Historical simulations of temperature and precipitation from the CORDEX Africa model in the Wabi Shebele Basin

Rising global temperatures and shifting precipitation patterns have significant socio-economic consequences if not properly studied and predicted. Regional climate models (RCMs) are utilized to assess local-scale climate change. However, the reliability of individual models must be validated due to inherent limitations and methodological constraints. This study evaluates the performance of CORDEX Africa RCMs using observed rainfall and air temperature data from 1986 to 2005. Model performance was evaluated using statistical indicators such as bias, RMSE, r, MAE, and a concise plot of the statistical indicators which is Taylor’s diagram. In rainfall simulation, the RACMO22T performed admirably in the upper parts of the basin (region of high rainfall and cold temperature) and lower regions of the basin (region of low rainfall and hot temperature) with bias −8.64% and 6.19% respectively. HIRHAM5 and CCLM4-8 simulate well the maximum temperature in the upper parts with biases of (0.14 °C and −0.14 °C respectively), whereas RCA4 is well performed in the lower parts of the basin. CCLM4-8 is good for minimum temperature simulation in the upper parts, but HIRHAM5 and RCA4 are good in the lower parts of the basin. In rainfall simulation, all models are slightly good in dry months than in wet. All models underestimated the maximum temperature and overestimated the minimum temperature in the study area as compared to the observed.


Introduction
Over recent decades, climate change has had significant and far-reaching effects on natural systems and human activities across the globe (IPCC 2013, 2014, Pachauri et al 2014).The impacts of climate change, particularly on the natural system, have been profound and widespread (Parry et al 2007, Langat et al 2017).Unless greenhouse gas emissions are reduced, further warming and long-term alterations in all aspects of the climate system can be expected.Developing countries like Ethiopia, whose livelihoods heavily rely on agriculture, are particularly vulnerable to the detrimental effects of climate change (Conway andSchipper 2011, Jilo et al 2019).Consequently, climate change has become a pressing concern for policymakers who are actively devising strategies to mitigate and adapt to the future impacts of climate change.
To study the impact of climate change around the world, climate models play a crucial role in comprehending and predicting the intricate nature of the Earth's climate (Kamworapan and Surussavadee 2019).Numerous global climate models (GCMs) have been developed by research centers worldwide, aiming to simulate the planet's climate on a global scale.However, due to their coarse spatial resolution, GCMs fall short of capturing local climate details (Doblas-Reyes et al 2021).The demand for climate change information at regional and local levels is a central concern in the global change discourse as it is vital for assessing the impacts of climate change on both human and natural systems and developing suitable national-level adaptation and mitigation strategies.
To address this need, high-resolution regional climate models (RCMs) are employed to downscale information, enabling local-scale impact assessments such as river flooding or crop production (Dosio 2018).The Coordinated Regional Climate Downscaling Experiment (CORDEX) is widely utilized to study the impact of climate change around the world.
The primary objective of CORDEX is twofold: to establish a framework for assessing and benchmarking model performance (known as the model evaluation framework) and to design a series of experiments that generate climate projections for impact and adaptation studies (Giorgi et al 2009, Endris et al 2013).Several studies have been conducted to evaluate the ability of CORDEX Africa RCMs in simulating rainfall and temperature across Africa.Nikulin et al (2012) examined the performance of these models and found that they generally provided satisfactory simulations of temperature and precipitation.However, certain regions or seasons exhibited uncertainties and biases.Another study by Ayugi et al (2020) assessed the performance of CORDEX Africa RCMs in simulating temperature and precipitation specifically in East Africa.The findings indicated that the models also offered reasonable simulations of temperature and precipitation in the region.
In a study conducted by Ilori and Balogun (2021) that assessed the performance of CORDEX Africa RCMs in West Africa, it was determined that the models achieved reasonably accurate simulations of temperature and precipitation at a regional level.Similarly, Taguela et al (2020) evaluated the performance of CORDEX Africa RCMs in simulating temperature and precipitation in Central Africa and found that the models provided reasonable simulations of temperature.These studies emphasize that RCMs exhibit varying levels of accuracy when simulating rainfall across different regions.While the models generally provided satisfactory simulations of rainfall variability in the regions under study, there were noted biases and uncertainties in certain areas or seasons.These discrepancies could be attributed to variations in topography and land use/land cover (LULC).Consequently, the studies recommended the careful selection and evaluation of models based on the specific characteristics of the region and variables of interest.
Hence, it is crucial to evaluate by comparing model results with observational data before utilizing climate simulation data from the CORDEX RCMs.This evaluation serves to assess the RCMs' ability to replicate climate conditions in specific locations and to quantify uncertainties in climate model simulations.This information is essential for studying the impacts of climate change, informing policymaking, and adapting strategies according to local conditions (Jacob et al 2012, Endris et al 2013, Takhsha et al 2018, Sørlad et al 2021).
In the context of Ethiopia, research conducted by Worku et al (2018), Dibaba et al (2019), and Demissie and Sime (2021) have shown that the performance of RCMs varies depending on the land characteristics, both on an annual and seasonal basis.It was observed that RCMs do not consistently perform equally well across all locations.Therefore, this research underscores the significance of testing the performance of multiple models in each location to determine the most appropriate model.
This study aims to evaluate the performance of RCMs in simulating air temperature and rainfall in the Wabi Shebele region of Southeast Ethiopia.The Wabi Shebele Basin is characterized by diverse terrain and climate conditions, and the local population heavily relies on agricultural activities for their daily livelihoods.Consequently, the region is particularly vulnerable to the impacts of climate change, wherein uncertain water availability and the increasing frequency of extreme conditions contribute to a higher risk of hydrological disasters such as floods and droughts.On the other hand, the Wabi Shebele Basin possesses significant potential for irrigation development and biomass production, making it a key area for growth and development under the country's Growth and Transformation Plan (GTP).
To provide robust assessments for impact analysis, utilizing multiple models is preferred over-relying on a single model.In this particular investigation, the ICHEC-EC-EARTH model was selected from the African CORDEX projections.While several RCMs within CORDEX have demonstrated good performance in various African studies (Kisembe et al 2018, Vanderkelen et al 2018, Dibaba et al 2019, Demissie and Sime 2021), their performance has not been specifically tested in the Wabi Shebele Basin.Therefore, this study evaluated the performance of four RCMs (RACMO22T, CCLM4-8, RCA4, HIRHAM5) and their ensemble in accurately representing rainfall patterns and air temperature in the Wabi Shebele Basin.
The evaluation in this study specifically examined the performance of RCMs in the Wabi Shebele Basin.Although not all RCMs were compared, the focus was placed on the commonly used and well-performing models in Africa (Ogega et al 2020, Tumsa 2021, Olusegun et al 2023).The study aimed to demonstrate the relationship between the simulated rainfall and temperature by these models and the observed data in the study area.The outputs of the well-performing models will serve as valuable inputs for water resources planning and management strategies.

Description of the study area
Wabi Shebele is one of the 12 river basins in Ethiopia.It is a transnational river basin that Somalia and Ethiopia both share.It came from eastern Ethiopia's high plateau.The river basin is located between latitudes 4 • 45 ′ N and 9 • 45 ′ N and longitudes 38 • 45 ′ E and 45 • 30 ′ E. The overall area of the Wabi Shebele Basin is about 192 960.38 km 2 .The Oromia region is located in the basin's northwest, the Somali region is located in the southeast, and the Harari region is located entirely in the center of the basin.About 38% and 60%, respectively, of the basin area are covered by the regional states of Oromia and Somali.The basin is located topographically between an elevation of 4320 m to 166 m.It rises about 4320 m above sea level in the Bale Mountain ranges of the Galama and Ahmar and drains into the Indian Ocean through Somalia.About 72% of the catchment is lying in Ethiopia (Abebe and Foerch 2006).In figure 1, figure 1(a) representing the Ethiopian major river basin, figure 1(b) representing the location of selected stations, and figure 1(c) shows the elevation ranges in the Wabi Shebele Basin.

Climate of Wabi Shebele Basin
The climate classification in both Ethiopia and the study area is primarily determined by altitude and mean temperature, resulting in five climate zones: hot, warm, tepid, cool, and very cool (table 1).In the highland areas of Wabi Shebele, the temperatures are cooler, making them suitable for human settlement, whereas the lowland regions are arid and unsuitable for human habitation (Adane 2009).The rainfall pattern in the Wabi Shebele Basin exhibits variability, with precipitation levels ranging from 267 mm in the dry and arid parts of the basin to 1044 mm in the upper sections.This variability in rainfall is largely influenced by differences in altitude.The selection of meteorological stations within the basin was based on this climate zoning.For instance, Degahabur, Gode, and Imi experience hot climate conditions, while Adaba, Badesa, and Girawa have very cool climates.The upper parts of the basin have high rainfall with mono-modal rainfall type (Adaba, Badesa, and Girawa stations were found in this part) and the lower parts of the basin have low rainfall with bimodal rainfall type (Degahabur, Gode, and Imi stations were found in this part).All the considered stations are missing data of less than 4%.Therefore, missing rainfall data were filled in using linear regression, while missing temperature data were filled in using the simple average approach.

Data used and data source
Daily rainfall and temperature data were collected from the National Meteorological Service Agency (NMSA) of Ethiopia.Six stations were selected based on data availability and data length.Observed data were used to verify the performance of the RCMs.
The CORDEX African rainfall and air temperature data for this study were downloaded from the Earth System Grid Federation infrastructure (ESGF).The ESGF home website is freely available at (https://esgfnode.llnl.gov/projects/esgf-llnl/). For both rainfall and air temperature, the selected RCMs were: RACMO22T, HIRHAMS, CCLM4-8, and RCA4 models derived by (ICHE EARTH-EC CORDEX Africa) (table 2).The study by (Giorgi et al 2012, Nikulin et al 2012, Demissie and Sime 2021, Sime and Demissie 2022) suggest that RACMO22T, HIRHAMS, CCLM4-8, and RCA4, among other CORDEX Africa RCMs, can be useful tools for simulating climate conditions over Africa and Ethiopia.Indicated that these models are well performed in Ethiopia.Sime and Demissie (2022), show the performance of these models and used them for an impact study over South West Ethiopia.(Dibaba et al 2019) also show the performance of these models over two different watersheds.Other studies (Kisembe et al 2018, Vanderkelen et al 2018, Tumsa 2021) also checked the performance of these models and used them for an impact study.The CORDEX Africa RCMs have spatial grid resolutions for both longitude and latitude of 0.44 • (Luhunga et al 2016).The list of models used were shown in table 2.
The station data can be interpolated from the grid data.The bilinear interpolation technique is the most widely employed to interpolate grid data to the station (Zhou and Zhao 2015).Bilinear uses information from neighboring cells adjacent to a point for interpolation in both weighted and non-weighted systems.Four nearest neighbors are used in bilinear interpolation (Teegavarapu et al 2012).The Climate Data Operator (CDO) program was used to do these tasks.The CDO software is made up of several operators that work together to handle climate and forecast model data consistently.In this study, the downloaded CORDEX-RCMs data was interpolated into point rainfall data using a bilinear interpolation technique.
The Spatial analysis of the Climatological data are conducted using Inverse Distance Weighting (IWD).According to a study by (Amjad andYılmaz 2017, Fagandini et al 2023), rainfall data interpolated with IDW provide more accurate values, demonstrating that IDW is a good method to interpolate average rainfall.IDW's interpolation methods produce accurate results with realistic calculations.

Performance RCM data
To assess the performance of climate models in replicating rainfall patterns in the study area, specific criteria were employed.The first criterion involved evaluating the RCMs' capability to reproduce the climatology of rainfall and the characteristics of rainfall events in comparison to observed data.This entailed comparing various factors such as the magnitude of mean annual and seasonal rainfall, the pattern of mean monthly rainfall, the distribution and frequency of rainfall events, and the return period between the outputs of the RCMs and the observed data (Worku et al 2018).Additional evaluation techniques utilized statistical metrics such as BIAS, root mean squared error (RMSE), and correlation coefficient (CC) (r).These metrics were applied to assess the agreement between the areal average rainfall from the RCMs and the observation data.
To evaluate the performance of RCMs in simulating rainfall and air temperature, metrics such as mean Bias (BIAS), RMSE, and Pearson correlation (r) were employed consistently in multiple studies (Dibaba et al 2019, Demissie and Sime 2021, Sime and Demissie 2022).A smaller absolute value of RMSE and Bias indicates better model performance, while a higher absolute value suggests poorer performance.The value of r ranges from −1 to 1, with −1 indicating a perfect downhill (negative) linear relationship, 1 indicating a perfect uphill linear relationship, and a value close to 0.0 indicating poor performance or correlation of the model.Equations ( 1)-(3) were utilized to calculate BIAS, RMSE, and r, respectively (Pai and Saraswat 2011) Si and Oi represent simulated and observed values of RCMs, respectively, while O denotes the observed value of the climate variable.'I' refers to the simulated and observed pairs, 'n' signifies the total number of pairs, and 'm' represents the mean.

Taylor diagram
For a long time, Taylor diagrams have been employed to evaluate climate and hydrological models and data (Taylor 2012).These diagrams facilitate a straightforward comparison between simulations and observations by considering the patterned CC, the centered pattern root mean square difference (RMS), and the ratio of spatial standard deviations (SDs) between the modeled and observed values (Taylor 2001).By assessing correlations, RMS, and variance ratios, a graphical representation is generated to provide a rapid and concise evaluation of how closely patterns resemble each other (Ghorbani et al 2023).A model is considered better at replicating the spatial pattern of observed precipitation when the centered RMS difference approaches zero and the ratio of spatial correlation to spatial SDs approaches one.

Statistical model performance
The statistical model analysis (RMSE, Pbias (for rainfall), Bias (for temperature), MAE, r, and Taylor diagram techniques were used to verify the model's performance in simulating annual average rainfall and temperature over all the study areas.The performance of all the individual models and their ensemble has been considered in the performance analysis of rainfall and temperature relative to the observed data in both the upper and lower parts of the basin.

Performance of RCM in simulating annual rainfall
Table 3 indicated statistical model performance in both regions, figure 2(a) shows the performance of the rainfall in the upper region of the basin, while figure 2(b) shows the model's performance in simulating rainfall in the lower part of the basin using the Taylor diagram.
In the upper region of the basin, RCA4 and HIRHAM5 overestimate rainfall, while CCLM4-8 underestimates rainfall.The Ensemble and RACMO22T simulated rainfall with slight overestimation and underestimation respectively.In terms of the RMSE statistics, the smallest error was produced by Ensemble and RACMO22T.Using the Taylor diagram (figure 2(a)), the RACMO22T simulated data have low SD as represented by radial coordinate and low values of root mean square difference (RMSD) as represented by the inner semi-cycle.It also has a high CC, as represented by angular coordinates.This indicated that RACMO22T performed well in the upper part of the basin.
In the lowlands of the Wabi Shebele Basin, a region of low rainfall amounts and hot climatic conditions, RACMO22T, and RCA4 outperformed others.The HIRHAM5 and CCLM4-8 underestimated the annual average rainfall (table 3).The CCLM4-8 models only show a positive correlation with the observed data (table 3 and figure 2(b)).Similarly using the Taylor diagram, there is also a high correlation between the observed and simulated data from the RACMO22T and RCA4.However, the standard division ensemble is  far from the observed data than that of RACMO22T, suggesting that RACMO22T is a better fit than the ensemble.CCLM4-8 has large values of RMSD and small values of CC followed by HIRHAM5 (figure 2(b)).
All models simulate rainfall significantly over lowland parts of the basin than in the upper region of the basin.
The study by (Dibaba et al 2019) evaluated the performance of different RCMs for temperature and precipitation in Ethiopia and found that the RACMO22T model performed better than the other models in most of the stations, followed by the ensemble.This suggests that RACMO22T may be a suitable choice for climate change impact assessments involving temperature and precipitation in Ethiopia.
Similarly, the study by Girma et al (2022) evaluated the performance of different RCMs for temperature and precipitation in a river basin in Ethiopia and found that RACMO22T and HIRHAM5 performed well, while the ensemble did not always perform well.The authors recommended careful evaluation of model performance and selection based on the specific characteristics of the region and variables of interest.The study by Demessie et al (2023) evaluated the performance of different RCMs for rainfall in the Guder watershed of Ethiopia and found that RACMO22T performed well in simulation.
Figure 2. The average annual rainfall performance in the upper region of the Wabi Shebele Basin

Performance of air temperature in simulating annual average temperature 3.2.1. The performance the RCMs' in simulating annual average maximum air temperature
In the maximum average temperature simulation, in terms of bias, except CCLM4-8, all models including their ensemble overestimated the maximum temperature in the upper parts of the basin.HIRHAM5 and CCLM4, on average, outperformed the three RCMs models and their ensembles in the upper basin when simulating maximum air temperatures.CCLM4-8, HIRHAM5, RACMO22T, RCA4, and their ensemble showed RMSE values of 2.57 • C, 2.61 • C, 4.73 • C, 3.65 • C, and 3.41 • C respectively in the upper Wabi Shebele Basin.In terms of r, the simulated and observed maximum temperature is positive for all models over the upper region of the basin (table 4 and figure 3(a)).
In the lowland region of the basin, all models slightly underestimated the maximum temperature.In terms of RMSE and MAE, RCA4 performed better than the three RCMs and their ensembles in the lowlands of the basin in simulating maximum air temperatures.CCLM4-8, HIRHAM5, RACMO22T, RCA4, and their ensemble showed RMSE values of 10.51 • C, 3.2 • C, 3.3 • C, 1.7 • C, and 2.8 • C respectively in the lowland of Wabi Shebele Basin (table 4).CCLM4-8 is more biased than all other models in maximum temperature simulation in lowland regions.In the lowland region ensemble and RCA4 is the better.Table 4 shows the statistical indices in the annual maximum temperature simulation.

The performance the RCMs' in simulating annual average minimum air temperature
All models overestimated the average annual minimum air temperature over the upper region of the Wabi Shebele Basin with bias values of 0.8 • C, 0.3 • C, 0.6 o C, 0.6 • C, and 0.7 • C for HIRHAM5, CCLM4, RACMO22T, RCA4, and the ensemble respectively.CCLM4 better performed than all models and ensembles with a Pbias of 0.3 • C (table 5).The r in simulating average annual minimum air temperature is positive for most models over the upper region.Simulations of minimum air temperature have shown a bias in the range of 0.3 • C to 0.8 • C in the upper region of the basin.HIRHAM5 failed to perform in the upper region minimum temperature simulation.
In the lowland region of the basin, the ensemble performed better in simulating the average annual minimum air temperature as indicated in table 5. CCLM4 showed a high value of RMSE compared to other models when simulating the average annual minimum air temperature in the lowland region of the Wabi Shebele Basin.HIRHAM5 slightly overestimated rainfall, whereas all others underestimated it.The ensemble means of the models performed well followed by RCA4 in the lowland region of the basin (table 5).

The performance of the models in simulating annual average air temperature using the Taylor diagram
Figures 3(a) and (b) shows the models' performance in simulating the annual average maximum air temperature in the upper and lower region of the Wabi Shebele Basin respectively; figures 3(c) and (d) show the performance of the models in simulating the annual average minimum air temperature in the highland and low land region of the basin respectively.
In the upper region of the basin in terms of RMSE, RACMO22T, and RCA4 simulated maximum annual average temperature apart from the observed (figure 3(a)).The ensemble has a high correlation with the observed.In terms of SD, HIRHAM5 is well performed followed by the ensemble.In the maximum annual average temperature simulation in the lowlands of the basin, the RCA4 values are close to the observed values (red) concerning SD, followed by Ensemble (figure 3(b)).
When simulating the minimum annual average temperature in the upper region (figure 3(c)), the SD and centered root mean square of the CCLM4-8 model is very close to those observed values, and this shows that it performed well and this is supported by the statistical techniques discussed in (table 5).In terms of SD and root mean square error ensemble is well close to the observed values whereas the CCLM4-8 has a low correlation with the observed (figure 3(d)).
The study by Demessie et al (2023) evaluated the performance RCMs in simulating maximum and minimum annual average air temperature in the Guder watershed of Ethiopia and found that the HIRHAM5 model performed well, indicating its potential usefulness in climate change impact assessments for this region.Similarly, the study by Yonas and Dadi (2022) evaluated the performance RCMs and found that the RCA4 and CCLM4 models performed well in simulating temperature in the Omo Gibe basin of Ethiopia, suggesting their potential usefulness in climate change impact assessments for this region.Figure 3 shows the performance of the models in simulating annual average temperature simulations using the Taylor diagram.

Mean monthly cycle of rainfall
The mean monthly cycle showed the prominent features of the basin's rainfall.The upper region (Adaba, Badesa, and Girawa) have high rainy months from June to September, small rainfall months from March to April, and dry months from November to February (figure 4), and the lowland region (Degahabur, Gode, and Imi) have high rainy months start from March to May and dry months from June to September (figure 5).Variations of stations in climate zone contributed to identifying the more performing RCMs at different stations of the basin specifically.Some models constantly simulate the observed pattern over most stations in the region while others simulate it differently.CCLM4-8 underestimated rainfall data over the upper region of the basin, especially during rainy months.HIRHAM5 dynamically simulates observed data over the two regions of the Wabi Shebele Basin and it simulates the peak value of rainfall in the both upper and lowland regions of the basin.RCA4 simulates better in the lowland region than in the upper region.The mean monthly cycle of observed and simulated rainfall at the upper stations of the Wabi Shebele Basin was shown in figures 4 and 5 shows the mean monthly cycle of observed and simulated rainfall over the lowland of the Wabi Shebele Basin.

Mean monthly cycle of maximum air temperature
The monthly mean maximum air temperature cycle indicated that most RCMs overestimated the annual mean maximum temperature in the upper region of the basin.The CCLM4 underestimates the maximum temperature in the upper parts of the basin (figure 6) and failed to simulate the maximum air temperature.The RACMO22T and RCA4 simulate maximum air temperature data that differ/deviate from the observed data with high values.The HIRHAM5 and ensemble simulated values are somewhat good when compared with other models.Figure SM3 indicated the mean monthly cycle of maximum air temperature in the upper parts of the basin.
In the lowland parts of the basin, RACMO22T is highly biased in simulating maximum air temperature in April and October, while CCLM4-8 is biased in May, June, July, August, and September months (Figure SM4).RCA4 performed better than other RCMs in the low region of the basin in simulating maximum temperature.In the low region, all models underestimated the mean maximum air temperature except the RCA4 model for January to August in the basin (figure 7).

Mean monthly cycle of minimum air temperature
Based on the analysis, it was found that all models tended to overestimate the mean monthly minimum air temperature in the upper parts of the basin throughout the year (as seen in figure 8).Specifically, HIRHAM5 exhibited a significant overestimation of the minimum temperature, while CCLM4-8 consistently underestimated it relative to the other models.Among the models investigated, CCLM4-8 performed comparatively better in terms of accurately representing the minimum temperature (as shown in figure 8).
In the hot regions of the basin, HIRHAM5 consistently overestimated the minimum temperature for all months, while the other models consistently underestimated it throughout the year.In this particular region, the ensemble of models performed more accurately compared to individual models.Notably, the models produced better simulations with more accurate values during June, July, and August.However, they displayed substantial biases during January, February, and March.The mean monthly cycle of minimum air temperature in the lowland area of the Wabi Shebele Basin can be observed in figure 9.

Spatial rainfall distribution
Spatial analysis was done using yearly average rainfall over the period (1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005).The spatial distribution of the rainfall shows that there is a high amount of rainfall over the Upper parts of the basin and a low amount of rainfall downstream of the basin.The yearly average rainfall shows decreasing from the Upper parts to the downstream parts of the basin.HIRHAM5 and RCA4 simulated a large amount of rainfall in the upper parts than all other models (figure 10).Observed, Ensemble and RACMO22T have a similar distribution of rainfall over the basin.

Conclusion
This study evaluated the performance of the individual models and ensemble average over the upper and lowland regions of the Wabi Shebele Basin in simulating rainfall and air temperature.The study examined how well the RCMs simulated the mean monthly and annual variability in rainfall and air temperature.
Specifically, the study found that based on both mean annual statistical performance and Taylor diagram performance checking techniques, the RACMO22T and ensemble better-simulated rainfall over the upper regions with bias values of −8.64% and 5.21% and RAMSE values of 0.37 and 0.30 respectively.CCLM4-8 and RCA4 shows biases with −27.94% and 34.40% respectively this indicated that these models failed to perform in the region of high rainfall with cold temperature.RACMO22T is also better-simulated rainfall in the lowland regions with a bias of 6.19% and RMSE of 0.33.This suggests that RACMO22T has potential usefulness in climate change impact assessments for different regions and variables of interest.
The study indicates that the HIRHAM5 and CCLM4 models performed well in simulating maximum temperature in the cold region or upper parts of the basin with the bias of 0.14 • C and −0.14 • C respectively.While the ensemble and RCA4 models were better suited for simulating maximum temperature in the hot region of the basin with biases (0.09 • C and −0.11 • C) and RMSE (2.82 and 1.73) respectively.The CCLM4 model was effective in simulating the upper minimum temperature in the basin, while the ensemble model performed well in simulating the minimum temperature in the hot region of the basin.
The monthly cycle showed the prominent features of the basin's rainfall.The upper region (Adaba, Badesa, and Girawa) showed main rainy months from June to September, small rainfall months from March to April, and dry months from November to February, and the lowland region (Degahabur, Gode, and Imi) showed rainy months start from March to May and dry months from June to September.The monthly mean maximum air temperature indicated that most of the RCMs overestimated the observed annual mean maximum temperature on average in the upper region of the basin.The RCMs underestimated the minimum air temperature in both upper and lowland regions of the study area throughout the months of the year in historical periods.The best-performed models, CCLM4 and RACMO22T will be used to evaluate and assess rainfall change impact over the Wabi Shebele Basin.Furthermore, the Ensemble average of the RCMS is recommended to estimate the variability of the minimum air temperature over the lowland region of the basin.
In general, it is important to carefully evaluate the performance of climate models in specific regions and for specific variables of interest.This can help identify the most suitable models for climate change impact assessments or other applications and support evidence-based decision-making and planning for climate change adaptation.Overall, these findings underscore the importance of model evaluation and selection in climate change impact assessments, taking into account the specific characteristics of the study region and the variables of interest.

Figure 1 .
Figure 1.Location of the study area.

Figure 2 .
Figure 2. The performance of the models in simulating rainfall using the Taylor diagram.

Figure 3 .
Figure 3.The performance of the models in simulating air temperature using the Taylor diagram.

Figure 4 .
Figure 4. Mean monthly cycle of observed and simulated rainfall at upper stations of Wabi Shebele Basin.

Figure 5 .
Figure 5.The mean monthly cycle of observed and simulated rainfall over the lowland of the Wabi Shebele Basin.

Figure 6 .
Figure 6.The mean monthly cycle of maximum air temperature in the upper parts of the Wabi Shebele Basin.

Figure 7 .
Figure 7.The mean monthly cycle of maximum air temperature in the lowland parts of the Wabi Shebele Basin.

Figure 8 .
Figure 8. Mean monthly cycle of minimum air temperature in the upper parts of the Wabi Shebele Basin.

Figure 9 .
Figure 9. Mean monthly cycle minimum air temperature in the lowland of the Wabi Shebele Basin.

Figure 10 .
Figure 10.Spatial distribution of annual average rainfall over Wabi Shebele Basin.

Table 1 .
Location of the station with their rainfall and temperature data.

Table 3 .
Statistical indices in annual rainfall simulation.

Table 4 .
Statistical indices in annual maximum temperature simulation.

Table 5 .
Statistical indices in annual average minimum air temperature simulation.