that play a pivotal role in hydrogeology (Odeh et al. 2009, 2016) by facilitating groundwater movement, leading to rising permeability and are hence crucial for identifying prospective groundwater zones (Reddy et al. 2000; Mukherjee et al. 2012; Murmu et al. 2018). Especially in a hard rock terrain, the groundwater occurrence is monitored by secondary porosity, which is related to the presence of fractures/lineaments that enable maximum subsurface percolation (Kumar and Krishna 2018). Their intersection points and neighbouring areas are favourable sites for higher infiltration and thereby groundwater storage (Gupta and Srivastava 2010; Mukherjee et al. 2012). The lineament density obtained was categorised into five classes (Fig. 3b), viz. Very Poor (0–0.10 km/km2), Poor (0.11–0.25 km/km2), Moderate (0.26–0.40 km/km2), Good (0.41–0.58 km/km2), and Very Good (0.59–1.0 km /km2), respectively (using the Jenks natural breaks classification). Lineament density has a positive relationship with groundwater recharge as the highest lineament density values (0.59–1.0 km/km2) are feasibly the most ideal sites for groundwater percolation and these were hence assigned the maximum weightage in the AHP analysis.

Geomorphology

Topographic features/landforms and their alignment/shape play a significant role in helping discern groundwater availability, by influencing surface runoff and percolation and subsurface groundwater movement (Karanth 1987; Kumar et al. 2008; Muralitharan and Palanivel 2015; Kumar and Krishna 2018; Arulbalaji et al. 2019). The Tadri Basin has five major geomorphological units (Fig. 4b), of which the pediment-pediplain complex comprises 39.96% of its area, which can contribute only moderately towards groundwater infiltration (Basavarajappa et al. 2014). The second most dominant geomorphic unit is the moderately dissected hills and valleys that covers 39.52%, which enables minimum infiltration and has thence been assigned a lower weightage. Quarry and mine dumps or the anthropogenic landscape units have negligible areal coverage (< 1%). The active floodplains (3.61% coverage) present along the estuarine and coastal tracts of the Tadri Basin received the highest weightage as they can feasibly contribute towards groundwater recharge through hyporheic exchange. This zone basically comprises of river/marine deposits of silt, sands and clays (that are unconsolidated sediments), which can effectively enable infiltration as well as subsurface storage. (Chowdhury et al. 2009; Biswas et al. 2020). The low dissected hills and valleys (16.91% coverage) adjoining the main Tadri channel and its principal tributaries were presumed to be the next best in respect of their proportionate contribution towards groundwater recharge and accorded the relevant weightage.

Slope

The land surface slope is a decisive criterion that conditions both runoff and infiltration, thereby affecting the groundwater potentiality/recharge in any region (Bera et al. 2020). This was categorized into five classes (Fig. 3d), viz. flat (0°–5°), gentle (6°–10°), medium (11°–20°), steep (21°–30°) and very steep (31°–68°). Steeper slopes are present along the edges of the pediment-pediplain complex as well as on the flanks of the residual hills, while gentler slopes abound in the lower part of the basin and along the coast. Steeper slopes enable more runoff than infiltration, providing less time for percolation (Murmu et al. 2018; Kumar et al. 2020). Hence, the relatively flat coastal areas and floodplains were assigned a higher weightage while the steeper cliffs of the central highlands and the hills in the central part of the basin were given a lower value due to its dual runoff-enhancing and infiltration-limiting influence.

Drainage density

The drainage density (DD) is quintessentially used to detect the erosional evolution of a watershed (Mukherjee and Patel 2022). In groundwater studies, DD is a parameter that is known to negatively affect the permeability of water (Agarwal and Garg 2015). Thus, high DD values signify high runoff and lowered infiltration. The DD was obtained as the ratio of the total stream length in the basin to its area (Horton, 1932) and segmented into five classes (Fig. 3a), viz. Very Poor (0–0.84 km/km2), Poor (0.85–1.5 km/km2), Moderate (1.60–2.30 km/km2), Good (2.40–3.30 km/km2), and Very Good (3.4–5.5 km/km2), respectively (the Jenks natural breaks classification was again used here). Higher drainage density zones (3.4–5.5 km/km2) that facilitate only restrictive groundwater recharge were assigned lower weights and such zones were found in the central portion. On the contrary, the low-lying areas along the coast had low drainage densities and were assigned higher weights because of their ability to allow more infiltration (Halder et al. 2020). In this respect, Bera et al. (2020) and Biswas et al. (2020) have highlighted a direct correlation between the delineated GWPZs and DD values, despite explaining the inverse relationship between DD and permeability, which seems quite contradictory.

Rainfall

Higher rainfall enhances the infiltration possibility and subsequently heightens groundwater recharge. Low intensity and longer duration rainfall events increase infiltration while short-duration high-intensity rainfall is more likely to augment runoff. Based on this premise, the average annual rainfall was classified into five categories (Fig. 3c): Very low (1990–2300 mm), Low (2301–2500 mm), Moderate (2501–2700 mm), High (2701–2900 mm) and Very High (2901–3200 mm), with the highest rainfall zones being accorded a higher weightage and vice-versa.

Soil type

The local soil character affects the infiltration rate and thus influences the GWPZs mapped. Soil permeability depends on several independent factors like soil texture, structure and saturation and many studies have used these factors for GWPZ delineation (see Table 1). The Tadri Basin comprises three principal soil types- coastal alluvial, lateritic and red soils, which cover around 9.75%, 22.98% and 67.26% of its area, respectively. Due to its maximum infiltration capacity and low runoff potency, coastal alluvial soils are ideal for groundwater recharge, while lateritic and red soils have hardpan duricrust formation which inhibits infiltration (CGWB 2014; Joji et al. 2021). Higher weightage was thus accorded to the coastal alluvial soils while the least was assigned to red soils (Fig. 4c).

LULC

The ambient LULC components influence groundwater occurrence and recharge (Muarlitharan and Palanivel 2015), and also dictate its utilization and requirement. The LULC map was prepared from Landsat OLI-8 images via supervised image classification. Six different LULC classes are present (Fig. 4d)—saltmarsh/estuary (covering 2.65% of the basin area), forests (57.79%), agricultural lands (20.80%), fallow landuse (17.11%), barren lands and rocky outcrops (0.21%) and settlements (1.45%). Dense forests, mixed forests and plantations constitute the largest areal share in the basin (57.79%), followed by agricultural lands (20.8%). Given their high potentiality for facilitating ground water recharge, water bodies, forest, plantation and agricultural lands were given higher weightage (Nair et al. 2017). However, this may not be applicable always and location-specific settings may augment or retard groundwater rejuvenation (Li et al. 2018). Barren lands and settlements were assigned a lower weightage because of the higher runoff generation from such surfaces and their impervious nature (for settlements) (Khan et al. 2020). To understand the accuracy of the LULC layer, 65 random points were selected throughout the basin and an error matrix (Table S5) was produced to assess errors in final classification. The overall accuracy for the LULC layer was found to be 73.85% with a kappa coefficient of 0.68.

Distribution of delineated GWPZs

The delineated GWPZ map (Fig. 5) shows significant variations in the ascertained groundwater potential across the basin. About 7.2% of the basin has very good potential for groundwater, especially along its western margins near the coast where the land is relatively flat and is covered by coastal alluvial soils or lateritic soils (CGWB 2014), along with in the easternmost part of the basin, where there is high lineament density in the harder rock areas (Fig. 3b) that also favours better groundwater potentiality. This western coastal tract of the Tadri Basin is not very densely populated and previous studies have shown that the limit of intrusion of saline water landward along the Karnataka coast is quite less (Prusty and Farooq 2020). Thus, the threats that the local GWPZs can face from subsurface saline water inflow (e.g. Hussain et al. 2019; Dhakate et al. 2020) is quite low and this zone can provide a viable source of groundwater. The central and greater portion of the eastern section of the Tadri basin are primarily dominated by a flat-topped plateau surface with shallow soil cover and impervious sub-stratum (CGWB 2012a). Hence, even though this area is ideal for groundwater recharge because of the flat slope of  < 10°, the sub-stratum retards groundwater infiltration and leads to the formation of several springs along the Western Ghats escarpment, which seep through the fissures, cracks or joints, as was observed while delineating the spring/groundwater points from the topographical maps. Around 16.4% of the basin has good groundwater potential, mainly due to the wide distribution of dolerites (Fig. 4a) in the central part and pockets of high lineament densities. Moderate potential is found within the pediment-pediplain complex and in the dissected hills and valleys, where dolerite and hornblende formations are present and the slope varies from  ~ 11° to 20°. A substantial portion lies under the low GWPZ area (36.32%), due to the undulating topography and presence of impervious lithological units therein. However, the remaining 15.57% of the basin was identified as being groundwater scarce due to steeper slopes, impervious surfaces, paucity of joints and fractures and presence of red/lateritic soils. Overall, the derived GWPZ map revealed that higher GWPZs could be detected in the low-lying areas along the western section comprised of coastal plains and around the narrow floodplains of the Tadri River, while the most of the eastern parts of the basin, because of its undulating topography and presence of impervious surfaces, has a low groundwater potential. Furthermore, it is also pertinent to note that a surface waterbody cannot be a potential groundwater extraction zone, even though such a feature may be identified as a GWPZ due to the GIS-based combinations of the other parameters present in that location. For practical purposes these are just surface waterbodies, from which extraction of sub-surface water is not feasible.

Fig. 5
figure 5

GWPZs depicted within the Tadri River Basin

Full size image

Validation of GWPZ delineation

The ROC curves (Fig. 6b and d) were created for validating the GWPZs using two parameters, one by using the average depth to water level database from the CGWB monitoring wells (Fig. 6a) and the other from the SoI topographical map derived well locations (Fig. 6c). The inverse distance weighting (IDW) based average water level depths and well density across the basin shows that the model has been quite successful in correctly predicting a large proportion of the data, as is indicated by the high instance of true positive rates. The computed AUC values were 79% and 82.1%, respectively, which indicated an acceptable accuracy of the AHP method to separate between zones having high and low groundwater potential (Razandi et al. 2015; Rajasekhar et al. 2021). Additionally, the delineated GWPZ were further validated with groundwater depth data obtained from CGWB monitoring wells herein (Fig. 7). The pre- to post-monsoon groundwater fluctuation in these wells (Table 6) and the average depth to groundwater was negatively correlated with the derived GWPZ values at those locations. This implied that areas with higher groundwater potential experienced lower fluctuation of the groundwater table (Gupta and Patel 2021) and moreover the depth to water table was less (i.e., it was closer to the surface). This would enable greater extraction and utilization of the groundwater in these locations, with a higher degree of certainty about its year-round availability. Similar findings were also noted by Patra et al. (2018). Conversely, very low GWPZ values were found in areas where the groundwater level is high, which means that groundwater is found at a considerable depth from the surface. Therefore, the prepared GWPZ map corroborated well with the present status of groundwater depth and fluctuation in the studied basin. The relationship between the individual parameters’ sub-classes and the GWPZ zones (Fig. 8) have also been depicted via several cartograms and scatter plots.

Fig. 6
figure 6

a and c Average Depth to Water Level (CGWB monitoring wells) and Well Density (SoI maps) within the Tadri River Basin, respectively; b and d The ROC curve exhibits the validation of the AHP model with AUC values of 79% and 82.1%, respectively

Full size image
Fig. 7
figure 7

a Average Depth to Water Level extracted from the CGWB monitoring wells from 2009 to 2019. Correlation between Fluctuation in water level, and b Depth to Groundwater Level (DtGWL) with the GWPZ values, where c Annual Average DtGWL, d Pre-Monsoon DtGWL, e Post-Monsoon DtGWL, in mbgl units

Full size image
Table 6 CGWB groundwater monitoring station data within Tadri River Basin, 2009–2019
Full size table
Fig. 8
figure 8

Relationship between GWPZ and the different parameters—a rainfall, b lineament density, c drainage density and d slope, e geomorphology, f soil, g lithology, and h LULC. Note: The sub-categories of the different parameters are as follows—Geomorphology (MDHV Moderately Dissected Hills and Valleys, LDHV Low Dissected Hills and Valleys, FP Flood Plain, PPC Pediment Pediplain Complex, QM Quarry and Mine Dump); Soil (LS Lateritic Soil, RSS Red Sandy Soil, CA Coastal Alluvial); Lithology (CS Chlorite Schist, L Laterite, BIF Banded Iron Formation, VTM Vanadiferous Titano-magnetite, HCS Hornblende-actinolite-chlorite Schist, DL Dolerite); LULC (BL Barren Land, FL Fallow Land, ST Settlement, ALS Agricultural Land/Scrub, WB Water body/Salt Marsh/Estuary, F Dense Forest/Mixed Forest)

Full size image

Sensitivity analysis

The sensitivity analysis for the selected thematic layers used in delineating the GWPZs revealed that the lithology, LULC, DD and lineament density parameters are the most sensitive in the study area. The highest mean variation index of 3.7% was found for lithology (Table 7), which was expected given the higher empirical weight assigned to this layer. The LULC was the second-most important sensitive parameter for the identification of GWPZs, with a mean variation index of 1.46%. As forest cover is the predominant LULC type in this basin, therefore removing this layer had a higher influence in GWPZ identification. It also highlights the role of forested areas and plantations in groundwater recharge, as these tracts are more conducive to infiltration (van Dijk and Keenan 2007). The drainage and lineament density parameters were also highly sensitive to GWPZ identification (mean variation index 1.28% and 1.18%, respectively). However, the influence of the annual rainfall and slope attributes are moderate (mean variation index of 0.99% and 0.97%, respectively). Furthermore, the delineated GWPZs were apparently less sensitive to the local soil (0.79%) and geomorphology (0.75%) attributes. Analysis of the parameter-wise sensitivity indicated a similar influence of the selected parameters (Table 8) towards GWPZ identification, wherein lithology was the most important parameter followed by lineament density and geomorphology. The relative rankings of the parameters, based on the effective weights, closely followed the order of the empirical weights. Lithology was the most important parameter with corresponding empirical and effective weights of 33% and 38%, respectively. The empirical weights assigned to the soil, rainfall, DD and LULC classes were almost similar to their mean effective weights. However, some deviations were observed between the assigned empirical and effective weights for the lineament, geomorphology and slope classes.

Table 7 Map removal sensitivity of GWPZ obtained by removing one thematic layer at a time
Full size table
Table 8 Summary statistics of parameter-wise sensitivity analysis
Full size table

Conclusion

Integrated GIS and MCDM methods (AHP) were used in this study to delineate the GWPZ within the Tadri Basin. This is a cost-effective approach, which can be applied in any region prior to intensive exploratory surveys, which are quite expensive. As per the sensitivity analysis, lithology was the most sensitive parameter in delineating the GWPZ, with a mean variation index of 3.7%. Parameter sensitivity analysis also revealed a similar importance of the lithology, wherein the mean effective weight was found to be 38.39%. However, some deviations between the empirical and effective weights are noted for attributes like basin lineaments, geomorphology and slope. The resultant GWPZ map was validated using the local well density map, employing the ROC method and the obtained result was acceptable as it showed the model accuracy to be 73%. Thus, we have not just restricted the analysis to merely a geospatial examination, as often occurs in such studies, but have also used available field/measured information to ascertain the validity of our results. This lends further credence to the assessments undertaken. The resultant GWPZ map output was categorized into five units, depending on their respective potential groundwater condition. Very high and high GWPZs were observed in the floodplains and coastal plains of the basin while the sudden changes of slope along the cliff edge/escarpments of the spurs are excellent sources of springs. The soils near the river mouth are also seemingly proficient in absorbing the surface water and storing it. The middle portion of the basin is not rich in groundwater resources and only has pockets of high groundwater reserves, a fact mainly determined by the local slope and the sparse fluvial deposits present along the main trunk of the Tadri River. The rugged topography of the Western Ghats does not facilitate much percolation and thus the steeper slopes contain sparse amounts of groundwater. In the eastern part of the basin, on the plateau surface wherein the terrain is relatively flatter, there are some areas of high groundwater potential near Hirekai village. The methods used here, particularly the well-density based validation analysis, can be feasibly used else to conduct similar assessments.

The novelty of this study lies in the validation part as we have used both groundwater depth data (obtained from CGWB) and the number of wells found in the watershed (obtained from the topographical map sheets) to authenticate and confirm the delineated potential zones of groundwater. In support of this evidence, we have also plotted the GWPZ against each raster/vector based thematic layers in order to understand their inter-relationship. We have also tabulated the recent literatures in this domain (Table S1 in the Supplementary Information file) in order to present a clear view on the usage of various parameters used in GWPZ analysis.

Addressing the uncertainty in spatial data integration models, like the one we have used here for estimating groundwater mapping is important, as the thematic layers or the spatial datasets used to produce the potential groundwater zones are not error free. As the principle objective of this paper was to identify GWPZs over a complex geomorphological setting located in the Western Ghats, hence the uncertainty analysis is not within its current scope. Future research can be conducted to examine the uncertainty components in estimating potential groundwater zones arising from the stochastic modelling of groundwater and elaborately inter-compare the different techniques of estimation and sources of uncertainties in estimating groundwater zones. Similarly, the aspect of groundwater quality in the region, particularly in the areas along the coastline that may be affected by saltwater intrusion, can be another focused area of research, along with the history of groundwater use and extraction in the region and how this has affected the above aspect.