Impact of climate change on the streamflow of the Arjo-Didessa catchment under RCP scenarios

In this study, the impact of climate change on the streamflow of the Arjo-Didessa catchment, Upper Blue Nile basin, is evaluated. We used the outputs of four climate models for two representative concentration pathway (RCP) climate scenarios, which are RCP 4.5 and RCP 8.5. Streamflow simulation was done by using the HEC-HMS rainfall-runoff model, which was satisfactorily calibrated and validated for the study area. For the historic period (1971–2000), all climate models significantly underestimated the observed rainfall amount for the rainy season. We therefore bias-corrected the climate data before using them as input for the rainfall-runoff model. The results of the four climate models for the period 2041 to 2070 show that annual rainfall is likely to decrease by 0.36 to 21% under RCP 4.5. The projected increases in minimum and maximum temperature will lead to an increase in annual evapotranspiration by 3 to 7%, which will likely contribute to decreasing the annual flows of Arjo-Didessa by 1 to 3%. Our results show that the impact is season dependent, with an increased streamflow in the main rainy season but a decreased flow in the short rainy season and the dry seasons. The magnitudes of projected changes are more pronounced under RCP 8.5 than under RCP 4.5.


GRAPHICAL ABSTRACT INTRODUCTION
The fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) indicates that rainfall over Eastern Africa has decreased between March and June in the last three decades, while there has been an increase in temperature over East Africa since the beginning of the 1980s. Climate projections also indicate that there will be a likely increase in rainfall amount and extreme rainfall in the region by the end of the 21st century. There will be higher rates of evaporation in Ethiopia due to warming over the country. Such changes are expected to impact the economy of East African countries, including Ethiopia, by high climate sensitivity (Change ). under the RCP climate change scenarios. The study considered the RCP 2.6, RCP 4.5, and RCP 8.5 scenarios using the second-generation Canadian Earth System Model (CanESM2) and concluded that, due to climate change, the streamflow of the watershed is found to be increasing by 4.06, 3.26, and 3.67% under the RCP 2.6, RCP 4.5, and RCP 8.5 scenarios, respectively. Mengistu et al. () used the CCLM regional climate model (RCM) to assess the climate change impacts on the UBN River basin by projections of increases in mean annual temperature and decrease in precipitation in most parts of the basin. They concluded that the total water yield of the basin is estimated to decrease by 1.7 to 6.5% and 10.7 to 22.7%, for simulations forced by the RCP 4.5 and RCP 8.5 scenarios, respectively. This indicates that climate change has different impacts on a small scale than on a large scale. Although some studies were done on the UBN and its sub-basin using the RCM, different projected streamflow magnitudes were observed between them. This is mainly due to most studies using a single climate model. River basin and reviled that the average annual temperature and evapotranspiration will increase by 19.1 and 12.6%, respectively, while rainfall is expected to decrease by 15.3% in 2100 and the flow is also likely to decrease. Similarly, Tariku et al. () showed the impact of climate change on hydrology and hydrologic extremes of the Upper Blue Nile River basin using three hydrological models and four climate models under the RCP 4.5 and RCP 8.5 scenarios. The author concludes that the potential global warming impact on the extreme and mean streamflows of the UBN River basin was projected to decrease by 7.6% in 2050s and in the 2080s.

Evaluation of simulated rainfall by climate models
For the time period (1981 to 2000), the downscaled climate rainfall data from the selected climate models were

Bias correction
In this study, simulated precipitation and temperature data were bias-corrected before being used as an input to runoff modeling. The temperature was bias-corrected by the linear shifting and scaling method (Terink et al. ).
where T corr is the corrected daily temperature ( C); T rcm is the simulated daily temperature from the RCM; and T obs is the observed daily temperature, while T obs is the mean observed temperature and T rcm is the mean simulated temperature.
The simulated precipitation data were also bias-corrected by using a nonlinear correction method. This method results in the mean and standard deviation of the daily precipitation distribution becoming equal to those of the observed distribution (e.g. Lafon et al. ). The equation reads: where P* is the corrected value of the variable (precipitation), b is the scaling exponent, and a is the coefficient that is determined from the mean of observed rainfall data and the mean of P b .

HEC-HMS model calibration and validation
In this study, 2 years of daily data (1981 to 1982)   Muskingum X parameter was set at the default value of 0.2, since it was found that the model outputs were not sensitive to this parameter.

Model performance evaluation
The most straightforward possibility of evaluating model performance is a visual inspection of the observed and simulated hydrograph. This graphical technique provides an initial general overview. In this case, the inspection of the graph first focused on the hydrograph pattern, which was followed by an inspection of the base flow and then of peak flow. The NSE values can range between 1.0 (perfect fit) and À∞. An NSE value of less than zero indicates that the mean value of the observed time series would have been a better predictor than the model. The performance of the model is commonly considered very good (NSE > 0.8), good (0.6-0.8), satisfactory (0.5-0.6), and unsatisfactory (NSE < 0.5) (Pachepsky et al. ). NSE is defined as follows: where Q O,i is the observed discharge at the time step i, Q O is the mean of the observed discharge, Q S,i is the simulation discharge at the time step i, and n is the number of observations. RVE is estimated by where all terms are as defined previously. The RVE values range between À∞ and þ∞. The model performance is considered very good if the RVE value is between À5 and 5% and satisfactory when it is between 5 and 10% and À10 and À5%.
Standard periods are often defined for the impact study. Since the short-term is already happening and the long-term is too far, our impact analysis focused more on the middleterm period. Climate change impact on streamflow was analyzed statistically at monthly, seasonally, and annual scales.

Model sensitivity to parameters
The HEC-HMS model sensitivity to its parameters was eval-

Calibration and validation
The simulated and observed hydrographs for the calibration period are shown in Figure 3. Overall, the observed streamflow hydrograph is well simulated by the model. The rising and recession limbs of the simulated hydrograph are also well simulated, but the timing of the simulated hydrograph is not optimal. The main limitation of the model is in reproducing peak flows with noticeable underestimation for some years. We assume that the mismatch between the observed and the simulated hydrographs, particularly for high flows, will also be affected by the rating curve. We therefore suggest that the rating curve of the station must be revisited.
Model performance was found good in capturing the observed hydrograph pattern when evaluated by using NSE (NSE ¼ 0.65). The RVE for the calibration period is 5.1%, which suggests that the model is highly efficient in estimating the observed streamflow volume. We consider that the percentage error in peak flow is 19%, which is relatively large. The model limitation and quality of observed data for peak flows may have contributed to the performance in simulating peaks.

Evaluation of rainfall estimates from climate models
All four climate models failed to satisfactorily reproduce the observed annual rainfall of the Arjo-Didessa catchment ( Figure 4). These models mostly underestimated the observed rainfall amount with notable underestimation for the rainy season. The peak monthly rainfall amount was captured only by HadGM2-ES. However, this model noticeably misses the start of the rainy season and ends earlier  The annual rainfall of the Arjo-Didessa catchment is underestimated by as much as À34.8 to À72% with an average bias of À46.5% (Table 5). The largest bias was shown by the CM5A-MR model. The simulated annual rainfall is mostly more variable than the observed annual rainfall.
HadGM2-ES showed the worst performance in reproducing the temporal variability (CV ¼ 17.2%) of the annual rainfall amount (CV ¼ 7.6%). The performance of the models was also unsatisfactory when evaluated by RMSE and CC.
These values indicate that the simulated rainfall over the catchment is not fit for direct use for our climate change impact study, but that bias correction is needed.

Climate change impact
The climate models do not agree in terms of both the mag- However, there will be a significant decline (up to 21%) in annual rainfall of the study area according to MPI-ESM-      However, the results of all models show that streamflow will increase in Kiremet even though there is a significant difference in the magnitude of change between the models. In Kiremt, runoff over the catchment is projected to increase by up to 18%.     The climate models did not satisfactorily capture the monthly rainfall pattern, volume, and peak of the study area. The annual rainfall amount was significantly underestimated, while there was also a weak linear relationship between the simulated and the observed annual rainfall amount (correlation ¼ À0.08 to À0.354). We, therefore, applied bias correction to the rainfall and temperature data of the climate models before further analysis. The daily maximum temperature is projected to increase by 1.17 to 1.39 C under the RCP 4.5 scenario. The minimum temperature is projected to increase by 0.98 to 1.24 C. Consequently, the annual PET will increase by 3 to 5%. The magnitude of PET change in this study is much smaller than that in some previous studies (Nawaz et al. Under RCP 4.5, the annual streamflow of the study area is projected to decrease by small amounts (<3%). There will also be a decrease in the flows of the dry season and the small rainy season. However, the future flow will be higher than historic flows (by up to 5%) in the main rainy season. The results are also similar to those for RCP 8.5 despite some inconsistencies between the projections of the climate models. The annual streamflow magnitude that we report in this study is different in both direction and mag-