For a country whose economy is highly reliant on rainfed agriculture, it is particularly important to study and evaluate water resources at both spatial and temporal scales. This includes assessing not only the current availability of water but also considering future conditions as much as possible. Ethiopia, for instance, boasts an estimated water potential of around 122 billion cubic meters (BCM), with the Awash basin alone holding 4.9 BCM, accounting for approximately 4% of the country’s total potential (Awulachew et al. 2007). Within the broader Awash Basin, the Upper-Awash Sub-Basin (UASB) emerges as a key player. It contributes significantly to the main Awash River due to its comparatively higher magnitude of rainfall (Daba and You 2020)(Mulugeta et al. 2019).
The UASB holds significant strategic importance for the country due to its geographic location and vibrant socio-economic activity. The basin, housing the capital city and serving as a hub for most of the largest industries in the countries, witnesses intense socio-economic activity. The basin is home to over 6 million people, with the majority residing in the capital city and nearby towns (CSA 2013). In these areas, people are predominantly engaged in industries, both public and private sectors, while farming remains the dominant economic activity outside the cities and towns. The economic disparity results in high growth and a related population increase in cities and towns. Despite various water resource developments and future plans in the sub-basin, mainly focused on domestic water consumption, any development in these areas is expected to create water stress downstream, emphasizing the importance of studying past, present, and future water resources in the sub-basin.
Several approaches are commonly employed to study water resources in specific areas. The first approach involves analyzing long-term variations in runoff and meteorological elements. This can be achieved through statistical analysis of the relationship between runoff and other meteorological variables or by examining past extreme events. For instance, Vicente-Serrano et al. (2021) utilized SPI indices developed from monthly precipitation data to observe long-term drought trends in western Europe. Similarly, Han et al. (2016) utilized 54 years of water level data from a lake in China to assess long-term trends using various statistical tests. Further, Abraham and Kundapura (2022) used 68 years of rainfall and temperature data to study the temporal rainfall and temperature trend by applying extreme indices together with trend tests at annual and seasonal time scales.
The second approach employs a water balance method over an extended period. Daly et al. (2019) proposed a framework describing long-term catchment-scale water balance based on key measured variables like rainfall, streamflow, and potential evapotranspiration, demonstrating its applicability for quantifying water balance in ungauged catchments. Additionally, Wang et al. (2021) employed the water balance approach within the SWAT model to analyze the impact of different proportions of impervious urban areas on long-term water balances in the Yamoto River catchment, Japan.
The third approach involves studying estimates of changes in climatic and hydrologic characteristics for large regions using Global Climate Model (GCM) outputs combined with hydrological models. For example, Gebrechorkos et al. (2023) used meteorological variables from seven CMIP6 Global Climate Models (GCM) together with Variable Infiltration Capacity and Vector-Based routing models to simulate runoff and streamflow for 68,300 river reaches in East Africa.
Another common alternative nowadays is the use of deterministic hydrologic models. Baker and Miller (2013), for instance, employed SWAT model to understand watershed response to land use changes. Similarly, Knebl et al. (2005) utilized HEC-HMS to estimate overland flow and channel flow for a catchment area in Central Texas, USA, covering 10,000 km2. Additionally, Data Driven Models (DDM) ranging from the classical models like Autoregressive Moving Average (ARIMA) to Artificial Neural Networks (ANN’s) which has high importance in quick runoff prediction and forecasting are also considered deterministic models.
Historically, numerous efforts have been made to study water resource conditions in the basin. The earliest, by Kinfe (1999), used a water balance model to explore the impact of climate change on runoff estimation in the Awash basin using climate models under the CMIP3 archive. The study concluded a decrease in future water resource potential for all scenarios and time periods. Subsequently, Taye et al. (2018) investigated water availability using climate models under CMIP5 until the end of the 21st century. This study estimated changes in water resource availability using a change factor (CF) developed by subtracting evaporation from precipitation at the monthly level. An increase in water availability was predicted for recent and mid-time periods for all scenarios, while a decrease was predicted for the far-time period. Heyi et al. (2022) applied the HBV model together with the HadCM3 climate model, an output of earlier works under CMIP3, indicating a decrease in rainfall during JJAS and an increase in MAM for two scenarios and all time periods. A recent study by Chelkeba et al. (2023) applied the SWAT model with four CORDEX RCMs from the CMIP3 archive on the Akaki Catchment, aligning with the results obtained by Taye et al. (2018).
In this study, similar efforts will be undertaken to study future monthly streamflow availability in UASB by using SWAT and NARX models together with an ensemble of climate model outputs obtained from the CMIP6 archive. It is thought that models incorporated under CMIP6 show comparable or improved capabilities in terms of simulating global climate compared to the older CMIP5. Therefore, the aim of this study is to shed light on the future water resource conditions over the sub-basin with the help of new climate model outputs under CMIP6 and two hydrological models with different modeling assumptions. It is known that SWAT is a more complex hydrological model requiring a greater number of input variables than NARX model. Therefore, this study in addition will also try to evaluate the NARX model’s capability in generating comparable future streamflow projections.