Estimating groundwater abstraction for irrigation in a data-scarce semi-arid region using optical and InSAR time-series analysis: the Tensift catchment, Morocco

In Morocco, the expansion of irrigated agriculture has intensified groundwater abstraction and increased pressure on overexploited aquifer systems. Moreover, the substantial lack of pumping measurement data represents a critical limitation for water resources management. To address this gap, the present study quantifies groundwater abstraction at the perimeter scale during agricultural season (2015–2016) in the Tensift catchment, using an integrated methodology combining optical and radar remote sensing (Sentinel-1,-2) with field-based observations. The results show that the total groundwater abstraction reached 4.13 × 10 6 m 3 with a monthly average of 344,515 m 3. During the April–August period, groundwater supplied 36.1% of crop water requirements, while dam releases contributed 63.9%. These estimates were validated against pumping records from monitoring wells, where field measurements quantified groundwater and dam release contributions of 45% and 55%, respectively. The water balance assessment indicated a total agricultural water supply of 18.8 × 10 6 m 3.yr −1, consistent with crop water requirements estimated at 17.8 × 10 6 m 3.yr −1. InSAR-derived land subsidence analysis revealed marked spatial heterogeneity, with maximum subsidence rates reaching 6.5 mm.yr −1, concentrated in the northern sector and isolated central clusters. Overlap analysis showed a co-occurrence rate of 70.6% between high-abstraction and subsidence zones. The spatial Pearson correlation between land subsidence and groundwater abstraction ranges from 0.39 to 0.71. This study demonstrates that the proposed approach provides a robust and scalable framework for groundwater abstraction monitoring in data-scarce environments.

Arid and semi-arid regions cover approximately 30% of the Earth’s land surface and face increasing water scarcity due to climate variability and growing water demand (Okin et al., 2006; Scanlon et al., 2006; Bierkens and Wada, 2019; Wang et al., 2021). Population growth increases water demand for food security, driving the expansion of irrigated areas; global water requirements for food production are projected to rise significantly to meet the demands of a 50% increase in global population, from approximately 6 billion in 2000 to 9 billion in 2050 (Scanlon et al., 2007). Consequently, groundwater withdrawals have progressively intensified, leading to overexploitation of groundwater resources (Scanlon et al., 2006; Smith et al., 2016; Nassah et al., 2018).

Although the combined use of surface water and groundwater is ideal, the imbalance is evident: as surface supplies fail to meet the baseline requirements for expanding irrigation and urbanization, groundwater serves as a buffer, leading to depletion rates that often exceed natural recharge capacities (Bierkens and Wada, 2019).

Given this pressure, groundwater is becoming as a critical driver of socio-economic development in semi-arid regions, where surface water resources are often limited (Bierkens and Wada, 2019; Jasechko et al., 2024). It constitutes the primary water source for domestic, urban, agricultural, and industrial uses worldwide (Aeschbach-Hertig and Gleeson, 2012; Famiglietti, 2014; Hoogesteger, 2022). Globally, groundwater accounts for approximately 33% of total water withdrawals worldwide (Siebert et al., 2010; Famiglietti, 2014). Irrigated agriculture remains its dominant consumer, accounting for approximately 70% of total freshwater withdrawals, and constitutes the economic pillar of many developing countries in these regions (Fischer et al., 2007; Schultz et al., 2009; Saeed et al., 2021).

Groundwater plays a key role in sustaining water supply during drought periods (De Stefano et al., 2015; Scanlon et al., 2016; MacDonald et al., 2019), as it is more available and less susceptible to drought impacts than surface water (Aeschbach-Hertig and Gleeson, 2012). Under prolonged drought, groundwater often becomes the main or only available source, meeting both basic human needs and production requirements (Bourque et al., 2019; Nunes et al., 2022). However, increasing dependence on groundwater as a buffer against surface water scarcity often leads to overexploitation, as abstraction rates frequently exceed natural recharge capacities, threatening the long-term availability of the resource.

Groundwater depletion is a critical global challenge, particularly in arid and semi-arid regions, where populations depend largely on this resource for domestic use and irrigation. Over the past century, the volume of water abstracted for human needs has increased from approximately 500 to about 4,000 km 3 per year (Oki and Kanae, 2006; Wada et al., 2014; Bierkens and Wada, 2019). In these regions, low groundwater recharge rates combined with limited alternative water sources, mean that groundwater abstraction exceeds recharge, leading to aquifer depletion largely attributed to agricultural withdrawals (Scanlon et al., 2007; Gleeson et al., 2010; Walsh, 2022). This persistent overexploitation causes continuously declining water tables, long-term depletion of deep aquifers, land subsidence, and downstream water scarcity (Siebert et al., 2010; Madani et al., 2016; Ashraf et al., 2021; Le Page et al., 2025).

In response to severe aquifer depletion, many governments have implemented various mechanisms to regulate and control groundwater abstraction (Shah et al., 2003; Theesfeld, 2010; Hoogesteger and Wester, 2015). However, sustainable groundwater management depends on accurate and reliable monitoring of groundwater withdrawals. Such monitoring data are essential for detailed assessments of groundwater balances and for evaluating the sustainability of alternative management strategies (Zekri et al., 2017; Esteban and Dinar, 2013; Wang et al., 2020). Despite their central role, groundwater resources remain poorly monitored in many regions, and pumping withdrawals are frequently unrestricted and unrecorded, complicating the estimation of actual consumption rates and underscoring the absence of a comprehensive global monitoring framework (Daly et al., 2008; Famiglietti et al., 2011).

Direct monitoring methods, such as mechanical or smart water meters installed at pump outlets, are the simplest and most cost-effective ways to track extraction(Mohapatra and Mitchell, 2009; Ursitti et al., 2018). Indirect methods based on proxy indicators such as irrigated area and irrigation duration are more accessible but often yield unreliable estimates, as water application rates can vary considerably between fields even under similar climatic and soil conditions (Wang et al., 2020).

Individual well metering is further constrained by institutional and logistical challenges: smallholder-dominated user communities make monitoring and enforcement costly (Ross and Martinez-Santos, 2010; Vos and Vincent, 2011; Ursitti et al., 2018), metering is often perceived as an intrusion into traditional water-use practices or an infringement of historically inherited water rights (d'Elgin, 2016; Walsh, 2022), and equipment remains vulnerable to vandalism and theft in remote areas.

Estimating groundwater abstraction remains an under-quantified component of the water balance, particularly in semi-arid regions where surface water supply is increasingly insufficient to meet growing agricultural demand. In Morocco, the Green Morocco Plan (2008–2018) and its successor Generation Green (2018–2028), have rapidly expanded irrigated agriculture, intensifying uncontrolled reliance on already overexploited aquifer systems. In this context, high-resolution remote sensing data combined with field observations offer a promising and robust alternative for large-scale, near-real-time monitoring of water resources. For example, the Gravity Recovery and Climate Experiment (GRACE) mission (Tapley et al., 2004) provides estimates of variations in total water storage which, when combined with auxiliary hydrological datasets, can yield groundwater storage change estimates of sufficient accuracy to support water management (Yeh et al., 2006; Zaitchik et al., 2008; Amazirh et al., 2024).

The Ouled Gaid perimeter is one of the irrigated schemes managed by the Regional Office of Agricultural Development of Haouz (ORMVAH), which oversees approximately 473,000 hectares of irrigated land across the region. The water management constraints documented here are representative of those prevailing across the wider ORMVAH fields of action, conferring the methodology a direct transferability to comparable perimeters. The representativeness of the study area rests on two complementary dimensions. From an agronomic standpoint, the farming structure dominated by smallholdings (<5 ha) under olives, cereals, fodder crops, and trees with intercropped annual crops, reflects the prevailing model across irrigated plains, including the Haouz, Tadla, and Doukkala perimeters. From a climatic standpoint, the 2015–2016 season, recorded as a dry year with a mean annual rainfall of 154.8 mm, is representative of the recurrent water deficit conditions characterizing the semi-arid context, ensuring that abstraction estimates reflect demand patterns under the stress conditions most relevant for regional water resource planning.

The near-total lack of direct measurements and monitoring of groundwater abstraction represents a critical and persistent gap in water balance assessment and water resource management. The remote sensing approach developed here, combining Sentinel-1/-2 with Interferometric Synthetic Aperture Radar (InSAR) using the Small Baseline Subset (SBAS) technique, is designed to bypass the need for ground-based pumping records, remaining applicable in data-scarce environments subject to appropriate parameterization accounting for local agricultural practices and surface water availability. Beyond quantification, this work introduces groundwater abstraction as an underquantified component of the water balance, which is often unrecorded or crudely approximated in hydrological and hydrogeological models. Integrating groundwater abstraction into these models would improve their realism and predictive capacity, and better represent the interactions between agricultural water demand, aquifer dynamics, and climatic variability in semi-arid irrigated systems.

In this framework, the present study provides a spatio-temporal estimation of groundwater abstraction at the scale of the Ouled Gaid perimeter over the 2015–2016 agricultural season. To address this gap, two interconnected objectives are pursued: (i) to quantify groundwater abstraction at the irrigation perimeter scale and assess the contribution of pumping to water supply relative to surface water resources in meeting crop water requirements; (ii) to explore land subsidence as a spatial proxy for delineating high-abstraction zones, providing an indirect yet robust approach for mapping pumping patterns where ground-based records are unavailable.

These objectives are addressed within a semi-arid smallholder farming context through an integrated framework combining Sentinel-2 optical imagery with a crop water balance model to estimate spatiotemporal groundwater abstraction volumes. Abstraction estimates were validated at two complementary levels: directly against groundwater pumping monitoring data, and indirectly by comparing the reconstructed water supply (pumping, rainfall, and dam releases) with crop water requirements derived from land use mapping, offering an independent test of water balance coherence. The approach is further strengthened by coupling estimated abstraction volumes with land subsidence derived from InSAR data, enabling a quantitative link between aquifer exploitation and differential land subsidence, an articulation rarely achieved in the semi-arid Mediterranean context.

Several studies have highlighted a strong relationship between groundwater dynamics and land subsidence. Longer and more severe groundwater droughts were associated with greater land subsidence, with wavelet coherence analysis revealing significant correlations between groundwater drought and InSAR-derived deformation (Shahnazi et al., 2025). Similarly, InSAR-derived subsidence rate maps revealed a strong relationship between land subsidence and groundwater-level variations, highlighting the role of groundwater dynamics in driving ground deformation (Chen et al., 2016). Excessive groundwater extraction was strongly associated with groundwater-level decline and land subsidence (Abidin et al., 2009). Long-term excessive groundwater extraction was identified as a major driver of land subsidence between 1976 and 2006 in areas of intensive extraction (Feng et al., 2008). Increasing groundwater extraction from deep confined aquifers was associated with significant settlement, primarily driven by the consolidation of thick soft clay layers (Chen et al., 2003). Building on this established relationship, the contribution of the present study lies in the specific remote sensing-based framework developed to quantitatively estimate groundwater abstraction volumes in a data-scarce environment, by combining InSAR-derived subsidence with a simplified water balance approach. While previous studies have primarily investigated the relationship between groundwater dynamics and land subsidence, our approach integrates these observations with crop water requirements and surface-water inputs to derive spatially explicit estimates of groundwater abstraction. In this framework, subsidence patterns are used as an indirect indicator of groundwater abstraction rather than as a direct quantitative proxy, acknowledging that subsidence magnitude is also influenced by geological, hydrogeological, and other environmental factors.

Study area

The Ouled Gaid perimeter is one of the irrigated schemes managed by the ORMVAH, which oversees approximately 473,000 ha of irrigated land within the Tensift basin. The Ouled Gaid agricultural sector forms an integral part of the upper Tessaout irrigation perimeter (52,000 ha), located approximately 70 km east of Marrakech. The study area covers 2,361 ha and has been equipped with a modernized gravity-fed irrigation system since 1977 (Figure 1). Water supply in the sector is primarily sourced from the Tessaout basin, regulated by the Moulay Youssef Dam. This dam, with a total storage capacity of 175 mm 3 supports the irrigation of 30,000 ha and generates approximately 60 GWh per year of hydropower for the cities of Marrakech and Ouarzazate. A secondary dam Timinoutine, located 2 km downstream of Moulay Youssef, enables inter-monthly regulation of water releases to match agricultural demand and mitigate excess flows. Oulad Gaid is composed of smallholdings under 5 ha, dominated by olives, cereals, fodder crops, and intercropping, a land-use structure that represents the prevailing model across irrigated plains.

The study area has a semi-arid Mediterranean climate with limited water resources, characterized by successions of dry years’ representative of recurrent water deficit conditions. Annual rainfall ranges between 250 and 500 mm and exhibits high spatial and temporal variability (Chaponnière et al., 2008; Hajhouji et al., 2018). Streams are often ephemeral, with intermittent or zero discharge (Bouimouass et al., 2020; Hajhouji et al., 2022), and potential evapotranspiration rates are high, reaching approximately 1,600 mm.yr −1 (Er-Raki et al., 2010). The study area is located within the Haouz aquifer, an unconfined aquifer hosted in Plio-Quaternary alluvial deposits and Neogene formations. It constitutes the primary and most heavily exploited groundwater reservoir in the region. Measured transmissivity values vary from 5.0 × 10 −5 to 9.0 × 10 −2 m 2 /s, and hydraulic conductivity values vary from 4.6 × 10 −6 to 5.5 × 10 −3 m/s (Sinan and Razack, 2006). The deposits of the Ouled Gaid sector belong to the Middle Quaternary; the dominant facies is characterized by a rubified soil with scattered pebbles and an encrusted silty-clay matrix. Previous studies have shown evidence of groundwater recharge from streamflow, particularly during flood episodes in the piedmont area (Boukhari et al., 2015; Hajhouji et al., 2022), as well as from irrigation return flows (Jarlan et al., 2015). Despite being the most productive aquifer in the region, it is also subject to intense groundwater exploitation for agricultural and urban water needs (El Faïz, 2002). Consequently, groundwater levels in the region have been steadily declining since the 1980s, with annual drawdowns of 0.5 to 1.5 m and occasional declines of up to 3 m.yr −1 (Abourida et al., 2009; Le Page et al., 2012; Jarlan et al., 2015). In our study area, which is part of the Haouz aquifer, the depth of the observation wells used for validation, ranges from 80 m to 140 m.

Data and methodology

Data

Precipitation data were recorded at the Bour meteorological station (Latitude: 31.701°; Longitude: −7.352°; Elevation: 730 m). From a climatic standpoint, the 2015–2016 agricultural season was a dry year, with a mean annual rainfall of 154.8 mm. Monthly precipitation ranged from null values in December to a maximum of 50 mm in March. Monthly dam allocations for the Ouled Gaid irrigation sector, supplied by the Moulay Youssef Dam, were obtained from ORMVAH for the 2015–2016 agricultural season. Deliveries at the head of the perimeter ranged from 0.47 × 10 6 m 3 in February to 1.51 × 10 6 m 3 in October, totaling 14.53 × 10 6 m 3 /year. These monthly volumes were used as the dam allocation term (Idam) in the water balance calculation. Field data comprised groundwater abstraction measurements from monitoring wells and an in-situ crop survey conducted by ORMVAH across the study area. These field observations were used to validate the land-use classification and independently assess the reliability of the estimated groundwater abstraction.

Methodology

This study aims to estimate groundwater abstraction in the Ouled Gaid sector during the 2015–2016 agricultural season using a remote sensing-based approach that combines optical Sentinel-2 data and InSAR data (Figure 2). The first step involved calculating the Normalized Difference Vegetation Index (NDVI) from Sentinel-2 imagery to produce monthly NDVI maps, allowing assessment of temporal variations in vegetation cover and identification of distinct vegetation types across the study area. The second step was to develop a land use map for 2015–2016 by combining NDVI temporal profiles with field surveys. NDVI was determined for each satellite image throughout the 2015–2016 season, allowing its temporal evolution to be tracked in each pixel for the entire year. These NDVI temporal profiles were used to identify vegetation classes and crop growth stages, since a single NDVI image cannot reliably discriminate between crop types that may share similar spectral values at a given date but follow distinct phenological cycles (Simonneaux et al., 2007). In addition, the field dataset consists of an in-situ plot-level survey and investigations conducted by ORMVAH across the study area, with crop types verified and geolocated. These field observations were paired with NDVI time series to derive baseline spectro-temporal signatures for each main crop category and to improve classification accuracy. A supervised minimum distance classification was then applied; chosen for its robustness with multi-temporal spectral profiles; combining the NDVI temporal variations with the field-identified agricultural plots to produce the final land use map (Hajhouji et al., 2024). Finally, contextual filters were applied to refine the surface quantification estimates. Crop Evapotranspiration (ETc) maps were developed by combining crop coefficients (Kc), land use maps and reference evapotranspiration (ETo). ETo was estimated using the FAO Penman-Monteith method (Howell and Evett, 2004), while ETc was calculated by multiplying ETo by Kc values following Allen et al. (1998) (Equation 1). The Kc values were derived from NDVI using the Kc–NDVI relationship. The relationship between Kc and NDVI used in the Haouz region for the different types of crops was established by Er-Raki et al. (2008).

Groundwater abstraction volumes were estimated using a simplified crop water balance approach, where the equilibrium between water inputs and outputs provides an indirect assessment of groundwater abstraction. The objective was to determine the potential volume of groundwater abstracted from the aquifer to compensate for deficits between crop water requirements and surface water supplies. This study follows the simplified crop water balance (Equation 2), in which groundwater abstraction is estimated as the residual deficit between crop water requirements (ETc) and available surface water inputs: precipitation and irrigation supplied by groundwater abstraction and dam allocations. This approach is conceptually consistent with basin-scale water balance methods in which groundwater pumping is treated as a residual term to close the water balance (Ruud et al., 2004). Water balance models can range from complex conceptual formulations with numerous parameters to simplified models retaining only a few key parameters (2 to 5 parameters), especially at the monthly water balance scale (Xu and Singh, 1998). In line with this simplified modeling approach, and given the objective and the data available for this study, we opted for a simplified formulation retaining only the most influential terms of the water balance: crop water requirements, precipitation and irrigation (groundwater abstraction and dam allocations). Several simplifications were applied. Runoff was considered negligible at the basin scale given the essentially flat topography (Ruud et al., 2004; Catarino et al., 2025). Deep percolation was not explicitly represented in this simplified water balance (Abourida et al., 2009), which constitutes a limitation of the approach; however, given the dominant clay-loam soil texture in the study area and the depth of the water table (exceeding 80 m at monitoring wells), its impact on the overall water balance is expected to be minor. Quantifying deep percolation as a boundary condition remains a recognized challenge for this type of model and represents a priority for future refinement.

At the monthly and sector scales, crop water requirements (ETc), estimated from the crop type identified on the land-use map, were compared with available water inputs, comprising precipitation (P) and irrigation allocations from the dam (dam release). When P + dam release ≥ ETc, crop water requirements were considered fully satisfied by the available inputs, and no groundwater abstraction was required (GW abstraction ≈ 0). Conversely, when P + dam release < ETc, the residual deficit was assumed to be entirely compensated by groundwater abstraction, such that GW abstraction = ETc − (P + dam release). This assumption is supported by the operation of the studied irrigation system, in which dam allocations are regulated and set monthly according to reservoir availability, while precipitation during the study period (2015–2016) represents an observed water input. Thus, within this simplified approach, groundwater abstraction represents the residual term required to meet crop water requirements when surface-water inputs are insufficient.

To evaluate the groundwater abstraction volumes, the estimated extraction rates were validated against field data collected from 14 observation wells distributed throughout the study area. Groundwater abstraction at these control points was systematically monitored from April to August 2016. For the agricultural plots supplied by these wells, precipitation inputs and dam water allocations were also assessed to differentiate abstracted groundwater volumes from surface water contributions (Figure 2).

To assess the consistency of the results, crop water requirements at the study area scale were compared with total water supply, including precipitation, dam releases and groundwater pumping. Requirements were weighted by the relative area of each crop type from the land use map. This comparison provides an order of magnitude estimate of total water supply and serves as an independent validation of water balance coherence across the irrigated perimeter.

Surface displacement was measured with the Small Baseline Subset (SBAS) InSAR method applied to Sentinel-1A/B IW SLC images from a single descending track (relative orbit 154). Two stacks were processed: an intensive stack of [N1] acquisitions covering the year 2015, and a decadal stack of 11 acquisitions (one summer scene per year) covering 2015–2025. Only sequential pairs were formed, with temporal baselines of 12–24 days for the intensive stack and about 365 days for the decadal stack, and perpendicular baselines |B⊥| < [Bp] m. The full baseline network is shown in Supplementary Figure S1. SBAS is widely recognized as a robust multi-temporal approach because it limits decorrelation and topographic phase errors by using only interferograms with short spatial and temporal baselines.

The whole chain was run with open-source software: PyGMTSAR v2/GMTSAR for interferometry, SNAPHU for phase unwrapping, and xarray/rioxarray/Dask for the time-series analysis. Interferograms were multilooked (3 × 12) and coherence was estimated on the multilooked data; no amplitude dispersion or stability index was applied. Phase unwrapping was done interferogram by interferogram with SNAPHU (statistical-cost, network-flow), using coherence as the cost weight, and each unwrapped field was re-referenced to a stable reference area before time-series accumulation. The operational unwrapping mask was γ ≥ 0.075. This permissive threshold was chosen to keep the unwrapping network spatially connected across the irrigated perimeter; low-coherence pixels therefore contribute only weakly to the solution. Because C-band coherence drops over densely cropped fields during the growing season, the full analysis was repeated with a conservative mask (γ ≥ 0.30). Where the two results differ, the conservative result is retained, and pixels that do not pass the conservative mask are blanked in the figures rather than interpolated.

Line-of-sight (LOS) displacements were converted to vertical land motion (VLM) using d_v = d_LOS / cos θ, with the scene-average incidence angle (41° / 50° for the two stacks). VLM carries a sign (positive = uplift, negative = subsidence), and the term “land subsidence” is used only for negative VLM. Since a single track was used, horizontal and vertical motion cannot be separated; an ascending-track stack was processed as an independent consistency check. The LOS rates and displacement time series are made available in an open data repository.

Within this framework, highly localized, quasi-circular patterns of negative VLM (“subsidence bowls”) were used as indicators of the position of active pumping wells, and were compared with well locations reported by well owners.

Spatial comparison between abstraction and VLM

To examine the spatial relationship between groundwater abstraction and vertical land motion, two independently produced datasets were combined: The InSAR-derived cumulative VLM field (full distribution, including both subsidence and uplift) and the groundwater abstraction volumes estimated from the satellite-based water balance model. Both layers were resampled to a common 100 m × 100 m grid in the same coordinate system (UTM Zone 29 N). A threshold at the 30th percentile was applied to both fields to isolate zones of high pumping (≥ 3 mm yr −1) and marked subsidence (≥ 0.73 mm), and the overlap was calculated as the proportion of the high-pumping area located inside high-subsidence zones (Equation 3). Because a single threshold could appear arbitrary, a sensitivity analysis was performed: the overlap was recomputed for every percentile from the 10th to the 50th, in steps of 5, applied jointly to both fields (Supplementary Figure S1). The overlap fraction remains stable across this range, which shows that the reported value does not depend on the choice of the 30th percentile. In addition, the comparison was done both for the decadal (2015–2025) VLM field and for the period-matched 2015–2016 deformation from the intensive stack, so that the abstraction estimate (2015–2016 season) is compared with deformation of the same period. Finally, this spatial correspondence is interpreted as spatial consistency between two independent datasets—a necessary but not sufficient condition for a causal link—and not as proof that pumping is the dominant driver of the observed deformation.

Spatiotemporal patterns of NDVI

During the 2015–2016 agricultural season, analysis of NDVI time series derived from Sentinel-2 imagery enabled the characterization of the phenological dynamics of major land use classes within the study area. Generated maps highlight monthly NDVI variations at the regional scale, reflecting fluctuations in vegetation cover throughout the agricultural year (Figure 3). The lowest NDVI values were observed at the beginning and end of the season, corresponding to bare or sparsely vegetated soils, whereas peak NDVI values indicated dense vegetation cover, particularly in areas with forage crops such as alfalfa and berseem, which exhibited values exceeding 0.6 over an extended period.

Hajhouji et al. (2024) investigated the temporal dynamics of NDVI profiles for the dominant land-cover classes and demonstrated that each vegetation type exhibits a distinctive curve, indicative of its unique phenological behavior. Cereal crops show a pronounced NDVI peak in spring (≈0.59), whereas olive trees maintain a stable profile around 0.4 throughout the season. Trees and annual crops exhibit intermediate dynamics, with trees maintaining a relatively stable NDVI around 0.4, reflecting their inherent canopy reflectance, while seasonal increases associated with intercropped annual crops can raise NDVI values above 0.6. Bare soils maintain consistently low NDVI values year-round. This spatiotemporal analysis underscores the reliability of NDVI as a valuable indicator for crop type discrimination, growth stage monitoring, and assessment of agricultural practices.

Land-use map and spatiotemporal patterns of evapotranspiration

The land-use distribution in the Ouled Gaid sector during the 2015–2016 agricultural season revealed that cereals and olive trees were the predominant crops, covering 792 ha (33.56%) and 530 ha (22.46%), respectively. Areas combining trees with intercropped annual crops and fodder crops occupied approximately 451 ha (19.11%) and 434 ha (18.39%), respectively, while bare soil accounted for 153 ha (6.48%) of the total area (Figure 4). The obtained land-use was adequately simulated using field data acquired by the ORMVAH. The results showed a Kappa coefficient of 0.91 and an overall accuracy of 93.3% (Hajhouji et al., 2024).

The Kc values derived from NDVI using the Kc-NDVI relationship and ET0 datasets were combined with the established land-use map to generate monthly ETc maps that highlighted both temporal and spatial variability (Figure 5). Temporally, ETc showed clear seasonal contrasts between dry and humid periods, while spatially it varied according to crop type, with values ranging from 816 mm for fodder crops (alfalfa and berseem) to 493 mm for cereal crops (wheat). The temporal dynamics of ETc can be divided into two distinct phases. A moderate phase, characterized by values below 120 mm, occurred from October 2015 to March 2016, this was followed by a pronounced phase extending from April to September, during which ETc increased markedly, ranging from 176 mm to a maximum of 303 mm.

The Spatial patterns of ETc strongly influenced by crop type. Fodder crops exhibited the highest ETc values reaching approximately 816 mm. Tree-based systems (fruit trees and wild trees) intercropped with annual vegetables showed intermediate values, around 700 mm. Lower ETc values were observed for cereals and olive trees, with totals of 493 mm and 647 mm, respectively (Figure 5).

Spatiotemporal patterns of groundwater abstraction

The potential volumes of groundwater abstracted were quantified both spatially, across the 2,361 ha irrigated area, and temporally over the 2015–2016 agricultural year, using a remote sensing approach combining optical Sentinel-2 data with field observations for validation.

The results show pronounced seasonal variability in groundwater pumping, closely associated with climatic conditions, crop types, agricultural calendar and availability of surface water resources. Total volume of groundwater abstracted within the study area was estimated at 4.13 × 10 6 m 3, corresponding to a monthly average of 344,515 m 3. At the beginning of the agricultural season in September, the pumped groundwater volume was approximately 425,700 m 3, then decreased sharply during the wet period between October and March. During this phase, groundwater abstraction remained low, with minimum values recorded in March (4,800 m 3), November (6,300 m 3) and October (13,650 m 3) (Table 1), respectively. These low pumping rates coincide with high dam releases, which reached an allocated volume of 1.50 × 10 6 m 3 in October and November, and with the maximum rainfall observed in March at 50 mm. This inverse relationship between surface water availability and groundwater abstraction confirms that farmers reduce their reliance on groundwater when surface resources are sufficient to meet irrigation requirements. From April the groundwater pumping increased significantly, reaching 293,700 m 3 and continued to rise throughout the dry season. The highest pumping values were recorded in July with 1.07 × 10 6 m 3 and August 0.95 × 10 6 m 3. This intensification reflects the combined effect of dry climatic conditions, the near absence of rainfall, and the peak evaporative demand.

Despite relatively high dam allocations during this period with 1.31 × 10 6 m 3 in July and 1.36 × 10 6 m 3 in August, the surface water alone was insufficient to meet crop water needs, particularly for perennial crops such as olive trees, trees intercropped with annual crops and fodder crops.

The spatial distribution of groundwater pumping reveals strong heterogeneity across the study area (Figure 6), with the highest abstraction rates concentrated in the central and northern parts of the region. These high-abstraction zones correspond spatially to areas dominated by olive trees, mixed zones of trees combined with annual crops fodder crops, as shown in the land use map. In contrast, areas covered by cereals tend to coincide with moderate to low pumping values. The mixed zones of trees combined with annual crops appear to drive localized groundwater abstraction, reflecting the coexistence of perennial and seasonal irrigation demands. Overall, this spatial correspondence between land use and pumping intensity suggests that crop type is a key determinant of groundwater abstraction patterns, with water-intensive annual crops exerting the greatest pressure on the aquifer.

The results of groundwater abstraction showed marked seasonal dynamics, remaining limited during the wet season when surface water availability was adequate, and increasing substantially during the dry period despite relatively significant dam water allocations. Spatially, pumping was concentrated in areas dominated by perennial and mixed cropping systems, highlighting the key influence of crop type on groundwater demand. Overall, these findings emphasize the strategic role of groundwater as a resource water-scarce periods, ensuring crop water supply when surface water allocations alone were insufficient to meet full irrigation requirements.

To evaluate the reliability of the mapped groundwater pumping estimates, field-measured abstraction (converted from volumetric measurements using the irrigated area of each well) was compared against the corresponding class predicted by the remote sensing-simplified water balance approach for monitoring wells. Of the 14 monitoring wells initially instrumented, 2 were stolen during the experiment, and their data could therefore not be used for validation; the analysis was consequently conducted on the remaining 12 wells. Measured values fell within their assigned mapped class in 100% of cases, confirming the categorical consistency of the classification. To assess agreement in magnitude, each class was represented by a reference value (the class midpoint) and compared against the field-measured abstraction. This comparison showed a strong linear relationship (R 2 = 0.97) and a significant Spearman rank correlation (ρ = 0.98, p < 0.001), confirming the consistency of the classification. A small positive bias was observed (+2.40 mm) but was not statistically significant (p = 0.91), indicating no systematic over- or under-estimation. Although the validation sample was limited, the strength and consistency of this agreement across multiple metrics provide robust evidence supporting the validity of the proposed method. Together, these results (Figure 7) indicate that the remote sensing-simplified water balance approach reliably estimates groundwater pumping, both in class assignment and in relative magnitude across wells.

Seasonal dynamics of water supply components

During the 2015–2016 agricultural season, the water supply for agriculture in the Tensift region revealed the structural dependence of local farming systems on multiple water sources, namely rainfall, dam allocations and groundwater abstraction. Rainfall exhibited spatiotemporal variability and remained largely insufficient, with an annual total of 154.8 mm. Monthly rainfall ranged from 0 mm to a maximum of 50 mm in March, with a monthly average of 12.9 mm (Figure 8). Dam allocations totaled 14.53 × 10 6 m 3, with a mean of 1.21 × 10 6 m 3 per month, and monthly volumes ranging from 0.47 × 10 6 m 3 in February to 1.51 × 10 6 m 3 in October. Allocations were relatively high in autumn, decreased during winter, and gradually increased in spring and summer to meet crop water requirements. This monthly variability was mainly driven by the dam filling rate and rainfall conditions. Total groundwater abstraction was estimated at 4.41 × 10 6 m 3, with monthly withdrawals ranging from 4,800 m 3 in March to 1.07 × 10 6 m 3 in July. Groundwater abstraction volumes remained relatively low during autumn and winter, corresponding to periods of higher natural precipitation and lower irrigation demand. Pumping gradually increased in spring and peaked in summer, driven primarily by crop water requirements and limited availability of surface water from rainfall and dam releases (Figure 8).

Irrigation water supply in the study area relies on three main sources: rainfall, dam water allocations, and groundwater abstraction. Analysis of their respective contributions reveals strong seasonal dynamics and functional complementarity among these sources. Rainfall is concentrated in winter, mainly between January and March, during which it constitutes the primary source of irrigation. Rainfall contributes up to 58% of the water mobilized for agriculture in February and 46.8% in March, highlighting its significant yet limited role in the semi-arid context of the region. In contrast, dam water allocations provide the major part of irrigation supply throughout the year, generally exceeding 50%, except in February when they represent only 36% due to higher rainfall reducing dependence on dam releases. Groundwater abstraction shows an inverse pattern to rainfall, it decreases when precipitation increases and intensifies during dry periods, particularly from May to August, supporting the heightened irrigation demand in summer, with contributions reaching up to 44.3% in July (Figure 9).

The variability and seasonal dynamics of water supply underscore the limited contribution of rainfall under semi-arid conditions, characterized by low and irregular precipitation. Within this framework, surface water from the dam is the principal source of irrigation, while groundwater plays an important complementary role by compensating for surface water deficits during periods of increased irrigation demand or reduced dam allocations. In semi-arid contexts, groundwater acts as a safety valve, ensuring water availability for irrigation during peak demand periods or droughts, particularly when rainfall is scarce and dam releases are insufficient.

Validation of groundwater pumping using field measurements

Groundwater abstraction results obtained were validated using field data collected through a monitoring network of 14 wells, established in collaboration with ORMVAH. This number reflects the maximum monitoring network achievable under the field conditions encountered: despite considerable efforts, significant challenges arose in identifying suitable monitoring wells and accurately measuring groundwater abstraction volumes, compounded by resistance from well owners, additional sensors initially installed on other parcels were subsequently stolen, reducing the effectively available monitoring points. This constraint is consistent with the broader difficulty of implementing dense, individual-well metering in smallholder-dominated groundwater systems, as discussed in the Introduction. Well locations were nonetheless selected to span the dominant land-use classes of the perimeter (cereals, olive trees, and intercropped tree/fodder systems; Figure 4), ensuring coverage of the main agronomic and hydraulic conditions encountered in the study area.

Field measurements were conducted over an irrigated area of approximately 250 ha between April and August 2016. Data were collected from 14 observation wells distributed across the study area. During this period, the measured water supply consisted of 343,800 m 3 from dam releases and 69,700 m 3 from groundwater abstraction. The crops grown on the sampled plots were primarily cereals, olive trees and fodder crops. The study area was dominated by cereals and olive trees, while fodder crops occupied a relatively small area. Within the study area, cereals are generally harvested between May and June, rendering their water requirements negligible during the second half of the study period. In contrast, olive trees, require regular and controlled irrigation inputs throughout the agricultural season. Over the April–August 2016 period, crop water requirements were estimated at 2,500 m 3.ha −1, corresponding to a total volume of 625,100 m 3 for the sampled area. The total measured water supply estimated at 413,500 m 3, accounts for approximately 66.2% of the total crop water requirements.

However, the remote sensing analysis was conducted over a larger area of 2,361 ha and covered the entire 2015–2016 agricultural year (September 2015 to August 2016), assuming full satisfaction of crop water requirements (100%). To enable comparison with field measurements from the control wells, the total water supply over the corresponding period (April–August 2016), under the assumption of full crop water requirement coverage, was estimated at 5,825,316 m 3 from dam releases and 3,291,045 m 3 from groundwater pumping.

To ensure consistency between the simulated and measured groundwater pumping, the datasets were harmonized to a common spatial and temporal scale. Dam water allocations for agriculture are governed by a predefined irrigation schedule and remain constant regardless of the level of water requirement coverage. Similarly, rainfall constitutes a constant input over the study period. Consequently, achieving 100% satisfaction of crop water requirements under field conditions necessitates an increase in groundwater abstraction, which serves as the adjustable component of the irrigation water supply. The estimated groundwater pumping required to achieve full crop water requirement coverage was 281,300 m 3, resulting in a total irrigation water supply of 625,100 m 3.

In conclusion, field measurements indicated that the relative contributions of dam releases and groundwater pumping to the total irrigation water supply were 55% (343,800 m 3) and 45% (281,300 m 3), respectively. In comparison, the Sentinel-2 remote sensing approach estimated the dam’s contribution at 63.9% (5,852,315 m 3) and groundwater pumping at 36.1% (3,291,045 m 3) (Figure 10).

The difference between field measurements and remote sensing estimates can be attributed to several factors. The monitored plots were predominantly cultivated with cereals, olive trees, and alfalfa, crops with relatively high water demands. Consequently, the observed 45% groundwater abstraction reflects local conditions and cannot be extrapolated to the entire irrigated perimeter. Additionally, due to the lack of comprehensive data on dam water allocations across the full study area, these plots are likely located in hydraulically disadvantaged sectors with restricted surface water access, resulting in groundwater pumping rates that exceed the estimated perimeter-wide average.

In addition, the Sentinel-2 approach inherently integrates both the variability in surface water allocations (hydraulic conditions) and the diversity of land use patterns (agronomic conditions) across the entire perimeter. The spatial variability of water supply, including areas with adequate access to surface water reduces the relative contribution of groundwater pumping at the overall scale compared to localized field measurements. Moreover, the spatial diversity of land-use patterns ranging from water-intensive crops such as fodder crops to water-efficient crops such as cereals, directly influences the spatial distribution of irrigation water requirements and the degree of local reliance on groundwater pumping.

Consequently, Sentinel-2-based estimates reflect the average hydraulic and agronomic conditions across the entire perimeter, as a result reducing the relative weight of groundwater abstraction compared to localized measurements. The latter remain susceptible to spatial sampling bias and may overestimate pumping contributions due to the disproportionate representation of hydraulically disadvantaged areas and highly water-consumptive crops.

Assessment of crop water requirements and irrigation water supply

The land-use map of the Ouled Gaid sector is dominated by cereal crops, olive trees, tree-crop intercropping, and fodder crops. Bare soils covering 153 ha were excluded from the calculation of water requirements; only the irrigated agricultural area was considered in the analysis. Consequently, the irrigated agricultural considered area amounts to 2,207 ha. Crop water requirements were estimated using values from previous studies conducted in the study area and from institutional references provided by the regional irrigation water management authority (ORMVAH). The adopted water consumption values for each crop type are presented in Table 2.

To estimate overall crop water requirements, a weighted average of water consumption was calculated based on the relative area of each crop class. The resulting water consumption was estimated at 8097 m 3.ha −1.yr −1. At the scale of the total irrigated agricultural area, the annual crop water requirement is estimated to range 17.87 × 10 6 m 3.yr −1.

Irrigation water supply in the study area relies on three main sources: rainfall, irrigation allocations from the Moulay Youssef dam, and groundwater pumping. The annual rainfall was estimated at approximately 154 mm.yr −1. Applied to the total irrigated area (2,207 ha), this corresponds to an annual volume of about 3.40 × 10 6 m 3.yr −1. Effective rainfall was estimated by applying a rainfall efficiency coefficient of 0.9, resulting in an effective rainfall volume of 3.06 × 10 6 m 3.yr −1. Irrigation allocations from the Moulay Youssef dam were estimated at approximately 14.52 × 10 6 m 3.yr −1. Accounting for an average conveyance efficiency of 20% for the canal distribution network, as adopted by ORMVAH, the total volume effectively delivered to farm plots was estimated at 11.61 × 10 6 m 3.yr −1. Additionally, groundwater volumes mobilized through pumping were estimated using a remote sensing–based approach, yielding an annual volume of approximately 4.13 × 10 6 m 3.yr −1.

The total water supply for agriculture in the Ouled Gaid sector, derived from rainfall, dam releases, and groundwater abstraction, and accounting for conveyance efficiency coefficients, amounts to approximately 18.8 × 10 6 m 3.yr −1 at the sector scale, corresponding to an average water input of approximately 8,519 m 3.ha −1.yr −1. This total volume is consistent with the crop water requirements estimated at 17.87 × 10 6 m 3.yr −1, particularly given that the 2015–2016 period was classified as a dry year (Hajhouji et al., 2022).

Detection of groundwater pumping-related land subsidence using InSAR time series

The over-extraction of groundwater resources causes severe environmental consequences beyond simple water table depletion, notably worsening water scarcity and promoting the salinization of coastal aquifers. Furthermore, intensive groundwater overexploitation is also a primary driver of land subsidence (Foster and Shah, 2012; Le Page et al., 2025; Madani et al., 2016; Zekri et al., 2015). To comprehensively monitor groundwater pumping and evaluate its associated subsidence impacts, Sentinel-1 InSAR data from 2015 to 2025 were utilized. Time-series InSAR techniques, particularly SBAS, are highly effective in mapping millimeter-scale ground deformations associated with high-volume extraction, enabling precise identification of both localized subsidence bowls and potentially unmapped or illegal pumping sites (Ahmad et al., 2022; Khan et al., 2022).

The InSAR-derived land subsidence map for the 2015–2016 season (Figure 11a) reveals a marked spatial heterogeneity across the Ouled Gaid irrigated perimeter. Maximum subsidence rates, locally reaching 6.5 mm.yr −1, are concentrated in the north part of the perimeter and in isolated clusters across the central-eastern domain. Conversely, the southern sector exhibits substantially lower subsidence values (< 3 mm·yr −1). The groundwater abstraction map (Figure 11b) displays a spatially coherent pattern, with extraction fluxes exceeding 80–100 mm·yr −1 in the northern and north-central sectors. This spatial distribution is consistent with the dominance of water-demanding horticultural and arboricultural crops identified in the land use map.

The spatial overlap map (Figure 11c), generated by thresholding both distributions at the 30th percentile (≥ 3 mm·yr −1 for subsidence; ≥ 0.73 mm for abstraction), reveals a co-occurrence rate of 70.6% between high-abstraction and high-subsidence zones. This high spatial correspondence indicates a strong coupling between groundwater extraction and InSAR-detected land subsidence across the perimeter. The spatial dominance of overlap areas (green pixels) in the north and central sectors confirms the predominantly anthropogenic origin of the observed subsidence, driven by agricultural groundwater pumping.

Pixels classified as high pumping only (blue), specially located in the central and eastern parts of the perimeter, attributed to the presence of rural settlements (douars) and built-up areas, as well as the prevalence of olive trees and arboricultural plantations (Figure 1). These surface features with dense canopy cover and anthropogenic structures act as decorrelation sources in SAR imagery, limiting the InSAR-SBAS ability to detect ground deformation despite significant groundwater abstraction in these sectors.

Given the geological homogeneity of the study area (Middle Quaternary deposits dominated by a rubified soil with scattered pebbles and an encrusted silty-clay matrix), the observed spatial variability in subsidence rates cannot be attributed to lithological contrasts. Collectively, these results demonstrate that land subsidence within the Ouled Gaid perimeter is primarily driven by groundwater abstraction, notably in the north-western and central sectors. The 70.6% spatial co-occurrence between high-abstraction and high-subsidence zones confirms that InSAR-SBAS land subsidence constitutes a reliable proxy for monitoring groundwater overexploitation, offering a spatially continuous alternative to sparse piezometric networks in semi-arid irrigated plains.

InSAR subsidence-groundwater abstraction coupling under dry and wet years

The spatial Pearson correlation between the Sentinel-1 InSAR-derived vertical land motion (VLM) and the estimated groundwater abstraction over the Ouled Gaid study area ranges from 0.39 to 0.71, with a mean value of 0.52 (Figure 12). This correlation was computed on the full VLM distribution, including both subsiding and uplifting pixels, and not on the negative (subsidence) part only. In addition, the analysis was repeated using only the pixels that pass the conservative coherence mask (γ ≥ 0.30), and the correlation values remained of the same order, which indicates that the result is not controlled by low-quality pixels. These values show a consistent, moderate spatial association between the two independently derived fields. However, following the arguments presented above, this association is interpreted as spatial consistency between abstraction and land subsidence; a necessary but not sufficient condition for a causal link; and not as proof that groundwater pumping is the dominant driver of the observed deformation.

Several factors can explain why the correlation is moderate rather than strong. First, land subsidence depends not only on pumping intensity, but also on hydrogeomechanical controls that are not included in this analysis, such as aquifer compressibility, sediment thickness, effective stress evolution, and delayed compaction, which can shift or smooth the deformation response in space and time. Second, the C-band coherence loss over densely cropped fields biases the VLM sampling towards field margins, tracks and bare soils, so the deformation signal is not measured exactly where abstraction is strongest. Third, there is a partial temporal mismatch between the single-season (2015–2016) abstraction estimate and the multi-annual deformation record, since subsidence is a cumulative response of the aquifer system rather than an immediate reaction to seasonal pumping. Finally, uncertainties in both input fields (crop coefficients, evapotranspiration, irrigation allocation, and InSAR displacements) further reduce the measurable spatial correspondence. Taken together, these factors mean that a moderate correlation is the expected outcome, even in a situation of active and spatially organized abstraction.

Correlation values are systematically higher during dry years determine using the standardized precipitation index (SPI) (McKee et al., 1993) (SPI < 0: 2015–2016, 2018–2020), reaching up to 0.59–0.71, reflecting intensified groundwater abstraction as the dominant irrigation source when surface water availability collapses and the allocation from the canal network is substantially reduced. Conversely, wet years (SPI ≥ 0: 2017, 2021–2022) coincide with notable drops in r (down to ~0.39–0.42), During wet years, multiple converging mechanisms contribute to reducing the spatial correspondence between pumping zones and active subsidence. Abundant precipitation provides a direct supplementary water source for irrigated agriculture, while simultaneously increasing dam filling rate, thereby reducing dependence on groundwater abstraction for irrigation supply. Concurrently, enhanced aquifer recharge during wet periods, reflected by piezometric level rises of up to 1.13 m recorded across the Haouz plain (Hajhouji et al., 2022). consistent with reduced pumping pressure when aquifer recharge and reservoir inflows partially substitute well extraction. This inverse relationship between SPI and the subsidence-abstraction correlation provides observational evidence that precipitation variability is a primary driver of the interannual modulation of subsidence patterns across the Haouz irrigated plain.

Vegetation cover is an additional factor limiting the InSAR measurement itself: C-band signals decorrelate over densely cropped fields, so the VLM field is undersampled precisely where irrigation—and thus abstraction—is concentrated. A vegetation correction was not applied here because no robust method exists to recover the deformation signal under low-coherence canopy at C-band; instead, unreliable pixels were masked (γ ≥ 0.30) rather than interpolated. This structural limitation, and the use of L-band sensors (ALOS-2, NISAR, ROSE-L) as the appropriate solution for vegetated perimeters, is further discussed in the Limitations section.

The results of our study reveal a structural dependence of the farming systems in the Ouled Gaid semi-arid irrigated perimeter on multiple and complementary water sources, including rainfall, dam allocations and groundwater abstraction. The combined dynamics of these sources determine the water supply available for agriculture.

Rainfall is the most unpredictable component, characterized by high spatiotemporal variability. It remains largely insufficient with an annual total of 154.8 mm. Monthly rainfall ranged from complete absence to a maximum of 50 mm recorded in March, with a monthly average of 12.9 mm. These figures confirm the semi-arid nature of the study area and underline the structural inadequacy of rainfall alone to sustain agriculture.

Dam allocations represented the primary regulated surface water input, with a cumulative annual volume of 14.53 × 10 6 m 3 and a mean monthly release of 1.21 × 10 6 m 3.month −1. Monthly volumes fluctuated between a minimum of 0.47 × 10 6 m 3 in February and a maximum of 1.51 × 10 6 m 3 in October, reflecting a seasonal pattern with relatively high allocations in autumn, declining through winter, and progressively increasing in spring and summer in response to rising crop water demand. This variability primarily depended on inter-annual variations in rainfall and was driven by dam reservoir filling rates, highlighting the inherent dependency of surface water availability on climatic inputs and reservoir management decisions.

Groundwater abstraction emerged as a complementary and adaptive resource, compensating for the deficit between surface water availability and actual crop water requirements. Total annual groundwater abstraction was estimated at 4.41 × 10 6 m 3, corresponding to an average of 344,515 m 3 per month. Groundwater withdrawals ranged from 4,800 m 3 in March to a peak of 1.07 mm 3 in July. Pumping volumes remained comparatively low during autumn and winter, coinciding with humid periods characterized by high rainfall and reduced irrigation demand, before increasing steadily in spring and reaching maximum intensity during summer.

Crop water requirements were estimated based on values from previous studies conducted in the study area and from institutional references of the regional irrigation water management authority. Yielding a total of 17.87 × 10 6 m 3.yr −1, equivalent to 8,097 m 3.ha −1.yr −1, consistent with the range of variation reported for the Tensift region. Irrigation demand was met through a combination of rainfall, dam allocations and groundwater abstraction. Effective rainfall contributed approximately 3.06 × 10 6 m 3.yr 1, while allocations from the Moulay Youssef dam, adjusted for canal conveyance efficiency, supplied 11.61 × 10 6 m 3.yr −1. Groundwater abstraction added approximately 4.13 × 10 6 m 3.yr −1, resulting in a total water supply of around 18.8 × 10 6 m 3.yr −1 at the perimeter scale, equivalent to an average of 8,519 m 3.ha −1.yr −1. This total closely matches the estimated crop water requirements of 17.87 × 10 6 m 3.yr −1.

The results of our study show that water resources management is conditioned by agronomic, hydrological, and socio-economic factors. Agronomic conditions depend on the types of crops cultivated and their phenological stages, resulting in temporally variable water requirements for the same plot. Hydrological conditions mainly depend on the availability of surface water resources, notably rainfall, which exhibits strong spatiotemporal variability in semi-arid contexts. Water supply from the dam is constrained by both reservoir filling rates and collective management rules within the traditional irrigation canal network (seguias), including rotational irrigation schedules, in addition irrigation scheme preferentially favors water supply to upstream parcels at the expense of those located downstream. Socio-economic conditions shape irrigation strategies, particularly the limited financial capacity of smallholder farmers, which affects their ability to abstract groundwater, including access to energy sources (electricity or diesel) necessary for pumping. Under these constraints, groundwater abstraction plays a complex and critical role in bridging the gap between surface water availability and actual crop water requirements. It serves as a complementary resource during sensitive growth stages and periods of water scarcity when surface water allocations are insufficient to meet irrigation demand, ensuring a continuous water supply. Groundwater serves as a safety valve, providing a water source that is directly controlled by farmers, consistently available, readily accessible, and largely independent of climatic variability and institutional constraints.

Groundwater abstraction estimated using remote sensing approach was validated against field measurements from monitoring wells irrigating a total area of 250 ha. For the period from April to August 2016, field data indicated that surface water supplied from the dam accounted for 55% (343,800 m 3) of the total water supply, while groundwater abstraction contributed 45% (281,300 m 3). In comparison, results obtained using the remote sensing approach over a larger area of 2,361 ha estimated the dam contribution at 63.9% (5,852,315 m 3) and groundwater pumping at 36.1% (3,291,045 m 3).

The underestimation of groundwater abstraction contribution by the remote sensing approach can be attributed to two complementary factors. First, the remote sensing methodology inherently integrates spatial variability across the entire agricultural perimeter, covering areas with adequate surface water supply where reliance on groundwater abstraction is consequently limited. This spatial averaging effect reduces the relative weight of groundwater abstraction estimated at the perimeter scale compared to localized field measurements. Second, the agronomic heterogeneity of the perimeter ranging from water-intensive to water-efficient crops, introduces variability in irrigation water requirements and in the degree of local dependence on groundwater. Conversely, field measurements remain susceptible to spatial sampling bias, as monitoring points may disproportionately represent hydraulically disadvantaged areas with water-intensive crops (olive trees and fodder crops), potentially leading to an overestimation of groundwater abstraction contributions at the local scale.

Despite these quantitative differences, both measured and estimated groundwater abstraction converge toward a consistent structural conclusion, dam water supply is the primary source of irrigation, while groundwater pumping represents a significant supplementary contribution ranging from 36% to 45% depending on the scale of analysis. This agreement confirms the overall reliability of the remote sensing approach for large-scale water accounting, while highlighting the indispensable role of field measurements for local calibration and model validation.

However, beyond its numerical contribution, groundwater abstraction is a strategic resource whose functional importance far exceeds its quantitative share. It becomes indispensable when surface water sources are unavailable or insufficient, particularly during critical phenological stages and drought episodes, when water supply reliability becomes crucial to safeguard agricultural production.

The InSAR-derived land subsidence reveals marked spatial heterogeneity across the Ouled Gaid irrigated perimeter. Maximum subsidence rates, locally reaching 6.5 mm, are concentrated in the northern part of the perimeter and in isolated clusters across the central-eastern parts dominated by water-intensive crops, while the southern sector exhibits substantially lower values. The spatial overlap analysis yields a co-occurrence rate of 70.6% between high-abstraction and high-subsidence zones.

InSAR time series analysis over the 2015–2022 period reveals a consistent spatial coupling between land subsidence and groundwater abstraction, with Pearson correlation coefficients ranging from 0.39 to 0.71 (mean r = 0.52). The interannual variability of this coupling, modulated by SPI-derived hydroclimatic conditions, demonstrates that dry years (SPI < 0) amplify the subsidence-abstraction correspondence through intensified groundwater dependency (r up to 0.71), while wet years (SPI ≥ 0) attenuate it via aquifer recharge and reduced pumping pressure.

Despite the influence of confounding factors including spatial heterogeneity in well density, variability in aquifer transmissivity, seasonal vegetation cover, and the presence of rural settlements and built-up areas affecting InSAR coherence, both the spatial overlap results and the interannual variability of the Pearson correlation consistently confirm the predominantly anthropogenic origin of the observed subsidence, driven by agricultural groundwater abstraction. InSAR detected land subsidence thus constitutes a reliable proxy for monitoring groundwater overexploitation at the irrigated perimeter scale.

This study provides a multifaceted contribution to the understanding of groundwater dynamics in water-stressed semi-arid environments. By integrating remote sensing-derived estimates with available field observations, it addresses a critical knowledge gap regarding groundwater pumping in semi-arid regions of Morocco, where the absence of systematic pumping records has long impeded reliable water balance assessments. Beyond its regional application to the Tensift basin, the proposed methodological framework, combining spatio-temporal abstraction mapping with land subsidence as a qualitative proxy for pumping activity, constitutes a transferable and scalable approach applicable to comparable irrigated perimeters in data scarcity environment.

The relationship between groundwater abstraction and land subsidence is not strictly causal: subsidence magnitude is also strongly modulated by others factors that are independent of pumping intensity itself. Land subsidence should therefore be regarded as an indirect indicator of groundwater abstraction rather than a direct, quantitative proxy. This is consistent with our own findings, where a significant spatial overlap between abstraction and subsidence hotspots. Building on this established but indirect relationship, the novelty of the present approach lies in combining InSAR-derived subsidence with a simplified remote sensing water balance to quantitatively estimate groundwater abstraction volumes rather than simply detecting subsidence hotspots as a qualitative proxy for pumping activity, as in most prior studies.

Several limitations of the present study should be acknowledged. Groundwater abstraction was validated against field measurements from 14 monitoring wells over 2,361 ha, providing an independent assessment of the proposed estimates; however, the limited and field-constrained sampling density may affect their statistical representativeness. The simplified water-balance approach does not explicitly account for percolation, although its contribution is expected to be relatively limited given the clay-loam soils and deep water table (>80 m). Finally, the relationship between groundwater abstraction and land subsidence is not strictly causal, as subsidence is also influenced by local geological and hydrogeological conditions, including aquifer compressibility and the presence of compressible fine-grained layers. Therefore, land subsidence should be interpreted as an indirect indicator of groundwater abstraction rather than a direct quantitative proxy. Estimating groundwater abstraction remains challenging in areas of intensive pumping because of the heterogeneity of wells and pumping practices (Zekri et al., 2015; Xu et al., 2017; Mustafa et al., 2018). In this study, the uncertainty associated with the estimated groundwater abstraction; the central component of the proposed approach; was empirically assessed by comparing the estimates with field measurements from monitoring wells. The resulting RMSE was 12.63 mm, corresponding to approximately 20.5% of the mean measured value. This result provides support for the reliability of the proposed remote-sensing-based estimation method, while also indicating the uncertainty associated with well-scale estimates.

A formal probabilistic propagation of uncertainty was not performed because empirical error distributions and covariance information were unavailable for several historical datasets. Independently applying generic uncertainty percentages to these data could have produced confidence intervals that appeared quantitative but were not supported by observations. Therefore, uncertainty was assessed using empirical data and complementary sensitivity analyses.

In the absence of dedicated ground-based monitoring networks enabling equivalent site-specific validation, indicative uncertainty values were assigned to the other water-balance components. An uncertainty of ±5% was assigned to the instrumental component of precipitation measurements, based on the achievable uncertainty reported by the WMO for liquid precipitation under specified operating conditions (WMO-No. 8, 2018). This value does not account for spatial representativeness, gauge exposure, wind-induced undercatch, spatial interpolation, or other systematic errors. For dam-released irrigation water, an uncertainty of ±10% was assigned to the annual released volume as an indicative bound for volumes derived from well-calibrated stage–discharge rating curves under median-flow conditions (Kiang et al., 2018). This value should be regarded as a pragmatic literature-based assumption rather than as a site-specific uncertainty estimate.

Additional empirical checks provided further context for the uncertainty of the estimates. For the period from April to August 2016, the satellite model estimated the contributions of groundwater and dam water at 36.1% and 63.9%, respectively, compared to 45% and 55% for the monitored plots. The difference in the groundwater contribution was thus 8.9 percentage points. This comparison indicates the potential magnitude and direction of the discrepancy, but is not considered a formal estimation error, as the monitored plots covered approximately 250 ha, while the remote sensing estimate represented the entire perimeter of 2,361 ha. At the perimeter scale, the reconstructed water input was 18.80 × 10 6 m3, compared with estimated crop water requirements of 17.87 × 10 6 m 3, corresponding to a relative difference of 5.2%. This comparison provides an indication of water-balance closure at the perimeter scale, but does not constitute a confidence interval or a formal uncertainty bound for groundwater abstraction.

For the InSAR component, sensitivity to measurement quality was assessed using two consistency thresholds, while sensitivity to the spatial overlap between withdrawal and subsidence was examined using percentile thresholds ranging from 10% to 50%. These analyses evaluated whether the observed spatial correspondence was strongly dependent on a particular consistency mask or classification threshold. Land subsidence derived from InSAR was not included as an additive water-balance component because it was used as an independent indicator of groundwater-system response.

The main remaining uncertainties concern the Kc-NDVI relationship, reference evapotranspiration, the spatial representativeness of precipitation, the conversion of dam allocations to water actually distributed to plots, irrigation efficiency, and the omission of variations in runoff, deep percolation, and water storage in the root zone. Accordingly, the estimated groundwater volume should be interpreted as a preliminary estimate of additional groundwater irrigation requirements rather than a direct measure of gross withdrawal from the aquifer.

The present study conducted in a semi-arid area characterized by limited water resources and dominated by smallholder farmers, in the Ouled Gaid irrigated perimeter, groundwater abstraction constitutes a critical and adaptive component of the agricultural water supply, bridging the structural gap between surface water availability and crop water requirements. Total annual groundwater abstraction reached 4.13 × 10 6 m 3.yr −1, with monthly withdrawals ranging from 4,800 m 3 in March to a peak of 1.07 × 10 6 m 3 in July, reflecting a strong seasonal pattern aligned with climatic conditions and crop demand. Within the overall water supply structure, combining rainfall, dam allocations, and groundwater abstraction, the total water input reached 18.8 × 10 6 m 3.yr −1 at the perimeter scale, equivalent to an average of 8,519 m 3.ha −1.yr −1, closely matching estimated crop water requirements estimated at 17.87 × 10 6 m 3.yr −1. Groundwater contributed 36.1% of this supply, functioning as a safety valve that is directly controlled by farmers, consistently available, and largely independent of climatic variability and institutional constraints, particularly during sensitive growth stages and drought episodes when dam allocations prove insufficient. The InSAR-derived land subsidence analysis further reveals marked spatial heterogeneity across the perimeter, with maximum subsidence rates locally reaching 6.5 mm.yr −1 concentrated in the northern part of the perimeter and in isolated clusters across the central domain, The overlap analysis reveals a 70.6% spatial co-occurrence between high-abstraction and high-subsidence zones, while Pearson correlation coefficients between land subsidence and groundwater abstraction over the 2015–2022 period range from 0.39 to 0.71 (mean r = 0.53).