Hydrologic responses to climate and land-use/land-cover changes in the Bilate catchment, Southern Ethiopia

The likely effects of climate and land-use/land-cover (LULC) changes on hydrologic processes in Bilate catchment, Ethiopia were evaluated. The study emphasizes the evaluation of individual and combined impacts on hydrologic responses of climate and LULC changes. Climatic scenarios included a downscaled regional climate model from CORDEX-Africa. The CA–Markov model was used to project LULC. The results revealed that distinct changes on hydrologic responses occurred which follow the direction of climate and LULC changes. A 30.87% decline in rainfall resulted in about 4.09, 1.43 and 3.57% decline in runoff, groundwater and water yield, respectively. A rise in mean temperature by 1.3 °C resulted in a 7 and 0.8% increase in potential and actual evapotranspiration, respectively. Runoff, groundwater and water yield are projected to decrease by 11.24, 12.54 and 11.54%; however, potential and actual evapotranspiration are projected to increase by 19 and 14.7%, respectively, under combined climate and LULC changes. The joint effects of climate and LULC changes on hydrologic responses in the forthcoming were higher than the variation trend of climate or LULC change alone. Climate change compared with LULC change has a higher impact on hydrologic responses. The results obtained provide further insight into future water balance, and assistance in water resources planning and management.


INTRODUCTION
Water managers and planners are facing substantial uncertainties in the future demand and availability of water. Land-use/ land-cover (LULC) and climate changes and their probable hydrological effects are increasingly contributing to this uncertainty. LULC and climate changes are expected to alter the timing and magnitude of hydrological processes. LULC and climate changes are important factors causing combined impacts on catchment hydrologic processes.
Many studies from different parts of the world show that LULC and climate changes significantly impact hydrologic processes. For example, Liu et al. (2020) found that the combined effect of LULC and climate changes increased runoff by 0.22 m 3 /s, in which LULC change caused the runoff increase by 0.07 m 3 /s attributed to the decreased area of arable land and the increased area of urban land in the Coastal Basin, Northern China. Besides, four future hypothetical LULC scenarios and climate change simulation showed a decrease in streamflow by 11.8 and 26.72%, while sediment yield decreased by 7.54, 19.4, 11.1 and 9.01% in the Upper Ebonyi River watershed, Nsukka, South-East Nigeria (Ndulue & Mbajiorgn 2018). Other studies have discussed the impacts of LULC and climate changes on catchment hydrologic processes at different places (Li et al. 2009;Zuo et al. 2016;Bessah et al. 2020). Studies in Blue Nile Basin, Ethiopia indicated that water resources are vulnerable to both LULC and climate changes (Woldesenbet et al. 2018;Gebresilassie et al. 2020;Tigabu et al. 2020;Getachew et al. 2021;Negesse 2021;Teklay et al. 2021). Besides, a study at the Rift Valley Basin, Ethiopia showed that the combined effect of LULC and climate changes significantly decreases streamflow by 11.8% and increases evapotranspiration by 2.2% at Meki River Basin (Legesse et al. 2010).
To keep up water resources in the desired manner, it is important to predict and evaluate the future water balance tendencies at the catchment scale. Therefore, the assessment of anthropogenic activities on LULC and climate changes on hydrologic processes is highly desired, which requires effective hydrologic modeling. Hydrologic models have been advanced, as they provide comprehensive analysis between LULC, climate and water resources (Germer et al. 2009). Zhang et al. (2020) employed the SWAT model to evaluate the influence of LULC change on the hydrologic cycle, in Su-Mi-Huai, China investigated the response of surface runoff, recharge, and provided the basis for rational water resources planning. The soil and water assessment tool (SWAT) has been widely employed for the evaluation of impacts of LULC and climate changes on hydrologic processes (Shooshtari et al. 2017;Yin et al. 2017;Berihun et al. 2019;Lamparter et al. 2019).
In several previous studies, the impacts of changing climate on hydrologic processes are investigated by using downscaled regional climate model data in the SWAT model (Boru et al. 2019;Worner et al. 2019;Yan et al. 2019). Regional climate models at their finer resolution allow the simulation of local climate conditions in detail and provide future predictions. A coordinated regional climate downscaling experiment has been planned to simulate the climate of Africa based on a variety of climate regimes (Giorgi et al. 2009). Some countries are vulnerable to future climate change, for example Ethiopia is often cited as one of the most extreme examples (Conway & Schipper 2010). The expected effects in Omo Gibe Basin, Ethiopia are already having a profound effect on hydrology in the Omo River flows (Mohammed 2013). Despite the vulnerability to climate change, few attempts have been made to indicate water scarcity for different sectors.
Bilate catchment is found in the Rift Valley Basin, Ethiopia where small-scale irrigation projects are planned to improve the livelihoods of the growing population. In Bilate catchment, there are some districts whose population density exceeds six hundred individuals per kilometer square (CSA 2013;Adugna 2014). Population growth aggravates LULC change because of very high pressure on the land resources. Sloppy terrain areas and grazing fields were converted into cultivation lands in the catchment. As a result reduction in natural vegetation, soil erosion, sediment and nutrient loads to water bodies became common problems. In the catchment, there is high competition for irrigation water. A few state farms, individual investors and farmers use Bilate River and its tributaries without water management. Besides, subsistence farming in the catchment is exposed to fluctuating rainfall, and the growing population depends on limited land and water resources. The Ethiopian economy is rain-fed agriculture, which is heavily sensitive to climate variability and change (Zerga & Gebeyehu 2018). The short rainy season (February-May) crop of 2021 had completely failed due to the highest decline in February, March and April rainfall in the catchment. However, there is inadequate information about LULC and climate change effects on water resources in the catchment. Bilate catchment has been exposed to problems, and the result of this study would be important to supplement the water and LULC management strategies. Therefore, the aim of this study is intended to explore the combined and individual impacts of LULC and climate changes on the hydrologic responses in Bilate catchment, Ethiopia. LULC and climate change scenarios were used for hydrologic simulations using the SWAT model for two future periods.

Study area
Bilate catchment covers 562,560 ha of land area and is located in the Rift Valley Lakes Basin, Ethiopia ( Figure 1). It is bounded by Omo Gibe Basin to the west, Abiyata and Shala Lakes to the east. The catchment draws out across different climate zones ranging from the high to the lowlands of the Rift Valley. The altitude of the catchment ranges from 1,174 meters at Lake Abaya to 3,330 meters above sea level. The elevated terrains are the Ambarcho, Duguna, Mugo and Damota mountains which made the catchment undulating and characterized by steep slopes. Toward the center and the south, the morphology is quite gentle slopes. The Bilate River and its tributaries drain to Lake Abaya. Its maximum discharge is provided by most of the perennial streams from the western high lands of the catchment and the eastern intermittent streams provide fewer amounts.

Data source and methodology
A 30 mÂ30 m resolution DEM of Bilate catchment was downloaded from USGS at https://earthexplorer.usgs.gov site and used to delineate the watershed using SWAT and to analyze the drainage patterns of the land surface terrains. The data was projected to UTM zone 37 north on the spheroid of D-WGS 1984 datum (Figure 2(a)). The inclusive physico-chemical properties of soil are required by the SWAT model. A 250 m resolution of soil grids was assessed from the world digital soil map of the Food and Agricultural Organization (FAO 2002). The extraction was performed using ArcGIS 10.3 software. The classification used is based on the FAO classification system and customized in the way the SWAT model requires (Figure 2(b)). The dominant soils of this catchment include Vitric Andosols, Pellic Vertisols, Chromic Vertisols, Orthic land-use changes. The model recognizes spatial land-use distributions and transitions (Gidey et al. 2017). Temporally, the model forecasts change probability from one particular class to another class (Al-sharif & Pradhan 2014). LULC maps of 1986, 2002 and 2018 were used for land-use analysis in IDRISI 17.02 software. The procedure involves the following steps: transition probability matrix, transition area matrix and transition suitability maps. LULC maps of 2002 and 2018 were used in this procedure for calibration. The transition probability matrix contains the changing probability of an individual land-cover class to other classes, while the transitional area matrix contains the pixel number that is expected to change from each land-use class over the specified time frame (Eastman 2016). Suitability maps were developed for each LULC class based on potential driving factors with multi-criteria evaluation. Factors such as DEM (aspect, altitude and slope), urban centers, distance from streams and roads were utilized to generate suitability maps. Finally, the LULC maps for 2038 and 2058 were projected by using basic LULC maps, transition areas and transition suitability images.
Validation is the measure of the reliability of a model in predicting the future LULC by using datasets. LULC of 1986 and 2002 maps were used to simulate the 2018 LULC map. Markov transitions areas and transition suitability images were developed using 1986 and 2002 LULC maps. The simulated 2018 land-use map was compared with the reference 2018 land-use map using Κ indexes (Pan et al. 2017). The quantification and location errors are resolved by cause-dependent K-indexes such as Κ as overall accuracy of simulation (Kno), Κ for location (Klocation), Κ for standard (Kstandard) and Κ for stratum-level location (Klocationstrata) (Kim et al. 2011;Pontius & Millones 2011;Mosammam et al. 2016). The combination of K-indexes is considered for a comprehensive evaluation of the overall accuracy in the validation of the model. Kno measures the overall fraction of correctly classified versus expected values. Klocation evaluates the measure of spatial accuracy due to the correct Journal of Water and Climate Change Vol 12 No 8,3754 assignment of values between the simulated and reference maps. Kstandard estimates the proportion assigned correctly versus a proportion that is correct by chance. Klocationstrata quantifies spatial accuracy within predefined strata, and how well the grid cells are situated within the strata. The precision and consistency values of the κ coefficient below 0.4 indicates less, 0.4 κ 0.75 is moderate and κ . 0.75 is a high level of consistency between the simulated and reference land-use maps (Viera & Garret 2005).

Climate change model
Regional climate model data extraction and bias correction Regional climate model simulations for multi-period employed downscaling under the lateral boundary conditions are from a completely coupled GCM and components of the climate system (Castro et al. 2005). The downscaled regional climate model data were obtained from https://esg-dn1.nsc.lpu.se/projects/esgf.liu/cordex. The CORDEX-Africa model data at longitude 0.44°and latitude 0.44°horizontal resolution and a multi-model ensemble of regional climate models with their driving GCMs providing the boundary conditions were used in the study (Nikulin et al. 2012). Owing to the diversity of climate regimes and to improve simulations and projections, the coordinated regional climate downscaling experiment CORDEX-Africa was set (Giorgi et al. 2009). The model selection was built on the choice of likely variations in climatic means, changes in climatic limits and the model's ability to reproduce the past climate (Lutz et al. 2016). These five regional climate models with their driving GCMs employed CCLM4-ICHEC, RCA4-CanESM2, REMO2009-MPI, HIRHAM5-ICHEC and RACMO22T-ICHEC (Mascaro et al. 2015). The selected regional climate models were further evaluated using statistical parameters to check their association with observed data from the weather stations. Due to doubts in the model forecasts, it is necessary to assess the model's performance in simulating the climate in comparison to observations (Flato et al. 2013;Tinghai et al. 2013). The climate model data were evaluated against the observed rainfall and temperature data through the root-mean-square error (RMSE) and the Pearson correlation coefficient (r) (Chai & Draxler 2014). The association has been performed with the historical model data from 1981-2010 with that of observed data within the identical period. Climate parameters such as rainfall and temperature were evaluated for two weather stations. The r and RMSE between models and observed data of maximum temperature are in between 0.64 and 0.86, and 2.1 and 9.2, respectively. Similarly, the r and RMSE of minimum temperature are in between 0.45 and 0.63, and 2.2 and 5.10, respectively. Besides, the r and RMSE of rainfall are in between 0.55 and 0.81, and 4.7 and 11.5, respectively, of the models and observed data. The downloaded regional climate model data were in the NetCDF file format. The regional climate model data of rainfall, minimum and maximum temperatures were extracted by ArcGIS 10.3 software through a multidimensional tool and NetCDF table view. The stations' data were extracted through their latitudes and longitudes. The bias is the long-term average difference between the observed and model data. Bias correction was employed on the regional climate model's rainfall and temperature in Bilate catchment.
Climate model data downloaded from CORDEX-Africa through RCP4.5 and RCP8.5 emission scenarios were used to project the future rainfall and temperature over Bilate catchment. An individual climate model is liable to uncertainties that may arise from climate models. To resolve this problem, the ensemble average of five climate models was used in the analysis of rainfall and temperature.

Precipitation correction
The power transformation algorithm has been used to correct the spatial distributional biases in rainfall outputs in regional climate models. The power transformation corrects both the mean and coefficient of variation to normalize peak daily and monthly rainfall amounts at rain gauge stations. This method was employed in Hare watershed, Southern Rift Valley of Ethiopia (Biniyam & Kemal 2017). Similarly, the power transformation was employed for the basin of the river Meuse upstream of Borgharen in the Netherlands (Leander & Buishan 2007); in the Rhine Basin, Western Europe (Terink et al. 2009). This method matches the coefficient of variations of the stations' and regional climate models' rainfall data. The daily rainfall quantity p converted to a modified rainfall amount P* is given by the following equation (Leander & Buishan 2007;Leander et al. 2008): where a is the transformation coefficient and b is a scaling exponent. In this study, coefficients of variations (CV) of the observed rainfall data were matched with the climate model (RCM) rainfall data with some numerical iteration processes.

Temperature correction
The mean and standard deviation adjustment method was employed to correct the temperature. The mean and standard deviation adjustment is calculated according to the following equation (Leander & Buishan 2007): where T c is the bias-corrected future temperature, T̅ obs is the mean observed temperature in the baseline period, T̅ model is the mean RCM temperature in the baseline period, T model is the RCM temperature in the future period, Std obs is the standard deviation of observed temperature in the baseline period and Std model is the standard deviation of climate model temperature in the baseline period.

SWAT model
The physically-based, semi-distributed SWAT model was developed to simulate the effect of land-use management practices on hydrologic processes taking place in basins (Arnold et al. 1998). The water equilibrium approach of the SWAT model simulates surface runoff, infiltration, percolation, channel routing, shallow and deep aquifer flow. SWAT uses the water balance approach to simulate watershed hydrologic partitioning (Neitsch et al. 2011). The hydrologic routines replicated by the SWAT model are established on the water equilibrium equation: where SW t is the last soil water content (mm), SW o is the primary soil water content on the day i (mm), t is the time (days), P day is the quantity of precipitation on the day i (mm), Q surf is the quantity of surface runoff on the day i (mm), E a is the quantity of evapotranspiration on the day i (mm), W seep is the quantity of water entering the vadose zone from the soil profile on the day i (mm) and Q gw is the quantity of return flow or base flow on the day i (mm). The SWAT run was performed on a daily climatic data basis. The data consist of rainfall, solar radiation, relative humidity, wind speed, minimum and maximum temperatures from two stations located in Bilate catchment.

Parameter sensitivity analysis, model calibration and validation
Model parameters exert the highest influence on model calibration and predictions. The models respond in a different way to different parameters, and as a result, the outputs become different. Model sensitivity is scrutiny of the change in model output per change in a parameter of input. Sensitivity analysis is one version of the SWAT model tool to show the mean relative sensitivity of the best parameters through SWAT-CUP (Feyereisen et al. 2007). Model calibration is the adjustment of model parameters, within the recommended ranges, to optimize the simulated output so that it matches with the observed data. Calibration allows the provision of different parameters for the adjustment during operation. Validation is the process of testing the calibrated parameters with an independent set of observed data without further changes to parameters. Flow data collected at the gauging station near Halaba was used for Bilate River flow simulation for the period 2006-2017. The river flows from 2006-2017 were split into warm-up (2006-2008), calibration (2009-2014) and validation (2015-2017) periods. The Sequential Uncertainty Fitting version two (SUFI-2) algorithm set in SWAT-CUP 2012 was used during model calibration and validation (Abbaspour et al. 2007).

Model performance evaluation
The fitness of SWAT model simulation with the river flow was conveyed by statistics like the coefficient of determination (R 2 ) and the Nash-Sutcliffe simulation efficiency (NSE) (Nash & Sutcliffe 1970). Their values range from 0 to 1; the closer the value to 1, the higher the agreement between the simulated and the measured flows. R 2 describes the proportion of the total variance in the observed and simulated data that can be explained by the model and given by the following equation: where, Qob is the observed/measured value; Qob av is the average measured value. Qsi is the predicted value, and Qsi av is the average predicted value. NSE shows the degree of fitness of observed and simulated data and given by the following equation: where Qob is the measured value, and Qob av is the average measured value. Qsi is the predicted value. The measured, simulated and average values were measured in m 3 /s. NSE shows how well the plot of observed versus simulated value fits the 1:1 line. The NSE and R 2 values are greater than 0.5 and 0.6, respectively, the simulated value is a better predictor than the mean measured value and is viewed as acceptable performance (Santhi et al. 2001). Parameter uncertainty is caused by parameter non-uniqueness and input uncertainty as a result of errors in input data such as rainfall, land-use, soil-type and observed data. The model simulation uncertainty was quantified by 95% prediction uncertainty (95PPU) (Abbaspour et al. 2007). The quality of calibration and prediction uncertainty is judged based on the closeness of p-factor to 100% (all observations bracketed by the prediction uncertainty) and r-factor to 1 (achievement of a rather small uncertainty band) (Talebizadeh et al. 2009).

Sensitivity analysis
Model calibration requires the identification of key parameters and parameter precision. The sensitivity analysis of Bilate catchment by the SWAT model utilized 12 parameters. The global sensitivity analysis of a multiple regression system was employed against Nash-Sutcliffe objective function. Twelve parameters are found to be sensitive with relative sensitivity values, for example, runoff curve number to moisture condition II (CN2), groundwater 'revap' coefficient (GW_REVAP), effective hydraulic conductivity in the main channel (CH_K2), groundwater delay (GW_DELAY), deep aquifer percolation fraction (RCHRG_DP), baseflow recession constant (ALPHA_BF), threshold depth of water in the shallow aquifer for 'revap' to occur (REVAPMN), soil evaporation compensation factor (ESCO), threshold depth of water in the shallow aquifer required for return flow to occur (GWQMN), plant uptake compensation factor (EPCO), surface runoff lag coefficient (SURLAG), an available water capacity of the soil layer (SOL_AWC) and effective hydraulic conductivity in main channel alluvium (SOL_K). These sensitive parameters are considered the high effect on Bilate River flow during calibration and validation in SWAT-CUP (Table 1). The assessment of sensitive parameters was measured using the t-stat and p-values. The sensitive parameters were selected on the basis where the t-stat values are more sensitive for larger absolute t-stat values and p-values are closer to 0.

Flow calibration and validation
The calibration and validation were performed by using Bilate River flow at Halaba gauging station from 2009 to 2014 and 2015 to 2017, respectively. The periods were used for river flow calibration and validation by employing the 2018 LULC. The river flow was calibrated by auto-calibration and manual procedures. Model simulations during calibration and validation periods were evaluated on the monthly basis of the NSE and R 2 . Based on procedures suggested by Moriasi et al. (2007), calibrated and validated river flows were agreeable at the monthly time scale. Peak flow values exhibited by observed and simulated flows during model calibration and validation. River flow calibration and validation were performed on a monthly basis. Six years for calibration and 3 years for validation were used. Flow hydrographs were derived to evaluate observed and simulated flow values for calibration and validation for 2018 LULC (Figure 4).

Hydrological model performances
The mean monthly simulated flows seem to have replicated the observed flows as shown in the calibration and validation results (Figure 4). The calibration result indicates that the SWAT model can capture the observed flow with R 2 and NSE of 0.79 and 0.71, respectively. Validation indicates R 2 and NSE of 0.81 and 0.73, respectively. Besides, p-factor and r-factor showed good agreement with 0.82 and 0.75 during calibration and 0.83 and 0.73 during validation, respectively. Both the calibration and validation indicate that the SWAT model achieved a relatively good fit between observations and simulations.

Simulation of 2018 LULC
The LULC maps of 1986 and 2002 were used to simulate the LULC map of 2018 using the CA-Markov model. The 2018 map was compared with the simulated map 2018 to appraise the performance of the CA-Markov model using κ indexes. The model validation results on cause-dependent Κ indexes with Κ for overall accuracy (Kno), Κ for location (Klocation), Κ

LULC projection
The LULC is likely to change in the 2038 and 2058 periods compared to 2018 (Table 2; Figure 5). Compared with the LULC of 2018, water bodies, built-up areas and forest land are continuously increasing by 0.22-0.45%, 1.30-2.17% and 5.34-5.94%, respectively, while cultivation land is decreasing by 8.60-9.84%. Grazing and barren lands will increase from 2018 to 2038 by 0.12 and 1.53%, while the decrease from 2038 to 2058 is by 0.08 and 0.24%, respectively. The highest increment of forest/ woodland is mainly related to Eucalyptus plantation at the expense of cultivation and grazing lands, and the practice of homestead agroforestry in the study catchment. This result was consistent with a recent study in Majang Forest Biosphere Reserves of Gambella, Southwestern Ethiopia, in which the CA-Markov model was employed (Tadese et al. 2021).

Climate change projections
Climate change projection was performed by using the ensemble mean of the models for the present ( (2018) and Kebede et al. (2013) on climate change country profiles shows that the mean annual temperature is projected to increase by 1.1-3.1°C by the 2060s and 1.5-5.1°C by the 2090s. Similarly, the study by Kassie et al. (2013) showed that the annual mean temperature is expected to increase in the range of 1.4-4.1°C by the 2080s in the Central Rift Valley of Ethiopia.

Rainfall projection
The projected change in rainfall during the present (2021-2050) and mid (2051-2080) centuries relative to baseline  under RCP4.5 and RCP8.5 scenarios of Bilate catchment was performed by using a multi-model ensemble of regional climate models. It seems that in the catchment, the rainfall is likely to feature increased and decreased under RCP4.5 and RCP8.5 scenarios (Figure 7). This is true especially over the catchment where rainfall is expected to decrease in the ranges of 134.4-659.4 and 38.7-822.5 mm under the RCP4.5 scenario, respectively, in the current (2021-2050) and mid  centuries. Similarly, rainfall is expected to decrease in the ranges of 74.9-624.2 and 52.0-436.3 mm under the RCP8.5 scenario, respectively, in the current (2021-2050) and mid (2051-2080) centuries. Rainfall is expected to decrease in January, March, May, June, July and August under RCP4.5 in two centuries. Besides, decreasing rainfall is projected in November under RCP4.5. Rainfall is expected to decrease in February, March, July, August and November under RCP8.5 in two centuries. In the mid-century, rainfall is projected to decrease in January, May and September under RCP8.5. Projected decreases in rainfall are more pronounced in the two centuries. For example, under RCP4.5 projected increase in rainfall is featured in September, October and December. Similarly, under RCP8.5 increased rainfall is featured in April, June and October. The projected decrease in rainfall during the short rainy season (February-May) and the long rainy season (June-September) affects the catchment, because of subsistence rain-fed agriculture. Ethiopia's heavy dependence on rain-fed and subsistence agriculture increases its vulnerability to the adverse effects of climate changes (Kassa 2015;Zerga & Gebeyehu 2018). Water shortage is more pronounced in the forthcoming together with increasing temperature in the catchment. Studies in Ethiopia, climate trend analysis from 2010 to 2039 revealed rainfall decline ranges from 50 to 150 mm and is related to lower harvest and poor pastoral rangelands across the southern and eastern parts (Funk et al. 2012). Similarly, Kassie et al. (2013) reported that annual rainfall will change in the range of þ10 to À40% by 2080, in which it will increase outside the growing season and decline during the growing seasons. Studies in Ethiopia reported unstable rainfall patterns and magnitudes in the future. For example, the future rainfall (2040-2059) of the long rainy season (June-September) tends to decrease over the southern part of the country (Li et al. 2015).

Hydrologic responses to 2018 LULC
The hydrologic components were simulated by the SWAT using climatic data from 2006 to 2017 and the 2018 LULC map. The SWAT model performed well in both the calibration and validation periods, accurately simulating the outlet flows at Halaba gauge station agreeing with model performance criteria (R 2 and NSE) after 12 parameters were optimized. During the calibration (2009)(2010)(2011)(2012)(2013)(2014) and validation (2015-2017) periods, the time series plots of observed and simulated were matched in the periods (Figure 4). The model has been calibrated by river discharge; the hydrologic components such as potential evapotranspiration (PET), actual evapotranspiration (ET), surface runoff (SUR_Q), groundwater (GW_Q) and water yield (WYLD) were quantified. Notably, the simulated PET, ET, surface runoff, groundwater and water yield were 3,751.9, 1,537.98, 671.95, 637.16 and 1,401.34 mm/year, respectively, in 2018. The R 2 and NSE values in this study were generally good in calibration and validation. The result suggested that the SWAT model performed well and can replicate the catchment.

Future hydrologic responses to land-use change
To assess the impacts of LULC alteration on water balance, the calibrated model was used to simulate two LULC scenarios in 2038 and 3058 keeping the weather did not change. The water balance comparison was made between the 2018 and simulations under the future LULC scenarios ( Table 3). The increase in surface runoff is 3.06, 3.78% for land use of 2038 and 2058, respectively. This increase in runoff can be attributed to a trend in cultivation land transformation to built-up areas and barren land increment. Groundwater and water yield are decreased by 1.85, 2.84%, and 1.83, 2.83%, respectively. The increase in PET is 6.34, 7.49% for land use in 2038 and 2058, respectively.
Many studies from different parts of the world confirmed that LULC changes significantly impact hydrologic processes. For example, Lu et al. (2015) found that in the land-use transformation, increased forest land area contributed more than any other land-use class to water yield, but built-up areas had the most impact due to small initial loss and infiltration. The finding agrees with Berihun et al. (2019) who reported that in Kecha and Kasiry, Upper Blue Nile Basin, the influence of land use has increased runoff by 4 and 28.7%, respectively.

Future hydrologic responses to climate changes
Changes in rainfall and temperature in the current (2021-2050) and mid (2051-2080) centuries were used to forecast the effects of climate change on hydrologic components of Bilate catchment. The calibrated and validated SWAT model was used to simulate future hydrologic components under RCP4.5 and RCP8.5 scenarios. The simulated hydrologic responses of the future periods and 2018 were compared to show their variations. SWAT simulations in two future periods showed a decrease in surface runoff, groundwater and water yield, while there was an increase in PET and ET (Table 4). The projected change in climate indicated a decline of rainfall and an increase of temperature in the catchment. Variations in the rainfall and temperature were associated to the variations in hydrologic response in Bilate catchment. Water yield, groundwater and surface runoff are likely to decrease under RCP4.5 and RCP8.5 in two future periods. These variations follow the path of rainfall changes (Table 4). An alteration in rainfall could result in variations in hydrologic responses. This is associated to the decrease in rainfall. For example, a 15.39% reduction in the annual rainfall caused about 9.22, 11.11 and 10.25% reductions in water yield, groundwater and surface runoff, respectively, under RCP4.5 in the mid (2051-2080) century. This result was similar to a recent study in Lake Tana Basin (Tigabu et al. 2020). PET and ET are expected to increase under RCP4.5 and RCP8.5 scenarios in two future periods. These changes follow the direction of temperature changes (Table 4). A 2.0°C increase in the mean annual temperature resulted in about 19 and 13.9% increase in PET andET, respectively, under RCP8.5 in the mid (2051-2080) century. This result agrees with a research report in Fincha watershed, Western Ethiopia (Dibaba et al. 2020).
The sensitivity of hydrologic components to climate change is different in future periods, and the impacts can be explained in the combination of the components. For example, the decline in rainfall reduced surface runoff, groundwater and water yield, while increasing temperature augmented an increase in PET and ET. The relationship between climate change and hydrologic components, ascertained by rainfall and temperature changes, varies with future periods under RCP4.5 and RCP8.5 scenarios. For example, Mengistu et al. (2021) found that climate change in the Upper Blue Nile Basin in the 21st century showed an increase in PET by up to 27% under RCP8.5 and a decrease in water yield of 1.7 to 6.5% and 10.7 to 22.7% under RCP4.5 and RCP8.5, respectively. Berihun et al. (2019), in Kecha and Kasiry, Upper Blue Nile Basin, Ethiopia reported that the effect of climate change has increased evapotranspiration by 33.6 and 42.1%, respectively. Besides, a finding by Gebremeskel & Kebede (2018) in Tekeze River Basin reported that the effects of climate change will decrease surface runoff by 13 and 14%. Climate change had a higher influence on water balance relative to that resulting from LULC change. Compared to LULC change, climate change leads to an increase in the average annual PET by 5.85 mm. However, the average annual SUR_Q, GW_Q and WYLD declined by 3.37, 3.33 and 2.57 mm, respectively. This result agrees with Kahsay et al. (2018) which reported a decline in groundwater followed by decreasing annual rainfall in Tekeze Basin, Ethiopia. Therefore, the hydrological processes are impacted more by climate change than the LULC change in Bilate catchment.

Future hydrologic responses to land-use and climate changes
The changes in LULC, temperature and rainfall were used to forecast the effects on water balance components in Bilate catchment. The hydrologic responses of the catchment to LULC and climate variations were investigated with two LULC scenarios. The first LULC scenario is from 2018-2038 and the second from 2018-2058 with the LULC of 2018. Two climate change periods are from 2021-2050 and 2051-2080 proposed under the RCP4.5 and RCP8.5 scenarios. The hydrologic responses of the Bilate catchment as considered to the annual PET, ET, SUR_Q, GW_Q and WYLD (Table 5).
The increasing temperature and decreasing rainfall under changing LULC in Bilate catchment resulted in a decline of runoff, groundwater and water yield, while there was an increase in PET and ET except under RCP4.5 of 2058 LULC. When LULC and climate variation scenarios are combined, the effects of water balance components are more noticeable. The highest annual surface runoff decline of 14.4% is projected under RCP4.5 in the mid-century, the highest groundwater decline of 16.8% is projected under RCP8.5 in the current century and the highest water yield decline of 14.1% is projected under RCP8.5 in the current century.
Studies conducted at different catchments agree with these results. The study by Yan et al. (2019) reported that the combined effect of LULC and climate has increased surface runoff, water yield and evapotranspiration. Similarly, Berihun et al. (2019) reported that the effect of climate and land-use variation had significantly increased surface runoff and evapotranspiration. Gebresilassie et al. (2020) found that the impact of land-use and climate variation has increased surface runoff and evapotranspiration under RCP8.5 from 2045 to 2055. In this study, the hydrologic components caused by combined effects of climate and land-use variations were higher than individual variation. This was further supported by Marhaento et al. (2018) that climate and land-use variations individually will cause water balance, but that more marked changes are likely if the factors are joined. The spatial distribution of the hydrologic processes under projected climate and LULC changes of Bilate catchment is presented in Figures 8 and 9. The SWAT model showed that surface runoff, groundwater and water yields are high, while PET declined in the upstream areas in Bilate catchment. On the downstream areas, a significant amount of surface runoff, groundwater and water yield reduction was observed due to climate and land-use changes, although PET is significantly increasing. Moving downstream the water potential is reduced due to high temperature increase coupled with decreasing rainfall that could lead to reduced soil moisture, which is required for plant growth. Besides, semi-aridity is another problem for the downstream areas in Bilate catchment. This could imply the reduced availability of water for crop production, which will become a continuing problem for the farmers in the catchment whose livelihood is based on agriculture.

CONCLUSION
This study investigated the effects of climate and LULC variations on the hydrologic components in Bilate catchment, Rift Valley Basin of Ethiopia. Two climate and LULC alteration scenarios were established, and the hydrologic components were simulated using the SWAT model. LULC scenario prediction maps of 2038 and 2058 were generated using the CA-Markov model. Climate changes for the current (2021-2050) and mid (2051-2080) centuries were projected using the ensemble mean of regional climate models under RCP4.5 and RCP8.5 scenarios. The model was calibrated, validated for 2018 and used to simulate the forthcoming water balance under two scenarios. The agreement between the measured and replicated river flows indicated that the SWAT model could be used to simulate the hydrologic responses to climate and LULC variations in the catchment. Besides, the forthcoming water balances were associated with those of 2018. The historical and forthcoming hydrologic responses were simulated under the monthly scales. Runoff, water yield and groundwater decreased in the current (2021-2050) and mid (2051-2080) centuries under the RCP4.5 and RCP8.5, but water yield increased in the current century under the RCP4.5 scenario. PET and ET increased in two periods, but decreased in the mid-century under the RCP4.5 scenario. The hydrologic responses follow the path of rainfall. This is related to the decline in rainfall amount in Bilate catchment. The decline in the rainfall resulted in a significant decline in surface runoff, water yield and groundwater. Similarly, the hydrologic responses follow the path of temperature in the catchment. This is attributed to the increasing temperature in Bilate catchment. The likely increase in the mean temperature resulted in a substantial increase in ET and PET. The combined impacts of climate and LULC variations on hydrologic responses in the forthcoming are greater than the variation trends of climate or LULC change alone. Besides, the effect of climate change on water balance is higher than that of LULC change.
Furthermore, the decreasing rainfall and increasing temperature in the catchment may cause hot and dry years. Accordingly, the catchment is highly vulnerable to climate change and its impacts could range from warming to crop failures. The change in water balance in Bilate catchment is likely to be more challenging in the future.
The study suggests that the assessment of the combined impact of climate and LULC changes on hydrologic responses in Bilate catchment is well represented by the SWAT simulations. However, the feature of hydro-climatic data and the lack of local LULC evidence in the region need urgent attention to improve our understanding of the variations in the future.
The results provided evidence on how the hydrological processes in Bilate catchment respond to the changes in climate and LULC. This could help to plan water resources and land management interventions. If the degraded sloppy lands are rehabilitated, tree plantations are based on decisions as to how soil and water conservation works properly implemented the groundwater recharge increase. Accordingly, the surface runoff which washes the topsoil and nutrients into the water bodies is reduced. Furthermore, Lake Abaya in the catchment should be buffered with proper management strategies.