Assessing the influence of land use/land cover dynamics and climate change on water resources in Upper Blue Nile, Ethiopia

Various land uses, such as agriculture, livestock, urbanization, forestry, and industry (Abbas et al. 2021; Li et al. 2023; Elahi et al. 2024), affected water resources (Waseem et al. 2022). Land use and land cover (hereafter, LULC) and climate change (hereafter, CC) are recognized as the most critical causes of a change in the environment (Kundzewicz 2008; Chanapathi & Thatikonda 2020; IPCC 2022a). LULC alterations and CCs are interconnected (Zhang et al. 2016; Li et al. 2019). LULC change impacts regional, national, and global climatic systems. Land and climate interact in complex ways through changes in forcing and multiple biophysical and biogeochemical feedbacks across different spatial and temporal scales. Changes in land conditions resulting from human activity or CC affect regional and global climates. Alterations in local land cover and land use can either alleviate or exacerbate regional CC (Jia et al. 2022). Incorporating terrestrial climate processes into climate projections enables a better understanding of how land responds to CC and allows for a more accurate estimation of the potential effectiveness of land-based strategies for mitigating CC (IPCC 2014).

It is thought that LULC alterations and CC have an impact on hydrological processes (Chanapathi & Thatikonda 2020). Land use changes and CC can accelerate the global depletion of freshwater supplies, which are essential for sustaining human populations and ecological services (Kaushal et al. 2017). As a result of CC, environment, recreation, industry, agriculture, and energy production risks and costs are rising. CC also changes the quality and amount of water that people and ecosystems can use across the world (Hayhoe et al. 2018; USGCRP 2018). Restoring global water security requires comprehensive consideration of the escalating interaction between land use and CC across all phases, along with our expanding impact on the water cycle from degradation to ecosystem restoration (Kaushal et al. 2017).

A variety of land uses, such as agriculture, urbanization, forestry, and industry, affected water resources (Li et al. 2023; Elahi et al. 2024). Due to the loss of natural resources, population growth, poverty, and rapid urbanization have all been made worse by CC (Awulachew et al. 2008). Water scarcity is expected to worsen dramatically because of increased human demand for water resources. Changes in land use affect how runoff and percolation work, how groundwater is refilled, how much sediment is transported, and how much water is lost through evaporation (Gosling & Arnell 2016). For instance, Hu (2021) reported that changes in river discharge due to CC are highly uncertain, and a recent study used a global streamflow dataset to assess whether such trends are detectable. Streamflow changes occurred more often in basins impacted by human disturbances than in pristine ones, and there was no clear signal from CC alone. This study highlights the importance of incorporating land use change in hydrological studies.

Globally, CC has led to alterations in the amount of rainfall, its seasonal distribution, and its interannual variability (Easterling et al. 2000). These changes in rainfall patterns would have the most significant impact in arid and semiarid regions, where the timing and availability of water play crucial roles in governing primary productivity, biogeochemical cycles, growth and reproduction phenology, and agricultural output (Feng et al. 2013). Predictions suggest that global warming will amplify the hydrological cycle, potentially increasing the intensity of extreme precipitation events and the likelihood of flooding (Tabari 2020).

Nations all across the world are at risk from climate-related hazards (Getachew & Melesse 2012), but the global south is especially vulnerable due to its limited capacity for adaptation and substantial dependence on climate-vulnerable industries, such as agriculture (Aryal et al. 2021). East Africa's agricultural productivity has already been negatively impacted by CC, and things are only going to get worse. By the end of the 21st century, CC is simulated to increase temperatures by 1.4–5.5 °C and precipitation by −2 to 20% in East Africa (Adhikari et al. 2015; IPCC 2022b). Flooding in East Africa will become more likely due to the variability of climate, which will also hurt human health and cause infrastructure damage (Serdeczny et al. 2017).

Global climate models or general circulation models (hereafter, GCMs) are based on the general principles of fluid dynamics and thermodynamics (Stute et al. 2000). The primary approach for forecasting variability and changes in climatic variables involves the use of GCMs. It provides the most reliable information for assessing global change of climate variability. However, the low spatial resolution of GCMs limits their usefulness for assessing local impacts (Fowler 1999; Wilby & Harris 2006; Lee & Singh 2018; Shrestha et al. 2021). Also, it doesn't take into account subgrid-sized things like terrain and clouds, which are very important for figuring out local-scale weather conditions (Fowler 1999; Wilby et al. 2002). For instance, Getachew et al. (2021b) used only a single GCM to assess the impacts of climate and land use change on the hydrology of the Lake Tana Basin (LTB). However, the application of an ensemble of multi-models improves the effectiveness of GCM simulations (Seiller et al. 2012; Broderick et al. 2016; Liu et al. 2017, 2019; He et al. 2018; Das et al. 2019; Pooralihossein & Delavar 2020; Srivastava et al. 2020; Jose et al. 2022). Therefore, downscaling and multimodel ensemble mean are required to minimize uncertainty about CC projections that arise from internal climate variability (Getachew & Manjunatha 2021; Getachew et al. 2021b).

The Upper Blue Nile River Basin (UBNRB) has the highest discharge volume of 1,548 m³/s (Melesse et al. 2014; Tikuye et al. 2023), the second-largest area of 325,000 km², and the main source of the Nile River (Taye et al. 2015). Due to its political and economic importance, many studies were conducted to find out how the water balance of the UBNRB and the LTB affected by LULC and CCs (Kim et al. 2013; Kuhn 2014; Zeleke & Damtie 2017; Berihun et al. 2019; Getachew et al. 2021b; Getachew & Manjunatha 2022b). Thus, knowledge about these water balance components is the key to assessing the water resource potential, besides planning and executing the water resource development schemes in the basin (Woldegebriel et al. 2022).

Much of the available research investigated the impacts of changing climate (Mekonnen & Disse 2018; Woldesenbet et al. 2018b; Berihun et al. 2019; Mengistu et al. 2020) effects of LULC and CC change on water balance on a certain type of water balance components such as surface runoff, evapotranspiration, soil water (Legesse et al. 2003; Kim et al. 2009, 2013; Mahmood & Jia 2016; Woldesenbet et al. 2018a; Torabi et al. 2020). Furthermore, many previous studies (Kim et al. 2013; Kuhn 2014; Zeleke & Damtie 2017; Berihun et al. 2019; Getachew et al. 2021b; Getachew & Manjunatha 2022b) have used climate data obtained from national meteorological stations for baseline hydrological simulations as well as CC projections. However, this data has many missing values, which is not suitable for hydrological modeling. Thus, this study is designed to use the Climate Hazards Group Infrared Precipitation with Stations Data (CHIRPS), which minimizes missing data problems in the study area. This study's goal is to evaluate the integrated and separate impacts of changing LULC and CC on the water resources in the LTB, Ethiopia.

The Blue Nile River originates from Lake Tana, a natural reservoir (Stave et al. 2017; Getachew et al. 2021b). The Lake, which is 66 km wide and 84 km long, is found in the nation's northwest mountains. Lake Tana has a 3,000–3,600 km² area and is situated at a height of 1,784 m (Getachew et al. 2020). Lake Tana is a highland lake that is just 15 m deep. The Gumera, Gilgel Abay, Ribb, and Megech are Lake Tana's four main tributaries (Abtew & Dessu 2019; Getachew et al. 2021b). These rivers account for >93% of the yearly water budget of the lake.

The basin experiences an average annual precipitation of approximately 1,280 mm. It undergoes a single wet season, spanning from mid-June to mid-September, with the climate primarily influenced by the tropical highland monsoon (Getachew & Manjunatha 2022b). Despite notable daily temperature fluctuations and minimal annual and seasonal variations, the average annual temperature in the basin remains around 20 °C (Ademe et al. 2022). The basin experienced unimodal rainfall distributions.

Baseline land use maps for the research region were generated using Landsat 4–5, Landsat 7, and Landsat 8 satellite data for the years 1990, 2005, and 2020, respectively, with a spatial resolution of 30 m. Before classifying the image datasets, preprocessing of the imagery was conducted. Further details on the preprocessing can be found in Tikuye et al. (2023). Subsequently, the IDRISI Selva 17.0 software program was employed, utilizing the maximum likelihood technique, to classify the LULC data. The basin comprises seven LULC classes: Grazing lands, shrublands, forests, built-up areas, bare lands, water bodies, and farmlands. Simulations of LULC changes for the 2030s and 2070s were performed using the baseline LULC classes from 1990 to 2020.

The cellular automata (CA)-Markov chain model was employed to simulate the LULC changes for the 2030s and 2070s. Currently, LULC simulation using the CA-Markov chain model is widely used in different parts of the world (Ghosh et al. 2017; Gidey et al. 2017; Hamad et al. 2018; Beroho et al. 2023). Markov chain models assume that transition probabilities are stationary, meaning they remain constant over time, and are typically derived from historical data without considering underlying drivers of change, such as economic incentives, land suitability, and policy interventions. Additionally, they heavily depend on the quality and resolution of historical land use data. To address these limitations, the study integrated Markov chain models with CA, which can incorporate spatial dependencies, non-stationary transition probabilities, and the influence of underlying drivers and feedback mechanisms.

The preparation of various inputs was conducted before the projection of the future LULC transition. The creation of two separate basic LULC maps served as the initial input and should be utilized to create the probability of transition suitability maps (Ghosh et al. 2017). The next step was the creation of constraint and factor maps to build the necessary raster images. Details of the steps for preparation of constraints and factor maps can be found in Getachew et al. (2021b).

Future CC data were downscaled using the medium-emission Representative Concentration Pathway (RCP4.5) and high-emission RCP8.5 from the Coupled Model Intercomparison Project Phase Five (CMIP5). The study relies on climate projections from the CMIP5 rather than the more recent CMIP6. CMIP6 includes updated models and scenarios that reflect advancements in climate science, potentially offering more accurate and reliable climate projections, which provide a more robust foundation for impact assessments, policy-making, and adaptation strategies (Bouramdane 2023)s. However, while CMIP6 represents the latest in climate modeling, CMIP5 remains crucial for ensuring continuity, comparability, and accessibility in climate research. Its established use and comprehensive documentation make it a valuable resource for ongoing and comparative studies. The CMIP5 project relies on standardized greenhouse gas concentration trajectories known as RCPs. It assumes that external forcing factors – such as solar radiation, volcanic activity, and anthropogenic emissions – are accurately represented and drive the climate system's response (Bouramdane 2023). Additionally, the models assume they can adequately represent natural climate variability (Moss et al. 2010; van Vuuren et al. 2011).

The SWAT model incorporated various datasets including the digital elevation model (DEM), soil, discharge, weather, LULC, and soil information. Hydrological response units (HRUs), slope reclassification, and watershed delineation were all defined using the Shuttle Radar Topographic Mission (SRTM) 30-m dataset. The input soil layer data was obtained from the Ministry of Agriculture (MoA). On the other hand, baseline weather datasets were obtained from CHIRPS, while future weather datasets were downscaled using LARS-WG. The Ministry of Water, Irrigation, and Electricity (MoWIE) supplied discharge data from four stations (Gumera, Ribb, Gilgel Abay, and Megech) for the period between 1980 and 2007, which was used for the calibration and validation of the SWAT model. To address gaps in the hydrological data, linear interpolation techniques were employed to fill in the missing values. This comprehensive data collection and preprocessing ensure that the assessment is based on accurate and high-resolution information, which is crucial for the reliability of the study's outcomes. Furthermore, both the hydrological and climate models underwent careful calibration and validation steps before running the SWAT model.

In this study, the components of the water balance were determined using the SCS Curve Number and Hargreaves techniques. The model was employed to assess the potential impact of various land management techniques on water, chemical, sediment, and agricultural fields. It operates on a daily basis (Arnold et al. 1998; Arnell & Freeman 2011; Gassman et al. 2014; Döll et al. 2015; Abbaspour et al. 2018).

The SWAT model assists in making decisions regarding various water management techniques, land use and management practices, as well as pollution control measures (Berihun et al. 2019; Getachew et al. 2021a, 2021b). This is accomplished by carefully calibrating and reducing uncertainty in the SWAT model. As a result, the Integrated Parameter Estimation and Uncertainty Analysis Tool (IPEAT) was used. IPEAT is a tool for automatically calibrating the SWAT + model (Yen et al. 2014, 2019). Thus, the model was automatically and manually calibrated over 20 years (1985–1999) using monthly streamflow at gauging stations in the Gumera, Ribb, Gilgel Abay, and Megech river basins, and data from 2000 to 2007 were used to validate the model.

The effectiveness of hydrological models is typically assessed using metrics such as the correlation coefficient (R²), percent bias (PBIAS), and Nash–Sutcliffe efficiency (NSE) coefficient (Nash & Sutcliffe 1970; Mengistu & Sorteberg 2012; Abbaspour et al. 2018). According to (Gassman et al. 2007; Arnold et al. 2010; Mengistu & Sorteberg 2012; Abbaspour et al. 2018) the model is adequate for simulating the water balance components when NSE and R² are >0.5 and PBAIS is <± 25%.

Daily data spanning from 1981 to 2010 was utilized for each station to calibrate and validate the model. Additionally, baseline data from the same period was employed to conduct site analysis and generate synthetic time series data using LARS-WG. Statistical tests such as the Kolmogorov–Smirnov (K–S) test for probability distributions, T-test for means comparison, F-test for standard deviations comparison, and R² coefficient of determination were employed to evaluate the performance of LARS-WG in generating synthetic data compared to the baseline.

Figures 5 and 6 also show a visual comparison of the minimum and maximum temperatures at the stations in Gonder, Debre Tabor, Dangila, and Bahir Dar. The model was capable of simulating these characteristics exceptionally well, and the generated data closely matched the observed historical data across all months.

Moreover, for the Bahir Dar and Gonder stations, the standard deviation between simulated and observed maximum and minimum temperatures showed good performance. However, the performance of the LARS-WG for Dangial and Debre Tabor showed relatively poor performance in terms of the standard deviation between the observed and simulated maximum and minimum temperatures. Details of the coefficient of determination (R²) and p-values on the performance of the LARS-WG for simulating maximum and minimum temperature are found in Figures 5 and 6.

Tables 1 and 2 evaluate the performance of LARS-WG in replicating seasonal and daily rainfall distribution for the LTB. The K–S test results for seasonal wet/dry series distribution of observed data across four stations Bahir Dar, Dangila, Debre Tabor, and Gonder indicate a perfect fit in all cases. For each station, data were evaluated for both wet and dry conditions across all four seasons (DJF, MAM, JJA, SON). The number of observations (N) was consistently 12 for each category. The K–S values ranged between 0.01 and 0.14, with all corresponding P-values equal to 1, leading to an assessment of a perfect fit for every season and condition at all stations. This consistency suggests that the observed data aligns extremely well with the expected distributions for each seasonal wet and dry series at these locations. Table 2 presents the K–S test results for daily rain distribution. The evaluation reveals that, except for January, February, March, and December at the Bahir Dar station, as well as January and February at the Gonder station, LARS-WG performed exceptionally well in replicating daily rainfall distributions.

Details of the LARS-WG performance using the Kolmogorov–Smirnov (K–S) test statistics and p-value in precipitation and temperature can be found in Tables 1–4. In line with this study, Chisanga et al. (2017) reported that LARS-WG was more capable of simulating the seasonal distributions of the wet/dry spells and the daily precipitation distributions as well as maximum and minimum temperature distributions in each month.

The study utilized five distinct GCMs under different RCPs to project future CC trajectories. To mitigate the divergence in results due to varying emissions scenarios, the mean of these five GCMs was used for the impact analysis scenarios. For instance, Lovino et al. (2018) highlighted the importance of using an ensemble of the nine most skillful models, noting that this approach enhances the performance compared to using individual GCMs.

The individual and ensemble means of five GCMs have revealed an increase in the minimum temperature at the four meteorological sites, much like the maximum temperature. For instance, the ensemble average minimum temperature in Gonder, Debre Tabor, Dangila, and Bahir Dar tends to rise to 0.5, 1.8, 1.8, and 1.8 °C by 2030s and 2.0, 2.9, 2.8, and 3.4 °C by 2070s, respectively, under RCP4.5 compared to the historical period from 1981 to 2010 (Figure 8). Furthermore, under RCP8.5, the ensemble mean minimum temperature likewise increased up to 2.6, 1.4, 1.4, and 1.9 °C by the 2030s and 4.7, 5.6, 5.4, and 6.2 °C by the 2070s in Gonder, Debre Tabor, Dangila, and Bahir Dar, respectively.

Under all RCPs and models, there is no clear pattern in the yearly rainfall percentage change. Two GCMs (CanESM2 and MRI-CGCM3) overestimated the annual average rainfall compared to the baseline at Gonder, Debre Tabor, Dangila, and Bahir Dar stations by the 2030s and 2070s under RCP4.5 radiative forcing scenarios, while three GCMs (CSIRO-MK3.6.0, MPI-ESM-MR, and IPSL-CM5A-MR) underestimate the annual average rainfall at Gonder, Debre Tabor, Dangila, and Bahir Dar under RCP4.5 by the 2030s and 2070s.

By the 2030s, under RCP8.5, two GCMs (CSIRO-MK3.6.0 and MPI-ESM-MR) underestimate the mean annual rainfall, while three GCMs (CanESM2, IPSL-CM5A-MR, and MRI-CGCM3) overestimate the annual average rainfall at Bahir Dar and Debre Tabor. Besides, at Gonder and Dangila stations, one GCM (CSIRO-MK3.6.0) underestimates, while four GCMs (CanESM2, IPSL-CM5A-MR, MPI-ESM-MR, and MRI-CGCM3) overestimate the mean annual rainfall under RCP8.5 by the 2030s compared to the baseline periods. By the 2070s, three GCMs (CanESM2, IPSL-CM5A-MR, and MRI-CGCM3) exceeded the annual average rainfall at Gonder, Debre Tabor, Dangila, and Bahir Dar stations, while two GCMs (CSIRO-MK3.6.0 and MPI-ESM-MR) underestimated it.

Under the RCP4.5 compared to the baseline rainfall, the ensemble mean annual precipitation is simulated to decrease up to 3.5% and 0.63 at Debre Tabor and Bahir Dar and to increase up to 0.31 and 1.52% at Dangila and Gonder stations by the 2030s. Furthermore, under the RCP4.5, the ensemble mean annual precipitation is projected to decline by the 2070s by up to 1.1, 4.2, 2.8, and 0.41% at the Gonder, Debre Tabor, Dangila, and Bahir Dar stations, respectively.

During the classification of LULC in the basin (Figure 10), seven classes were identified: water bodies, shrublands, grazing lands, forests, built-up areas, barren lands, and agricultural lands. As indicated in Table 5, the percentage of built-up, cultivated land, and forested land increased, while the percentage of barren land, water bodies, and grazing lands decreased (Getachew & Manjunatha 2022b). A baseline land use map for 2020 was projected using data from the 1990 to 2005 maps. Additionally, major drivers such as open water bodies, rivers, highways, elevation, slope, and urban variables were considered when simulating future land use changes.

Before projecting two future periods, the simulated LULC maps were calibrated and validated using the research area's 2020 baseline/reference LULC maps. The projected and baseline LULC classes showed minimal differences. Various Kappa indices (KIA) and related statistics were utilized to quantify the degree of agreement between the baseline and projected maps.

Based on the Kappa values obtained during the validation procedure (K standard = 0.81, Kno = 0.87, and K location = 0.88), there was a strong agreement between the projected and baseline maps, indicating the accuracy of the model prediction (Beroho et al. 2023). Future simulation maps for the 2070s and 2100s were generated using the CA-Markov chain model. The projection indicated that an increase in cultivated land and settlements will lead to a reduction in certain land use types, such as forests, grazing lands, and shrublands (Getachew et al. 2021a).

The LULC data from 1990 to 2100 show substantial changes across different classes. Water bodies increased slightly from 20.48 in 1990 to 21.08 in 2005 and remained relatively stable through 2020 at 21.06, but are projected to decrease to 20.42 by 2100. Grazing lands experienced a sharp decline from 12.44 in 1990 to 8.09 in 2005, and further down to 3.50 by 2020, with a continued decrease to 2.85 by 2100. Forests initially decreased from 2.41 in 1990 to 0.82 in 2005, but are expected to recover to 2.23 by 2020 and increase to 2.88 by 2100.

Built-up areas expanded significantly from 0.04 in 1990 to 0.91 in 2005, and surged to 3.41 by 2020, with projections reaching 4.06 by 2100. Cultivated lands steadily increased from 56.69 in 1990 to 63.23 in 2005 and 63.94 in 2020, maintaining this level through 2100 (Figure 10 and Table 5). These trends reflect ongoing urbanization and agricultural expansion, alongside efforts in reforestation, impacting ecosystem management and land use planning in the future.

The model performance assessment focused on the parameters REVAL_MIN, REVAL_CO, FLOW_MIN, ALPHA, DELAY, ESCO, AWC, and CN2. Sensitivity analysis of these parameters is essential for reducing parameter dimensions and calibration time. Sensitive hydrological parameters were calibrated through sensitivity analysis to achieve satisfactory agreement between model-simulated and observed values. Calibration used streamflow data, and the SWAT model's performance was evaluated using percent bias (PBIAS), the correlation coefficient (R²), and the NSE.

The model's output at all four gauge stations is satisfactory. The R² and NSE values revealed the model was adequate for the Gumera, Megech, Ribb, and Gilgel Abay rivers. The negative PBIAS result demonstrated that the model's projected values for the four rivers were greater than those observed during validation and calibration. The calibrated model could replicate streamflow in the LTB, and a model with more parameters might help modify how the streamflow will react to LULC and CC changes (Table 6 and Figure 11). Consequently, earlier research has supported the effectiveness of the SWAT paradigm (Chu et al. 2004; Setegn et al. 2008; Dile et al. 2013; Gebremicael et al. 2013; Döll et al. 2016; Fentaw et al. 2018; Gebrechorkos et al. 2019).

The water balance of the LTB to CC and LULC change was evaluated for three future time windows: the 2030s (2021–2050) and the 2070s (2071–2100). The 1981–2010 time frame serves as the baseline/ reference. Three modeling scenarios were created. Thus, S1 just examines LULC, S2 only considers CC, and S3 analyses both (the combination of S1 and S2).

Throughout the dry season (ONDJF), streamflow increased by 30.5–32.7%; however, it decreased by 6.8 and 2.8% through the short rainy season (MAM) and by 2.8 and 3.8% through the major rainy season (JJAS) by the 2030s and 2070s, respectively (Figure 12). During the MAM and JJAS months of the basin, the water flow in the stream dropped, whereas it increased during the ONDJF months. During the dry and short rainy season, runoff increased by 21.4 and 24.0%, 10.4 and 16.9%, respectively, while during the rainy season, it decreased by 2.0 and 1.8%, respectively, by the 2030s and 2070s periods (Figure 12).

Under a potential LULC change scenario, runoff, lateral flow, water yield, and percolation follow similar patterns. Hence, these hydrological components decrease through the wet season, whereas they increase through the dry and short wet seasons. All of these variables have a direct relationship with the projected variation in precipitation in the basin. Similarly, precipitation would increase throughout the dry and short seasons while decreasing through the wet season.

The average streamflow, runoff, lateral flow, water yield, and soil water in a particular section of the basin tend to increase by 10, 10, 10, 60, and 20%, respectively (Figure 13). Additionally, streamflow, runoff, water yield, lateral flow, and soil water tend to decline up to 10, 10, 5, 40, and 0.5% in other parts of the basin, while percolation tends to decline up to 2% in all parts of the LTB. Unlike the other portions of the water balance, PET and AET tend to increase by 0.25 and 10%, respectively, in the whole basin.

Figure 14 depicts the mean seasonal anomaly of water yield, lateral flow, streamflow, and runoff under S2 scenarios. The aforementioned hydrological parameters will increase in the future throughout dry and short rainy seasons but relatively decline through the rainy season of the basin. Even though the pattern of change has a similar trend, the degree of change is different for each hydrological component. For instance, the flow of stream tends to increase from 0.6 to 42.0% in the 4.5 RCP and from 1.7 to 55.3% in the 8.5 RCP by the 2030s and 2070s periods under both the dry and short rainy seasons of the basin but decline through the main rainy season.

The basin's dry and main wet seasons reduce lateral flow and percolation. These components rise throughout the basin's short rainy season due to increased rainfall. However, all RCPs and periods decrease basin soil water compared to the baseline. All RCPs and periods enhance actual and PET during the small, main, and dry season months. Evapotranspiration increased in all months due to higher temperatures.

Figure 16 displays the seasonal variation in AET and PET because of S3 change. The results indicated that evapotranspiration would rise in all seasons and periods. For instance, AET and PET tend to rise by 25 and 16%, respectively, under the RCP8.5 at the end of the 21st century.

This study investigated the separate and dual effects of LULC and CC using the SWAT model. Consequently, model input parameters were also generated using LARS-WG for climate and CA-Markov chain for LULC simulation. Multiple data sources were used to get input factors like hydrometeorological variables, soil data, DEM-SRTM data, and LULC, which were applied to simulate the hydrological response of the LTB as closely as possible. GCMs are employed to examine changes in various precipitation attributes, including spatial extent, timing, duration, frequency, and intensity of precipitation events. Nevertheless, biases and uncertainties in GCMs limit their accuracy and performance in simulating these attributes across different regions (Getachew et al. 2021b; Abbas et al. 2022).

To develop small-scale adaptation plans, large-scale GCM data were downscaled for the basin to account for the effects of changing climate on water resources. In this study, Wilby & Dawson (2013) said that downscaling GCMs to a small-scale level enhances the effectiveness of modeling the hydrological effects of changing climate and enables the implementation of suitable adaptation methods. This research also uses CHIRPS data for the simulation of the baseline hydrological model as well as the downscaling of GCMs CMIP5 models. The finding indicated that there is good agreement between baseline climate data and downscaled historical climate which indicates highly dependable satellite data products for hydrological analysis. In this respect, Du Plessis & Kibii (2021) indicate that CHIRPS data correlate well with observed monthly rainfall data for all stations used, having an average coefficient of determination of 0.6 and bias of 0.95 and recommended its applicability for hydrological analysis in South Africa.

Significant changes have occurred in the mean annual maximum and minimum temperatures, as well as precipitation, during the 2030s and 2070s. Under the RCP8.5, the ensemble mean annual maximum temperature tends to rise by up to 1.2 °C by the 2030s and 4.3 °C by the 2070s while the minimum temperature likewise increased up to 5.5 °C by the 2070s in the basin, respectively. Precipitation is also simulated to rise to 16% under RCP8.5 by the 2070s periods in the LTB. The results are in line with other studies conducted in Ethiopia and East Africa (EPCC 2015; Birara et al. 2018; Gebrechorkos et al. 2019). This finding aligns with other findings made by (WWF 2006; IPCC 2013; Gebrechorkos et al. 2019), in East Africa which projects rainfall might increase by up to 20%. A decline in precipitation along with an increase in temperature would have a significant effect on energy, water, and agriculture. On the other hand, an increase in rain during the short rainy season is needed to prepare the soil and seed crops that take a long time to mature and enhance irrigation in the LTB.

This finding also confirmed that seasonal variability in rainfall has shown considerable changes in the water balance components of the basin rather than cumulative CC. In line with this study, Silva-Júnior et al. (2021) indicated that the effects of changes in hydrology show strong spatial and temporal variations due to variability in climatic factors such as precipitation and watershed heterogeneity. According to Konapala et al. (2020), patterns of water availability, which have implications for ecosystems and human civilization, are influenced by both seasonal and annual mean precipitation and evaporation.

Bekele et al. (2021) reported that temperature and precipitation have been increasing, resulting in constant streamflow variability. Moreover, there is a notable seasonal shift from the existing primary rainfall period (June to September) to an earlier onset (January to May) with less pronounced peaks. This shift results in a longer duration of the rainfall season in the Blue Nile Basin under current CC scenarios. Cherie (2013) and Dile et al. (2013) reported that since most of Ethiopia's crops are grown in the summer and spring, CC may help the country's rain-fed agriculture, even though it hurts evapotranspiration because both the highest and lowest temperatures are rising. Furthermore, Abbas et al. (2022, 2023) indicated that the reduction in winter precipitation in Pakistan has substantial consequences for crop water needs, river flow, and various sectors, including industrial and domestic applications. According to Singh & Kumar (2018), the UBNRB will likely grow wetter and warmer in the 2050s due to CC. Beyene et al. (2010) revealed that, contrary to the results of other studies, summer rainfall may increase. Gebrechorkos et al. (2020) also indicated that streamflow and soil–water balance conditions are impacted by projected increases in evaporation in Ethiopia, which in turn affect agricultural output, hydropower output, and ecosystem services.

These findings suggest that streamflow, soil water, runoff, and lateral flow are projected to increase in future periods due to the influence of climate and land use change in both medium- and high-emission scenarios (RCP 4.5 and RCP 8.5) in the LTB, particularly during dry and short rainy seasons. However, during the basin's main rainy season, streamflow, soil water, surface runoff, and lateral flow are expected to decrease. Additionally, AET, PET, percolation, and soil water are predicted to increase during the dry, short, and main rainy seasons.

This study revealed that changes in climate and LULC had substantial impacts on the response of hydrological components of the LTB. Hence, the integrated impacts of CC and land use change on hydrological response are higher than the separate impacts of each scenario. According to this research, Woldesenbet et al. (2018a) and Aredehey et al. (2020) reported that CC would have a larger effect on hydrological components than the LULC change. On the other hand, Cuo et al. (2013) stated that changes in LULC are more likely to have an effect on basin hydrology than CC. Tan et al. (2022) also reported that both CC and LULC change reduced the streamflow in all eight estuaries, and the impact of LULC change was slightly larger than that of CC. Singh & Kumar (2018) claim that low flow is the result of climate and LULC change-related consequences. Paul & Rashid (2017) also suggested that the dual effects of LULC and climate will amplify hydrological changes in the James River. Furthermore, it revealed that the dual impacts of climate and LULC change would have greater effects on hydrological components than each cause acting alone (Setegn et al. 2011; Talib & Randhir 2017; Tariku et al. 2020).

Increasing water yield, soil water, percolation, and lateral flow are favorable for the basin to support its agriculture-based economy and to supply the energy demand of its fast-growing population. During the dry season, lateral flow is large and one of the primary sources of streamflow. Lateral flow is essential during the dry season for maintaining ecological water needs and irrigation, which can aid in mitigating drought (Liu et al. 2015; Getachew et al. 2021b). According to the findings, it is likely that lateral flow, water yield, soil water, and percolation will rise in the LTB, especially during the short and dry seasons. During the short wet and dry seasons, baseflow will likely contribute more to streamflow. Thus, it will likely contribute more to streamflow during a short and dry season.

Using the SWAT model, the study examined the combined and independent effects of LULC changes and climate on the hydrology of the LTB. LULC and CC projections were generated using CA-Markov chains and LARS-WG, respectively. Both models accurately replicated the climate and land use characteristics of the basin. The results indicated a tendency for the mean annual maximum and minimum temperatures in the basin to increase in the upcoming periods, particularly when considered alongside precipitation.

Using the CA-Markov chain, a future LULC change simulation indicated that cultivated and built-up areas are projected to increase, whereas forests, shrubland, and grassland areas are likely to decline in the future. This will increase evaporation, baseflow, streamflow, water yield, soil water, and AET and PET of the basin.

Changes in LULC are simulated to have a bearing on the water balance of the LTB in the 2030s and 2070s, possibly increasing streamflow, lateral flow, water yield, and soil water while decreasing AET and PET. In addition, CC is projected to enhance AET, PET, lateral flow, streamflow, percolation, water yield, and soil water by the 2030s and 2070s. Furthermore, the integrated effects of CC and LULC have a larger effect on the response of hydrological components of the basin than their separate impacts. Therefore, understanding the integrated impacts of climate and land use change on hydrological responses should be a prerequisite for the planning and implementation of water resource management strategies in the basin.

Uncertain predictions regarding land use change can affect various hydrological processes, including runoff, infiltration, and groundwater recharge. Consequently, this uncertainty impacts the quantity, timing, and quality of water available in river basins and aquifers. Moreover, uncertainties in climate projections increase the vulnerability of water resources to extreme weather events like storms, floods, and droughts. In essence, the uncertainty surrounding projections of land use change and CC significantly hinders the assessment and management of water resources.

The SWAT model effectively simulated the separate and integrated impacts of climate and land use change on the hydrology of the LTB. Even though the research covered both individual and combined effects on the water balance of the LTB, CMIP5 GCMs were considered. The whole impact of CC must therefore be determined through the latest model simulations, such as CMIP6, of shared socio-economic scenarios from the SWAT simulations. It is thought that the multiple and most recent GCM ensembles will make models less uncertain about how water will react. When managing water resources, it is essential to examine how LULC and the changing climate may impact the water balance of the basin. To offset the impacts of the rapidly changing climate on the river basin, climate-resilient management practices must also be used, as follows:

Promote the adoption of water-saving strategies such as rainwater collection, water recycling, and efficient irrigation to mitigate the impacts of reduced precipitation. Diversify water sources by investing in alternatives like desalination, groundwater recharge, and wastewater reuse to reduce dependence on precipitation-dependent sources. Additionally, prioritize the restoration and protection of natural ecosystems such as watersheds, wetlands, and forests, as they play a crucial role in regulating water flow, preventing floods, and replenishing groundwater.

Implement land use planning regulations that prioritize sustainable land management practices, including measures to reduce urban sprawl, conserve farmland, and protect riparian areas. Invest in green infrastructure solutions, such as artificial wetlands, green roofs, and permeable pavements, to mitigate the impacts of changing land use on water quantity and quality. Enforce laws and policies, such as stormwater management guidelines, riparian buffer zone standards, and erosion control strategies, to address the effects of land use on water resources and ensure sustainable management practices are upheld.

Furthermore, the findings of this study validate that the CHIRPS produces hydrological simulations that are more reliable. Thus, it is recommended to be used for any research efforts that could be used to model hydrological responses, drought analysis, and water resources management.

We extend our gratitude to the National Meteorological Services Agency of Ethiopia and the Hydrology Department of the Ministry of Water, Irrigation, and Electricity of Ethiopia, as well as the Ministry of Agriculture, for generously providing the meteorological data, discharge data, and soil data, respectively. We also appreciate the assistance of the United States Geological Survey (USGS) and https://earthexplorer.usgs.gov/ for facilitating the download of elevation and Landsat data.

B.G.T. has contributed to the design, processing, and analysis of data, sample design, evaluation, interpretation of results, and manuscript revisions. The manuscript was edited, commented on, and given ideas along the writing process by M.T. and K.G. The final manuscript was read and approved by all writers.

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

The authors declare there is no conflict.