Establishing irrigation potential of a hillside aquifer in the African highlands

Feeding 9 billion people in 2050 will require sustainable development of all water resources, both surface and subsurface. Yet, little is known about the irrigation potential of hillside shallow aquifers in many highland settings in sub‐Saharan Africa that are being considered for providing irrigation water during the dry monsoon phase for smallholder farmers. Information on the shallow groundwater being available in space and time on sloping lands might aid in increasing food production in the dry monsoon phase. Therefore, the research objective of this work is to estimate potential groundwater storage as a potential source of irrigation water for hillside aquifers where lateral subsurface flow is dominant. The research was carried out in the Robit Bata experimental watershed in the Lake Tana basin which is typical of many undulating watersheds in the Ethiopian highlands. Farmers have excavated more than 300 hand dug wells for irrigation. We used 42 of these wells to monitor water table fluctuation from April 16, 2014 to December 2015. Precipitation and runoff data were recorded for the same period. The temporal groundwater storage was estimated using two methods: one based on the water balance with rainfall as input and baseflow and evaporative losses leaving the watershed as outputs; the second based on the observed rise and fall of water levels in wells. We found that maximum groundwater storage was at the end of the rain phase in September after which it decreased linearly until the middle of December due to short groundwater retention times. In the remaining part of the dry season period, only wells located close to faults contained water. Thus, without additional water sources, sloping lands can only be used for significant irrigation inputs during the first 3 months out of the 8 months long dry season.

the Robit Bata experimental watershed in the Lake Tana basin which is typical of many undulating watersheds in the Ethiopian highlands. Farmers have excavated more than 300 hand dug wells for irrigation. We used 42 of these wells to monitor water table fluctuation from April 16, 2014 to December 2015. Precipitation and runoff data were recorded for the same period. The temporal groundwater storage was estimated using two methods: one based on the water balance with rainfall as input and baseflow and evaporative losses leaving the watershed as outputs; the second based on the observed rise and fall of water levels in wells. We found that maximum groundwater storage was at the end of the rain phase in September after which it decreased linearly until the middle of December due to short groundwater retention times. In the remaining part of the dry season period, only wells located close to faults contained water. Thus, without additional water sources, sloping lands can only be used for significant irrigation inputs during the first 3 months out of the 8 months long dry season.

K E Y W O R D S
Ethiopia, groundwater, highlands, irrigation, Lake Tana, water budget, water storage, water table fluctuation method 1 | INTRODUCTION As the world's population grows to around 9 billion, demand for agricultural products puts growing pressure on scarce water resources, nowhere more so than in sub-Saharan Africa (SSA). Food security in SSA is closely related to water resources variability (Tucker & Yirgu, 2011). Dry season irrigation is one of the best coping mechanisms for improving food security in a changing climate (Alberti, Cantone, Colombo, Oberto, & La Licata, 2016;Karimov et al., 2013). Other potential mechanisms include drought tolerant crop varieties, planting trees, soil conservation, and changing planting dates (Alemayehu & Bewket, 2017;Bryan, Deressa, Gbetibouo, & Ringler, 2009). However, surface and groundwater resources are limited in most of sub-Saharan Africa (Altchenko & Villholth, 2015;Xie, You, Wielgosz, & Ringler, 2014;You et al., 2011) and are likely to become even more scarce in the future (Gan et al., 2016;Vörösmarty, Ellen, Green, & Revenga, 2005).
Previous studies have explored the viability of small-and largescale irrigation in sub-Saharan Africa (SSA) through use of surface and groundwater resources. The predictions are highly variable between studies as they are dependent on the calculation of groundwater levels and other water demands (e.g., domestic, livestock, industry and environmental flows) (Pavelic, Smakhtin, Favreau, & Villholth, 2012).
According to Xie et al. (2014), by combining surface and groundwater, small-scale irrigable land could increase potentially by 30 million ha in sub-Saharan Africa. Schmitter, Kibret, Lefore, and Barron (2018) estimated that 6.3 million ha in Ethiopia could be irrigated using solar photovoltaic based lifting technologies to extract groundwater up to 25 m depth. Worqlul, Collick, Rossiter, Langan, and Steenhuis (2015) showed that in the Gilgel Abay basin less than 0.1% of the potential irrigable land could be irrigated with the available surface water resources. The greatest weakness in these studies is whether sufficient surface or groundwater is available for planned irrigation (Otoo, Lefore, Schmitter, Barron, & Gebregziabher, 2018).
Groundwater availability for irrigation depends on both the recharge and the loss to stream by baseflow and interflow. In Ethiopia, Kebede (2013) estimated the annual recharge of groundwater based on annual baseflow measurements in the range 0-400 mm/year. Mechal, Wagner, and Birk (2015) using the SWAT model predicted groundwater recharge in the south-eastern Main Ethiopian Rift, from 410 mm/year in the highlands to 25 mm/year on the rift floor. In semi-arid Tigray, Walraevens et al. (2009Walraevens et al. ( , 2015 (Guzman, Anibas, Batelaan, Huysmans, & Wyseure, 2016;Walraevens et al., 2009).
Several methods have been used for determining groundwater availability. Theis (1935) developed an analytical solution for finding the transmissivity and specific yield of an aquifer based on the drawdown of a well by pumping. He based his solution on the analogy of Darcy flow in an aquifer with heat flow. Goes (1999), Healy andCook (2002), and Leiriao, He, Christiansen, Andesen, and Bauer-Gottwein (2009) developed and refined the water table fluctuation (WTF) method in which the seasonal and daily fluctuation of the water table were used to determine the recharge; Brutsaert and Nieber (1977) used the rate of change in river flow during periods without rain to determine the aquifer properties by the change in discharge to short and long time analytical solutions of the aquifer outflow; Pool and Eychaner (1995) performed sophisticated temporal-gravity surveys to measured changes in the differences in gravity between two reference stations on bedrock and six stations at wells. Steenhuis, Jackson, Kung, and Brutaert (1985) used the instantaneous profile method to measure the recharge by measuring simultaneously the change in moisture content and matric potential at several depths within the 2 m soil profile. Finally, Steenhuis and van der Molen (1986) found that the Thornthwaite and Mather (1955) procedure matched the recharge measurements of the instantaneous profile method. This method is based on a water balance (WB) of the rootzone where the actual evaporation is linear function of the potential evaporation at field capacity and zero at wilting point.
From the above list of methods, WTF method is according to Healy and Cook (2002) and Leiriao et al. (2009) the preferred method for determining recharge for unconfined aquifers when specific yield and time series water level data are available. Specific yield needed to convert water table height measurements into water availability can be approximated from soil moisture retention curve (Meinzer, 1923;Veihmeyer & Hendrckson, 1931), from WB analysis (Walton, 1970), or from pumping tests (Barker & Herbert, 1989;Walton, 1970).
Finally, the results of Walker et al. (2018) implied that the WTF method should be modified for sloping terrains because the lateral movement of water downslope is significant.
With the increasing interest of agricultural transformation through irrigation both in Ethiopia and Africa, it is important to determine the temporal availability of groundwater storage in highland areas. In the Ethiopian highlands, the land is sloping, and groundwater has relatively short retention times (Alemie et al., 2019, in press). It is important to obtain accurate predictions of the available seasonal groundwater storage both in space and time to support sustainable agricultural intensification through irrigation (Kvaernø, Engebretsen, Barneveld, & Deelstra, 2016;Sopocleous, 1991). However, traditional methods such as the WTF method are not suitable without modification. The objective of this study is, therefore, to find effective ways to calculate the spatial and temporal distribution of groundwater storage in sloping shallow aquifers to predict the potential for sustainable groundwater use accurately. The selected Robit Bata watershed in the Lake Tana basin is representative of watersheds in the highlands of Ethiopia where smallholder irrigation takes place to grow chat (a mild narcotic crop) and vegetables.

| Description of the study area
The Robit Bata experimental watershed is just under 1,000 ha and is located at the south-eastern edge of Lake Tana, in the headwaters of the Blue Nile, Ethiopia (Figure 1). The area has a sub-humid climate with annual rainfall of 1,450 mm, of which 85% falls from June to September.
Elevation ranges between 1,800 and 2,030 m with slopes up to 44%. Forty-five percent of the area has slopes above 10%. Shallow soils over bedrock can be found in the upper areas while the lower part of the watershed is underlain by an aquifer over basalt bedrock.
Massive basalt rock outcrops are visible in riverbeds and occasionally on steeper slopes and quarry sites. Faults can be identified by the Acacia tree growing in lines (Ayenew, Demile, & Wohnlich, 2008). Red soils overlay the basalts of which 31% are Luvisols, 24% Alisols and 19% Nitisols (Figure 1).
Mixed agriculture is the predominant land use accounting for 80% of the area. The livestock graze within the watershed on lands available. Crop production during the rainy season is dominated by cereal crops and to lesser extent legumes. During the dry season small-irrigated plots of Khat, Mango trees and vegetables can be found along the river course and near water bearing faults. Water is Over recent years, smallholder irrigation has expanded in the watershed resulting in an increase in the number of shallow groundwater wells used for irrigation and domestic water supply. The water levels of these wells are on average 4 m depth from the ground surface during the end of rain and up to 8 m depth on average in the dry season.
This has led to the drying up of the river in the dry season.

| Conceptual model for the watershed hillslope aquifers
The watershed has two distinct parts. The eroded upper watershed consists of shallow soils of less than 1 m depth over basalt. Saturation excess runoff prevails when the soil becomes saturated during rainstorms, as infiltration in the subsoil is limited (Tilahun et al., 2013. The remaining lower half of the watershed has deep soils up to 15 m over basalt containing an aquifer that is the origin of the baseflow and interflow in the rivers. All wells are in the lower half of the watershed (Figure 1).

| Conceptual model for hillslope aquifers
A schematic cross section of the lower part of the watershed with the hillslope aquifer is given in Figure 2. The deep soil (maximum depth 15 m) overlays volcanic basaltic rocks with faults. Due to soil degradation in the highlands, a dense layer with a thickness between 40 and 60 cm is formed below the ploughed rootzone due to leaching of fine soil particles from the surface (Figure 2). This dense layer restricts root growth but has sufficient capacity for rainwater to percolate downward (Tebebu et al., 2015). As a result, the rootzone is often restricted to 10-20 cm (Tebebu et al., 2015). As the ploughed rootzone soil is very permeable and little surface runoff occurs, the percolation is equal to the amount of rainfall minus the water that transpires from the crops and water directly evaporated from the soil surface.
The groundwater behaviour in the highlands is very dynamic during the rainy season. When there is both more rainfall (p) than evaporation (e a ) and the soil is above field capacity, excess rain (r) percolates downward so that the soil will be at field capacity. Consequently, a water table builds up in the vadose zone which in hillslope areas flows laterally downward through the soil matrix (q d ) and through the faults (q f ) into the river as interflow (Figure 2). Initially there is more recharge than lateral flow and groundwater builds up. However, as the rainfall decreases the lateral flow is greater than the recharge and the groundwater table decreases. The water table in the vadose zone disappears first at the top of the hill and then gradually moves downhill. As we will see later in the result section, the perched aquifer in the study area has dried up completely by December and the interflow ceases.
The river becomes dry and so are most of the wells. Only wells near faults hold water and are located close to Acacia trees which are in line with the faults in the landscape.

| Calculating shallow groundwater storage in highland areas
We applied two independent methods to calculate the availability of groundwater for irrigation. The first method is based on the WB of the calculated recharge from precipitation and potential evaporation while the second method is a modification of the WTF (Goes, 1999;Healy & Cook, 2002;Leiriao et al., 2009) method using 40 of the 42 monitored shallow that had between 60 and 88 weekly well depth measurements in the period from April 2014 to the December 2015.

| WB method
In the WB method, the groundwater storage can be determined from the mass balance by subtracting the recharge (i.e., the excess water that cannot be stored in the root zone of the portion of the watershed where the aquifer is located) to the groundwater in the area underlain by the aquifer from the baseflow and interflow generated from this aquifer at the outlet, for example, where S wb is the average aquifer storage expressed as a depth (L) over the whole watershed calculated with the WB method (hence subscript F I G U R E 2 Conceptual model of a hillside aquifer with actual evapotranspiration rate, e a ; rainfall rate, p; percolation rate, r, to the hillslope aquifer; lateral flow through the soil matrix, q d ; the rate of later inflow to the faults, q fi ; and the rate of lateral outflow from the faults, q fo The proportional area was computed with the PED model as explained in the Supplementary material S1 and S2. Recharge r t (L/T) to the groundwater (sometimes called potential recharge Walker et al., 2018) was calculated from the widely used Thornthwaite Mather method (Dunne & Leopold, 1978;Steenhuis & van der Molen, 1986;Thornthwaite & Mather, 1955;Tilahun et al., 2013;Walker et al., 2018). The method first calculates the actual evapotranspiration and the change in actual water content of the root zone and then assigns the excess water that cannot be stored in the root zone to either percolation or surface runoff. Actual evapotranspiration was calculated assuming a linearly decrease in evapotranspiration with moisture content from field capacity to wilting point (Steenhuis & van der Molen, 1986). At field capacity, the evapotranspiration is assumed to be equal to the potential evapotranspiration and zero when permanent wilting point is reached. Formally, this can be written as: where p t (L/T) is the precipitation on day t, e p (L/T) is the potential evapotranspiration rate, e at (L/T) is the actual evapotranspiration rate, is the available water in the root zone at the previous time step and AWC (L) is the maximum volume of water in root zone (i.e., water holding capacity) per unit area.
The available water in the root zone per unit area on day t, AW t (L), is either equal to AWC when the rainfall exceeds the storage capacity of the soil or for smaller amounts of rainfall the previous day's moisture content plus the difference of precipitation and evaporation and can be simply written as: For the Robit Bata watershed assuming an average root depth of 20 cm, average field capacity of 0.35 cm 3 /cm 3 and average permanent wilting point of 0.2 cm 3 /cm 3 resulted in an AWC of 30 mm (Hune, 2016 andGetachew, 2018).
The excess water above the field capacity, r t (L/T), percolates downward and recharges the groundwater. When the soil is below field capacity, recharge is negligible. Formally, this can be expressed as: As lateral flow is significant in hillside aquifers, the WTF method, recommended for unconfined aquifers for individual wells without lateral flow, was modified by using the average change in water table height for the aquifer underlying the watershed and including the baseflow at the outlet to account for the lateral flow component. Consequently, groundwater storage at the watershed scale was computed by product of the relative average groundwater water level and the drainable porosity (also called specific yield) of the watershed aquifer and areal proportion of watershed area contributing to the aquifer: where S mWTF (L) is the aquifer storage averaged expressed as a depth of the whole watershed, η (L 3 /L 3 ) specific yield or drainable porosity; h (L) average water table height of the aquifer; and h o (L) is the average groundwater level in the wells (which is 1.7 m as we will see later) when the baseflow ceases to flow (or starts to flow). Groundwater tables below h o are associated with the faults.
In this method, the specific yield or drainable porosity (η) is a critical variable to convert water table height in amount of groundwater available for irrigation (Delleur, 2010). Enku et al. (2016) found that in the nearby Fogera Plain the soil moisture content decreased linearly with height above the water table and therefore expressed the drainable porosity as: with η the drainable porosity, θ s (L 3 /L 3 ) volumetric soil moisture content at saturation and θ fc (L 3 /L 3 ) the volumetric soil moisture content at field capacity. The probability of exceedance of rainfall intensity was calculated for the 5-min rainfall intensity and was compared with the steady state infiltration rate.

| Infiltration rates and main runoff mechanisms in the watershed
To confirm that the surface soil has sufficient infiltration capacity when unsaturated, steady state infiltration rates were measured using 30 cm diameter single ring infiltrometer inserted to a depth of 10 cm on 10 locations with different land uses located at different landscape position during rainy season of 2014 (

| Determination of specific yield
Soil moisture at saturation and field capacity were determined taking undisturbed soil samples from side wall of the wells at a depth below 4 m using a 49 mm diameter and 51 mm high cylindrical core sampler.

T A B L E 1 Steady state infiltration rates measured under different land use classes and topographic locations
Four soil samples (A-D, Figure 1) were taken from selected wells along a transect in the watershed: A was taken near to the river outlet, where the basalt bed rock reaches the surface; B was taken upstream representing transported clay loam soil texture mixed with highly weathered rock fragments; C was taken upstream of A near to a tributary stream with a slightly strong weathered rock formation and D taken near to the second mainstream, in a weak highly weathered rock area that is saturated during the wet season. Soil moisture content of the four samples at saturation were determined from the soil porosity using the bulk density samples. Field capacity was determined at 33 kPa (−3.3 m) using the 15 bar plate extractor apparatus of model 1500F2, vessel wt. 39.7 kg and vessel volume 11 L. Median specific yield of 6.5% was estimated using Equation (6) as shown in Table 2. This value was used to estimate the groundwater storage from the groundwater levels data.

| Rainfall, rainfall intensity and infiltration capacity as function of land use and topography
The  (Table 1).
In Figure 4, the 5-min rainfall intensities are compared with the medium infiltration rates of the various slope positions. Downslope, the rainfall intensities exceeded minimum infiltration rate (6--8 mm/hr) only 20% of the time during the wet season when these soils are near or at saturation. Since in the remaining parts of the watershed the median infiltration rate of 60 mm/hr or above was exceeded by rainfall intensity only 2-2.5% of the time (Figure 4), surface runoff will be minimal, and water will infiltrate in the soil and recharge the perched groundwater table in the upslope soils with basalt at shallow depths or in the perched hillslope aquifer in the lower parts. This lack of surface runoff was also reflected in anecdotal field observation.

| Spatial and temporal groundwater levels near the surface
The rise and fall of the monthly average kriged water tables in the northern lower part of the watershed upstream of the outlet is shown in Figure 5.   Figure 1 In August, the water table was at its maximum level above the bedrock and ranged between 3 and 5 m from the ground surface over most of the lower part of the watershed (i.e., the dark blue coloured area in Figure 5). In September near the end of the rainy season, the water table decreased slightly but continued to be elevated. Once the rains stopped in October, the water table remained elevated near the outlet but decreased most rapidly farthest away from the watershed outlet (light green areas in Figure 5) and slightly less rapid at interme-

| Contribution of interflow and baseflow to discharge
Discharge was small in early June but then increased towards the end of the rainy season. The runoff coefficients as ratio of total streamflow to rainfall were computed to understand the temporal distribution of the runoff coefficients. In 2014 and 2015, the runoff coefficients increased with time when the interflow compared to the surface runoff increased. The coefficient reached 0.8 in October 2014 ( Figure 6).
Both the water table depth in the lower part of the Robit Bata watershed and runoff coefficient increased as the rainy season advances (Figures 5 and 6). This is expected since according to Darcy law the interflow can be calculated as the product of the water table height near the river and the hydraulic gradient. For a hillslope, the hydraulic gradient is nearly equal to the slope of watershed and thus constant. Hence, interflow is directly proportional to the height of the water table for sallow hillslope aquifers. This is consistent with other watersheds in the Ethiopian highlands (Liu et al., 2008;Tilahun et al., 2015).

| mWTF method
The average water table of 40 of 42 wells are shown in Figure S3 in the Supplementary material. Using the mWTF method, the water table data is converted to the perched groundwater storage using the following parameters in Equation (5) Figure S2). This height also corresponds, approximately, when the stream flow ceases in the middle of January. The drainable porosity is given in Table 2 and equals 0.065 cm 3 /cm 3 . The proportion underlain by the aquifer was similar to that in previous section using the WB method and is equal to 0.52. The resulting storage calculated by the mWTF method is shown in Figure 9 with orange symbols.
The results show that the maximum groundwater storage is reached in the second week of September which is near the end of the rainy season. The maximum storage ranges from 147 mm in 2014 to 106 mm in 2015, a week after the peak storage calculated using the mass balance.

| Interflow explains temporal variations in groundwater storage
The two independent methods to estimate the groundwater storage on sloping deep soils in an experimental watershed in the sub-humid Ethiopian highlands agreed well (Figure 9). The input in the WB method was the daily precipitation, potential evaporation and the baseflow separated hydrograph (Figures 7 and 8). The mWTF method used the average weekly observes groundwater levels ( Figure S2) together with the drainable porosity.
The PED model estimated that 52% of the watershed (17% saturated valley bottoms and 35% of the hillslope area) contributes to the shallow aquifer present in the watershed (Table S1). The contribution of the hill- The shallow hillside aquifers behave differently than the major aquifers in Africa. While pumping at rate greater than the recharge for the major aquifers will cause a permanent and unsustainable decline in the groundwater (Tilman, Cassman, Matson, Naylor, & Polasky, 2002) such as experienced in China, India and Middle East to North America, this is not case for hillside aquifers with travel times shorter than the dry monsoon phase. Over irrigation of crops might actually help keep the water longer in the vadose zone and thereby possibly extend the irrigation season. In addition, investing in water saving agriculture will not result in more water available for the crop of the farmer later in the season, because the velocity of the drying front is independent of the height of the water table. The benefit of water savings agriculture is for the downstream water users because it will increase the downstream baseflow. Finally, mitigating climate change by storing more water in the season so that the aquifer is recharged as proposed by Alberti et al. (2016) and Karimov et al. (2013) will not be feasible.

| Importance of temporal shallow groundwater storage for small-scale irrigation
Detailed understanding of spatial and temporal variation of groundwater availability and levels can support smallholder irrigation initiatives in designing and implementing feasible smallholder solutions. We found that for a sloping watershed typical for the Ethiopia highlands that shallow groundwater levels peaked throughout the valley bottoms and mid-slopes in September reaching on average 3-5 m below the surface depending on their location ( Figure 5). From October to December, groundwater levels rapidly dropped in the up-and mid-slope areas as interflow receded. In January wells in the mid-slope areas started to dry up which is approximately 100-120 days after the rainy season had finished. As a result, the average groundwater depth in the majority of the watershed drops below 10 m in the mid-slope and below 7 m in the lowland with the exception of wells located near the faults.
Understanding the temporal variability of the groundwater storage and the spatial variability of water levels provides useful information for smallholder farmers in selecting suitable dry season crops and lifting technologies. Especially, farmers who have newly made the transition from rainfed systems to intensified irrigated agriculture in the dry season could considerably reduce risks of crop failure through information on available shallow groundwater. The sloping upland areas will have limited to no shallow ground water available from December to January onwards in the dry monsoon phase. Hence, dry season irrigation will be possible only from October to December/January which translates in one high value crop season with a cultivation period of approximately 80-100 days such as onion, lettuce, carrots and beans (Allen, Pereira, Raes, & Smith, 1998). A second crop (e.g., chilli or tomato) or a perennial crop (e.g., khat) will be possible only on a small part in the valley bottom where the groundwater in the faults can be accessed. All of this will contribute to increased food security and potentially introduce new lines of income to diversity agricultural livelihoods.

| CONCLUSION
The groundwater availability in a sloping watershed in the Ethiopian highlands was evaluated with two independent methods. The two methods agreed well and indicated that the recharge to the portion of the watershed with aquifer was significant (up to 500 mm/year when averaged over the watershed and 933 mm in the deep soil portion).
However, subsurface outflow rates were high as well due the slope of the aquifer and most of the water that was recharged during the rain phase was lost as interflow and baseflow within a first 3-month period of the dry phase. Since interflow amount depend on the slope, the smaller the slope the longer the ground water will be available. The practical significance is that hillside aquifers in deep soils can only be used for irrigating crops in a 3-to 4-month period after the monsoonal rain season ends. The only limited water available during the remainder of the dry phase is from wells in valley bottoms associated with geological faults.
Since donors and governmental agencies consider extending smallholder groundwater irrigation vital for increase food production, our results are directly relevant to planning of smallholder irrigation on rainfed agricultural lands in sub-Saharan Africa. Many of these potential areas are hillside/sloping lands with challenging access to surface water, often implying expensive uphill pumping. Our results suggest that for planning purpose lateral subsurface flow on even the slightly sloping lands should be considered in estimating the quantity of water that can be stored in shallow groundwater for small-scale irrigation interventions. By simply considering the recharge only and ignoring the lateral movement of water, irrigation potential will be unrealistic in actual practice. Moreover, practices such as recharging additional water during the recharge period as advocated for level lands, do not apply to the shallow hillslope aquifers.
Future research in specifying the water availability and water yields in hillside might focus especially in answering two issues that this study was not able to address: First of all, the origin of the water in the wells near faults from January through the beginning of the rain phase should be investigated. We expect that this water comes from the unsaturated vadose zone, percolating slowly downward as unsaturated flow and then flows sideways over sloping bedrock where it collects in the soil near the faults. Second, the effect of slope on water availability should be examined further. We expect that when the slope becomes smaller, the groundwater will last longer after the rain phase ends because the velocity of the water is slower because the hydraulic gradient becomes smaller. In the nearby Fogera Plain with a slope of 0.1%, the water is available year around. It would be helpful for planning purposes to know the maximum slope and at the type of soils that will assure year around availability of groundwater.

DATA AVAILABILITY STATEMENT
The data that support the findings are available in the supplementary material.