Quantifying water loss in leaky micro-dam reservoir through water balance analysis and high-resolution water level data modeling

Water is an essential resource for sustaining life and agricultural production (Khan & Hanjra 2008). Understanding the various components of the water balance is crucial for effective water resource management. The partitioning of precipitation into different variables of the water balance is important for sustainable water quality and quantity management (Sophocleous 1991; De Vries & Simmers 2002; Singh 2014), as well as for the planning and design of water harvesting (Biazin et al. 2012) and flood control structures. It is particularly significant in arid and semi-arid regions where the impact of climate change is significant.

Hydrological phenomena, such as runoff, flow regime, sediment transport, and surface water–groundwater interaction, play a significant role in land degradation. On the other hand, hydrological interventions can aid in the restoration of catchments and the planning of water harvesting schemes. Understanding the surface water–groundwater interactions of an area is crucial for ecological and water resource development and management (Ferone & Devito 2004; Jolly et al. 2008).

Climatic factors, especially precipitation, are particularly important for the recharge of shallow aquifers. The hydrological cycle, land degradation and management, and environmental sustainability are all interconnected (Khan & Hanjra 2008).

Leakage-related problems in reservoirs have been observed worldwide. For example, the McMilan Reservoir in the USA dried up after 12 years of operation (Pearson 1999). In the UK, approximately 30% of embankment dams constructed between 1854 and 1960 experienced incidents primarily due to leakage. Additionally, nearly 173 reservoirs in the UK have been abandoned due to various types of failures (Tedd et al. 2000). A compilation of 900 dam failure cases from around the world showed that the majority of failures were on earth dams and occurred within the first 5 years of service. Leakage and piping were the main causes of dam failures (Zhang et al. 2007).

The livelihoods of people in Ethiopia depend heavily on land resources for food and other necessities, with the majority of the population engaged in agricultural activities (Central Statistics Authority CSA 2008; Adimassu & Haile 2011; Adimassu et al. 2014). However, water availability is a limiting factor for agricultural and economic activities in many parts of the country due to low and erratic rainfall patterns. Rain-fed agriculture is common in Ethiopia, and food self-sufficiency is still in its early stages. To address this issue, the government of Ethiopia is constructing various water harvesting structures (WHSs), including micro-dam reservoirs (MDRs) to store runoff water for domestic and irrigation purposes. However, these initiatives have faced challenges related to leakage, siltation, insufficient runoff, and inefficient water management (Desta 2005; Yazew 2005; Haregeweyn et al. 2006).

To address the existing problems and develop a sustainable strategy, it is crucial to gain a comprehensive understanding of the rainfall–runoff–recharge system in the area. The previous research has utilized various approaches, including engineering geological, geophysical, and hydro-geochemical methods, to assess the leakage phenomenon. Detailed studies involving geological, geophysical, and hydro-geochemical analyses have been conducted at selected MDRs, including the Arato MDR (Berhane et al. 2013, 2016). The Arato MDR, one of the 92 MDRs in Tigray, has been significantly affected by leakage problems since its construction in 1997. The results of previous studies have emphasized a significant hydraulic connection between the reservoir and the leakage zones, particularly within a limestone–shale–marl intercalated rock unit (Berhane et al. 2013, 2016). Based on the findings, it has been determined that the reservoir is experiencing leakage through the subsurface foundation and left flank materials.

In light of this, the objective of this paper is to build upon the previous work by employing a water balance approach and utilizing high-resolution water level sensors. The paper aims to evaluate reservoir leakage by analyzing the water balance and recharge–runoff processes, while also comparing the results with previous research and different techniques. By doing so, it aims to contribute to a better understanding of the leakage issues and provide insights for developing effective strategies to address them.

The geology of northern Ethiopia can be classified based on the stratigraphic sequence. Starting from the youngest to the oldest, the geological formations are as follows: Flood Basalt (Paleo-Neogene/Quaternary Volcanics); Amba Aradam (Upper Sandstone); Agula Shale; Antalo Limestone; Adigrat Sandstone (Lower Sandstone); Enticho Sandstone and Edaga Arbi Tillite and Upper Complex Metamorphic (Basement) rocks. In addition to these formations, there are also Quaternary soil deposits found in depressions and on flat landforms.

The occurrence of groundwater in the Mekelle Outlier region is closely associated with fracturing and joints (faults) as well as the impact of dolerite intrusion on the surrounding rock. The Mekelle Outlier is a nearly circular area spanning approximately 8,000 km², where the Mesozoic sedimentary succession has been preserved from erosion. Detailed geological information about the Mekelle area, including the Outlier and its stratigraphy, can be found in the works of Beyth (1972), Levitte (1970), and Gebreyohannes (2009).

Girmay et al. (2015) identified three groundwater flow systems in the study area: shallow/local, intermediate, and deep/semi-regional. The shallow groundwater flow is concentrated in the highland plateau areas, particularly in the Agula Shale and dolerite formations, which exhibit characteristics of shallow and localized groundwater flow systems. For detailed information about the geological and hydrogeological conditions of the site, including maps and sections, refer to the studies conducted by Berhane et al. (2013, 2016).

The study area is overlain by Mesozoic sedimentary rocks and Paleo-Pliocene Volcanics. The mineralogical composition of limestone and dolerite has been summarized based on thin-section analysis conducted at the Geological Survey of Ethiopia (Addis Ababa) as part of the present study (Table 1).

The average annual rainfall in the area, based on data from the Mekelle meteorological station, slightly exceeds 600 mm, with the highest rainfall occurring in July and August (Figure 1(c)). The majority of the precipitation, around 70–80% annually, falls during the ‘Kiremt’ (summer) season, which spans from June to September. In certain years, rainfall may commence later, in July, and the months of March, April, May, and June experience negligible rainfall.

The mean minimum temperature varies from approximately 9 °C in December to 13 °C in May and June, while the mean maximum temperature ranges from 22 °C in December to around 27 °C in June.

To protect the pressure transducers from damage, they were installed inside a pipe. Since the MDR lacks a tower to vertically lower the sensor, a pipe was laid along the upstream slope of the dam to house the sensor. During installation, the proper functioning of the sensor was tested to ensure accurate readings (see Figure 3(a)). The SHDW is an open, circular-shaped well with a diameter of 4 m, a depth of approximately 6 m, and composed of alluvial deposits (including silt, clay, and some sandy materials toward the bottom).

The water level sensor measures the total pressure, from which the piezometric head can be determined after compensating for atmospheric pressure. To compensate for atmospheric pressure, a barometric pressure transducer was also installed at the meteorological station, solely recording atmospheric pressure data.

To estimate the runoff of the area, the Soil Conservation Service Curve Number (SCS-CN) method (USDA-SCS 1985) was utilized. This method relies on actual meteorological records obtained at the site. CN is a dimensionless parameter indicating the runoff response characteristic of a watershed or drainage basin. It is derived from established tables based on the characteristics of the site. The major factors that determine CN are (a) the hydrologic soil group (HSG), (b) land cover type, (c) catchment treatment, (d) hydrologic conditions, and (e) runoff condition. A weighted average CN was determined for the catchment based on factors such as land use, hydrologic conditions, and HSG. The land use map and soil condition/soil group of the catchment were established through field observations, Google Earth, aerial photographs, and existing laboratory tests.

The estimation of reference evapotranspiration was conducted on a daily basis for a period of 182 days using the FAO ETo calculator/program (FAO 2009). In addition to the parameters recorded at the local station, default values based on the geographic location were utilized.

The SMB model is a lumped model that tracks soil water over time, treating the entire watershed as a single unit. Water is stored in the soil reservoir until the soil water content (SW) exceeds the field capacity, at which point recharge occurs. The soil water balance requires monitoring the accumulated potential water loss (APWL) and the amount of water in the soil (SW). The model can be applied at daily, weekly, or monthly time steps. In this study, a daily time step was used for a duration of 182 days.

Calculations to determine SW and APWL were performed for each day using daily precipitation (P), reference evapotranspiration (ET_o), and runoff (RO) in an Excel sheet. Detailed procedures can be found in Steenhuis & Van Der Molen (1986).

The model requires local rainfall data, ET_o, plant available water (PAW) (the difference between volumetric water content at field capacity and permanent wilting point), and runoff (RO) as inputs. PAW was estimated based on vegetation cover and soil texture using the method proposed by Thornthwaite & Mather (1957). For this study, a PAW value of 200 mm was used, considering moderate-rooted cereals and clay loam as the dominant cover and soil type, respectively.

Additionally, groundwater recharge was estimated using the environmental tracer – chloride mass balance (CMB) and water table fluctuation (WTF) methods for comparison.

Effective precipitation refers to the amount of precipitation that is actually added and stored in the soil after runoff is removed. Rainwater and groundwater samples were collected from the area in 2014, directly from raindrops and from SHDW where the water level sensor was installed, respectively. All water samples were analyzed at the Laboratory for Applied Geology and Hydrogeology of Ghent University, Belgium.

The WTF method is most suitable for shallow water tables that exhibit sharp water level rises and declines. The study site has a typical shallow aquifer that experiences water level rises during the wet season due to direct natural groundwater recharge and leakage from the MDR.

The model considers various factors over a specific time period, such as a month or hydrological year. These factors include inflows in the form of runoff (RO) from the watershed and direct rainfall (PR) on the reservoir's surface. Outflows are accounted for through evaporation from the water surface (ER), water consumed by livestock (Live), water released through the spillway (Spill) and outlet (Out), as well as losses due to leakage (Leak) (refer to Figure 4).

In accordance with the principle of conservation of mass or volume, the change in water volume within the reservoir can be calculated by subtracting the output volume from the input volume (Rodríguez-Huerta et al. 2020). For Arato MDR, this simplified conservation of mass or volume can be further expanded.

To calculate the reservoir water loss caused by evaporation from the surface of the reservoir, potential evapotranspiration is used due to the absence of pan evaporation data at or near the study site. It is important to note that using data from pans located far from the water body under investigation can lead to significant errors, as suggested by Winter (1981).

To convert reference evapotranspiration rates to an open water surface, Penman (1948) provides factors ranging from 1.25 to 1.67. Additionally, Doorenbos & Pruitt (1984) present empirical factors, also known as crop coefficients, for converting reference evapotranspiration to open water body evaporation. For dry environments with strong wind (comparable to the present study site), an empirical factor of 1.2 is suggested for all types of water bodies.

Indeed, calibration is necessary when employing the water balance model approach in isolation, without supplementing it with other methods. In the present scenario, it utilized the CMB and WTF methods to cross-validate the outcomes of the water balance model. Additionally, the model was executed using the measured volume of water stored in the Arato Reservoir. Consequently, the reliability of the Arato MDR water balance model is already presumed, as it has been verified through independent methods and recorded data at the reservoir.

The various software applications utilized for constructing, drawing figures, and analyzing models were ArcMap 10.4 (ArcGIS), Microsoft Excel 2010, GrapherTM 8, Surfer 10, and Microsoft Power Point 2010.

Throughout the 182-day period, the total rainfall amounted to approximately 633 mm, with only 55 days experiencing rainfall. The individual rainfall totals ranged from 2 to 37 mm. The reservoir and groundwater levels rose to a maximum of about 6 and 2 m, respectively. However, the rise in groundwater level in the SHDW was incomplete due to the well starting to overflow caused by leakage and the presence of a confining clay layer. During the period of measurable rise in the SHDW (before overflowing), a cumulative rainfall of 446 mm was recorded, while for the reservoir, it was about 530 mm. The WTF method was applied only during the period when the well did not overflow. The CMB and SMB methods were used to account for the discrepancy caused by the well overflowing.

The hydraulic gradient between the reservoir and the SHDW, where the water level sensor was located, was estimated to be approximately 0.043. This estimation was based on the elevation of the reservoir and groundwater levels, which were measured at 2,424 and 2,398 m, respectively, and a horizontal distance of about 600 m.

At the study site, there was no device specifically designed to measure runoff. Therefore, the estimation of surface runoff was carried out using the SCS-CN method. A weighted CN was used for the entire watershed based on the land use and soil type characteristics.

To estimate the CN, the specific characteristics of the clay loam soil were taken into account; including its poor to fair hydraulic condition and its classification as HSG C, as outlined in Table 3. In addition, the farming practices in the area involve plowing and treating the plots in accordance with the topography, following contour lines. This contoured treatment or practice was considered for the entire area. Minor terraces were also observed in the grassland areas. The CN value was calculated by weighting the surface area of different land uses and resulted in a value of 79.6 for the entire watershed, as shown in Table 4.

Considering the weighted CN for the entire catchment, along with a corresponding potential maximum retention (S) value of 65.1, the runoff generated from the catchment over a period of 182 days amounted to 48.8 mm. This runoff volume represents approximately 7.7% of the total rainfall received during that period, which was 633 mm. This can be expressed as an RC of 0.077. It is worth noting that out of the 182 days, only 13 days experienced rainfall that resulted in runoff, with runoff amounts ranging from 0.2 to 6.5 mm. The remaining 55 rainy days had precipitation levels ranging from 2 to 37 mm, but did not generate any significant runoff. To provide a broader context, Table 5 presents various study findings from Tigray and other regions of Ethiopia for comparison.

Moreover, recharge was estimated using CMB. The chloride concentration in rainwater was around 0.98 mg/l, which is comparable to previous studies in the Mekelle area (0.8 mg/l) (Kahsay 2008) and in Mugher (0.88 mg/l), Jema (0.97 mg/l), and Upper Awash (0.84 mg/l) areas (Berehanu et al. 2017). However, it is lower than the weighted average reported for Mendae Plain (Tigray, Ethiopia) (2.6 mg/l) (Walraevens et al. 2015), Abu Delaig in Sudan (weighted average 4.6 mg/l), the Sahel zone in Senegal (weighted average 2.8 mg/l), and East Africa (Rodhe et al. 1981). Nevertheless, it is a reasonable value for the study site considering its distance from the coast (over 300 km from the Red Sea and 1,400 km from the Indian Ocean) and low atmospheric dust during the rainy season. The chloride concentration in groundwater was 6.17 mg/l, and the effective rainfall (after subtracting runoff) during the observation period was 584.2 mm. By using Equation (6), the estimated recharge was determined to be 92.8 mm, accounting for approximately 15% of the total precipitation. In a study conducted in Indonesia (Yogyakarta City), Buana et al. (2023) reported a groundwater recharge of approximately 126 mm using the same method. On the other hand, Rodríguez-Huerta et al. (2020) estimated a recharge ranging from 43 to 143 mm in their study in Mexico, employing different methods. Considering a reference value from FAO (2009), the recharge was estimated to be around 72 mm. Furthermore, the recharge was estimated using actual measurements of WTF in a hand-dug well using a water level data logger (Figure 5). Only a portion of the observation period was considered for this method, as recharge is no longer reflected by rising water levels once groundwater starts to overflow. Based on the soil types in the recharge area and considering an average specific yield of clay and silt (S_y = 0.05) from Johnson (1967), the estimated recharge during the water table rise over a 44-day period (19 July 2014 to 31 August 2014) was 100 mm. This accounts for approximately 22.5% of the precipitation (P_rise = 446 mm) during the water table rise or 15% of the total precipitation during the entire observation period. In a study conducted in the upper Blue Nile Basin (Ene-Chilala watershed, 4.4 km²) in Ethiopia, Addisie (2022) utilized the WTF method and obtained a recharge value of 89.7 mm, which accounted for approximately 16.9% of the precipitation. This finding is comparable to the results obtained in the present study.

The water balance of Arato MDR can be analyzed by referring to Figure 4. By separating the known and unknown variables of the model, it is possible to solve the unknown components separately for the dry and wet seasons.

During its existence, Arato MDR has never reached its maximum reservoir level, leading to no spillway overflow. Additionally, during the observation period from 19 July 2014 to 16 January 2015, there were no irrigation or outlet releases, and livestock consumption was negligible. Given these conditions, in the water balance equation with all other components known, the direct inflow (runoff, RO) and leakage loss (Leak) remain as unknown (refer to Table 7). The estimation of RO using the CN method will be considered later to determine leakage for the entire observation period.

The leakage component can be determined during the dry period when there is no inflow to the reservoir. To estimate leakage, an observation period from 17 October 2014 to 16 January 2015 was selected, considering that there was no surface or subsurface inflow and meaningful rainfall contributing to the change in reservoir volume (refer to Table 7). Continuous decline in water level during this period, attributed to evaporation and leakage, was observed. Firstly, evaporation loss and direct rainfall input on the reservoir were estimated using Equations (11) and (12), resulting in values of 5.77 × 10⁴ and 0m³, respectively. Using Equation (8), Leak was estimated at approximately 112,300 m³ over 83 days (equivalent to 1,353 m³/day). Considering the reservoir area (10 × 10⁴ m²), this corresponds to 13.5 mm/day.

Assuming constant leakage over time, the leakage amount over 182 days is estimated to be around 0.25 × 10⁶ m³. Once the leakage loss is determined from the dry observation period, the water balance for the wet season (refer to Table 7) can be solved to determine runoff (RO). For this purpose firstly, the evaporation loss and direct rainfall input on the reservoir were estimated using Equations (11) and (12), resulting in values of 2.85 × 10⁴ and 4.24 × 10⁴ m³, respectively.

Then, using Equation (8) and adjusting the leakage loss for the number of days in the wet season, the surface runoff (RO) is estimated to be approximately 0.64 × 10⁶ m³ (equivalent to 44.14 mm).

Therefore, the total inflow is the sum of direct runoff from the catchment and direct rainfall on the surface of the reservoir, which is equal to 0.68 × 10⁶ m³. Without considering the input from direct rainfall on the reservoir, the surface runoff (0.64 × 10⁶ m³) is slightly lower compared to the runoff estimated using the CN (0.71 × 10⁶ m³) and significantly lower than the initial design capacity of the reservoir (2.5 × 10⁶ m³). This observed runoff value obtained from water level sensor measurements should be considered reliable over the observation period. The slightly higher value obtained using the SCS-CN method may be attributed to various factors associated with catchment characteristics and the estimated parameters in the calculation of ET_o.

Now, with the runoff determined using the CN method and the leakage loss of the reservoir calculated using the reservoir water balance model, the leakage loss during the wet period was computed (Equation (8)).

Considering the values and unknowns presented in Table 8 and applying Equation (8), the reservoir loss due to leakage (Leak) is estimated to be 4,602 m³/day (equivalent to 0.24 × 10⁶ m³ in 53 days), corresponding to 57.5 mm/day. Therefore, during the wet season, approximately 34 and 38% of the runoff estimated from the CN method and water balance method, respectively, is lost through leakage.

The rise in water level in the reservoir is influenced by rainfall over the catchment area and direct rainfall on the reservoir surface. The water level rise in the SHDW does not show a strong relationship with individual rainfall events. The slope of the rise remains constant regardless of the amount of rainfall on a daily basis. This could be due to continuous replenishment of the aquifer through leakage from the upstream MDR. From September onwards, the reservoir level continuously declines, while the groundwater level in the SHDW remains constant. Field observations indicate that the well was full and overflowing to the ground surface, suggesting that during the dry period, the water in the SHDW comes directly from the Arato MDR. This conclusion was confirmed by Berhane et al. (2016) using manual water level measurements and hydrochemical analysis. In contrast, a different situation was observed at the non-leaking Tsinkanet MDR, where there was no interaction between the reservoir and the nearby shallow aquifer, resulting in a constant reservoir level for months while the groundwater level in the shallow aquifer declined.

Evaporation is a natural process by which water is lost from a basin or water body. In the case of a reservoir, evaporation contributes to the overall water loss. To estimate the amount of water lost due to evaporation from the reservoir, a reference evapotranspiration value is multiplied by an empirical factor of 1.33. This factor takes into account the lack of complete meteorological data at the reservoir site. For a period of 136 days, the estimated evaporation from the reservoir was 933 mm, extrapolating this value; the annual reservoir water evaporation is approximately 1,818 mm.

Various studies have estimated the annual evaporation for different lakes in Ethiopia. For Lake Haromaya (elevation range: 1,980–2,343 m a.s.l.), Setegn et al. (2011) estimated mean annual evaporation of 1,882 mm using the energy balance method and 1,784 mm using the Simple Abtew equation method. For Lake Tana (elevation: about 1,800 m a.s.l.), Chebud & Melesse (2009) obtained annual evaporation estimates of about 1,430 mm using the Penman method and 1,420 mm using the Meyer method. Vallet-Coulomb et al. (2001) estimated annual evaporation for Lake Ziway (elevation: 1,636 m a.s.l.) at approximately 1,780 mm using the lake energy balance method and 1,870 mm using the Penman method.

These results for different lakes in Ethiopia are comparable to the estimate obtained for the Arato reservoir. It is important to note that with additional meteorological data, more accurate and refined estimates can be obtained.

Understanding the hydrological processes and ensuring efficient water utilization is crucial for optimal design and operation of water harvesting schemes. However, many developing countries, including Ethiopia, face challenges in this regard due to the lack of reliable long-term data (Collick et al. 2009). This situation often leads to improper sizing and design of reservoirs, culverts, and storm pipes. For instance, in Tigray, out of the 92 MDRs constructed, 21 (22%) experience low inflow, and 56 (61%) suffer from siltation issues, primarily due to the lack of reliable hydrological data (Berhane et al. 2016).

Table 5 provides a summary of the runoff and recharge estimations using different methods in the region. The results obtained from the current study align with other research findings from the area (Table 5). It was observed that runoff starts to occur after a minimum rainfall of 13 mm. Collick et al. (2009) reported that runoff occurs on 20% of degraded areas in the Andit Tid and Yeku watersheds (from the Ethiopian highland area) after 10 mm of rainfall. Girmay et al. (2009) conducted field experiments in 2006 at Maileba MDR, also in the same region, and found runoff percentages of 3.8, 8, 10.8, and 21% for exclosure, Eucalyptus plantation area, grazing land, and cultivated land, respectively. Arefaine et al. (2012) estimated the surface runoff for the Illala sub-basin, which includes the Arato MDR, at approximately 7% of rainfall using WetSpass. The studies conducted in the Tigray region are consistent with the present results and with studies conducted in other regions (see Table 4).

Selamyihun (2004) estimated the RC for central Ethiopian highland Vertisols (in a different region) using calibrated CN. The RC values ranged from 23 to 51%, equivalent to 102–258 mm/year. Another study in the central highlands of Ethiopia by Adimassu et al. (2014) reported an annual runoff volume of about 145 mm (19% of rainfall) for a plot with soil bunds and 325 mm (28%) for fallow land. Adimassu & Haile (2011) reported 169.53 mm (32.3%) from a field experiment in a plot covered with wheat in the central highlands of Ethiopia. Implementation of soil and water conservation practices, as well as exclosures, reduce runoff and enhance local infiltration (Vandecasteele 2007; Walraevens et al. 2009, 2015; Nyssen et al. 2010; Vandecasteele et al. 2011), ultimately leading to land degradation reversal and forest regeneration in Tigray over the past three decades (De Mûelenaere et al. 2014; Belay et al. 2015).

During the observation period, the catchment yield from surface runoff was estimated at approximately 0.71 × 10⁶ m³, which represents only 34% of the initial design capacity of the MDR (2.5 × 10⁶ m³) as reported by Hagos (1995). This yield is higher than that obtained by Gebreyohannes (2009) using the WetSpass method, which was 0.096 × 10⁶ m³. It is important to note that the present estimation is based on site-specific data for a period of 182 days, without considering any runoff contribution during the remaining dry period of the hydrological year.

The recharge estimated using the SMB method can be considered as an annual estimate, assuming negligible recharge during the rest of the year. Similarly, the recharge from the CMB method is annual by nature, while the recharge from the WTF method represents a period of 44 days. Mekonnen et al. (2015a, 2015b) highlighted the significance of small sediment storage structures in relation to infiltration and sediment trapping. According to Stroosnijder (2009), who surveyed 181 medium-sized dams in the Eritrean Highlands, 31% were completely silted up, 52% were partially silted up, and only 17% did not suffer from siltation. These conditions clearly emphasize the need for integrated land and water management, tied with water harvesting planning, such as the MDRs.

The core focus of this research was to assess the leakage rate from the reservoir and confirm earlier conclusions using geological, geophysical, and hydrogeological approaches by Berhane et al. (2013, 2016). The leakage rate calculated from the water balance during the dry period was found to be high at 13.5 mm/day, compared to the results obtained by Yazew (2005) for the Gumselasa and Korir MDRs, which were 0.9 and 0.4 mm/day, respectively. Both the Gumselasa and Korir MDRs, located 35 km south and 26 km north of the Arato MDR, have low leakage rates. However, the leakage rate of the Arato MDR is approximately 15 and 34 times higher than that of Gumselasa and Korir MDRs, respectively.

During the initial design of the project, it was estimated that there would be an annual seepage loss of 9,965 m³. However, the actual leakage amount for 182 days exceeded this estimate by about 0.24 × 106 m³. Thus, approximately 15% of the inflow, as calculated using the CN method, is leaking during the 83-day dry period. The leakage rate during the wet period, considering the inflow from the CN method, was found to be 4,602 m³/day or 57.5 mm/day, which is higher than the values obtained from the dry season water balance model (1,353 m³/day) of the reservoir. It is observed that a rapid rise in water level in the reservoir triggers strong leakage, while leakage amounts decrease during the dry period as water levels recede.

It is important to note that the leakage estimate does not provide information about the specific location of the reservoir where the leakage is occurring. This can be better understood by considering the site geology and hydrogeology. The right abutment and central foundation of the MDR are based on dolerite, while the left abutment is on a limestone-shale-marl intercalation unit. The reservoir area is underlain by both units, with surficial Quaternary alluvial deposits found in depressions and along the river course. Based on reported hydraulic conductivities for different formations, it can be concluded that the dolerite and shale layers are less permeable compared to the bedded and fractured limestone. Thus, the limestone layers in the limestone–shale–marl intercalation unit are likely responsible for the leakage of the MDRs. Additionally, weathered and fractured top parts of all units, as well as the contact zone between the dolerite and the intercalation unit, were found to be permeable and prone to leakage.

Maps and sections providing further details on the site geo-hydrology can be found in Berhane et al. (2013, 2016). Conventional geological and geophysical techniques, such as vertical electrical sounding and profiling, were used to identify the leakage zone and understand its mechanisms, as reported by Berhane et al. (2016).

Interestingly, the leakage from the MDRs indirectly benefits local farmers by improving their livelihoods through small-scale irrigation from shallow hand-dug wells and diverted streams originating from the reservoir leakage.

By considering an average groundwater recharge of 98.9 mm from three methods, a runoff of 48.8 mm from the CN method, and an AET of 509 mm from the Thornthwaite and Mather SMB method, it is possible to compare these values with the total rainfall in the area for the given period. The total rainfall (P) was approximately 633 mm, while the sum of the three components (RO + AET + Rech) was 656.7 mm. These values are comparable and provide confidence in the application of the different methods in this area and in similar settings.

In conclusion, this research project underscores the importance of using site-specific climatic data and physical characteristics of a watershed, as well as employing different approaches and models in reservoir and water resource planning. Long-term annual water balance analysis can contribute to a better understanding of local conditions and improve the optimal planning of WHSs and other water resource management strategies.

The estimation of evaporation was indirect due to the lack of an open pan evaporimeter at the site, which could introduce uncertainties. However, the estimated evaporation was found to be comparable with estimations from other parts of the region with similar climatic and altitude conditions. Estimating groundwater recharge and reservoir leakage is challenging and associated with uncertainties. The uncertainties in the CMB method can be attributed to analytical precision and errors in determining chloride concentration. The ultimate goal of this research was to quantify leakage from the reservoir and understand the inflow, and the estimated leakage from the measured and monitored data can be considered a fairly good estimate for future planning and management in data-scarce areas.

The study site's groundwater recharge was estimated using multiple approaches. The natural recharge rates were determined to be 104, 92.8, and 100 mm using the SMB, CMB, and WTF methods, respectively. These estimates correspond to a total recharge of approximately 1.41 million m³ for the catchment area. Assuming negligible recharge during the dry period, the results from the SMB and CMB methods can be considered as annual recharge estimates, while the WTF method covers a period of 44 days during the rainy season.

The runoff was estimated using the SCS-CN and water balance approaches. The SCS-CN method yielded a runoff estimate of about 0.71 million m³, while the water balance approach, utilizing input data from water level sensors, resulted in a runoff estimate of approximately 0.64 million m³.

The leakage from the reservoir was estimated using the water balance model approach for the initial part of the dry period. The calculated leakage rate was found to be 13.2 mm/day, equivalent to a total of 112,300 m³ over the 83-day dry period (equivalent to 1,353 m³/day). Using the same water balance model and the runoff estimated by the CN method, the leakage during the wet period was determined to be 4,602 m³/day (equivalent to 0.24 million m³ over 53 days), corresponding to a rate of 57.5 mm/day.

The methods employed in this study, along with accurate time series local climatic input data, can be applied in other regions to forecast water resources, particularly in areas with water scarcity and data limitations. These results and approaches are valuable for dam planners and for identifying suitable WHS.

The authors would like to express their sincere gratitude to Ghent University, Belgium, and Mekelle University, Ethiopia, for generously providing funds for the data collection process. The authors would also like to extend their appreciation to the local administration of Arato village, as well as the data logger and meteorological station guards, for their invaluable support during the data collection period and for ensuring the proper care of all instruments in the field in our absence.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.