Enhancing groundwater potential mapping in arid regions through integrated microwave remote sensing, statistical analysis, and geophysical-based modeling

Figure 4 illustrates the structured workflow of the integrated methodology applied in this study. The groundwater-potential (GwP) assessment combines topographic, geological, structural, hydrogeological, and climatic variables within a hybrid statistical framework. To further refine the analysis, the approach incorporates data from DC-resistivity soundings and aeromagnetic surveys. The subsequent sections describe each methodological step in detail.

Datasets collection and preparation

The primary data sources include S1A VV and VH polarization imagery obtained from the Alaska Satellite Facility for lineament extraction and Rb extraction, the ALOS-DEM for topographic and hydrogeological parameter extraction, and IMERG data for rainfall distribution analysis. Additional datasets comprise groundwater well locations and a lithological map derived from the CONOCO⁵⁹ database provided by the European Soil Data Center (https://esdac.jrc.ec.europa.eu/content/geologic-map-egypt). All digital datasets were geometrically corrected and co-registered to the Universal Transverse Mercator (UTM) coordinate system, WGS 1984 Zone 36N, using ArcGIS Pro 3.6⁶². The subsequent sections provide an in-depth analysis of the dependent inventory and independent factor datasets, emphasizing their significance within the modeling framework. Table 1 summarizes all datasets used in this study, including the source, acquisition date, native resolution, preprocessing steps, and the final coordinate reference system to which all layers were standardized.

Potential controlling factors

Aly et al.⁶⁴ emphasized that preparing thematic data layers and selecting relevant factors are critical components of any modeling approach. In this study, ten thematic parameters were identified to assess groundwater potential within the SCAq region, as they either influence or exhibit correlations with groundwater occurrence^{8,24,48,50,60}. These parameters include slope, curvature, Topographic Wetness Index (TWI), Stream Density (SD), Distance to Stream (DTS), lithology, Radar backscattering (Rb), Lineament Density (LD), Lineament-Stream Intersection Density (LSID), and Average Annual Rainfall (AAR).

The selection criteria for these parameters are grounded in extensive literature that highlights their relevance to groundwater potential in the SCAq region, particularly in the context of the Eastern Desert’s complex terrain^50,60. The regional geology features diverse lithological formations shaped by prevalent faults and fractures, as evidenced by the linear boundaries of rock units (Fig. 3). These geological characteristics play a crucial role in influencing groundwater storage and movement, affecting aquifer recharge dynamics.

Furthermore, hydrological factors such as Distance to Stream (DTS), Stream Density (SD), and the Topographic Wetness Index (TWI) are essential for identifying potential groundwater accumulation zones^65,66. According to Sahour et al.⁶⁵, areas in proximity to natural drainage systems within watersheds tend to exhibit shallower groundwater tables, which are critical for sustainable water resources. In this way, these parameters collectively inform our understanding of groundwater behavior and enhance the reliability of our modeling approach.

Topographic factors

Key topographic parameters, particularly slope and curvature, play a critical role in aquifer recharge, especially in the highly trained terrains of the Eastern Desert. These parameters were derived from an ALOS/PALSAR-1 Radiometrically Terrain-Corrected (RTC) digital elevation model (DEM) product, acquired on June 11, 2009, by the Alaska Satellite Facility (ASF)⁶⁷. The dataset comprises twelve scenes in Fine Beam Single (FBS) polarization mode (HH), each with a native spatial resolution of 12.5 m. All scenes were downloaded in GeoTIFF format, mosaicked using the “Mosaic to New Raster” tool in ArcGIS Pro 3.6, and projected to the Universal Transverse Mercator (UTM) Zone 36N coordinate system with the World Geodetic System 1984 (WGS84) datum.

The spatial distribution of surface slope was calculated using the Slope tool (ArcGIS Pro V3.6), which calculates the maximum rate of change in elevation between each cell and its eight neighbors. The output slope grid was computed in degrees. This parameter is essential for understanding how water moves across the landscape. Additionally, the Curvature tool was utilized to assess deviations from a planar surface, generating an output grid where negative values represent concave, positive values represent convex, and zero values represent flat surfaces. These topographical features significantly influence hydrological processes, determining areas of water accumulation and influencing the efficiency of aquifer recharge.

Hydrological factors

Hydrological factors, including DTS, SD, and TWI, offer critical insights into potential groundwater accumulation zones⁶⁶. Consequently, the mosaicked 12.5 m ALOS-DEM was used as input for hydrological analysis. Areas in proximity to natural drainage systems within watersheds are more likely to exhibit a shallow groundwater table⁶⁵. Consequently, ALOS-DEM data were used to extract the natural drainage network maps. The spatial distribution of stream networks was obtained by applying the eight-direction (D-8) flow algorithm⁶⁸ using the Arc Hydro toolset (version 2.0) in ArcMap. The stream network was delineated using a flow-accumulation threshold of 2000 pixels, such that only cells draining at least 2000 upstream cells were included in the network. The SD, defined as the ratio of total stream length to drainage basin area, was derived using the Line Density tool, with a search radius of 1000 m and an output cell size of 10 m. The TWI is widely used to assess soil moisture distribution and saturation zones within a given terrain⁶⁹. The TWI was computed using the following Eq. (1), implemented in the ArcGIS Raster Calculator.

where (a) is the local upslope contributing area (m²/m), (b) is the slope gradient (measured in degrees) at that specific position, and “ln” is a constant that refers to the Napierian logarithm.

Geological factors

The variations in surface geology and soil distribution directly affect hydrological processes, determining how effectively water infiltrates into the ground and replenishes aquifers. Each rock type possesses distinct properties that significantly influence surface water movement and soil porosity within the basins. The lithological map used in this analysis was derived from the published geological map of CONOCO⁵⁹, scaled at 1:500,000 (Fig. 3). The geological map was scanned, georeferenced in ArcGIS using ground control points, and the geological polygons were digitized. This thematic vector layer was then converted to raster format at a 10 m cell size using the Polygon to Raster tool to standardize the layer, facilitating the assignment of weights and ranks.

Lineaments, observed as linear or curvilinear features in satellite imagery, encompass geological structures such as faults, fractures, and joints, playing a crucial role in groundwater flow and accumulation²⁷. Two high-resolution Ground Range Detected (GRD) scenes from S1A VV and VH polarizations, acquired in June 2021 using the Interferometric Wide (IW) mode, were used due to the availability of the images over the whole basin at no cost (Source of data). These images were captured in a descending orbit with a spatial resolution of 10 m × 10 m. The preprocessing included applying the orbit file, speckle filtering, radiometric calibration, thermal noise removal, and Range-Doppler Terrain Correction. The images were subsequently converted to a decibel scale to mitigate disturbances in the VV and VH bands. A 3 × 3 Lee filter was applied to minimize speckle noise while preserving structural integrity, particularly in homogeneous regions and along feature boundaries^70,71.

Lineaments, observed as linear or curvilinear features in satellite imagery, encompass geological structures such as faults, fractures, and joints²⁷. For their extraction, two high-resolution Ground Range Detected (GRD) scenes from Sentinel-1A (S1A) with VV and VH polarizations were acquired. The scenes, captured on June 21, 2021, in Interferometric Wide (IW) mode with a descending orbit, were downloaded from the Alaska Satellite Facility (ASF) vertex portal. Each scene had a native spatial resolution of 10 m × 10 m. Preprocessing was performed using the Sentinel-1 Toolbox in SNAP (ESA SNAP 9.0) and included: (1) Apply Orbit File, (2) Thermal Noise Removal, (3) Radiometric Calibration to sigma0, (4) Range-Doppler Terrain Correction using the SRTM 30 m DEM, and (5) Conversion to decibel (dB) scale. To mitigate speckle noise while preserving structural integrity, a 3 × 3 Lee filter was applied to the calibrated dB images^70,71.

Lineament extraction was performed using the LINE module in CATALYST PROFESSIONAL PCI Geomatica 2022 (version 23.0). The filtered VV polarization dB image was used as input. The following parameter values were set for the extraction algorithm, with GTHR and ATHR adjusted from default values to enhance geological feature detection, following Embaby et al.⁸

The extracted vector lineaments were manually reviewed and edited to remove non-geological features (e.g., roads, field boundaries). The Intersect tool in ArcGIS was subsequently applied to determine intersection points between the extracted lineaments and stream networks. Structural indices, including LD and LSID, were computed using the Kernel Density tool in ArcGIS Pro V3.6 with a search radius of 1000 m and an output cell size of 10 m.

Radar backscatter intensity has been widely recognized as an effective parameter for monitoring hydrological components^72,73. A static regional Rb image was generated from the preprocessed S1A VH polarization dB image. The image was resampled to a 10 m grid to analyze spatial variations in soil properties.

Climatic factors (average annual rainfall)

Rainfall serves as the primary water source in arid and semi-arid basins⁷⁴. Daily precipitation data were acquired from the NASA Global Precipitation Measurement (GPM) mission’s Integrated Multi-satellitE Retrievals for GPM (IMERG) Final Run product (Version 06). The data were accessed via the NASA Giovanni platform (https://giovanni.gsfc.nasa.gov/giovanni/) for the period from January 1, 1998, to January 31, 2019⁷⁵. This dataset provides half-hourly precipitation estimates at a native spatial resolution of 0.1° × 0.1° (approximately 10 km × 10 km). The daily data were aggregated to monthly, then to annual totals. A long-term Average Annual Rainfall (AAR) grid was computed by averaging the 21 annual grids. This coarse AAR raster was then projected to UTM Zone 36N, WGS84, and interpolated to a 10 m grid cell size to match other thematic layers using the Inverse Distance Weighting (IDW) geostatistical method in ArcGIS, with a power parameter of 2 and a variable search radius. AAR was identified as a relevant conditioning factor, given its applicability over extensive spatial scales⁷⁵.

Groundwater inventory data

The residents utilize springs, where natural groundwater is discharged from the NSAS⁴⁹, along with drilled groundwater wells for daily use and agricultural activities⁴⁵. A total of 40 productive groundwater (PGw) points, including productive well sites and a few spring locations, were identified using Google Earth Pro imagery and field observations within the study area. To satisfy the presence/absence requirements of the bivariate models and improve data distribution, absence points were generated from two complementary sources: (1) documented dry wells (shallow, non-productive boreholes) encountered in the field, and (2) bedrock exposures and high-terrain regions identified from satellite imagery and geological maps, which are characterized by high runoff and minimal overburden, making them inherently unfavorable for groundwater storage. To account for the influence of local faults and aquifer discontinuities, a 100-m buffer zone was established around all presence points. No absence points were sampled within these buffers to prevent conflict with productive structural areas.

For modeling, 28 PGw points (70%) and an equivalent number of absence points were randomly selected for training the WOE and EBF models, while the remaining 12 PGw points (30%) were reserved for validation. Final independent validation was performed using 17 Vertical Electrical Sounding (VES) points, which were excluded from the training process entirely.

Hybrid GIS-based bivariate analysis for groundwater potential mapping

Selecting an appropriate decision rule is crucial, as it determines the ranking and weighting methodology used to integrate the influencing factors under predefined weights, thereby significantly affecting the analysis outcomes. This study employed two bivariate geospatial models, Weight of Evidence (WoE) and Evidential Belief Function (EBF), to evaluate the ranking of sub-class factors within the investigated basin.

All raster layers, except for lithology and curvature, were reclassified into five scoring levels using the natural breaks (Jenks) classification method, which operates at a spatial resolution of 12.5 m. This classification approach enhances differentiation among classes by identifying optimal threshold values that group similar data points based on inherent patterns. The resulting framework effectively represents the diverse topographic, geological, structural, hydrogeological, and climatic characteristics of the study area, each of which plays a role in influencing the spatial distribution of groundwater potential (GwP) within SCAq.

To examine the intricate relationships between dependent and independent controlling factors, 70% of the groundwater inventory points were utilized. The Index of Entropy (IOE) method was deployed to assign weights to the conditioning factors affecting GwP. Furthermore, a hybrid modeling framework was developed by integrating IOE with WoE and the EBF method to generate predictive GwP maps.

The resulting groundwater potential models were subsequently validated using the Area Under the Curve (AUC) approach, with the remaining 30% of the dataset employed for model reliability assessment. These models were developed using the Spatial Data Modeller (SDM) extension in ArcMap 10.8, along with XLSTAT and IBM SPSS Statistics. A comprehensive description of the adopted methodology is provided in the subsequent sections.

Index of entropy (IOE) model

The Index of Entropy (IOE) method is commonly employed to assess the uncertainty and instability of a system⁷⁶. In this study, the IOE was utilized to determine the predictive weights (\({\text{W}}_{\text{j}}\)) of each controlling factor based on the following equations⁷⁶.

where the frequency ratio (FR) quantifies the spatial association between a specific factor class and the target phenomenon. For a given groundwater factor, (A) represents the area of a specific class within that factor, while (B) is the total area of the entire factor. Similarly, (C) denotes the number of pixels in that class area, and (D) is the total number of pixels in the study area. The term (b) is the percentage area of a specific class relative to the factor’s total area, whereas (a) is the percentage area of the factor relative to the entire study domain. The probability density is denoted by Pij. Entropy values are expressed as H_j (actual entropy for factor j) and H_jmax (maximum possible entropy for that factor). S_j is the number of classes within factor j, I_j is the information coefficient reflecting the factor’s predictive power, and w_j is the resultant weight value assigned to the factor as a whole, which ranges between 0 and 1.

GWP map of the entropy method was prepared by overlaying all the thematic layers in terms of the obtained weights of each model through the spatial analysis tool in the ArcGIS package using the following equation:

where GWP is the groundwater potential zone; Wi is each layer weight and Ri the class rates within a thematic layer.

Weight of evidence (WOE) model

The Weight of Evidence (WoE) model is a data-driven, Bayesian method that estimates the probability of groundwater occurrence given a specific evidential class. For each class, it computes two weights: a positive weight (W⁺) when the class is present and a negative weight (W⁻) when it is absent. The contrast, C = W⁺ − W⁻, summarizes the class’s overall spatial association with groundwater; larger positive contrasts indicate a stronger favorable relationship.

These weights are derived by comparing the density of presence points (groundwater occurrences) within a class to the density of absence points outside that class. As a bivariate statistical technique, WoE combines simple logical ratios with Bayesian probability to quantify the evidence provided by each factor class for predicting groundwater presence⁷⁷. This produces objective, class-level weights that reduce subjective bias and help reveal complex interdependencies among conditioning factors⁷⁸. The positive and negative weights for each sub-category are computed using the following equations^79,80.

where F and \(\overline{\text{F} }\) represent the presence and absence of conditioning parameters, respectively; P denotes probability; and G, and \(\overline{\text{G} }\) correspond to the presence and absence of groundwater, respectively.

Standard deviation of W is calculated as follows⁸⁰:

where S²W⁺ and S²W⁻ are defined as the variances of positive weights and negative weights, respectively.

The final WOE coefficients (\({R}_{WOE}\)) assigned to each factor class can be obtained as follows.

where \({W}_{min}^{-}\) represents an overall measure of spatial association between the training points and the evidential theme, combining the effects of the two weights. Sometimes, W⁺ can be close to zero, yet W^– is strongly negative. This situation arises if the presence of the theme is not particularly predictive of training points, but the absence of the theme provides strong evidence that points are unlikely to occur. Conversely, there can be an imbalance between the absolute values of W⁺ and W^– in the other direction, or the two weights can have absolute values in about the same range.

According to Bonham-Carter⁸⁰, the absolute values of weights provide a measure of predictive strength: values between 0 and 0.5 are mildly predictive; values between 0.5 and 1 are moderately predictive; values between 1 and 2 are strongly predictive; and values greater than 2 are extremely predictive.

The final \({GwP}_{\text{W}\text{O}\text{E}-\text{I}\text{O}\text{E}}\) map was generated by applying the inferred IOE weight (\({\text{W}}_{j}\)) to the ranked numerical values of factor classes, represented as attributes of \({R}_{WOE}\).

In standard WOE, classes with zero presence or zero absence points yield infinite or undefined weights, causing numerical instability. In the study area, for example, the basement bedrock lithological class contained zero productive groundwater points compared to the Nubian Sandstone and Alluvium deposits. To address this, a correction factor of 0.5 was added to all presence and absence counts following Bonham-Carter⁸⁰, using the ArcSDM toolbox. This correction yielded a finite negative weight (W⁺ = − 1.25) for the basement bedrock class, correctly indicating unfavorable conditions while preserving numerical stability.

Assessment of predictor independence

A fundamental assumption of the Weight of Evidence model is that evidential themes are conditionally independent with respect to the training points⁸⁰. Violation of this assumption can lead to overestimation of posterior probabilities and unstable weight estimates⁸¹. To evaluate the degree of dependence among the ten conditioning factors, a multicollinearity assessment was conducted using pairwise Pearson correlation and Variance Inflation Factor (VIF).

Pearson correlation coefficients (r) were calculated between all factor pairs (Table S1), with |r|> 0.7 indicating a strong correlation⁸². VIF was computed as VIF = 1/(1 − R²), where values exceeding 5 indicate moderate multicollinearity concerns (Table S2)⁸³. Strong correlations were observed between physically coupled pairs: Slope–TWI (r = − 0.82), LD–LSID (r = 0.86), DTS–SD (r = 0.71), and Slope–Rb (r = 0.73). VIF values remained within acceptable ranges (82,83

These correlations reflect the inherent physical relationships among the factors rather than data artifacts. Although all factors were retained due to their conceptual distinctiveness, WOE is known to be sensitive to conditional dependence. To address this, the ten factors were organized into five conceptual groups based on their hydrogeological roles: topographic (Slope, TWI, Curvature), structural (LD, LSID), hydrologic (DTS, SD, Rainfall), geologic (Lithology), and surface (Radar backscattering). Within each group, moderate to strong correlations are expected because the factors capture different dimensions of the same subsurface process. For example, LD measures the frequency of lineaments, while LSID targets their intersections. Together, their combined contribution more effectively represents structural control on groundwater flow than either factor alone. Rather than treating correlated factors as independent predictors, this grouping framework interprets their joint contribution as reflecting an integrated hydrogeological process^82,83.

Evidential belief function (EBF) model

The Evidential Belief Function (EBF) is a knowledge-driven spatial modeling framework used here to assess groundwater potential by integrating multiple conditioning factors⁸⁴. In this approach, thematic layers representing factors that control groundwater occurrence (e.g., lithology, slope, land use) are treated as distinct pieces of evidence. The model then synthesizes this evidence to produce predictive groundwater potential maps⁸⁰. This synthesis is achieved by quantifying the spatial correlation between each evidential layer and the locations of existing groundwater wells. By evaluating how strongly each factor class is associated with well presence, the model derives weights that form the basis for the final integration. This data-driven correlation enhances the objectivity and reliability of the predictive map.

A key advantage of the EBF model is its capacity to not only delineate prospective groundwater zones but also to explicitly quantify the uncertainty inherent in those predictions⁸⁴. This dual output is particularly valuable for decision-making, as it provides stakeholders with a measure of confidence in the mapped zones, thereby supporting more nuanced and effective groundwater management strategies.

The model is formalized within the Dempster-Shafer theory of evidence, which generalizes traditional Bayesian probabilities into lower and upper probability bounds⁸⁵. In this framework, the lower and upper probabilities correspond to the belief (Bel) and plausibility (Pls) measures, respectively. The framework also explicitly quantifies related measures, including the degree of disbelief (Dis), which represents support against a proposition, and the degree of uncertainty (Unc), which reflects a lack of knowledge or conflict between pieces of evidence⁸⁶. Together, these four functions (Bel, Dis, Unc, and Pls) provide a comprehensive representation of the evidential support for groundwater occurrence across the study area.

A detailed explanation of the algorithm is provided by Carranza et al.⁸⁵. In the context of groundwater potential mapping using EBF^86,87, the frame of discernment is defined by Eqs. (14) and (15):

where:

λ(Tp)E_ij = Weight of evidence for the presence of groundwater given the class (j) of factor (i). This is a dimensionless ratio.

N (G ∩ E_ij) = Number of pixels containing groundwater wells that also fall within class (j) of factor (i).

N(G) = Total number of groundwater well pixels in the entire study area.

N(E_ij) = Total number of pixels comprising class (j) of factor (i).

N(C) = Total number of pixels in the entire study area.

The belief measure is given by:

The degree of disbelief is defined as:

Uncertainty is calculated as:

Plausibility (Pls, representing the maximum possible support for groundwater occurrence. It is the sum of belief and uncertainty, and is dimensionless. Pls is expressed as:

The final GwP map generated using the EBF-IOE model was obtained by applying the inferred IOE weight (Wj) to the ranked numerical values of factor classes represented as attributes of Bel:

The GwP maps were classified into five susceptibility levels: very low, low, moderate, high, and very high, employing the natural breaks classification method in ArcMap.

Validation of groundwater potential (GwP) models

The validation of GwP models was conducted using a combination of statistical, spatial, and field-based approaches to ensure robust evaluation of predictive performance and generalizability. Given the relatively small dataset and the influence of spatial autocorrelation, multiple complementary validation techniques were employed.

A repeated k-fold cross-validation (CV) framework was implemented to overcome the limitations of a single random data split. The dataset, comprising 40 productive groundwater points and 40 absence points (n = 80), was subjected to 5 iterations of fivefold CV. In each iteration, the data were randomly divided into five subsets, with four folds (64 points) used for training and the remaining fold (16 points) used for testing. This repeated resampling generated a distribution of performance metrics, allowing the computation of mean Area Under the Curve (AUC) values and corresponding standard deviations, thereby providing a reliable estimate of model accuracy and associated uncertainty.

To further address the effects of spatial autocorrelation, spatially blocked cross-validation was applied as the primary test of model generalization. The study area was divided into five spatially contiguous blocks based on natural geographic boundaries. Models were iteratively trained on four blocks and evaluated on the remaining block until each block had been used as a test set. Importantly, the spatial blocks used for validation were defined independently of the groundwater occurrence clustering employed during model development. The blocked cross-validation was applied only during model evaluation and did not influence the generation of training samples or the spatial distribution of productive and non-productive groundwater points. Consequently, the blocked validation provides an independent assessment of model generalization capability in previously unsampled areas and offers a more conservative estimate of predictive performance than conventional random cross-validation.

Model discrimination ability was evaluated using the Receiver Operating Characteristic (ROC) curve analysis^30,88. For each validation iteration, ROC curves were generated by plotting the True Positive Rate (TPR; sensitivity) against the False Positive Rate (FPR; 1 − specificity) across a range of classification thresholds. The Area Under the ROC Curve (AUC) was calculated using the trapezoidal rule^85,86:

where TPRi and TPRi + 1 represent the true positive rates, and FPRi and FPRi + 1 represent the false positive rates at consecutive threshold values (i) and (i + 1). Final reported AUC values represent the mean across all CV iterations with associated standard deviation. AUC values range from 0 to 1; values above 0.7 are considered acceptable, and those above 0.9 indicate outstanding predictive accuracy⁸⁸.

In addition to statistical validation, frequency ratio analysis was employed to assess the spatial agreement between observed groundwater occurrences and predicted suitability classes. This evaluation was based on two assumptions: (i) high and very high GwP zones should occupy a relatively smaller proportion of the study area, and (ii) a reliable model should assign the majority (> 50%) of observed groundwater points to at least moderate suitability zones, with higher concentrations in high and very high classes. This spatial validation provides an intuitive measure of model consistency with observed patterns.

Finally, independent ground truth validation was conducted using 17 VES points and field observations from existing wells and springs located within high and very high potential zones. These VES data were not included in model training, ensuring independence of validation. The geophysical data provided critical insights into subsurface conditions, including lithological variations, aquifer thickness, depth to bedrock, groundwater levels, salinity, and structural features such as faults and fractures.

To further support model validation, detailed geophysical investigations were carried out in high and very high GwP zones of Wadi Hodein. The integration of aeromagnetic data and VES measurements enabled the identification of subsurface lithological variations and structurally controlled aquifer systems. Aeromagnetic data delineated faults and fractures, while VES data characterized resistivity variations associated with groundwater saturation and salinity. The strong agreement between geophysical evidence and predicted high-potential zones confirms the reliability of the GwP models in identifying structurally controlled groundwater systems.

Geophysical investigations

Detailed geophysical investigations were conducted within the high and very high GwP and the surrounding zones of Wadi Hodein to assess the role of subsurface structures in controlling aquifer systems (Fig. 5a). Drilled wells and natural springs show significant variability in depth, yield, and water quality over short distances, reflecting the complexity of the subsurface and the presence of diverse aquifer systems.

A DC resistivity survey (17 VESs) was intentionally targeted within these zones to address subsurface complexities, specifically fault impacts and basement relief, that remote sensing and WOE modeling cannot fully capture. Low-potential zones were not surveyed due to logistical constraints and the focused goal of validating high-potential targets.

To overcome the surface limitations of remote sensing and RadarSat, aeromagnetic and VES data were integrated. Aeromagnetic data delineated lithological variations and structural features (faults and fractures), while VES provided resistivity profiles to identify aquifer boundaries, saturation, and water salinity. Together, these geophysical approaches validated and strengthened the understanding of structurally controlled groundwater systems in the study area.

Processing of aeromagnetic data

Magnetic anomalies occur due to the uneven distribution of magnetic minerals in the Earth’s upper crust, which leads to variations in the magnetic field. The spatial patterns and distribution of these anomalies offer valuable insights into the presence of magnetite-bearing rock formations and the depth of the basement surface¹⁹. This information helps clarify the underlying subsurface hydro-structures¹⁸.

The original airborne magnetic survey was conducted through a collaborative effort between the Egyptian government and the Aero-Service Division of Western Geophysical Company, USA⁶³. Flight lines were flown in the NE-SW direction with 1.5 km spacing, while tie lines were oriented NW–SE with 10 km spacing, at a nominal flying altitude of 120 m above ground level⁶³. Preprocessing included diurnal variation correction, International Geomagnetic Reference Field (IGRF) removal (IGRF 1975), and micro-leveling using standard tie-line adjustment procedures⁶³.

The Reduced-to-Pole (RTP) transformation was applied to compensate for the Earth’s magnetic field inclination, relocating anomaly maxima directly above their causative sources⁸⁹. Using Geosoft Oasis Montaje V. 8.3.3, the RTP was computed with an inclination angle of 35.5° and a declination angle of 2.5°, applied to the gridded data (100 m cell size) via a wavenumber domain filter.

The Tilt Angle Derivative (TAD), an edge detection technique, was employed to delineate subsurface magnetic source boundaries, with fault locations identified from zero-value contours⁹⁰. The TAD was calculated from the RTP grid using a 3 × 3 convolution kernel in Geosoft Oasis V. 8.4. Additionally, the Analytical Signal (AS) technique was utilized to enhance near-surface features, with the AS grid filtered using a 3 × 3 median filter to reduce noise while preserving structural boundaries.

Vertical electrical sounding (VES)

DC resistivity surveys are widely used in groundwater exploration to delineate aquifer geometry, lithological variations, and fracture networks via resistivity contrasts⁹¹. When integrated with remote sensing-derived lineament density and surface structural data, they better characterize recharge systems by linking surface lineaments with subsurface conductive zones and improving 3D aquifer modeling⁹².

At the local scale, a Schlumberger DC resistivity survey was conducted to delineate potential groundwater zones by analyzing vertical resistivity variations. Seventeen VES points were measured using a four-electrode Schlumberger configuration with a maximum electrode spacing of 1000 m, enabling investigation of subsurface layers to depths of about 200–250 m for groundwater potential assessment (Fig. 5b). These soundings were acquired along three profiles to assess how subsurface structures and geology influence the aquifer within the identified high groundwater potential (GwP) zone. The survey was also designed to improve the lateral mapping of the Nubian Sandstone (NSAS), especially along the fault zone where natural springs occur between basement rocks and the NSAS. All 17 VES soundings were collected before the WoE and EBF modeling, and were positioned within the NSAS and adjacent basement exposures to support the conceptual hydrogeological model and evaluate structural controls on groundwater distribution.

These profiles were defined using spatial analysis of lineament density and drainage patterns, field-observed lithological variations, and data from available boreholes and springs. Profile AA’ (NW–SE) connects the Abraq spring to the Wadi Dif well, crossing Wadi Hodein and multiple east–west lineaments. Profile BB’ (E–W) follows Wadi Hodein through Wadi Abu Saafa, where variations in drilled-well depth and discharge indicate the role of basement relief and faults. Profile CC’ (N–S) links upstream drainages of Wadi Abu Saafa, supporting recharge along Wadi Abu Saafa and Wadi Hodein.

The sounding curves were inverted using a 1D Newton-based inversion scheme in IPi2win Ver. 3.0 software⁹³. To reduce non-uniqueness (equivalence), the inversion was constrained using nearby boreholes: each profile contains at least one well, and its lithology was used to build the initial reference model near each VES, guiding the layer number, initial resistivities, and water-table depth from observed stratigraphy. The RMS% was limited to 3–5%, and the resulting uncertainty was used to estimate bounds on interpreted resistivity and thickness for each VES.

Finally, geoelectrical results were calibrated with lithological information from well logs (Fig. 5b), enabling mapping of vertical and lateral variations in true subsurface resistivity and associated structures (Fig. 5c). Figure S1 (Supplementary) presents representative VES points and their geological interpretations.