Modelling the Impacts of the Changing Climate on Streamflow in Didesa Catchment, Abay Basin, Ethiopia

. Te Didesa catchment, which is the second largest subbasin of the Abay basin, signifcantly contributes to the Blue Nile’s outfow. Understanding the dynamics of water availability under the changing climate in such a basin assists in the proper planning of land use and other development activities. Tis study presents changes in climatic elements such as rainfall, temperature, and evapotranspiration using observation data and regional climate models (RCMs) under two representative concentration pathways (RCPs) for three future periods. We use a calibrated hydrological model to further assess climate change’s efects on streamfow. We select three RCMs and their ensemble’s mean by evaluating their performance with respect to observations. We apply the modifed Mann–Kendall test to detect trends in each dataset. Te result shows that annual mean maximum and minimum temperatures increase in the catchment for the 2021–2040, 2041–2070, and 2071–2100 periods as compared to baseline (1989–2018) under both RCP4.5 and RCP8.5 scenarios. Annual mean maximum temperature and potential evapotranspiration experienced a signifcant decreasing trend during the year from 1989 to 2018. Furthermore, there was an increasing trend in annual rainfall from 1989 to 2018, which could be related to the cooling of sea surface temperature over the equatorial Pacifc. We detect an increasing trend in temperature in both scenarios and all periods; however, no clear trend pattern is found in rainfall. Te result from hydrological model simulations reveals that the mean monthly streamfow slightly increases in the winter season while it decreases during the main rainy season. Further study of detailed weather systems, which afect the subbasin’s climate, is recommended.


Introduction
About half of the global population is experiencing severe water scarcity due to climatic and nonclimatic factors for some part of the year. Te hydrological cycle is intensifed due to anthropogenic climate change, which is afecting the physical aspects of water security. Tis has exacerbated water-related vulnerabilities caused by other socioeconomic factors [1]. Te recent drying trends since the 1980s match the warming observed across the continent of Africa [2], including Ethiopia [3].
Te change in climate is a major challenge that afects the hydrological cycle. Besides, the Intergovernmental Panel on Climate Change (IPCC) assessment report shows that the global average temperature would rise with an increasing total carbon dioxide concentration. Furthermore, climate change induces changes in tributary fow characteristics and changes in rainfall patterns, afecting the interception process and changing the evapotranspiration process [4,5].
Developing countries are likely among the most vulnerable to the impacts of the changing climate due to the lack of economic development and institutional capacity [6]. Te impacts of climate change have the potential to weaken and even reverse the progress made in improving the socioeconomic well-being of African countries [7].
Ethiopia and other developing nations will be more susceptible to the efects of climate change. A large part of the country is arid and semiarid, making it highly prone to drought and desertifcation. Terefore, the country should be concerned about climate change and its efects. According to Wara et al. [8], the change in climate is projected to cause more frequent and intense ENSO events, leading to widespread drought in the area. Hence, assessing vulnerability to climate change impact and preparing adaptation options as a part of the national program are very crucial [9].
Despite the observed drying trends and frequent droughts in recent years in Ethiopia (e.g., [3,10], and [11]), only a few studies assessed the impact on water availability in watersheds which is caused by climate change. In addition, there has been some inconsistency in the fndings of these investigations. For example, Dile et al. [12] and Adem et al. [13] projected an increase in streamfow at the end of the 21st century in the Gilgel Abay catchment, Abay basin. On the contrary, Setegn et al. [14] projected a decrease in the streamfow of the Lake Tana catchment for the same period. Te Didesa catchment is a subbasin of the Abay basin having the second largest area and contributing about 10.9% to the outfow of the Blue Nile River. Terefore, a detailed analysis of the impact of climate change using recent climate models and appropriate hydrological models is essential.
Many earlier researchers used diferent climate models to understand climate change's efects on surface water availability both for the baseline and future periods [15]. Te diference in results of the climate change's efects on streamfow for the same basin is due to the choice of the general circulation model (GCM) and downscaling technique, and the selection of the hydrological model [16]. Terefore, the selection of recently developed climate models with appropriate downscaling techniques is vital.
Te impact of climate change can vary in diferent catchments with relatively small areas due to local climate and catchment characteristics. Altered frequency and distribution of rainfall events, with associated consequences, have efects on discharge rates and streamfow characteristics of the river basin [17]. Climate change impact includes changes in the magnitude of runof, changes in the frequency of foods and droughts, rainfall patterns, extreme weather events, and the amount of surface water availability [18].
Tis study aims to assess the change in climatic elements such as rainfall, temperature, and potential evapotranspiration and evaluate their impacts on streamfow in current and projected future periods in the Didesa catchment. Terefore, we identify changes and trends from observations and modelling. Tis study strongly assists in planning land use and other development activities in a way that fts the dynamism of water availability under the changing climate.

Study Area.
Te study is carried out in the Didesa catchment, which is part of the Abay basin and lies between latitudes of 7.7°N and 10°N and longitudes of 35.5°E and 37.3°E. Its catchment area is 17,645.5 km 2 , with an average length of 1368.3 km and a width of 1453.1 km. Te elevation varies between 852 m and 3041 m above mean sea level (amsl), with both fat and steep slopes (Figure 1). Te catchment is characterized by warm, humid tropics, with a long rainy season (locally known as Kiremt) lasting from June to September with an average annual rainfall between 1450 mm and 2050 mm.
Te dry season, which is locally called Bega, extends from November to February. Te annual maximum and minimum temperatures in the subbasin vary between 21°C to 30°C and 10°C to 20°C, respectively, during the period of 1989-2018. Annual mean rainfall peaks in the months of June to September, which is the main rainy season for the area (Figure 2).
From a hydrogeological point of view, sedimentary rocks are exposed to the land surface due to regional tectonic activity in the Didesa catchment. Alluvial soils and eluvial soils are developed from granitoid, basalt, Mesozoic sandstone, and Paleozoic sediments in the catchment. Te yields of the wells in these aquifers range from 0.5 to about 10 l/s [19].

Methods.
We use the physically based soil and water assessment tool (SWAT) model [20] to determine climate change's efects on streamfow due to its simple and powerful tools for modelling [21]. We employ measured fow from 2000 to 2013 years for model calibration and validation. Te soil, land use/land cover, DEM, and climate data are inputs to the model for simulating catchment surface runof.
During the data preparation phase, we convert the point rainfall to its corresponding area estimate using the Tiessen polygon method [22]: where R i is the rainfall measure at a station i, A i is the area of subcatchment covered by a station i, and A is the total area of the catchment. We compute the potential evapotranspiration using the Hargreaves method as in Droogers and Allen [23] and Byakatonda et al. [24]: where T mean is the daily mean temperature in°C, T max is the daily maximum temperature in°C, T min is the daily minimum temperature in°C, and R a is the extraterrestrial radiation (in MJ·m − 2 ·day − 1 ). Te mean extraterrestrial radiation (R a ) is estimated from the latitude of the station and the month of the year.   Outputs of regional climate models cannot be directly used for impact assessment as the computed variables may difer systematically from the observed ones [25,26]. Terefore, we apply bias corrections to compensate for any tendency to overestimate or underestimate the downscaled variables. We use power transformation for rainfall bias correction. It is a nonlinear method, which corrects both the mean and variance of rainfall as explained by Terink et al. [27]. We apply the correction method by comparing the daily observed rainfall at each station with the outputs of RCMs. On the other hand, we apply the variance scaling method to correct biases in temperature [28]. Te temperatures are bias corrected by the following equation: where T corr is the bias-corrected temperature; T rcm is the raw temperature from the RCM model; σT obs and σT rcm are the standard deviations of the observed and the RCM model output temperature while T obs and T rcm are the mean temperatures from the observation and the RCM model, respectively. We evaluate the RCMs using statistical measures such as bias, root mean squared error (RMSE), and coefcient of variation (CV) and their performance to reproduce the annual cycle of rainfall. Te equations of these statistical measures are as follows: where M is the model output, O is observation, and N is number of observation. A value of 1 is the perfect score. A bias value above/below 1 indicates an aggregate model overestimation/ underestimation.
where σ is the standard deviation and x is the mean of the data under consideration. We employ the modifed nonparametric MK trend test to assess trends in diferent climatic elements. Studies widely use the MK trend test as it does not require the data to be normally distributed [24]. It requires the time series data to be serially independent [29]. Tus, an autocorrelation test is applied to determine the presence of serial correlations in the time series data. To remove autocorrelation from time series data, trend-free prewhitening (TFPW), the most used technique, was used [29,30]. Te computation of MK statistics uses S statistics. Te S statistics is given by the following equation: where X j and X i are the time series observations in chronological order, and n is the length of the time series, while sgn is given by the following equation: Positive values of the S statistic indicate an increasing trend while negative values indicate a decreasing trend [24]. Te variance (V) of S is calculated as follows: where n is the length of time series, t p is the number of ties for p th value, and q is the number of tied values [29]. Te standardized test statistics Z is then computed using S and the variance V(S) as given by equation 19: Positive Z values designate an increasing trend in the time series whereas negative Z values indicate a negative trend. Given a confdence level, α a statistically signifcant trend is said to be existing in a time series data if |Z| > Z 1− (α/2) . Te level of signifcance employed in this study is α � 0.05, and from the standard normal table, the value of Z 1− (α/2) at the level of signifcance α � 0.05 is 1.96.
Tis study classifes the catchment into multiple subwatersheds, which are further subdivided into a hydrologic response unit (HRU) with unique characteristics of land use management, topography, and soil ( Figure 3). SWAT simulates hydrological parameters at each HRU using the water balance equation: where SW o and SW t denote the initial and fnal volumes of water in the soil (mm); R day , Q surf , E a , W seep , and Q gw are the rainfall amounts, surface runof amount, evapotranspiration amount, infltration amount, and return fow amount on day i (mm) in respective order; and t is the time in days. By using the Soil Conservation Service-Curve Number (SCS CN) method [31], SWAT can be used to evaluate the relative impact of climate change at the catchment level [32,33].
Te subbasin runof is routed to obtain the total runof for the entire basin. Surface runof is computed by using SCS CN as in the following equation: where Q surf is the daily surface runof (mm), R da y is the rainfall depth for the day (mm), and S is the retention parameter (mm). Te retention parameter (S) is calculated as in the following equation: where S is drainable volume of soil water per unit area of saturated thickness (mm/day), and CN is curve number. Water yield is the aggregate sum of water leaving the HRU and entering the main channel during the time step [34]. Te total water yield is computed as follows: where W yld is the measure of water yield (mm), Q surf is the surface runof (mm), Q lat is the lateral fow contribution to streamfow (mm), Q gw is the groundwater contribution to streamfow (mm), and T loss is the transmission losses (mm) from tributary in the HRU by means of transmission through the bed. After having all the necessary spatiotemporal data required by the SWAT, we quantify model sensitivity to parameter changes, as it is a vital step before model calibration. Hydrological parameters are selected for sensitivity analysis for the simulation of streamfow with default lower and upper bound parameter values. In addition to hydrological parameters, the observed monthly fow values are used.
Te SWAT model is calibrated and validated with the streamfow observed at the gauging station during the baseline period. We use two-thirds of the data for calibration (2000-2009) and one-third for validation (2010-2013). During the calibration process, the model parameters are subjected to adjustments to obtain model results that correspond better to the measured data sets. Calibration and validation stages are done under SWAT-CUP software support. We evaluate the model performance using the coefcient of determination (R 2 ), the Nash-Sutclife efciency (NSE), and the percent bias (PBIAS), which are time series-based metrics during the calibration and validation periods.

Advances in Meteorology
where Q obs and Q sim represent observation and simulated discharge, respectively,

Evaluation of Climate Model Performance.
We present the evaluation results of the climate models' outputs with respect to the observed data in Table 1 and Figure 4. All RCMs reasonably predicted the monthly rainfall distribution patterns with gauged rainfall in the catchment with only a slight underestimation of the observation (Figure 4). We also use the models' ability to reproduce the long-term mean annual rainfall (annual RF) of the catchment for the best model selection. Te observed mean annual rainfall has been 1977.7 mm, while it is 1775.1 mm, 1794.9 mm, 1807.4 mm, and 1852.2 mm for the REMO2009, RACMO22T, the ensemble's mean, and CCML-4, respectively (Table 1). Te comparison shows good agreement with a slight underestimation. Te accuracy of the models is not the same in reproducing the rainfall, with CCML-4 performing best (BIAS � − 6.3%) while REMO2009 performs the worst (BIAS � − 10.2%). Te result indicates that the degree of rainfall variability is almost similar in all models; CCML-4 model (CV � 10.8%), RACMO22T model (CV � 11.2%), and REMO2009 model (CV � 13.0%). As there are some differences between model output and observation, the biases should be corrected before applying for impact assessment. Tese climate models were reported to perform well in the wide area of East Africa in a previous study [35].

Trends in Observed Data.
In the Mann-Kendall trend test, when P < α (α � 0.05 in this study), the null hypothesis (H o ) is rejected; this indicates the existence of a trend in the data under consideration. However, when P > α, the null hypothesis is accepted, which shows that the trend is insignifcant. Sen's slope is employed to compute the magnitude of the trend, where positive and negative values indicate increasing and decreasing trends, respectively. Te trends are signifcant when they are associated with P values less than 0.05. Te trend analysis for the annual total rainfall amount indicates an increasing value of 7.3 mm per year from 1989 to 2018 (Figure 5(c)). Te MK test statistics (S) indicates that there is a nonsignifcant increasing trend in rainfall at a signifcance level of 0.05 with a P value of 0.54 (Figure 5(c)). Te increasing trend could be related to the cooling of sea surface temperature over the equatorial Pacifc, which is reported to have a positive impact in increasing rainfall amount over west Ethiopia [36,37]. Diro et al. [36] found that a warm (cold) sea surface temperature anomaly over the equatorial Pacifc is associated with rainfall defcit (excess) in western Ethiopia during the main rainy season. Similarly, Dufera et al. [37] indicated that there is a negative correlation between sea surface temperature over the equatorial Pacifc and drought magnitudes in western Ethiopia during the same season. However, further investigation of the weather systems is required to fully address the cause of the current wetting trend in the subbasin.
Mean annual maximum temperature shows a signifcant decreasing trend with a negative Sen's slope value and a P value of 0.02 at a signifcance level of 0.05 ( Figure 5(a)) over the period from 1989 to 2018. In addition, the minimum temperature shows a nonsignifcant decreasing trend over the same period ( Figure 5(b)). Te statistics for the trend of minimum temperature is − 0.19, 0.30, and − 0.02 for Kendall's tau, P value, and Sen's slope, respectively. Te weather systems afecting the area's climate should be further investigated in future studies. Table 2, the trend in annual rainfall is not clear. Te selected models do not consistently show similar trend patterns. Furthermore, the identifed trends are insignifcant in both scenarios and all periods except RACMO22T under RCP8.5 in the nearterm period. Te RACMO22T shows an increasing signifcant trend under the mentioned scenario and period.

Trends in Projected Climate. As indicated in
Te mean monthly changes in maximum temperature for the future period are shown in Table 3. Te trend in maximum temperature shows an increasing trend in both scenarios and all future periods. Te mean minimum temperature also shows similar trend patterns (Table 4). An increasing rate of evapotranspiration, a decreasing availability of water resources, and an increasing water demand are expected when the temperature is increasing [38].

Future Climate Change Patterns.
Te projected climate can be used as input for process-based hydrologic models to assess the impact of climate change on streamfow. We compute the projected change in temperature, rainfall, and potential evapotranspiration under RCP4.5 and RCP8.5 scenarios in three diferent future times. In the near-term period, the projected change of rainfall from CCLM-4 and REMO2009 indicates a decrease in the amount for the months from February to December under the RCP4.5 scenario ( Figure 6). However, it is not consistent under RCP 8.5.
Te mean change of potential evapotranspiration is mostly positive ranging from 5 to 80 percent in both scenarios and all time periods except some cases during the months of January to May, where the change in potential temperature reaches up to 40 percent (Figure 7). Tis could be related to the increase in temperature.

Sensitivity Parameters Analysis.
In the analysis, we identify the sensitive parameters of the streamfow. We select parameters for sensitivity analysis to simulate streamfow with default lower and upper bound parameter values as in [39]. Te parameters are ranked according to the magnitude of P value and the corresponding t-stat (Table 5). For the global sensitivity analysis, curve number (CN2), base fow alpha factor (ALPHA_BF), groundwater delay time (GW_DELAY), threshold depth of water in the shallow aquifer required for return fow to occur (GWQMN.gw), groundwater re-evaporation coefcient (GW-REVAP.gw), saturated hydraulic conductivity (SOIL_K), soil evaporation compensation factor (ESCO), depth from the soil surface to the bottom of layer (SOIL_Z), deep aquifer percolation fraction (RCHRG_DP), initial depth water in the aquifer (SHALLST_N), manning roughness value for the main channel (CH_N2), channel efective hydraulic conductivity (CH_K2), threshold depth of water in the shallow aquifer for "revap" to occur (REVAPMN.gH), and maximum canopy storage (CANMX) are highly sensitive parameters and ranked from 1 up to 15, respectively. Terefore, curve number (CV), base fow alpha factor (ALPHA_BF), and GW_DELAY are the most sensitive parameters. Te t-stat provides a measure of the sensitivity parameters (large absolute values are more sensitive), whereas the P value indicates the signifcance of the sensitivity. Te P value closer to zero means the parameters have more signifcance.
3.6. Calibration. Te parameter ranges are modifed automatically based on the correlation between the simulated and observed streamfow while ensuring sufcient parameter space as well as fast convergence. After automatically calibrating, the fnal results of the calibration are obtained by multiplying, adding, or subtracting the default values by a necessary factor guided by a manual calibration helper. Te values of the parameters are varied iteratively within the allowable range until the simulated fow matches the observed streamfow. Figure 8 Figure 9.

Climate Change Efects on Water Balance Components.
Te mean monthly rainfall and evapotranspiration in the Didesa catchment during the base period are 156 mm and 113 mm, respectively ( Figure 11). Te surface runof amount, lateral fow, groundwater contribution, and transmission loss are about 60 mm, 10 mm, 45 mm, and 7 mm, respectively, for the catchment resulting in a total water yield of 108 mm.
Under a short-term period, the change in average annual components of water balance including rainfall, groundwater fow, and potential evapotranspiration      Advances in Meteorology signifcantly increases by 8.8%, 11.1%, and 8.6%, respectively. Similarly, the other components such as lateral fow, percolation of water, and transmission loss signifcantly increase by 12.5%, 15.0%, and 34.9%, respectively, while surface runof and total water yield decrease by − 2% and − 4%, respectively ( Figure 12). All water balance components increase during the mid-term except surface runof and total water yield.

Climate Change Efects on Mean Streamfow.
During the short-term period, the monthly streamfow change in the main river is positive for January, February, April, September, October, and November months, while it is negative for the other months. During the mid-term period, the change in monthly streamfow is positive in all months except the months of February, March, and December. Furthermore, the change in streamfow is positive only in the months of January, February, and April during the longterm period ( Figure 13). Te change in streamfow in the Didesa catchment is positive for the annual and winter seasons, while it is negative in other seasons for short-term period as compared to the base period. Te change in streamfow is negative during summer and positive during winter in all time periods. Tis is not a good sign as summer is the main growing season around the catchment (Figure 14). Tere are   irrigation activities that have been practiced in the catchment. Moreover, about 16 and 68 percent of the area are found highly suitable and moderately suitable, respectively, for surface irrigation [41].

Conclusions
Tis study presents the evaluation of the current and future climate change's efects on streamfow in the Didesa catchment for 2021-2040 (near-term), 2041-2070 (midterm), 2071-2100 (long-term) relative to 1989-2018 using the ensemble's mean of four RCMs output under RCP4.5 and RCP8.5 emission scenarios. Te bias correction is performed for future rainfall and temperature before directly using it as input to the hydrological model. Te modifed Mann-Kendall (MK) trend test for rainfall shows an increasing trend for short-term and mid-term in all periods under both scenarios considered except for RCP4.5 during the mid-term period for RACMO22T and REMO2009 models. However, the CCLM-4 rainfall projection contrastingly shows a decreasing trend in RCP4.5 and RCP8.5 except during mid-term under RCP 4.5, where it shows an increasing trend. Te mean change of rainfall is negative for CCLM-4 in the short-term period for the months from February to December, with values ranging from − 1.6% to − 69.5% under the RCP4.5 scenario. REMO2009 model shows a negative rainfall change in all months under RCP 4.5. Te mean minimum temperature change per annum over the catchment for the short-term, mid-term, and longterm periods increases by 0.93°C, 1.8°C, and 2°C, respectively, from the baseline under the RCP4.5 scenario, while it is 1.05°C, 1.5°C, and 2.4°C under RCP8.5 scenario. July and August are months with mean peak fow in the catchment, which corresponds to periods of high rain. Te Te percent changes in average annual components of water balance are mostly positive, while total water yield shows a decrease during short-term period. Most water balance components increase except surface runof and total water yield during mid-term period. Under long-term period (2071-2100), the percent changes in average annual water balance component show a signifcant increase except surface runof and lateral fow.
In general, a decreasing trend in mean maximum and mean minimum temperature, a decreasing trend in evapotranspiration, and an increasing trend in total annual rainfall are detected during the 1989-2018 period resulting in increasing amount of mean streamfow. While a detailed study of the weather systems of the catchment is recommended, the increase in rainfall could be related to the frequent cooling of sea surface temperature in the equatorial Pacifc. On the other hand, an increase in temperature and evapotranspiration with an indefnite pattern of rainfall is expected in future periods. A decrease in streamfow during the main rainy season is expected. Terefore, water resource availability should be included in any planning, and the resources should be used wisely. Te result from this study is crucial to plan irrigation projects and to develop drinking water and other activities.

Data Availability
Te data supporting the current study are available from the corresponding author upon request.

Conflicts of Interest
Te authors declare that there are no conficts of interest.

Annual
Spring Summer Annual