Ensemble modeling of the hydrological impacts of climate change: A case study of the Baro River sub-basin, Ethiopia

The Earth's climate has been experiencing a warming trend over the past 50 years (IPCC 2013), and this warming has had a considerable impact on the availability of water, human health, and food security (Gelete et al. 2020). One of the most noticeable effects of climate change is the increase in the frequency and severity of floods and droughts, as well as alterations in water supplies brought on by changes in temperature, precipitation patterns, soil moisture, humidity, runoff, and other hydrological cycles (Worku et al. 2021). Elevated temperatures accelerate the global water cycle, increasing the likelihood of extreme hydrological events and changing water resources are distributed globally at various scales (Chen et al. 2015). These hydrological changes have far-reaching consequences on many aspects of human well-being such as the availability of water, agricultural productivity, and the production of energy.

Recently, the impacts of climate change on hydrological processes and the availability of water at various spatial scales have been the subject of several research works. According to Coulibaly et al. (2018), the impact of climate change on water resources in the transboundary watershed of West Africa is projected to decline in the 21st century as a result of decreasing precipitation. In their 2018 study, Bodian et al. (2018) looked at the effects of climate change on the two main river basins in the Senegal, West Africa. The findings from multiple models revealed a decline in annual streamflow under both RCP4.5 and RCP8.5 scenarios. In the Blue Nile Basin of Ethiopia, it is predicted that precipitation will increase by 7–48% and the streamflow is estimated to undergo an increase ranging from 21 to 97% (Roth et al. 2018). Likewise, Tigabu et al. (2021) investigated how climate change may affect different aspects of the water balance in the Ethiopian watershed of Lake Tana. The study reported a slight change in rainfall but a decline in the number of rainy days in the Tana sub-basin. Overall, the various impacts of the climate change can be linked to differences in geographical locations, climatic features of river basins and differences in emission scenarios used.

The way that different studies use simulations from climate models varies. Some studies used the outputs of regional climate models (RCMs) and global climate models (GCMs) directly (Beyene et al. 2010; Setegn et al. 2011). The output from a climate model, however, must first go through some sort of preprocessing to remove any biases that may be present before being used directly for hydrological simulation (Haerter et al. 2011). Some studies have utilized bias-adjusted and downscaled climate model results before incorporating them into hydrological evaluations (Bodian et al. 2018; Galata et al. 2021). To improve the accuracy of their simulations, climate models should, therefore, undergo downscaling and bias correction before being used to evaluate hydrological implications (Pierce et al. 2015).

Ethiopia experienced climate-related hazards such as floods, droughts, intense rainfall, and heatwaves based on historical records. For example, the country experienced significant flood events resulting in the loss of life and damage to property in various regions during the years 1988, 1993, 1994, 1995, 1996, and 2006 (NMA 2007). Numerous studies have been carried out at the watershed and basin scales to determine how climate change may affect water resources. For example, Taye et al. (2018) used three global climate models (GCMs) from the Combined Inter-Model Comparison Project Phase 5 (CMIP5) to examine the impacts of climate change on water resources of the Awash Basin. The study found that there is an increase in water deficiency in all seasons due to the projected decrease in precipitation and increase in temperature. Chaemiso et al. (2016), on the other hand, projected an increase in surface runoff and total water yield for the 2030 and 2090s in the Omogibe River Basin, Ethiopia. The varying effects of climate change on water resources across basins highlight the significance of understanding the hydrological implications using appropriate climate change scenarios. Multiple studies have evaluated the Upper Blue Nile Basin's water resources and projected both positive and negative effects of climate change on the basin. For instance, Roth et al. (2018) used three models with the Soil and Water Assessment Tool (SWAT) hydrological model and reported an increase in precipitation and streamflow from the Blue Nile. Similarly, Wagena et al. (2016) reported an increase in mean annual flows for the Tana and Beles basins under climate change. On the contrary, the reduction in annual precipitation and increase in temperature reported on Jemma sub-basin of the Upper Blue Nile Basin projected a reduced surface runoff and water yield (Worku et al. 2021). Overall, it was observed that the hydrological process would vary as a result of the projected changes in temperature and precipitation. However, the degree of the prevailing change in temperature and precipitation determines how much has changed.

The Baro River sub-basin experiences significant climate variability, leading to frequent fluctuations in hydrological process due to the impacts of climate change (Muleta 2021; Mengistu et al. 2023). Both studies reported increasing temperature projections, however, precipitation projections have not shown a systematic increase or decrease (Muleta 2021) and Mengistu et al. (2020) reported a decrease in precipitation. Moreover, both studies lack up-to-date information on the current climate scenarios. For example, the climate change studies are developed based on the Special Report on Emission Scenario (SRES) based on A1B storyline (Muleta 2021) and RCP4.5 and RCP8.5 under CMIP5 (Mengistu et al. 2020). It is evident that the precipitation projections from the two studies show discrepancies for the same area. It will be difficult to understand the hydrological process of the Baro-Akobo River basin under such discrepancy. Therefore, comprehensive investigation is required to establish the projections of climate variables selecting the best-performing multi-model outputs. Especially, in areas with complex topography, a single model is not representative as precipitation is highly affected by orography. One way to reduce climate model discrepancies is by using multi-model GCMs with the most recent state-of-the-art climate models, properly evaluating the strengths and weaknesses of individual and ensemble models. Therefore, to completely understand how the basin's hydrology responds to various climate scenarios, it is necessary to simulate the Baro River sub-basin's hydrological processes using the most recent Coupled Model Intercomparison Project Phase 6 (CMIP6) developments.

In the Baro River sub-basin, rainfall projections show significant seasonal variation, with summer and autumn experiencing a significant increase and winter and spring seeing a decrease. An evident consequence of this fluctuation will be the likelihood of hydrologic extremes, such as drought and flooding, over the basin due to the significantly increased evapotranspiration (ET) (Muleta 2021). The livelihoods of most of the inhabitants are based on rain-fed agriculture and are often affected by climate change. This suggests that smallholder farmers' livelihoods would be significantly affected by climate change, and thorough climate impact assessments are needed to inform climate change adaptation strategies. In order to reduce the uncertainties associated with particular GCMs, this study used models that are considered atypical or least realistic within a specific study area as demonstrated in Gebisa et al. (2023). By utilizing spatial, seasonal, and multivariate statistics, Gebisa et al. (2023) employed the multi-mean ensemble to select representative models.

The significance of this study stems from its use of the advanced Phase 6 CMIP6 model for the Baro River sub-basin. Initially, the research carefully selected the top-ranked GCMs, as described by Gebisa et al. (2023). This combination of modeling approaches provides a thorough understanding of the hydrological dynamics in this specific region, deepening the study's findings. In view of the challenges caused by climate change in various regions, this study is especially important because it provides a thorough examination of the hydrological components of the Baro River sub-basin. The assessment of climate change impacts on hydrological components at the watershed and sub-basin levels will provide valuable assistance to both governmental and non-governmental organizations engaged in watershed planning and management.

The meteorological stations utilized in this study include the following nine locations: Alemteferi, Yubdo, Gimbi, Dambidollo, Bure, Uka, Gore, Masha, and Matuhospital. However, three stations have a poor temperature record with missing values of more than 10%. Therefore, only six stations are used for temperature (Alem Teferi, Gore, Masha, Matu Hospital, Yubdo, and Dambidollo), while all stations are used for precipitations. Detailed information on the meteorological stations, including latitude, longitude, and elevation, along with weather variable records, is presented in Table 1. Solar radiation, wind speed, and relative humidity data are available only at one station (Gore).

The topography of the watershed was established using the 30 × 30 m digital elevation model (DEM) in order to identify the watershed boundary, define the stream network, and generate sub-basins. Moreover, DEM facilitated the extraction of essential sub-basin parameters such as terrain slope length, slope gradient, stream network, slope classes, and channel lengths. The DEM data used in this study was obtained from the United States Geological Survey (USGS) at https://earthexplorer.usgs.gov/.

Daily flow data are required to evaluate the performance of the SWAT model. The daily streamflow data of the Baro River sub-basin at the Gambella gauging station from 1985 to 2009 were collected from the Ministry of Water and Energy's Hydrology Department, and observed meteorological data were collected from the National Meteorological Agency (NMA) of Ethiopia from 1985 to 2014. The observed streamflow data were used for calibration and validation from 1990 to 2002 and 2003 to 2009, respectively, with a warm-up period of three years (1987–1989).

Following the evaluation of global climate models (GCMs), three models that exhibited superior performance were chosen (Gebisa et al. 2023). According to Gebisa et al. (2023), a detailed analysis of the GCMs evaluations revealed that the ensemble mean of three GCMs outperformed the individual GCMs and the ensemble of all GCMs. The authors downscaled the GCMs to better suit specific applications, selecting the best-performing model based on multi-evaluation criteria. Accordingly, GCMs used in this study are GFDL-CM4, INM-CM4-8, and INM-CM5-0 for precipitation and minimum temperature and CMCC-ESM2, MRI-ESM2-0, and INM-CM4-8 for maximum temperature under the SSP2-4.5 and SSP5-8.5. In this study, the hydrological responses of the basin to climate change impacts were modeled for the future period spanning 2031–2060, with the baseline period covering 1985–2014. While considering future scenarios, other weather variables like solar radiation, relative humidity, and wind speed in the historical period were maintained without alteration as changes in these variables might not have a substantial impact on the local hydrology compared to the precipitation and temperature (Arnell 2003; IPCC 2014). Moreover, baseline data for solar radiation, wind speed, and relative humidity are available only for limited time periods and from a few stations in Baro River sub-basin.

Despite the ensemble mean and individual GCMs' satisfactory performance, it was found that three of them showed clear biases. Therefore, it is essential to examine and reduce these biases before beginning a study on how climate change will affect hydrology. Linear scaling (LS) for precipitation and distribution mapping (DM) for temperature were used for bias correction in this study. More comprehensive information regarding the GCM models, downscaling method, and bias correction technique can be found in the work conducted by Gebisa et al. (2023).

where SW_t is the soil's final soil water content (mm), SW_o is the initial soil water content on day i (mm), t is the time (days), R_day is the amount of precipitation on day i (mm), Q_surf is the amount of surface runoff on day i (mm), is the amount of ET on day i (mm), W_seep is the amount of water entering the vadose and Q_gw is the amount of return flow on day i (mm) (Neitsch et al. 2005).

The SWAT model's simulation of the hydrological process begins with the definition of watersheds and the construction of streamflow networks. Using a 30-m resolution DEM, the Baro River sub-basin was divided into 21 sub-basins, and a multiple HRU was established. Consequently, 309 HRUs were produced. With the input of weather data from five stations, the SWAT model setup was then prepared for the first simulation, which was utilized for model evaluation.

Assessment of the impact of climate change on water resources is made possible by coupling the SWAT hydrologic model with the global climate models (Saharia & Sarma 2018). The model's performance for the river basin was examined before looking at the results of the SWAT simulation. The parameters of SWAT model were calibrated and validated using historical data, and they were assumed extrapolative under future scenarios.

Finding the most important factors among the model parameters is essential for enhancing the calibration of the hydrological model in SWAT. Sensitivity analysis is essential in identifying the extremely sensitive variables that have a big effect on the flow process. For the SWAT, a number of calibration techniques have been developed with the SWAT Calibration and Uncertainty Program (SWAT-CUP), with Sequential Uncertainty Fitting 2 (SUFI-2) being the most extensively used. This calibration process incorporates a variety of objective functions to enhance the accuracy of the model (Sao et al. 2020).

The process of model calibration involves adjusting model parameters to align with observed data while allowing for a certain level of variability. To reduce uncertainty in calculating process parameters, a variety of strategies are used during calibration. Validation includes the comparison of model outcomes with an independent dataset, without making any additional adjustments to the parameter values. A warm-up period from 1987 to 1989, a calibration period from 1990 to 2002, and a validation period from 2003 to 2009 were used to evaluate the performance of the hydrological model.

A set of 23 hydrological parameters were selected to identify the sensitive parameters for the calibration and validation of stream flow in the SWAT-CUP. The 23 parameters were run through a program with 500 simulations using the global sensitivity method to identify which ones were the most influential ones. The programme used a t-test and p-value to pinpoint the parameter that had a direct impact on the stream flow response. A p-value close to zero and a larger absolute value of the t-value indicates higher sensitivity of the parameter. Following the iterations, 12 parameters were identified as the most sensitive parameters (Table 2).

Curve number value (r_CN2.mng) followed by Plant Uptake Compensation factor (v_EPCO.hru) and Deep Aquifer Percolation fraction (v-RCHRG_DP.gw) are the top three most sensitive parameters.

Based on the results of the global sensitivity study, the 12 most sensitive parameters, listed in Table 2, were optimized using the SUFI2 algorithm in the SWAT-CUP. The majority of observed data is covered by the SUFI2 method, which includes all uncertainties and translates them onto parameter ranges to achieve a 95% prediction uncertainty. These uncertainties encompass various aspects such as model input, conceptualization, model parameters, and measured data (Abbaspour et al. 2004).

Model validation was carried out following model calibration. Model validation was done to determine how accurate the model was while replicating times that were not covered during calibration. By performing model validation, the suitability and reliability of the hydrological parameters derived during the calibration process were determined. For model validation, a 7-year period from 2003 to 2009 was analyzed. The average monthly flow validation results, shown in Table 3, include R², NSE, PBIAS, and RSR values.

This demonstrates that the validation outcomes for stream flow simulation with the SWAT model were satisfactory. Although the data used to evaluate the SWAT model during calibration and validation showed very good agreement between simulated and observed streamflow, the model was unable to reproduce peak flows in some of the years considered for calibration and validation. On the other hand, the SWAT model overestimated base flow in the majority of the calibration and validation periods considered. A similar case was reported by Dibaba et al. (2020), who noted that the SWAT model was unable to accurately simulate peak flow due to the second storm effect. In this study, we established a comprehensive model evaluation range for a monthly time step by incorporating performance ratings from the recommended statistics as proposed by Moriasi et al. (2007).

To assess the potential influence of climate change on the hydrological process of the basin, changes in temperature and rainfall patterns reported in Gebisa et al. (2023) were used. The projected temperature and rainfall values were used as inputs into the previously calibrated and validated SWAT model to simulate the impacts of climate change on the hydrological processes. The increase in the projected annual temperature and precipitation were found to have a substantial impact on the hydrological process of Baro River sub-basin under both SSP2-4.5 and SSP5-8.5. Even though the magnitude varied, both climate scenarios projected an increase in surface runoff, water yield, ET and PET and a decrease in groundwater. Table 4 shows that under the SSP2-4.5 scenario, there is an annual change of +100.43 mm (30.33%) in SURQ, −24.43 mm (−13.17%) in GWQ, +57.99 mm (6%) in WYLD, +34.8 mm (4.85%) in ET, and +45.1 mm (4.49%) in PET. Conversely, under the SSP5-8.5 scenario, the annual change shows +186.11 mm (44.67%) in SURQ, −5.39 mm (−2.64%) in GWQ, +199 mm (18.01%) in WYLD, +76.7 mm (10.1%) in ET, and +68.1 mm (6.63%) in PET. The findings regarding the impact of climate change on various hydrological processes are summarized in Table 4.

In light of the implications of future climate change, it is projected that the overall water yield will increase both annually and seasonally as shown in Figure 5(b). The total annual water yield is projected to increase by +6 and +18% under SSP2-4.5 and SSP5-8.5, respectively. In the JJA and SON seasons, the projected water yield is expected to increase by 6.87 and 8.6% under SSP2-4.5, and by 16.3 and 20.01% under SSP5-8.5, respectively. Similarly, the water yield in the DJF and MAM seasons is projected to change by −4.63 and 0.1% under SSP2-4.5, and by 14.17 and 20.76% under SSP5-8.5, respectively. The projected precipitation under SSP5-8.5 is greater than under SSP2-4.5. Consequently, the water yield during the SON season under SSP5-8.5 is almost similar to the water yield during the MAM season under SSP5-8.5. Therefore, the change in precipitation is the most significant factor influencing these variables. The total water yields are impacted by elements including surface runoff, groundwater flow, and water loss reduction when groundwater flow drops in particular months. The expected increase in rainfall and streamflow could have a number of outcomes. On the one hand, it might offer an abundance of water resources to support the basin's expanding population. While a decrease in rainfall during the minor rainy season increases the likelihood of water scarcity and stress, projections for increased rainfall during the major rainy season raise concerns about a heightened risk of flooding. This study closely resembles the work of Tessema et al. (2021) conducted in the Kesem sub-basin and similar areas in Ethiopia, providing valuable insights to support current and future water resource management in the Baro River sub-basin. In a comparable scenario, this study exhibits similarities with the research conducted by Wagena et al. (2016), emphasizing both the significant variability in climate change impacts and the potential for positive outcomes in the region.

The projection of seasonal and annual groundwater (GWQ) showed a decreasing trend under both SSP2-4.5 and SSP5-8.5 scenarios. The annual projected GWQ decline under SSP2-4.5 and SSP5-8.5 scenarios showed a decrease by 13.17 and 2.64%, respectively (Figure 5(c)). Although the decrease in GWQ with rising precipitation may seem contradictory, increased SURQ with higher precipitation may reduce GWQ recharge depending on soil conditions. Furthermore, larger increases in temperature and PET may result in higher rates of ET which contributes to the high rate of water loss. This could lower the amount of accessible water for GWQ recharge. Consequently, the increase in surface runoff and decrease in infiltration could lead to a reduction in groundwater. This finding is consistent with the study by Tigabu et al. (2021), who reported that increased runoff, along with reduced infiltration, led to reduced groundwater.

While the direct measurement of temperature's impact on groundwater may be challenging, it can be deduced from other hydrological factors such as ET. On the other hand, one of the factors contributing to the decrease in groundwater is soil moisture dynamics. Increased temperatures can alter these dynamics, leading to drier soils, which in turn can reduce the infiltration rates of precipitation into the ground and hence lower groundwater recharge (Vereecken et al. 2008). Positive or negative changes in one hydrological component might have an impact on other components in the system. Although temperature does not directly affect ET, it does have an impact on PET. However, the actual ET also depends on the availability of water, which can be limited, particularly during certain seasons. Additionally, higher rainfall intensity exacerbates the situation by promoting rapid overland flow and reduced infiltration. While overall precipitation may increase, changes in its distribution, intensity, and frequency can impact groundwater recharge. More frequent heavy rains can result in increased surface runoff, which does not effectively infiltrate the ground (Vereecken et al. 2008). Similar findings related to the impacts of climate change have been reported in different parts of Ethiopia (Mengistu et al. 2020; Tigabu et al. 2021). According to Tigabu et al. (2021), the projected decline in groundwater flow was reported along with an increase in surface runoff in Lake Tana. Mengistu et al. (2020), on the other hand, found that the total water yield of the Upper Blue Nile basin was projected to decrease despite the increase in surface runoff. In future scenarios, higher temperatures and changing precipitation patterns may increase water demand for agricultural, industrial, and domestic uses. This could lead to higher rates of groundwater extraction, further depleting groundwater reserves in the river basin.

PET and actual ET can be influenced by the projected changes in temperature and precipitation. As a result, a rise in temperature may cause ET and evaporation rates to increase, which could hasten the depletion of groundwater supplies. Figure 5(e) shows that annual PET was increased by +4.49 and +6.63% under SSP2-4.5 and SSP5-8.5, respectively.

The annual ET was increased by +4.85 and +10.1% under SSP2-4.5 and SSP5-8.5, respectively, as shown by Figure 5(d). The rise in PET and ET is more pronounced under SSP5-8.5 compared to SSP2-4.5, due to the higher temperature change associated with SSP5-8.5. ET is a vital component of the hydrological cycle. It is crucial to account for this natural process when assessing agricultural water demand and determining the practical storage capacity of a dam (Adib et al. 2023). The current study shows an increase in annual PET across all periods and SSPs, possibly due to a rise in temperature. Similarly, ET is expected to increase alongside a rising temperature, although this is limited by water availability. The increase in ET during JJA is 2.61% under SSP2-4.5 and 4.43% under SSP5-8.5, while in DJF, ET increases by 11.55 and 27.02% under SSP2-4.5 and SSP5-8.5, respectively. ET is greater during the dry season than the wet season as shown in Figure 5(d). The results are consistent with the findings of various researchers in Ethiopia (Daba & Nageswara 2016; Seneshaw et al. 2020). For instance, Seneshaw et al. (2020) reported that ET is projected to rise under all scenarios as temperatures increases. Another study in the Upper Blue Nile basin showed that higher precipitation and a rising temperature contributed to an increase in evaporation (Getachew et al. 2021). Overall, the interaction between precipitation, temperature, and ET serves to illustrate the impact of climate change on a basin's hydrology. An increase in precipitation amplifies surface runoff, while a rise in temperature leads to increased PET and decreased groundwater levels within the hydrological system.

Although the impacts of climate change is projected to increase surface runoff, water yield, and ET, while decreasing groundwater in the Baro River sub-basin, each sub-basin exhibits distinct yield characteristics. While precipitation changes have direct impact on the hydrological process, changes in temperature are predicted to have an indirect impact.

The highest rise in ET is shown by SSP2-4.5 around the south and northwest of the basin, while south and the east of the basin was projected under SSP5-8.5. The lowest increase was projected around the north under SSP2-4.5 and in the west under SSP5-8.5. In the projected climatic change for this study, ET increased at a higher rate under SSP5-8.5 than under SSP2-4.5. The projected future climate was marked by higher rates of evaporation and more seasonal water loss, which affects the groundwater and water yield. This is in agreement with previous studies. Earlier studies show that the ET increases (Ungtae & Kaluarachchi 2009; Dibaba et al. 2020). The study by Dibaba et al. (2020) reported that the increase in ET was higher under RCP8.5 compared to RCP4.5. The expected temperature rises in future climate scenarios are probably what caused the observed hydrological alterations.

Overall, this study assessed the effects of anticipated climate change, assuming constant LULC for both historical and future periods. When examining future scenarios, essential weather variables such as solar radiation, relative humidity, and wind speed from the historical period were held constant without modification. It is crucial to note that in reality, variations in these weather variables and alterations in LULC occur due to both natural processes and human activities. Consequently, a more comprehensive understanding can be achieved by conducting research that considers changes in these weather variables and LULC, thereby enhancing the reliability of the study's findings.

Following an evaluation of the model performances of the global climate models (GCMs), which are components of CMP6, this paper evaluated the ensemble modeling of the hydrological impacts of climate change on the Baro River sub-basin. For the analysis of the climate change impacts, GFDL-CM4, INM-CM5-0, and INM-CM4-8 were used for precipitation and minimum temperature while CMCC-ESM2, MRI-ESM2-0, and INM-CM4-8 were used for maximum temperature.

The study demonstrated the significance of using the models that are perceived as atypical or least realistic in a given study region in order to reduce the uncertainty associated with the individual GCMs. Consequently, the multi-mean ensemble of three GCMs was employed in the study after the representative models had been determined based on spatial, seasonal, and multivariate statistics. Using a bias-corrected ensemble of the GCMs, the hydrology impacts of climate change were conducted with a calibrated and validated SWAT model. Although the SWAT model was capable of simulating the rising and recession limbs of the hydrographs, it was unable to simulate the peak flow. The results of the ensemble GCMs showed an increasing precipitation and temperature under both climate scenarios used in this study. Consequently, annual and seasonal surface runoff is projected to increase under both SSPs. The increasing surface runoff is related to the increase in precipitation. Contrarily, the annual decrease in groundwater flow (GWQ) was evident under SSP2-4.5 and SSP5-8.5 scenarios. Both ET and PET were increased, and raising PET resulted in a decrease in soil moisture due to rising temperatures.

The special distribution of the hydrological process under the climate change shows that the projected increase in surface runoff and water yield is highest in the northern and southern parts of the basin, under both SSP2-4.5 and SSP5-8.5 scenarios. Groundwater decline is significantly more pronounced in the southeast under both SSP scenarios.

The study finding provided crucial details on the relative influences on how the hydrological process responds to climate change scenarios. This is an important contribution as it could help to plan proper water resources management interventions. The outcome emphasises the necessity of regional cooperation and development to promote effective management techniques that are climate resilient and to slow down the study area's accelerating climate change.

The authors express their gratitude to Bule Hora University and Jimma University for hosting and supporting the study. We would especially like to acknowledge the Ethiopian National Meteorological Agency for providing the data on the observed climate and the Ministry of Water and Energy for providing the data on the observed stream flow. Additionally, the authors extend their appreciation to the Working Groups of the World Climate Research Program for making CMIP6 available in the Earth System Federation (ESGF) Archive.

The authors confirm that they did not receive any funds, grants, or other support during the preparation of this manuscript.

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

The authors declare there is no conflict.