A pragmatic approach to the combined effect of climate change and water abstraction on Lake Ziway water balance, Ethiopia

Greenhouse gases have increased the heat stored in the earth's atmosphere, contributing to global warming and climate change (IPCC 2014). According to preliminary observations and climate projections, climate change significantly affects freshwater resources, with far-reaching consequences for human societies and ecosystems (Arnell 2004). Climate change affects temperature trends, evapotranspiration, precipitation, runoff, and streamflow. Temperatures will rise because of climate change, affecting catchment evapotranspiration and lake evaporation (Nigatu et al. 2016), and consequently, lake levels fluctuate due to climate change (Seyoum et al. 2015). The IPCC (2014) projected a general increase in precipitation in East Africa in the future. Simultaneously, the IPCC's Fifth Assessment Report revealed that increased global precipitation in the long term is associated with significant seasonal variations in precipitation (Collins et al. 2013). Arnell (2004) also reported an increase in annual streamflow in East Africa based on climate change projections, which is consistent with the findings of other researchers (IPCC 2014). The changes/variability in hydro-climatic variables yield frequent floods, droughts, and reduced food production, which eventually influences social well-being. Because water is used for a variety of purposes, increased human interference harms freshwater resources. As a result, it is critical to investigate the combined effect of climate change and anthropogenic influences on a catchment's water resources. The Lake Ziway catchment is one of the most important catchments in Ethiopia's rift valley lakes basin. Climate change and human influences may jeopardize the availability of water resources in the future. However, the magnitude of change caused by these factors has yet to be quantified.

Lake Ziway and its environs are vital to the livelihoods of about 2 million people, according to Central Statistical Authority (CSA) report in 2014. According to Hayal et al. (2015), Lake Ziway serves as a source of drinking water for local towns, irrigation supplies, and the supply of fish to the country's largest market center. However, climate change/variability and the intensification of agricultural operations in the catchment are putting a strain on the lake. Water withdrawal from the lake and Feeder Rivers for irrigated farms is rising (Ayenew 2004). Rivers feeding into Lake Ziway have been diverted for irrigation recently. Furthermore, farmers and investors around the lake regularly pump water from the lake for irrigation resulting in a decreased outflow into the Bulbula River (Herco et al. 2007; Desta et al. 2020). Climate change and anthropogenic factors may have a folded effect on the lake's hydrological and ecological integrity, which could be harmful (Ayenew 2004; Seyoum et al. 2015; Desta et al. 2020; Goshime et al. 2019a). However, the combined effects of future climate change and water withdrawal from the lake and tributary rivers are poorly understood.

Variations in main climate variables such as temperature, precipitation, and evapotranspiration are one of the repercussions of climate change. This, in turn, could cause changes in the hydrological cycle, reducing lake level and ultimately affecting the catchment's water availability. Several researchers revealed the presence of both climate variability/change and a strong anthropogenic disturbance on Lake Ziway's water levels (Zeray et al. 2006; Herco et al. 2007; Hayal et al. 2015; Desta et al. 2017, 2020; Damtew et al. 2021). However, most of these studies focused solely on the water balance under natural conditions, water abstraction, or land-use changes. Therefore, this study uses copious Global and Regional Climate Models as input to the Hydrologiska Byråns Vattenbalansavdelning (HBV) hydrological model to evaluate and quantify the implications of climate change and future planned water abstraction for irrigation scenarios from Lake Ziway.

The climate in the Lake Ziway catchment ranges from humid to sub-humid. The lowland catchment areas surrounding the lake have arid or semi-arid climates, while the mountains have dry-humid to humid climates (Tenalem et al. 2007). The annual temperature ranges between 12.5 and 25.8 °C. The yearly oscillation of the inter-tropical convergence zone has a significant impact on rainfall patterns in the study area, resulting in warm and wet summers and dry, cold, and windy winters. According to time series data from 1986 to 2016, the catchment's mean annual rainfall ranges from 713 mm near the lake floor to 1,146 mm on the plateaus at the Ziway and Butajira stations. Agriculture/cultivated area is the most common land use/cover, accounting for 81% of the total land use area. The remaining areas are covered by forests, bare surfaces, and water bodies.

We collected meteorological data such as temperature, rainfall, sunlight hour, wind speed, and relative humidity from the Ethiopian Meteorology Institute (EMI). The period of time-series data spans from the years 1980 to 2018. Data quality measures such as homogeneity, outlier, and consistency tests were used to find complete and consistent datasets. The streamflow and lake level data were obtained from the Hydrology Department of the Ministry of Water and Energy (MoWE) and used to calibrate and validate the hydrological model. MoWE Ethiopia also provided the bathymetric survey data of Lake Ziway conducted in 2013.

The Digital Elevation Model (DEM), having a resolution of 30 m × 30 m from ASTER GDEM V2, was downloaded from https://lpdaac.usg.gov.data access and was used to delineate the catchment. The land use/land cover (LULC) map of Ethiopia for 2016 was used to acquire land use and land cover statistics. The research area's soil map was also received from Ethiopia's Ministry of Water and Energy. The dominant crops grown in the area and cropping pattern data were collected using a field survey in the irrigation area. Throughout the year, the study area has two cropping seasons. A total of 4,099 hectares of the Lake Ziway sub-catchment was surveyed for data on commercial and large-scale farms managed by the government and private investors. Moreover, the Meki-Ziway large-scale irrigation project-II, having the potential of a 2,000 ha irrigable area, not yet functional, was considered in this study for future scenarios. Earlier studies did not consider this scheme and new development plans in this study area.

We used dynamically downscaled regional climate model (RCM) output data from the Coordinated Regional Climate Downscaling Experiment for the Regional Climate Change Assessment Project (CORDEX) from https://esgf-node.llnl.gov/projects/esgf-llnl/ in the current study. Using downscaled climate data with a resolution of 0.44° × 0.44°, CORDEX Africa was used to investigate future precipitation and temperature projections (Musie et al. 2020). This study used data from six (6) GCM and four (4) RCM models (Table 1). With a temporal resolution of one day, the climate model output includes precipitation, maximum and minimum temperature, humidity, wind speed, sunshine hours, and solar radiation for the reference and future periods from RCP 4.5 and RCP 8.5.

Several climate models were used to forecast future climate variables. Climate model data for hydrologic modeling (CMhyd) was used to extract data from global and regional climate models (Rathjens et al. 2016). We compared climate models before bias correction to identify high-performing models capable of capturing basin climatic aspects. The climate model that meets a set of statistical criteria (Musie et al. 2020) was chosen. Statistical metrics such as the root mean square error (RMSE), percentage bias, correlation coefficient (CC), and coefficient of variation (CV) were used to evaluate the effectiveness of the climate models. Table 2 shows the statistical measures used to evaluate model performance.

Table 2

Statistical measures used for performance evaluation of the GCM/RCM products

Statistical Measure .	Equations .	Unit .	Best fit value .
Bias		%	0
¥RMSE		mm	0
CV		%	0
CC		–	1

Statistical Measure .	Equations .	Unit .	Best fit value .
Bias		%	0
¥RMSE		mm	0
CV		%	0
CC		–	1

^¥RMSE, Root mean square error; CV, Coefficient of variation; CC, Correlation Coefficient; R_R and R_ob is RCM simulated rainfall and observed rainfall, respectively, over bar symbol denotes the mean of the statistical value over the analysis period (n); σR refers to the standard deviation of either RCM or observed rainfall.

View Large

The contents of the model, i.e., the routines for each variable, the procedures of simulations for mass exchange between the upper and lower boundaries, and other detailed descriptions of the model are documented in recent publications such as in Habib et al. (2014), Dessie et al. (2015) and Goshime et al. (2019b). Daily rainfall, temperature, potential evapotranspiration, river flow, land use/land cover, and the research area's Digital Elevation Model (DEM) were used as input variables for the HBV hydrological model. The measured river discharge at Meki and Katar was used to calibrate and validate the HBV hydrologic model. Eight model parameters, K4, Khq, Alfa, CFLUX, FC, BETA, LP, and PREC, from the earlier studies were selected to calibrate the model (Wale et al. 2009; Goshime et al. 2019b, 2020).

Table 3 presents the descriptions of the selected calibrated model parameters and their initial, minimum, and maximum values.

The model was initialized for a 1-year warm-up period to conduct the model calibration process. Then, model sensitivity analysis was carried out to obtain model-sensitive parameters so that attention is given to these parameters during model calibration. The default values of model parameters were used as a benchmark in deciding sensitive parameters. The sensitivity analysis was conducted by varying a model parameter and investigating the overall model run-up performance using statistical indices such as relative volume error (RVE) and Nash-Sutcliffe efficiency (NSE). The parameters that produce significant variation in the statistical indices for varying model runs were selected. Parameter sensitivity is measured by varying one parameter at a time and fixing all other parameters during the model run. A parameter whose rate of change is high compared to others during model runs is considered a sensitive parameter. Finally, the model was calibrated using streamflow data from 1985 to 1991 and verified using daily flow data from 2000 to 2007.

The areal precipitation over the lake's catchment area was estimated using data from four stations (Meki, Ziway, Adami Tulu, and Ogolcho). The areal precipitation was calculated using the Thiessen polygon method. After calculating the depth of rainfall over the surface area, the volume of rainfall was calculated by multiplying the average lake area by the rainfall depth. Finally, the potential evapotranspiration (PET) was calculated using a combination of the Penman-Monteith (Allen et al. 1998) and Hargreaves (Hargreaves & Allen 1985) methods.

The streamflow from the two gauged main rivers (Meki and Katar) was simulated using the calibrated HBV model. Then, the rainfall in the ungauged catchment was translated into runoff using the area-ratio approach for the two sub-catchments located in similar rainfall regime areas (Nruthya & Srinivas 2015). The area-ratio method was chosen due to similarities in the hydrological, geological, and land cover of the gauged and ungauged catchments.

The water demand for irrigation was estimated using the CROPWAT 8.0 method (Allen et al. 1998) by applying site-specific crop data. The CROPWAT model was used to compute the crop's water requirements (CWR) using the study area's climatic, soil, and crop data. Following the FAO protocol, crop and soil data were utilized to assess consumptive usage. Field data were collected to identify common crops grown and cropping patterns of the study region. Accordingly, onion, tomato, green beans, cabbage, maize, pepper, papaya, and wheat are the most commonly grown crops in the study area. We also used crop coefficients, growth stage, depletion, and soil data from FAO (Allen et al. 1998) and other documents, including the basin master plan and irrigation system design report to compute the irrigation water demand. The water demand was estimated by considering the following:

• i.

  Irrigable area expansion with the current precipitation and temperature condition

• ii.

  Area expansion with climate change in the future climatic variables

• iii.

  The potential irrigable area is reaching 30,090 ha in future periods

After determining all water balance components, a spreadsheet model was created to mimic lake water levels using Equation (7) (Goshime et al. 2020). Lake bathymetric data was utilized to obtain the observed lake-level data from Ethiopia's Ministry of Water and Energy in 2013. The simulated lake level was then compared to the observed lake level to investigate feasible differences. Using the bathymetric data, the lake volume was converted to lake level. The simulated lake level's accuracy was then compared to the observed lake level.

When comparing the observed mean annual rainfall with model counterparts, there exists an overestimate and underestimation for the respective RCMs. Seven models (IPSL-RCA4, ICHEC-CCLM4.8, ICHEC-ReMo2009, CNRM-RCA4, MPI-ReMo2009, GFDLG-RCA4, and GFDL2M-ReMo2009) show higher mean annual precipitation value than the observed mean annual values. GFDLM-RCA4, IPSL-ReMo2009, and CNRM-RCA4 climate models highly overestimated the rainfall in the wet season, while GFDL2G-ReMo2009 and ICHEC-ReMo2009 slightly overestimated the rainfall during the same season. The ICHEC-CCLM4.8 model underestimated wet-season rainfall over the catchment area. However, MPI-ReMo2009 and IPSL-RCA4 models revealed a more consistent and closer pattern with observed precipitation values (Figure 2 and Table 4). As a result, these two models were chosen for this study's intended purpose.

Table 4 shows the statistical indices used to evaluate the performance of the climate models. The mean annual value comparison of precipitation variables between observed and climate model outputs revealed that MPI-ReMO2009 (bias = 16.48%) and IPSL-RCA4 (bias = 15.13%) performed well. In contrast, GFDLM-RCA4 (bias = 55.78%) showed a significant deviation from the observed value based on the bias (Table 4). In addition, MOHC -CCLM4.8 shows a negative association with observed precipitation in terms of correlation coefficient (CC). The MPI-ReMO2009 and IPSL-RCA4 climate models estimated precipitation values closer to those observed values. Therefore, these models were used to assess the effects of climate change on the water balance of the study area.

This study's result agrees with the study by Kumar et al. (2020) that revealed the RCM (REMO2009) and its derivative GCM (MPI) model portrayed the rainfall value near the Global Precipitation Climatology Centre (GPCC) observed rainfall. The authors reported that the biases in REMO2009 and GCM (MPI) are comparable in amplitude, making them suitable for future projection of climate variables under Representative Concentration Pathways (RCPs) 4.5 and 8.5 at 99 and 95% confidence levels. According to Ogega et al. (2020), when compared to the other 24 multiple climate models runs from five Coordinated Regional Climates Downscaling Experiments, the RCM (ReMo2009) forced by MPI has the best performance in simulating East Africa's spatio-temporal precipitation characteristics (CORDEX).

Figure 5 shows the projected changes in monthly, seasonal, and annual precipitation of the Meki and Katar sub-catchments compared to the baseline period (1971–2000) under the RCP 4.5 & 8.5 scenarios.

The mean annual maximum temperature over the Lake Ziway catchment will rise by 1.6 degrees Celsius for all models under the RCP 4.5 emission scenario and 2.5 degrees Celsius for all models under the RCP 8.5 scenario between 2021 and 2080, compared to the baseline period (1971–2000). All models show that the monthly maximum temperature will rise in all months except June and August, which showed the smallest increase compared to the other months under both scenarios. However, due to increased CO₂ emissions, the RCP 8.5 scenario (2051–2080) has shown the greatest increase of all models.

The results of model sensitivity analysis of the HBV model parameters in the Katar and Meki catchments revealed that the BETA, FC, and LP parameters were the most sensitive, determining the catchment storage. The response of the NSE statistical index indicated that the parameters that determine the shape of the hydrograph, such as K4, PERC, and Khq, were moderately sensitive. When compared to other parameters, the PERC and Alfa parameters were found to be less sensitive. Goshime et al. (2019b; 2020) reported similar findings over the Lake Ziway area, as did Dessie et al. (2015) over the Lake Tana sub-basin in Ethiopia. Hence, the calibration process should take caution while calibrating volume and shape-controlling parameters in the study area.

The HBV model performs well when measured quantitatively using the NSE and RVE statistical indices (Table 5). Table 5 shows that the model-calibrated parameter values are within the acceptable range. Using data from the Katar sub-catchment, the model's performance was found to be very good, with R² = 0.8 and NSE = 0.63 during the model calibration period from 1985 to 1999, and R² = 0.78 and NSE = 0.62 during the validation period from 2000 to 2007. Similarly, when the Meki catchment data was used, the model performed very well, with R² = 0.81 and NSE = 0.64 during calibration and R² = 0.79 and NSE = 0.54 during validation for the same period as the Katar sub-catchment. The model's performance was slightly lower during the validation period than during the calibration period. However, its performance during the validation period is still very good, indicating that the model can be used to project future streamflow into Lake Ziway.

According to CROPWAT model results, the total annual irrigation water demand is 39.4 Mm³ (Table 6). The current irrigation water requirement was calculated using average climate data from 1980 to 2009. Because the streamflow of the Meki and Katar rivers has decreased and there is a scarcity of water for irrigation using these rivers, the Meki and Katar catchments have only one cropping season. Therefore, water withdrawals for other purposes or sectors, aside from agricultural water use, have not been considered in this study. According to Ayenew (2004), increased irrigated farming in the sub-catchments has contributed to a significant decrease in the flow of the feeder rivers and the water level of Lake Ziway.

From 2021 to 2050, if 40% of the total irrigable area is used, the annual irrigation water requirement will be 81.1 Mm³, doubling from the baseline. Between 2051 and 2080, the yearly irrigation water needs for 60% of the total irrigable area will increase to 121 Mm³, a threefold increase over the baseline. The RCP 4.5 scenario projects annual irrigation demand of 85.6 Mm³ (2021–2050) and 124.4 vMm3 (2051–2080), respectively, whereas the RCP 8.5 scenario projects 102.8 Mm³ (2021–2050) and 125.6 Mm³ (2051–2080). As a result, if all of the intended irrigated areas are constructed, the annual water requirement is expected to be 150 Mm³. Such irrigation practices may cause a 3 m drop in Lake Ziway's level, dramatically dropping Lake Abiyata's level and the Bulbula River drying up (Ayenew 2007). Previous research found that abnormally dry years and extensive irrigation near Lake Ziway caused the lake to drop 1.5 meters below the long-term average (Tenalem 2002).

We considered the existing yearly total water abstraction of 42 Mm³ in this analysis. Earlier studies put the water abstraction at 38 Mm³ (Goshime et al. 2021), 28 Mm³ (Ayenew 2004), and 41 Mm³ (Desta & Lemma 2017). The dataset and methodology used in each of these studies may differ slightly. For example, under the RCP 4.5 scenario, the lake's annual total inflow and outflow are estimated to be 1,471.6 Mm³ and 1,268.4 Mm³ for 2021–2050, respectively, and 1,751.1 Mm³ and 1,321.9 Mm³ for the period 2051–2080. In contrast, the lake's inflow and outflow volume under RCP8.5 are 1,384.7 Mm³ and 1,256.8 Mm³ for 2021–2050 and 1,972.8 Mm³ and 1,352.8 Mm³ for 2051–2080, respectively (Table 7).

Under the RCP 8.5 scenario, the lake water level will rise by 62.4 cm, while the surface area and storage will grow by 25.4 km² and 293.3 Mm³, respectively, for the period 2051–2080. On the other hand, the lake's water level might rise by 56.7 cm above the average baseline lake level under the RCP 4.5 scenario, with a surface area of 23 km². This occurrence could be linked to the use of GCM models and climate model uncertainties, the amount of data used and the methodology used in the study. We believe that the accuracy of this study's findings was enhanced by using two climate models and an ensemble.

Finally, the combined effect of irrigation expansion and climatic change was assessed, and the results revealed a decrease in lake water level, volume, and surface area (Table 8). Under the RCP 4.5 scenario, the lake water level, surface area, and storage loss might be as much as 25 cm, 10.3 km², and 101.7 Mm³, respectively, between 2021 and 2050. The highest lake level reduction under RCP 8.5 might be up to 23 cm. This suggests that water abstraction, rather than climate change, will have the most detrimental impact on Lake Ziway's water level. Seyoum et al. (2015) found similar results for Lakes Abiyata and Ziway, informing that lake level rise is attributed to increased river inflow and over-lake precipitation. However, extensive irrigation water abstraction reduced the volume of both lakes.

In contrast to earlier studies, this analysis combined rainfall-runoff and water balance models with observed and projected climate data from different model output datasets. The HBV conceptual rainfall-runoff model was used to simulate the streamflow of the Lake Ziway gauged catchment. The amount of streamflow from ungauged catchments was calculated using area-ratio approaches. Unlike the earlier studies, which used only one station's data, evaporation was calculated using four stations' air temperature data. The use of various climate model outputs has aided in the estimation of lake area rainfall. This study is also relatively comprehensive because it employed the CROPWAT model to incorporate the irrigation water demand considering site-specific crops to calculate the water balance study area. Prior studies, however, relied on few specific crops with the highest water requirements.

It should be noted that the calculation of the lake's water balance components may be subject to much uncertainty. The first source of uncertainty is groundwater's contribution to the lake's water balance, which was overlooked in this study. Estimating lake evaporation and lake areal precipitation is another source of uncertainty. Finally, runoff simulation from gauged and ungauged catchments has inaccuracies. As a result, future studies should consider uncertainties in the water balance by combining the advantages of multiple methodologies in assessing each water balance component.

The influence of climate change on water balance and the water level was explored in this study utilizing outputs from different climate models for currently feasible scenarios and water resource development near Lake Ziway. Future research should consider water balance uncertainties by employing improved techniques, enhanced model calibration procedures, and more advanced bias correction algorithms.

The combined effect of climate change and irrigation water abstraction has been explored in this study under the RCP 4.5 and RCP 8.5 emission scenarios for 2021–2050 and 2051–2080 periods. Annual precipitation, mean annual maximum, and minimum temperatures over the Katar and Meki catchments are projected to increase in the two future periods. As a result, the average yearly potential evapotranspiration in the sub-catchments and across the lake will rise. Despite climate change-induced increases in lake level and volume, both RCP scenarios project a net decline in lake level and concomitant lake volume reduction in the future. From 2051 to 2080, the consequences of water abstraction may result in a 257.2 Mm³ loss in annual lake volume, a 77.5 cm drop in lake level, and a 25 km² surface area decline. The combined effects of climate change and water use could result in a 25 cm drop in lake level, with annual reductions of 10 km² surface area and 101 Mm³ volume. Therefore, strict monitoring procedures for water abstraction and proper lake ecosystem management policies should be in place for the Lake Ziway catchment. We suggest future studies to assess and evaluate various water management scenarios to mitigate the adverse effects of water withdrawal and inform integrated water resource management among all stakeholders.

We acknowledge the Ministry of Water and Energy, MoWE, and the National Meteorology Institute, NMI, for providing this study's necessary hydrology and climatic data. In addition, we recognize the central Rift Valley lakes basin development office for providing information during field data collection. Finally, the authors also gratefully acknowledge the Swedish Meteorological and Hydrological Institute (SMHI) for allowing the HBV model free of charge.

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

The authors declare there is no conflict.