Performance assessment of bias correction methods using observed and regional climate model data in different watersheds, Ethiopia

The behavior of the climate system, its components, and their interactions can be studied and simulated using tools known as climate models. Each component or group of components of the climate system can be represented by models of varying complexity. Climate models use a quantitative method to simulate the interactions of the atmosphere, oceans, land surface, and ice. They are used for a variety of purposes from the study of the dynamics of the climate system to projecting the future climate. These are designed mainly for studying climate processes and natural climate variability. Climate models are also used to compute the consequences of greenhouse gas emission scenarios, and to simulate and quantify the climate response to present and future human activities. The climate model is a mathematical description of the Earth's climate system, broken into a number of grid boxes and levels in the atmosphere, ocean, and land. At each of these grid points, equations are solved which designate the large-scale balances of the momentum, heat, and moisture. Based on this, a wide range of climate models are developed (Roeckner et al. 2004).

The modeling approach and the resolution of the model vary from one model type to another. The climate models driven by global climate model (GCM) projections can be used to assess the possible effects of future climate on hydrology. Although GCMs limit the exact simulations of regional climatology owing to the inability to accurately simulate features of local or regional climate including topography, orthography, cloudiness, and land use due to the inherent coarse spatial resolution ranging between 100 and 250 km, the resolution of regional climate models (RCMs) is tens of kilometers, in the range of 12–50 km, in the proximity of the watershed scale (Mearns & Hulme 2001; Dibaba et al. 2019; Gunathilake et al. 2020).

The RCM is also called the limited-area model. It is the best tool for dynamic downscaling of climate data from the output of a GCM and enables us to make predictions for a particular region. It provides a new opportunity for climate change effect analysis since they have a higher spatial resolution and more reliable results on a regional scale compared to GCMs (Buonomo et al. 2007; Chen et al. 2013; Turco et al. 2017).

The basic strategy in regional modeling is to rely on the GCM to reproduce the large-scale circulation of the atmosphere, and to simulate sub-GCM scale regional patterns of climate, such as precipitation, and temperature over the region of interest. The atmospheric fields such as surface pressure, wind, temperature, and humidity are simulated by the GCM and are fed into the vertical and horizontal boundaries of the RCM. The RCM simulations for multi-period employed downscaling under the lateral boundary conditions are from a completely coupled GCM and components of the climate system (Kuma et al. 2021).

The advantages of the RCM over global climate models are as follows: it provides a high spatial or temporal resolution and highly resolved information, the information is derived from physically-based models, its ability to model atmospheric processes and land cover changes explicitly, many variables are available and better representation of some weather extremes than in GCMs (Mearns & Hulme 2001).

The green house gas (GHS) emission scenarios constructed on technological growth, socioeconomic development, and demographic growth add uncertainty to climate change impact evaluations. To reduce such uncertainties, using multi-RCMs is better compared to the use of a single RCM (Gunathilake et al. 2020; Kuma et al. 2021). Furthermore, the RCM simulations are dependent on the regions. Moreover, the RCM rainfall simulations differ along the regions, performing well in some regions and poorly in others (Endris et al. 2013). The recent studies by Dibaba et al. (2019) and Van Vooren et al. (2019) indicate that RCM estimation is sensitive to elevation, producing higher biases for higher elevation.

The performance of the COordinated Regional Downscaling EXperiment (CORDEX)-Africa regional climate simulations in representing the hydrological cycle of the Niger River basin was studied (Mascaro et al. 2015). The results demonstrate that the water balance is mostly sensitive to the parameterized land surface and atmospheric processes of the nested RCMs, with less influence on the driving general circulation model. Moreover, the performance of the RACMO22T and RCA4 to simulate the daily observed precipitation and minimum and maximum temperatures of the Gelana watershed was investigated by Daniel & Abate (2022), and the results indicate that the RACMO22T climate model is performed better than the RCA4 model. Similarly, evaluation of the CORDEX RCM’s performance in simulating climate conditions of two catchments in the Upper Blue Nile Basin, Ethiopia, was studied (Dibaba et al. 2019). The results suggested that the RACMO22T model simulates precipitation over most stations better than the other models.

The resolutions of the RCMs are highly resolved, it is a better one. Yet, RCM projected climatic variables should be handled with caution because they contain significant biases due to imperfect conceptualization, discretization, internal climatic variability, and spatial averaging within grid cells (Gunathilake et al. 2020). These reasons limit the direct use of RCMs in hydrological models. The output of climate models simulates with a little overestimation and underestimation discrepancies in values over the catchment (Legesse et al. 2015). This bias should be corrected by using the bias correction methods. It is used to compensate for any tendency to overestimate or underestimate the mean of downscaled variables. Bias correction factors are computed from the statistics of observed and historical simulated variables for the same duration. Bias correction methods are assumed to be stationary, i.e., the correction algorithm and its parameterization for current conditions are assumed to be valid for future conditions as well. This approach is commonly used to adjust simulated climate data at appropriate spatial and temporal scales in hydrological modeling (Teutschbein & Seibert 2012; Gadissa et al. 2019).

Biases may be amplified when climate change effects are included, such as in hydrological effect studies. Bias correction should be considered before direct use of GCM-RCM precipitation projections over complex territories (Mendez et al. 2020). Therefore, bias correction of RCM data is the prerequisite step to the data being used in any climate change effects analysis (Luo et al. 2018). The bias correction methods are used for significantly improved simulated precipitation and temperature time series, as indicated by goodness-of-fit statistics of both frequency and time series-based statistics (Zhang et al. 2018). The bias correction can be done through linear scaling (LS), power transformation (PT), local intensity scaling, distribution mapping (DM), variance scaling, delta change, and quantile mapping (QM) (Teutschbein & Seibert 2012).

The bias correction for simulated precipitation was carried out by the LS, PT, and DM methods, whereas LS, variance scaling (VARI), and DM methods were used for adjusting the maximum and minimum temperatures of the Gelana watershed (Daniel & Abate 2022). The results suggest that, based on their range of variability and their ability to bring the raw RCM simulations closer to observations, the PT method performed better for precipitation, while the DM method performed better for maximum and minimum temperatures. A parallel study was carried out in the Bilate catchment, Southern Ethiopia (Kuma et al. 2021). Analogous supposition was also drawn by Teutschbein & Seibert (2012), supported by Geleta & Gobosho (2018) and Dibaba et al. (2020) in the Finchaa watershed in Ethiopia.

To manage water resources, the critical issue is analyzed, quantifying and modeling the different elements of hydrologic processes. Moreover, hydrological models have been set up for the watershed by inputting the climatic parameters (Eyasu 2022). Thus, correcting the bias of climatic parameters is necessary for modeling a watershed.

Some parts of the Ethiopian Rift Valley basin and Omo Gibe basin are located in arid and semiarid regions and situated in Horn Africa, and are extremely vulnerable to climate change effects since most micro-macro water projects are constructed and under structure in these basins (Shiferaw et al. 2018; Daniel & Abate 2022). Thus, it is essential to select suitable bias correction methods for providing credible inputs to evaluate climate change impacts over the regions. Therefore, the Gelana watershed from the Rift Valley basin and Deme watershed from Omo Gibe basin were selected as a case study, where the economies are heavily reliant on rain-fed agriculture, undoubtedly making them victims of the climate change. The objective of the study is to investigate the performance of RCMs and bias correction methods for downscaling climatic outputs of CORDEX-Africa and provide a better combination of bias correction methods for precipitation and temperature on the study areas. The four precipitation and temperature correction methods are included for the evaluation of bias correction methods that are comprised of the three RCMs in this study. The effect of bias correction methods on climate data simulations is studied by comparing with observed meteorological data. Furthermore, the study outputs are intended to be used for the evaluation of the climate change impact on hydrology, water resources, and environments, and other related studies on the Deme and Gelana watersheds.

The study was conducted in Gelana and Deme watersheds located in the southeastern and southern parts of Ethiopia, in Rift Valley and Omo Gibe basin, respectively.

Climate data of the six and seven meteorological stations in and around the Deme watershed and Gelana watershed, respectively, from the period 1988 to 2019 were collected from the national meteorological agency of Ethiopia. Rainfall in the Deme watershed has strongly varied seasonally and by elevation. The wet months are extending from April to May, and August to September. In addition, the mean annual rainfall of the watershed is computed to be 1,341 mm. Also, the elevation of the watershed region ranges from 1,138 to 3,269 m above the sea level.

The topographic feature of the Gelana watershed has diverse altitudinal difference which ranges from 1,171 to 3,167 m above the mean sea level. The rainfall pattern of the Gelana watershed is the bimodal profile with an absolute peak in May and a relative peak in October, with the main rain occurring from March to May (Belg), and from August to November (end of Kiramet to Tseday). The mean monthly rainfall in Gelana watershed varies from 11.1 mm in January to 243.4 mm in May for the period from 1987 to 2019. Moreover, relatively intensive rainfall was received in April, May, September, and October. The minimum rainfall occurred from December to February but started to increase in March.

The climatic condition of Ethiopia is classified into five climatic zones based on altitude and temperature variation. It ranges from the high cold area known as Wurch to the highly hot climatic condition area known as Berha. These zones are Wurch (cold to moist, altitude > 3,200 m), Dega (cool to humid with altitude between 2,300 and 3,200 m), Weyenadega (cool to sub-humid in between 1,500 and 2,300 m), Kola (warm semiarid, between 500 and 1,500 m in altitude), and Berha (hot arid type, altitude < 500 m) (Melesse et al. 2013). Therefore, the climatic condition of the Deme watershed and Gelana watershed ranges from humid in the highland to semiarid in the lowland of the watershed.

This study was done by analyzing some data: spatial data (Digital Elevation Model (DEM)), meteorological data (precipitation, maximum and minimum temperatures), and the climate data from the RCMs-CORDEX-Africa. DEM is used to obtain spatial information and delineate the watershed. Meteorological data (precipitation, maximum and minimum temperatures) were used for performance checking of the climate models and bias correction methods.

Data were collected from the National Meteorological Agency, CORDEX-Africa-RCMs, Alaska Satellite Facility, and United States Geology Survey websites. Furthermore, their sources and descriptions are summarized in Table 1.

The climate data, including rainfall, maximum and minimum temperatures of both watersheds are summarized in Table 2 and Table 3. There are seven and six meteorological stations in and around the Gelana and Deme watersheds, respectively.

The collected data were effectively analyzed by using the software and materials. These are Arc GIS, RCMs, Excel, and Climate Model data for hydrologic modeling (CMhyd) software. Furthermore, the major materials and software used and their purposes are tabulated and summarized in Table 4.

CMhyd software is designed to work with the CORDEX data archive, which provides downscaled RCM data. It is obtained from https://swat.tamu.edu/software/ and is used for the extraction of CORDEX-NetCDF and bias correction of precipitation, minimum and maximum temperatures (Dibaba et al. 2020). It is also used to correct the bias of the extracted climate data from the CORDEX model, such as precipitation, minimum and maximum temperature simulations for historical and future periods of analysis under different scenarios (Geleta & Gobosho 2018).

The choice of baseline can highly influence the result of a climate change impact assessment in hydrology. According to Carter (2007), the baseline period is usually selected based on the following criteria:

• Representative of the current-day or recent average climate in the study region.

• Includes sufficient duration of a range of climatic variations (severe drought and flood season).

• Encompass a period for which data on all major climate variables are adequately distributed over space and readily available, and includes data of high quality.

• Consistent with the climatological baseline used in other impact assessments.

Therefore, the baseline periods from 1988 to 2019 were selected for the study based on the representative of the recent year, sufficient duration of a range of climatic variations, readily available for all major climate variables and including data of high quality.

The CORDEX model is an initiative of the World Climate Research Program, aimed at regional climate modeling to produce an ensemble of high-resolution historical and future climate projections through dynamically downscaled Coupled Model Inter-comparison Project 5 atmospheric Ocean General Circulation Model outputs using multiple RCMs (Gunathilake et al. 2020). The past practices indicated that projects such as Atmospheric Model Intercomparison Project (AMIP) and Coupled Model Intercomparison Project (CMIP) are precious for the global modeling community. However, nowadays, CORDEX is essentially structured to play a similar role for the regional climate downscaling community (Giorgi et al. 2009; Daniel & Abate 2022).

CORDEX-Africa is one of the special concerns of the CORDEX program. CORDEX-Africa ensembles with multi-GCM/multi-RCM help to evaluate the climate variation signal, to identify and quantify the many foundations of uncertainty (Dosio & Panitz 2016). The CORDEX-Africa model data at longitude 0.44° and latitude 0.44° horizontal resolution and a multi-model ensemble of RCMs with their driving GCMs provide the boundary conditions (Kuma et al. 2021). Africa is nominated as the target area of the CORDEX program for three major reasons. These reasons are: the high vulnerability of this region in many sectors follows from climate variability, the relatively low adaptive capacity of its economies, and significant changes in temperature and precipitation patterns (Giorgi et al. 2009; Dibaba et al. 2019).

CORDEX promotes international downscaling coordination, and facilitates easier analysis by scientists and end-user communities at the local level of regional climate changes with many methods of RCMs for Africa. It is a bridge between the existing gap of the climate modeling and the end-user communities (Giorgi et al. 2009; Trzaska & Schnarr 2014). Several studies used CORDEX to downscale the different climate models simulations for climate change effect analysis in the world. For instance, the future climatic data from the CanRCM4; one of the RCMs used in the CORDEX, to quantify the impacts on streamflow dynamics for two major rivers of the Northern Lake Erie Basin in Canada (Zhang et al. 2018). A similar study was done by using RegCM4 in Central America by Montecelos et al. (2018). The performance evaluation of the climate model outputs to CORDEX simulations over Africa has been carried out in many studies including Luhunga et al. (2016), Pinto et al. (2016), Akinsanola & Ogunjobi (2017), Mutayoba & Kashaigili (2017), and Näschen et al. (2019), which explain the better performance of CORDEX models. Climate model outputs together with CORDEX simulations in Ethiopia have been carried out in some studies including Geleta & Gobosho (2018), Dibaba et al. (2020), and Daniel & Abate (2022), which assessed the impact of climate change by using the RCMs from the CORDEX-Africa for the future climate scenarios.

RCMs were used due to high resolution and resolved information, and their ability to model atmospheric processes and land cover changes explicitly, wider range of variables, and better representation of some weather extremes than in GCMs (Mearns & Hulme 2001). In addition, based on the vintage, resolution, validity, and representativeness, and among the several RCMs, RACMO22T, HIRHAM5 and RCA4 models were selected. Furthermore, their performances were discussed by Mascaro et al. (2015), Dibaba et al. (2019), and Daniel & Abate (2022). Generally, in this study, from CORDEX-Africa, the three RCM simulations for the baseline periods were downscaled under historical scenarios, as shown in Table 5.

The downloaded RCM data from CORDEX are in the NetCDF file format. Therefore, the precipitation, minimum and maximum temperatures are extracted by ArcGIS software through a multidimensional tool and NetCDF table view. The grid points were extracted, which were the nearest for observed meteorological stations by using the latitude and longitude of observed stations (Kuma et al. 2021; Daniel & Abate 2022).

Four precipitation and temperature bias correction methods were applied and compared in both watersheds in this study, as shown in Table 6. The precipitation correction methods are LS, DM, empirical quantile mapping (EQM), and PT, whereas DM, EQM, LS, and variance scaling were used for temperature correction by using CMhyd software. These are the main types and cover major categories of existing bias correction methods. All of them are conducted on a daily basis for each calendar month during the period 1988–2019 in each watershed.

The LS approach assumes that the correction algorithm and parameterization of historical climate will remain stationary for future climatic conditions. Climate change studies done in different regions of the globe validate that the LS approach performs well for coarse temporal scale analysis. Similarly, the performance of the LS method is evaluated by using statistical indicators (Gunathilake et al. 2020). LS adjusts RCM time series with correction values based on the relationship between long-term monthly mean observed and RCM control run values. It adjusts both variables (precipitation and temperature). For climate projection, the RCM output bias corrected by the LS approach has a very good capability to replicate the historical maximum and minimum temperature and precipitation for the observed period (Luo et al. 2018). It corrects the mean, and the variability of corrected data is more consistent with the original RCM data. The limitations are standard deviation, wet-day frequencies, and intensities are not corrected, also all events are adjusted with the same correction factor (Lenderink et al. 2007). The LS method implements a constant corrected factor that is estimated by the difference between the original RCM simulations and the observations for each calendar month.

The Local Intensity Scaling (LOCI) method aims to simultaneously correct the precipitation intensity and frequency. Initially, the rainfall intensity threshold for each month is confirmed. Accordingly, the number of wet days in RCM precipitation that exceed this threshold matches the number of days for which observed precipitation is determined. This approach can effectively eliminate the drizzle effect because too many drizzly days are often included in original RCM outputs (Luo et al. 2018).

While LS accounts for a bias in the mean, it does not allow differences in the variance to be corrected. Therefore, PT can be used to specifically adjust the variance statistics of a precipitation time series (Teutschbein & Seibert 2012). The PT method utilizes a nonlinear approach in an exponential form to correct the mean and variance of the precipitation series (Luo et al. 2018; Zhang et al. 2018).

The described PT method is capable of correcting both the mean and variance of precipitation while being restricted to correct temperature in the use of the power function. A viable alternative is offered by the VARI method, which is developed to correct both the mean and variance of temperature (Luo et al. 2018).

The DM method is applied to correct the distribution function of the RCM outputs and to align them with the observed distribution function. This can be done by creating a transfer function to shift the occurrence distributions of precipitation and temperature. It is based on the assumption that both the RCM-simulated and observed climatic variables obey a specific frequency distribution (Teutschbein & Seibert 2012).

Gamma distribution with shape parameter and scale parameter is often considered to be appropriate for the precipitation probability distribution, which previous studies have shown to be effective (Zhang et al. 2018). Concerning temperature, the Gaussian distribution (normal distribution) with location parameter and scale parameter is often assumed to agree with the optimal temperature distribution (Luo et al. 2018).

The DM method uses the transferring of function to adjust the cumulative distribution of estimated data to the cumulative distribution of rain gauges, and it reproduces precipitation and temperature very well (Dibaba et al. 2020).

The QM technique is among the most effective and common bias correction methods. It implements statistical transformations for the post-processing of climate modeling outputs. It has the ability to overcome systematic errors from the climate model simulations (Enayati et al. 2021). The QM method is a nonparametric bias correction method and is generally applicable for all possible distributions of precipitation without any assumption on precipitation distribution. It can effectively correct bias in the mean, standard deviation and wet-day frequency as well as quantiles (Fang et al. 2015). QM bias correction algorithms are commonly used to correct systematic distributional biases in precipitation outputs from climate models. It can be applied to any kind of climatic variable. Its principle is based on pointwise daily constructed empirical cdfs (Luo et al. 2018), although the degree of corruption in the climate models' trends by QM is particularly large for changes in long period return values. Furthermore, it has been argued that projected trends should be preserved following bias correction so that the underlying model's climate sensitivity is respected (Cannon et al. 2015). However, all QM methods may not necessarily have a similar capability in the correction of the bias of the RCM outputs. This is especially the case with distinct topographic features, since it has been an already challenging task for RCMs to simulate the climatic variables under such conditions (Enayati et al. 2021).

Relative to QM, detrended quantile mapping (DQM) incorporates additional information about the climate model outputs in the projected period, in the form of the projected mean. Depending on the degree of extrapolation still required after detrending (and the means by which extrapolation is handled), the climate change signal from DQM will tend to match that of the underlying climate model. This applies to the mean, but does not necessarily apply to all quantiles, such as those in the tails of the distribution that define climate extremes (Cannon et al. 2015).

Quantile delta mapping (QDM) is compared on synthetic data with DQM, which is designed to preserve trends in the mean, and with standard QM. Performance is assessed based on precipitation extremes indices and results from a generalized extreme value analysis applied to annual precipitation maxima. QM can inflate the magnitude of relative trends in precipitation extremes with respect to the raw climate models, often substantially, as compared to DQM and especially QDM (Cannon et al. 2015). The QDM for precipitation preserves model-projected relative changes in quantiles, while at the same time correcting systematic biases in quantiles of a modeled series with respect to observed values. Preservation of relative changes follows directly from the quantile delta change and quantile perturbation methods, both of which apply simulated relative changes in quantiles on top of the observed historical series. The idea is similar to DQM, except that relative changes in all modeled quantiles are accounted for rather than only relative changes in the modeled mean (Cannon et al. 2015). Systematic distributional biases are relative to observations in a historical baseline period and are corrected by first detrending all projected future quantiles from a model and then applying QM to the detrended series, thus ensuring that the climate sensitivity of the underlying climate model, at least so far as quantiles are concerned, is unaffected by the bias correction (Cannon et al. 2015).

QM and QDM are applied to simulated daily temperature and precipitation over China from an RCM projection. The study outcomes show that both the QM and QDM methods are effective in removing the systematic model biases. Moreover, for future changes, the QDM preserves the temperature change signals well, in both magnitude and spatial distribution, whereas the QM artificially modifies the change signal by decreasing the warming and modifying the patterns of change. In addition, for precipitation, both methods preserve the change signals well but they produce a greater magnitude of the projected increase, especially the QDM (Tong et al. 2021).

The ability of the selected RCMs and bias correction methods to simulate climate variables at a particular location can be evaluated using various statistical parameters techniques. There is no single best evaluation technique for the performance model. Rather, it is the combined use of many techniques and measures that provides a comprehensive overview of model performance (Geleta & Gobosho 2018; Zhang et al. 2018). Variables from the selected climate model are evaluated against recorded variables using some of the statistical measures recommended by the World Meteorological Organization.

The performance of each bias correction method is assessed on the basis of the capacity to generate precipitation and temperature (Luo et al. 2018).

In this study, the performances of the selected climate models and bias correction methods were evaluated by using four statistical indicators: Pearson correlation coefficient (R), root mean square error (RMSE), Nash–Sutcliffe efficiency (NSE), and percent bias (PBIAS) (Dibaba et al. 2019; Daniel & Abate 2022).

Rainfall, maximum and minimum temperatures are the key climate variables in Ethiopia, particularly in Deme and Gelana watersheds, where the socioeconomic activities of these regions are principally based on rain-fed agriculture.

In this study, the main focus was to evaluate the performance of three RCMs regarding their capability in simulating precipitation and temperatures. The ability of the selected RCMs to simulate the daily precipitation, minimum and maximum temperatures of the watersheds was statistically evaluated in the base period (1988–2019). The four statistical approaches such as Pearson correlation coefficient, RMSE, NSE, and PBIAS were used. The climate models' simulations were assessed relative to observed climate variables data.

The statistical parameters indicate that selected RCMs can reproduce the observed precipitation, maximum and minimum temperatures with a considerable performance before the bias correction, as shown in Table 7. Moreover, the outputs indicate that RACMO22T, RCA4, and HIRHAM5 models can simulate with slight overestimation and underestimation before the bias correction methods were applied in both watersheds.

Furthermore, the RCM performances were also evaluated after bias correction was applied to the climate variables such as precipitation, maximum temperature, and minimum temperature, as summarized in Tables 8 and 9.

Pearson correlation coefficient (R) and NSE are the best fit when they approach 1, while RMSE and PBIAS are a better fit when close to zero. Now also, the results indicate that the selected RCMs have the ability to reproduce observed daily precipitation, maximum and minimum temperatures of both watersheds.

Therefore, all the selected statistical parameters such as R, RMSE, NSE, and PBIAS by considering the mean monthly simulation of all stations before and after bias correction indicate that the RACMO22T and HIRHAM5 climate models were performed better than the RCA4 model in reproducing daily precipitation, maximum and minimum temperature in the Deme and Gelana watersheds. A similar result was done by concluding that the RACMO22T climate model simulating rainfall over most stations was better than the other CORDEX RCMs in the upper Blue Nile basin in Ethiopia (Dibaba et al. 2019). Similar results were concluded by Geleta & Gobosho (2018) in the Finchaa watershed in Ethiopia. In addition, RACMO22T shows better adjustment at the simulation of both precipitation and temperatures despite their significant bias (Tumsa 2021).

In addition, the Gelana watershed is located in the southeastern part of Rift Valley basin, while the Deme watershed is located in southern part of Ethiopia, Omo Gibe basin. Thus, all RCMs performed better in the Deme watershed than in the Gelana watershed. This shows the RCM simulation depends on the regions, and watersheds. Generally, the models do not have similar performance in all watersheds and regions. In this case, selecting the appropriate climate model gives better outputs. The related study concluded that the performance of RCMs was different at different watersheds along with the specified locations and topographies (Tumsa 2021).

Therefore, to pave the way for supplementary study and understanding of the effect of climate change on water resources and hydrology, it is very important to incorporate and select the watershed-location-based RCMs by considering their performance in simulating daily precipitation and temperature.

The pre-processing of the RCM outputs such as removing uncertainty (bias) should be the first step of climate change effect studies to further estimate and predict the future climate change consequences and suggest expectation mechanisms throughout the global community. The bias corrections of climate models simulations were carried out by the LS, PT, EQM, and DM for adjusting the precipitation, whereas LS, VARI, EQM, and DM were used for adjusting the maximum and minimum temperatures.

The parallel conclusions were also done by Teutschbein & Seibert (2012) and Luo et al. (2018). They were also supported by Geleta & Gobosho (2018) in the Finchaa watershed in Ethiopia. Moreover, Dibaba et al. (2020) concluded that there are differences between the correction methods, but the DM reproduces very well for temperature bias correction.

Generally, selecting a suitable bias correction method is important to provide reliable inputs for the evaluation of climate change impact. Referring to this evidence, selecting bias correction methods is critical for fixing the variation (over and under estimation) between RCMs simulated and observed climate variables in order not to delude the hydrological and climate model outputs for policymakers to better mitigate climate change consequences. Really, the selection and application of appropriate bias correction methods to semiarid climate regions is the most challenging issue due to the behavior of the rainfall in such regions being very scarce, erratic, and infrequent in nature (Mendez et al. 2020). Thus, most of the Gelana and Deme watersheds are semiarid regions, so that the using the bias correction method in these watersheds requires evidence-, technical- and study-based selection.

This paper compares the performances of four precipitation and temperature bias correction methods and three RCM performances in simulations of climate variables in the Deme and Gelana watersheds, Ethiopia. The performance of the climate models and bias correction methods were evaluated using Pearson correlation coefficient (R), RMSE, NSE, and PBIAS. There is no single best performance evaluation technique. Thus, the combined use of many techniques provides a complete overview of model and bias correction methods’ performance.

The raw RCM outputs are very biased, and their direct use in the analysis amplifies the error in the results. All RCMs have the ability to simulate and reproduce the observed precipitation and temperatures, with some overestimation and underestimation. Similarly, the four bias correction methods have the potential to condense the gaps between simulated and observed climate variables in both watersheds.

The performance of different precipitation and temperatures-corrected methods was shown to be good. However, the LS and PT show a poor performance of the corrected precipitation in both watersheds. Correspondingly, the LS and variance scaling (VS) show a poor performance of the corrected temperatures relative to other methods. Generally, the EQM method performed best in reproducing precipitation and maximum temperature, while the DM method performed extremely well for minimum temperature correction.

Overall, this paper accentuates the necessities of using several bias correction methods and RCMs to crosscheck their performances before selecting the methods and models for further climate and hydrological analysis. Moreover, there is no single best climate model, rather the combined use of many models provides a comprehensive overview of models' simulations. Climate modeling by using multi-RCMs helps to minimize uncertainties comparatively to the use of single RCMs. In addition, using an ensemble mean of climate models is preferable in the analysis of rainfall and temperatures.

Furthermore, the RCM simulation is dependent on the regions, it performs well in some regions and poorly in other regions. Besides, the selection of suitable bias correction methods for semiarid climate regions is the most challenging issue due to the behavior of rainfall.

Therefore, to pave the way for supplementary study and understanding of the effect of climate change on water resources and hydrology, it is very significant to incorporate and select the watershed-location-based RCMs by considering their performance in simulating daily precipitation and temperatures. In future, this study should be extended by considering numerous bias correction methods by combining the use of many RCMs for climate change analysis, since it reduces the uncertainties in studies, outputs. Moreover, the selection of RCMs is highly dependent on the regions and climate patterns, therefore climate change studies in different watersheds or basins and countries must check their performance before directly using the different models.

The author would like to acknowledge the National Meteorological Agency of Ethiopia for providing the meteorological data.

All relevant data are included in the paper.

The authors declare there is no conflict.