Introduction

Climate change is indeed the alteration of climatic variables. In the field of climate studies and hydrology, precipitation and temperature are the two main important climate change variables (IPCC 2007). The high concentration of greenhouse gases in the atmosphere promotes a rise in the temperature and variability in rainfall distribution (Houghton et al. 2001). The concentration of atmospheric greenhouse gas, mainly \({\mathrm{CO}}_{2}\), has increased significantly over the last centuries, raising the prospect of rising temperatures (UNFCCC 2007).

Climate fluctuation is also a significant barrier for Ethiopia, inhibiting poverty reduction and sustainable development initiatives (NMA 2007). Ethiopia is ranked as the most vulnerable country to the consequences of climate change due to its economy's reliance on rainfed agriculture, inadequate adaptation ability, and increased reliance on natural resources for subsistence (NMA 2007; World Bank 2010; EPCC 2015). Because climate change has a significant impact on the world's revenue at present moment, it has become the primary subject of study to develop the best adoption strategies based on scientific discoveries.

Changes in rainfall and temperature influence surface hydrology and water resources and can alter regional water balances and hydrological regimes. Increasing temperatures impact the occurrence, quality, distribution, and flow of water resources, as well as the frequency of hydrological events such as floods and droughts at the watershed level. Rainfall variability is the most frequently noted factor in nations whose economies rely on rainfed agriculture for socioeconomic challenges such as food security (Cheung et al. 2008). As a result, examining the spatiotemporal dynamics of these climate change variables is crucial for giving feedback to policymakers and practitioners to assist them to make educated decisions.

The Upper Blue Nile River Basin plays a significant role in the streamflow of the Nile River, which contributes above 60% of the Nile’s water (Conway 2000; Conway, 20,005). Disputes are rising between the upstream and downstream water users of the Nile River. Ethiopia is building the Grand Ethiopian Renaissance Dam (GERD), but Egypt is dissatisfied due to concerns about reduced water flow. It is true downstream countries will impact due to runoff variability, and it is expected it causes conflict if a water shortage occurs. Hence, studying how future climate change affects the streamflow of the Nile River helps to set possible solutions to reduce the conflicts between upstream and downstream water users.

Global climate models (GCMs) and regional climate models (RCMs) are used to better understand and predict climate behavior on longer time scales (Reddy et al. 2018). Various climate models are available at different spatial and temporal scales and provide climatological data that is useful in the hydrological modeling under climate change impact assessment. There is a lack of detailed information on the impacts and vulnerabilities of the Upper Blue Nile Basin to climate change by applying ensemble new shared socioeconomic pathways (SSPs) scenarios and the Coupled Model Intercomparison Panel (CMIP-6).

The newly introduced shared socioeconomic pathways (SSPs) is the latest iteration of scenarios, used for Coupled Model Intercomparison Project 6 (CMIP-6) and the IPCC Sixth Assessment Report (IPCC 2021). But, most of the previous studies in the Upper Blue Nile Basin used CMIP-3 and CMIP-5. For example, Mekonnen and Disse (2018) used HadCM3 (Hadley Centre Climate Model 3) from CMIP-3 using A2a and B2a scenarios and canESM2 from CMIP-5 GCMs under RCP2.6, RCP4.5, and RCP8.5 scenarios to analyze the future climate change in the Upper Blue Nile River. The Gilgel Abay River, Upper Blue Nile Basin, hydrological response to climate change studied by Dile et al. (2013) used HadCM3 Global Circulation Model (GCM) scenario data. Worqlul et al. (2018) used HadCM3 of A2 and B2 emission scenarios to study the influence of climate change on streamflow hydrology at the Beles basins in the Upper Blue Nile Basin. Roth et al. (2018) studied the effects of climate change on water resources of the Upper Blue Nile Basin, Ethiopia using CMIP3. Gebre and Ludwig (2015) used the emission scenarios based on IPCC’s fifth assessment report (AR5) to study the hydrological response to climate change of the Upper Blue Nile River's four catchments—Gilgel Abay, Gumer, Ribb, and Megech. Chakilu et al. (2022) studied climate change responses of streamflow in the four gauged watersheds (Gilgel Abbay, Ribb, Gumara, and Megech) of the Lake Tana Basin in the Upper Blue Nile Basin used CMIP-5 and used only a single RCP8.5. Ayalew et al. (2021) studied the potential impact of climate change on the hydrology of Ribb catchment, Lake Tana Basin, Ethiopia through using the CMIP-5 under RCP4.5 and RCP8.5 scenarios. Takele et al. (2022) studied future climate change and its impacts on water resources in the Upper Blue Nile Basin by using regional climate model (RCM) under the RCP4.5 and RCP8.5 scenarios. Worku et al. (2021), Wubneh et al. (2022), and Bekele et al. (2021) also used RCM under Assessment Reports 5 (AR5).

Only a few authors are used the newly introduced CMIP-6 in the research area. Alaminie et al. (2021) investigated trend analysis of future rainfall and temperatures under four CMIP6-SSPs scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5, and SSP3-7.0) but only a single model output (MRI-ESM2-0 for temperature and BCC-CSM-2MR for precipitation) is used. Besides, no research works have been done at our case study watershed by using the newly introduced CMIP-6 under the shared socioeconomic pathways (SSPs).

In the current study, precipitation and temperature scenarios were generated from three downscaled climate model data outputs: ACCESS_ESM1-5, FGOALS_g3, and GFDL_ESM4) with two shared socioeconomic pathways: SSP2-4.5 and SSP5-8.5. The shared socioeconomic pathways (SSPs) is the latest iteration of scenarios, used for Coupled Model Intercomparison Project 6 (CMIP6) and the IPCC Sixth Assessment Report (IPCC 2021).

Hydrological modeling is used to study the future climate change impact on hydrology in comparison to the reference line. The purpose that the model must fulfill has a significant role in the selection of an appropriate model structure. The best hydrological model for a given issue may be selected using a variety of factors. Due to the unique demands and objectives of each project, these criteria are always project-specific. Additionally, some criteria depend on the user, making them subjective. According to Cunderlik (2003), four standard, essential questions must always be addressed among the different project-dependent selection criteria.

  • The model should estimate the required model outputs since they are crucial to the project (does the model forecast the variables needed for the project?)

  • To accurately predict the intended outputs, it is necessary to represent some aspects of hydrology (e.g., can the model simulate discrete or continuous processes?).

  • Accessibility of input data (can all the inputs needed by the model be supplied within the project's time and budget constraints?)

  • Price (does it seem like the investment will be worthwhile given the project's goals?) The model was chosen by taking into account the aforementioned factors, as well as the data's availability, degree of application, purpose, needed precision, space and time scale, catchment area, simplicity, prior trends (studies) in the neighborhood, and Ethiopia overall. Physical-based models (SWAT), which took into account all of the aforementioned criteria, were chosen for this investigation.

The following factors also led to the SWAT model being chosen for the Kessie Watershed:

  • The model is calibrated and simulated with good results in the Upper Blue Nile Basin, according to information provided and tested by several academics and journal publications (for example, Takele et al. 2022; Chakilu et al. 2022; Ayalew et al. 2021; Roth et al. 2018; Dile et al. 2013; Setegn et al. 2008). The case study watershed, the Kessie Watershed, is located in the Upper Blue Nile Basin which has which does not have a significant difference in land use, climatic, and topographical conditions.

  • The model has been shown to have a clear benefit as a tool for hydrological modeling, including modularity, computational effectiveness, the capacity to forecast long-term effects as a continuous model, and the capacity to employ easily accessible worldwide information. One of the most commonly used hydrological models globally is made possible by the availability of trustworthy user and developer support.

  • The model is easily and publicly accessible, less demanding on input data, and mimics the main hydrological processes in the watersheds.

A physically based semi-distributed hydrological model called SWAT—Soil and Water Assessment Tool—has gained international recognition as an interdisciplinary watershed modeling tool for runoff simulation. It is now adopted in more than a hundred nations (Gassman et al. 2007). The model is applicable in Mediterranean basins (Pulighe et al. 2021), semi-arid regions (Maria et al. 2022), Africa (Faramarzi et al. 2012), and Thailand (Shrestha 2014).

The main objective of this study is to assess climate change's impact on the streamflow of the Kessie Watershed in the Upper Blue Nile Basin by using the newly introduced shared socioeconomic pathways (SSP2-4.5 and SSP5-8.5) scenarios. This study aimed to answer the following questions

  • What are the general future trends of temperatures and precipitation variations one could expect compared to the present situation?

  • How temperature and precipitation change reflect on the stream flow of the Kessie Watershed in the future?

Methods and materials

Study area description

The Upper Blue Nile (Abbay) Basin is located in Ethiopia's northwestern region. The topographic features of the basin are highlands, hills, valleys, and occasional rock peaks (Alem et al. 2019). Locally, the climatic seasons are classified as the dry season “Bega” from October to February; a short rainy period called “Belg” from March to May; and a long rainy season, “Kiremt,” from June to September, with the highest rainfall occurring in July and August (Alem et al. 2019). The Upper Blue Nile is the largest contributor to the Nile River's streamflow, which accounts for more than 60% of the total flow (Conway 2000, 2005). The Kessie Watershed (case study watershed) is part of the Upper Blue Nile Basin (Fig. 1 left bottom). It is located in the northern part of the Upper Blue Nile Basin and the area coverage is about 23, 58 km2. The geographic location of the case study watershed is between 10° 40′–12° 45′ N and 36° 40′–38° 15′ E. The elevation of the study area is from 1249 to 4103 m above mean sea level (Fig. 1 right). Afro-alpine, bare land, cultivation, grassland, forest land, shrub land, urban areas, and water bodies are the main land use land cover types (Fig. 2b). The dominant land use type is cultivation land (59.25%). The case study watershed has a large water body (12.97%) due to Lake Tana, the largest lake in Ethiopia, is included. The case study watershed is characterized by twelve soil groups, namely: Chromic Luvisols, Eutric Cambisols, Eutric Fluvisols, Eutric Leptosols, Eutric Vertisols, Haplic Acrisols, Haplic Alisols, Haplic Luvisols, Haplic Nitisols, Haplic Alisols, and Lithic Leptosol (Fig. 2c). The dominant soil type of the case study area is Eutric Leptosols (22.14%) and followed by Eutric Vertisols (15.49%). The case study watershed also has a different slope (Fig. 2d); the dominant slope is < 5% which counts for about 64.59% of the total area. The 5–15% slope counts about 24.75% and greater than 15% counts about 10.66%.

Fig. 1
figure 1

Locations of the study area with elevation and metrological stations

Full size image
Fig. 2
figure 2

a Climate classification, b Land use types, c soil types and d slope data

Full size image

According to the traditional climate classification based mainly on altitude variation, the case study watershed has four climate groups, namely Kola, Woyna Dega Dega, and Wurch (Fig. 2a). It covers 35% Dega, 31% Weyna-Dega, 25% Kola, and 9% Wurch. The Lake Tana Basin and closer areas to the basin are characterized by Kola. The area close to the Upper Blue Nile River also has kola climatic conditions.

Ethiopia's climate is primarily influenced by the seasonal migration of the Inter-tropical Convergence Zone (ITCZ) and accompanying atmospheric circulation, as well as the country's varied geography. The Upper Blue Nile Basin's climate is regulated by the seasonal migration of the ITCZ. The basin receives a significant quantity of yearly rainfall, ranging from 800 to 2200 mm. The basin receives four major seasons, which are Kiremt (June to August), Bega (December to February), spring (Tseday) (March to May), and autumn (Belg) (September to November). Based on the rainfall pattern shown in Fig. 3, the watershed has higher rainfall in July and August and lower rainfall from December to February. The watershed receives intermediate rainfall, lower rainfall than in July and August but higher rainfall than the remaining months, in June and September. From the total annual rainfall, 70% to 90% occurs from June to September. The range of average annual precipitation in the watershed was found to be 1049–2310 mm per year. It was examined for the rainfall data simulation period between 1985 and 2014. The information obtained from the observed minimum and maximum temperature data of the stations clearly showed the temperature of the area is spatially, temporally, and seasonally variable (Fig. 4).

Fig. 3
figure 3

Mean monthly rainfall distribution (a) mean monthly observed areal rainfall–runoff distribution (b)

Full size image
Fig. 4
figure 4

Mean annual maximum and minimum temperatures

Full size image

From the thirty years (1985–2014) temperature data, the mean daily minimum and maximum temperature of the catchment is 6.78 °C and 29.86 °C, respectively. In the maximum temperature records, all the stations in the area record the higher value in March and April. The lower maximum temperature is recorded in July and August. In the minimum temperature, the higher record occurred in April and May, and the lower record is shown in November, December, and January. The mean annual maximum and minimum temperatures of the catchment are illustrated in Fig. 4.

Data description

Observed hydro-meteorological and spatial data

The observed meteorological data used for this study were temperatures (both minimum and maximum), rainfall, wind speed, wind direction, relative humidity, sunshine hours, and solar radiation. These data (1985–2014) were collected from the Ethiopian National Meteorological Agency (NMA), Addis Ababa, and Bahir Dar Branch. The meteorological stations where temperature and rainfall data were collected are mentioned in Figs. 3a and 4. Solar radiation data, which is not possible to obtain through direct observation, were prepared from daytime sunshine hour data using the Angstrom method. The weather generator, which is used to provide data for stations with missing data, was prepared manually by using PCPSTAT and dewPOINT2.0 version programs on 30-year rainfall, temperature, and wind data. The spatial data used for this study include the digital elevation model (DEM) (20 m * 20 m) downloaded at http://earthexplorer.usgs.gov/, land use and land cover data (1: 250,000) obtained from the Ministry of Irrigation, Energy and Water Resource, and soil type data (1: 250,0000) obtained from Ministry of Irrigation, Energy and Water Resource. The meteorological and spatial data mentioned were used as input to the SWAT model to simulate the streamflow of the case study watershed. The hydrological (discharge) data was collected at the Ministry of Irrigation, Energy, and Water Resources and used for the SWAT model calibration and validation. In addition, DEM data were used to develop the elevation and slope of the study area and the soil data were used to develop the soil map of the study area. The land use data was used to develop the land cover map of the study area. The materials and tools used for the study were ArcGIS4.1, ArcSWAT10.4, XLSTATA2014, Google Earth Pro., SWAT-weather database-v01803, Minitab 18, and SWAT-CUP.

Global climate model (GCM) data

The global climate model data outputs (Table 1) with two representative shared socioeconomic pathways (SSPs) (SSPs 2-4.5 and SSPs 5-8.5) were downloaded from CMIP-6 data set website at https://esgf-node.llnl.gov/projects/cmip6/. These data were, after the best-fit climate model outputs to the observed data have been selected, used for the trend analysis of future temperature and precipitation change and used as input for the SWAT model to simulate the future streamflow data.

Table 1 The GCMs data outputs
Full size table

Methods

Observed data quality testing

Performing the data quality test is necessary to provide accurate and precise data to the model and to get a reasonable simulated result. Hydrological studies are usually started with data quality assessments such as checking for outliers, missing data, homogeneity and consistency tests, etc. On hydro-meteorological data, missing data were filled in, a homogeneity and consistency test was performed, and a trend analysis was carried out. For the rainfall and temperature data, missing data were filled by selecting the best method by investigating and comparing their minimum standard error. The double mass curve technique was adapted for data consistency checks. The homogeneity of the rainfall and temperature time series was tested by Pettitt at a 95% significance level using XLSTAT software. The absence of a trend test was carried out using the Mann–Kendall trend test using XLSTAT integrated with Excel.

GCMs data bias correction

There are various methods used for bias correction in hydrological studies. Linear scaling, power transformation, the variance of scaling, and quantile mapping are among the common methods. Linear scaling approach (Teutschbein 2012) was used for this study due to its simplicity and suitability for bias correction at the daily bias of precipitation and temperature data (Nigatu 2013). It is the most straightforward bias correction technique employed in several studies. Precipitation and temperature model data are corrected by the multiplier and additive terms, respectively.

$$ P_{{\text{RCM, daily}}}^{*} = P_{{\text{RCM, daily}}} X \left( {\frac{{\mu \left( {P_{{\text{observed, daily}}} } \right)}}{{\mu (P_{{\text{RCM, daily)}}} }}} \right) $$
(1)
$$ T_{{\text{RCM, daily}}}^{*} = T_{{\text{RCM, daily}}} + \left( {\mu \left( {T_{{\text{observed, daily}}} } \right) - \mu \left( {T_{{\text{RCM, daily}}} } \right)} \right) $$
(2)

where \(P_{{\text{RCM, daily}}}^{*}\) is the future daily bias-corrected precipitation, \(P_{{{\text{RCM}},{\text{ daily}}}}\) the daily RCM simulated precipitation, \(T_{{\text{RCM, daily}}}^{*}\) the future daily bias-corrected temperature, and \(T_{{{\text{RCM}},{\text{ daily}}}}\) the daily RCM simulated temperature data \(\mu\): mean value.

Evaluations of CMIP-6 climate model outputs

After bias correction, global climate model data outputs were evaluated to select the best-fit models for this study. Performances of twelve GCMs models were evaluated using statistical indicators. The statistics used for the evaluation of climate data sets were root mean square error (RMSE), percent of bias, and correlation coefficients which are also used for this study.

$$ {\text{Root mean square error }}\left( {{\text{RMSE}}} \right) = \sqrt {\frac{1}{n}\sum \left( {{\text{Simulated}} - {\text{Observed}}} \right)^{2} { }} $$
(3)
$$ {\text{Percent of}} {\text{Bais }}\left( {{\text{PBIA}}} \right) = \frac{1}{n}\sum \left( {{\text{Simulated}} - {\text{Observed}}} \right)^{2} { } \times { }100{ } $$
(4)
$$ {\text{Correlation coefficient }} = \frac{{{\text{covariance }}\left( {{\text{simulated}},{\text{observed}}} \right)}}{{\text{St simulated *St observed}}} $$
(5)

Historical data were compared with observed rainfall, maximum temperature, and minimum temperature, for all global climate models.

Trend analysis of future temperatures and rainfall change

To find trends and other changes in climatic and hydrological variables, a variety of statistical techniques have been applied in research. Both parametric and nonparametric methodologies fall within this category. In contrast to nonparametric approaches, parametric methods presuppose a normal distribution of variables. The Mann–Kendall test does not make any assumptions about the distribution of the data, it is robust against outliers, and it does not need the elimination of outliers before trend discovery (Takele et al. 2022). Hence, the nonparametric Mann (1945) and Kendall (1970)—Mann–Kendall trend test—was utilized in this study. Trend analyses of rainfall and temperature data were evaluated by using the Mann (1945) and Kendall, (1970)—Mann–Kendall trend test—in XLSTAT 2016 and Sen’s slope estimator. The algorithm of this method is based on the principle that when the computed P value is greater than the alpha value (0.05), it shows there is no significant trend (Arun et al. 2012). If a data value from a later period is higher than a data value from an earlier period, statistic S is increased by 1. On the other hand, if the data value from a later period is lower than the data value sampled earlier, S is decreased by 1. The net result of all such increases and decreases yields the final value of S. Mann–Kendall S Statistics is computed as follows

$$ S = \mathop \sum \limits_{i = 1}^{n - 1} \mathop \sum \limits_{j = i + 1}^{n} {\text{sign}} \left( {X_{j} - X_{i} } \right) $$
(6)
$$ {\text{sign}}\left( {X_{j} - X_{i} } \right) = \left\{ {\begin{array}{*{20}l} 1 \hfill & {\quad {\text{if}}\;X_{j} - X_{i} > 1 } \hfill \\ 0 \hfill & {\quad {\text{if}}\;X_{j} - X_{i} = 1} \hfill \\ { - 1} \hfill & { \quad {\text{if}}\;X_{j} - X_{i} < 1} \hfill \\ \end{array} } \right. $$

A positive and negative value of S indicates an upward and downward trend, respectively (Arun et al. 2012).

Soil and Water Assessment Tool (SWAT) modeling

The SWAT model is a semi-distributed, physically based, and time-continuous hydrological model that was developed by the United States Department of Agriculture’s Agricultural Research Service. This model requires spatial data, meteorological variables, and hydrological data. The model divides a basin into sub-basins that include topographic information, and the sub-basins are further divided into minimum hydrologic response units (HRUs) that uniquely combine land use, slope, and soil type. The main components of SWAT are watershed delineation, hydrological response units (HRU) creation, opening the SWAT editor to complete input preparation and execute SWAT, and visualization of results. SWAT simulates the streamflow based on the following water balance equation (Arnold et al. 1998).

$$ SW_{t} = SW_{o} + \mathop \sum \limits_{i = 1}^{t} \left( {R_{{{\text{day}}}} - Q_{{{\text{Surf}}}} - E_{{\text{a}}} - W_{{{\text{seep}}}} - Q_{Qw} } \right) $$
(7)

where \(SW_{t}\): final soil water content (mm); \(SW_{o}\): initial soil water content on a day i (mm H2O); t is the time (days); \(R_{{{\text{day}}}}\): the amount of precipitation on a day i (mm H2O); \(Q_{{{\text{Surf}}}}\): the amount of surface runoff on a day i (mm); \(E_{{\text{a}}}\): the amount of evapotranspiration on a day i (mm H2O); \(W_{{{\text{seep}}}}\): the amount of water entering the vadose zone from the soil profile on a day i (mm H2O); and \(Q_{Qw}\): the amount of return flow on a day i (mm H2O).

SWAT input data preparation and model setup

The meteorological and spatial data were used for the SWAT input data. The spatial data—a digital elevation model, land use and land cover maps, and soil maps—were projected into a common projection system for the model setup. The hydro-meteorological data were prepared based on the model input format. The data were supplied to the model to simulate the streamflow of the case study watershed. The SWAT divides the watershed into sub-watersheds based on the topography and stream networks and the hydrological response unit (HRU), which is based on the soil, slope, and land use of the area.

Sensitivity analysis, model calibration, and validation

Model sensitivity analysis, calibration, and validation were done by the SWAT Calibration and Uncertainty Fitting Program (SWAT-CUP). There are five algorithms in SWAT-CUP, which are sequential uncertainty fitting 2 (SUFI-2), generalized likelihood uncertainty estimation (GLUE), parameter solution (ParaSol), Markov chain Monte Carlo (MC-MC), and particle swarm optimization (PSO) (Abbaspour et al. 2004). SUFI-2 has gained the most popularity among users for carrying out parameterization, sensitivity analysis, calibration, validation, and uncertainty analysis of hydrological parameters (Saleh et al. 2000; Abbaspour et al. 2004), which was used for this study. Model sensitivity analysis was used to select the most sensitive parameters that can affect the runoff formation of the case study watershed. Based on previous studies on the region and the literature received, 27 parameters were selected for model sensitivity analysis. The most sensitive parameters were selected based on the P value and T test values. The t test in the analysis is a sensitivity measure, with a higher value indicating greater sensitivity and a lower value indicating less sensitivity. A P value indicates the significance of sensitivity; a value close to zero indicates greater sensitivity, while a value far from zero indicates less sensitivity. Twelve more sensitive parameters have been selected for model calibration and validation.

Model calibration and validation are necessary to make sure the hydrological model is correctly representing the hydrology of the watershed. Model calibration is the process of proving models for their reliability, accuracy, and predictive ability. The best possible correspondence between observed and simulated runoff from a catchment must be determined by selecting model parameters. The total available observed flow data at the gauged station from 1985 to 2014 were used, with 20 years (1989–2006) for calibration and 9 years (2007–2014) for validation.

Model performance evaluation

Coefficient of determination (R2) (krause et al. 2005) and Nash–Sutcliffe efficiency (NSE) (Nash and Sutcliffe 1970), and percent of bias (Gupta et al. 1999) statistical parameters have been used to evaluate the SWAT model's performance evaluation. This is due to these statistical parameters being the most commonly used methods (Takele et al. 2022). Besides, these statistical parameters are the most commonly used in many research works (for example Takele et al. 2022; Chimdessa et al. 2018; Roth et al. 2018; Worqlul et al. 2018; Alaminie et al. 2021).

Nash–Sutcliffe efficiencies (\(\mathbf{N}\mathbf{S}\mathbf{E})\)

Nash–Sutcliffe efficiencies (\(\mathrm{NSE})\) range from infinity to one and are calculated by comparing the model's goodness of fit to the variance of the observed data. The efficiency of one is equivalent to a perfect match between the modeled discharge and the measured data. It is defined as one minus the sum of the absolute squared differences between the predicted and observed values, normalized by the variance of the observed values during the period under investigation.

$$ {\text{NSE}} = 1 - \frac{{\sum { }\left( {{\text{Simulated flow }}\left( i \right){ } - {\text{Observed flow}}} \right)^{2} { }}}{{\sum \left( {{\text{Observed flow }}\left( i \right) - {\text{Average of Observed average }}} \right)^{2} { }}} $$
(8)

Coefficient of determination (R 2)

The R2 ranges between 0 and 1 which describes how much of the observed dispersion is explained by the prediction. A value of zero means no correlation at all, whereas a value of 1 means that the dispersion of the prediction is equal to that of the observation.

$$ \left( {R^{2} } \right) = 1 - \frac{{\sum {\text{(Simulated flow}} - {\text{Average simulated flow)}}^{2} }}{{\sum \left( {{\text{observed flow}} - {\text{Average observed flow}}} \right)^{2} { }}} $$
(9)

Percent of bias (PBIAS)

The PBIAS measures the average tendency of the simulated values to be larger or smaller than their observed counterparts. The optimal value of PBIAS is zero. PBIAS is the deviation of data being evaluated, expressed as a percentage. A positive PBIAS value indicates the model is under-predicting measured values, whereas negative values indicate over predicting.

$$ {\text{PBIAS}} = \sum { }\frac{{{\text{Simulated flow}} - {\text{observed flow}})}}{{\text{Obseevd flow}}} \times 100 $$
(10)

The SWAT performance rate. Source: Moriasi et al. (2007)

No

Performance rating

NSE

R 2

PBIAS

1

Excellent

0.75 ˂ NSE ≤ 1.00

0.7 ˂ R2 ≤ 1.00

PBIAS ≤  ± 10

2

Good

0.65 ˂ NSE ≤ 0.75

0.6 ˂  R2 ≤ 0.7

± 10 ≤ PBIAS ˂ ± 15

3

Satisfactory

0.50 ˂ NSE ≤ 0.65

0.50 ˂  R2 ≤ 0.6

± 15 ≤ PBIAS ˂ ± 25

4

Unsatisfactory

NSE ≤ 0.50

R2 ≤ 0.50

PBIAS ≥ 25

The calibrated and validated SWAT model was used to simulate the streamflow of the river by using the best-fitted climate model data outputs (temperature and rainfall) with the two representative shared socioeconomic pathways (SSP2-4.5 and SSP5-8.5) scenarios. It was simulated on two future time horizons: 2050 (2041–2070) and 2080s (2071–2100) (Fig. 5).

Fig. 5
figure 5

A graphic representation of SWAT modeling

Full size image

Results

Best-fitted climate models

The best-fitted climate model outputs have been selected based on statistical indicators as stated in the method flowed part above. It was selected from the 12 climate models. The ACCESS_ESM1-5, FGOALS_g3, and GFDL_ESM4 climate model outputs were selected due to having less RMSE (4.157, 7.177 7.311) and BIAS (− 0.547, − 0.764, and − 391) and a large correlation coefficient (0.634, 0.397, and 0.354) (Table 2).

Table 2 Model evaluation using statistical parameters
Full size table

Trend analysis of future temperature and precipitation

The projected future rainfall for the near term, the 2050s (2041–2070), and long term, the 2080s (2071–2100) time horizons, showed an increasing trend (S showed positive) (Table 3) in all model output data and both scenarios. However, the change is not significant (P values greater than the alpha value in Table 3) in all model outputs, and both scenarios and time horizons.

Table 3 Projected mean annual rainfall and temperature trend analysis result
Full size table

Like rainfall, the projected future maximum temperature for the near term, the 2050s (2041–2070), and long term, the 2080s (2071–2100), showed an increasing trend (Table 3) in all model output data and both scenarios. The minimum temperature also showed an increasing trend (Table 3). However, the change is not significant in both maximum and minimum temperatures (P values greater than the alpha value, 0.05) in all model outputs and both scenarios (Table 3).

The figurative representation (Figs. 6, 7) of the mean annual precipitation and temperatures (maximum and minimum) showed an increasing trend from 2041 to 2100 in all climate model outputs data and both scenarios.

Fig. 6
figure 6

Future rainfall change (2041–2100)

Full size image
Fig. 7
figure 7

future maximum temperature (a) and minimum temperature (b)

Full size image

Annual, seasonal, and monthly changes in projected rainfall

In the 2050s, mean annual precipitation may increase by 5% and 4.89% based on ACCESS_ESM1-5 model data, 4.7% and 3.8% based on FGOALS_g3 model data, and 4.67% and 3.81% based on GFDL_ESM4 model data in SSP2-4.5 and SSP5-8.5 scenarios, respectively. The relative change of maximum monthly average precipitation is increased up to 15.49% and 17.63% based on ACCESS_ ESM1-5, 80.25% and 73.16% based on FGOALS_g3, 46.84% and 79.31% based on GFDL_ESM4 in SSP2-4.5 and SSP5-8.5 scenarios, respectively.

In the 2080s, the relative change of projected precipitation may increase by 10.13% and 6.8%in based on ACCESS_ESM1-5, 4.3% and 4.84% based on FGOALS_g3, and 2.94% and 4.77% based on GFDL_ESM4 model data in SSP2-4.5 and SSP5-8.5 scenarios, respectively. The relative change of maximum monthly average precipitation is increased up to 46.73% and 31.63% based on ACCESS_ESM1-5, 63.0% and 58.13% based on FGOALS_g3, 55.35% and 64.35% based on GFDL_ESM4 in SSP2-4.5 and SSP5-8.5 scenarios, respectively (Fig. 8).

Fig. 8
figure 8

Change of annual, seasonal, and monthly precipitation a in the 2050s and b 2080s

Full size image

Annual, seasonal, and monthly changes in projected temperature

In the 2050s, the mean annual maximum temperature may increase by 3.62 °C and 1.87 °C based on ACCESS ESM1-5 model data, 1.76 °C and 1.25 °C based on FGOALS-g3 model data, and 2.15 °C and 3.83 °C based on GFDL-ESM4 model data in SSP2-4.5 and SSP5-8.5 scenarios, respectively. The relative change of the maximum monthly mean maximum temperature is increased up to 2.72 °C and 2.57 °C based on ACCESS ESM1-5, 2.75 °C and 2.15 °C based on FGOALS-g3, 2.05 °C and 2.79 °C based on GFDL-ESM4, in SSP2-4.5 and SSP5-8.5 scenarios, respectively.

In the 2080s, the mean annual maximum temperature may increase by 3.31 °C and 2.99 °C based on ACCESS ESM1-5, 3.44 °C and 2.61 °C based on FGOALS-g3, and 1.37 °C and 2.66 °C based on GFDL-ESM4 model data in SSP2-4.5 and SSP5-8.5 scenarios, respectively. The relative change of the maximum monthly average maximum temperature is increased up to 2.97 °C and 2.78 °C based on ACCESS ESM1-5, 2.64 °C and 2.66 °C based on FGOALS-g3, 2.24 °C and 2.40 °C based on GFDL-ESM4, in SSP2-4.5 and SSP5-8.5 scenarios, respectively (Fig. 9).

Fig. 9
figure 9

Mean annual, seasonal, and monthly maximum temperature change a the 2050s and b the 2080s

Full size image

In the 2050s, the mean annual minimum temperature may increase by 2.73 °C and 1.90 °C based on ACCESS ESM1-5 model data, 3.04 °C and 2.43 °C based on FGOALS-g3 model data, and 2.31 °C and 3.29 °C based on GFDL-ESM4 model data in SSP2-4.5 and SSP5-8.5 scenarios, respectively. The relative change of maximum monthly mean maximum temperature is increased up to 4.43 °C and 4.64 °C based on ACCESS ESM1-5, 4.4 °C and 4.81 °C based on FGOALS-g3, 4.31 °C and 5.36 °C based on GFDL-ESM4 in SSP2-4.5 and SSP5-8.5 scenarios, respectively.

In the 2080s, the mean annual minimum temperature may increase by 5.63 °C and 4.52 °C based on ACCESS ESM1-5 model data, 3.55 °C and 4.36 °C based on FGOALS-g3 model data, and 3.16 °C and 3.87 °C based on GFDL-ESM4 model data in SSP2-4.5 and SSP5-8.5 scenarios, respectively. The relative change of maximum monthly mean minimum temperature is increased up to 5.14 °C and 4.78 °C based on ACCESS ESM1-5, 4.81 °C and 4.63 °C based on FGOALS-g3, 4.69 °C and 4.64 °C based on GFDL-ESM4 in SSP2-4.5 and SSP5-8.5 scenarios, respectively (Fig. 10).

Fig. 10
figure 10

Annual, seasonal, and monthly minimum temperature change a the 2050s and b the 2080s

Full size image

Hydrological modelling results

The total available observed river flow data at gauged stations from 1985 to 2014 was used for 20 years (1989–2006) with a two-year warming-up period for calibration and 9 years (2007–2014) for validation. Sensitivity analysis should be carried out to identify which parameter is most important for runoff formation. Griensven et al. (2006) characterize that sensitive parameter's global rank 1 as “very important,” rank 2 up to 6 as important, rank 7 up to 9 as “slightly important,” and ranks 27 and 28 are “not important”. The t test provides a measure of sensitivity, with a large value being more sensitive, and the P value provides a measure of significance, with a value close to zero being more significant, in sensitivity ranking and analysis. The sensitive parameters with their sensitivity ranks are listed in Table 4.

Table 4 Rank for P value and t-stat value for most sensitive parameters
Full size table

The performance evaluation values from calibration fulfilled the requirement of R2 > 0.6 and ENS > 0.5, which is recommended by Santhi et al. (2001) and P-factor and R-factor are also used to determine the strength of model calibration, P-factor is a measure of the percentage of observed data is within 95PPU have value greater than 0.65 and R-factor is a measure of the thickness of the envelope (95PPU) have to value less than 1.5 were recommended for model calibration. After analysis, the objective functions to be used for each model are Nash–Sutcliffe (NSE) (0.7), coefficient of determination (R2) (0.84), and also P-factor and R-factor having values 0.7 and 1.4, respectively (Fig. 11, Table 5).

Fig. 11
figure 11

Hydrograph of simulated and observed flow for calibration (a) and validation (b) in daily calibration (c) and validation (d) in monthly

Full size image
Table 5 Summary of statically model performance indicators from the above graph
Full size table

Climate change impact on monthly streamflow

Using the SWAT hydrological model, the climate change impact on monthly streamflow was analyzed by comparing the baseline observed data with future simulated streamflow data for the 2050s (2041–2070) and 2080s (2071–2100). The monthly streamflow volume in the near term (the 2050s) compared to the baseline showed an increase in all months (Table 6). In this period, the streamflow volume increases up to 37% in August based on SSP2-4.5 and 57% in November based on SSP5-8.5 for ACCESS-ESM1-5, 32% in March based on SSP2-4.5 and 58% in October based on SSP5-8.5 for GFDL-ESM4, and 58% in April based on SSP2-4.5 and 29% in October based on SSP5-8.5 for FGOALS-g3.

Table 6 Monthly streamflow change in (%) the near term (2050s)
Full size table

The monthly streamflow volume in the long term (the 2080s) compared to the near term (2040s) showed an increase in all months (Table 7). In this period, the streamflow volume increases up to 38% in May based on SSP2-4.5 and 57% in November based on SSP5-8.5 for ACCESS-ESM1-5, 48% in May based on SSP2-4.5 and 59% in January based on SSP5-8.5 for GFDL-ESM4, and 65% in March based on SSP2-4.5 and 75% in August based on SSP5-8.5 for FGOALS-g3.

Table 7 Monthly streamflow change in (%) for the long term
Full size table

Climate change impact on seasonal and annual streamflow

To predict how climate change may affect the socioeconomic situation in the Upper Blue Nile Basin, the effects of climate change on the seasonal and annual flow volumes are also explored. The study area has four distinct seasons: Kiremt (June to August), Belg (March to May), Tseday (September to November), and Bega (December to February).

In the 2050s, the average annual stream flow may increase up to 21.8% and 15.4% for ACCESS-ESM1–5, 12.1% and 13% for GFDL-ESM4, and 18.5% and 9.8% for the FGOALS-g3 model in SSP2-4.5 and SSP5-8.5 scenarios, respectively. In the 2080s, the average annual stream flow may increase by 17.35% and 13.08% for ACCESS-ESM1-5, 15.14% and 21.46% for GFDL-ESM4, and 24.02% and 40.95% for FGOALS-g3 in SSP2-4.5 and SSP5-8.5 scenarios, respectively. The seasonal variation for Kiremt, Belg, Tseday, and Bega for ACCESS-ESM1-5, GFDL-ESM4, and FGOALS-g3 for both scenarios for the near term and short term is described in Tables 8 and 9 as well as Fig. 12.

Table 8 Change in Annual and Seasonal flow volume (%) the 2050s
Full size table
Table 9 Change in annual and seasonal flow volume 2080s (%)
Full size table
Fig. 12
figure 12

Percentage change in seasonal and annual flow volume a the 2050s and b 2080s

Full size image

Discussion

The future mean annual precipitation and simulated streamflow, as well as the mean annual temperature (both maximum and minimum temperatures) of the Kessie Watershed, are expected to increase in all model outputs data (ACCESS ESM1-5, FGOALS-g3, and GFDL-ESM4), in both scenarios (SSP2-4.5 and SSP5-8.5), and time horizons (the 2050s and the 2080s). But, increasing streamflow with increasing temperature seems irrational. But, the dominant land use in the case study watershed is agriculture which covers about 59.25% of the total area. It may cause less infiltration rate and high runoff occurrence. Besides, the forest land has very small area coverage which case for less interception and transpiration loss. Both these together may cause the streamflow to be more governed by precipitation change than the temperature in the case study watershed.

The mean monthly streamflow data also showed an increasing trend from all months based on all model outputs data, both time horizons and also both scenarios. But, when we saw the rainfall and temperatures, it is not shown to increase in all months and all models as well as both scenarios. From April to August, both maximum and minimum temperatures are expected to decrease in the near term (the 2050s) and long term (2080s) in all model output data and both scenarios but rainfall doesn’t show a consistent increase in the above-mentioned months. So, increasing the streamflow in these months may be following the decreasing of temperature (decreasing temperature means less evapotranspiration loss) and not many significant changes in rainfall. From October to February, both rainfall and temperature showed an increasing trend almost in all model output data in both scenarios and time horizons. The streamflow is also increasing. It may be due to the aforementioned reason which is land use of the area cases for high runoff occurrence following the rainfall is increasing.

On a seasonal basis, the streamflow is expected to increase in all seasons (summer, winter, spring, and autumn). In winter and spring, both maximum and minimum temperatures showed to increase in all model output data in both scenarios and time horizons. The precipitation also shows the same increasing trend except for a very slight decrease in model ACCESS_ESM1-5 output data in the 2050s in both scenarios' results. Rainfall is expected to increase at a high rate (for example in winter, it is expected to increase up to 66.3% based on the FGOALS_g3 model output data in the 2050s on SPP2-4.5 scenario and up to 97% based on the FGOALS_g3 model output data in the 2080s on SPP5-8.5 scenario). But, the temperature is not increasing at as high a rate as rainfall (for example the highest maximum temperature change is about 5.7% based on the ACCESS ESM1-5 in SPP5-8.5 in the 2050s which happened in spring and the highest maximum temperature change is about 6.36% based on the ACCESS ESM1-5 in SPP2-4.5 in the 2080s which is also happened in spring). The increasing rate of minimum temperature is also not as high as rainfall. Besides, the model is highly rainfall dependent due to the aforementioned reasons. Hence, increasing streamflow with increasing rainfall and temperatures in winter and spring may have come from these facts.

In summer and autumn, both minimum and maximum temperatures show a decreasing trend based on all model output data in both scenarios and time horizons. Due to these, the streamflow of the River seems to increase due to the decreasing effect of temperatures.

Previous study results in the Upper Blue Nile Basin support our findings. For example, the ensemble mean of six GCMs, namely HadCM3, GFDL-CM2.1, ECHAM5-OM, CCSM3, MRI-CGCM2.3.2, and CSIRO-MK3, revealed a rising trend for precipitation and maximum and minimum temperatures in Upper Blue Nile Basin. The relative change in precipitation varied from 1.0 to 14.4%, whereas the change in mean annual maximum temperature could range from 0.4 to 4.3 °C and the change in a mean annual minimum might range from 0.3 to 4.1 °C (Mekonnen and Disse 2018). Again in the Upper Blue Nile Basin, Alaminie et al. (2021) investigated trend analysis of future rainfall and temperatures under four CMIP6-SSPs scenarios (SSP1-2.6, SSP2-4.5, SSP5-8.5, and SSP3-7.0) at two time zones—near and long term, or 2031–2060 and 2071–2100, respectively. The results show that under the four SSPs, the mean annual maximum temperature increased by 1.1 (1.5), 1.3 (2.2), 1.2 (2.8), and 1.5 (3.8) oC throughout the near- (long-) term period. Contrarily, under four SSPs scenarios, precipitation forecasts indicate a marginally (statistically insignificant) upward trend over the near (long) term periods of 5.9 (6.1), 12.8 (13.7), 9.5 (9.1), and 17.1 (17.7)%. The Gilgel Abay River, Upper Blue Nile Basin, hydrological response to climate change study by Dile et al. (2013) indicated that the annual mean precipitation may increase from 2040 to 2100, which may increase streamflow. Additionally, it states that during the Belg (small rainy season) and Kiremit (main rainy season) periods, the seasonal mean flow volume may rise by more than twice as much and by 30–40%, respectively. The researchers also conclude that flow volume would rise annually as a result of climate change. According to Worqlul et al. (2018), the influence of climate change on streamflow hydrology at the main and Gilgel Beles basins in the Upper Blue Nile Basin will rise throughout the future time horizons (2020–2100). The minimum and maximum temperatures will rise by 3.6 °C and 2.4 °C, respectively, by the end of the twenty-first century, but trends in annual rainfall do not show statistically significant trends between years and concluded streamflow will rise by up to 64% in dry seasons by the end of the century. Using five GCMs, the hydrological response to climate change of the Upper Blue Nile River's four catchments—Gilgel Abay, Gumer, Ribb, and Megech—in the 2030s (2035–2064) and 2070s (2071–2100) Runoff is anticipated to increase in the future; at 2030s, the average annual runoff projection change may increase up to + 55.7% for RCP 4.5 scenarios and up to + 74.8% for RCP 8.5 scenarios. Projections showed significant average monthly and seasonal precipitation change variability. For the RCP 4.5 and RCP 8.5 emission scenarios, the average annual runoff percentage change will increase by + 73.5% and + 127.4%, respectively, by 2070s (Gebre and Ludwig 2015).

Chakilu et al. (2022) studied climate change responses of streamflow in the four gauged watersheds (Gilgel Abbay, Ribb, Gumara, and Megech) of the Lake Tana Basin in the Upper Blue Nile Basin. The study uses the highest emission scenario Representative Concentration Pathway (RCP8.5). The result reveals the temperature and the rainfall are expected to increase in the future. It also affirms the streamflow is expected to increase by 5.89%, 5.63%, 4.92%, and 4.87% in Ribb, Gumara, Megech, and Gilgel Abbay Watersheds, respectively. Roth et al. (2018) studied the effects of climate change on water resources of the Upper Blue Nile Basin, Ethiopia. The study result showed the precipitation will increase from 7 to 48% and that streamflow could increase by 21–97%.

On the other way, some studies disagree with our results. For example, Ayalew et al. (2021) concluded that future temperatures will increase but there is no consistent trend in rainfall in the Ribb catchment, Lake Tana Basin, Ethiopia. As a result, the streamflow is expected to decrease in both scenarios (RCP4.5 and RCP8.5). Similarly, Takele et al. (2022) studied future climate change and its impacts on water resources in the Upper Blue Nile Basin. The study result affirms the projected rainfall shows a decreasing trend, while the temperature shows an increasing trend. The decline in rainfall, increase temperatures and rise in evapotranspiration cause a decline in the streamflow. The reasons may study result of Ayalew et al. 2021 is based on the single RCM climate model output data, the model is selected without justifications as well as the authors use the RCP scenarios data. The study result of Takele et al. (2022) is based on RCP scenarios and the climate model outputs have not been selected with reasons.

Coming to our study result, the climate model outputs have been selected based on the results of statistical indicators which were done by comparing the observed and model output data. The best-fit models were selected from twelve GCMs models. The other improvement is our study has been using updated Coupled Model Intercomparison Project Phase 6 (CMIP-6) data outputs, three climate model outputs; ACESS_ESM1-5, FGOALS_g3, and GFDL_ESM4 with two shared socioeconomic pathways (SSP2-4.5 and SSP5-8.5) scenarios.

Conclusion

The study aimed to examine the performance of the new six-generation CMIP-6 climate output with two representative shared socioeconomic pathways (SPP2-4.5 and SSP5-8.5) scenarios to use for future temperature and precipitation change trend analysis and to estimate the effect of climate change on streamflow of the Kessie Watershed, Upper Blue Nile Basin, Ethiopia. The best-fit GCM climate data outputs were selected by comparing them to the observed data from the twelve climate models and hence three best-fit models outputs are (ACESS_ESM1-5, GFDL_ESM4, and FOGALS_g3). Trend analysis of future temperature and precipitation was carried out by using the Mann-Kendal trend test and Sen’s slope estimator. The catchment hydrology was represented by the SWAT model through calibration and validation. The trend analysis and simulation of future streamflow were done on the two time horizons: the near term, the 2050s (2041–2070), and the long term, 2080s (2071–2100). The observed data (1985–2014) were used as a reference period. The result revealed precipitation and temperature are expected to slightly increase from the reference period to the long term based on all climate outputs data and both scenarios. Following In the 2050s, there may be an average annual increase of streamflow of up to 22% and in the 2080s, there may be an increase of up to 41%. The projected change in different seasons showed an increase in streamflow in the future 2041–2100.