Water-Energy Nexus 7 (2024) 87–102

Contents lists available at ScienceDirect

Water-Energy Nexus
CHINESE ROOTS
GLOBAL IMPACT
journal homepage: www.keaipublishing.com/en/journals/water-energy-nexus/

Research Paper

Validation of CORDEX Regional climate models to simulate climate

trends and variability across various agro-ecological zones of North
Shewa, Ethiopia
Biruk Getaneh Ayele a,⇑, Ayele Desalegn Woldemariam a, Hailu Kendie Addis b
a
Department of soil and water management, Debre Brihan Agricultural Research Center, Po.BOX 112, Debre Brihan, Ethiopia
b
Amhara Regional Agricultural Research Institute, Po.Box. 527, Bahir Dar, Ethiopia

a r t i c l e i n f o a b s t r a c t

Article history: Climate variability significantly affects the overall agricultural production and productivity over space
Received 14 September 2023 and time. To quantify the effect of climate variability on natural systems on Earth in the future, the
Revised 28 November 2023 use of regional climate models (RCMs) as a tool is mandatory. Thus, the use of well-validated RCM pro-
Accepted 2 January 2024
vides efficient climate data; and predicts future climate variability over agroecological zones (AEZs).
Available online 26 January 2024
Eleven RCMs from CORDEX-Ethiopia were evaluated to simulate maximum temperature (Tmax), mini-
mum temperature (Tmin) and rainfall variability over space and time, to validate their performance and
Keywords:
to select the best-fit models. The temporal variability Tmax and rainfall was smaller with a CV of 10 %,
Agro-ecological zones
CORDEX-Ethiopia
indicating uniform distribution over stations. However, CV for Tmin over stations shows satisfactory dis-
Regional Climate Model tributions. Most of the evaluated RCMs have a better capability to capture the spatial variability of rain-
Validation fall,Tmax, and Tmin in some stations. However, some of the RCMs can handle the spatial variability of Tmax
while others are not. None of the RCMs can handle the spatial variability of Tmin in Debre Birhan, Majete,
and Kewat AEZs. In general, most RCMs can capture the spatial and temporal variability of rainfall and
temperature. It might be better to use the ensemble average of the output from these RCMs for precise
estimation of data than specific RCMs. Therefore, this work could suggest the use of high-resolution RCMs
for the projection of temperature and rainfall over different AEZs.
Ó 2023 The Authors. Production and hosting by Elsevier B.V. on behalf of KeAi Communications Co., Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-
nd/4.0/).

1. Introduction planning agricultural management, and farm decisions is quite

important.
Climate change is a major concern of this world as it affects the Temperature and rainfall are the major climate parameters that
overall systems on the earth both with spatial and temporal pat- have a great impact on overall ecosystems (Li et al., 2019). There is
terns (Gebrehiwot and Veen, 2013). However, its effect is more sig- a consensus among scholars that rainfall variability is increasing
nificantly observed in developing countries than in developed and trends decreasing temporarily, and increasing of temperature
countries (Asfaw et al., 2021). It has considerably affected African trends in many parts of the of world, and the same is true in Ethio-
agriculture and it might increase exponentially in the future pia. These would have a strong implication on crop production and
(Thornton, 2010). Because developed countries have established be a challenge for adaptation planning and management
tools to reduce the existing and emerging impact of climate (Kassahun, 2011). The use of RCMs is important to generate histor-
change. Nonetheless, developing nations hardly formulate such a ical and future climate data and to assess the impact of climate
kind of tool in local conditions. Ethiopia has diverse agroecological change on agriculture (Wilks., 1999). Thus, they are a standard
zones (AEZs) with a variety of micro-climatic, soil, and topographic component of decision support systems in agriculture (DSSAT),
conditions (Astawsegn, 2014). But, the availability of climatic data environmental management, and hydrology (Munangv., 2010).
at different local scales is insignificant. Therefore, the use of RCMs Moreover, the projection of temperature and rainfall with precise
to generate climatic information for simulating crop growth, and validated RCMs is needed to set up adaptation mechanisms
on climate change.
Validation of RCMs about real observed data provides the best-
⇑ Corresponding author. fit model for the given locality. According to the (IPCC, 2007), the
E-mail address: [email protected] (B. Getaneh Ayele). Intergovernmental Panel on Climate Change’s (IPCC) fourth report

https://doi.org/10.1016/j.wen.2024.01.001
2588-9125/Ó 2023 The Authors. Production and hosting by Elsevier B.V. on behalf of KeAi Communications Co., Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 1. Geographical location of the study area.

Fig. 2. Agro-ecology based classification of North Shewa zone.

confirmed that some extremes are changing continuously due to would be used to project future trends of temperature and rainfall
human interventions. According to the IPCC 5 report, the variabil- (Gutowski et al., 2016).
ity of climate conditions would be increased exponentially (Birch, The simulation of rainfall and temperature from CORDEX out-
2014). Regional downscaling of GCMs is used to provide climate put is assessed by using the taylor diagram (Taylor, 2001), root
information at the smaller scales (Flato et al., 2013). There is lim- mean square error (RMSE), correlation coefficient (R), and coeffi-
ited information in the literature regarding AEZ-based validation cient of determination (R2). Based on previous studies, RCMs are
of RCMs in North Shewa. Since the launch of the Coordinated well reproduce rainfall and temperature in some regions but not
Regional Downscaling Experiment (CORDEX) Africa programme, a captured in other areas of Ethiopia. For example, the use of an
few studies have evaluated the RCMs’ performance over parts of ensemble average of RCMs with bias correction is important to
Ethiopia (Alemu and Sisay, 2021). Now a time, the output from capture the variability (Kim et al., 2014). Hence, to select the best
GCMs is of insufficient spatial and temporal resolution (Giorgi RCMs for a specific location, evaluating their performance is neces-
et al., 2009). In this study, about 11 RCMs from CMIP 5, CORDEX sary. Therefore, this proposal was initiated to 1) describe the tem-
for Ethiopia, were employed. The RCMs which have high precision poral and spatial distribution of temperature and rainfall in the

88
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Table 1 three AEZs 2) validate the performance of RCMs and select the
List of regional climate model and their drived global climate model. best-fit RCM.
No. Name of institutions GCM name RCM name
1 Canadian Centre for Climate CCCma-CanESM2 CRCM4 2. Materials and methods
Modeling and Analysis, Canada
2 Canadian Centre for Climate CCCma-CanESM2 RCA4 2.1. Description of the study areas
Modeling and Analysis, Canada
3 The Atmosphere and Ocean MIROC-MIROC5 RCA4 The study was conducted in the three AEZs of North Shewa, one
Research Institute (The University of of the administrative zones in Amhara Regional State, Ethiopia. The
Tokyo), National Institute for study area includes two metrological stations from each AEZs,
Environmental Studies, and Japan
Agency for Marine-Earth Science
Kewat (KW) and Majete (MJ) stations, Alem Ketema (AK) and Arerti
and Technology, Japan (AR), and Debre Birhan (DB) and Mehal Meda (MM) stations from
low lands, mid-altitude, and highlands respectively. The geograph-
4 Bjerknes Centre for Climate NCC-NorESM1-M RCA4
Research, Norwegian Meteorological ical locations of the study area is presented in (Fig. 1).
Institute, Norway
5 Max Planck Institute for M
MPI
ESM
LR RCA4 2.2. Source of observed and modeled data
Meteorology, Germany
6 Max Planck Institute for MPI- CFCM5 The observed maximum and minimum temperature, and rain-
Meteorology, Germany M
MPI
ESM
LR fall data (1990–2020) were obtained from National Metrological
7 Centre National de Recherches CNRM-CERFACS- RCA4 Service Agency (NMSA). Before using the data, data quality man-
Météorologiques, Centre Européen CNRM-CM5 agement like data arrangement, outlier detection, and handling
de Recherche et de Formation
of missing data was performed. Outliers were detected by using
Avancée en Calcul Scientifique
France
an interquartile range test on an Excel sheet treated as missing val-
ues and filled by normal ratio methods. Modeled monthly observa-
8 Geophysical Fluid Dynamics NOAA-GFDL- RCA4
Laboratory, USA GFDL-ESM2M
tional and reanalysis of temperature and precipitation data were
retrieved from CORDEX- Ethiopia, which is a CMIP 5 product with
9 Irish Centre for High-End Computing ICHEC-EC-EARTH HIRHAM5
Ireland European Centre Earth global
the same period as the actual observed data. A statistical compar-
climate model system ison between observed and modeled data is a major part of overall
validation but provides valuable information on the behavior of the
10 Irish Centre for High-End Computing ICHEC-EC-EARTH RACMO22T
Ireland European Centre Earth global data.
climate model system
11 Irish Centre for High-End Computing ICHEC-EC-EARTH RCA4 2.3. Area analysis on agro-ecological zones
Ireland European Centre Earth global
climate model system Agroecology can be classified based on altitude and rainfall pat-
terns. Fortunately, Ethiopia exhibits all AEZs and the same is true
for the Amhara region. AEZ-based classification of the area is the

Table 2
Temporal variability of temperatures and rainfall over the stations.

Stations Parameters Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Ave RME CV
DB Tmax 20 21 21 21 22 22 19 18 19 19 19 19 20 1 6
Tmin 5 6 8 9 8 8 9 9 7 4 4 4 7 2 32
RF 9 12 33 36 32 41 239 316 64 16 4 3 804 101 13
MM Tmax 19 19 19 19 19 20 18 17 17 17 17 18 18 1 6
Tmin 6 7 8 8 9 9 8 8 8 6 5 5 7 1 19
RF 16 23 55 45 40 41 268 252 74 24 7 5 850 91 11
AK Tmax 26 27 28 28 28 27 22 21 23 25 25 25 25 2 10
Tmin 13 14 14 15 16 15 12 12 12 12 12 12 13 1 10
RF 7 19 46 48 55 70 280 308 141 24 10 9 1016 105 10
AR Tmax 27 28 29 29 30 29 25 26 26 26 26 25 27 2 7
Tmin 9 11 13 14 16 15 14 15 14 13 11 8 13 3 21
RF 10 29 51 62 32 76 232 233 94 42 15 9 886 79 9
KW Tmax 29 30 32 33 34 36 33 31 31 31 30 29 31 2 7
Tmin 10 13 15 15 17 18 18 14 12 12 11 11 14 3 21
RF 28 48 58 86 81 24 144 182 84 39 14 48 836 50 6
MJ Tmax 26 27 29 30 31 33 31 29 27 27 25 24 28 3 10
Tmin 12 13 15 16 17 18 17 16 16 14 12 12 15 2 15
RF 31 40 72 99 64 25 221 297 104 45 28 23 1049 86 8

DB-Debre birhan, MM-Mehal meda, AK-Alem Ketema, AR-Arerti, KW-Kewat, MJ-Majete.

89
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 3. The spatial map of rainfall extremes distribution across locations.

spatial grouping of the landscape into areas of similar agricultural between zero and one, with ‘‘000 indicating no agreement and ‘‘1”
and ecological characteristics (Hurni., 1998). As coated by Hurni indicating a perfect agreement. It powerfully expresses the mod-
(1998), Dove (1890) reported major agricultural zones in Northern eled data with the observed data based on R, RME, and d. The
Ethiopia more precisely as being ‘low land’ at altitude below RME, CV, and d value was determined by the equation ((1)–(4))
1800 m.a.s.l, ‘midaltitude’ for altitude between 1800 m.a.s.l to
 X 0:5
2400 m.a.s.l, and ‘high land’ for areas above 2400 m.a.s.l. In this 1 n 2
study, the areal classification of North Shewa was done by using RME ¼ ð pi
O i Þ ð1Þ
n i¼1
Arc GIS software based on the above traditional altitudinal classifi-
cation methods. sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P
1 ðpi
Oi Þ2
CV ¼   100 ð2Þ
2.4. Validation criteria on regional climate models ^
OiA
N

The generated data from RCMs are compared with the observed P
ðP i
Oi Þ2
data with statistical indicators like: - root mean square error d¼1
P ð3Þ
^
j þ jOi
Oi A
ðjPi
Oi A ^
jÞ2
(RME), coefficient variations (CV), correlation coefficient (R), coef-
ficient of determination (R2), and index of agreement (d). A lower
RME indicates that better performance of the model and vice versa. Pn


i¼1 ðPi
pi Þ  ðOi
Oi Þ
The CV values are considered as excellent if smaller than 10%, good ffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r ¼ sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð4Þ
P n

2 P1

2
if between 10 and 20%, fair if between 20 and 30%, and poor if lar- ðPi
pi Þ  ðOi
Oi Þ
ger than 30% (Mekonen and Berlie, 2019). The higher the CV value i¼1 i¼1

indicates more variability of the data and vice versa. The index of
agreement (d) was proposed to measure the degree to which the Whereas: pi is the model predicted value, Oi is the observed individ-
observed data come up to the predicted data (Willmott, 1982). It ual value, n is the number of observations, Oiis the mean of the
is a measure of the predictive ability of the models. It ranges observed value, and Pi is the mean of the predicted value.

90
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 3 (continued)

3. Result and discussions 1800 m, 4359 km2 (28%) being as weyna Dega (midaltitude) for alti-
tudes between 1800 and 2400, and 5890 km2 (37.8%) considered as
3.1. Area classification of North Shewa based on agro-ecology Dega (high land) for above altitudes of 2400 m (Fig. 2). This classifica-
tion is crucially important to understand which RCMs would well
North Shewa covered a total area of 15,568 Km2, of which is reproduce on which agroecology, which crop potentially grown and
5319 km2 (34.2%) categorized as Kolla (low land) for altitudes below guide to undertake measures on the existing and emerging situations.

91
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 4. The spatial map of maximum temperature distribution across locations.

3.2. Temporal variability of temperature and rainfall The RCMs, which have the closest values to the observed, indi-
cate the better capability of the model to interpolate the rainfall
As presented in Table 1, the temporal variability and distribu- distributions over the area. For instance, in mid-AEZ (AR and AK)
tion of rainfall under the three AEZs of North Shewa show uniform all RCMs can capture the spatial variability of rainfall (Fig. 3). Of
with a smaller CV. This shows a sound effect on agricultural pro- the all RCMs, ICECH-HIRHAM, ICHEC-RACMand MIROC5 were well
duction and productivity by optimizing the soil moisture and tem- represent the rainfall pattern of MM and CCCma-CRCM4, ICHEC-
perature. The monthly Tmax and rainfall were in the acceptable RCA4, CRCM5, LRRCA4, MIROC5 were well reproduce the rainfall
threshold value within all stations. However, rainfall is highly vari- in DB station. While other RCMs either under estimate or over esti-
able in the lowland than highland AEZ, Tmin shows a slight variabil- mate the rainfall extremes. Similarly, for Majete (CCCma-RCA4 and
ity and satisfactory distributions in all stations except AK (Table 2). CCCma-CRCM4) and KW stations (CNRM-CM5, HIRHAM, ICHEC-
RACM, MIROC5) well reproduce the rainfall patterns respectively.
3.3. Spatial analysis of monthly rainfall and temperatures According to (Alemu and Sisay, 2021), RCA4 was well performed
in both simulating rainfall and air temperature over the Wabi She-
The spatial distribution of the annual rainfall, Tmax, and Tmin was bele basin. However, in this study, RCMs have variable perfor-
determined by inverse distance weighted (IDW) interpolation mance over AEZs of North Shewa. The CV value for observed
methods. Since North Shewa has a diverse climate with tempera- rainfall data mapping was smaller (5.9%), indicating smaller rain-
ture and rainfall variability, the ability of RCMs to handle the spa- fall variability in North Shewa. For instance, CCCma-CRC4, ICHEC-
tial variability in contrast to the observed value was examined. The RCA4, and MIROC5 models showed relatively close agreement with
developed spatial map from average rainfall from different RCMs the observed having a CV of 6.6%, 7.2%, and 7.6%, respectively.
indicated below shows that most of the RCMs’ performance was The spatial variability of Tmax (Fig. 4) for the observed data was
moderate to good agreement with the observed spatial distribu- in the acceptable range (12%). The CV values for models ICECH-
tion. The RCMs’ rainfall and air temperature data were downloaded HIRHAM and CCCma-CRCM4 were 11.1% and 8.9%, which was sim-
from a public website, with the required data in Arc GIS. The spa- ilar to the observed spatial distribution of the Tmax value. These
tial map for rainfall in Fig. 3 indicates that the spatially interpo- models can handle the spatial Tmax variability. However, for other
lated value of distributed rainfall from the gauging stations of RCMs the observed CV values were smaller than the observed, indi-
each RCM across locations was quite different from the observed cating the inability of the models to catch up with the variability in
values. Tmax. According to Dong et al. (2018), of the evaluated five RCMs

92
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 4 (continued)

Fig. 4 (continued)

none of them can capture the spatial patterns of the temperature Majete station only CCCma- CRCM4 well reproduce the patterns
and precipitation, either overestimated or underestimated. In gen- of Tmax.
eral, the spatial distribution of mean monthly Tmax for most of The spatial variability of observed Tmin (Fig. 5) across locations
RCMs looks less variable across locations as the CV values are smal- was 20.5%, which indicates higher variability and uneven distribu-
ler. However, for mid-AEZ all RCMs can capture the spatial pattern tions. The CV values for all RCMs were nearly 11%, which is smaller
of Tmax except CCCma-CRCM4, ICHEC-HIRHAM, and MPI-CRCM5 than the observed value by half. This shows no RCMs are capable of
(Fig. 4). For highland AEZ, all RCMs capture the spatial pattern of reproducing the Tmin across locations in the observed spatial distri-
Tmax except ICHEC-HIRHAM and MPI-CRCM5. However, for Kewat butions. This primarily influences all agricultural activities either
stations all RCMs do not capture the spatial patterns, and for positively or negatively. As shown in Fig. 5, all RCMs can handle

93
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 5. The spatial map of minimum temperature distribution across locations.

the spatial patterns of Tmin for all stations in mid-AEZs. For Debre value for Tmax observed for all RCMs under the stations of MM,
Birhan station all RCMs can capture the spatial pattern of Tmin. KW, and MJ was relatively larger. Similarly, the MIROC 5 model
But, for the stations of Mehal Meda, Majete, and Kewat neither of for Tmin recorded the smaller RME value in AK station. All climate
the RCMs is able to handle the pattern. In general, all RCMs are models showed smaller RME values in Tmin for AK and larger
simulated Tmax and Tmin variably over locations. However, Kim RME values for MJ stations. The RME value obtained for Tmin in
et al., (2014) found that RCMs well reproduce Tmax than Tmin in mid altitudes is relatively smaller than that of the highlands and
Ethiopian highlands. lowlands for most of the climatic RCMs. This indicates that most
RCMs performed well in mid-altitudes (AK and AR) for the simula-
3.4. Validation metrics of regional climate models tion of Tmax and Tmin. Whereas for rainfall, the minimum RME value
was observed in AR and KW stations nearly for all models. How-
The fitness of each RCMs is determined based on how the root ever, the value of RME in the AK station was relatively larger than
mean square error (RME) value is smaller and indicates close esti- others in the case of all models. The RME value for DB and MM in
mates to the historical observed values. The minimum RME for simulating rainfall was relatively larger in most of the RCM. The
Tmax was observed for CNRM-CM5 and the maximum RME was RME value of rainfall for highlands is larger than mid and low lands
observed in ICHEC-EARTH1 at DB station. Almost all of the RCMs in most of the RCMs. That means, there is greater rainfall variabil-
account for a smaller RME value for Tmax in DB and AK. The RME ity in the highland areas. The mid altitude agro-ecology relatively

94
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 5 (continued)

shows a high RME value for rainfall and low RME for temperature and all models reproduced Tmin poorly in the highland stations
in all models (Fig. 6). (MM and DB (Fig. 7).
The CV for Tmax in AR station was found smaller (values close to The CV for rainfall in AR, MJ, and KW stations shown an accept-
5%) of all stations and followed by AK stations (values between 5% able range of 8% to 10% nearly for all RCM. This shows smaller vari-
and 10%). This indicates the variation between the observed Tmax ations of the RCMs produced output rainfall and the observed
and the produced output from models is smaller, thereby nearly rainfall data, thereby the models can predict the future rainfall
all models are well performed in generating historical Tmax for events. However, there was no significant variability among mod-
areas that have no meteorological station. Nevertheless, the differ- els output in simulating rainfall. For stations of AK, MM, and DB the
ences amongst the models in simulating Tmax were insignificant. CV values were also in the acceptable ranges (10% to 13%), and pro-
Similarly, nearly for all models in KW and MJ stations, the results vided a smaller variation in rainfall between the observed and
show relatively moderate CV values for Tmax, which is between model output.
10% and 15%. This indicates the models can predict the Tmax for The observed Tmax and the produced Tmax from all models for
lowland areas. DB station shown smaller CV values (under 10%) stations of AK and AR were in good agreement, as the index of
for Tmax in the case of all models except ICHEC-EARTH1 and MPI- agreement was in the range of 85% and 90%. However, for sta-
LR2 models (above 30%). All models predicted Tmax poorly in the tions of MM, KW, and MJ the index of agreement was smaller.
highland stations (MM stations and DB for some RCMs). From this For DB station, the index of agreement value was satisfactory
output, the produced Tmax values were not significantly different (35–40%) for most RCMs except CNRM-CM5 model. This indi-
from model to model; but varied from station to station. All RCMs cates for mid altitude stations nearly most of the models can
are well reproduced Tmax in mid-altitude stations (AK and AR) as produce Tmax efficiently, and for lowlands and highlands, the
the CV values are smaller (Fig. 7). RCMs poorly reproduce Tmax, as the index of agreement value
The CV values of Tmin for all models are smallest in the AK and is smaller. In line with this (Worku et al., 2018) reflected that
AR stations, which is in the acceptable range of 15% to 20%. The the RCMs show bias at higher elevations, but work well for
maximum CV values for Tmin (above 30%) were observed in MM low-elevation regions. The higher index of agreement for repro-
and followed by DB stations. The CV values for low-altitude sta- ducing Tmin was observed in AR station (80–85%), and followed
tions (KW and MJ) in reproducing Tmin were higher (23–30%). by DB stations (75–80%). The index of agreement for KW station
The variation between the observed values and the produced val- in most of the RCMs was moderate (45% to 50%). Unlikely, for
ues are significantly different, hence the models are not well- stations of AK, MM, and MJ, the index of agreement was smaller
performing for prediction purposes especially for lowlands and (30% to 40%), indicating the satisfactory agreements of most
highland stations. The obtained CV values in the case of Tmin were RCMs to the observed records.
relatively higher than those of Tmax in all stations for all models. However, there was an insignificant difference among the
This indicates the variability of Tmin is higher than Tmax in all sta- RCMs. The results are in line with Kumar, et al. (2008), who con-
tions. However, all models are well reproduced Tmin and Tmax in cluded that better agreements of observed and simulated monthly
mid-altitude stations (AK and AR) as the CV values are smaller climate data in the majority of the weather parameters. A majority

95
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 6. The Root mean square value of RCMs for Tmax, Tmin, and rainfall.

of RCMs can simulate the most reliable annual and seasonal rain- agreement with observed Tmin values (Fig. 8). The index of agree-
fall and air temperature estimations (Worku et al., 2018). In gen- ment values of all models to reproduce monthly rainfall was close
eral, nearly most of the RCMs for a station of AR showed the best to 100%, which indicates there was a highest agreement between

96
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 7. The coefficient of variations of RCMs for simulating Tmax, Tmin, and Rainfall.

RCMs and the observed monthly rainfall. Even though there were The higher correlation coefficient (R) values were observed in
insignificance variations among RCMs, they were efficient for stations of AK and AR (90–100%) nearly for most of the RCMs.
reproducing the rainfall in all stations. For the DB station, the observed R-values were also highest in most

97
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 8. The index of agreement of RCMs in simulating Tmax, Tmin, and rainfall.

98
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 9. The correlation coefficient of RCMs in simulating Tmax, Tmin, and rainfall.

99
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 10. Taylor diagram of RCMs in simulating Tmax, Tmin, and rainfall in highlands area.

Fig. 11. Taylor diagram of RCMs in simulating Tmax, Tmin, and rainfall at mid-altitudes.

of the RCM (above 90%) except for ICHEC-HIRHAM and MPILR- observed Tmin, Tmax, and rainfall over most stations (Philbert
CFCM5 models (R-values of 57% and 47% respectively). This indi- et al., 2016).
cates a strong relationship between observed Tmax and RCMesti- The R-values observed in stations of DB, MM, AK, and AR were
mated Tmax values. The R-value for stations of MM, KW, and MJ near 100% (above 95%) for most of the RCM in reproducing
was in the acceptable value (70% to 90%) for most of the RCM monthly rainfall except MPILR-CFCM5 at AK station (70%). This
except CCCma-CRCM4at KW stations (R-value of (
27%)). In gen- shows a positive correlation of most RCM with the observed rain-
eral, most of the RCMs have a positive correlation with the fall data. As per the investigation of Abiy et al., (2021), the ensem-
observed Tmax in most of the stations, especially in AK. ble of RCMs fits with the observed data than single RCMs. Similarly,
The value of R for stations DB and MM as most of RCM produced for stations of KW and MJ, the observed R-value was in the range of
Tmin was nearly 100% (above 95%). Similarly, for stations MJ, KW, 75% to 95% for most of RCM, which is in the acceptable range. There
and AR, the R-value was observed as higher (85–95%) for most of was no significant difference in RCMs for most stations in repro-
the RCM, indicating a strong positive relation with the observed ducing the monthly rainfall (Fig. 9).
Tmin. Nevertheless, for a station of AK the observed R-value of all Taylor diagram indicates the RCMs’ ability to simulate the pat-
models to produce Tmin was relatively smaller, moderate values tern and amplitude of the observed inter-annual variation of tem-
(50% to 60%). However, the degree of the relation of RCM with perature and rainfall. As the Taylor diagram showed (Figs. 10–12),
the observed Tmin does not reflect a significant difference. The sim- for each parameter the distance of all points (RCMs) from the red
ulated Tmin, Tmax, and rainfall by all RCMs compared well with point (observed value) was nearly equal. That means, there is no

100
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Fig. 12. Taylor diagram of RCMs in simulating Tmax, Tmin, and rainfall at lowlands.

clear picture to identify which RCMs exhibit the shortest distance concluded that from the above analysis, most of the RCMs simulate
(smallest standard deviation) and the highest R-value compared Tmin more than Tmax over stations. All parameters simulated by
with the observed value. most of theRCMs are positively correlated with the observed val-
For a station of DB, the SD and R values were smaller for some of ues. This result is in close agreement with that of Philbert, et al.
the RCMs and larger for some of the RCMs. MPI-CRCM5 simulates (2016), ofound a better simulation of Tmin than Tmax; and RCMs
Tmax differently from others with a SD value between 0.5 and 1.0 °C simulate Tmin and Tmax better than rainfall in most stations they
and exhibits an R-value of 0.57. In the case of MM, the RACM model were studied. They suggested the use of CORDEX RCMs in repre-
shows relatively the shortest distance from the red point and has senting the variability of Tmax and Tmin over the Tanzania region.
the highest R-value. For other RCMs, there is no clear visualization Similarly, the majority of RCMs are highly correlated with observed
to select which models are better. or MM and D B, the SD value was and ERA-Interim data in most areas of South Africa and they ade-
found between 0.5 & 0.6 Co and the R value of nearly one., general, quately capture precipitation (Mxolisi, et al., 2015).
most of the RCMs are outperformed in simulating Tmin for high- The index of agreement of most of the RCMs is found as sound
lands areasrather. Unlike Tmax and Tmin, for the simulation of rain- on stations of AK and AR for simulating Tmax; DB and AR for simu-
fall, the Taylor diagram powerfully explained the performances of lating Tmin and all stations in the simulation of Rainfall. In line with
RCMs. In DB and MM stations, ICHEC-HIRHAM relatively exhibits these results Ifeka, et al., (2021), evaluated two RCMs and found
the shortest distance (smallest SD of 20 mm) and the largest R- both CCCma-CanRCM4 and MPILR-CFCM5 agrees with the
value (0.99), followed by the RCMs of MPI-CRCM5, CNRM-CM5, observed data. Based on the Taylor diagram, most of the RCMs
ICHEC-RACM, ICHEC-RCA4. For other models, the SD value of 20 are outperformed to simulate Tmax and Tmin. However, there is a
to 40 mm, and R-value between 0.95 to 0.99 was observed. slight difference among RCMs in simulating the seasonal changes
For AK and AR stations, the SD was found between 0.8 & 1 °C of rainfall. Hussen et al. (2013), support this as some of the evalu-
and 1.3 to 1.5 °C respectively; and the R value was in the range ated RCMs have been Outperformed concerning the observed GPCC
of 0.9 to 0.92. As shown in Fig. 11, most of the RCMs exhibit nearly and others have not yet; however, the multimodal ensemble is
equal SD and R-value for the simulation of Tmax. For instance, for found generally closer to the GPCC than individual models. Accord-
AK and AR, SD ranges between 1 & 1.5 Co. But, the R-value for AK ing to (Tamene and Chala, 2021), all four RCMs simulated the sea-
is in the range of 0.5 and 0.6, for AR between 0.8 & 0.9. On the other sonal rainfall, and to some extent biased around a warm climate
hand, at the station of AK, ICHEC-HIRHAM, ICHEC-RACM, ICHEC- zone in simulating Tmax and Tmin.
RCA4, CNRM-CM5, MPILR-RCA4 shows an SD of 20 mm and R- This result is supported by Kiya et al., (2020), who investigated
value of 0.99. Other RCMs account for variable SD and relatively that all models were found best in capturing the observed rainfall,
lower R-value (Fig. 11). At AR, most of the RCMs show an SD under but no individual model is best in all performance measures. The
20 mm and an R-value of above 0.95. ensemble of models is better at simulating the characteristics of
most of the RCMs can capture the temporal variability of Tmax rainfall than the individual models and is best in all performance
and Tmin For MJ, the SD of Tmax ranges between 1.8 and 2 °C for measures. The RCMs from the CORDEX indicated a better perfor-
most of the RCMs except CCCa-CRCM4. For KW stations, nearly mance in reproducing the rainfall characteristics over the Mbarali
all RCMs are unable to capture rainfall variability (all RCMs points River catchment (Mutayoba and Kashaigili, 2017). These RCMs
fall out of the taylor diagram), indicating to high SD and R value. In were recommended for use for climate change impact and adapta-
MJ stations, CHEC-HIRHAM, ICHEC-RACM, ICHEC-RCA4, CNRM- tion studies in water resources management in the selected cities.
CM5, MPILR-RCA4, MPILR-CFCM5, NORESM1-M shows an SD of
20 mm and R-value of above 0.95. However, other RCMs exhibit 4. Conclusion and recommendation
an SD value of 40 to 60 mm and an R-value from 0.85 to 0.9. With
the Exception of the KW station, all RCM points fall out of the Tay- In this research agro-ecological classification, temporal and spa-
lor diagram, indicating the largest SD and R-value. tial variability, trends analysis of rainfall, maximum and minimum
The performance of RCMs in simulating Tmax, Tmin, and rainfall temperature, and validation of eleven RCMs under various climatic
over different stations is presented in (Figs. 10–12), and it can be agro-ecological zones of North Shewa were addressed. North
101
B. Getaneh Ayele, A.D. Woldemariam and H.K. Addis Water-Energy Nexus 7 (2024) 87–102

Shewa exhibits the three AEZ categories of 34.2% Kolla, 28% Weyna comparative analysis in NorthCentral Ethiopia (Woleka Sub-basin). Heliyon 7,
(4) e06761.
Dega, and 37.8% Dega. The temporal variability of observed Tmax
Astawsegn, Z., 2014. Effects of Climate Change and Variability on Rural Livelihoods
and rainfall is smaller with a CV of 10%, and shows uniform distri- and Responses: The Case of Soro Woreda, Hadiya Zone, SNNPR. MA Thesis in
butions over stations. Whereas, the CV value for Tmin over stations Climate Change and Adaptation. Addis Ababa University, Ethiopia.
shows satisfactory distributions. The spatial variability of rainfall is Birch, E., 2014. A Review of ‘‘Climate Change 2014: Impacts, Adaptation, and
Vulnerability” and ‘‘Climate Change: Mitigation of Climate Change”. J. Am. Plan.
significantly higher than that of Tmax and Tmin. Since agriculture is Assoc., 184–185
directly impacted by climate change (an increase in temperature DongLin Guo, J.-Q. S.-T., 2018. Evaluation of CORDEX regional climate models in
and rainfall variability), well-validated RCMs are needed. simulating temperature and precipitation over the Tibetan Plateau. Atmos.
Oceanic Sci. Lett., 219–227
The performance of the RCMs were found to be variable in the Flato, G., Marotzke, J., Abiodun, B., Braconnot, P., Chou, S.C., Collins, W., Cox, P.,
simulation of rainfall and temperature in all AEZs of North Shewa. Driouech, F., Emori, S., Eyring, V., Forest, C., 2013. In: Stocker, T.F., Qin, D.,
Most of the RCMs can capture the spatial variability of rainfall and Plattner, G.-K., Tignor, M., Allen, S.K., Boschung, J., Nauels, A., Xia, Y., Bex, V.,
Midgley, P.M. (Eds.), Evaluation of Climate Models. Climate Change 2013: The
temperature in mid-AEZs. However, some of the RCMs can slightly Physical Science Basis. Contribution of Working Group I to the Fifth Assessment
handle the spatial variability of Tmax and Tmin while others not. Report of the Intergovernmental Panel on Climate Change. Cambridge
None of the RCMs captured the spatial variability of Tmin in the University Press, Cambridge, United Kingdom and New York, NY, USA.
Gebrehiwot, T.G., Veen, A., 2013. Assessing the Evidence of Climate Variability in the
highland and lowland AEZs. This work highlights the use of high- Northern Part of Ethiopia. J Dev Agric Econ 5 (3), 104–119.
resolution RCMs for the projection of Tmax, Tmin, and rainfall across Giorgi, F., Jones, C., Asrar, G.R., 2009. Addressing climate information needs at the
locations and to develop strategic plans to alleviate the impacts of regional level: the CORDEX framework. World Meteorol. Organ. Bull. 58, 175–
183.
future climate change in all AEZs of North Shewa and other similar
Gutowski, W.J., Jr., Giorgi, F., Timbal, B., Frigon, A., Jacob, D., Kang, H.S., Krishnan, R.,
regions. Lee, B., Lennard, C., Nikulin, G., O’Rourke, E., 2016. WCRP Coordinated regional
downscaling experiment (CORDEX): a diagnostic MIP for CMIP6. Geosci. Model
Dev. (GMD) 9, 4087–4095.
5. Funding Hurni, H., 1998. Agroecological Belts of Ethiopia. Soil Conservation Research
Programme Ethiopia, Addis Abeba.
There are no funding options for publication. Ifeka, A.B.I., 2021. Assessment of projected temperature over west africa using
CORDEX regional climate models. J. Environ. Earth Sci. 11, 1–12.
IPCC, 2007. Climate Change: Synthesis report. Geneva, Switzerland, p: 104: Working
6. Authors’ contributions Group I, II, and III contributions to the Fourth Assessment Report.
Kassahun, A., 2011. The impact of climate variability on crop production in Ethiopia:
Which crop is more vulnerable to rainfall variability? EEA/EEPRY, Addis Ababa.
Biruk Getaneh Ayele did Conceptualization, Data curation, Kim, J., Waliser, D.E., Mattmann, C.A., Goodale, C.E., Hart, A.F., Zimdars, P.A.,
Writing – original draft, Writing – review & editing, Visualization, Crichton, D.J., Jones, C., Nikulin, G., Hewitson, B., Jack, C., 2014. Evaluation of the
Investigation, Formal analysis, Methodology. Ayele Desalegn CORDEX-Africa multi-RCM hindcast: systematic model errors. Climate
dynamics 42, 1189–1202.
Woldemariam did Conceptualization, Data curation, Writing – Kiya, K., Boja, M., Negese, R., 2020. Evaluation the performance of regional climate
review & editing, Visualization, Investigation, Formal analysis, models in simulating rainfall characteristics over upper awash sub-basin,
Methodology, Supervision, Resources and Hailu Kendie Addis did Ethiopia. Int. Res. J. Adv. Eng. Sci. 5 (1), 134–138.
Kumar, Choudhury, B., Sood, A., Jalofa, S. 2008. Evaluation of the Climgen Model to
Funding acquisition, Writing – review & editing, Visualization, generate weather parameters under different climatic situations in Punjab.
Supervision and Project administration. Li, Y., Li, Z., Zhang, Z., Chen, L., Kurkute, S., Scaff, L., & Pan, X. 2019. High-resolution
regional climate modeling and projection over western Canada using a weather
research forecasting model with a pseudo-global warming approach. 4635–
Declaration of competing interest 4659.
Mekonen, A.A., Berlie, A.B., 2019. Spatiotemporal variability and trends of rainfall
The authors declare no conflict of interest. and temperature in the northeastern highlands of Ethiopia. Model. Earth Syst.
Environ. 6, 285–300.
Munangv, T., R. M., 2010. Assessment of the ClimGen Stochastic Weather Generator
Acknowledgment at Cameroon Sites 1(3), 1–7.
Mutayoba, E., Kashaigili, J.J., 2017. Evaluation of the performance of the
CORDEX regional climate models in simulating rainfall characteristics over
The authors would like to acknowledge Debre Birhan Agricul- mbarali river catchment in the rufiji basin, Tanzania. J. Geosci. Environ. Protect.
tural Research Center (DBARC) for the financial support. Moreover, 5, 139–151.
the authors also acknowledged the National Metrological Agency Mxolisi, E.S.C.-A., 2015. An evaluation of CORDEX regional climate models in
simulating precipitation over Southern Africa. Atmos. Sci. Lett. 16, 199–207.
of Ethiopia for providing the necessary climatological data. Philbert, L.J.B., 2016. Evaluation of the performance of CORDEX regional climate
models in simulating present climate conditions of Tanzania. J. Southern
Hemisphere Earth Systems Sci. 66, 32–54.
Tamene, A. D., Chala, H. S., 2021. Assessment of the performance of CORDEX
References regional climate models in simulating rainfall and air temperature over
southwest Ethiopia. Heliyon 7, e07791:1 10.
Abiy, G.M., Tekalegn, A.W., Yihun, T.D., 2021. Evaluation of the performance of bias- Taylor, K.E., 2001. Summarizing multiple aspects of model performance in a single
corrected CORDEX regional climate models in reproducing Baro Akobo basin diagram. J. Geophys. Res. 106, 7183–7192.
climate. Theor. Appl. Climatol. 144, 751–767. Wilks, D.W., 1999. The weather generation game: a review of stochastic weather
Alemu, S.G., 2021. Performance evaluation of the cordex Africa regional climate models. Prog. Phys. Geogr. 23 (3), 329–357.
models in simulating rainfall and air temperature in the case of Wabi shebele Willmott, C.J., 1982. Bulletin. American Meteorological Society 63, 1309–1313.
Basin. Jimma University open access Institutional Repository, Ethiopia. Worku, G., Teferi, E., Bantider, A., Dile, Y.T., Taye, M.T., 2018. Evaluation of regional
Asfaw, A., Bantider, A., Simane, B., Hassen, A., 2021. Smallholder farmers’ livelihood climate models performance in simulating rainfall climatology of Jemma sub-
vulnerability to climate change-induced hazards: an agroecology-based basin, Upper Blue Nile Basin. Ethiopia. Dynam. Atmos. Oceans 83, 53–63.

102