Assessment of future climate change impact on water balance components in Central Rift Valley Lakes Basin, Ethiopia

There is no doubt that climate change is the most important environmental challenge facing the world in the 21st century, owing to its wider-reaching impacts on human society. There are two causes of global climate change. Natural processes, such as changes in the sun's energy, variations in Earth's orbit, and volcanic eruptions and human activities, change the climate of the earth (US EPA 2016). Since the industrial revolution, humans added large amounts of greenhouse gases (GHGs) into the atmosphere by burning fossil fuels, heating, and cooling buildings and vehicles as well as land use land cover change, and environmental degradation (Oroud 2008; US EPA 2016; Chakilu et al. 2022). Human activities significantly increase the release of large amounts of carbon dioxide (CO₂) and other GHGs into the atmosphere. For example, the global CO₂ concentration in the atmosphere increased from 288 parts per million in 1750 to 415 parts per million in 2021 (Jin et al. 2021) and causing for warming of the Earth's atmosphere (Chakilu et al. 2022). According to IPCC (2007), the global surface temperature increased by 0.74 °C over the last 100 years. The rise of global and regional temperatures changes the hydrological cycle which is likely to enhance the frequency and severity of extrema climate events of floods and droughts (Machiwal & Jha 2012; Ye et al. 2013; Dowlatabadi & Ali Zomorodian 2015). As temperatures increase, more water evaporated into the atmosphere and intensifies the hydrologic cycle which results in severe showers in the coming years (Bao et al. 2012; Ye et al. 2013; Zhao et al. 2014; Xin et al. 2019). Many studies have shown the impacts of climate change and variability (Chien et al. 2013; Kundu et al. 2017) and the rate of potential evapotranspiration on the hydrology cycle (Wang et al. 2017). As a result of the cycling effects of climate change, the spatial and temporal distribution of runoff is altered, finally which is resulting in the most significant consequences of alteration of access to water, increase dry spells, and food shortage in many parts of the world (Yuan et al. 2016). This is more common in regions where the population is heavily dependent on natural resources and has a limited capacity to adapt to climate change impacts (Wagesho et al. 2012).

Understanding the current and future water resources under various management scenarios could provide a solid foundation for the sustainability of water resources. One of the most fundamental aspects of climate change research is the ability of global climate model output to simulate local climate conditions and downscaling techniques. A coordinated regional climate downscaling experiment (CORDEX) has been planned to simulate the climate of Africa based on a variety of climate regimes (Geleta et al. 2022). Regional climate models (RCMs) at their finer resolution allow the simulation of local climate conditions in detail and provide future predictions. As the study indicated, East African countries are vulnerable to future climate change and Ethiopia is often cited as one of the most extreme examples (Bekele et al. 2021). The availability of fresh water is expected to be among the most significant consequences of the projected climate change scenarios (Du et al. 2021; Galata et al. 2021). Over the last few decades, hydrological and general circulation models (GCMs) have been coupled to extensively study the climate change impact on water resources under different scenarios (Galata et al. 2021). The potential impact of climate change on hydrologic processes is widely investigated using the soil and water assessment tool (SWAT) developed by the US Department of Agricultural Service (Arnold et al. 2012a, 2012b), and the dynamical downscale RCMs data (Kuma et al. 2021; Emiru et al. 2022). Many studies have been conducted in Ethiopian basins on climate change's impact on water resources, and much significant work has been done on stream flow responses to climate change (Nigatu et al. 2016; Abraham et al. 2018; Worqlul et al. 2018; Bekele et al. 2021; Chakilu et al. 2022; Takele et al. 2022).

All these mentioned studies have used one or more GCM-RCM and emission scenarios, various downscaling methods (statistical or dynamical), and either bias-corrected or raw (uncorrected) data without evaluation of the ability of the climate model to simulate local climate conditions. These methodological differences can have a significant impact on the outcome of an impact study. For example, Chakilu et al. (2022) investigated the impact of climate change on the streamflow the Upper Nile of the basin of Gumera watershed and they conclude that there will be an increase of streamflow in future scenarios, while Mengistu et al. (2021) concluded that there would be a decreasing stream flow up to 22.7% in the same basin. This suggests that there is no common agreement on the effects of climate change on water resources result may result due to uncertainty of the GCMs and RCMs inherited from model structure, parametrization of the climate system, initial and boundary conditions, spatial resolution, emission scenarios (Taylor et al. 2012; Kim et al. 2014), and availability of multiple climate model outputs. For example, Balcha et al. (2022a) have evaluated the ability of five (5) RCMs to downscale 23 GCMs outputs over the study regions and concluded that each GCM is simulating differently the climate condition of the study regions and there was uncertainty between the downscaling institutions while downscaled the same GCM outputs. The details are found in Balcha et al. (2022a).

Additionally, the utilization of water resources has proceeded without knowing the hydrologic components and climate variability, which is a critical problem in water resource management. The growing demand for irrigation water, food security, policy enforcement for import substitution of wheat, expansion of horticulture crops, the establishment of a new agro-processing industrial zone (Bulbula and Hawassa), and population pressure will create concerns about the sustainability of water, as predicted by unsettling future climate conditions (Goshime et al. 2021). The supply of a more accurate explanation regarding the timing of the development of stream flow, lake level, and precipitation in connection with climate change could help to convey critical information to water resources planners, decision-makers, and the local community. The main objective of this study was to assess the impact of climate change on the water resources of the Central Rift Valley (CRV) using the SWAT model and best RCMs. This study creates climate change scenarios for 2021–2050 (the 2040s) and 2051–2070 (the 2070s) and examines climate change trends of rainfall and temperature changes. The methodology presented in this document could be a useful approach for studying climate change's impact on water resources. Section 2 describes the research topic, data sets, and methods, while Section 3 presents the findings and related discussions. Section 4 concludes with conclusions.

The Ethiopian National Meteorological Agency (NMA) provided daily climate data of precipitation and maximum and minimum temperatures for 16 stations from 1997 to 2017. The data were prescreened for unrepresentative or wrongly entered values and checked for negative rainfall with a minimum temperature greater than the maximum temperature. For stations that do not have temperatures (maximum or minimum or both), values from the nearest stations were used. Additionally, Climate Hazards Group Infrared Precipitation with Station (CHIRPS) was downloaded from http://climateserv.servirglobal.net/ website from 1983 to 2017. The spatial correlation between observed and CHIRPS datasets was greater than 0.8, which falls in the range of very good. Hence, the CHIRPS datasets were used for SWAT model setup, calibration, validation, and bias correction (BC) of climate model output.

The output of the Coordinated Regional Downscaling Experiment (CORDEX) of the African domain was used to simulate the study regions’ past and future climate conditions. Three CMIP5 GCMs’ outputs were selected after extensive evaluation for their performance to simulate the climate condition of the study regions. To address the uncertainty in future concentrations of GHGs and emissions of aerosols, we used two representative concentration pathways (RCPs), which compress diverse possible future greenhouse gas and aerosol emissions scenarios (RCP 4.5 and RCP 8.5). The RCP 4.5 is similar to the lowest scenarios (B1) of ERES of IPCC AR4 and it represents a ‘medium’ emissions scenario where total greenhouse gas emissions are reduced by 2050s, while the RCP 8.5 is similar to with high emission scenarios of A1F1 of SRES of IPCC AR4 and characterized by increasing gas emissions greenhouse with more of a ‘business as usual’ scenario until the end of 2100 (San José et al. 2016). Additionally, according to Ethiopia's medium-term and long-term for sustainable development (FDRE 2011) and the word developing level, we select mid and high typical standard RCP scenarios, which were RCP 4.5 and RCP 8.5. The historical and future runs of middle and high (RCP 4.5 and RCP 8.5) datasets of rainfall and temperatures were downloaded from https://esgf-node.llnl.gov/search/esgf-llnl/ and used for climate change impact studies on water balance components. The datasets are available in a curvilinear or rotated projection with a horizontal resolution of 0.44°. The datasets were re-projected bilinearly into latitude and longitude and units of rainfall were converted from kg/m² into mm/day and the temperature from Kelvin to degrees Celsius using the Climate Data Operator (CDO) on the Ubuntu operating system. Then, the datasets were extracted for each station considered in this study using the Python programming language version 3.9. Furthermore, the datasets were grouped as baseline run from 1983 to 2005; future runs from 2021 to 2050 and 2051 to 2080, referred to as baseline, and future scenarios of the 2040s and 2070s, respectively.

Following the development of the computing power of the computer, various hydrological models have been developed to investigate the changes in climate, land use land cover, and soil properties on the hydrological cycles (Islam 2011; Devi et al. 2015; Ghonchepour et al. 2021). Hydrological models are classified based on the physical processes involved in modeling (Refsgaard & Abbott 1996). Although event-based models were developed in the 1930s, the first hydrological models for continuous simulation of rainfall-runoff processes appeared in the 1960s, when computing power allowed for a simplified ‘conceptual’ representation of the relevant land-phase processes. Later, in the 1970 and 1980s, ‘physically based’ hydrological models could be developed (Ghonchepour et al. 2021). According to Islam (2011), physically based hydrological models are based on known scientific principles of energy and water fluxes.

Nowadays, various hydrological models are ranging from lumped models such as the unit hydrograph concept to semi or fully distributed physical models, which include: Hydrologiska Byråns Vattenbalansavdelning (HBV; Bergström & Lindström 2015), Systeme Hydrologique Européen (MIKE SHE; Abbott et al. 1986), hydrologic engineering center-hydrologic modeling system (HEC-HMS; HEC-HMS 2022), and soil and water assessment tool (SWAT; Neitsch et al. 2011) are some of the widely used hydrological models. Hydrologic modelers are challenged to determine which model best is suited to a specific catchment for a given modeling exercise. According to Marshall et al. (2005), there is no single model that can be identified as ideal over the range of possible hydrological situations. There is a range of criteria to select the best hydrological model (Marshall et al. 2005; Ghonchepour et al. 2021). However, user preferences and their ability to use a particular model, the aim of the modeling, and the available time to develop and execute a model are a few of them. Additionally, the quantity and quality of hydrometeorological data as well as data on physical basin properties are associated with the selection of a hydrological model (Hughes et al. 2010). Based on this, the SWAT model was selected for this particular study of the climate change impact on the water balance of the Katar and Meki subbasins.

The SWAT is a physically semi-distributed model that is computationally highly effective and capable of modeling continuous simulation over a long period. The SWAT model was developed to predict the impact of land management practices on water, sediment, and agricultural chemicals in large and intricate watersheds with varying soils, land use, and management (Kim et al. 2008; Gassman et al. 2010; Winchell et al. 2010; Arnold et al. 2012a, 2012b). The SWAT model divides a given watershed into multiple sub-watersheds and further subdivides into the smallest unit called Hydrologic Response Units (HRUs) constituting homogeneous land use or management, soil, and slope distributions. Empirical and physical equations are embedded into the SWAT model to calculate surface runoff, evapotranspiration, infiltration, percolation, and flow of shallow and deep aquifers from each HRU and finally route the surface runoff through channels, ponds, and/or reservoirs and flow into the watershed outlet (Arnold & Fohrer 2005; Zhang et al. 2015). The SWAT model has been used for over three decades to simulate various hydrologic and environmental simulations. Nowadays, it is extensively recognized as one of the top hydrological models applied in research focused on hydrologic assessment, soil erosion and sediment transport, water quality analyses, climate, and land use changes, and watershed management impact studies (Arnold et al. 2012a, 2012b; Akoko et al. 2021). The water balance equation of the SWAT model includes precipitation, surface runoff, actual evapotranspiration (AET), lateral flow, percolation, base flow, and deep groundwater components (Neitsch et al. 2011). The model uses the modified Soil Conservation Service Curve Number (SCS-CN) method (USDA-NRCS 1986) to determine the surface runoff based on the combination of soil hydrologic group, land use, and slope and the antecedent moisture content for each HRU. In this study, the SCS-CN method for surface runoff simulations, the Hargreaves method for estimation of potential evapotranspiration, and the variable storage routing method for monthly stream flow routing were used.

The SWAT model was built based on the geospatial data of the Digital Elevation Model (DEM), soil and land use maps, physical and chemical properties of the soil, and hydrological and meteorological data. Using DEM, the watershed and sub-watershed were delineated after various steps passed through, including DEM setup, stream definition, watershed outlet selection, definition, and calculation of subbasin parameters in the SWAT model. Furthermore, the resulting sub-watersheds were divided into HRUs based on a combination of land use, soil, and slope based on the threshold values of 20% for land use, 10% for soil, and 10% for slope, which was recommended for the Ethiopian case study by Setegn (2008). Due to high topographic variability in the study regions, five slope classes of 0–5%, 5–10%, 10–15%, 15–20%, and >20% were defined.

In hydrological scientific research, the uncertainty is summarized into input data uncertainty, model uncertainty, and parameter uncertainty (Abbaspour et al. 2015). Any source of uncertainty is a bottleneck to the accuracy of the hydrological model. Several studies have shown that the ability of hydrological models to produce satisfactory predictions is based on an adequate sensitivity analysis followed by model calibration and uncertainty analysis (Lenhart et al. 2002; van Griensven & Meixner 2003; White & Chaubey 2005; van Griensven et al. 2006; Saltelli et al. 2008). The Sequential Uncertainty Fitting Version 2 (SUFI-2) algorithm can map all sources of uncertainties within the 95%PPU and is used for SWAT model calibration, validation, sensitivity, and uncertainty analysis (Abbaspour et al. 2007, 2015). In this study, the global sensitivity analysis method was used to identify the most sensitive parameters, and the significance of the sensitivity was judged by the t-value and p-value (Khalid et al. 2016). Calibration is the process of estimating model parameters by comparing the model simulation results with observed or measured data for the same conditions (Moriasi et al. 2007).

After the entire watersheds were simulated for the period 1997–2014, the first simulation was compared with the observed stream flow and the parameter range was adjusted manually. Then, automatic model calibration techniques were used until the model's performance met the rating criteria specified by Moriasi et al. (2007) for Nash–Sutcliffe efficiency (NS_E), coefficient of determination (R²), and percent of bias (PBIAS). Additionally, the P-factor and R-factor were used to measure the strength of SWAT to simulate monthly stream flow in both subbasins. The P-factor can measure how much the data is bracketed within the 95PPU and the values are varied from 0 to 1, where 1 is 100% bracketed by the measured variables within the model prediction uncertainty. The R-factor is another metric that is calculated as the ratio of the average width of the 95PPU to the standard deviation of the measured variables (Abbaspour et al. 2015). In addition to model performance metrics, these two uncertainty measures were used until the values were within the acceptable range recommended by Abbaspour et al. (2015) for flow simulation. After achieving the objective function through calibration, validation of the model followed to check the calibrated model for estimating the effectiveness of future potential management practices by testing the model against an independent set of measured data from the 2008–2014 periods.

The historical datasets of five RCMs available in the Coordinated Regional Downscaling Experiment (CORDEX)-Africa database are evaluated against ground-based observed rainfall in the Katar and Meki subbasins. The goal of the evaluation is to see how successfully RCMs recreate climate conditions in the research region, such as monthly, seasonal, and yearly rainfall cycles, and to quantify the uncertainty in downscaling the same global climate model outputs between RCMs. The ability of the RCM outputs was assessed using the normalized root mean square (NRMSE), bias, and correlation coefficient (CC). In addition to evaluating the performance, a multi-criteria decision approach of compromise programming was applied to weight the assessment metrics and rank the GCMs for each station. The details are found in Balcha et al. (2022a).

Dynamic downscaling using RCMs can eliminate some of these biases due to their smaller resolutions, which allows topography to be more accurately represented, and these models are currently considered convection-permitting at the highest resolutions. Considerable biases can persist in many circumstances, either from the driving GCM or from the RCM itself (Sørland et al. 2018). According to Mehrotra et al. (2018), simulations of climate models require BC before use in impact assessments or for statistical or dynamic downscaling to finer scales. Several BC methods are available in the literature, which outperform differently under different conditions. The performance of BC methods has been evaluated by Teutschbein & Seibert (2013), Mehrotra et al. (2018), and Enayati et al. (2021). In this study, distribution mapping and power transformation for rainfall and distribution mapping and linear scaling for temperature BC were compared for their performances, with the best being selected to correct future climate conditions.

The ability of the SWAT model to simulate monthly discharge was tested against measured discharges at Abura (Katar subbasin) and Meki Village (Meki subbasin) stations from 1997 to 2014. To optimize the SWAT model, a sensitivity analysis was performed using the global sensitivity analysis in the SWAT-CUP of the SUFI-2 algorithm (Abbaspour 2012). The most sensitive parameters with a larger absolute t-stat value and a smaller p-value (0.05) were selected for calibration of the streamflow. After identifying these parameters, streamflow calibration and validation were performed for each subbasin separately. The model performance was evaluated using Nash–Sutcliff efficiency (NS_E), coefficient of determination (R²), and percent of bias (PBIAS) during the calibration and validation periods. The simulated streamflow results show good agreement with the observed streamflow, as per the statistical indices suggested by Moriasi et al. (2007) for hydrological model performances. Specifically, for the calibration period, the values of NS_E, R², and PBIAS were 0.68, 0.73, and −17.5% at the Katar subbasin and 0.83, 0.85, and −1.6% at the Meki subbasin, respectively. Furthermore, for the validation period, the values of NS_E, R², and PBIAS were 0.67, 0.72, and −22.7% at the Katar subbasin and 0.75, 0.75, and −1.9% at the Meki subbasin, respectively. The details of the model setup, calibration, and validation processes are found in Balcha et al. (2022b).

The performance of five RCMs, which were downscaled from 22 GCM outputs, was evaluated by Balcha et al. (2022a) for stations in the Katar and Meki subbasins. Three GCMs, two of them downscaled by RCA4 and one by CRCM5, the top-ranked, were used for climate change impact assessment on the surface water resources of the study region (Katar and Meki subbasins). The details of the evaluations were presented by the authors, and the top five of the GCMs–RCMs are summarized in Tables 1 and 2 for the Katar and Meki subbasins, respectively. Three GCMs/RCMs such as MPI-ESM-LR/CRCM5, EC-EARTH/RCA4, and MIROC5/RCA4 were common for the two watersheds selected for climate change impact assessment on surface water (Tables 1 and 2).

Tables 3 and 4 show the average change in monthly maximum and minimum temperatures in the study regions for the RCP 4.5 emission scenario. Monthly maximum temperatures in the Katar subbasin will rise by 0.32–1.55 °C by the 2040s and by 0.97–2.35 °C by the 2070s. Similarly, for the Meki watershed, the maximum temperature will increase in the range of 0.28–1.57 °C by 2040 and in the range of 0.66–1.94 °C by 2070. Over the Katar subbasin, average monthly minimum temperatures will rise by 1.05–1.74 °C by 2040 and by 1.28–2.59 °C by 2070. In the Meki subbasin, the future minimum temperature will increase in the range of 1.01–2.00 °C by 2040 and 1.31–2.06 °C by 2070, respectively. For the RCP 8.5 emission scenario, monthly average maximum and minimum temperatures will increase over Katar and Meki watersheds for the periods of 2040 and 2070 (Tables 5 and 6). Averagely, monthly maximum temperatures will increase in the range of 1.29–3.8 °C by 2040 and in the range of 3.46–5.6 °C over the Katar subbasin. The monthly maximum temperature in the Meki subbasin will rise to 2.72–3.56 °C by 2040 and 3.97–5.55 °C by 2070 (Table 5). Similarly, monthly average minimum temperatures will increase in the range of 1.31–2.02 °C by 2040 and in the range of 2.69–3.5 °C by 2070 over the Katar subbasin. As well, for the Meki subbasin, the monthly average temperature will increase in the range of 0.85–2.20 by 2040 and in the range of 2.32–3.85 by 2070 (Table 6). The study result has been confirmed by similar studies by NMA (2007) and Adem & Bewket (2011) specifically as indicated by Mcsweeney et al. (2010), the average temperature will be expected to increase to 5.1 °C by 2090. Notwithstanding the increasing trend of both climatic variables, the increase in rainfall seems obscured by increases in temperature.

A non-parametric Mann-Kendall test and Sen's slope estimator are used to locate the trend and quantify the magnitude of change for observed rainfall and temperatures. From 16 stations of rainfall data, stations at Dagaga, Iteya, Bui, and Butajera exhibited decreasing trends, while stations found at Kulumsa and Ejerese-Lele presented an increasing trend in monthly rainfall. However, the rest stations have shown no significant increasing or decreasing trends in the monthly observed rainfall. Similarly, as with observed rainfall, the trend of climate model outputs was examined for the baseline and future climate conditions under RCP 4.5 and RCP 8.5. The trends of future rainfall and temperatures were tested using the MK test (Z_s) and the magnitude of the trend was quantified using the Sone's slope estimator (β) for the time series of the 2040s (2021–2050) and 2070s (2051–2080). Table 7 shows the values of Z_s statistics and the magnitude of change β for baseline and future climate scenarios. Based on the Z_s statistics, the baseline and future rainfall show increasing and decreasing trends in the annual and seasonal rainfall. However, the trend was statistically insignificant at 95% of the confidence interval except for summer rainfall under RCP 4.5 by 2070 for the EARTH climate model. Similar to rainfall, the Z_s statistics of all climate models show an increasing trend in the projected temperature under RCP 4.5 and RCP 8.5 by 2040 and 2070. Annual maximum temperatures in the Katar and Meki watersheds increased by 0.02–0.07 °C and 0.02–0.0.09 °C, respectively, by 2040 and 2070. In the summer season, maximum temperatures will increase in the range of 0.04–0.13, in winter in the range of 0.03–0.06, and in spring in the range of 0.02–0.09 °C (Table 8). The annual minimum temperature will significantly increase in the range of 0.02–0.09 °C for the climate scenarios of RCP 4.5 and RCP 8.5 by 2040 and 2070 over both watersheds. As well, the seasonal minimum temperature significantly increased during the summer season in the range of 0.02–0.06 °C, the winter season in the range of 0.01–0.06 °C, and the spring season in the range of 0.02–0.07 °C by 2040 and 2070 (Table 9).

Under this section, the main water balance component such as AET, PERC, and WYLD were described under future climate conditions. Table 10 summarizes the climate change scenarios of RCP 4.5 and RCP 8.5 to evaluate the sensitivity of water balance components to future anticipated climate changes. In this study, it was assumed that future land use change and all other variables would be constant to investigate the impact of climate on SWAT water balance components. In general, regardless of the applied scenarios, the predicted annual average water balance components decreased and increased in comparison to the baseline for Katar and Maki subbasins. The AET revealed a relative decrease or increase depending on the applied scenarios. As a result, the annual water yield (WYLD) is expected to fall in the future for EARTH and MIROC5 climate models and increase for the climate MPI-ESM-LR model over the Katar subbasin under RCP 4.5 and RCP 8.5. However, for the Meki subbasin, the annual water yield (WYLD) is likely to increase due to a decrease in AET, and deep percolation loss (PERC).

As can be seen from the table, the annual water yield for the baseline at Katar and Meki subbasins is 670 and 632 mm, respectively. It was very clear that an increase in future rainfall and temperatures would lead to a decreasing trend of the WYLD in the range of 4.9–15.3% at the Katar subbasin and 6.7–7.4% at the Meki subbasin for applied scenarios. Furthermore, the annual water yield will be increased in the range of 0.38–57.1% and 6.57–49.9% for Katar and Meki subbasins, respectively, for the RCP 4.5 and RCP 8.5 climate scenarios. As far as the knowledge of the authors, there are no studies on climate change impacts on water balance components in the study area. However, the findings of this study were consistent with previous research at the Upper Blue Nile Basin by Sitotaw et al. (2022) and in the Awash basin by Taye et al. (2018) which predicted that the annual water yield would decrease the climate scenarios used. For example, the result in Table 10 indicated a 2.57% change in rainfall (Pr), resulting in a 4.9% change in WYLD for the EARTH under RCP 4.5 and a scenario period of the 2040s at the Katar subbasin. Similarly, for the Meki subbasin, a 4.4% change in rainfall results in a 10.4% change in WYLD for the same conditions. In general, the change in rainfall depth is directly correlated with the change in the depth of water yield of the subbasins.

In the preceding section, we have discussed the spatial and temporal variability of rainfall and temperature in the study regions. Furthermore, the trends and their magnitude of change were investigated using the MK test and Sen's slope estimator for future scenarios under RCP 4.5 and RCP 8.5. Future climate change impacts on stream flow are investigated as follows. Under RCP 4.5 of the future climate scenario, the average annual stream flow will decrease by 14.1% in the 2040s and increase by 1.0% in the 2070s for the Katar subbasin. In the Meki subbasin, the average annual stream flow may decrease by 1.9% in the 2040s and increase by 6.5% in the 2070s in comparison to the baseline period. About the RCP 8.5 climate scenario, the average annual stream flow may decrease by 10.7 and 23.8% in 2040 and 2070, respectively, for the Katar subbasin, and by 2.4 and 9.5% in 2040 and 2070, respectively, for the Meki subbasin.

This study investigated the climate change on water balance components and surface water availability in the Katar and Meki subbasins, which are important river basins of Lake Ziway of the Central Rift Valley Lakes Basin of Ethiopia. After the performance evaluation of RCMs, the three best climate model outputs were selected to assess future climate change impact studies. The bias-corrected RCMs have been integrated into a semi-distributed SWAT model to assess the future climate and water resources of the Katar and Meki subbasins for the periods of 2021–2050 (the 2040s) and 2051–2070 (2070s). The results show that rainfall and temperatures in the study region are anticipated to be increased by the 2040 and 2070s under the climate scenarios of RCP 4.5 and RCP 8.5. However, the average annual stream flow will decrease by 2040s and increase by 2070s for the Katar subbasin under the RCP 4.5 scenario. Also in the Meki subbasin, the average annual stream flow may decrease by the 2040s and increase by the 2070s in comparison to the baseline period under the RCP 4.5. About the RCP 8.5 climate scenario, the average annual stream flow may decrease for both subbasins by the 2040 and 2070s.

The average annual streamflow could decrease in future periods compared with the baseline period. In the dry season, however, the possibility of evaporation in the basin is high, the amount of rainfall is small, and the stream flow is small, resulting in a small total amount of water. Overall, the declining trend of water resources in the study area suggests that water resources are insufficient to support the development of the subbasin. As a result, strategies for managing the basin's water resources are required to avoid future impacts of climate change on water availability. Climate change scenarios and simulated impacts are used to develop water resource management strategies. The study's limitation is that it does not account for future land-use and land-cover changes or the impact of socioeconomic conditions in the subbasins on water resources. Therefore, future studies should consider the combined effects of climate change, land-use/land-cover change, and socioeconomic conditions in the subbasins.

The authors would like to thank GCRF, the Ministry of Water and Electricity (MoWE), Water and Land Resource Center (WLRC), National Meteorological Service Agency (NMSA), Rift Valley Lakes Basin Authority, towns’ water supply offices (Asella, Bokoji, Sagure, Abura, Arata, and Meki), Silte, East and West Arsi zone Agriculture, and Natural Resources offices for providing relevant data.

This work was supported by the Water Security and Sustainable Development Hub funded by the UK Research and Innovation's Global Challenges Research Fund (GCRF) [grant number: ES/S008179/1].

All authors made substantial contributions to the development of this manuscript. S.K.B. oversaw the conceptualization, data collection, data analysis, and writing of the original draft. T.A.H. and G.A. reviewed, edited, and improved the manuscript. A.A.A. and A.B. supervised the overall research work of this study. All authors read and approved the final manuscript.

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

The authors declare there is no conflict.