Modeling the Impact of Climate Change on Hydrological Responses in the Lake Tana Basin, Ethiopia

Hydrologic systems have been changing due to the impact of climate change and climate variability. The impacts of climate change are set to increase in the future due to the rise of global warming. Quantifying the impact of climate change on the spatial and temporal hydrological processes is important for integrated water resource management. The Lake Tana basin, which is the source of the Upper Blue Nile, is vulnerable to climate change and variability. This study was carried out in the four major tributary watersheds of the Lake Tana basin: Gilgel Abay, Gumara, Ribb, and Megech. The climate model and hydrological model was used to (i) to evaluate the performance of the Soil and Water Assessment Tool for study watershed, (ii) to assess the future rainfall and temperature variability and change in the study watershed, and (iii) to examine the impact of climate change on future watershed hydrology. The study used dynamically downscaled climate data for the baseline (2010–2015) and future period (2046–2051) under two Representative Concentration Pathways (RCP4.5 and RCP8.5). The climate scenarios were simulated using the Weather Research and Forecasting (WRF) model, with a 4-km horizontal resolution. A linear scaling method was applied to minimize model biases. The SWAT model was used to estimate the baseline and future hydrology using the bias-corrected climate data. both with between 0.71 to The vary

under both scenarios. On average, annual rainfall may increase by 9.8% and 21.2% under RCP4.5 and RCP8.5 scenarios, respectively. Minimum temperature may rise by 1.68 °C and 2.26 °C while maximum temperature may increase by 1.65 °C and 2.75°C under RCP4.5 and RCP8.5 scenarios, respectively. The changes in climate may cause an increase in surface runoff by 20. 9% and 46.5% under RCP4.5 and RCP8.5 scenarios, respectively, while the evapotranspiration increase by 4.7% and 12.2% under RCP4.5 and RCP8.5, respectively.

Conclusion
The findings provide valuable insights to implement appropriate water management strategies to mitigate and adapt to the negative impacts of climate change and variability on the Lake Tana basin, and other regions which have similar agroecology.

Background
Climate change is expected to happen due to the anthropogenic alterations of atmospheric composition through the burning of fossil fuels, such as oil and coal, which emit greenhouse gases into the atmosphere (IPCC 2014). Several studies indicated that climate change is one of the most important global environmental challenges to humanity affecting social and ecological systems (Ngongondo et al. 2013;Clifton et al. 2018). The hydrological cycle will be directly altered by the impacts of climate change, subsequently affect local water resource availability in most regions in the world (Faramarzi et al. 2013;Mechal et al. 2015;Li et al. 2016;Aduah et al. 2017;Awal et al. 2018;Jin et al. 2018). However, the magnitude and direction of the impact vary from region to region; for the impact severe to Ethiopia as its economy is rain-fed agriculture, which is highly vulnerable to climate change and variability (Conway and Schipper 2011). Understanding the impacts of climate change is vital especially in regions where the social and ecological systems are highly dependent on climate and water resources such as in Africa.
The potential impacts of climate change and variability on water resources have been studied in different regions in Africa. For example, Faramarzi et al. (2013) analyzed the impact of climate change on freshwater availability in Africa at a coarser resolution for the period of 2020-2040. The results showed that the available water in Africa may increase for entire Africa. Aich et al. (2014) assessed the impact of climate change on streamflow of four large representative African river basins (Niger, Upper Blue Nile, Oubangui and Limpopo basin) and showed a statistically significant increase in streamflow for the period 2070 to 2099 compared to the baseline period of 1970 to 1999. Liersch et al. (2018) also showed that climate change may decrease streamflow of the Upper Blue Nile River for the months June and July, but an increase for the period August to November. Localscale climate change studies have also been conducted to assess the impacts of climate change and variability on the hydrology of the Upper Blue Nile watersheds (Abdo et al. 2009;Setegn et al. 2011;Dile et al. 2013;Nigatu et al. 2016;Abebe and Kebede 2017;Birhan and Worku 2018). For example, Dile et al. (2013)  The climate data was used as input for the General Water Loading Function (GWLF) hydrological model to simulate runoffs. The results showed that the future (2020-2039) runoff may increase in both watersheds under all the six GCM outputs.
Besides, Melke and Abegaz (2017) estimated the impact of climate change on streamflow in Gumara watershed using SDSM downscaled HadCM3A2a and HadCM3B2a climate outputs in which they found different trends of streamflow between the studied GCMs. Although the studies used different emission scenarios, GCMs, downscaling techniques, and hydrological models, the findings attest that climate change may unequivocally affect the water availability in Africa.
Although there is a consensus that climate change may affect the water resources in the Lake Tana basin, most of the previous studies in the basin assumed empiricalstatistical relationships between large-scale predictors (i.e. GCM-derived atmospheric parameters) and local predictands (e.g. rainfall or temperature) (Abdo et al. 2009;Setegn et al. 2011;Dile et al. 2013;Enyew et al. 2014;Ayele et al. 2016). However, the climate has a strong association with local soil conditions, topography, and land use dynamics (Haile et al. 2009;Collow et al. 2014;Cao et al. 2015), which is not properly accounted for in a statistical predictor-predictand relationship. Although most of these concerns are better addressed in the Fifth generation IPCC Assessment Report (IPCC-AR5) Representative Concentration Pathways (RCPs) (van Vuuren et al. 2011;Knutti and Sedláček 2013;), most of the climate change studies in the Lake Tana basin used GCM based IPCC AR4 Special Report on Emissions Scenarios (SRES). Therefore, estimates from these studies are less reliable since they do not account for local details such as land use, topography, and soil due to the coarse spatial resolution of the GCMs.
Although watershed studies in other regions showed that regional climate models (RCMs) provide more reliable results in assessing climate change impacts than GCMbased estimates (Fang et al. 2015;Yhang et al. 2017;Wulong et al. 2018), such studies are lacking in the Lake Tana basin.
This study, therefore, applied a high-resolution RCM with RCP emission scenarios to study the impact of climate change on the hydrology of the Gilgel Abay, Gumara, Ribb, and Megech of the Lake Tana basin. These watershed contribute more than 93% of the inflow to Lake Tana (Kebede et al. 2006). This study is the first of its kind to assess the impact of climate change on the seasonal and annual water resources of the Lake Tana watersheds at finer spatial and temporal resolution. The findings of this study will have paramount importance for the policymakers in Ethiopia since the Lake Tana basin is one of the main growth corridors in the country because of its significance for hydropower generation and irrigation.
Besides, the employed methodology can be replicated to investigate the impact of climate change on other basins.

Study Area
The study is conducted in the Lake Tana basin, which is located in the Amhara region of Ethiopia ( Figure 1). The Lake Tana basin is the headwater of the Upper Blue Nile Basin having a catchment area of ~15,000 km 2 . The elevation of the basin ranges between 1780 and 4100 meter above sea level (masl). The Lake Tana, which is the largest freshwater body in Ethiopia, covers 20% of the basin area. Although more than 40 tributary rivers flow into Lake Tana, the major tributaries are Gilgel Abay, Gumara, Ribb, and Megech, which account for more than 93% of the flow to Lake Tana (Kebede et al. 2006). The Gilgel Abay river is the largest tributary that originates from the southern part of the basin and has a catchment area of 4553 km 2 having its outlet at the Lake. The Ribb river is the second largest tributary with a catchment area of 1970 km 2 joining the Lake from east. The Gumara river flows from the eastern part of the basin draining 1780 km 2 of the Lake Tana basin. The Megech watershed is the smallest of the four major tributaries of the Lake Tana basin with a catchment area of 860 km 2 . The Megech river joins the Lake Tana from the north.
The climate in the Lake Tana basin is monsoon tropical highland with a rainy season from June to September and a dry season from October to March. The rainfall in the basin differs widely between mean annual rainfall of 800 mm in the northeastern part of the basin and 2000 mm in the southern. The temperature is warmer around Lake Tana with mean annual temperature between 18.6°C and 21.2°C but gets cooler moving away from Lake Tana.

Modeling Approach
This study involved climate and hydrological modeling to estimate the impact of climate change in the Lake Tana basin. Initially, the quality of all observed climate and streamflow data were checked using basic statistical error checking approaches. Thereafter, the baseline and two future climate scenarios (RCP4.5 and RCP8.5) were dynamically downscaled using the Weather Research and Forecasting (WRF) model. The downscaled climate datasets were corrected from model biases using the linear scaling method. The Soil and Water Assessment Tool (SWAT) model was calibrated and validated using observed climate and streamflow data. Finally, the calibrated and validated SWAT model was used to estimate the baseline and future hydrology using the bias-corrected climate data. The detailed procedures for the climate and hydrologic modeling are presented as follows.

Climate Simulation
Rainfall and temperature datasets of this study were estimated using WRF (version 3.8) regional climate model. The WRF model assumes a fully compressible, and a terrain-following hydrostatic pressure coordinate (Skamarock et al. 2008 for the baseline run, and from 1 January 2046 to 31 December 2051 for the future (RCP4.5 and RCP8.5) runs. The simulation outputs were daily rainfall, and daily maximum and minimum temperature at 2 meter above the ground over the respective simulation periods.
Direct application of the RCM outputs for assessing the impact of climate change at the local scale is not recommended due to biases resulting from limitations in the global forcing fields and regional climate models (Hurkmans et al. 2010;Smitha et al. 2018). Therefore, the linear scaling bias correction method was applied to minimize biases from rainfall and temperature data. This method was selected due to its simplicity and satisfactory performance in correcting climatological biases (Fang et al. 2015;Sisay et al. 2017). The baseline rainfall and temperature data were corrected with a multiplier and an additive term on a 10 days' average basis (Fang et al. 2015;Sisay et al. 2017). The correction factors were applied for the future climate of the RCP4.5 and RCP8.5 emission scenarios. The baseline and projected rainfall biases were corrected using Equation 1 and 2, respectively.
The subscripts refer to similar meaning as that of the equations in the rainfall correction equations.
To verify the bias correction method and to show the improvement obtained after bias correction, the average daily observation and WRF simulation (before and after bias correction) were compared using statistical measures such as the correlation coefficient (r) and root mean square error (RMSE).

Hydrological Model
The Soil and Water Assessment Tool (SWAT) model was used to assess the impact of climate change on the Gilgel Abay, Gumara, Ribb, and Megech watersheds. The SWAT model is developed to predict the impact of land use, land management practices, and climate change on water balance, nutrient cycling, and sediment transport at a watershed to a river basin and regional scale. It is a physically-based semi-distributed model that functions on a continuous-time step. In the SWAT model, a watershed is divided into sub-basins where each sub-basin will have one or more Hydrologic Response Units (HRUs). HRU is a unique combination of land use, soil, and slope classes. Most of the biophysical processes are calculated at the HRU level and aggregated at the sub-basin level (Neitsch et al. 2011). The SWAT model has been widely applied in the Upper Blue Nile basin and showed satisfactory results (Van Griensven et al. 2012;Dile et al. 2016;Fentaw et al. 2018).

Input data
The model requires spatio-temporal data such as topography, land use, soil, weather, and land-management interventions to simulate different processes in the watershed. The model also requires hydrological data (e.g. streamflow) to calibrate and validate the performance of the model. Digital Elevation Model (DEM) is used to delineate the watershed and generate other topographic characteristics. The study used DEM data from the Shuttle Radar Topographic Mission (SRTM), which has a spatial resolution of 90 m. Land use data is required in the SWAT model to define the HRUs and link the landscape with the crop database. The land use data was produced from 30 m Landsat image using a supervised classification method for the 2016 Ethiopia land cover (Figure 2). The classification was satisfactory with an overall accuracy of 87.9% and a Kappa coefficient of 0.79 (Kindu et al. 2013). The original raw data was obtained from the Regional Centre for Mapping of Resources for Development. The soil data was used to create HRUs and provide physiochemical information about the soils in the landscape. The soil map was obtained from the chemical soil properties required by the SWAT model were extracted from the International Soil Reference and Information Center. Daily rainfall, minimum/maximum temperature, solar radiation, wind speed and humidity data are required to simulate processes in the SWAT model. Climate data within and around the Lake Tana basin were collected from the Ethiopian National Meteorology Services Agency. The rainfall and temperature data were available for 27 climatic stations (Figure 1) for the period 1982 to 2015. Missing values in climate data were filled using the SWAT built-in weather generator. Streamflow data is often used to calibrate and validate the performance of the SWAT model. This study used monthly streamflow data of Gilgel Abay, Gumara, Ribb and Megech for model calibration and validation. Observed daily and monthly streamflow data for the period 1990-2015 was obtained from the EMWIE.

SWAT model setup, calibration and validation
The watershed boundary was delineated by considering the hydrometer station as an outlet (Figure 1). The watershed was discretized using a threshold area of 25 km 2 , which provided 31, 26, 27, 15 sub-basins for Gilgel Abay, Gumara, Ribb, and Megech watersheds, respectively. The HRUs were defined using accounting all the slope, land use, and soil features in the sub-basins. Angereb reservoir is located in the Megech watershed, which was included in the SWAT model setup. Angereb dam was constructed in 1997 to supply potable water to Gondar town (Haregeweyn et al. 2012). The reservoir has a total area of 50 ha and 60 ha at its principal spillway and emergency spillway, respectively. The volume of the reservoir at the principal and emergency spillway is 3,530,000 m 3 and 5,160,000 m 3 , respectively. SWAT has several options to simulate biophysical processes. The surface runoff was estimated using the Soil Conservation Service Curve Number (SCS CN) method (Betrie et al. 2011;Fentaw et al. 2018). Flow within the channels was routed using variable storage method. The evapotranspiration was estimated using the Penman-Monteith method (Gebre and Ludwig 2015).
The model parameters were calibrated using the Sequential Uncertainty Fitting version 2 (SUFI-2) in the SWAT-CUP (SWAT Calibration and Uncertainty Program) (Abbaspour et al. 2004). The models were calibrated and validated using observed monthly streamflow data at the Gilgel Abay, Gumara, Ribb, and Megech river gauging stations (Figure 1). The calibration and validation were conducted for the periods 1900-2004 and 2005-2015, respectively. The model was warm-up for three years (1987)(1988)(1989) to properly initiate the biophysical processes. The model calibration considered 18 hydrological parameters (Table 1), which were selected based on the literature (Setegn et al. 2008;Gebremicael et al. 2013;Dile et al. 2016;Halefom et al. 2018;Teklay et al. 2018).

Model Evaluation
Model performance was evaluated by comparing the observed and simulated streamflow using the Nash-Sutcliffe efficiency (NSE) and percent bias (PBIAS). The NSE is a normalized statistic that estimates the relative magnitude of the residual variance compared to the observed data variance (Nash and Sutcliffe 1970). PBIAS compares the average tendency of the simulated data to the corresponding observed data (Gupta et al. 1999). The NSE value theoretically ranges from negative infinity to 1; a value of 1 corresponds to a perfect match between observed and simulated values (Nash and Sutcliffe 1970). While the PBIAS value can be positive or negative, in which a value of zero represents the best model simulation performance (Moriasi et al. 2007). A positive PBIAS value indicates model underestimations, and a negative value indicates model overestimations. The NSE and PBIAS are computed using Equation 5 and 6, respectively. Moriasi et al. (2007) suggested that a model simulation that provides an NSE value of >0.5 and PBIAS value of is considered as satisfactory.
The uncertainty of the simulations was estimated using a p-factor and r-factor (Abbaspour et al. 2007). The p-factor measures percent of observed data bracketed by the 95% prediction uncertainty (95PPU). While the r-factor measures the thickness of the uncertainty band. It is estimated by dividing the average thickness of the 95PPU band by the standard deviation of the observed data. The p-factor value ranges between 0 and 1, while the r-factor ranges between 0 and infinity. A pfactor of 1 and r-factor of 0 represent a simulation that exactly matches the observed data. According to Abbaspour et al. (2007), a model providing a p-factor ≥ 0.75 and r-factor ≤ 1.5 is considered robust for streamflow simulation.

Bias Correction of Climate Variable
Daily rainfall and temperature data were extracted from the grid-based WRF simulations using coordinates of the 27 meteorological stations (Figure 1). Figure 3 shows observed and WRF simulated rainfall and temperature data, averaged over the Lake Tana basin, for the baseline period (2010 -2015). The WRF model without bias correction performed well in simulating rainfall only during the dry season (January to May); while it considerably overestimated rainfall during the rainy season (June to September) (Figure 3a). Besides, the model showed consistent cold biases of maximum temperature (Tmax) across all the months (Figure 3b While bias correction improved overall performances of the WRF model in simulating rainfall and temperature; there were differences in the model evaluation statistics (Table 2). For temperature, bias-corrected values were very close to the observed.
After bias correction, the average cold bias was substantially reduced from 1.42 ºC to 0.02 ºC for maximum temperature and the warm bias decreased from 1.45 ºC to 0.03 ºC for minimum temperature (Table 2). Moreover, the RMSE was considerably reduced by 1.27 ºC and 1.36 ºC for maximum and minimum temperatures, respectively. However, RMSE for rainfall was decreased from 5.43 to 3.88 mm/day.
In addition, bias-corrected simulations showed good agreement with the observed data. Correlations between observed and simulated data were improved from 0.65, 0.76, and 0.60 to 0.78, 0.89, and 0.84 for rainfall, maximum and minimum temperatures, respectively. These results suggest that bias correction output is acceptable and consistent with the previous studies (Gebre and Ludwig 2015;Sisay et al. 2017;Goshime et al. 2019).

Projected temperature
Temperature projections using the WRF model were averaged by season to investigate overall trends of future climate in Gilgel Abay, Gumara, Ribb, and Megech watersheds. The baseline (2010 -2015) and projected (2046 -2051) seasonal temperature distributions for the four watersheds are presented in Figure   4. The baseline and projected minimum and maximum temperatures show the same seasonal pattern. The simulated maximum temperature is high during spring, whereas minimum temperature is high during summer. Compared to the baseline period, both RCP4.5 and RCP8.5 seasonal maximum and minimum temperatures were predicted to increase in all the watersheds. In the Gilgel Abay watershed, the mean annual maximum temperature is projected to increase by 1.64 °C and 2.76 °C under RCP4.5 and RCP8.5 scenarios, respectively, while minimum temperature is expected to increase by 1.65 °C under the RCP4.5 and 2.16 °C under the RCP8.5.
The seasonal increase in maximum temperature showed a large variation with a range from 0.28 °C to 3.19 °C under RCP4.5 and 1.98 °C to 3.48 °C under RCP8.5 ( Figure 4a). The maximum temperature increase is projected to be higher during autumn under both scenarios (Figure 4a). However, the minimum temperature change during summer is projected to be as high as 2.17 °C and 3.12 °C under RCP4.5 and RCP8.5 scenarios, respectively (Figure 4b). In the Gumara watershed, the mean annual maximum temperature is projected to increase by 1.85 °C and 2.94 °C and minimum temperature by 1.77 °C and 2.33 °C under RCP4.5 and RCP8.5, respectively. The seasonal increase in maximum temperature displayed a large variation with a range from 0.63 °C to 3.23 °C for RCP4.5 and 2.05 °C to 3.84°C for RCP8.5 (Figure 4c). The highest minimum temperature increase is projected to occur in summer with 2.16 °C and 3.15 °C for RCP4.5 and RCP8.5 projections, respectively (Figure 4d). In the Ribb watershed, the mean annual maximum temperature increase by 1.49 °C and 2.67 °C and minimum temperature by 1.66 °C and 2.25 °C under RCP4.5 and RCP8.5, respectively. Compared to the baseline period, the RCP4.5 maximum temperature seasonal increase is projected between 0.19 °C and 3.02 °C, while under RCP8.5 between 1.62 °C and 3.65 °C ( Figure 4e). The simulated minimum temperature under RCP4.5 did not show considerable variation among seasons (Figure 4f). In the Megech watershed, the mean annual maximum temperature is projected to increase on average by 1.62 °C and 2.63 °C while the average increase in minimum temperature is projected to increase by 1.63 °C and 2.28 °C under RCP4.5 and RCP8.5, respectively. The seasonal maximum temperature increase is expected to vary between 0.30 °C to 3.18 °C and 1.85 °C to 3.60 °C under for RCP4.5 and RCP8.5, respectively (Table   4g).

Projected rainfall
The seasonal distributions of rainfall under the two scenarios were similar to the baseline period with the highest rainfall (705 -1547 mm) occurring during summer while the smallest rainfall (4 -33 mm) is during winter ( Figure 5). The mean annual rainfall under RCP4.5 and RCP8.5 scenarios are projected to be higher than the baseline period. Similar to temperature, simulation of rainfall under RCP8.5 is expected to be substantially greater than the RCP4.5 simulations in most of the watersheds. Under RCP8.5, the mean annual rainfall is projected to increase by 668 mm, 337 mm, 275 mm, and 33.9 mm in Gilgel Abay, Gumara, Ribb, and Megech watersheds, respectively. The greatest increase (34.6%) is expected in Gilgel Abay

SWAT Model Calibration and Validation
Statistical analysis of the SWAT model results indicates that the monthly streamflow was well simulated for most of the calibration period, with NSE > 0.72 for each watershed and PBIAS well within the satisfactory range of ±25% according to Moriasi et al. (2007) (Table 3). Evaluation of the average monthly streamflow during calibration revealed that the SWAT model overestimated the streamflow in the Gilgel Abay, Ribb, and Megech watershed by 1. 1%, 15.4%, and 15.1%, respectively, while the model underestimated the flow in the Gumara watershed by about 10.1%.
The agreement between observed and simulated flow was slightly better for the Gilgel Abay and Gumara watersheds, compared to the Ribb and Megech watersheds.
Validation results also showed overall very good model performances in all watersheds, with NSE values > 0.75 and PBIAS < ±10 (Table 3). During the validation period, the SWAT model slightly overestimated average streamflow by about 6.1%, 7.2%, and 8.3% in the Gilgel Abay, Gumara, and Ribb watersheds, respectively, whereas it underestimated the stream flow by about 8.4% in Megech watershed.
The uncertainty bands of the model captured most of the observed streamflow with p-factor > 0.82 for Gilgel Abay and Gumara watersheds. The 95PPU relative width was slightly small (0.75-0.78) in Megech and Ribb watersheds, but it was under the acceptable model estimates (Abbaspour et al. 2007). Overall, model simulation results demonstrated that the calibrated and validated model replicated observed streamflow very well in all study watersheds (Moriasi et al. 2007;Arnold et al. 2012). Our findings are in agreement with previous studies in the region (Dile et al. 2016;Halefom et al. 2018;Teklay et al. 2018). Therefore, it is safe to conclude that the SWAT model is reliable to simulate the watershed hydrological processes in the Ethiopian highlands.
Moreover, the graphical comparison results showed that simulated streamflow was in a very good agreement with observations at the four gauging stations ( Figure 6). However, the model tends to underestimate peak flows while overestimating low flows depending on the study watershed. The majority of the peak events were not captured adequately over the study watersheds owing to the poor performance of the SWAT model for peak flow simulations (Polanco et al. 2017;Lee et al. 2018).

The Impacts of Climate Change on Hydrological Responses
The major hydrological components were simulated using the calibrated SWAT model and bias-corrected climate date. The simulated hydrological components for the projected period (2046-2051) under RCP4.5 and RCP8.5 scenarios were compared to the corresponding values in the baseline period (2010)(2011)(2012)(2013)(2014)(2015). Overall, the mean annual rainfall, streamflow, and evapotranspiration are projected to increase compared to the baseline period in all the four study watersheds ( Table 4).
The magnitude and direction of seasonal streamflow and evapotranspiration changes were different among watersheds except for Gumara and Ribb.
In the Gilgel Abay watershed, Overall, the mean annual streamflow may increase by 38.9% and 58.2%, which corresponds to 22.1% and 34.7% increases in rainfall under RCP4.5 and RCP8.5 scenarios, respectively (Table 4). On a seasonal scale, the streamflow under RCP4.5 and RCP8.5 is projected to increase in all seasons except for winter (Figure 7a). The streamflow changes were mostly related to rainfall change over the watershed. Under RCP8.5, the changes in summer rainfall clearly showed the largest increases (337 mm) in comparison to other seasons (Figure 5a), with a maximum streamflow increase of 245 mm during summer. Similarly, the RCP4.5 autumn rainfall increase by 131 mm (30.3%), thereby considerably increasing streamflow by 180 mm (61.2%). However, the winter season streamflow is projected to decrease slightly (2.9 mm) under RCP4.5, despite projected increases in rainfall during the same time, since the rainfall increase substantially contributed to soil moisture increase (Table S2). The mean annual evapotranspiration under RCP8.5 is expected to increase by 10.7% due to the average temperature increase of 2.5 °C and rainfall increase of 34.7%. However, the mean annual evapotranspiration under RCP4.5 shows only a marginally increase (0.1%). The evapotranspiration may increase in spring and summer under both emission scenarios. This may be due to the increase in average temperature and rainfall in these seasons. Although the temperature increase may occur in winter, evapotranspiration may decrease by 7.1 mm (14.1%) and 15.8 mm (31.1%) under RCP4.5 and RCP8.5 scenarios, respectively. These results indicated that all the temperature increases may not contribute to evapotranspiration increases.
As can be seen in Figure 7, the direction of hydrologic response changes in Gumara and Ribb watersheds were similar, since the hydrologic model in both watersheds used almost the same meteorological stations that are located in the eastern part of the Lake Tana basin (Figure 1). Therefore, the results of the Gumara watershed are presented here for clarity purposes. The mean annual streamflow increase under RCP4.5 is very small (0.1%), while the mean annual streamflow increase under RCP8.5 is considerable large (26.7%). This may be due to a larger increase mean annual rainfall under RCP8.5 than under RCP4.5. Under RCP4.5, the streamflow is projected to decrease in most season, with the largest decrease of 141.5 mm (26.1%) during summer. However, the streamflow increase during spring is extremely large (195.2 mm), which may cause flooding. The RCP8.5 streamflow may increase in all seasons with a range between 33.8 mm (195.5%) during spring and 158 mm (29.2%) during summer. Such considerable increases were driven by projected increases in rainfall during summer (126.4 mm). The mean annual evapotranspiration is projected to increase by 9.6% and 16.5% under RCP4.5 and RCP8.5 scenarios, respectively, which corresponds with the projected increases in average annual temperature by 1.81 °C and 2.64 °C and rainfall by 3.9% and 22.1%, under RCP4.5 and RCP8.5, respectively (Table 4). The RCP4.5 projections show reductions in evapotranspiration during summer and autumn, while the RCP8.5 scenarios show increases in evapotranspiration during all four seasons except for winter.
In the Megech watershed, the mean annual streamflow may increase in the future period under both scenarios. Unlike other watersheds, the projected increase in streamflow under RCP4.5 is slightly greater than under RCP8.5. This could be due to the grater temperature increases under RCP8.5 (0.84 °C more than RCP4.5) while rainfall was more or less the same (Table 4)

Discussion
The Lake Tana basin is one of the growth corridors in Ethiopia where large and small scale irrigation and hydropower projects are under construction. Many studies have indicated that climate change has a substantial impact on streamflow from the major tributaries of the Lake Tana basin, and the studies have suggested that the streamflow variability of tributary rivers was the result of rainfall variability and temperature increase. These studies used rainfall and temperature output from coarse spatial resolution. However, few studies have evaluated the impact of climate change on streamflow response in the major tributaries rivers of the Lake Tana basin using dynamically downscaled climate data.
The projected mean annual maximum and minimum temperatures under RCP4.5 and RCP8.5 scenarios are higher than the baseline period. The mean annual temperature increase under RCP8.5 is considerably larger than that under RCP4.5 because the RCP8.5 scenario represents a higher radiative forces (Wulong et al. 2018). The climate projections results showed that overall warming over the Lake Tana basin, with an average temperature increase of 1.66 °C and 2.50 °C under RCP4.5 and RCP8.5 scenarios, respectively. These results are comparable with the previous studies (Dile et al. 2013;Nigatu et al. 2016;Liersch et al. 2018;Mekonnen and Disse 2018) in the Upper Blue Nile basin. Besides, the results agree with the study by Ayele et al. (2016) in Gilgel Abay and Gumara watersheds. They reported that projected (2020-2039) temperature in the A2/B2 scenarios will increase by 1.2/1.1°C and 0.8/0.7°C in Gilgel Abay and Gumara watershed, respectively. The simulated mean annual rainfall under RCP4.5 and RCP8.5 is larger than the baseline period.
The result showed that the RCP4.5 rainfall simulation increased rates were greater in spring than other seasons, which is in agreement with Worqlul et al. (2018) findings in the Upper Blue Nile basin. Moreover, rainfall under RCP8.5 is expected to increase in all four seasons in all the watersheds. These results are in agreement with results from previous studies in the Lake Tana basin (Gebre and Ludwig 2015;Ayele et al. 2016;Nigatu et al. 2016;Melke and Abegaz 2017), though variations exist in the magnitude of change, which could be due to differences in GCM models, emission scenarios, and downscaling methods used between studies. However, this result contradicted with the study by Setegn et al. (2011) in the Lake Tana basin who reported decreasing trends in future rainfall. A decrease in seasonal rainfall in Gumara, Ribb, and Megech watershed resemble with the results by Beyene et al. (2010), who also showed a decrease in the mean monthly rainfall in 2040-2069 in Nile river basin, where our study watersheds are located.
Several researchers (Dile et al. 2013;Gebre and Ludwig 2015;Ayele et al. 2016) have shown that the streamflow and evapotranspiration were projected to increase in the major tributary rivers in the Lake Tana basin, and this is very similar to the results of this study. The seasonal streamflow variation mainly associated with seasonal rainfall variation. However, the streamflow increase rate was larger than the rainfall increases rate, which is consistent with the findings of Aich et al. (2014) from four African River basins. They reported that a 25% increase in rainfall leads to a 50% increase in the river discharge over Upper Blue Nile Basin. The increase in streamflow due to rainfall increment was also discussed by Pandey et al. (2017); they revealed that a 28% increase in annual rainfall may result in streamflow increased by approximately 49% from their study in the Armur watershed in Godavari river basin, India. The decrease in streamflow during the rainy season associated with the rainfall reduction corresponds with the findings of Worqlul et al. (2018) in the Upper Blue Nile Basin, Ethiopia. This highlights climate variation during the rainy season could have a profound effect on the annual hydrological responses. Although the rainfall under RCP4.5 slightly decreases during summer and autumn, the streamflow did not show a reduction trend in these seasons. This could be due to the excess soil moisture in the previous season (i.e. during spring, RCP4.5 soil moisture increase by 67.4 mm Table S2). Similar findings were reported in the Upper Blue Nile Basin (Roth et al. 2018) and Northern Lake Erie Basin in Canada . The evapotranspiration was expected to increase at rates similar to those of the temperature and rainfall dynamics, which is in agreement with findings (Gebre and Ludwig 2015;Lemann et al. 2017;Teklesadik et al. 2017;Birhan and Worku 2018) in the Upper Blue Nile Basin. The evapotranspiration, however, is expected to decrease during winter under RCP8.5, contrary to most of the studies conducted in the surrounding watersheds (Enyew et al. 2014;Ayele et al. 2016).
Although the climate variables were simulated using the RCM and updated emission scenarios, there are several uncertainties associated with the climate and hydrological models. Both hydrological and climate model simulations have not considered land-use dynamics; but land use has a significant role in the partitioning of the hydrologic components into infiltration, interception, and evapotranspiration (Wagner et al. 2016;Marhaento et al. 2018;Sunde et al. 2018;Woldesenbet et al. 2018). Besides, land use dynamics will directly influence the land surfaceatmosphere interaction and consequently alter the climate processes and patterns . Therefore, future studies should consider the combined impact of climate and land use change on hydrological responses. Moreover, climate simulation should incorporate a very fine spatial resolution to resolve the complex topography.

Conclusion
This study assessed the response of hydrological processes in four major tributaries of Lake Tana basin (Gilgel Abay, Gumara, Ribb, and Megech watershed) to a possible future climate change scenario using the WRF model and a physical distributed hydrological model (SWAT). Overall, WRF simulations reproduced average monthly observation but large biases were found in some cases. The linear scaling bias correction method was capable of correcting most biases with the WRF model. Climate projection showed that rainfall and temperature may increase in the mid-21 st century (2046)(2047)(2048)(2049)(2050)(2051) in all the four study watersheds. Compared to the baseline period (2010)(2011)(2012)(2013)(2014)(2015), the average temperature is likely to increase by 1.66°C and 2.50 °C under the RCP4.5 and RCP8.5 scenarios, respectively. Similarly, the mean annual rainfall is likely to increase by 9.8% and 21.1% under the RCP4.5 and RCP8.5 scenarios, respectively.
In all the watersheds, the mean annual streamflow is predicted to increase by 0.1 -38.9% and 4 -58.2% under RCP4.5 and RCP8.5 scenarios, respectively, which implies that the overall amount of the water resources may not decrease in the future period over the Lake Tana basin. However, streamflow seasonal variation is large under both scenarios in all four watersheds. Projected annual evapotranspiration in the watersheds will vary on average from 0.1 to 9.6% under RCP4.5 and 8.9 to 16.5% under RCP8.5. While the magnitude and direction of the changes of streamflow and evapotranspiration can vary, impacts of climate change on water resources in the Lake Tana basin need to be taken in to account for future water resources planning. Moreover, the results obtained from this modeling study provide critical insights, potentially helping resource managers and government policymakers to develop effective water resource management strategies in the face of climate change. Based on the results, flood controlling practices should be considered in Gilgel Abay watershed because streamflow may increase by 245 mm (46.1%) during the main rainy season. In the Gumara and Ribb watersheds, a considerable amount of streamflow is expected to occur during spring which could be used to support irrigation practices. However, the projected climate and hydrologic components of Megech watershed highlight the need to put in place appropriate adaptation measures to meet the future water demand for domestic and agricultural production. T a b l e 2. The statistical measures of climate variables before bias correction (WRF) and after bias correction (WRF-BC) for the Lake Tana basin from 2010-2015. The Lake Tana basin showing its four major watersheds (Gilgel Abay, Gumara, Ribb, and Meg Lake Tana basin land use and soil map subjected to hydrological model simulation.

Figure 3
Monthly mean observed and WRF simulated (before and after bias correction) rainfall (a), ma Seasonal and annual maximum temperature (a, c, e, and g) and minimum temperature (b, d,