Groundwater recharge estimation in data-limited water-scarce regions Open Access

Ethiopia faces significant challenges in water resource management, primarily due to its complex hydrological dynamics. A key factor contributing to these challenges is the uneven distribution of rainfall across the country, with the highlands receiving more precipitation compared to the lowlands (Funk et al. 2012). This disparity in rainfall leads to a high degree of water variability, often resulting in extreme and recurrent droughts and floods, severely impacting both agricultural productivity and water availability (Conway 2000; Nafchi et al. 2022). Moreover, Ethiopia's rapidly growing population intensifies the water demand, putting additional strain on already limited resources. Alongside drinking water security, one of the most critical implications of this water scarcity is its impact on food security. A recent study by the World Bank alarmingly noted that in 2022, it was estimated that approximately 20.4 million people in Ethiopia were acutely food insecure, desperately requiring assistance (World Bank 2022).

In Ethiopia agriculture has been traditionally rainfed; however, in recent years climate change has significantly contributed to making rainfall patterns increasingly unreliable (Li et al. 2023). A direct consequence of this is the inability to consistently cultivate crops, which together with a growing population has underlined the urgent need for effective water management strategies (Gebul 2021). One promising approach to alleviate water stress and concurrently address hunger issues is via the sustainable utilization of groundwater resources through shallow well drilling for irrigation (Li 2024; Li et al. 2024). However, the key to ensuring the sustainability of such practices lies in a thorough understanding of groundwater recharge. The estimation of groundwater recharge rates is essential to prevent over-exploitation of this precious resource and to guarantee that groundwater abstraction remains within sustainable limits, thereby fostering long-term water security and agricultural productivity in Ethiopia. Additionally, an effective understanding of recharge may also lead to better decision-making, regarding where not to irrigate in favor of a clear focus on dryland agriculture, using organic techniques. Further, knowing the recharge situation can inform managed aquifer recharge (MAR) projects to replenish depleted aquifers, often by surface water injection, to help support the natural environment. This may include reforestation, but also entertains the possibility of remediating an aquifer, perhaps to allow sustainable abstraction to resume in the future (Ostad-Ali-Askari & Shayannejad 2021; Mann 2023).

There are various methods for estimating groundwater recharge, including the water table fluctuation (WTF) method, chemical and isotopic methods, and soil water balance models. These methods come with inherent uncertainties, primarily due to the complex and often hidden nature of groundwater systems. This intrinsic uncertainty in estimating groundwater recharge poses a significant challenge for validation. As a result, multiple methods are often employed to provide a broader, yet still uncertain, understanding of groundwater behavior. Despite these limitations, insights from these methodologies are invaluable for guiding sustainable groundwater use, particularly in agricultural planning and management.

The WTF method is a widely used approach for estimating groundwater recharge by observing changes in the groundwater level (Healy & Cook 2002). This method assumes that rises in the water table, following precipitation events, are directly indicative of recharge. While it offers a relatively straightforward and cost-effective means of estimation, the WTF method has inherent limitations. It often assumes a homogeneous aquifer and consistent specific yield, which may not hold true in varied geological settings (Scanlon et al. 2002). Additionally, this method may not accurately capture recharge over longer timescales or in areas with deep water tables. To address these limitations, recent studies have explored the integration of machine learning techniques with traditional methods like WTF. Machine learning can enhance the analysis of hydrological time-series data, potentially leading to more accurate recharge estimations by capturing complex, non-linear relationships in the data that traditional methods might overlook (Pourghasemi et al. 2020). However, the effectiveness of machine learning in this context is highly dependent on the quality and quantity of available training data, and its integration with hydrogeological understanding.

Chemical and isotopic methods are pivotal in groundwater recharge estimation, offering detailed insights into the dynamics of aquifer systems. The chloride mass balance (CMB) method, utilizing chloride as a conservative tracer, calculates recharge based on the chloride concentration in precipitation and groundwater (Wood & Sanford 1995). This method is particularly effective in arid and semi-arid regions where chloride inputs and outputs can be clearly delineated. Isotopic methods, including age tracers like tritium, deuterium, and oxygen-18, help in understanding the time scales of groundwater movement and recharge. These tracers provide valuable information on the origin and age of groundwater, allowing for more accurate modeling of recharge processes (Clark & Fritz 1997). However, these methods are not without limitations. Data availability is a significant constraint, especially in remote or under-studied regions where obtaining consistent and widespread data samples can be challenging (Scanlon et al. 2002). Moreover, the accuracy of these methods relies heavily on precise and reliable measurements of tracer concentrations, which necessitates sophisticated and relatively expensive laboratory equipment and techniques. The representativeness of samples is another concern; samples collected may not adequately represent the entire study area, potentially leading to inaccuracies in recharge estimation. Despite these limitations, chemical and isotopic methods remain critical in groundwater studies, offering valuable insights into hydrological processes and contributing significantly to the understanding and management of water resources.

The soil water balance model is a widely utilized approach for estimating groundwater recharge, based on the principle of balancing the inputs and outputs of water within a soil profile. This model calculates recharge by accounting for precipitation as input and factors such as evapotranspiration, runoff, and changes in soil moisture as outputs (Healy & Cook 2002). One significant advantage of this model is its direct utilization of measurable meteorological and soil data, allowing for adaptability across diverse environmental settings (Scanlon et al. 2002). Moreover, it can provide localized recharge estimates by incorporating specific site characteristics. However, the model's accuracy is contingent on the precision of each component's estimation. A notable limitation arises from the cumulative errors associated with measuring or estimating these components, especially evapotranspiration and soil moisture changes, which are often challenging to quantify accurately (Sophocleous 1991). The process of sequentially subtracting outputs (like evapotranspiration and runoff) from inputs (precipitation) can amplify and accumulate biases, leading to less accurate recharge estimates. This issue is particularly pronounced when applying the model over large or heterogeneous areas, where spatial variability in soil properties and meteorological conditions can further exacerbate the inaccuracies (Healy 2010). Furthermore, the model's reliance on detailed local data can be a constraint, as such data may not be readily available or may require extensive field measurements, making the model less feasible in some regions. Therefore, while the soil water balance model is a valuable tool in hydrological studies, its application necessitates a careful assessment of the available data quality and an understanding of the potential error sources in the estimation of each water balance component.

The rest of the paper is organized as follows. Section 2 describes the study area, data sources, and methodology used in this study; Section 3 shows the results of groundwater recharge estimation, a temporal and spatial analysis of the estimated recharge and precipitation, and the relationship between LCLU and estimated recharge; Section 4 provides the comparative model validation with existing literature; Section 5 concludes the paper.

The agricultural practices in the Rift Valley region predominantly follow an agro-pastoralism system, a combination of crop production and pastoralism. This integrated approach combines crop cultivation with livestock rearing, fostering a diversified and sustainable agricultural model. As reported by the US Department of Agriculture (USDA 2016), key crops cultivated in the basin from 2011 to 2016 encompass barley, corn, sorghum, and wheat. This synergy of farming and herding optimizes resource utilization and enhances the resilience of the agricultural sector in the region.

In this study, the data product we used to estimate the groundwater recharge comes from NASA's Global Land Data Assimilation System (GLDAS), which is a land surface modelling framework that integrates observational data with LSM simulations to provide detailed insights into soil moisture, evapotranspiration, and energy fluxes. This system combines satellite and ground-based data with hydrological and meteorological models, improving our understanding of the Earth's water and energy cycles for applications in weather forecasting, drought analysis, and climate research. There are three main LSMs from GLDAS including the community land model, the Noah model, and the variable infiltration capacity model (Rodell et al. 2004). In this research, the monthly data from GLDAS Noah model version 2.1 with a spatial resolution of 0.25° × 0.25° from 2000 to 2022 was used for groundwater recharge estimation and analysis.

The variables we used and extracted from GLDAS Noah include surface runoff, subsurface runoff, and precipitation (Table 1). According to the GLDAS data document, surface and subsurface runoff, as accumulated variables, are the average 3-h accumulation over all 3-h intervals for a particular month. Precipitation represents the average 3-h mean rate overall 3-h intervals for a particular month. To make the units consistent, the runoff data and precipitation have been converted into monthly accumulated data with a unit of kg/m²/month. Based on the guidance in the GLDAS data document, the average 3-h accumulated surface and subsurface runoff have been converted to monthly accumulated data by multiplying 8 times the number of days in a particular month. In addition, the average 3-h mean rate of precipitation has been converted into monthly accumulated data with the same units of kg/m²/month. Note that the uncertainties in runoff estimations of the GLDAS products have been reported in a previous study (Qi et al. 2020). These uncertainties may influence the absolute magnitude of recharge estimates, but we believe the general trends and spatial patterns remain robust for regional-scale analysis.

For a more granular analysis, each of these three regions was further subdivided into two, resulting in a total of six distinct regions. In the northern area, the grid cells near the western boundary exhibited a mix of cropland, shrubland, tree cover, and built-up areas. These cells, showing a similar pattern on the map, were aggregated and designated as region 1. The remaining cells in the northern area were classified as region 2. In the middle area, the eastern section, primarily consisting of tree cover and shrubland, was aggregated as region 4, while the other cells near the western boundary were grouped into region 3. The western portion of the southern region, characterized by grassland, shrubland, bare or sparse vegetation, and herbaceous wetland, was labeled as region 5. The eastern part of the southern region, comprising mainly shrubland, tree cover, and grassland, was identified as region 6.

The simplicity and ease of application of the LSM model stand out as its primary advantages, unlike the water balance models, which involve the subtraction of multiple variables and are prone to cumulative errors and biases. The LSM model offers a straightforward method for estimating groundwater recharge, especially valuable for estimation in large regions. By balancing simplicity with accuracy, this model provides a pragmatic approach for hydrologists and environmental scientists, especially in scenarios where detailed data might be limited or in large watershed analyses where a more complex modeling approach might not be feasible. Moreover, the model's adaptability to varying proportions of runoff types makes it a flexible tool for different hydrogeological settings.

A positive value of indicates that the data tends to increase with time, while the negative indicates a decreasing trend. If the value of Z is less or greater than the value of ⁠, where is percentile of the standard normal distribution, the null hypothesis is rejected at α significance level indicating that there is a significant upward or downward trend of the variable over time. In R, the ‘Kendall’ package is utilized to conduct the test. The output of this test comprises a tau value and a p-value. Tau, derived from the S statistic, is scaled to range between −1 and 1, representing the strength and direction of the observed trend. The p-value, on the other hand, is used to assess the statistical significance of the results. The Mann–Kendall test is particularly useful in environmental and climatological studies for its ability to handle non-normal and censored datasets, offering a reliable tool for trend analysis in these fields.

These spatially explicit recharge estimates can play a critical role in informing groundwater management policies in Ethiopia. By identifying areas with varying recharge capacities, such as the south eastern Rift Valley, policymakers can be guided by multidisciplinary teams towards sustainable water development. Recognizing this need, government universities like the Arba Minch University Water Technology Institute, and organizations such as Global MapAid, are planning ‘Sustainable Knowledge Centres’ (SKCs) across Ethiopia. These SKCs will be instrumental in prioritizing effective water capture, conservation, and utilization strategies on small farms, and in developing targeted training programs for farmers to implement these practices. Moreover, the model can support the identification of optimal locations for MAR interventions, helping to enhance groundwater storage during periods of surplus. These insights can also contribute to broader climate adaptation planning by promoting resilience in water-scarce regions through data-driven resource allocation and long-term sustainability strategies.

In this section, we present the findings from our analysis of temporal and spatial variability in precipitation and estimated recharge. The discussion is organized into subsections that cover the Rift Valley at large, its six constituent regions, and individual grid cells within the Valley. The first two subsections concentrate on examining autocorrelation and trends, as well as exploring the relationship between precipitation and estimated recharge. For the final subsection concerning grid cells in the Rift Valley, rather than detailing autocorrelation and trends for each cell, we instead highlight the conversion rates of rainfall into groundwater recharge across the grid cells, visualized through a map.

Similarly, we examined the auto-correlation of the monthly estimated recharge and precipitation data for the six distinct regions in the Rift Valley, as illustrated in Figures S1 and S2 in the Supplementary material. The outcomes revealed wave-like patterns in these regions, indicating a degree of seasonality in the data. Conversely, when analyzing the yearly estimated recharge and precipitation, no significant autocorrelation was observed (Figures S3 and S4 in the Supplementary material).

Given the absence of autocorrelation in the yearly data, the Mann–Kendall trend test was conducted to analyze the annual trends within the six regions. For precipitation, statistically significant downward trends were detected in regions 1 and 4. For estimated recharge, statistically significant downward trends were detected in regions 1, 2, 4, and 6. The tau values and p-values, which quantify the trend direction and strength, and statistical significance, respectively, for both precipitation and estimated recharge, can be found in Table 3a. In light of the observed significant downward trends, Sen's slope was utilized to quantify the magnitude of these trends. The Sen's slope estimates for precipitation in regions 1 and 4 are −0.87 and −0.94, respectively, as detailed in Table 3b. These results demonstrate a pronounced and significant decline in precipitation within these areas. Furthermore, Sen's slope estimates for estimated recharge in regions 1, 2, 4, and 6 are −0.33, −0.07, −0.13, and −0.04, respectively, signifying a moderate to mild yet statistically significant reduction in recharge across these regions. The downward trends can also be observed and verified in the STL in Figures S5 and S6 in the Supplementary material.

To better understand the relationship between the estimated recharge and precipitation, Spearman's rank correlation coefficients were determined, alongside the conversion percentages of rainfall to groundwater recharge for each region. The correlation analysis yields coefficients of 0.88, 0.85, 0.8, 0.89, 0.69, and 0.76 for Regions 1 through 6, respectively, indicating a relatively strong association between precipitation and estimated recharge across the regions. The conversion percentages from rainfall to groundwater recharge are found to be 6.2, 2.5, 8.8, 3.3, 6.7, and 2.1% for these regions. The western regions of the Rift Valley, encompassing regions 1, 3, and 5, demonstrate better conversion rates compared to their eastern area. Our analysis revealed that the western regions exhibit a moderately dense, more diverse, and more evenly distributed LCLU pattern than the eastern regions. The higher recharge efficiencies in the west could be a result of more permeable soils, the beneficial characteristics of LCLU, or more effective land management strategies.

In examining the grid cells within the Rift Valley, our focus shifts from analyzing autocorrelation and trends within each cell towards a comprehensive understanding of precipitation patterns, estimated recharge, and the efficiency of precipitation-to-groundwater recharge conversion. Figure S7(a) in the Supplementary material displays the spatial distribution of monthly average precipitation across the grid cells, revealing higher precipitation rates centrally within the Rift Valley and along its northwest boundary, with moderate levels observed in the northern region. The northeast and southern regions are characterized by lower precipitation levels. Figure S7(b) in the Supplementary material presents the estimated recharge map, which delineates the conversion rates of rainfall into groundwater recharge for each grid cell. The color coding of each grid cell in Figure S7(b) visually represents the estimated recharge volume, while numeric percentages within each cell indicate the proportion of rainfall converted into groundwater recharge for that specific cell. We can observe that grid cells located near the northwest boundary in the north, along the western boundary in the center, and towards the southwest boundary in the south exhibit superior conversion rates compared to other regions, highlighting areas of enhanced recharge efficiency.

To analyze the relationship between estimated recharge from the LSM model and LCLU, we aligned the resolutions of their respective maps by rescaling the LCLU resolution from 10 m to match the 27.75 km grid cells. Subsequently, we employed k-means clustering in a quantum geographic information system (QGIS) to categorize the grid cells into 10 distinct LCLU classes (Figure S8(a) in the Supplementary material). These classes include NearlyAllTree, CropTreeShrubGrass (CTSG), CropTreeShrubGrassNearWater, ModeratelyTree (ModT), MostlyCrop (MC), MostlyShrub (MS), MostlyTree, ShrubGrass (SG), ShrubGrassTreeCrop (SGTC), and SparseVegetation (SV). This process provided each of the 103 cells in the Rift Valley with specific estimated recharge values and a reclassified LCLU category.

The initial exploration of this data is represented in a box plot (Figure S8(b) in the Supplementary material), which shows clear differences among the groups. Most notably, SGTC is much more variable and has a higher median than all the other groups. To rigorously test for significant differences in the estimated recharge medians among the 10 LCLU groups, we conducted a Kruskal–Wallis test. This non-parametric statistical approach tests the null hypothesis that the group medians are equal. Our analysis yielded a p-value of 1.577e-06, decisively lower than the 0.05 threshold for statistical significance, leading us to reject the null hypothesis. Furthermore, we executed a nonparametric Fligner–Killeen test to assess the homogeneity of variances, operating under the null hypothesis that the variances across sample groups are equal. The findings, yielding a p-value of 0.003, suggest a significant disparity in variances among the groups. These findings indicate significant variations in the estimated recharge medians and variability across different LCLU categories.

Furthermore, for the pairwise comparisons, we utilized the Wilcoxon rank sum test, another non-parametric method, to assess whether there is a difference in the median values between groups (Table S1 in the Supplementary material). This test does assume neither normality nor equal variances. The median estimated recharge for the MS group differs significantly from the medians of other LCLU groups. Additionally, the median estimated recharge for the CTSG group exhibits a significant difference with most of the groups except the SGTC and SV. The SGTC group's median is also significantly distinct from most other groups, with the exceptions of CTSG and SV. Lastly, the median for the SV group shows a significant difference in comparison to both MC and MS groups and a marginally significant difference compared to SG.

Validating estimated groundwater recharge poses a significant challenge, particularly in data-limited and water-scarce areas like the Rift Valley in Ethiopia. To assess the accuracy and reliability of our LSM model in this context, we conducted a comparative analysis with findings from existing literature, focusing on studies within similar geographical and climate features including Rift Valley Ethiopia, other regions in Ethiopia, and semi-arid regions in the world.

In 2019, Molla et al. conducted a study to quantify the shallow groundwater resources in the Abaya-Chamo basin, part of the Rift Valley in Ethiopia, using the WetSpass hydrologic model. Their findings indicated that approximately 9.7% of precipitation contributes to groundwater recharge in the basin. Moreover, Molla et al. compared their findings with four previous studies, demonstrating general consistency in the estimated groundwater recharge rates (Molla et al. 2019). In our study, which encompasses a broader region within the Rift Valley, including the Abaya-Chamo basin, we estimate that 5.2% of precipitation is converted to groundwater recharge. This percentage represents the average recharge calculated across four subregions within the Rift Valley (Regions 1–4). Although our estimate is lower than the 9.7% reported by Molla et al., both studies emphasize the limited groundwater recharge from precipitation in this semi-arid region. The differences may stem from variations in modeling approaches, spatial scales, or data periods, yet both results highlight the low recharge efficiency and the challenges of groundwater replenishment in the Rift Valley.

Dereje & Nedaw (2019) conducted a study to estimate groundwater recharge in the Upper Bilate catchment, part of the Ethiopian Rift Valley, using the WetSpass hydrologic model. Their findings indicated that approximately 9.4% of annual precipitation contributes to groundwater recharge in the catchment (Dereje & Nedaw 2019). They identified the western and northern areas as recharge zones, which aligns well with our results: the western cells (#11 and #19) show conversion rates of 7 and 5%, respectively, while the northern cell (#5) demonstrates a 12% conversion rate (Figures S7(b) and S8(a)). In our study, areas overlapping with the Upper Bilate catchment show that approximately 6% of precipitation contributes to groundwater recharge. While this is lower than the 9.4% estimated by Dereje and Nedaw, the overall findings remain generally consistent, highlighting the limited recharge potential in this semi-arid region.

Ferede et al. (2020) undertook a comprehensive study to estimate shallow groundwater recharge in the Eshito micro-watershed, located approximately 25 km northwest of Arba Minch city, in southern Ethiopia. This watershed, a relatively small area within the middle west of the Rift Valley, spans 70.4 ha, compared to the vast expanse of the Rift Valley basin, which covers about 13,142,300 ha. Utilizing methods like CMB, water-level fluctuation (WLF), and baseflow separation, their research suggested that 20–35% of annual rainfall contributes to recharge, according to CMB and WLF (Ferede et al. 2020). Their CMB results align closely with findings from a previous study, while their WLF results are consistent with those of two prior studies. Additionally, baseflow separation indicated varying recharge rates of 38% to 28% from upstream to downstream. They found that the baseflow separation results aligned with findings from four previous studies conducted in Ethiopia. In our study, we divided the Rift Valley into 103 grid cells, matching the GLDAS Noah data resolution. Our model estimated that 22% of rainfall contributes to groundwater recharge in the grid cell encompassing the Eshito watershed. This finding closely aligns with the CMB and WLF results of Ferede et al., although it shows a lower recharge rate compared to their baseflow separation estimates.

In 2021, Yifru et al. investigated the Meki Basin in the Main Ethiopian Rift Valley using the coupled SWATMODFLOW approach, which integrated the Soil and Water Assessment Tool and Newton Modular Finite Difference Groundwater Flow. The basic inputs include a digital elevation model (DEM), soil properties, climate (precipitation, temperature, humidity, solar radiation, and wind speed), slope classes derived from DEM, and LULC. Their results revealed that groundwater recharge accounts for roughly 10% of precipitation (Yifru et al. 2021). Our model applied to the same area, estimated the recharge to be about 7%, slightly lower but generally consistent with Yifru et al.'s findings.

In another relevant study, Demissie et al. (2023) employed the Water and Energy Transfer between Soil, Plants, and Atmosphere under quasi-Steady State (WetSpass) model to estimate groundwater recharge in the Upper Gelana Watershed of the South-Western Ethiopian Rift Valley. Their results indicated that about 7% of mean annual rainfall converts to groundwater recharge (Demissie et al. 2023). Additionally, they found agreement between their findings and five existing studies. In contrast, our LSM model estimated that only about 2% of rainfall leads to recharge in the same area, which is lower than 7%.

These comparisons with established studies in similar regions demonstrate that our estimates are in line with, but in general lower than, the estimates for the other regions in Rift Valley. The deviations can be attributed to methodological differences and the inherent uncertainty in hydrological processes. This comparison exercise suggests that our model provides conservative, but credible estimates of groundwater recharge, and also provides valuable insights for future refinement and application in similar water-scarce and data-limited regions.

In 2009, Zeleke and Merkel conducted a study to estimate the long-term average groundwater recharge in Dire Dawa, a semiarid region in central-east Ethiopia, covering an area of 1,333 km². They employed the WetSpass model, utilizing data on precipitation, potential evapotranspiration, temperature, wind speed, land use, elevation, slope, soil texture, and permeability. Their findings indicated that about 5% of the average annual rainfall (626 mm) contributed to groundwater recharge, amounting to 28 mm (Zeleke & Merkel 2009). They also found that their results align well with three earlier reports.

Later, in 2019, Kahsay et al. undertook a similar study in the Raya Valley of the Tigray region in northern Ethiopia, covering an area of 2,500 km². They used a spatially distributed water balance model (WetSpass) and prepared grid maps incorporating meteorological, hydrological, and geographical data. Their research concluded that approximately 8% of the average annual rainfall (710 mm) was converted into groundwater recharge (57 mm) (Kahsay et al. 2019). They also noted that their findings were consistent with two previous studies.

In the same year, Gebru and Tesfahunegn estimated groundwater recharge in the Illala catchment, also in the Tigray region, but with a smaller area of 340 km². They applied the CMB method, supported by geological field surveys and satellite imagery. Data collection included field surveys from 2010 to 2011, documenting wells, springs, runoff collection sites, and meteorological stations for rainfall sample collection. Additionally, Landsat Thematic Mapper Plus (ETM+) satellite images from January and February 2007 were used to delineate geological structures and units. They found that 11.7% of the rainfall in the area contributed to groundwater recharge (Gebru & Tesfahunegn 2019), a result consistent with findings from four previous studies. In the same Illala region, Teklebirhan et al. (2012) estimated groundwater recharge using the WetSpass model and determined that it accounts for approximately 12% of the annual precipitation (Teklebirhan et al. 2012).

Focusing on the Rift Valley located in southern Ethiopia, our model estimates that around 5% of the region's rainfall is converted into groundwater recharge. This rate aligns closely with Zeleke and Merkel's findings in Dire Dawa. When comparing our findings to the recharge rates observed in other Ethiopian regions like the Raya Valley and Illala catchment, our results seem both reasonable and conservative. This is especially pertinent considering the Rift Valley's larger geographic area and more diverse environmental conditions compared to the more localized study areas previously mentioned.

In 1996, Bazuhair and Wood conducted an estimation of groundwater recharge utilizing the chloride mass-balance technique within small alluvial aquifers located in the wadi systems of the Asir and Hijaz mountains in western Saudi Arabia. Their findings indicated that recharge rates were between 3 and 4% of total precipitation, aligning with recharge rates observed in other arid and semi-arid regions globally (Bazuhair & Wood 1996). Additionally, they found their results align well with four previous studies. This suggests that in arid or semi-arid environments, it is typical for only a small fraction of precipitation to contribute to groundwater recharge.

Ping et al. (2014) assessed groundwater recharge in the North Okanagan region, a semiarid mountainous area located in the southern interior of British Columbia, Canada, utilizing the CMB method. For this assessment, the chloride concentration in precipitation was calculated using historical data collected by Environment Canada. Additionally, the chloride concentration in groundwater, particularly that associated with surface water, was determined through an analysis of groundwater samples and findings from previous studies. The results of their study revealed a range of estimated recharge rates: in the valley bottoms, recharge was found to be between 10 and 16 mm/year, accounting for 1.1–1.9% of total precipitation, while in the mountain areas, it ranged from 7 to 11 mm/year, making up 1.8–2.7% of precipitation (Ping et al. 2014).

In 2017, Coelho et al. employed a combination of remote-sensing data and a geographic information system (GIS)-based water balance method to estimate groundwater recharge in the semi-arid Ipanema River Basin, situated in the Brazilian Northeast. Covering approximately 6,217 km², the study area presented unique challenges typical of semi-arid regions. Alongside this approach, they also implemented the WTF method, with a primary focus on evaluating the effectiveness of remotely sensed data in estimating groundwater recharge. The results of their study revealed that in 2011, 14.89% of the region's rainfall contributed to groundwater recharge as estimated by the WTF method, while the water balance approach indicated a slightly lower contribution of 13.53% (Coelho et al. 2017).

These findings provide valuable insights into the variability of groundwater recharge rates in different topographical areas within semiarid regions. Our analysis indicates that recharge rates in the Rift Valley range from 1 to 22% across different grid cells, with most areas near the eastern boundary showing only about 1% of precipitation converting to groundwater recharge. Conversely, select locations in the northwest, middle-west, and southwest of the Rift Valley exhibit higher percentages of precipitation contributing to recharge, ranging from 5 to 22%. This variability in recharge rates across the Rift Valley not only echoes the findings from similar semi-arid regions but also underscores the heterogeneity in hydrological processes within such terrains.

In this study, we developed a straightforward LSM model to estimate groundwater recharge in the large, data-limited, and water-scarce Rift Valley basin of Ethiopia. This model primarily utilizes publicly accessible data from the NASA GLDAS Noah product, offering a cost-effective solution for groundwater recharge estimation in expansive regions. Our motivation stemmed from the observation that areas suffering from water scarcity and limited data availability are often in dire need of developing groundwater resources for sustainable irrigation. To address the challenge of validating the model in such data-constrained environments, we compared our results with existing studies from areas with similar geographic and climatic characteristics. The findings indicate that our model tends to be in line with but generally lower than the WTF and baseflow separation methods. Nevertheless, our estimates align well, particularly with studies conducted within the same Rift Valley basin, demonstrating the model's applicability and relevance.

To the best of our knowledge, our framework for the estimation of groundwater recharge has not been proposed by previous studies in rainfed agricultural regions. A significant benefit of our LSM model lies in its simplicity, requiring only basic variables such as surface and subsurface runoff. This streamlined approach contrasts with more complex water balance models, which involve the subtraction of multiple variables and are prone to cumulative errors and biases. Moreover, our model stands out for its cost-effectiveness and practicality, especially in data-scarce and water-limited regions. Traditional methods like WTF, CMB, and baseflow separation often demand extensive data and expensive equipment. In contrast, our approach leverages readily accessible data and open-source software tools, making it an ideal solution for regions facing acute water shortages and limited data resources. These areas, particularly reliant on rainfed agriculture, can greatly benefit from our model in developing sustainable groundwater resources for agricultural use.

The primary limitation of this study lies in the challenge of model assessment. Although we have employed indirect validation methods, such as comparing our findings with previous research in regions with similar climate types, a more thorough validation of the model would be feasible with access to additional data. However, it is important to note that difficulties in assessing models are a common issue across all recharge estimation approaches, largely due to the inherently unobservable nature of groundwater recharge. This limitation underscores the need for more comprehensive data collection and the development of diverse estimation methods to enhance the accuracy and reliability of groundwater recharge models. Another limitation of this study is its reliance on GLDAS data as the primary data source for recharge estimation. While GLDAS provides comprehensive and globally consistent datasets, it has inherent constraints, such as potential biases in model outputs, resolution limitations, and uncertainties in representing local hydrological processes accurately. These factors can impact the precision of recharge estimates, especially in regions with complex terrains or unique hydrogeological characteristics not fully captured by the model.

Future research should concentrate on diversifying the methodologies for groundwater recharge estimation and comparing them with the proposed LSM model to better understand their strengths and weaknesses. Incorporating detailed geological analysis can improve model accuracy, particularly in regions with complex subsurface structures. Additionally, studying the impacts of climate change, including shifting weather patterns and extreme events, is vital as they significantly influence recharge rates. These efforts will enhance resource management and offer valuable guidance for sustainable water use and conservation amid environmental changes.

This work represents a collaboration among George Mason University, Arba Minch University, and Global Map Aid with support from the Czech Geological Survey. We appreciate the editor and the two anonymous reviewers for their constructive feedback.

This research was partially supported by a graduate research fellowship to W.L. from the Czech Geological Survey and George Mason University's Center for Resilient and Sustainable Communities.

W.L. and P.H. conceptualized the study, and W.L. wrote and edited the article. K.L., P.H., M.F., and R.D. reviewed and edited the article.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.