Remote Sensing-Based Multiscale Analysis of Total and Groundwater Storage Dynamics over Semi-Arid North African Basins

Abstract

This study evaluates the use of remote sensing data to improve the understanding of groundwater resources in climate-sensitive regions with limited data availability and increasing agricultural water demands. The research focuses on estimating groundwater reserve dynamics in two major river basins in Morocco, characterized by significant local variability. The study employs data from Gravity Recovery and Climate Experiment satellite (GRACE) and ERA5-Land reanalysis. Two GRACE terrestrial water storage (TWS) products, CSR Mascon and JPL Mascon (RL06), were analyzed, along with auxiliary datasets generated from ERA5-Land, including precipitation, evapotranspiration, and surface runoff. The results show that both GRACE TWS products exhibit strong correlations with groundwater reserves, with correlation coefficients reaching up to 0.96 in the Oum Er-rbia River Basin and 0.95 in the Tensift River Basin (TRB). The root mean square errors (RMSE) were 0.99 cm and 0.88 cm, respectively. GRACE-derived groundwater storage (GWS) demonstrated a moderate correlation with observed groundwater levels in OERRB (R = 0.59, RMSE = 0.82), but a weaker correlation in TRB (R = 0.30, RMSE = 1.01). On the other hand, ERA5-Land-derived GWS showed a stronger correlation with groundwater levels in OERRB (R = 0.72, RMSE = 0.51) and a moderate correlation in TRB (R = 0.63, RMSE = 0.59). The findings suggest that ERA5-Land may provide more accurate assessments of groundwater storage anomalies, particularly in regions with significant local-scale variability in land and water use. High-resolution datasets like ERA5-land are, therefore, more recommended for addressing local-scale heterogeneity in regions with contrasted complexities in groundwater storage characteristics.

1. Introduction

Groundwater is the main driver of socio-economic development in semi-arid areas, where surface water resources are often limited [1]. In contexts where drought periods are frequent, groundwater is heavily extracted to maintain a constant or increasing agricultural production to meet the growing food demand and domestic water supply needs [2,3]. In addition, decreasing surface water supply will negatively impact groundwater recharge [4,5], known to occur through indirect processes in semi-arid areas (e.g., streamflow losses) [6,7,8,9]. Reduced recharge will increase the gap between inputs and withdrawals from aquifer systems leading to their depletion [1]. Such unsustainable water use is a serious threat for some countries where the agricultural sector plays a key role in the national economy [10].

The northwestern African region, more specifically Morocco, is expected to be a worldwide hotspot of climate change, where increased temperature and reduced and irregular precipitation will significantly reduce surface water supply [11,12,13,14]. Hence, understanding the hydrological system dynamics under the ongoing global environmental change and unsustainable use of water resources is as mandatory as it is urgent. However, efficient water resources management in data-scarce regions without accurate and timely data on water storage changes in reservoirs can be challenging. Furthermore, even when data are available, there may be significant gaps in the quality and discontinuities in both location and time of the in situ observations [15,16].

Considering today’s challenges, groundwater storage variability must be understood to enable sustainable water resource management. This insight becomes essential for shaping socio-economic development and ensuring environmental sustainability, particularly in countries such as Morocco where the economy is highly dependent on agricultural productivity [13,15,17]. Even though groundwater plays such a pivotal role in a variety of fields, its monitoring suffers from a serious discrepancy and limited spatial and temporal coverage. Among the major constraints is the sparse distribution of piezometers and monitoring wells [10,17], which results in a significant shortage of relevant and consistently recorded observations. However, leveraging remote sensing data obtained from multiple satellite sensors to complement traditional ground-based measurements is promising. Indeed, this integration provides a viable pathway for improving our understanding of groundwater dynamics, offering insights for effective groundwater resource management and planning strategies [18].

To cope with these challenges, the availability of different remote sensing data and the advance in computer memory allow the monitoring of water bodies in terms of surface and volume without using many in situ measurements. For example, the Gravity Recovery and Climate Experiment (GRACE) data provide an addition to local in situ measurements and might be a useful tool for monitoring large scale water supplies [19,20,21,22]. Few studies have used GRACE to investigate the temporal variations of terrestrial water storage (TWS) and groundwater storage (GWS) across Morocco [15,17,23]. Ahmed et al. (2021) used GRACE and GRCAE-FO to derive TWS and GWS in northern (403,678 km²) and southern Morocco (270,296 km²), and Ouatiki et al. (2022) used GRACE to investigate TWS and GWS in TRB and OERRB jointly. Even though they yielded important findings, these studies did not consider the significant differences in water availability and groundwater extraction, which may lead to the misinterpretation of the sustainable use of water resources in Morocco. This may be justified by GRACE’s coarse spatial resolution, however, it may yield further uncertainties in TWS and GWS in these semi-arid regions with significant local variabilities in water and land use [7,13]. To cope with this challenge, other higher resolution data may be associated with GRACE to capture local-scale variabilities in TWS and GWS. Recently, there has been growing interest in reanalysis datasets [24,25]. For example, the ERA-Interim [26], ERA5 [27], and ERA5-Land [28] reanalysis datasets, developed by the European Centre for Medium-Range Weather Forecasts (ECMWF), are widely considered to be state-of-the-art within various disciplines.

This study aims to present an overview of groundwater dynamics in a water-scarce region of Morocco, based on different data sources and resolutions. The investigation uses the GRACE and ERA5-Land products to analyze temporal changes in TWS and GWS in two major and contrasted river basins in Morocco. A comparative assessment of the two products is done through water balance analysis, encompassing variables such as precipitation, evapotranspiration, and runoff. The focus is then turned to determining the appropriate product accurately accounting for local spatial heterogenies of total and groundwater storage. Then, a faltering test is carried out in order to differentiate between the long-term trend and seasonality effect on both used products.

2. Material and Methods

2.1. Study Area

In this study, we focused on two of the major river basins of Morocco—the Oum Er-rbia River Basin (OERRB) and the Tensift River Basin (TRB) (Figure 1). These two neighboring basins are bordered from the east by the High-Atlas culminating at 4167 m.a.s.l and the Middle-Atlas Mountains. The basins also host two of the largest plains of the country, the Haouz and the Tadla plains, depending almost entirely on the Atlas Mountains for their water supply. In this section, we present a brief description of physiographic, climatic, and hydro(geo)logic characteristics of these basins, in addition to the main water resource-related issues.

2.1.1. Climatic and Hydrologic Settings

The OERRB is one of the largest river basins in Morocco, covering an area of around 38,000 km². It is situated between 31°15′N and 33°22′N latitudes, and 5°00′W and 9°20′W longitudes. The basin is characterized by a semi-arid climate with low precipitation (around 250 mm/year), heterogeneous over the space, with the lowlands receiving the least amount of rainfall, while the mountains receive the most [17,29]. In contrast, potential evapotranspiration (ETP) ranges from 2300 mm/year close to the mountains to 1600 mm/year along the coast [30]. The basin is mainly drained by the OER river originating from karst springs in the Middle Atlas to the east of the OERRB Basin. In addition, an important contribution to the OERRB flow comes from Oued El Abid and Oued Lakhdar, two of its tributaries originating from the High-Atlas to the west of the OERRB. The OERRB flows along 550 km until it reaches the outflow in the Atlantic Ocean near El Jadida city. The OERRB held 19% of the surface water potential in the country, with nearly 3410 mm³ [30]. This important surface water supply is conditioned by 15 dams with a total capacity of 5300 mm³ (33% of the national capacity).

The TRB is located in central Morocco, covering an area of around 30,000 km², between latitudes 30.75° and 32.40 N, and longitudes 7.05° and 9.9 W. The TRB has a semi-arid climate, tending to be sub-humid in the High-Atlas [31]. The annual potential evapotranspiration (ETP) is about 1600 mm/year [32,33,34]. The surface water supplies in the TRB come mainly from the streams draining the northern flank of the High Atlas. All the streams in the TRB are ephemeral, with no perennial surface water source. Total surface water volumes in the TRB are estimated to be 800 mm³/year [35], significantly lower than OERRB.

2.1.2. Hydrological Settings

The study area hosts five main aquifer systems constituting the main water supply for socio-economic activities. Three main aquifer systems extending from the High- and Middle-Atlas Mountain-fronts to the Atlantic Ocean dominate the OERRB: the Tadla aquifer system, the Bahira aquifer, and the coastal Sahel-Doukkala. The TRB Basin contains two main aquifer systems: the Haouz-Mejjate aquifer system and the Meskala-Kourimate aquifer system.

The multilayered Tadla aquifer system is situated in the Tadla plain syncline, with geological formations spanning from the Palaeozoic to the Quaternary periods. It comprises four aquifers: (i) the Mio-Plio-Quaternary, (ii) the Eocene, (iii) the Senonian, and (iv) the Turonian, with the Turonian being the most productive (up to 1300 L/s). The unconfined alluvial Mio-Plio-Quaternary aquifer, divided into Beni Amir (600 km²), Beni Moussa (885 km²), and Tassaout (500 km²) units. The sandy limestone Eocene aquifer reaches a thickness of 100 m and extends over 6400 km². The marly limestone Senonian aquifer extends over 9100 km² and varies in thickness from 40 to 60 m in the north, and to over 200 m near the Atlas Mountains. The Turonian aquifer, formed by intensive karst limestone and dolomite, transitions from being limestone-dominated in the north to dolomite in the south, underlain by Cenomanian marl, with thicknesses ranging from 50 to 100 m over 10,000 km². In the north, at the phosphate plateau, the Eocene, Senonian, and Turonian aquifers are unconfined but become confined toward the Atlas Mountains. The Tadla aquifer system is extensively used for irrigation and drinking water supply, with well depths ranging from a few meters to over 1000 m. Groundwater generally flows from the northeast and northwest to the west in the deep aquifers, and from the northeast to the southwest in the Mio-Plio-Quaternary aquifer along the Oum Er-Rbia River [36]. Vertical hydraulic exchanges occur through low permeability layers or sub-vertical faults.

More to the south of the OERRB, the Bahira aquifer system, partly shared with the TRB, is formed of multiple aquifers from the Palaeozoic to the Plio-Quaternary, extending over 5000 km². Discontinuous groundwater flow exists in the fractured ancient schist basement and the fractured Lutetian limestones as outcropping over the Gantour plateau, and is overlain by Plio-Quaternary alluvial deposits in the plain. The latter form a quite heterogeneous alluvial aquifer over the Bahira plain, which consists of a clay complex with interbedded gravel and crushed stone. Surface water infiltration, particularly from nearby mountains, is the primary source of natural recharge for the western and central Bahira aquifer. Within the plain, recharge occurs through the direct infiltration of rainwater. Groundwater flows from the Ganntour, Jbilet, and Rehamna recharge areas towards the closed depressions of Sed Elmejnoun and Zima Lake, where significant evaporation occurs [37]. Overexploitation of groundwater happens mainly during the dry season due to salt extraction at Lake Zima and the irrigation of crops around Sed Elmejnoun [37].

The Doukkala-Sahel aquifer system, located downstream of the OERRB on the Atlantic coastline, consists of two deep aquifers from the Lower Cretaceous and Upper Jurassic, and a surficial Plio-Quaternary aquifer extending over 6350 km². The Upper Jurassic and Lower Cretaceous formations are generally formed of thick karst limestones. The deep aquifers contained in these formations are poorly known and less exploited due to their high depth below the surface, and poor water quality because of the dissolution of gypsum [38]. Groundwater flows from East to West and discharges into the Atlantic Ocean [38]. The Plio-Quaternary deposits are more permeable and constitute the main groundwater supply for the region. The geological nature and texture of the Plio-Quaternary formations (marine and dune calcareous sandstone) give them the characteristics of an aquifer with interstitial and micro-fissure permeability. The Plio-Quaternary is likely to be karstified. Surface and deep karstic forms can significantly improve its permeability. The Plio-Quaternary is the most accessible aquifer in the studied region and is tapped by most of the wells dug by the inhabitants of Doukkala [38]. Groundwater in the Sahel-Doukkala region is mainly recharged by streamflow losses within streambeds and direct infiltration of rainwater, and discharges through pumping and to the Atlantic Ocean through diffuse and submarine springs [39]. The Haouz aquifer system in the TRB is formed of a large unconfined Plio-quaternary aquifer and several confined deep aquifers [40]. The deep aquifers are poorly known in the central part of the Haouz plain because of their high depth and limited extension [7]. However, these deep aquifers composed of fractured carbonates from Upper Jurassic, Cenomanian-Turonian, and Eocene are important in the western part of the Haouz-Mejjate area, and important springs emerge from these aquifers [13]. The unconfined Haouz aquifer is the most important alluvial aquifer in Morocco, extending over the Houz plain for 6000 km² [41]. Groundwater in the Tensift Basin is the sole source of drinking water for rural populations and some urban agglomerations. It also constitutes the main supplier for irrigated agriculture, the main socio-economic activity in the basin, with more than 80%, particularly when surface water is not readily available [42]. Hence, the Haouz aquifer is undergoing a severe groundwater depletion due to excessive pumping [7,13].

The Meskala-Kourimate aquifer system is located in the Essaouira Basin, and the coastal area is formed by a group of independent but very similar hydrogeological systems, which correspond to synclinal basins. Generally, there are two main aquifers, the Plio-Quaternary phreatic aquifer and the Cenomanian-Turonian carbonate aquifer. The Plio-Quaternary phreatic aquifer consists of conglomerates, alluviums, colluvium, and sandstone matrix [43]. Its substratum comprises Senonian gray marls, and it can directly contact Triassic and Cretaceous formations [44]. The Cenomanian-Turonian carbonate aquifer is primarily composed of limestone and dolomitic–limestone layers. The Cenomanian, notable for its thickness exceeding 220 m, consists of gray marls, sandy marls, gypsiferous marls, limestones, sandy limestones, and dolomitic limestones [45]. The Turonian, at the surface, features micritic limestones and dolomites rich in silica, with marly bed intercalations at depth [46]. It exhibits lateral facies variations, though less pronounced than those of the Cenomanian. Groundwater primarily flows from southeast to northwest upstream, and from east to west downstream, discharging naturally into the Atlantic Ocean.

2.1.3. Land and Water Use

Agriculture is the main socio-economic activity in the OERRB, with 464,530 ha of irrigated land and 90% of water demand [47]. Citrus fruits, olives, wheat, and barley are the principal crops farmed in this basin [48]. Currently, the water demand largely exceeds the available volumes due to the growing population and increasing agricultural production [30]. As a basin depending more on surface water, the most important challenge for water resources is the impacts of climate change such as the reduction in precipitation and frequent droughts [49]. Groundwater is also facing rapid depletion due to overexploitation resulting in dropping groundwater levels and drying springs [30].

Groundwater resources in the Tensift Basin are subject to intensive pumping, leading to their depletion [7,13]. This overexploitation is due to the extension of modern groundwater-based irrigation such as olive trees and other water-consumptive crops promoted by the government through large funding programs [13]. In addition, frequent drought during the recent decades, and decreasing surface water resources, are putting more pressure on groundwater and reduce groundwater recharge [13,50].

2.2. Data Used

2.2.1. GRACE

The GRACE satellites are a pair of orbital missions designed to measure Earth’s gravity field and its variations. They use various sensors and instruments to provide data on Earth’s gravity field [51], which has been publicly available since 17 March 2002. These data enable hydrologists to estimate terrestrial water storage (TWS) changes on various regional and global scales [52]. This study uses two TWS solutions, the CSR Mascon and the JPL Mascon (RL06), both based on the “Mascon” principle, which divides Earth’s surface into small boxes to estimate mass changes within each box over time [53,54] (Figure 2). We note that the JPL and CSR datasets have their original native resolution of roughly 3 × 3 degrees, and they come from the same GRACE data. Moreover, the CSR dataset and the JPL 0.5 × 0.5-degree dataset are downscaled using comparable techniques. Despite their similarities, the CSR and JPL models have a few differences.

The CSR Mascon model was developed by the Center for Space Research (CSR) at the University of Texas at Austin. This model uses a slightly different method for estimating the mass changes within each Mascon, which involves using a combination of satellite data and hydrological models to estimate changes in surface water, soil moisture, and groundwater storage. The CSR Mascon model also includes additional post-processing steps to improve the accuracy of the estimated mass changes. The CSR Mascon depends on spherical harmonic coefficients and is accessible in 1° × 1° resolution, although it is also available at 0.25° resolution [55]. The JPL Mascon (RL06) uses a different type of gravitational field basis function, with prior limitations in space and time to lessen the impact of uncertainties [56]. The JPL Mascon model was developed by the Jet Propulsion Laboratory (JPL) at the California Institute of Technology. This model also uses a Mascon approach, but it uses a different method for estimating the mass changes within each Mascon. The JPL Mascon model is based purely on a data-driven approach, which involves using the GRACE data directly, without the need for additional information from hydrological models or other sources.

2.2.2. ERA5-Land

ERA5-Land is a high-resolution dataset of land surface variables produced by the European Centre for Medium-Range Weather Forecasts (ECMWF) [28]. The dataset is part of the ERA reanalysis project, which is a global dataset that provides a consistent and accurate representation of Earth’s climate and weather conditions. ERA5-Land includes a wide range of variables related to the land surface, such as precipitation, runoff, evapotranspiration, soil moisture, and vegetation cover. These variables are derived from a combination of observations and model-based simulations, and are available at a spatial resolution of 0.1 degrees (approximately 10 km × 10 km) and an hourly temporal resolution. The dataset covers the period from 1950 to the present, and is updated regularly with new data. It is widely used by researchers and policymakers for various applications, including climate modeling, weather forecasting, and land surface hydrology. Figure 3 shows the spatial variability of the main water budget components derived from ERA5-Land for the studied area.

The graph in Figure 4 displays the monthly time series of three essential water budget elements—precipitation, evapotranspiration, and runoff—for two basins: the TRB (top panel) and the OERRB (bottom panel) from January 2002 to December 2019. The graphic depicts the seasonal and interannual variations in these water budget components. For example, it demonstrates how precipitation peaks coincide with increases in runoff, showing possible groundwater recharge based on surface water availability. Similarly, evapotranspiration maxima frequently coincide with dry periods, indicating the region’s high evaporation rates owing to meteorological circumstances. We also conclude that, over the two basins, principal components are P and ET, while Runoff is low compared to other water balance components. A consistent variation is observed where components are responding to the water supply adequately.

By combining these three components, the image sheds light on the total water balance dynamics of the two basins, stressing the significance of climatic variability and its influence on water availability and management. Understanding these processes is critical for successful water resource management, especially in areas prone to drought and water shortages, such as the TRB and OERRB.

2.2.3. Groundwater Data

In this study, monthly groundwater-level records were obtained from the Hydraulic Basin Agencies (ABH) for both TRB and OERRB (

Table 1. Number of wells and the time period of the in situ groundwater measurements of the TRB and OERRB.

Number of Wells	Basin/Aquifer	Time Period	Origin (Reference)
11	TRB/Haouz aquifer	2002–2021	Hydraulic Basin Agencies of Tensift (ABHT)
40	OERRB/Oum Er-rabia aquifer	2002–2017	Hydraulic Basin Agencies of Oum Er-rbia (ABHO)

). In order to have the best time series overlap, a missing data check was performed. In TRB, out of 15 shallow piezometers, we selected 11 with the best continuous monthly series between 2002 and 2021 in the Haouz alluvial aquifer, and 40 in OERRB to present the validation data on the northern border of the High-Atlas Mountains. The selection process concentrated on piezometers that had the highest number of continuous monthly records, reducing data gaps to provide a dependable time series for the research period. In order to provide appropriate temporal overlap with GRACE data, we gave priority to piezometers that had the largest continuous spans during this timeframe or spanned the full research period. We also evaluated the quality of the data by looking for abnormalities or irregularities, and choosing those piezometers that produced accurate and consistent results. These monthly records were averaged over the basin scale and used to validate the groundwater spatiotemporal evolution derived from GRACE and ERA5-Land water balance. The data were standardized by their mean, and standard deviation was computed for the full timeframe, to represent the groundwater-level anomalies at the basin scale. This approach was selected to guarantee uniformity over the whole dataset and to precisely depict the data’s long-term patterns. To convert groundwater level to groundwater storage, a standardization procedure was adopted, given that aquifer characteristics (storativity) were unavailable on the study site. The standardization procedure we used involved transforming each groundwater level from observation wells to obtain a mean of zero and a standard deviation of one. This was done to allow for the direct comparison of the temporal patterns across different datasets, despite their differing units and scales. Specifically, we standardized each time series by subtracting the mean value of the dataset and dividing it by the standard deviation. Then, standardized data were used to validate the estimated GRACE groundwater storage anomalies.

2.3. Methods

Typically, remote sensing sensors capture a reflected signal from the surface that is often disrupted by noise, causing perturbations in satellite data. This interference arises from various factors such as obstructed or weakened signals due to clouds, aerosols in the atmosphere, fluctuating lighting conditions, and the angle at which the satellite observes the surface at any given time. To enable the effective analysis of time series data by minimizing noise, data processing becomes imperative. Thus, temporal data smoothing serves as a preliminary step before delving into time series analysis. In our specific case, the Savitzky–Golay filter (or least-squares smoothing filter) was employed on raw time series data to generate a clearer and more representative signal [57].

Following the smoothing of the collected data, two methodologies were adopted. These included the closure of the water balance budget, used for assessing TWS at a high resolution (~10 km), employing ERA5-Land data. We used ERA5-Land data to estimate different components of the water balance at a high resolution to assess Total Water Storage (TWS) changes over the study area. These TWS changes were then compared with the GRACE-derived TWS anomalies. The purpose of this comparison was to evaluate the ability of ERA5-Land data to capture the same GRACE TWS variations at a finer spatial resolution. Additionally, the residual approach was employed to evaluate groundwater level using both ERA5-Land and GRACE data.

2.3.1. Water Balance Closure for Water Storage Changes

Water balance closure is a fundamental concept stipulating that the total volume of water entering a system should equate to the volume leaving the system, encompassing any alterations in water storage within that system. This principle serves as a cornerstone in hydrology and water resource management, offering insights into the various facets of the water cycle within a defined region. Among the critical aspects of the water balance is the assessment of water storage changes, denoting fluctuations in water volumes stored within a specific system, such as groundwater, surface water bodies, or soil moisture.

In our study, we leveraged data from the atmospheric reanalysis ERA5-Land, encompassing precipitation, evapotranspiration, surface runoff, soil moisture, and snow water equivalent. These variables helped us to understand the dynamics of the water balance within the studied basins. The terrestrial water balance equation on a regional scale enabled the representation of TWS changes in selected river basins or areas, as outlined below:

$${\displaystyle \frac{\partial \mathrm{T}\mathrm{W}\mathrm{S}}{\partial \mathrm{t}}}=\mathrm{ }\mathrm{P}-\mathrm{E}\mathrm{T}-\mathrm{Q}$$

(1)

where, P is precipitation, ET is evapotranspiration, Q is surface runoff, and all are expressed in mm/month. t represents a given month. The term ∂TWS/∂t represents the change in total water storage over time, which is calculated from ERA5-Land data using the central difference formula to obtain monthly differences in the anomalies of total water storage. This calculation is designed to be comparable to the TWS anomalies observed by GRACE, not to imply that these values are directly obtained from GRACE data. We clarify that our computed TWS changes are aligned with the GRACE observations to enable a meaningful comparison of the TWS dynamics across different datasets. The central difference formula is described below in Equation (2):

$${\displaystyle \frac{\mathrm{d}\mathrm{T}\mathrm{W}\mathrm{S}}{\mathrm{d}\mathrm{t}}}={\displaystyle \frac{\left(∆\mathrm{T}\mathrm{W}{\mathrm{S}}_{\mathrm{t}+1}-∆\mathrm{T}\mathrm{W}{\mathrm{S}}_{\mathrm{t}-1}\right)}{2}}$$

(2)

2.3.2. Residual Approach

The residual method involves subtracting surface water and soil moisture components from GRACE-resulted total water storage to estimate changes in groundwater storage. By leveraging satellite-derived information, changes in surface water and soil moisture storage can be estimated and subsequently subtracted from the overall water storage signal recorded by GRACE. This approach offers distinct advantages over other methods by eliminating the need for assumptions regarding aquifer properties or water movement rates within the aquifer, which can be challenging to estimate accurately. The residual approach equation has been used by several researchers [58,59] and is represented as follows:

$$∆\mathrm{G}\mathrm{W}\mathrm{S}=∆\mathrm{T}\mathrm{W}\mathrm{S}-∆\mathrm{S}\mathrm{M}-∆\mathrm{S}\mathrm{W}\mathrm{E}-∆\mathrm{S}\mathrm{R}$$

(3)

where ΔGWS is the change in ground water storage, ΔTWS is the change in total water storage, ΔSM is the change in soil moisture, ΔSWE is the change in snow water, and ΔSR is the change in surface water. In Equation (3), the canopy water storage (CWS) component is omitted, based on the distinctive attributes of the semi-arid location in which this study was conducted. This semi-arid region has little vegetation, which means that the amount of water stored in the canopy is minimal compared to other components. Also, the impact of CWS on long-term TWS changes is minimal compared to other components. This assertion has been verified by prior research (e.g., [17,20,58,60]) in which CWS was omitted in comparable situations due to its negligible influence on TWS in such settings.

By combining Equations (1) and (3), we were able to derive ΔGWS using ERA5 data. Then, GWS was inverted from the central difference formula as the one described in Equation (2). The GWS was initiated with the first value of GWS derived from GRACE (Figure 5).

2.3.3. Groundwater Storage’s Trends and Seasonality

To discern groundwater trends and seasonality, we applied the Hodrick–Prescott (HP) filter [61] to both in situ and GRACE Groundwater Storage (GWS) data. The HP filter, a statistical tool introduced by economists, is used to decompose time series into trends and cycles. By employing the HP filter on standardized GWS data, we aimed to differentiate the long-term trend associated with changes in groundwater mass distribution from the short-term fluctuations linked to seasonal variations and other periodic events. This approach facilitated a more comprehensive analysis of the underlying trends in groundwater depletion processes.

The optimal selection of the smoothing parameter is crucial for correctly applying the HP Filter, with its value, denoted as λ, defining the trade-off between trend matching and smoothing. The smoothing parameter λ was set to 129,600 for monthly time series, as suggested by [62], to strike a balance between trend accuracy and effective smoothing, selected through cross-validation or other optimization techniques. A flowchart showing the main datasets and processing steps is shown in Figure 5.

3. Results and Discussion

3.1. GRACE and ERA5-Land-Derived Water Storage: Temporal Variability and Comparison

The comparison between two GRACE products (JPL and CSR-Mascon) was demonstrated on a monthly timescale from 2002 to 2020 (Figure 6). In general, both the CSR Mascon and the JPL Mascon models provide similar estimates of changes in water storage over time, but there are some differences between the two models, particularly in periods where there are large changes in water storage. A Pearson correlation coefficient (R) up to 0.96, with a root mean square error (RMSE) of 0.99 cm (mean bias error (MBE) = −0.72 cm), was observed in OERRB, and 0.95 with an RMSE of 0.88 cm (MBE = −0.60 cm) in TRB. From these results, the two products show a similar trend and variation over the selected basins.

Note that we were able to capture the whole range of variability, including both short-term swings and longer-term trends, by comparing the two datasets prior to eliminating the seasonal cycle. Analyzing the products in their original, without form adjustment, gave a thorough knowledge of how they performed over multiple time periods, ensuring that the chosen product appropriately reflected the region’s natural hydrological dynamics. Following this first evaluation, we removed the seasonal cycle to concentrate on long-term trends and interannual variability, which are crucial for understanding how groundwater storage varies over time.

From Figure 7, the change in TWS estimates derived from the ERA5-Land reanalysis product (Equation (1)) fit the temporal variations in the GRACE data. Overall, when comparing TWS changes from 111 km-GRACE-JPL to 10 km-ERA5-Land, Pearson correlation coefficients were 0.79 (RMSE = 0.01) and 0.87 (RMSE = 0.02) in the TRB and OERRB, respectively. However, the visual inspection of Figure 7 showed that TWS variations derived from ERA5-Land were larger than GRACE-derived ones during wet periods, especially in the OERRB. This discrepancy may be due to the ERA5-Land being capable of capturing local variabilities because of its high spatial resolution. Additionally, the differences can be explained by the leakage effect inherent in GRACE data, where signals from surrounding areas can ‘leak’ into the target area, thereby smoothing out spatial variations. Surface water in both of these semi-arid basins is mainly derived from the mountain front and the main irrigated plains for agricultural irrigation purposes [7,13], which results in a highly heterogeneous spatial distribution of surface water that is difficult to capture with high-resolution products. These fluctuations are less marked in the TRB Basin, where surface water is less available, as can be seen in the difference between the streamflow magnitude of both basins (Figure 4). This indicates that ERA5-Land products may be more suitable for total water storage estimation in such areas where a strong spatial zonation of water resources exists.

3.2. GRACE and ERA5-Land Groundwater Storage versus Groundwater Levels

When it comes to comparing groundwater storage anomalies derived from GRACE satellite data and estimated from ERA5-Land to ground measurements, one common approach is to calculate the Z-score [17,23]. The Z-score is adopted to standardize the time series data from all datasets, including GRACE, ERA5-Land, and the in situ groundwater level observations by subtracting the mean and dividing by the standard deviation for each dataset. This approach is consistent with the approach used in [17,63]. Figure 8 depicts the comparison between the standardized GRACE and ERA5-Land-based estimations, and ground-based measurements (GWL), over the OERRB and TRB Basins. The ground-based measurements curve represents the mean value of the standardized groundwater levels from all observation wells used within each river basin. The GRACE-derived GWS has a moderate correlation with groundwater levels in the OERRB (R = 0.59, RMSE = 0.82) and a weak correlation in TRB (R = 0.30, RMSE = 1.01). The results for OERRB are similar to other studies for basins with a similar spatial extent [64,65].

ERA5-land-derived GWS is well correlated with groundwater levels in the OERRB (R = 0.72, RMSE = 0.51) and moderately correlated with groundwater levels in TRB (R = 0.63, RMSE = 0.59). ERA5-land results seem to outperform GRACE products in estimating groundwater storage anomalies. This is due to the difference in spatial resolution between both products. Low-resolution GRACE products (~111 km) may be influenced by the signal of nearby regions, and this could generate errors when compared to local piezometric measurements [66,67]. In addition, the obtained results may be further influenced by seasonality and spatial heterogeneity characterizing groundwater recharge in semi-arid areas, often biased towards the wet season [7,8,15,68]. Bouimouass et al [7] showed that groundwater recharge in the Haouz plain, in the TRB, is mainly sourced from High-Atlas surface water through streambeds, as well as from flood-irrigated areas. They also showed that recharge is highly localized in the mountain front [7,8,68,69]. In contrast, downstream areas rely more on groundwater for irrigation, and severe decline in groundwater levels were observed there, especially during the past two decades [13]. The thickening of the unsaturated zone causes lag time in net groundwater recharge [70], and may further complicate the interpretation of results from remote sensing datasets.

Results from both GRACE and ERA5-Land allowed the determination of some small periods of groundwater replenishment, especially during the exceptionally wet hydrological years. The first replenishment occurred in 2008/2009, corresponding to an exceptionally wet period [13]. Afterwards, groundwater levels started decreasing until the year 2014, when high magnitude floods were triggered by intense rainfall in the High-Atlas and replenished aquifers [8,71]. Another replenishment occurred in the year 2016. Despite the occurrence of these short periods of replenishment, the resulting time series of groundwater storage changes, and both basins have undergone an overall persistent decrease since 2002. Similar observations were made by [17]. However, according to [17], these changes follow the temporal distribution and seasonality of precipitation, with positive trends mostly occurring during the wet period between November and April.

The high spatial variability in groundwater recharge and depletion in the TRB influences the accuracy of estimating both TWS and GWS, often lower than OERRB, by both GRACE and ERA5-Land. However, the correlation between GWS against groundwater levels in TRB increases significantly from GRACE (R = 0.3) to ERA5-Land (R = 0.63), thus attesting to the efficiency of the latter in capturing local variabilities because of its high spatial resolution. Even though management strategies often require regional-scale groundwater estimation [72], important changes can occur on the local scale and need to be considered for better water resource management. Hence, the use of high-resolution datasets such as ERA5-land may be more suitable for dealing with local-scale heterogeneities in land and water use.

The analysis of groundwater dynamics in Moroccan basins provides insight into the challenges of monitoring groundwater resources, especially in semi-arid regions. The results acquired are in accordance with many research papers, including by Ouatiki et al. [17], who used GRACE to analyze the impact of climate variability on groundwater depletion in Morocco, highlighting the importance of understanding natural recharge events. The study identified intermittent depletion episodes, which were masked by natural recharge occurring in wet years (i.e., 2009), emerging as a critical factor influencing groundwater storage, thus pointing to the complex interaction between surface water and groundwater resources. Ferreira et al. [73], who evaluated hydrological regimes using remotely sensed and modeled data, underscored the need to integrate multiple data sources to improve the accuracy of groundwater storage estimates. Furthermore, they demonstrated the effectiveness of combining GRACE data with hydrological models to enhance water balance estimation. On the other hand, recent studies by Bouimouass et al. [7,74] highlighted the significance of local groundwater recharge process sources, particularly heterogeneous regions, such as Moroccan semi-arid mountain front areas. The findings of the current study are in line with the existing literature, particularly in recognizing the challenges and limitations of using GRACE in semi-arid regions, while highlighting the potential of high-resolution reanalysis datasets like ERA5-Land to improve groundwater monitoring and management in such complex and heterogenous areas.

3.3. Annual Trends and Seasonality in Groundwater Storage

The previous section’s findings prompt an exploration into whether the apparent alignment between GRACE and in situ groundwater storage anomalies merely stems from seasonal and yearly cycles. Consequently, the correlation is recalculated for each subcomponent once the total signal is disentangled into trend, seasonal, and residual patterns (de-seasonalized). From Figure 9a, a non-linear trend spanning from 2002 to 2020 is evident for both in situ and GRACE data over the OERRB. Notably, both positive and negative tendencies are observed within this timeframe, with GRACE tending to overestimate the trend. In Figure 9b, an encouraging alignment between in situ and GRACE seasonal components is depicted, achieving a correlation of 0.67. The peak occurrences, both positive and negative, manifest during winter/spring and summer/autumn, respectively.

Figure 9c illustrates the de-seasonalized signals that exhibit periodic patterns recurring over 12 months. These patterns are invaluable for identifying prolonged periods of excessive wetness or aridity spanning beyond a single hydrological year. Notably, significant drought episodes in the OERRB Basin during the periods 2008–2009 and 2015–2016 are clearly discernible. Similarly, observable anomalies during wet years, such as 2009–2010, stand out prominently. These extreme events are observable in both GRACE data and in in situ measurements across the study area.

A notable correlation coefficient of 0.57 was observed between ground-based measurements and the GRACE signal. These findings collectively indicate that the substantial alignment between GRACE and ground measurements transcends a reliance solely on annual cycles and seasonality. According to [62], extended-term deficiencies and periodic lengthy cycles are intertwined with low-frequency oscillations in precipitation levels, interconnected with climate patterns like the North Atlantic Oscillation and the East Atlantic pattern.

Furthermore, comparing residual components reveals intriguing insights into anthropogenic effects on groundwater. For instance, the negative peak observed during 2008–2009 is likely associated with a drought that affected the OERRB region. This drought-induced abstraction led to the overexploitation of groundwater for irrigation, consequently reducing water levels in dams [75]. Such anthropogenic impacts on groundwater dynamics underscore the importance of accounting for human activities when analyzing water resource fluctuations.

Certain interannual components are still present in the time series depicted in Figure 9b. This is most likely because the HP filter is sensitive to the smoothing parameter, which influences the degree of separation between the trend and cyclical components. In this situation, the filter may have maintained certain interannual fluctuations that do not perfectly adhere to seasonal cycles due to the intrinsic unpredictability in hydrological processes, such as multi-year droughts, abnormal precipitation occurrences, or changes in groundwater recharge and consumption. While the HP filter is reliable for distinguishing trends from cycles, it may not eliminate interannual variability, especially in complicated semi-arid settings.

Figure 10 presents the annual trend, seasonal cycle, and de-seasonalized components of standardized GWS, respectively, for the Tensift (TRB) Basin.

The GRACE time series trend exhibits a significant decrease of standardized groundwater over the entire period of study. The in situ trend indicates a decrease but it is not as steep as that of GRACE. The ERA5-Land trend declines first and then there is a slight increase at the end (Figure 10a). Different trends depict the discrepancies between satellite-based observations, in situ measurements, and reanalysis data. In fact, the decline trend is consistent in both GRACE and in situ data, which should indicate a reliable signal of groundwater depletion in the basin. This offset in the ERA5-Land trend may be caused by differences in the assimilation techniques and model representations.

In Figure 10b, all datasets show the seasonal variations, though the GRACE time series shows more pronounced fluctuations. In turn, in situ data moderately coincides with GRACE, while ERA5-Land is smoother in terms of seasonal variations. The correlation strength between in situ GWS and other datasets is demonstrated by Pearson’s correlation coefficients, according to which the GRACE indicators are 0.17 and ERA5-Land indicators are 0.38.

The seasonal cycle corresponds to the periodic variations in standardized groundwater, but, in turn, it might be controlled by climatic and hydrological variables, including precipitation and evapotranspiration. A much lower correlation (0.17) between GRACE and in situ data, respectively, indicates a few discrepancies present between the fields, perhaps caused by differences in spatial resolution and measurement uncertainties. A moderate correlation, 0.38, is obtained with ERA5-Land, which thus reflects decent agreement in the temporal evolution of seasonality but indicates some differences.

As clearly seen, the GWL anomaly over the GRACE data is highly variable. The in situ and ERA5-Land data would support even smoother variations than GRACE, considering the following: both its Pearson correlation coefficients for de-seasonalized components are higher. This implies that the de-seasonalization has withdrawn long-term trends and variability from the seasonal noise. The increased correlations of 0.39 and 0.56 suggest that both GRACE and ERA5-Land data somewhat reasonably capture long-term changes noted by in situ data.

This figure serves as a good example of how GRACE satellite data can be used in combination with in situ measurements and ERA5-Land reanalysis data to monitor changes in groundwater storage over the Tensift Basin. The analysis shows a systematic decrease in groundwater storage, and the significant seasonality in the result is captured in the long-term changes by these different datasets. The different correlation coefficients discussed reflect the strengths and weaknesses of each dataset, an assertion that indicates the reason that data must be integrated into more than one data source when monitoring groundwater comprehensively.

Overall, higher correlations between in situ and both GRACE and ERA5-Land data for the seasonal cycle and de-seasonalized components are observed in the Oum Er Rbia Basin. This means that the hydrological behavior may be more consistent within the Oum Er Rbia Basin than within the Tensift Basin. Both basins show a decreasing trend in GWS, but with differences in the magnitude and rate of decrease. The seasonal variation in the Oum Er Rbia Basin shows a better agreement between GRACE, in situ, and ERA5-Land data compared to that of the Tensift Basin. This might be because of the more predictable or stable seasonal patterns in precipitation and recharge in the Oum Er Rbia Basin. The ground resolution of GRACE data can thus capture the signals of hydrology much better when the groundwater storage changes in the basin are more spatially homogeneous. Differences in hydrological and climatic conditions in the two basins can be expected. For example, the Tensift Basin may be more affected by complex interactions between surface and groundwater; many land use changes and human activities, such as irrigation and urbanization, could lead to more varied standardized GWL anomaly changes. Naturally, there are differences in geological features between basins that could influence the flow and storage of groundwater. Differences in aquifer properties, soil types, and geological formations would cause different responses to climatic inputs and human activities.

In analyzing Figure 10b,c, a lag was noticed between ERA5 and in situ data. Therefore, we performed a cross-correlation analysis to determine the time lag between the standardized ERA5-Land ΔGWL and in situ groundwater-level changes (Figure 11).

The findings, which are displayed in Figure 11, suggest that there is a lag of around three months for both the cyclical and de-cyclical components between changes in in situ groundwater level and ERA5-Land. The Pearson correlation coefficients, after accounting for this lag, were 0.72 for the cyclical data and 0.64 for the de-cyclical data, indicating a substantial connection between the datasets. The reason for this delay might be due to variations in the datasets’ temporal and spatial resolution, or the groundwater levels’ delayed reaction to the climatic factors depicted in the ERA5-Land data. The discrepancy in the time lag between ERA5-Land ΔGWL and in situ groundwater-level changes in the TRB, as opposed to the OERRB, can be attributed to many variables. Both basins are semi-arid and depend on indirect groundwater recharge sources, mainly irrigation and streamflow losses [7]. When comparing TRB with OERRB, the latter enjoys more surface water than the TRB, and there is even a cross-regional transfer (Rocade Canal) from dams in OERRB to supply irrigation and the drinking water needs of the city of Marrakech located in TRB. More surface water in OERRB means more irrigation recharge and less pressure on groundwater, which is the opposite case in TRB where groundwater is being depleted in a drastic way with more than 90 m of decrease in some areas since early 2000. The decrease in surface water supply from the High-Atlas in TRB is shown by [13], as is the drop in water levels resulting in significant lag time and less effective groundwater recharge. Recently published studies [7,8,50,74] show that groundwater recharge in the TRB is limited to the mountain front areas and is expected to be low in the rest of the areas.

The observed changes in groundwater storage, especially the observed decreased tends in the TRB and OERRB Basins, have important implications for regional agriculture and water management. Agriculture in both basins is strongly reliant on surface and groundwater resources, due to their common semi-arid climate where water availability is a major restriction. The changes in groundwater storage might have a direct influence on agricultural output and sustainability.

The TRB and OERRB’s reduction in GWL anomalies, as identified by both GRACE and in situ data, shows a potential over-extraction of groundwater resources, which is frequently associated with crop irrigation practices in the region. Farmers may face higher expenditures for pumping water from greater depths, less access to irrigation water, and increased vulnerability to droughts if groundwater levels continue to fall. This might have a significant impact on crop output, particularly for water-intensive crops, necessitating a change to drought-resistant crop types or the use of more effective irrigation systems.

Because groundwater serves as an important factor during dry years, a decrease in standardized GWL anomalies may reduce the region’s agricultural systems’ tolerance to climatic variability and extended droughts. These findings underscore the critical necessity for integrated water resource management methods that incorporate both surface and groundwater resources. This involves supporting sustainable groundwater extraction techniques, increasing irrigation efficiency, improving water resource monitoring and management, and encouraging adaptive agricultural strategies to reduce the effects of variable water supply.

While our study uses GRACE data to assess groundwater and total water storage, we recognize the limitations of its coarse native resolution, which can obscure small-scale hydrological variations, as noted in previous studies (e.g., [76,77,78]). The ERA5-Land dataset, used for climatic inputs, may not accurately capture localized hydrological dynamics in smaller-scale research [28,79,80]. Despite these limits, these datasets give valuable insights, and we have attempted to reduce these concerns by using the 25° GRACE dataset. Subsequent studies ought to carry out the further development of these methodologies, including incorporating downscaling methods or higher-resolution data sources to enhance assessments at the local level [15,81].

4. Conclusions

The present work provided an in-depth perspective on groundwater dynamics in a water-scarce region within Morocco’s main basins by combining data from multiple resolutions. GRACE and ERA5-Land products were used to study changes in both terrestrial and groundwater storage across two large river basins with contrasting land and water use characteristics: the Tensift River Basin (TRB) and the Oum Er-rbia River Basin (OERRB). A water balance analysis was employed to compare these two products, considering variables such as precipitation, evapotranspiration, runoff, soil moisture, and snow contribution, with the aim of identifying the most appropriate product for capturing local spatial variabilities in terrestrial and groundwater storage.

Two GRACE TWS products, including CSR Mascon and JPL Mascon (RL06), were used alongside ERA5-Land to generate further datasets such as precipitation (P), evapotranspiration (ET), and runoff (Q). Both GRACE TWS outputs showed substantial interrelation in the examined basins. GRACE-derived GWS showed a moderate correlation with groundwater levels in the OERRB (R = 0.59), but a lower correlation in the TRB (R = 0.30). In contrast, GWS derived from ERA5-Land revealed a good agreement with groundwater levels in the OERRB (R = 0.72) and a reasonable correlation in the TRB (R = 0.63). In terms of assessing groundwater storage anomalies, ERA5-Land proved to outperform the GRACE products, suggesting that high-resolution datasets such as ERA5-Land may be better suited for capturing local-scale heterogeneity in groundwater-level changes, thereby enhancing our ability to monitor and manage water resources effectively at a regional level.

In analyzing the trends, seasonal cycles, and residuals of groundwater-level changes, our study revealed distinct patterns between the TRB and the OERRB. The results indicate that the ERA5-Land data showed a stronger correlation with the in situ groundwater-level data compared to GRACE data, particularly for the seasonal and de-seasonalized components. In the OERRB, the correlation coefficients for the seasonal (R = 0.64) and de-seasonalized (R = 0.48) components were higher for ERA5-Land than for GRACE (R = 0.67 and R = 0.57, respectively), indicating that ERA5-Land is more consistent with observed groundwater dynamics. In the TRB, the correlation of ERA5-Land data with the in situ seasonal cycle (R = 0.38) and de-seasonalized components (R = 0.56) was also stronger than that of GRACE data (R = 0.17 and R = 0.39, respectively). These findings suggest that, while both ERA5-Land and GRACE data capture the general trends in groundwater levels, ERA5-Land demonstrates superior performance in reflecting the seasonal and residual variations across both basins, likely due to its higher spatial resolution and better ability to capture local hydrological processes.

However, there was an observed time lag of approximately three months between ERA5-Land and in situ datasets in the TRB. This lag suggests a delayed response of groundwater levels to climatic variables, which is linked to drastic groundwater depletion due to the decrease in surface water from the High-Atlas and the drop in water levels. This requires further investigation to better understand local hydrological processes. Future research should consider integrating these remote-sensing datasets into groundwater flow numerical modeling within a Geographic Information System (GIS) framework. Such integration, as suggested by De Filippis et al. [82], could enable the development of time-variant groundwater budgets and improve our understanding of groundwater flow dynamics, particularly in regions with complex spatial zonation of water resources.

Furthermore, the results of this study have potential synergies with the Next Generation Gravity Mission (NGGM) [83,84,85], which is designed to provide Earth’s gravity field measurements at the highest accuracy, significantly improving our understanding of variability in terrestrial water storage. Aligning our research methodologies with the high-resolution data from NGGM will allow for a more detailed assessment of groundwater resources. These efforts underscore the importance of continued investment in remote-sensing technologies and collaborative approaches to addressing water scarcity in semi-arid regions.