Spatiotemporal recharge estimation in the upper Awash sub-basin, central Ethiopia

Sustainable groundwater development requires estimating recharge that could be stored in groundwater aquifers (Collenteur et al. 2021; Talebmorad & Ostad-Ali-Askari 2022). Understanding the rate of groundwater recharge goes beyond water availability studies. Mapping recharge distribution helps to delineate wellhead protection zones, analyze surface-water interactions, evaluate the effects of urbanization, and determine aquifer vulnerability to contamination (Scanlon et al. 2002).

There are numerous methods used to assess groundwater recharge quantities such as experimental methods, hydrological budget (HB), empirical methods, and water table fluctuation (Rushton & Ward 1979; Alley et al. 2002). Nonetheless, due to the spatial and temporal variabilities of the factors that control groundwater recharge (i.e., lithology, lineaments, slope, land-use land cover, soil, and precipitation), recharge is the most difficult hydrological component to estimate (Muthuwatta et al. 2010). Especially in arid and semi-arid regions, recharge assessment is a key challenge in determining the sustainable yield of aquifers (Crosbie et al. 2010; Coelho et al. 2017; AL-Badry & S. Shamkhi 2021). Though there are shortcomings inherent to each recharge estimation technique, several groundwater recharge estimation techniques considering different environmental conditions affecting recharge distribution have been developed (Scanlon et al. 2002; Healy & Scanlon 2010).

The WetSpass model (Batelaan & De Smedt 2001) helps to estimate long-term spatial patterns of groundwater recharge which enabled it to address some of the main challenges in spatially distributed accurate groundwater recharge estimation (Armanuos et al. 2016; Dereje & Nedaw 2019; Meresa & Taye 2019). The WetSpass-M model is a modified version of the original WetSpass model. With the rising popularity of the Python programming language, the availability of refined input datasets, and advancements in remote sensing and Geographic Information System (GIS) capabilities, a new version-WetSpass-M has been developed to accommodate estimation at monthly time steps (Abdollahi et al. 2017). Unlike the WetSpass model which is only capable of estimating either seasonal or annual recharge (Kahsay et al. 2019; Meresa & Taye 2019; Dawana 2020; Zeabraha et al. 2020), the WetSpass-M model is developed to address the constraints of the seasonal model. It achieves this by scaling down water balance processes to monthly intervals, while also maintaining a simple model structure to minimize data needs. The WetSpass-M model is capable of estimating the monthly basis water balance components (Achraf et al. 2017; Ashaolu et al. 2020; Nannawo et al. 2021) such as surface runoff, evapotranspiration, and groundwater recharge. Recently, a few studies applied the WetSpass-M method to characterize and analyze groundwater recharge around the world, for example in West Africa, Black Volta Basin (Abdollahi et al. 2017), Croatia (Karlović et al. 2021), Iran (Zarei et al. 2016) in North Africa, Moulouya Basin (Amiri et al. 2022), in west Africa, Nigeria (Ashaolu et al. 2020), and Makutupora Basin, Tanzania (Kisiki et al. 2023).

The upper Awash sub-basin is of strategic importance in Ethiopia where about 70% of the industries and major urban settlements are located (Birhanu et al. 2021). Booming urbanization exerts negative effects on ground-surface hydrological processes at different spatial scales, land-use types, and water balance, such as surface runoff, groundwater recharge, and evapotranspiration (Zhang et al. 2017). As per the Central Statistics Agency of Ethiopia, 2018 projected population of the sub-basin is 12.43 million which is about 12.1% of the population of Ethiopia (Eyassu & Ababa 2022). There are massive surface and groundwater developments for domestic, industrial, and irrigation purposes and 65% of the urban water supply is from groundwater (Eyassu & Ababa 2022). Urban expansions with population growth are resulting in human-induced land degradation and increased surface water and groundwater abstraction (Legesse & Ayenew 2006; Ayenew et al. 2008).

Groundwater abstraction for domestic, industrial, and irrigation uses makes the upper Awash sub-basin one of the most important groundwater abstraction sites in Ethiopia. Because of the importance of the groundwater in the basin in supplying multiple water demands, there were attempts to estimate recharge in the region using the groundwater level fluctuation method, water balance method, chloride mass balance method, base flow separation method, and distributed Hydrus 1 D model (Yitbarek et al. 2012; Azagegn et al. 2015; Birhanu et al. 2021). However, the existing attempts to estimate recharge in the upper Awash River Basin provided aggregated results with a lack of a comprehensive understanding of the spatial distribution and seasonal variations. As a result, it is difficult to conduct site-specific or catchment-based water budget computations for sustainable groundwater planning and management decisions.

The novelty of this study lies in its application of the WetSpass-M model to estimate the spatial and seasonal distribution of groundwater recharge in the upper Awash sub-basin. Unlike previous studies that provided aggregated results and simplified assumptions or empirical relationships, this research aims to provide a comprehensive understanding of recharge variability by considering the heterogeneity of factors such as soil, land-use types, topography and different climatic variables. By closing the existing gap in understanding the distributed recharge of the upper Awash sub-basin seasonally and spatially, this study provides valuable insights for sustainable groundwater planning and management decisions. The study area exhibits different climate types, varying topography, and different soil and land-use parameters, making the WetSpass-M model suitable for this analysis that allows for a more accurate representation of the hydrological processes that influence recharge rates. We evaluated the WetSpass-M model accuracy in simulating key hydrological variables by comparing model outputs with observed data from gauging stations using independently computed recharge using the baseflow separation method. Additionally, we investigated the sensitivity of the model to local (slope factor, land-use factor, and soil factor) and global (interception parameter, Landscape Parameterization (LP) coefficient, alpha, and average intensity factor) parameters and evaluated its strength under different climatic conditions.

The geology of the study area primarily comprises a Scoriaceous Basalt unit, which is mostly composed of fresh to slightly weathered scoria and Scoriaceous basalts. The Rift Basalt unit can be found in the north around the Addis Ababa area, as well as in the central and northeastern regions of the study area. Additionally, Wechecha, Furi, and Yerer Trachyte, Entoto ridge, and Becho area Rhyolites are found in the mountains, while lacustrine and alluvial deposits are exposed in the southern part of the study area.

The WetSpass-M model requires two sets of model input parameters: meteorological data (rainfall, temperature, wind speed and PET) and physical data (land-use land cover, soil, and slope and groundwater depth). The preparation of the Wetspass-M model data, data resolution, and data sources are described as follows.

The mean monthly rainfall was compiled from 15 years (2000–2014) of daily rainfall records of 27 evenly distributed meteorological stations collected from the National Meteorological Agency (NMA) of Ethiopia. A total of 27 stations provided rainfall data, 21 stations provided maximum and minimum temperature data, and 10 stations provided wind speed data.

The study area has dry weather from October to February and May followed by a small rainy season from March to the end of April. The main rainy season is between June and September. The mean annual rainfall of the upper Awash sub-basin is 1,032 mm.

The minimum, maximum, and mean annual temperature collected from the NMA of Ethiopia are 13.68, 22.2, and 18.8 °C, respectively. The maximum mean monthly temperature is observed in April and the minimum mean monthly temperature is observed in December. The mean annual temperature increases from north to south (Figure 2(b)). The highest average monthly wind speed occurs in November, while the lowest average monthly wind speed occurs in September. The mean annual wind speed in the study area is 3 m/s which is collected from the NMA of Ethiopia.

The PET of the upper Awash Basin was analyzed using Thornthwaite's (1984) method. The minimum, maximum, and mean PET of the study area are 665, 1,027 and 846 mm, respectively. As shown in Figure 2(d), the central and southern parts of the study area have high potential evapotranspiration (PET) values, while the northern and northwestern parts have the lowest values. This difference is due to variations in elevation. Higher elevation areas have lower temperatures and PET, while lower elevation areas have higher temperatures and PET. The estimated mean annual actual evapotranspiration (AET) using this method is around 533 mm.

Since most of the study area is covered by cultivated land, an empirical equation of the Hargreaves and Samani method was employed for estimating PET. This method is widely used for irrigation planning and design (Hargreaves & Allen 2003). This method calculates PET based on the minimum and maximum temperature data and sunlight hours of the available meteorological stations. The Hargreaves evapotranspiration, calculated over a long period of time for the meteorological stations, has an average annual value of 609 mm. The WetSpass-M model utilizes PET estimates based on the Thornthwaite method.

The upper Awash sub-basin is densely populated and has intensive agricultural activity (WWDSE 2008). It is most intensively utilized and characterized by high urbanization (Birhanu et al. 2021). European Space Agency (ESA) global land cover product was used at 10 m resolution (Zanaga et al. 2021). These data are a new baseline product for 2020 based on Sentinel-1 and 2. To determine the water balance of the study area, each raster cell in the model is divided into fractions of the four basic land-use/cover types: impervious area, bare soil, open water, and vegetated area. All vegetation types including forests, bushes, woods, croplands, and others, are represented by the vegetation land-use land cover class. Exposed surfaces and bare soil types are represented by the land cover class known as ‘bare-soil land use.’ All surface water bodies, including lakes, rivers, and reservoirs, are represented by ‘Open water land use.’ In metropolitan areas and settlements with poor infiltration, human-made structures are the predominant types of impervious land cover. The land-use land cover classification of the study area resulted in six classes comprising cultivated land (82.6%), grassland (9.5%), forest (2.5%), water bodies (2%), shrub land and bushland (1.9%), and built-up area/exposed land surface (1.4%) (In order to account for land-use variability inside the cell, the model has its code for each land use or land cover, and land-use fractions are utilized as weighting factors for the computation of the water balance at the grid cell level (Abdollahi et al. 2017).

The topographic slope variability of the study area was prepared from the Advanced Space Borne Thermal Emission and Reflection Radiometer Digital Elevation Model (ASTER-DEM) with 30 m resolution. The northern, eastern, and western portions of the basin have a fairly steep slope, whereas the middle and southern portions have a gentle to flat slope (Figure 2(g)). From north to south, the elevation decreases by more than 1,500 m.

The spatial distribution of soil texture was sourced from Africa Soil Grids. Textural class (defined according to the USDA system) at 6-depth intervals derived from sand, silt, and clay contents was predicted using the Africa Soil Profiles Database (AfSP) v1.2 (https://data.isric.org/). The upper Awash sub-basin typically has soil textures including clay, loam, loamy sand, and sandy loam (Figure 2(h)). Clay predominates (65%), followed by loam (27%), sandy loam (7%), and loamy sand (2%). The area is covered by comparatively black cotton clay soil in the locations where the topography is plain to moderately sloping, i.e., the central and southern parts.

The groundwater level data for the parameter of groundwater depth were obtained from 214 shallow wells. The data were sourced from the Ethiopian Water Works Design and Supervision Enterprises (WWDSE). The upper Awash sub-basin is known to have a complex geological setting with multilayer aquifer systems (Yitbarek et al. 2012; Azagegn et al. 2015). Hence, a proper investigation of groundwater level data was done, and only those representing relatively shallow aquifers (<350 m depth) were used for this study. Similar to the metrological data sets, geostatistical ordinary kriging techniques were used to interpolate and fill data gaps of the depth groundwater using Arc GIS version 10.8.

WetSpass was initially created to estimate seasonal and annual groundwater recharge patterns at long-term spatial scales (Batelaan & De Smedt 2001). Due to the growing popularity of the Python programming language, the accessibility of refined input datasets, and improvements in remote sensing and GIS capabilities a newer version of WetSpass-M mode was created to estimate the distributed water balance components (groundwater recharge, surface runoff, and evapotranspiration) at monthly time steps (Abdollahi et al. 2017) to estimate the distributed water balance components (groundwater recharge, surface runoff, and evapotranspiration) of the Black Volta Basin, Morocco at a monthly basis. The original WetSpass was executed as an extension for ArcView-GIS whereas the newer version is scripted in IronPython v2.7 (an open-source implementation of Python for Microsoft. NET Framework v4.0, www.ironpython.net) and was made as a standalone model. WetSpass-M uses the Hydrology and Hydraulic Programming Library (H2PL) (Abdollahi et al. 2017).

The model must first have an optimal spatial resolution selected for it depending on the properties of each input dataset and the study area's spatial extent. The model requires several input parameters, including elevation, slope, soil type, land use, land cover, groundwater depth, and climatic information such as rainfall, wind speed, PET, and temperature (Figure 2). Climate variables are provided on a monthly basis, while land use, land cover, soil texture, elevation, and slope are single-frequency input variables. Each input dataset for the model must be created using the same spatial coordinate system and spatial resolution, and every input dataset cell must perfectly overlap. Lookup tables have been created for land use, land cover, and soil texture to facilitate the processing of these input parameters.

The processing order of the model involves several steps, including reading the data, followed by the computation of each grid cell's water balance components. The processes include interception, surface runoff, evapotranspiration, and recharge. These processes are calculated in a logical order based on the amount of rainfall.

The WetSpass-M model results were calibrated using 15 years' mean monthly river flow data (2000–2014) from two major river gauging stations (Hombole and Mojo), which can represent the entire upper Awash sub-basin. The correlation was reasonable (R² = 0.77). The average intensity parameter and interception, which are the global model parameters to which the WetSpass-M model is sensitive, were changed until a reasonable correlation was seen.

The simulated annual recharge using WetSpass-M is 114 mm, whereas the actual annual recharge using the baseflow separation method is 120 mm. Both the baseflow separation approach and WetSpass-M estimated that the area's annual groundwater recharge takes 11% of the yearly rainfall. The baseflow technique estimated that 12% (15 mm) of the recharge occurs during the summer and 88% (105 mm) occurs during the winter. Similar to this, the current WetSpass-M model predicted a recharge of around 21.3% (24 mm) during the summer and 78.7% (90 mm) during the winter. In all techniques, the recharge rates are highest in August and September and lowest in November and December.

The total AET was calculated as the sum of evaporations from bare soil, impermeable surface area, open water, interception, and transpiration of the vegetated area (Abdollahi et al. 2017). The simulated long-term monthly average AET ranges from 4.98 to 77 mm/month. The mean and standard deviation are 39 and 29 mm, respectively. AET for the entire year is 467 mm. The average AET equals 47% of the average annual rainfall (1,032 mm), with the winter seasons accounting for an average of 343 mm (73.4%) and the remaining 124 mm (26.6%) in the summer seasons.

In the southeast and the central part of the upper Awash sub-basin, high AET rates are seen on an annual basis (Figure 8). Increases in temperature close to the rift may be the cause of the maximum evapotranspiration in the southeast of the study area. The WetSpass-M model recognizes that seasonal variations in rainfall are the main cause of the seasonal variation in AET. According to the model, winter seasons contribute more to AET compared to summer seasons, with high AET values commonly observed during the winter season due to increased rainfall and enhanced soil moisture availability. Research conducted by Amiri et al. (2022) supports this, showing that winter seasons account for high AET, approximately 60% of the AET, while the remaining 40% occurs during the summer seasons.

May is the month with the highest AET because of the high temperatures and low humidity. This showed that the high rate of radiation and the presence of strong dry winds were the main causes of evapotranspiration, which was the main process of water loss in the basin. The same notion has also been supported by recent research conducted in Tanzania's Makutupora Basin using the same methodology (Kisiki et al. 2023).

The impact of various land use, land cover, and soil combinations on AET was further evaluated (Table 1). Water bodies, grass, and cultivated land with clay and loam soil have more AET than built-up or exposed land with loam soil. Similar patterns of elevated AET have been observed in other study sites such as the Moulouya Basin in Morocco and the Omo Basin in Ethiopia, where open water bodies and shrubland show higher AET (Amiri et al. 2022; Gelebo et al. 2022). The high transpiration requirement of vegetation cover and the water availability of specific soil types may contribute to increasing AET in these land uses.

The AET estimated by the Wetspass-M model was compared with the AET estimated by the water balance equation, Thornthwaite (1984) method resulting in a strong correlation (R² = 0.94). High evapotranspiration is attained in May and low evapotranspiration in December in both methods. Besides, the Budyko aridity index was employed to evaluate the aridity or dryness of a region. This index was calculated for various relevant meteorological stations, using the long-term mean annual rainfall and mean annual PET computed by Thornthwaite. The aridity index ranged from 0.4 to 0.9 across the different stations. Hence, the area is classified as semi-arid to humid climatic classification.

With an average value of 36.5 mm/month and a standard deviation of 55 mm/month, the estimated monthly surface runoff ranges from 0.13 mm/month to a maximum of 155 mm/month (Table 2). The yearly surface runoff varies from 156 to 1,143 mm (Figure 8), with an average and standard deviation of 430 and 108 mm, respectively (Table 2). The average surface runoff constitutes about 42% of the annual mean rainfall. The winter season accounts for 94.6% of the average yearly surface runoff, while the remaining 5.4% occurs in the summer season. The highest runoff is observed at the peak of the wet season (July and August) where there is a substantial amount of rain available.

Surface runoff is influenced by several factors, including soil type, land use, and topography. The upper Awash sub-basin is primarily composed of clay and loam soils, leading to higher levels of surface runoff and evapotranspiration. In particular, the low permeability of clay soil and impervious surfaces such as built-up areas and exposed surfaces dominated by clay and loam soils contribute to the highest surface runoff. Recent studies, such as the one conducted by Salem et al. (2023), have confirmed this finding. Their study revealed that the highest amount of surface runoff was observed in built-up areas with clay soil texture, which has a lower infiltration capacity.

Urbanization has been identified as the largest contributor to changes in surface runoff (Zhang et al. 2017). Studies conducted in Mexico (Nie et al. 2011) also revealed an increase in surface runoff due to urban expansion. Similarly, other studies have found that the type of land use has a significant impact on the amount and distribution of surface runoff (Kahsay et al. 2019; Gebru & Tesfahunegn 2020; Amiri et al. 2022). According to Table 1, grassland and forest cover with loamy sand soil exhibit the lowest surface runoff. This finding is consistent with studies conducted by Zhang et al. (2017) that highlighted how forests, grasslands, and shrubs have higher evapotranspiration and recharge, resulting in reduced runoff. On the other hand, agricultural land and bare land have lower roughness at the ground level, leading to increased runoff and decreased evapotranspiration compared to built-up areas. These observations emphasize the significant impact of land-use types on the dynamics of surface runoff and other water balance components. Moreover, the topography of a region plays a crucial role in determining the surface runoff rate in the WetSpass-M model. Hilly highlands, such as the northern, northwestern, and northeastern portions of the region, have relatively steep slopes, which result in a higher surface runoff rate (Figure 8). This is because the steep slope causes water to flow more rapidly over the surface, increasing the likelihood of runoff. This finding is consistent with a study conducted in the Raya Basin in northern Ethiopia (Kahsay et al. 2019). On the other hand, the central and eastern parts of the study area have a gentler slope, which leads to a lower surface runoff rate. This is because water can infiltrate into the soil more easily, reducing the amount of water that runs off the surface.

For the upper Awash sub-basin, a physically based pattern of the spatial and temporal distribution of recharge was estimated, in which all significant factors impacting both the amount and the spatial pattern of recharge were reasonably accounted for. Monthly groundwater recharge varies from 2.27 to 27 mm per month, with a mean and standard deviation of 9.5 and 7.6 mm per month, respectively (Table 2). The groundwater recharge values estimated by the Wetspass-M model were maximum (488 mm), minimum (0 mm), and mean (114 mm). The average recharge takes 11% of the yearly average rainfall. The winter season accounts for approximately 90 mm (78.7%) of the annual groundwater recharge, whereas the summer season accounts for the remaining 24 mm (21.3%).

The spatial variation in soil texture and land-use types significantly influences the seasonal and monthly distributed recharge, as observed from the overall spatiotemporal pattern of recharge. Soil texture and land use have a significant impact on groundwater recharge (Zomlot et al. 2015). The analysis of the soil and land-use effect revealed that forests with loam soil (374 mm/year) and grassland with loamy sand soil (229 mm/year) have the highest rates of recharge, as shown in Table 1. On the other hand, open water bodies and built-up areas exhibit the lowest rates of recharge (9 and 30 mm/year), respectively. There is substantial recharge in forests with permeable soils, similar to earlier research in north Ethiopia (Zeabraha et al. 2020), the Osoman River Basin in West Africa (Ashaolu et al. 2020), Raya Basin, north Ethiopia (Kahsay et al. 2019), and Moulouya Basin, Morocco (Amiri et al. 2022).

The northern and northwestern parts of the upper Awash sub-basin, characterized by heavy rainfall and permeable sandy loam and loam soils, exhibited higher annual and seasonal groundwater recharge (Figure 8). On the contrary, the central part is characterized by less permeable clay soil, which has the effect of reducing the amount of seasonal and annual groundwater recharge. Low groundwater recharge is found in city centers because of the low permeability of soils and the impervious nature of the urban areas. There are specific regions with relatively high groundwater recharge patterns, such as the northern portion of the study area, specifically the northern and northwestern highlands, and the southwestern highlands of the study area. This is mostly caused by the highly elevated topography and comparatively increased rainfall distribution. The same view is shared by research conducted in the Omo Basin in southwestern Ethiopia (Gelebo et al. 2022) that the topography variation has a significant impact on the distribution of groundwater recharge.

Previous studies estimated the upper Awash sub-basin mean annual groundwater recharge by using different methods: the groundwater level fluctuation method (Yitbarek et al. 2012) – 81 mm, the water balance method (Azagegn et al. 2015) – 131 mm, chloride mass balance method (Azagegn et al. 2015) – 135 mm, base flow separation method (Ayenew et al. 2019) – 121 mm, and distributed Hydrus 1D infiltration model (Berehanu et al. 2017) – 157 mm. Compared to the water balance method, chloride mass balance method, and base flow separation method, recharge estimates from the baseflow separation method can be taken to be more reliable than the other two methods (Berehanu et al. 2017). As a result, we chose to use the TimePlot base flow separation approach to test the monthly outputs of the WetSpass-M model because they were calculated using enough of the representative river discharge data study area. Comparing the estimated recharges from various methods with the results obtained from the WetSpass-M model (114 mm) and baseflow separation method (120 mm), it is evident that they are highly comparable and agreeable. The WetSpass-M model exhibited excellent performance in simulating groundwater recharge, as evidenced by its consistency with observed data and previously estimated models. This demonstrates the model's accuracy and reliability in estimating recharge for the research area.

This study has successfully configured the WetSpass-M model for the upper Awash sub-basin using 15 years of meteorological data from about 27 stations to simulate spatially and seasonally distributed water balance components. The sensitivity of the model was tested to both global and local parameters, and the results indicate that the WetSpass-M model is highly sensitive to local parameters and moderately to less sensitive to global parameters.

The WetSpass-M model results were calibrated using 15 years' mean monthly river flow data with a correlation coefficient of 0.77 between river discharge records and surface runoff simulation. Additionally, the recharge simulation of the model was validated with independently estimated recharge using the baseflow separation TimePlot method, with a correlation of 0.87.

The study findings indicated that around 11% of the mean annual rainfall recharges the groundwater, with 42% running off the surface and 45% evaporates. Land-use land cover and soil texture variations all play a significant role in the spatial variance of recharge. The northern and northwest regions of the upper Awash sub-basin experienced increased annual and seasonal groundwater recharge due to higher rainfall, while the central portion of the area exhibits lower rates of yearly and seasonal groundwater recharge due to its less permeable clay soil. The combined effects of soil and land use significantly influence the decomposition of rainfall into various hydrological components. Forest and grassland areas with loamy sand soil showed the highest recharge, while built-up areas and exposed surfaces with clay soil exhibited the lowest recharge.

The sensitivity analysis of the WetSpass-M model revealed its significant sensitivity to soil texture, impacting various hydrological components such as evapotranspiration, surface runoff, and groundwater recharge. Additionally, the model showed sensitivity to different land-use types, emphasizing the influence of land-use changes on model outputs. Moreover, variations in water balance components were observed in response to topographical variations, indicating the impact of elevation and slope on parameters like precipitation and runoff. These findings emphasize the importance of accurately representing soil texture, land use, and topography to ensure reliable predictions of hydrological processes using the WetSpass-M model.

This study demonstrated that the WetSpass-M model is a suitable tool for estimating water balance distribution in space and time in areas with different hydrological regimes, such as Ethiopia's upper Awash sub-basin. Future numerical models can benefit from incorporating precise results of spatially and seasonally distributed groundwater recharge obtained from the WetSpass-M model. Although the WetSpass-M model proved to be effective for estimating groundwater recharge in this study, it has a limitation in accurately representing or generalizing land-use patterns in the study area. Nonetheless, its practicality for water resources management, water allocation planning, flood forecasting, and drought assessment remains evident.

The authors confirm their contribution to the paper as follows: T. T. W. conceptualized the whole article, rendered support in data acquisition was involved in modeling, analyzed the results, and wrote the original draft. T. A. supervised the work and reviewed and edited the manuscript. B. B. rendered support in GIS map production, and B. B. reviewed and edited the manuscript.

Mizan-Tepi University has sponsored the PhD study of the first author. The other authors have no funding grant for this research.

The first author wants to acknowledge Mizan-Tepi University for providing a grant for her Ph.D. research. The authors thank the National Meteorological Services Agency and the Water Works Design and Supervision Enterprise (WWDSE) for the provision of meteorological data and groundwater level data, respectively. Addis Ababa University, School of Earth Sciences is highly acknowledged for providing a Ph.D. fellowship for the first author. My special thanks also go to friends and colleagues for their support.

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

The authors declare there is no conflict.