Water Balance Estimates and Piezometric Level Lowering Based on Numerical Modeling and Remote Sensing Data in the Recife Metropolitan Region—Pernambuco (Brazil)

Abstract

A more precise definition of groundwater dynamics is an urgent issue for developing reliable plans to assist in the sustainable management of these resources. The combination of remote sensing input data with groundwater flow models emerges as a tool capable of representing these dynamics and simulating important conditions for developing adequate groundwater exploitation plans. These technologies allow for a more detailed and accurate analysis of the interactions between the factors influencing aquifers’ behavior, such as climate variability and anthropogenic pressure on water resources. Thus, the present study aims to develop a numerical model of groundwater flow in the aquifers of the Recife Metropolitan Region, state of Pernambuco, in Brazil, to evaluate the dynamics of these waters and the piezometric level drawdowns between 2004 and 2023. The FREEWAT platform, which applies the MODFLOW-2005 code, was used to simulate the study area. The results showed the entry of seawater into some formations and drawdowns that reached more than 100 m at some points, indicating the urgent need for management strategies to mitigate salinization and preserve the quality of the region’s groundwater resources.

1. Introduction

Aquifers are extensive groundwater reservoirs since approximately 68.5% of all the freshwater on Earth are stored in glaciers. In comparison, most of the remaining 31.5% are estimated to be stored as groundwater [1]. Therefore, groundwater has the potential to offer society enormous social, economic, and environmental benefits, including adaptation to climate change [2].

Groundwater is the source of one-third of all freshwater withdrawals and provides approximately 36%, 42%, and 27% of the water used for domestic, agricultural, and industrial purposes, respectively [3]. It is estimated that over the past 50 years, the global groundwater extraction rate has at least tripled and continues to increase annually at a rate of approximately 1% to 2% [4]. Given the increasing demand for food, the percentage of agricultural areas irrigated with groundwater is expected to rise over the years [5,6,7]. Additionally, scientific advancements in agriculture have increased productivity and crop production [8], leading to higher consumption of groundwater resources.

As with any other resource, increased exploitation can lead to a wide range of problems. The excessive use of groundwater has resulted in a significant decline in aquifer levels, causing various water-related issues and conflicts that threaten the sustainability of these waters [9,10]. Saltwater intrusion in coastal aquifers is another major consequence of overexploitation. According to Carrera et al., 2010 [11], the progression of saltwater intrusion is influenced by coastal topography, water management, aquifer geological heterogeneity, the initial distribution of salts, and relative sea level rise. Consequently, the lowering of deep aquifer levels can intensify the movement of saline sources from upper to lower layers [12,13,14,15]. According to the Intergovernmental Panel on Climate Change [16], the global mean sea level rise increased from 1.4 mm per year during the period 1901–1990 to 3.2 mm per year between 1993 and 2015. Wassef and Schüttrumpf 2016 [17] developed scenarios to assess the impacts of rising sea levels in the Mediterranean Sea and groundwater extraction on aquifer salinity in the Nile Delta region, Egypt. The results indicated that, in scenarios where the extraction rate is relatively low, sea level rise is the primary driver of groundwater salinization. Conversely, in scenarios dominated by groundwater extraction, salinity levels vary between 100 and 5000 mg/L, affecting approximately 10% of the study area, which covers around 12,000 km².

Land subsidence is another adverse effect of excessive groundwater exploitation. Numerous studies have examined regions affected by this phenomenon, such as the work of Ehteshami, Salari, and Zaresefat 2016 [18] on the Damghan Basin, Iran, which demonstrated that the average groundwater level in the basin declined by 5.88 m between 2000 and 2010, with a land subsidence rate of approximately 0.58 m per year, primarily due to overextraction. In the city of Changzhou, China, researchers recorded a maximum groundwater drawdown of up to 18 m between 1960 and 2002, affecting an area of more than 2000 km² [19]. A similar study reported subsidence induced by groundwater extraction in the Mekong Delta, Vietnam, with an average rate of 0.018 m per year between 1991 and 2016 [20].

It is a matter of extreme urgency to define groundwater dynamics to produce reliable plans for the sustainable management of this resource in terms of quantity and quality [21]. This dynamic can be represented through models that simulate natural processes in a simplified manner.

Numerical models serve as simplified representations of real groundwater systems, employing mathematical equations solved through specialized computer programs [22,23]. They can provide managers with a robust framework based on physical properties to evaluate complex hydrologic systems, which would be difficult or impossible to represent analytically [24].

By estimating the actual conditions of aquifer systems, modeling can be used to quantitatively determine groundwater conditions and simulate future scenarios, generating helpful information for developing management plans. Several studies have shown the potential for applying numerical models in groundwater resource management [25,26,27,28,29].

Groundwater modeling involves a series of steps, the first of which is defining the conceptual model. Developing conceptual models is a crucial first step, as they provide a foundational understanding that can be subsequently refined and converted into numerical groundwater models [30,31,32].

Groundwater models rely on spatially continuous and coherent hydrogeological properties datasets. These properties control water through the subsurface [33]. From this perspective, the influence of heterogeneity in numerical aquifer modeling is a crucial aspect. The spatial variability of hydraulic conductivity, for example, presents significant challenges for numerical modeling, making it essential to consider the heterogeneous structure of the medium for more accurate predictions [34]. However, according to the IPCC report [35], limitations in the spatiotemporal coverage of groundwater monitoring networks, exploitation data, and data that allow for the numerical representation of groundwater recharge processes continue to constrain the understanding of these systems. Therefore, the effectiveness of such models is often compromised by data scarcity, uncertainties in conceptual frameworks, and poorly constrained parameterization [36].

In this sense, the advancement of remote sensing in recent years, with new technologies and techniques, has enabled the wide availability of information for natural resource management and for different users [37]. In hydrogeology, this growth represents a unique technological opportunity to meet the need for better spatial and temporal data distribution regarding groundwater modeling.

Geoprocessing and remote sensing have been essential tools for studies on water resources. This use is associated with hydrological cycle modeling techniques [38,39,40,41], where processes such as soil water flow are also simulated. In other cases, they are applied to broader water resource management and planning efforts [42,43,44,45,46,47,48,49]. With respect to groundwater, one of the key parameters for modeling is the aquifer recharge rate. Determining recharge represents a substantial challenge, as it cannot be measured directly and is strongly influenced by inaccuracies, inappropriate assumptions, and the scarcity of reliable, continuous data—both spatially and temporally—required for its estimation [50,51,52]. Recharge estimation is commonly performed using methods based on direct water balance calculations [53,54,55,56,57], or through hydrological models that simulate the water balance [58,59,60,61,62].

Components of the water balance, such as precipitation and evapotranspiration, can be derived from remote sensing data and used to estimate recharge in groundwater modeling studies. González-Ortigoza et al., 2023 [62] used CHIRPS precipitation data and other global soil and temperature datasets in the Soil-Water-Balance (SWB) model, achieving results consistent with observed data, showing that remote sensing can be successfully used for modeling areas with limited data. El-Hadidy and Morsy 2022 [63] used modeling and combined the results with remote sensing data to produce a groundwater potential map in the Nile Valley region. Belay et al., 2024 [64] applied the distributed WetSpass model using spatial data from CHIRPS for precipitation and TerraClimate for evapotranspiration, and compared the recharge results with those obtained using traditional methodologies based on measured data. The authors reported a 72% correlation between the estimates derived from remote sensing data and those from observed data, indicating that remote sensing is a viable and reliable alternative for groundwater recharge studies.

GRACE and InSAR products have been widely used to monitor changes in groundwater storage and land subsidence, both individually and through an integrated approach. Fallatah et al., 2019 [58] estimated aquifer recharge rates using the water storage variation balance from GRACE data, finding a recharge rate of 5.21 km³/year. Santarosa et al., 2024 [65] obtained a recharge rate ranging from 114.5 mm/year to 284.5 mm/year based on GRACE data for the Guarani and Bauru aquifer systems in São Paulo, Brazil. Castellazzi et al., 2018 [66] integrated GRACE and InSAR data to generate mass distribution maps, which yielded results closely aligned with official groundwater balance estimates. Similarly, Ouyang et al., 2024 [67] explored the feasibility of combining GRACE and InSAR, concluding that land subsidence predominantly occurs in areas experiencing significant groundwater storage depletion.

Evidently, aligning the spatial scale of research with that of remote sensing products is crucial to ensure that model inputs accurately reflect the spatial variability of soil properties. At finer scales, soil heterogeneity is more pronounced, whereas at broader scales, hydraulic parameters tend to appear more homogeneous [68].

MODFLOW 2005 is one of the most popular numerical modeling software for simulating groundwater flow. Due to its modular structure and detailed documentation, its code has been successfully coupled with many other physical process models and programs, representing more complex hydrogeological processes, such as contaminant transport, saline intrusion, and climate change.

Integrating geographic information systems (GISs) and hydrological codes facilitates the use of complex modeling environments, allowing for the storage, management, and visualization of large spatial datasets [69]. In this context, FREEWAT emerges as an open-source platform integrated with QGIS, designed to simulate various hydrological processes and support decision-making in water resource management [70]. Built on the MODFLOW code, it provides a robust framework for modeling groundwater flow and related hydrological dynamics.

Although it can still be considered a new tool, the vast application capacity of FREEWAT has already been highlighted in several studies regarding coastal aquifer management, modeling interactions between groundwater and surface water, and impacts of climate change on groundwater resources [71].

Menichini and Doveri 2020 [27] defined a conceptual hydrogeological model for an important coastal aquifer in the Versilia Plain, Italy, using an integrated multidisciplinary approach based on stratigraphic, hydrogeological, and geochemical data. The conceptual model was then transformed into a numerical groundwater flow model using FREEWAT. The results of the model’s water balance, considering extreme events in the region and the prospect that they may occur more frequently, showed that these represent a real threat to the aquifer system, which could significantly increase saltwater intrusion.

Chrysanthopoulos et al., 2024 [21] modeled the aquifer system of the Argolis Plain in Greece using the FREEWAT platform to simulate groundwater flow within the unconsolidated aquifer. The results demonstrated significant potential for informing the development of effective management plans for the sustainable use of this aquifer.

In the state of Pernambuco, Brazil, underground water deposits are concentrated in areas of sedimentary basins, covering only 13.6% of the territory [72]. This limited extension results in a relatively low hydrogeological potential for the state. The Recife Metropolitan Region (RMR), located along the state’s coast, experiences the highest level of groundwater exploitation compared to all other regions.

Historically, the RMR public water supply has depended mainly on surface water and dams [15]. However, the aquifer systems are an important drinking water source for one of Brazil’s largest metropolitan regions. The water supply service provided by aquifers is estimated to represent approximately 18% of the region’s total demand [73]. Well drilling began in the 1970s and intensified between 1993 and 1998, when the RMR was subjected to a long period of water rationing due to a severe drought [74]. Excessive pumping rates can prevent the aquifer from reaching a new equilibrium condition, leading to continued depletion of groundwater storage [75]. According to [72], approximately 68% of the wells registered throughout the state are in the RMR.

Thus, through FREEWAT modeling in the RMR, this study aims to use remote sensing products in conjunction with in situ monitoring data as input for a distributed numerical model, to characterize the spatiotemporal dynamics of water fluxes in a multilayer coastal aquifer to provide a quantitative estimate of the water balance in such a system, and to evaluate groundwater drawdown in the study area.

2. Materials and Methods

2.1. Geological and Hydrogeological Characterization of the Study Area

The RMR is located in Pernambuco state, covering an area of 2761.45 km². Composed of 14 municipalities, the population of the RMR is approximately 3.7 million inhabitants [76]. Geologically, the Recife Metropolitan Region (RMR) can be divided into four primary domains: (1) the Paraíba Basin to the north, (2) the Pernambuco Basin to the south, (3) the Recife Plain Domain in the central area, and (4) the Crystalline Basement to the west. The Pernambuco Lineament (LPE), the region’s most prominent geological structure, is the boundary between the two coastal sedimentary basins, Paraíba and Pernambuco.

The Paraíba Basin predominantly comprises the Beberibe Formation, whereas the Pernambuco Basin is mainly characterized by the Cabo Formation [77]. Both basins consist primarily of Cretaceous siliciclastic sediments derived from the erosion of the regional crystalline basement [78]. Figure 1 illustrates the aquifer formations within the RMR.

The Paraíba Basin comprises Beberibe, Gramame, Maria Farinha, Barreiras, and Coberturas Quaternárias Formations. The Beberibe Formation has an average thickness of 180 m, with maximum values that can reach more than 211 m in the coastal zone between Olinda and Itamaracá [79]. Together with the Barreiras Formation and Quaternary Covers, it forms interstitial aquifers, such as the Beberibe aquifer.

In the Northern Zone of the RMR, the Barreiras Formation represents a significant aquifer, covering more than 50% of the area. It is characterized by a phreatic aquifer sequence with confined layers at greater depths [80]. According to Pfaltzgraff et al., 2003 [81], this formation is minimally exploited when overlain by the crystalline basement, with groundwater extraction primarily carried out through shallow wells and Amazonas wells. However, the Barreiras aquifer plays a crucial role as a recharge zone for the Beberibe aquifer, particularly along sandy pockets with high permeability [80].

The Beberibe is the RMR primary aquifer, exploited for domestic supply and mineral water companies [80]. It is limited at its base by the impermeable substrate of the crystalline basement and the top by the limestones of the Gramame and Maria Farinha Formations or by the sandy-clay sediments of the Barreiras Group. The Beberibe has an average thickness of 180 m that increases towards the coast, reaching more than 320 m in the municipalities of Olinda and Paulista [81]. The increasing exploitation of this aquifer has led to overexploitation conditions in some areas, which contributed to lowering the water level by a few tens of meters, consequently leading to local flow alteration. These scenarios begin to occur in the most potentiometrically depressed areas [80].

The Gramame and Maria Farinha Formations comprise limestone lithotypes, constituting a karstic-fissural aquifer with poor-quality water due to its excessive hardness. Thus, the found water is unfit for human consumption and other uses, reducing the number of wells in these formations [80].

The Pernambuco Basin includes the Cabo, Estiva, and Algodoais Formations, the rocks of the Ipojuca Formation that the Barreiras Group partially covers, and the Quaternary Covers, which occupy larger areas. Given their sandy and sandy-clayey nature, the Cabo, Algodoais Formations, the Barreiras Group, and the Quaternary Covers give rise to interstitial aquifers [80]. The Estiva Formation, of a limestone-silty-clayey nature, and the Ipojuca Formation, consisting of volcanic rocks, are characterized by very low or no productivity, which is why their interest as exploitable aquifers is irrelevant [80].

The Cabo aquifer is the primary groundwater source within the Pernambuco Basin. It features both unconfined and confined zones and is exploited at depths reaching the impermeable substrate formed by the basalt flows of the Ipojuca Formation or the granitic units of the crystalline basement [80]. Despite its low permeability, the Cabo aquifer presents problems of overexploitation since many wells were built to supply residential condominiums recorded in its southern coastal zone, comprising the neighborhoods of Pina, Boa Viagem and Setúbal (municipality of Recife), and Piedade and Candeias (municipality of Jaboatão dos Guararapes). As a result, many wells have been contaminated by saline water [82].

The Crystalline Basement comprises lithotypes from the gneiss-migmatite complexes of Belém do São Francisco and Vertentes, along with various granitoids. These rock formations may either outcrop or be overlain by deposits from the aforementioned sedimentary basins [79].

The Recife Plain is composed of sediments of continental and marine origin, with thicknesses ranging from 50 to 90 m, overlying the Cretaceous and Tertiary sediments of the two other sedimentary basins in the area. Costa et al., 1994 [83] referred to these subsurface sediments as the Boa Viagem aquifer. Although they serve as good local groundwater reservoirs, the aquifer’s significance is limited due to its high vulnerability to contamination from saline and/or polluted waters [79].

2.2. Conceptual Model

A conceptual model aims to define and understand the hydrogeological conditions and hydrodynamic behavior of the study area. In this paper, the distribution of hydrogeological formations led to their discretization into three layers (Figure 2). The first layer represents the Barreiras Formation. The second layer represents the Quaternary Formations (Alluvial Deposits, Coastal Deposits, and Mangrove Sediments), Gramame, Algodoais, Estiva, Maria Farinha, and Ipojuca, in addition to Cabo and Beberibe when they are outcropping. The third layer represents the Beberibe and Cabo Formations when they are underlying. The main discharge area is the ocean, which is essential for the model’s functioning and is simulated as one of the boundary conditions.

2.3. Numerical Model

Numerical models use a governing flow equation derived from two basic principles: the law of conservation of mass and Darcy’s law. According to both laws, groundwater flow in three dimensions can be described by the following partial differential equation:

$${\displaystyle \frac{\partial }{\partial \mathrm{x}}} \left({\mathrm{K}}_{\mathrm{x}}{\displaystyle \frac{\partial \mathrm{h}}{\partial \mathrm{x}}}\right)+{\displaystyle \frac{\partial }{\partial \mathrm{y}}} \left({\mathrm{K}}_{\mathrm{y}}{\displaystyle \frac{\partial \mathrm{h}}{\partial \mathrm{y}}}\right)+{\displaystyle \frac{\partial }{\partial \mathrm{z}}} \left({\mathrm{K}}_{\mathrm{z}}{\displaystyle \frac{\partial \mathrm{h}}{\partial \mathrm{z}}}\right)={\mathrm{S}}_{\mathrm{s}}{\displaystyle \frac{\partial \mathrm{h}}{\partial \mathrm{t}}}-\mathrm{W}$$

(1)

where h is the total head; K_x, K_y, K_z are the main components of the hydraulic conductivity tensor; S_s is the specific storage coefficient; and W represents water sources or sinks within the aquifer.

Hydraulic heads are calculated based on the areas of interest. In contrast to analytical solutions, numerical solutions are not continuous in space or time. Hydraulic heads are computed at discrete points, known as nodes, in space and at specified time intervals. These models allow for the resolution of problems with a high level of complexity.

MODFLOW is a computer code for simulating three-dimensional groundwater flow through porous media [84]. The first version of MODFLOW was developed by the United States Geological Survey (USGS) [84]. Since then, the code has been updated with new parameters and functionalities and is widely used around the globe. The numerical modeling developed with the MODFLOW-2005 code uses the finite difference method to simulate flow in three dimensions [23]. In the finite difference method, the system described in Equation (1) is replaced by a finite set of discrete points in space and time, and the partial derivatives are replaced by terms calculated from the differences in the hydraulic head values [23].

FREEWAT is a free and open-source plugin available in the QGIS 2.18 software, developed through a partnership of developers from European countries, using resources from the European Union’s Horizon 2020 Research and Innovation program [85]. It is a modeling platform that includes codes developed by the USGS, such as MODFLOW. Integrating the platform and QGIS allows for the pre-processing, post-processing, and spatial analysis of the results within a geographic information system (GIS), avoiding time-consuming file transfers between GIS and computer modeling programs.

The groundwater flow model for the RMR was implemented in FREEWAT using the MODFLOW-2005 code. The study area of 2791.93 km² generated a horizontal grid with an inclination of 11.9° composed of 19,504 cells measuring 200 m × 200 m. The model is subdivided into three vertical layers to include all the considered formation (Figure 2). All cells outside the domain in each layer were deactivated, leaving 2962 active cells in layer 1, 4294 active cells in layer 2, and 2979 active cells in layer 3.

Topographic data were obtained from the PE3D Digital Terrain Model, a laser profiling of the entire state of Pernambuco with altimetric accuracy greater than 25 cm (https://pe3d.pe.gov.br/, accessed on 13 April 2025) (Figure 3). These data were used to determine the upper surfaces of the aquifers in layer 1, and in layers 2 and 3 when not overlain by aquifers from the immediately overlying layer. Drilling data related to the depths of each aquifer formation, as presented by Leitão et al., 2017 [80], were interpolated and extrapolated to the RMR. By correlating these depth values with the upper surfaces, the boundaries between the various aquifer formations analyzed in this study were determined. Each entity generally presents variable thickness, which tends to be thinner in the west and thicker towards the ocean. Figure 4 shows the surfaces of the three layers considered in the model, with their top and base elevations.

The hydraulic properties are assigned according to each hydrogeological entity in each layer. Figure 5 shows the spatial distribution of the initial values of the hydrogeological parameters associated with each formation. These parameters were from wells measurements adapted from previous works [80,86,87]. The point data were interpolated using QGIS and converted into distributed layers for each aquifer to more accurately capture the heterogeneity of the geological formation. Nevertheless, the limited availability of data for each aquifer introduces a degree of uncertainty into the parameterization of the model.

In the numerical model, boundary conditions are determined by the hydrological conditions at the boundaries defined in the conceptual model. The flow directions calculated by a steady-state numerical model and most transient models are strongly influenced by the boundary conditions, which are an important component of a mathematical model [24]. MODFLOW-2005 has a series of packages to represent these conditions.

In the present work, two boundary conditions were used. The constant potential condition represents the coastal region in contact with the sea, using the Constant Head Package (CHD), where the piezometric elevation of 0 m was imposed. The rivers that pass through the RMR, such as the Capibaribe River, Beberibe River, and Tejipió River, are represented by their drainage network using the Drainage Package (DRN). Both boundary conditions are shown in Figure 6.

The recharge package (RCH) was used to simulate the direct infiltration of precipitation into the model domain. Recharge rates vary spatially and were applied considering the infiltration characteristics of the region. These rates were estimated using the BALSEQ model [88], which is based on precipitation and potential evapotranspiration data. Precipitation data were obtained from the CHIRPS database, a high-resolution (0.05° × 0.05°), daily precipitation dataset that integrates various remote sensing sources and has been widely used in recharge estimation studies [62,89] (http://data.chc.ucsb.edu/products/, accessed on 13 April 2025). Potential evapotranspiration was calculated using the Penman-Monteith method, based on meteorological data (maximum and minimum temperature, relative humidity, solar radiation, wind speed, and atmospheric pressure) from the Recife meteorological station managed by INMET.

The model also requires input data for the Curve Number (CN) and the maximum available water capacity in the soil for evapotranspiration (AGUT). These parameters were derived from the intersection of soil types and land use and land cover information. Soil data at a 1:100,000 scale were obtained from the Agroecological Zoning of Pernambuco (ZAPE) [90], while land use and occupation data were extracted from the MapBiomas Project—Collection 8, 2013 dataset (https://brasil.mapbiomas.org/, accessed on 13 April 2025). Recharge values were calculated for each unique soil–land use combination. The application of the BALSEQ model yielded a spatially distributed daily average recharge across the study area for each simulated year (Figure 7). This methodology improves the accuracy of recharge estimation by accounting for the spatial heterogeneity of soil characteristics and precipitation, thereby increasing the reliability of the groundwater model.

The exploration wells are simulated using the Well Package (WELL). Two hundred and ninety eight wells (Figure 3), cataloged in the work of Monteiro 2020 [86], were considered.

Simulation is conducted in two stages. The first step is the steady-state simulation, where the model was simulated with only one stress period lasting 19 years, representing the entire simulation period from 1 January 2004 to 31 December 2023. A stress period represents a time interval during which boundary conditions are assumed to remain constant, that is, recharge and extraction rates are held fixed throughout the period. The steady-state simulation is usually the first step in the modeling process since most transient models use a steady-state solution as initial conditions [21]. Next, the transient simulation is performed, divided into 17 stress periods (

Table 1. Stress periods and model time steps.

Stress Period (SP)	Length (Days)	Date	Recharge
1	731	1 January 2004–31 December 2005	Daily Average from 2004 to 2005
2	365	1 January–31 December 2006	Daily Average from 2006
3	365	1 January–31 December 2007	Daily Average from 2007
4	731	1 January 2008–31 December 2009	Daily Average from 2008 to 2009
5	365	1 January–31 December 2010	Daily Average from 2010
6	365	1 January–31 December 2011	Daily Average from 2011
7	366	1 January–31 December 2012	Daily Average from 2012
8	730	1 January 2013–31 December 2014	Daily Average from 2013 to 2014
9	365	1 January–31 December 2015	Daily Average from 2015
10	366	1 January–31 December 2016	Daily Average from 2016
11	365	1 January–31 December 2017	Daily Average from 2017
12	365	1 January–31 December 2018	Daily Average from 2018
13	365	1 January–31 December 2019	Daily Average from 2019
14	366	1 January–31 December 2020	Daily Average from 2020
15	365	1 January–31 December 2021	Daily Average from 2021
16	365	1 January–31 December 2022	Daily Average from 2022
17	365	1 January–31 December 2023	Daily Average from 2023

). The division of the 19-year period into 17 stress periods was based on an analysis of the annual average daily recharge. Years with closely similar averages were grouped into a single stress period.

Model calibration refines the numerical representation of the hydrogeological structure, hydraulic properties, and boundary conditions to achieve a desired degree of correspondence between the simulations and observations of the flow system [91]. This process helps establish the legitimacy of conceptual and numerical models and is therefore considered the most important step in modeling [24].

The degree of difficulty in calibration depends on the complexity of the simulated system and the quality and quantity of the observed data series. In complex systems with a large number of variables, which are often associated with uncertainties, the process of adjustment, calibration, and validation of numerical models is, in principle, the most effective way to correct the initial assumptions and bring them closer to reality [80].

Reilly and Harbaugh 2004 [22] states that parameter adjustment is only one aspect of model calibration. It can be performed either manually or automatically through non-linear regression. Poor model calibrations can often be attributed to an inadequate, inaccurate, or insufficient conceptual model [92].

In this study, the calibration process was carried out by trial and error by modifying each simulated layer’s hydraulic conductivity parameters and formation. Two hundred and sixty nine piezometric level points from the groundwater monitoring network (Figure 3) were used together with the model’s Head Observation Package (HOB). The observed data are available on the website of the Groundwater Information System—SIAGAS (https://siagasweb.sgb.gov.br/, accessed on 13 April 2025). These levels were compared with the simulated levels of the model. The limited number of observation wells and their poor spatial distribution—primarily concentrated in the central part of the study area—limit model calibration, introducing uncertainties into the results. Moreover, the absence of continuous data from these wells complicates the identification of hydrodynamic trends and patterns, adding further limitations to the modeling process. The flowchart of the methodology used in this study is shown in Figure 8.

2.4. Model Evaluation Statistics

Calibration quality can be assessed by visually comparing simulated results and applying statistical calculations. Residual errors are also useful for model evaluation through the Relative Root Mean Square Error (RRMSE) (Equation (2)) and Mean Absolute Relative Error (MARE) (Equation (2)) statistics. In transient models, other statistics, such as the coefficient of determination (R²) (Equation (3)) and Pearson’s Correlation (r) (Equation (4)), can also be used to compare simulations with observed data.

$$\mathrm{RMSE}={\displaystyle \frac{\frac{1}{\mathrm{n}} {\sum }_{1=1}^{\mathrm{n}}{({\mathrm{y}}^{\mathrm{obs}}-{\mathrm{y}}^{\mathrm{sim}})}^2 }{\overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}}}}}\times 100$$

(2)

$$\mathrm{MARE}={\displaystyle \frac{1}{\mathrm{n}}} \left({\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}\left|{\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}}-{\mathrm{y}}_{\mathrm{i}}^{\mathrm{sim}}\right|}{\left|{\mathrm{y}}_{\mathrm{i}}^{\mathrm{o}\mathrm{b}\mathrm{s}}\right|}}\right)\times 100$$

(3)

$${\mathrm{R}}^2={\left({\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}} - \overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}}}\right) \times \left({\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}} - \overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{sim}}}\right)}{\sqrt{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{({\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}} - \overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}}})}^2 \times \sqrt{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{({\mathrm{y}}_{\mathrm{i}}^{\mathrm{sim}} - \overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{sim}}})}^2}}\right)}^2$$

(4)

$$\mathrm{r}=\left({\displaystyle \frac{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}\left({\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}} - \overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}}}\right) \times \left({\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}} - \overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{sim}}}\right)}{\sqrt{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{({\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}} - \overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{obs}}})}^2 \times \sqrt{{\sum }_{\mathrm{i}=1}^{\mathrm{n}}}{({\mathrm{y}}_{\mathrm{i}}^{\mathrm{sim}} - \overline{{\mathrm{y}}_{\mathrm{i}}^{\mathrm{sim}}})}^2}}\right)$$

(5)

where n is the number of targets, y^obs is the observed head, and y^sim is the simulated head.

3. Results and Discussion

3.1. Model Calibration

Calibration was based on the average water levels from 2004 to 2023 at the 269 observed piezometric level points (Figure 3). Figure 9 shows the hydraulic properties obtained for each formation considered after the calibration in the transient regime.

The statistical parameters evaluated are shown in

Table 2. Parameter values after transient calibration.

Parameter	Value
R2	0.97
r	0.98
RRMSE	23.96
MARE	42.96

. The graph of simulated and observed levels is shown in Figure 10. The results indicate an excellent model performance in predicting the observed data, with high R² and r values, both close to 1, showing a strong correlation between the simulated and observed data. With an R² of 0.97 and r of 0.98, the coefficients indicate that the model can capture trends and patterns in the data. Overall, there is a good correspondence between the temporal patterns of simulated and measured piezometric levels, suggesting that the model is robust and has suitable parameters accuracy.

However, the relative error values—such as an RRMSE of 23.96% and a MARE of 42.96%—indicate that, despite a very good correlation, discrepancies remain between the simulated and observed values. While the RRMSE can be considered acceptable, it still reflects a relatively significant margin of error. The MARE, although within a reasonable range, is relatively high, suggesting that the model may produce predictions with considerable deviation from actual values, likely requiring further adjustments. These errors can be attributed to factors such as (1) uncertainties in the input data, (2) model limitations, (3) spatial and temporal resolution, and (4) aspects of the flow that were not fully represented.

These results also reflect the influence of piezometric measurements on the calibration process. The deficient spatial and temporal distribution of these measurements hindered the ability to establish a balance between the correlation of observed and simulated data and the spatial accuracy of the model’s responses. Additionally, the resolution of the numerical grid may not have been sufficient to accurately capture the aquifer’s heterogeneity, particularly considering the system’s complexity and the limited calibration data available. These factors restricted the determination of physically consistent K values and impeded the numerical model’s convergence. The interaction of these limitations not only increased the calibration complexity, but also constrained the model’s ability to adequately simulate the aquifer’s behavior, as evidenced by the RRMSE and MARE results.

Another contributing factor may be the limitations of the model itself. The consideration of variable-density flow is essential in the numerical modeling of coastal aquifers subject to saltwater intrusion, as groundwater density can be influenced by salinity gradients, altering flow patterns. Groundwater flow models based on Darcy’s equation assume constant density. To overcome this limitation, it would be necessary to incorporate the saltwater intrusion module available in FREEWAT. However, this would increase computational complexity and require additional measured data, such as salinity and temperature over time, which are also scarce.

In this context, technologies such as artificial intelligence (AI) and machine learning (ML), combined with the wide range of available remote sensing data, present promising solutions for filling data gaps and improving model predictions. The literature review by Ahmadi et al., 2022 [93] indicates that groundwater level modeling is the most common application of these technologies. For instance, Rafik et al., 2023 [94] integrated GRACE data with the SWAT model and machine learning techniques (Random Forest) to predict groundwater levels in unmonitored or data-scarce basins, achieving results that demonstrated minimal error and consistent water storage trends. Similarly, Malekzadeh, Kardar, and Shabanlou 2019 [95] simulated groundwater levels in the Kabodarahang aquifer, Iran, using three models: MODFLOW, Extreme Learning Machine (ELM), and Wavelet-Extreme Learning Machine (WA-ELM). Their findings showed that the WA-ELM AI model achieved the highest accuracy in groundwater level simulation, with MARE and RRMSE values of 0.34 and 0.0002, respectively.

Machine learning can also be a solution for enhancing groundwater monitoring networks. Teimoori, Olya, and Miller 2023 [96] proposed combining groundwater models and machine learning algorithms to identify observation wells and design optimal groundwater monitoring networks. In another study, Elmotawakkil, Sadiki, and Enneya 2024 [97] utilized data from the GRACE and MODIS satellites and advanced machine learning models, such as Random Forest (RF) and Gradient Boosting Regression (GBR), to predict groundwater levels in the Rabat-Salé-Kenitra region, Morocco. The GBR model performed exceptionally well, achieving an R² value of 0.99 and a mean absolute error (MAE) of 1.94. Evans et al., 2020 [98] employed the Extreme Learning Machine (ELM) to fill gaps in well time series, enabling the reconstruction of complete temporal datasets. Their study found that ELM-generated estimates were more accurate than those produced by the Kriging interpolation method.

The analysis of residuals between observed and simulated data, shown in Figure 11, reveals a highly variable distribution with positive and negative values. Residuals represent the discrepancies between measured values and those predicted by the model—that is, the extent to which the model deviates from observed reality. These fluctuations can be understood as a combination of systematic and random errors. The highest negative values, such as −19.43 m, −18.95 m, and −16.52 m, suggest substantial discrepancies between observed and simulated data, possibly reflecting model failures at measurement points not adequately represented by the model’s spatial resolution or in specific regions, where the piezometric level is underestimated. Conversely, positive residuals, such as 29.83 m, 26.68 m, and 22.12 m, indicate situations where the model tends to overestimate piezometric levels, possibly due to simplifications in the model or inadequate calibration parameters for these areas.

On the other hand, very low residuals, such as 0.35 m and 0.17 m, can be attributed to measurement errors or inaccuracies in the input data. Overall, the distribution of residuals indicates that the model performs well, with some areas, especially in the central region of the model, presenting more significant errors.

Another important aspect is the presence of significant deviations, with residuals of −46.86 m, −42.66 m, and −47.15 m, which are visible anomalies compared to other values. These deviations may result from specific errors in input data, such as piezometer data or complex geological features that were not represented with adequate precision.

3.2. Water Balance

One of the outputs generated by FREEWAT is the water balance of the modeled system. The water balance with average inputs and outputs, considering the period between 2004 and 2023, for the different aquifer formations modeled is shown in

Table 3. Water balance (10³ m³) from FREEWAT modeling of RMR formations (2004–2023).

Formations		Barreiras	Algodoais	Beberibe	Cabo	Alluvial Deposits	Coastal Deposit	Estiva
Input	Storage	1346.23	0.16	181.09	0.84	387.13	187.02	0.01
	Constant potential	1.74	0.00	0.02	0.00	0.15	6.84	0.00
	Well	0.00	0.00	0.00	0.00	0.00	0.00	0.00
	Drain	0.00	0.00	0.00	0.00	0.00	0.00	0.00
	Recharge	174.40	8.38	142.73	0.38	592.24	1050.09	0.26
	Sea	0.00	0.00	0.00	0.00	4.81	3.20	0.00
	Barreiras	0.00	0.04	51.41	0.00	355.69	140.92	0.00
	Algodoais	0.00	0.00	0.00	0.00	0.12	0.10	0.00
	Beberibe	4.17	0.00	0.00	0.00	59.26	0.86	0.00
	Cabo	0.00	0.03	0.00	0.00	6.97	7.46	0.00
	Alluvial Deposits	8.37	0.03	153.05	20.89	0.00	17.53	0.00
	Coastal Deposit	0.66	0.03	173.05	11.62	23.00	0.00	0.00
	Estiva	0.00	0.00	0.00	0.00	0.00	0.02	0.00
	Gramame	0.31	0.00	34.46	0.00	0.49	4.28	0.00
	Maria Farinha	0.00	0.00	1.09	0.00	0.00	0.13	0.00
	Mangrove Sediments	0.00	0.01	0.05	0.02	0.10	0.10	0.00
	Ipojuca Suite	0.00	0.00	0.00	0.01	0.08	0.04	0.00
	Total	1535.88	8.67	736.96	33.75	1430.04	1418.60	0.28
Output	Storage	631.87	8.28	440.01	2.43	1055.02	940.55	0.26
	Constant potential	6.50	0.00	0.00	0.00	4.81	4.03	0.00
	Well	116.32	0.00	214.13	16.85	110.21	217.74	0.00
	Drain	179.72	0.15	17.56	0.00	63.36	31.53	0.00
	Recharge	0.00	0.00	0.00	0.00	0.00	0.00	0.00
	Sea	1.26	0.00	0.00	0.00	0.15	6.66	0.00
	Barreiras	0.00	0.00	4.17	0.00	8.37	0.66	0.00
	Algodoais	0.04	0.00	0.00	0.03	0.03	0.03	0.00
	Beberibe	51.41	0.00	0.00	0.00	153.05	173.05	0.00
	Cabo	0.00	0.00	0.00	0.00	20.89	11.62	0.00
	Alluvial Deposits	355.69	0.12	59.26	6.97	0.00	23.00	0.00
	Coastal Deposit	140.92	0.10	0.86	7.46	17.53	0.00	0.02
	Estiva	0.00	0.00	0.00	0.00	0.00	0.00	0.00
	Gramame	47.60	0.00	0.54	0.00	1.01	6.13	0.00
	Maria Farinha	1.00	0.00	0.00	0.00	0.00	0.65	0.00
	Mangrove Sediments	0.15	0.01	0.40	0.01	0.02	0.06	0.00
	Ipojuca Suite	0.00	0.00	0.00	0.00	0.01	0.01	0.00
	Total	1532.47	8.67	736.95	33.75	1434.46	1415.73	0.28
	In–Out	3.41	0.01	0.01	0.00	−4.42	2.87	0.00
	Percentage of discrepancy	0.00	0.00	0.00	0.00	0.00	0.00	0.00
Formations		Gramame		Maria Farinha		Mangrove Sediments	Ipojuca Suite
Input	Storage	9.36		0.50		1.78	0.15
	Constant potential	0.00		0.00		0.00	0.00
	Well	0.00		0.00		0.00	0.00
	Drain	0.00		0.00		0.00	0.00
	Recharge	12.10		0.18		83.53	0.32
	Sea	0.00		0.00		0.00	0.00
	Barreiras	47.60		1.00		0.15	0.00
	Algodoais	0.00		0.00		0.01	0.00
	Beberibe	0.54		0.00		0.40	0.00
	Cabo	0.00		0.00		0.01	0.00
	Alluvial Deposits	1.01		0.00		0.02	0.01
	Coastal Deposit	6.13		0.65		0.06	0.01
	Estiva	0.00		0.00		0.00	0.00
	Gramame	0.00		0.21		0.07	0.00
	Maria Farinha	0.13		0.00		0.00	0.00
	Mangrove Sediments	0.02		0.01		0.00	0.00
	Ipojuca Suite	0.00		0.00		0.00	0.00
	Total	76.89		2.55		86.02	0.49
Output	Storage	36.19		1.20		75.05	0.31
	Constant potential	0.00		0.00		0.00	0.00
	Well	0.00		0.00		1.50	0.00
	Drain	0.87		0.00		9.17	0.04
	Recharge	0.00		0.00		0.00	0.00
	Sea	0.00		0.00		0.00	0.00
	Barreiras	0.31		0.00		0.00	0.00
	Algodoais	0.00		0.00		0.01	0.00
	Beberibe	34.46		1.09		0.05	0.00
	Cabo	0.00		0.00		0.02	0.01
	Alluvial Deposits	0.49		0.00		0.10	0.08
	Coastal Deposit	4.28		0.13		0.10	0.04
	Estiva	0.00		0.00		0.00	0.00
	Gramame	0.00		0.13		0.02	0.00
	Maria Farinha	0.21		0.00		0.01	0.00
	Mangrove Sediments	0.07		0.00		0.00	0.00
	Ipojuca Suite	0.00		0.00		0.00	0.00
	Total	76.89		2.55		86.02	0.49
	In–Out	0.00		0.00		0.00	0.01
	Percentage of discrepancy	0.00		0.00		0.00	0.00

. Main water inputs come from recharge, with some areas receiving contributions from external processes, such as the sea. Outputs are dominated by drainage from river networks.

The Barreiras and Beberibe Formations have a small positive balance, with inputs slightly higher than outputs, indicating a slight accumulation of water and a relatively balanced system. The discrepancy values are minimal, indicating a well-adjusted water balance. On the other hand, formations such as the Alluvial Deposits and Coastal Deposits present negative balances with more significant discrepancies. The water deficit of the Alluvial Deposits, with outflows exceeding inflows, may be due to factors such as low recharge and overexploitation.

The inflow of seawater has significant implications for aquifer formations. Formations such as the Cape, Alluvial Deposits, and Coastal Deposits have their water dynamics influenced by external contributions of saltwater. This inflow of seawater can lead to challenges for the water use of these aquifers, such as salinization, requiring greater control of exploration to ensure water quality.

Cary et al., 2015 [78] identified a local increase in the Cape groundwater salinity, with saltwater leaking from the superficial aquifer. The authors attributed the increase to downward flows favored by increased pumping. The observations of Chatton et al., 2016 [15], considering the long renewal times of the Cabo and Beberibe aquifer systems and the ongoing contamination and salinization, led the authors to conclude that the current exploitation of these aquifers is not sustainable.

The groundwater cycle is much slower than the pace of human development, making it essential to integrate this extremely slow cycle into water management concepts [15]. Therefore, tools such as groundwater flow models become relevant sources of information, such as the water balance.

3.3. Drawdowns

The graphs in Figure 12 show the average maximum and minimum values of drawdowns of the piezometric groundwater levels in each stress period. The spatial distribution of the drawdown in each stress period is shown in Figure 13 for layer 1, Figure 14 for layer 2, and Figure 15 for layer 3. The analysis of the drawdowns displays distinct dynamics between the existing formations in the three layers, which reflect the interaction between precipitation and groundwater levels at different depths and geological characteristics.

The most prominent intensity of drawdown in the most superficial layer (layer 1) shows a faster response to precipitation variations. In SP1, for example, with the highest precipitation (1721.44 mm), the maximum drawdown of layer 1 reaches 36.84 m, reflecting the rapid infiltration of water into the soil and the consequent impact on the shallowest layer. As precipitation decreases over time, maximum drawdown also decreases continuously, evidencing a direct response to reduced water recharge in the shallower layers of the aquifer.

Layer 2 follows a similar trend, although with a more moderate behavior. In SP1, the maximum drawdown in the layer reaches 124.44 m and gradually decreases in subsequent time steps. As precipitation decreases, the maximum drawdown also decreases, although in a more controlled manner compared to layer 1. This scenario suggests that formations represented in layer 2 have greater retention capacity and resilience to climate variations, reflecting the slower and more gradual response of this layer to water recharge and discharge. Although decreasing over time, the minimum drawdown remains more stable, indicating that layer 2 is less susceptible to rapid fluctuations than layer 1.

The formations represented in layer 3 exhibit more stable trends with minimal fluctuations over time. Despite experiencing significantly higher drawdowns, there is a marked reduction in the magnitude of the maximum drawdowns.

The piezometric level drawdown data analysis in the three aquifer layers from 2004 to 2023 (Figure 16) shows significant differences in the variations of each represented aquifer formation. In layer 1, the average drawdown between 2004 and 2023 was 10.21 m. The maximum drawdown value observed was 92.2 m, indicating areas of the aquifer that were significantly affected by low rainfall, reducing the value of aquifer recharge or even more intense exploitation.

The average drawdown of 63.9 m observed in the region modeled in layer 2 indicates greater exploitation of these aquifers, associated with lower values of effective recharge in these formations. In layer 3, the average drawdown of the area was 58.6 m. In some points, drawdowns greater than 100 m were observed, reflecting significant exploitation.

Due to the growing demand, especially since the drought of 1998, there has been a drop in piezometric levels in the RMR. In some neighborhoods of the municipality of Recife, such as Espinheiro, Torre, Madalena, and Boa Vista, the water level reaches 60 m deep [99].

In HIDRORECK II, Costa et al., 2002 [100] compared the evolution of water levels in the Beberibe and Cabo aquifers between 1988 and 1992 and from 1998 to 2002. They identified a generalized drop in water levels in both aquifers. Between 1998 and 2002, the levels varied between 25 and 40 m throughout the studied region. If we consider the period between 1998 and 2002, the drop increased by around 60 m in some regions. Costa and Costa Filho 2004 [101] reported a significant average drawdown of the Cabo aquifer. Based on data obtained from a monitoring program that began in 1975 in the Boa Viagem neighborhood, the drawdown values were 17 m between 1975 and 1985, 33 m between 1986 and 1995, 43 m between 1996 and 2000, and more than 90 m between 2001 and 2015.

Leitão et al., 2017 [80], in the HIDRORECK III study, made projections of the drawdown in the RMR based on modeling data from 2013 to 2015. Realistic, optimistic, and pessimistic scenarios were considered over 5, 12, and 20 years. The highest drawdown values were found in a Northern Zone, in the municipality of Abreu e Lima, with drawdowns that could reach 81.39 m in the next 20 years in the pessimistic situation.

The issue of declining groundwater levels in the aquifers of the Recife Metropolitan Region (RMR) is multifactorial. Excessive groundwater extraction, both for urban consumption and for industrial and agricultural use, has been the primary factor contributing to the drawdowns of these aquifers. A study conducted in the plain area of northern China reported a significant decline in groundwater levels, with average reductions ranging from 25 to 31 m. This decline was attributed to climate change and human activities, such as irrigation [102]. Factors such as the intense urbanization of the region also exacerbate the problem by reducing precipitation infiltration due to soil impermeabilization, thus hindering recharge and, consequently, the recovery of aquifer levels. Haq et al., 2021 [103] investigated the impact of urbanization on groundwater levels in Pakistan. Their study found that the urban area in the region increased by 37.89% between 1991 and 2017, while the groundwater level declined at a rate of 1.98 mm/year. According to the authors, if this trend continues, the groundwater level could reach approximately 160 m below the natural surface by the end of the century. The very formation of the aquifers in the RMR also contributes to the intensification of the drawdowns, as the most significant aquifers in the region are predominantly confined or semiconfined, making them more susceptible to intensive exploitation and climatic variations. The Beberibe aquifer, for example, being confined, experiences slower and more limited recharge, making it directly affected by the combination of urbanization and intensive exploitation.

Groundwater resources and their long-term recharge are regulated by long-term climatic conditions [104]. The reduction in rainfall and the increase in evapotranspiration, consequences of climate change, further decrease the recharge of these aquifers, impairing their recovery capacity, making them more vulnerable to overexploitation, and resulting in more pronounced drawdowns. In Australia, a study found that 80% of the observed groundwater level variation in 26% of the analyzed wells could be explained by climatic variability [105]. Similarly, the analysis of an Alluvial aquifer in Turkey, which is highly vulnerable to changes in precipitation patterns, revealed seasonal groundwater level fluctuations of up to 30 m [106]. Additionally, rising sea levels can lead to saltwater intrusion, compromising groundwater quality. According to Borba, Costa Filho, and Mascarenhas 2010 [107], overexploitation in the Recife Metropolitan Region (RMR) has already led to a significant decline in phreatic and piezometric levels, which may further intensify saltwater intrusion. In coastal aquifers such as the Boa Viagem aquifer, which is heavily exploited for urban and commercial use, excessive extraction reduces the freshwater pressure, facilitating the intrusion of seawater. This also occurs in the coastal region of the Beberibe aquifer, which, despite being deeper, experiences saltwater intrusion in its wells due to intensive extraction, especially in areas with geological faults or compromised recharge zones.

In this context, the development of effective strategies that reconcile urban growth and water security for sustainable groundwater management in the RMR is urgent. For this, increased control over groundwater extraction is essential. Makhlouf et al., 2024 [108] integrated MODFLOW simulations for future groundwater flow scenarios with machine learning models to estimate optimal extraction rates. Continuous monitoring of aquifer levels, along with stricter regulation and enforcement of well drilling and usage, is necessary to prevent excessive extraction that compromises natural recharge. Several studies explore the potential use of machine learning models based on remote sensing data to facilitate the monitoring and management of aquifer exploitation [96,108,109,110]. These measures are particularly important in coastal aquifers. Drawdown maps help visualize the most critical areas that require swift and effective intervention. In Figure 12, Figure 13, Figure 14 and Figure 15, it is evident that the most significant drawdowns occur in the northern portion of the study area, especially near the coast, affecting important aquifers like the Beberibe aquifer. The creation of hydraulic barriers through freshwater injection to prevent seawater intrusion into aquifers would also be an interesting strategy.

The implementation of green infrastructure, such as infiltration gardens and permeable pavements [111,112], and the protection of infiltration areas through reforestation and preservation of riparian forests, play a crucial role in mitigating the problem of reduced natural recharge caused by intensive urbanization.

Overall, integrated urban planning with water management is essential to mitigate the impacts of urbanization and exploitation on aquifers. Strengthening governance and enforcement, with the involvement of the Pernambuco Water and Climate Agency and the State Secretariat for Water Resources and Sanitation, is essential to ensure compliance with guidelines for the preservation and responsible exploitation of groundwater.

4. Conclusions

Groundwater modeling is a valuable tool for sustainable groundwater management, helping to define the dynamics of these waters and understand the impact of human intervention, especially in the current climate change scenario. However, groundwater modeling is challenging due to the scarcity of hydrological and hydrogeological measurements with adequate geographic and temporal spatiality. Therefore, integrating groundwater flow models with remote sensing data and GIS techniques enhances the achievement of modeling studies.

The results obtained from numerical modeling with remote sensing data showed the vulnerability of the RMR groundwater to marine intrusion, in addition to the significant drawdown of the aquifer level over the analyzed years, especially the Cabo and Beberibe aquifers. The correlation and determination coefficients obtained during calibration, despite indicating a good fit between observed and simulated data (r = 0.98 and R² = 0.97), suggest, through the relative error values (MARE = 42.96% and RRMSE = 23.96%), the presence of significant discrepancies in certain areas, potentially associated with limitations in the spatial distribution of calibration data and the heterogeneity of the aquifers.

However, gaps and inconsistencies in the hydrogeological data should be improved. Additionally, the numerical model should be updated with new piezometric level measurements to improve the spatial distribution of the measured information and, consequently, improve the calibration adjustment. Collecting more up-to-date data from wells using groundwater would also ensure that the model simulates reality more closely, leading to more accurate numerical representations.

Future research may deepen the subject through the exploration of innovative approaches, such as the use of machine learning and artificial intelligence techniques. The application of these resources could assist in addressing uncertainties, filling gaps in time series, and identifying patterns in large hydrogeological datasets. Furthermore, machine learning-based optimization algorithms could reduce computational costs and refine model parameters. The integration of remote sensing data and in situ measurements with these predictive models could bring significant advancements to numerical aquifer modeling studies. Incorporating water quality models into the flow model could also broaden the applicability of groundwater research findings, especially when climate change scenarios are essential for water resources management. A sensitivity analysis of the parameters could also be useful in identifying the influence of each parameter on the accuracy of the simulations. These multidisciplinary approaches have the potential to transform the understanding of aquifer dynamics, particularly in regions with data scarcity and complex systems.