Introduction

Groundwater resource security is one of the greatest issues in the twenty-first century (Hameed et al. 2019; Nastar et al. 2018). In Ghana, over 90% of the rural populace and practically about 30% of the metropolitan populace depend on groundwater; hence borehole drilling has intensified over the years (Grönwall and Oduro-Kwarteng 2018). Meanwhile, groundwater is not available everywhere and the quality varies from one geographical point to the other. Zoning areas in rural regions for groundwater extraction are a vital part of water planning and management. Due to the fast-growing population, industrialization and climate change, the issue of groundwater zoning to identify areas of high groundwater potential is considered crucial among groundwater managers (Hameed et al. 2019). Consequently, distinguishing the potential zones of suitable groundwater for drinking and other domestic uses in coastal aquifers is an essential piece for water security and management.

The availability of groundwater is influenced by a wide range of factors. These may include lithology, climatic condition, geomorphology, slope, and drainage patterns (Gumma and Pavelic 2013). However, the availability and quality of groundwater in coastal aquifers may be influenced by additional factors. These may include seawater intrusion and salinization coupled with poor groundwater management practices, more run-off, and evaporation limiting groundwater availability and quality (Ntanganedzeni et al. 2018; Rakib et al. 2020; Werner et al. 2013). The most preferred source of water in the Central Region of Ghana is groundwater, primarily due to the dispersed nature of the rural populace, as well as constant pollution of available surface waters mainly from poor farming practices and illegal mining activities.

Additionally, the coastal aquifer regions have issues of salinization and for reliable groundwater extraction and management, there is the need for the zoning the groundwater potential for drinking using multiple tools that consider the local geology. Studies in previous years have used many different methods such as geological and rainfall (MacDonald et al. 2012), geophysical (Arsène et al. 2018), hydrogeological-based weighted overlay analysis (Nsiah et al. 2018), GIS and remote sensing-based weighted overlay analysis (Mukherjee et al. 2012), and Geographic Information System (GIS)-based Dempster–Shafer model (Mogaji et al. 2015) for mapping groundwater potential zones. Groundwater potential zoning is a typical multi-criteria decision problem and is often addressed using two approaches namely the expert decision method and the machine learning method. The expert decision method which is based on human judgement with decision making support tools has been used in recent studies for example Analytical Hierarchy Process (AHP) (Khashei-Siuki et al. 2020), Fuzzy Analytical Hierarchy Process (FAHP) (Mallick et al. 2019) and multi-impact factor investigation (Abijith et al. 2020).

Recently, studies have also examined groundwater potential zoning using statistical tree algorithms, logistic regression, regression trees, artificial neural networks, and random forest techniques (Arabameri et al. 2021). The advantage of machine learning is lower procedural errors and less perceptual bias, whereas expert methods rely on an expert's eyes in assigning weight to the input variables (Díaz-Alcaide and Martínez-Santos 2019). However, in a data scare region, the expert method in groundwater potential zoning such as FAHP has shown better performance than machine learning (Kumar et al. 2021).

In Ghana, the weighted overlay analysis has been used in groundwater potential zone mapping using hydrogeological data (Nsiah et al. 2018), hydrogeological data with GIS (Appiah-Adjei and Osei-Nuamah 2018) and remote sensing and GIS (Owusu et al. 2017). In order to effectively manage groundwater resources, defining zones for possible siting of drinking water in rural communities in complex coastal aquifers is crucial. In this study, hydrogeological and hydrochemical data, remote sensing and GIS were used with the FAHP method and weighted overlay analysis to define groundwater potential zones in coastal aquifers in the Central Region of Ghana.

Materials and methods

Study area

The Central Region lies within longitudes 2.15888° W and 0.4805  W, and latitudes 5.11455° N and 6.3052° N (Fig. 1). Administratively, it is bordered by Greater Accra and Western Regions to the east and west, respectively, while the northern boarders are shared with the Eastern and Ashanti Regions. The southern border marks the coast line with the Gulf of Guinea. The region is approximately 9826 km2 in size representing 4.1% of Ghana’s total landmass. It is also among Ghana's most important regions in terms of social, academic, economic and agricultural developments.

Fig. 1
figure 1

Location map of the study area

Full size image

The area is drained by five main rivers, namely the Ochi, Kakum, Amisa in its southern portion, and Offin, along with the Pra in its northern portion. All of these rivers flow southward into the Mediterranean Sea.

Geomorphologically, the land is undulating with few inselbergs. Annual precipitation in the region ranges from 850 mm in the southeast to 1500 mm in the northwest. (Ghana Meteorological Agency). April through June is the primary rainy season, followed by September through November, while the dry season lasts for a period of 4 months between December and March. The most elevated monthly mean temperature is commonly recorded in March at about 33 °C, while August records a minimum temperature of about 23 °C. (Ghana Meteorological Agency).

Weighted index overlay model

The Weighted Index Overlay model is a knowledge-driven decision-making tool for multi-criteria conception of complex problems. Under the weighted index overlay model, weighted values are assigned to each layer based on the relative importance and classes of factors, with a full understanding of what affects groundwater potability and availability. The weighted index overlay model has been applied in several studies particularly, inland susceptibility studies (Feizizadeh and Blaschke 2013). It also has extensive use in groundwater mapping or zoning (Melese and Belay 2022; Ndhlovu and Woyessa 2021; Pan et al. 2011; Tolche 2021). The parameter used in the overlay analysis is site-specific. Weighted overlay analysis takes a linear combination model (groundwater potential index) computed using the weighted linear summation algorithm given by the following equation:

$${\text{GWPI}} = \mathop \sum \limits_{w = 1}^{m} \mathop \sum \limits_{i = 1}^{n} \left( {w_{j} \times x_{i} } \right),$$
(1)

where GWPI is the Groundwater potential index, \({x}_{i}\) is the thematic maps, and \({w}_{j}\) is the normalized weight of the jth theme, m is the total number of themes, and n is the total number of classes in a theme.

Data and delineation of groundwater zones

Table 1 summarizes the data used for the study and their sources. Data on a total of 318 test holes [comprising borehole coordinates, borehole yields, water strike, Static Water Levels (SWLs), overburden thickness, logging and hydrochemistry] were collected for this study for wells drilled from 2014 to 2016 under the sustainable rural water and sanitation project (SRWSP) facilitated by the community water and sanitation agency (CWSA). In the delineation of groundwater potential zones in the study area, the following datasets were used: conventional maps (geology and soil), hydrogeological data [overburden thickness (n = 318), and borehole yield, depth to water and transmissivity (n = 195)], hydrochemical data (n = 192) for water quality index, remote sensing data (Landsat-Shuttle Radar Topography Mission Digital Elevation Model (Lansat-SRTM DEM) for land use/land cover and lineament density maps) in addition to available rainfall data (from 1975 to 2012). Finally, 123 borehole yields data out of the total 318 test holes were used to validate the groundwater potential map (GWPM).

Table 1 Data for the groundwater potential mapping
Full size table

The thematic maps of the data were prepared as follows: (i) Landsat 8 satellite image with 30 m resolution acquired in March 2016 was used to derive land use/land cover map and lineament map from which lineament density was generated, (ii) geological map and soil map obtained from the Ghana Geological Survey Department, and (iii) overburden thickness, borehole yield, depth to water, transmissivity, water quality index and rainfall maps were generated using geostatistical techniques. Local hydrogeologists’ expertise was consulted for the relative importance of themes as well as feature classes of the thematic maps; this was used for the Fuzzy-AHP modelling. All these themes were generated using ArcGIS (10.3) software. A flowchart showing the processes used to create the GWPM is shown in Fig. 2.

Fig. 2
figure 2

Flowchart showing processes used to create the GWPM

Full size image

Geology

The aquifer capacity of groundwater is determined by the geology of the region; additionally, the nature of the rock type on the surface contributes significantly to the groundwater recharge (Shaban et al. 2006). Precambrian crystalline igneous and metamorphic rocks underlie almost 96 per cent of the Central Region of Ghana, which includes the Upper Birimian, Lower Birimian, Tarkwaian, Togo, and Sekondian sediments (Fig. 3). Along the coast are Sekondian and Tertiary sediments from the Cenozoic and Palaeozoic eras. The Tertiary sedimentary rocks are found at the middle section between Anomabo and Mankessim and to the west of the region is the Sekondian (GGS 2009). The Upper Birimian formations, which are characterized by metamorphosed tuffs and lava rock types, are associated with the Ashanti and Kibi-Winneba belts in the north, in a north-east south-west trend. The Lower Birimian formation, according to Dapaah-Siakwan and Gyau-Boakye (2000), is made up of isoclinal, folded, and metamorphosed schist, slate and phyllite with interbedded greywacke. Furthermore, the entire Birimian formation is intruded by batholithic masses of biotite granitoids (which generally panel groundwater availability), gneisses, and migmatites. In the Birimian formations which have strong foliation and fractures, water yields are high, showing an average of 12.7 m3/h and ranging between 0.41 m3/h and 29.8 m3/h (Dapaah-Siakwan and Gyau-Boakye 2000).

Fig. 3
figure 3

Geological map of the study area

Full size image

Transmissivity

Transmissivity is an aquifer characteristic that is usually computed from pumping test results. However, for this study, it was derived from the specific capacity (Sc) due to limited pumping test data using the empirical Eq. (1) by Mace (2001) given as follows:

$$T = 15.3 \times \left( {{\text{Sc}}} \right)^{0.67} .$$
(2)

The transmissivity (T) values were then classified into three and thereafter, the T values were interpolated using the inverse distance weighting method at a power of 1 in ArcGIS. This interpolation method was preferred because the root mean square error (RMSE) was lower (37.02) compared to that of the kriging method with a RMSE of 37.82.

Rainfall

One of the fundamental climatic factors in the depiction of groundwater potential zones in this study was precipitation. This factor is known to be the primary source of groundwater. Available rainfall information from 1975 to 2012 from five meteorological stations were obtained from the Meteorological Services Department, Central Region (Ghana Meteorological Agency 2019). The mean yearly rainfall was computed over the period to generate a rainfall pattern of the Region using the kriging method of interpolation in ArcGIS.

Land use/land cover (LULC) and soil

LULC is among the crucial factors for groundwater recharge and quality (Das 2017). For instance, irrigated agricultural ecosystems have been observed to be consistent with moderate to high groundwater recharge rates while natural rangeland ecosystems have no or less groundwater recharge (Pan et al. 2011; Scanlon et al. 2005).

In ENVI 5.3 software, data were classified into different LULC features using the Spectral Angle Mapper (SAM) supervised classification algorithm. To facilitate the implementation of the classification of the LULC from the imagery, a stratified random sample of approximately 650 representative training points from the image subset were selected. From the expert opinion, high weight was assigned to the water bodies, agricultural cropland, dense vegetation, and sparse vegetation, while low weights were assigned to the built-up class and bare soil.

Soil has significant influence on hydrology; it reflects the topography, geology and drift cover (Tetzlaff and Soulsby 2008). The soil map used was based on the World Reference Base (WRB) for classification of soil resources (WRB-IUSS 2015) (Fig. 4). The following WRB classes of soil were identified, namely Dystric Fluvisols, Dystric Leptosols, Eutric Leptosols, Eutric Regosols, Eutric Vertisols, Ferric Acrisols, Ferric Lixisols, Ferric Luvisols, Gleyic Arenosols, Gleyic Cambisols, Haplic Lixisols, Sodic Solonchark, and Umbric Leptosols. The classes of soils were grouped based on their water saturation and porosity properties as high, moderate and low. Gleyic Arenosols, Gleyic Cambisols and Dystric Fluvisols were grouped as high porosity (80–50%) soils, moderate porosity (50–20%) soils, Dystric Leptosols, Eutric Leptosols, Umbric Leptosols, Sodic Solonchark. Eutric Regosols, Eutric Vertisols, Ferric Acrisols, Ferric Lixisols, Ferric Luvisols and Haplic Lixisols were grouped as low porosity soils (WRB-IUSS 2015). Soils with high porosity were scaled with high values, while lower porosity soils were scaled with low values.

Fig. 4
figure 4

Soil map of the study area

Full size image

Lineament density

Lineaments describe the geological structures or features such as fracture, joint, fault, etc. represented as linear to curvilinear area surface weakness. They depict areas of faulting and fracturing within the bedrocks and therefore describe more secondary porosity and permeability; as these enhance well yields and groundwater movement. Lineaments map was generated using Landsat-8 satellite data (band 4, 8 and 1) with the line module of PCI Geomatica software. Before the generation, the atmospheric and geometric correction was employed for image error removal. Lineament density, \({L}_{\mathrm{d}},\) defined as the total lineaments lengths per unit area is given as follows:

$$L_{{\text{d}}} = \frac{{\mathop \sum \nolimits_{i = 1}^{n} L_{i} }}{A},$$
(3)

where \(\sum_{i=1}^{n}{L}_{i}\) is the total length of lineaments, and A refers to a unit area of the region.

Higher lineament density is linked with a higher permeable zone; this favours groundwater movement and occurrence and so reveals good groundwater potential zones (Mukherjee et al. 2012). Low weight was assigned to low lineament density and high weight to high lineament density. The study area lineament density mapping was classified into five.

Water-quality index

The water-quality index provides a single value to the quality of water of a source based on the hydrochemistry compared with recommended levels (Gorai et al. 2016). Hence, different samples can be compared for quality based on the index value of respective samples. The water-quality index proposed by Pei-Yue et al. (2010) which uses the entropy weighting model was employed to provide a single value to rate the quality of groundwater. This follows the calculation of eigenvalue matrix X of m water samples of n measured hydrochemical parameters:

$$X = \left[ {\begin{array}{*{20}c} {x_{11} } & {x_{12} } & {\begin{array}{*{20}c} \cdots & {x_{1n} } \\ \end{array} } \\ {x_{21} } & {x_{22} } & {\begin{array}{*{20}c} \cdots & {x_{2n} } \\ \end{array} } \\ {\begin{array}{*{20}c} \vdots \\ {x_{m1} } \\ \end{array} } & {\begin{array}{*{20}c} \vdots \\ {x_{m2} } \\ \end{array} } & {\begin{array}{*{20}c} {\begin{array}{*{20}c} \ddots \\ \cdots \\ \end{array} } & {\begin{array}{*{20}c} \vdots \\ {x_{mn} } \\ \end{array} } \\ \end{array} } \\ \end{array} } \right].$$
(4)

This was followed by data normalization using the efficiency function:

$$y_{ij} = \frac{{x_{ij} - \left( {x_{ij} } \right)_{{{\text{min}}}} }}{{\left( {x_{ij} } \right)_{{{\text{max}}}} - \left( {x_{ij} } \right)_{{{\text{min}}}} }}.$$
(5)

To obtain the matrix Y;

$$Y=\left[\begin{array}{ccc}{y}_{11}& {y}_{12}& \begin{array}{cc}\cdots & {y}_{1n}\end{array}\\ {y}_{21}& {y}_{22}& \begin{array}{cc}\cdots & {y}_{2n}\end{array}\\ \begin{array}{c}\vdots \\ {y}_{m1}\end{array}& \begin{array}{c}\vdots \\ {y}_{m2}\end{array}& \begin{array}{cc}\begin{array}{c}\ddots \\ \cdots \end{array}& \begin{array}{c}\vdots \\ {y}_{mn}\end{array}\end{array}\end{array}\right].$$
(6)

Next, the ratio of the index value, Pij, of the jth index in the ith sample is given as follows:

$$P_{ij} = y_{ij} /\mathop \sum \limits_{i = 1}^{m} y_{ij} .$$
(7)

Values of the information entropy ej, of the jth parameter, are estimated using the Eq. 8

$$e_{j} = - \frac{1}{\ln m}\mathop \sum \limits_{i = 1}^{m} P_{ij} \ln P_{ij} .$$
(8)

After calculating the information entropy ej, the entropy weight wj of the jth parameter can be calculated as

$$w_{j} = \frac{{1 - e_{j} }}{{\mathop \sum \nolimits_{j = 1}^{n} \left( {1 - e_{j} } \right)}}.$$
(9)

For IEBGQI calculations, the quality-rating scale was assigned for each parameter. The quality rating is given by

$$q_{j} = \frac{{C_{j} }}{{S_{j} }} \times 100,$$
(10)

where Cj are the measured chemical parameter in each water sample, and Sj the drinking water Ghana Standard Authority. The IEBGQI was calculated using Eq. 11 and was classified into five ranks.

$${\text{IEBGQI}} = \mathop \sum \limits_{j = 1}^{n} w_{j} q_{j} .$$
(11)

Weight assignments and normalization

In the assigning of weight, the analytic hierarchy process (AHP) model, a multiple criteria decision-making methodology (MCDMM) developed by Saaty (1980) was employed to obtain the pairwise comparison matrix from experts. The AHP integrates qualitative and quantitative information to evaluate and integrate different factors. In AHP, comparison between any two factors is based on a scale from 1 to 9, where 1 implies equal importance between two factors while 9 indicates a factor of extreme importance. However, AHP cannot accurately account for the uncertainty in human choice in the quantitative delivery of preferences (Krejčí et al. 2017). Hence, the Fuzzy-AHP method was used to fuzzify the crisp weights from 1 to 9, used for pairwise comparisons. In fuzzy AHP, pairwise comparison matrix, elements are usually represented by a triangular fuzzy number \(\tilde{b }\) whose membership function is determined by an array of triples of real numbers of \({b}_{1}\le {b}_{2}\le {b}_{3}\) in the following manner:

$${\mu }_{\tilde{b }}\left(x\right)=\left\{\begin{array}{c}\frac{x-{b}_{1}}{{b}_{2}-{b}_{1}},{b}_{1}<x<{b}_{2},\\ 1, x={b}_{2},\\ \begin{array}{l}\frac{{b}_{3}-x}{{b}_{3}-{b}_{2}}, {b}_{2}<x<{b}_{3},\\ 0, \mathrm{otherwise}.\end{array}\end{array}\right.$$
(12)

The following steps were considered for the fuzzy extension of the analytical hierarchy process:

Step 1 Using the 1 to 9 integer scale (Saaty 1980), ten experts were given the AHP excel template with multiple inputs developed by Goepel (2013). A threshold, α, of 0.1 was set for the acceptance of inconsistency, and experts with lower consistency ratios had their expertise weighted more, such that expert input with a lower Consistency Ratio (CR) is given more weight than the inputs from other experts. Expert's judgements were modelled using the following parameters or indicators, namely the consistency index (CI):

$$\mathrm{CI}=\frac{\left({\lambda }_{\mathrm{max}}-N\right)}{N-1},$$
(13)

where \({\lambda }_{\mathrm{max}}\) is the maximal eigenvalue of matrix B.

The geometric consistency index (GCI) is calculated using

$${\text{GCI}} = \frac{2}{{\left( {N - 1} \right)\left( {N - 2} \right)}}\mathop \sum \limits_{i < j} \ln b_{ij} - \ln \frac{{p_{i} }}{{p_{j} }} ,$$
(14)

where \(p_{i} /p_{j}\) is the ratio of priorities calculated using the row geometric mean method (RGMM) of comparison matrix B = \(b_{ij}\).

And the CR calculated using

$${\text{CR}} = \frac{{{\text{CI}}}}{{{\text{RI}}}} ,$$
(15)

where RI is a random consistency index.

Step 2 Fuzzifying pairwise comparison matrices developed using the linguistic principles (Table 2).

$$\tilde{B }=\left[\begin{array}{ccc}\tilde{1 }& {\tilde{b }}_{12}\dots & {\tilde{b }}_{1n}\\ {\tilde{b }}_{21}& \tilde{1 }& {\tilde{b }}_{2n}\\ \begin{array}{c}\vdots \\ {\tilde{b }}_{n1}\end{array}& \begin{array}{c}\vdots \ddots \\ {\tilde{b }}_{n2}\cdots \end{array}& \begin{array}{c}\vdots \\ \tilde{1 }\end{array}\end{array}\right]=\left[\begin{array}{ccc}\tilde{1 }& {b}_{12}\dots & {\tilde{b }}_{1n}\\ 1/{\tilde{b }}_{21}& \tilde{1 }& {\tilde{b }}_{2n}\\ \begin{array}{c}\vdots \\ 1/{\tilde{b }}_{n1}\end{array}& \begin{array}{c}\vdots \ddots \\ 1/{\tilde{b }}_{n2}\cdots \end{array}& \begin{array}{c}\vdots \\ \tilde{1 }\end{array}\end{array}\right] ,$$
(16)

where \({\tilde{b }}_{ij}\) signifies a pair of criteria \(i\) and \(j\), \(\tilde{1 }=\left(\mathrm{1,1},1\right)\) when \(i=j\); the measure that criterion \(i\) is relatively important to criterion \(j\) is signified as \(\tilde{1 }, \tilde{2 }, \stackrel{\sim }{ 3}, \tilde{4 }, \tilde{5 }, \tilde{6 }, \tilde{7 }, \tilde{8 }, \tilde{9 }\) while the measure that criterion \(j\) is relatively important to criterion \(i\) is represented as \({\tilde{1 }}^{-1},{\tilde{2 }}^{-1},{\tilde{3 }}^{-1},{\tilde{4 }}^{-1},{\tilde{5 }}^{-1},{\tilde{6 }}^{-1},{\tilde{7 }}^{-1},{\tilde{8 }}^{-1},{\tilde{9 }}^{-1}\) (Mallick et al. 2019).

Table 2 Fuzzy pair-wise comparison matrix of the ten thematic layers
Full size table

Step 3 Fuzzy weight calculation from the pairwise matrix using the Krejčí et al. (2017) approach was founded on the constrained fuzzy arithmetic concept. With this approach, the overall fuzzy weight is not falsely increased compared to the typical fuzzy arithmetic method in a fuzzy extension of AHP.

Step 4 For fuzzy weight defuzzification and normalisation, the centre of the area was used. Thus, for a triangular fuzzy weight \(\tilde{w }=\left({w}_{1}, {w}_{2}, {w}_{3}\right)\), de-fuzzified weight was computed as follows;

$$c\left(\tilde{w }\right)=\frac{{w}_{1}+{w}_{2}+{w}_{3}}{3}.$$
(17)

The weights were then normalised which is \(\sum_{i=1}^{n}{w}_{i}=1,\) where,\({w}_{i}\), i = 1, 2, …. n, are the de-fuzzified weights of the criteria. The Fuzzy AHP was performed using the Fuzzy AHP package application version 0.9.0 (Caha 2017) in R statistical software.

Results and discussion

Hydrogeological parameters

Summary statistics of the measured parameters and some regional aquifer characteristics are presented in Table 3. The average borehole depth measured in the study area was 58.11 m and ranges between 31 and 106 m with a standard deviation and coefficient of variation of 11.41 m and 19.64%, respectively. This suggests that there is less variability of borehole depths in the region. Borehole depths in the region are similar to those recorded by Nsiah et al. (2018) (depth range 27.6–99.0 m, average = 48.5 m) and Water Resources Commission (2011) (depth range 24.0–99.0 m, average = 44.0 m). In terms of access to groundwater, the spatial distribution of borehole depth to water (water strike) in the region (Fig. 5) suggests that deeper aquifers are found in the north-western and south-eastern parts of the region in the Birimian and Tarkwaian formations compared to aquifers in the granitoids.

Table 3 Statistical summary of the hydrogeological parameters
Full size table
Fig. 5
figure 5

Spatial distribution of depth to water level

Full size image

The region's overburden thickness ranges from 3 to 44 m, with an average of 20.57 m, a standard deviation of 10.25 m and a coefficient of variation of 49.82%. Nsiah et al. (2018) obtained similar values ranging from 2.7 to 40 m with a mean of 16.5 m in the Naboga basin. The spatial distribution of overburden thickness (Fig. 6) suggests that high overburden thicknesses are located at the northwestern part of the region around Diaso and Dunkwa-on-Offin and cover 5.64% of the total area. Medium overburden thicknesses (15–30 m) cover the largest portion of the area with a percentage coverage of 69.71% of the total area, while low thicknesses (< 15 m) are located in the southwestern and eastern portions, covering 24.65% of the area. The respective low overburden thickness may have resulted in stable clay or the presence of quartz, which are of slow weathering rates (MacDonald et al. 2005) in the study area.

Fig. 6
figure 6

Overburden thickness spatial distribution map

Full size image

The borehole yields in the Central Region range between 0 and 700 l/min (0 and 42 m3/h) with a mean of 42.87 l/min, standard deviation of 79.95 l/min and coefficient of variation of 186.50% indicating great variability in borehole yields across the region. Values reported in this work compared well with values reported by the Water Resources Commission (2011) in the Naboo basin with values within the range of yields but less than the mean yield (102.6 l/min) of the basin. This may reflect the significant number of unsuccessful boreholes recorded in this study. Additionally, the yields were higher compared with borehole yields reported by Ganyaglo et al. (2017) (7.5 to 179.7 l/min), but in agreement with the mean yield (32.1 l/min) in the Ochi-Narkwa Basin. Also, from the spatial distribution of the borehole yield map (Fig. 7), dry test holes (yield between 0 and 5 l/min) were mostly located at the eastern and southeastern parts of the region covering about 2.34% of the area. Medium to very high borehole yields (> 60 l/min) can be found to the northwest and as speckles at the middle portion of the region, while marginal to low yields cover predominantly the middle and south eastern portions of the study area.

Fig. 7
figure 7

Borehole yields spatial distribution map

Full size image

The static water levels (SWL) within the region vary from 0.01 to 33 0.21 m with a mean of 9.86 m, standard deviation of 6.32 m and CV of 65.48%, indicating great variability of SWLs across the region. Groundwater flow directions in the region inferred from a generated piezometric head map (Fig. 8) suggests a predominant groundwater flow direction from the north to the south towards the coast. Mineralisation of groundwater in the Central Region could follow the same trend.

Fig. 8
figure 8

Groundwater flow directions (→) map inferred from piezometric head

Full size image

The specific capacity of wells in the region ranges from 0.28 to 57.24 m3/day/m, with a mean of 5.48 m3/day/m, a standard deviation of 8.45 m3/day/m, and a coefficient of variation of 154.24 percent, suggesting that this well parameter varies significantly throughout the research area. The heterogeneities in the rock formations may be the cause of such large fluctuations.

The transmissivity values in the study area range from 6.59 to 230.33 m2/day with an average of 40.24 m2/day, a standard deviation of 39.08 m2/day and a coefficient of variation of 97.12% suggesting moderate variations of transmissivity in the region. This range of transmissivity values was higher than values (0.3 to 35.7 m2/day) recorded by Ganyaglo et al. (2017). However, they were consistent with values (0.3 to 270 m2/day) reported in other studies (Darko 2002). These values may suggest lithological and structural inhomogeneities in the rocks (Yidana and Koffie 2014). Transmissivity values were classified using Krasny's classification scheme (Fig. 9), which indicated that aquifers of high transmissivities are located at the north western part of the region covering 2.08% of the area. Aquifers of low ranges of transmissivity are situated as speckles at the southeastern and southwestern portions of the region covering 0.5% of the study area. About 97.41% of the area falls within the intermediate (10 to 100 m2/day) transmissivity range. This shows that the Central Region of Ghana is dominated by aquifers of intermediate transmissivities.

Fig. 9
figure 9

Transmissivity spatial distribution map

Full size image

Rainfall

The outcomes of the analysis of the rainfall data revealed that the mean yearly rainfall ranges between 867 mm to 1500 mm, (Fig. 10) with higher rainfall patterns (> 1300 mm) covering a larger area (36.58%) observed at the northwestern part of the region. Lower rainfall patterns were observed at a much small area (1.29%) located in the southeastern zone of the region towards the coast. Generally, there is a decline in rainfall patterns from the northwestern end of the region around Diaso to the south eastern end around Winneba.

Fig. 10
figure 10

Mean annual rainfall (1975 to 2012) spatial distribution map

Full size image

Land use and land cover

Among the numerous elements that influence groundwater recharge and potential, land use/land cover plays a key effect (Lerner and Harris 2009; Martin et al. 2017). The major types of land use/land cover of the central region of Ghana are bare land, built-up, close forest, cropland, herbaceous vegetation, open forest and water. The open forest is the dominating land use/land cover type in the region (51.91%). This is followed by herbaceous vegetation (25.33%) and water (12.81%). The least dominating land use/land cover type in the region was cropland (0.05%), mostly located in the eastern part of the region (Fig. 11).

Fig. 11
figure 11

Land use/land cover map

Full size image

Lineament density (LD)

Lineaments can provide reliable information on groundwater storage and movement (Henriksen and Braathen 2006). Lineament density in the study area ranges from 0 to 48 km/km2. This was reclassified into five classes (Fig. 12), and the results show that higher lineament densities are located at the north western part with some specks in the south eastern section of the region. These areas could epitomise highly promising precincts for groundwater occurrence. These zones together cover about 5.61% of the study area and are found in the Birimian and the Tarkwaian formations (Supplementary Table 1). According to Achu et al. (2020), areas of fairly higher lineaments density signify higher groundwater occurrence and this is in conformity with the groundwater situation in the Central Region. However, much of the study area (48.09%) is characterised by low lineaments density (0 to 10 km/km2). Generally, fractured zones (high lineament density) overlain by thick overburden are linked with good borehole yields.

Fig. 12
figure 12

Lineament density spatial distribution map

Full size image

Water quality index

This theme is significant to this study as the research aims at targeting potential groundwater zones for future development of the resource for drinking and other domestic purposes. Based on the IEBGQI values calculated (Table 4), groundwater quality in the study area was ranked as 59.4% excellent, 20.3% good, 7.8% average/medium, 2.6% poor and 9.9% extremely poor. Spatially the IEBGQI values towards the south with proximity to the Gulf of Guinea showed extremely poor groundwater quality covering 16.23% of the study area, while the northern portion showed excellent groundwater quality covering a percentage area of 46.10% (Fig. 13).

Table 4 Statistical summary of the hydrochemical parameters
Full size table
Fig. 13
figure 13

Information entropy-based groundwater quality index spatial distribution map

Full size image

Weight normalisation for thematic maps Fuzzy-AHP

Field knowledge, literature survey and expert judgement were used to assign relative importance of the respective themes in a crisp comparison matrix and then modelled using Fuzzy-AHP to assign weight to each theme and their classes (Table 5). A key feature of the FAHP is its ability to handle vagueness and uncertainties when assigning weights for the overlay analysis. Borehole yield and transmissivity were assigned the highest weight. Both were assigned the same weight (19%). These were followed by rainfall (18%), geology (12%), WQI (8%), soil (7%), lineament density (7%), depth to water (4%), overburden thickness (4%) and land use/land cover (2%). The Consistency Ratio CR computed from the crisp weighted matrix was 0.0597, which implies that the assigned weights were reliable with the anticipated results.

Table 5 Thematic layers normalized weights
Full size table

Analysis of GWPZ classification map and implications for policy

The groundwater potential zones (GWPZ) were classified into the following five classes: very poor, poor, moderate, high and very high based on the distribution frequency of the GWPZ data. In general, the region is characterised by moderate GWPZ class covering about 77.48% (7613.5 km2) of the region. This is followed by high GWPZ class covering about 11.78% (1157.4 km2), poor GWPZ class covering 10.58% (km2), very high GWPZ class covering 0.13% (13.2 km2) and very poor GWPZ class covering 0.03% (2.6 km2) of the region (Table 6). The groundwater potential zones in the region tend to decrease from the north towards the coast, with the least potential zones situated predominantly in the south eastern and southern part of the region (Fig. 14). High lineament density and rainfall were profound at the north western part of the region, and display high groundwater potential, while areas of low lineament density and rainfall describe poor groundwater potential zones. Additionally, the zones depicting very poor and poor zones are underlain by geological formations that exhibit fewer fractures and shallower weathering profiles.

Table 6 Summary of groundwater potential zone mapping
Full size table
Fig. 14
figure 14

Groundwater potential zone map

Full size image

The groundwater potential zone map was validated using 123 borehole yield data (Fig. 15), by analysing the Receiver Operating Characteristic curve (ROC) and area under the curve. The plot analysis estimates the area under the curve (AUC) to be 0.869 (Fig. 16), suggesting that the produced map efficiency is good. Rajasekhar et al. (2019) have shown that FAHP performed better than fuzzy logic model and with AHP. Similarly, Kumar et al. (2021) found that FAHP is more effective than the machine learning methods. Hence, the use of hydrochemical, hydrogeological, RS, GIS-based Fuzzy-AHP models for groundwater potential delineation in coastal aquifers seems reliable. This study is an essential piece for improving water security and sustainable development and management of groundwater resource for drinking and other domestic purposes in the Central Region, Ghana which is in line with the Ghana Government’s agenda and the United Nation’s Sustainable Development Goal Six (SDG 6) of providing save water for all by the year 2030. It will also contribute to the sustainable water and land management element of Agenda 2063 of the African Union (AU).

Fig. 15
figure 15

Groundwater potential zones with validation boreholes

Full size image
Fig. 16
figure 16

Receiver operating characteristics curve

Full size image

Besides, the outcome of this research will help address the high cost associated with groundwater interventions as experienced by CWSA and other stakeholders due to the current low success rate of drilling and poor water quality especially along the coast.

Conclusions

Groundwater potential zones for drinking and other domestic purposes has been identified in the Central Region using RS, GIS, hydrochemistry, hydrogeology and FAHP approach. Ten factors in all were considered for this study and these include rainfall, geology, soils, land use/land cover, lineaments density and hydrochemistry. The others are overburden thickness, depth to water (water strike), yield and transmissivity. Borehole depths are in the range of 31 to 106 m with an average of 58.11 m. Overburden thickness also varies from 3 m in the south western and eastern parts to 44 m in the north western part in the Birimian and the Tarkwaian formations with the largest portion of the region underlain by medium (15–30 m) overburden thicknesses. A spatial distribution of water strikes indicates that deeper aquifers are found in the north western and south eastern parts of the region in the Birimian and Tarkwaian formations compared to aquifers in the granitoids or Dahomeyan.

There is a great variability in borehole yields ranging from 0 to 700 l/min with a mean of 42.87 l/min, standard deviation of 79.95 l/min and coefficient of variation of 186.50%. Similarly, transmissivity values in the study area range from 6.59 to 230.33 m2/day with an average of 40.24 m2/day, a standard deviation of 39.08 m2/day and a coefficient of variation of 97.12% suggesting moderate variations of transmissivity in the region. About 97.41% of the area falls within the intermediate (10 to 100 m2/day) transmissivity range. This shows that the Central Region of Ghana is dominated by aquifers of intermediate transmissivities.

In general, the region is characterised by moderate GWPZ covering about 77.48% (7613.5 km2). This is followed by high GWPZ covering about 11.78% (1157.4 km2), poor GWPZ representing 10.58% (1039.3 km2), very high GWPZ class covering 0.13% (13.2 km2) and very poor GWPZ class covering 0.03% (2.6 km2) of the region. The groundwater potential zone map was validated using 123 borehole yield data by the area under the curve (ROC), which was calculated to be 0.869, showing that the groundwater potential zone map is effective. As a result, the map produced can be used by government agencies such as the water resource commission (WRC), community water and sanitation agency (CWSA), and Ghana Water Company Limited (GWCL), as well as development partners such as UNICEF, USAID, Plan Ghana, and other stakeholders involved in groundwater resource development and management for potable water supply in Ghana.